tcp_input.c revision 1.225
1 /* $NetBSD: tcp_input.c,v 1.225 2005/03/29 20:10:16 yamt Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 34 * 35 * NRL grants permission for redistribution and use in source and binary 36 * forms, with or without modification, of the software and documentation 37 * created at NRL provided that the following conditions are met: 38 * 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. All advertising materials mentioning features or use of this software 45 * must display the following acknowledgements: 46 * This product includes software developed by the University of 47 * California, Berkeley and its contributors. 48 * This product includes software developed at the Information 49 * Technology Division, US Naval Research Laboratory. 50 * 4. Neither the name of the NRL nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 55 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 56 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 57 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 58 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 59 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 60 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 61 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 62 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 63 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 64 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 65 * 66 * The views and conclusions contained in the software and documentation 67 * are those of the authors and should not be interpreted as representing 68 * official policies, either expressed or implied, of the US Naval 69 * Research Laboratory (NRL). 70 */ 71 72 /*- 73 * Copyright (c) 1997, 1998, 1999, 2001, 2005 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 * This code is derived from software contributed to The NetBSD Foundation 80 * by Charles M. Hannum. 81 * 82 * Redistribution and use in source and binary forms, with or without 83 * modification, are permitted provided that the following conditions 84 * are met: 85 * 1. Redistributions of source code must retain the above copyright 86 * notice, this list of conditions and the following disclaimer. 87 * 2. Redistributions in binary form must reproduce the above copyright 88 * notice, this list of conditions and the following disclaimer in the 89 * documentation and/or other materials provided with the distribution. 90 * 3. All advertising materials mentioning features or use of this software 91 * must display the following acknowledgement: 92 * This product includes software developed by the NetBSD 93 * Foundation, Inc. and its contributors. 94 * 4. Neither the name of The NetBSD Foundation nor the names of its 95 * contributors may be used to endorse or promote products derived 96 * from this software without specific prior written permission. 97 * 98 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 99 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 100 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 101 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 102 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 103 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 104 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 105 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 106 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 107 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 108 * POSSIBILITY OF SUCH DAMAGE. 109 */ 110 111 /* 112 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 113 * The Regents of the University of California. All rights reserved. 114 * 115 * Redistribution and use in source and binary forms, with or without 116 * modification, are permitted provided that the following conditions 117 * are met: 118 * 1. Redistributions of source code must retain the above copyright 119 * notice, this list of conditions and the following disclaimer. 120 * 2. Redistributions in binary form must reproduce the above copyright 121 * notice, this list of conditions and the following disclaimer in the 122 * documentation and/or other materials provided with the distribution. 123 * 3. Neither the name of the University nor the names of its contributors 124 * may be used to endorse or promote products derived from this software 125 * without specific prior written permission. 126 * 127 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 128 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 129 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 130 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 131 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 132 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 133 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 134 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 135 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 136 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 137 * SUCH DAMAGE. 138 * 139 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 140 */ 141 142 /* 143 * TODO list for SYN cache stuff: 144 * 145 * Find room for a "state" field, which is needed to keep a 146 * compressed state for TIME_WAIT TCBs. It's been noted already 147 * that this is fairly important for very high-volume web and 148 * mail servers, which use a large number of short-lived 149 * connections. 150 */ 151 152 #include <sys/cdefs.h> 153 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.225 2005/03/29 20:10:16 yamt Exp $"); 154 155 #include "opt_inet.h" 156 #include "opt_ipsec.h" 157 #include "opt_inet_csum.h" 158 #include "opt_tcp_debug.h" 159 160 #include <sys/param.h> 161 #include <sys/systm.h> 162 #include <sys/malloc.h> 163 #include <sys/mbuf.h> 164 #include <sys/protosw.h> 165 #include <sys/socket.h> 166 #include <sys/socketvar.h> 167 #include <sys/errno.h> 168 #include <sys/syslog.h> 169 #include <sys/pool.h> 170 #include <sys/domain.h> 171 #include <sys/kernel.h> 172 #ifdef TCP_SIGNATURE 173 #include <sys/md5.h> 174 #endif 175 176 #include <net/if.h> 177 #include <net/route.h> 178 #include <net/if_types.h> 179 180 #include <netinet/in.h> 181 #include <netinet/in_systm.h> 182 #include <netinet/ip.h> 183 #include <netinet/in_pcb.h> 184 #include <netinet/in_var.h> 185 #include <netinet/ip_var.h> 186 187 #ifdef INET6 188 #ifndef INET 189 #include <netinet/in.h> 190 #endif 191 #include <netinet/ip6.h> 192 #include <netinet6/ip6_var.h> 193 #include <netinet6/in6_pcb.h> 194 #include <netinet6/ip6_var.h> 195 #include <netinet6/in6_var.h> 196 #include <netinet/icmp6.h> 197 #include <netinet6/nd6.h> 198 #endif 199 200 #ifndef INET6 201 /* always need ip6.h for IP6_EXTHDR_GET */ 202 #include <netinet/ip6.h> 203 #endif 204 205 #include <netinet/tcp.h> 206 #include <netinet/tcp_fsm.h> 207 #include <netinet/tcp_seq.h> 208 #include <netinet/tcp_timer.h> 209 #include <netinet/tcp_var.h> 210 #include <netinet/tcpip.h> 211 #include <netinet/tcp_debug.h> 212 213 #include <machine/stdarg.h> 214 215 #ifdef IPSEC 216 #include <netinet6/ipsec.h> 217 #include <netkey/key.h> 218 #endif /*IPSEC*/ 219 #ifdef INET6 220 #include "faith.h" 221 #if defined(NFAITH) && NFAITH > 0 222 #include <net/if_faith.h> 223 #endif 224 #endif /* IPSEC */ 225 226 #ifdef FAST_IPSEC 227 #include <netipsec/ipsec.h> 228 #include <netipsec/ipsec_var.h> /* XXX ipsecstat namespace */ 229 #include <netipsec/key.h> 230 #ifdef INET6 231 #include <netipsec/ipsec6.h> 232 #endif 233 #endif /* FAST_IPSEC*/ 234 235 int tcprexmtthresh = 3; 236 int tcp_log_refused; 237 238 static int tcp_rst_ppslim_count = 0; 239 static struct timeval tcp_rst_ppslim_last; 240 static int tcp_ackdrop_ppslim_count = 0; 241 static struct timeval tcp_ackdrop_ppslim_last; 242 243 #define TCP_PAWS_IDLE (24U * 24 * 60 * 60 * PR_SLOWHZ) 244 245 /* for modulo comparisons of timestamps */ 246 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 247 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 248 249 /* 250 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 251 */ 252 #ifdef INET6 253 #define ND6_HINT(tp) \ 254 do { \ 255 if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \ 256 tp->t_in6pcb->in6p_route.ro_rt) { \ 257 nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \ 258 } \ 259 } while (/*CONSTCOND*/ 0) 260 #else 261 #define ND6_HINT(tp) 262 #endif 263 264 /* 265 * Macro to compute ACK transmission behavior. Delay the ACK unless 266 * we have already delayed an ACK (must send an ACK every two segments). 267 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 268 * option is enabled. 269 */ 270 #define TCP_SETUP_ACK(tp, th) \ 271 do { \ 272 if ((tp)->t_flags & TF_DELACK || \ 273 (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \ 274 tp->t_flags |= TF_ACKNOW; \ 275 else \ 276 TCP_SET_DELACK(tp); \ 277 } while (/*CONSTCOND*/ 0) 278 279 /* 280 * Convert TCP protocol fields to host order for easier processing. 281 */ 282 #define TCP_FIELDS_TO_HOST(th) \ 283 do { \ 284 NTOHL((th)->th_seq); \ 285 NTOHL((th)->th_ack); \ 286 NTOHS((th)->th_win); \ 287 NTOHS((th)->th_urp); \ 288 } while (/*CONSTCOND*/ 0) 289 290 /* 291 * ... and reverse the above. 292 */ 293 #define TCP_FIELDS_TO_NET(th) \ 294 do { \ 295 HTONL((th)->th_seq); \ 296 HTONL((th)->th_ack); \ 297 HTONS((th)->th_win); \ 298 HTONS((th)->th_urp); \ 299 } while (/*CONSTCOND*/ 0) 300 301 #ifdef TCP_CSUM_COUNTERS 302 #include <sys/device.h> 303 304 extern struct evcnt tcp_hwcsum_ok; 305 extern struct evcnt tcp_hwcsum_bad; 306 extern struct evcnt tcp_hwcsum_data; 307 extern struct evcnt tcp_swcsum; 308 309 #define TCP_CSUM_COUNTER_INCR(ev) (ev)->ev_count++ 310 311 #else 312 313 #define TCP_CSUM_COUNTER_INCR(ev) /* nothing */ 314 315 #endif /* TCP_CSUM_COUNTERS */ 316 317 #ifdef TCP_REASS_COUNTERS 318 #include <sys/device.h> 319 320 extern struct evcnt tcp_reass_; 321 extern struct evcnt tcp_reass_empty; 322 extern struct evcnt tcp_reass_iteration[8]; 323 extern struct evcnt tcp_reass_prependfirst; 324 extern struct evcnt tcp_reass_prepend; 325 extern struct evcnt tcp_reass_insert; 326 extern struct evcnt tcp_reass_inserttail; 327 extern struct evcnt tcp_reass_append; 328 extern struct evcnt tcp_reass_appendtail; 329 extern struct evcnt tcp_reass_overlaptail; 330 extern struct evcnt tcp_reass_overlapfront; 331 extern struct evcnt tcp_reass_segdup; 332 extern struct evcnt tcp_reass_fragdup; 333 334 #define TCP_REASS_COUNTER_INCR(ev) (ev)->ev_count++ 335 336 #else 337 338 #define TCP_REASS_COUNTER_INCR(ev) /* nothing */ 339 340 #endif /* TCP_REASS_COUNTERS */ 341 342 #ifdef INET 343 static void tcp4_log_refused(const struct ip *, const struct tcphdr *); 344 #endif 345 #ifdef INET6 346 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *); 347 #endif 348 349 #define TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next 350 351 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL); 352 353 struct ipqent * 354 tcpipqent_alloc() 355 { 356 struct ipqent *ipqe; 357 int s; 358 359 s = splvm(); 360 ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT); 361 splx(s); 362 363 return ipqe; 364 } 365 366 void 367 tcpipqent_free(struct ipqent *ipqe) 368 { 369 int s; 370 371 s = splvm(); 372 pool_put(&tcpipqent_pool, ipqe); 373 splx(s); 374 } 375 376 int 377 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen) 378 { 379 struct ipqent *p, *q, *nq, *tiqe = NULL; 380 struct socket *so = NULL; 381 int pkt_flags; 382 tcp_seq pkt_seq; 383 unsigned pkt_len; 384 u_long rcvpartdupbyte = 0; 385 u_long rcvoobyte; 386 #ifdef TCP_REASS_COUNTERS 387 u_int count = 0; 388 #endif 389 390 if (tp->t_inpcb) 391 so = tp->t_inpcb->inp_socket; 392 #ifdef INET6 393 else if (tp->t_in6pcb) 394 so = tp->t_in6pcb->in6p_socket; 395 #endif 396 397 TCP_REASS_LOCK_CHECK(tp); 398 399 /* 400 * Call with th==0 after become established to 401 * force pre-ESTABLISHED data up to user socket. 402 */ 403 if (th == 0) 404 goto present; 405 406 rcvoobyte = *tlen; 407 /* 408 * Copy these to local variables because the tcpiphdr 409 * gets munged while we are collapsing mbufs. 410 */ 411 pkt_seq = th->th_seq; 412 pkt_len = *tlen; 413 pkt_flags = th->th_flags; 414 415 TCP_REASS_COUNTER_INCR(&tcp_reass_); 416 417 if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) { 418 /* 419 * When we miss a packet, the vast majority of time we get 420 * packets that follow it in order. So optimize for that. 421 */ 422 if (pkt_seq == p->ipqe_seq + p->ipqe_len) { 423 p->ipqe_len += pkt_len; 424 p->ipqe_flags |= pkt_flags; 425 m_cat(p->ipre_mlast, m); 426 TRAVERSE(p->ipre_mlast); 427 m = NULL; 428 tiqe = p; 429 TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq); 430 TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail); 431 goto skip_replacement; 432 } 433 /* 434 * While we're here, if the pkt is completely beyond 435 * anything we have, just insert it at the tail. 436 */ 437 if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) { 438 TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail); 439 goto insert_it; 440 } 441 } 442 443 q = TAILQ_FIRST(&tp->segq); 444 445 if (q != NULL) { 446 /* 447 * If this segment immediately precedes the first out-of-order 448 * block, simply slap the segment in front of it and (mostly) 449 * skip the complicated logic. 450 */ 451 if (pkt_seq + pkt_len == q->ipqe_seq) { 452 q->ipqe_seq = pkt_seq; 453 q->ipqe_len += pkt_len; 454 q->ipqe_flags |= pkt_flags; 455 m_cat(m, q->ipqe_m); 456 q->ipqe_m = m; 457 q->ipre_mlast = m; /* last mbuf may have changed */ 458 TRAVERSE(q->ipre_mlast); 459 tiqe = q; 460 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 461 TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst); 462 goto skip_replacement; 463 } 464 } else { 465 TCP_REASS_COUNTER_INCR(&tcp_reass_empty); 466 } 467 468 /* 469 * Find a segment which begins after this one does. 470 */ 471 for (p = NULL; q != NULL; q = nq) { 472 nq = TAILQ_NEXT(q, ipqe_q); 473 #ifdef TCP_REASS_COUNTERS 474 count++; 475 #endif 476 /* 477 * If the received segment is just right after this 478 * fragment, merge the two together and then check 479 * for further overlaps. 480 */ 481 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 482 #ifdef TCPREASS_DEBUG 483 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 484 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 485 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 486 #endif 487 pkt_len += q->ipqe_len; 488 pkt_flags |= q->ipqe_flags; 489 pkt_seq = q->ipqe_seq; 490 m_cat(q->ipre_mlast, m); 491 TRAVERSE(q->ipre_mlast); 492 m = q->ipqe_m; 493 TCP_REASS_COUNTER_INCR(&tcp_reass_append); 494 goto free_ipqe; 495 } 496 /* 497 * If the received segment is completely past this 498 * fragment, we need to go the next fragment. 499 */ 500 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 501 p = q; 502 continue; 503 } 504 /* 505 * If the fragment is past the received segment, 506 * it (or any following) can't be concatenated. 507 */ 508 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) { 509 TCP_REASS_COUNTER_INCR(&tcp_reass_insert); 510 break; 511 } 512 513 /* 514 * We've received all the data in this segment before. 515 * mark it as a duplicate and return. 516 */ 517 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 518 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 519 tcpstat.tcps_rcvduppack++; 520 tcpstat.tcps_rcvdupbyte += pkt_len; 521 tcp_new_dsack(tp, pkt_seq, pkt_len); 522 m_freem(m); 523 if (tiqe != NULL) { 524 tcpipqent_free(tiqe); 525 } 526 TCP_REASS_COUNTER_INCR(&tcp_reass_segdup); 527 return (0); 528 } 529 /* 530 * Received segment completely overlaps this fragment 531 * so we drop the fragment (this keeps the temporal 532 * ordering of segments correct). 533 */ 534 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 535 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 536 rcvpartdupbyte += q->ipqe_len; 537 m_freem(q->ipqe_m); 538 TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup); 539 goto free_ipqe; 540 } 541 /* 542 * RX'ed segment extends past the end of the 543 * fragment. Drop the overlapping bytes. Then 544 * merge the fragment and segment then treat as 545 * a longer received packet. 546 */ 547 if (SEQ_LT(q->ipqe_seq, pkt_seq) && 548 SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 549 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 550 #ifdef TCPREASS_DEBUG 551 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 552 tp, overlap, 553 pkt_seq, pkt_seq + pkt_len, pkt_len); 554 #endif 555 m_adj(m, overlap); 556 rcvpartdupbyte += overlap; 557 m_cat(q->ipre_mlast, m); 558 TRAVERSE(q->ipre_mlast); 559 m = q->ipqe_m; 560 pkt_seq = q->ipqe_seq; 561 pkt_len += q->ipqe_len - overlap; 562 rcvoobyte -= overlap; 563 TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail); 564 goto free_ipqe; 565 } 566 /* 567 * RX'ed segment extends past the front of the 568 * fragment. Drop the overlapping bytes on the 569 * received packet. The packet will then be 570 * contatentated with this fragment a bit later. 571 */ 572 if (SEQ_GT(q->ipqe_seq, pkt_seq) && 573 SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 574 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 575 #ifdef TCPREASS_DEBUG 576 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 577 tp, overlap, 578 pkt_seq, pkt_seq + pkt_len, pkt_len); 579 #endif 580 m_adj(m, -overlap); 581 pkt_len -= overlap; 582 rcvpartdupbyte += overlap; 583 TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront); 584 rcvoobyte -= overlap; 585 } 586 /* 587 * If the received segment immediates precedes this 588 * fragment then tack the fragment onto this segment 589 * and reinsert the data. 590 */ 591 if (q->ipqe_seq == pkt_seq + pkt_len) { 592 #ifdef TCPREASS_DEBUG 593 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 594 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 595 pkt_seq, pkt_seq + pkt_len, pkt_len); 596 #endif 597 pkt_len += q->ipqe_len; 598 pkt_flags |= q->ipqe_flags; 599 m_cat(m, q->ipqe_m); 600 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 601 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 602 tp->t_segqlen--; 603 KASSERT(tp->t_segqlen >= 0); 604 KASSERT(tp->t_segqlen != 0 || 605 (TAILQ_EMPTY(&tp->segq) && 606 TAILQ_EMPTY(&tp->timeq))); 607 if (tiqe == NULL) { 608 tiqe = q; 609 } else { 610 tcpipqent_free(q); 611 } 612 TCP_REASS_COUNTER_INCR(&tcp_reass_prepend); 613 break; 614 } 615 /* 616 * If the fragment is before the segment, remember it. 617 * When this loop is terminated, p will contain the 618 * pointer to fragment that is right before the received 619 * segment. 620 */ 621 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 622 p = q; 623 624 continue; 625 626 /* 627 * This is a common operation. It also will allow 628 * to save doing a malloc/free in most instances. 629 */ 630 free_ipqe: 631 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 632 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 633 tp->t_segqlen--; 634 KASSERT(tp->t_segqlen >= 0); 635 KASSERT(tp->t_segqlen != 0 || 636 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 637 if (tiqe == NULL) { 638 tiqe = q; 639 } else { 640 tcpipqent_free(q); 641 } 642 } 643 644 #ifdef TCP_REASS_COUNTERS 645 if (count > 7) 646 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]); 647 else if (count > 0) 648 TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]); 649 #endif 650 651 insert_it: 652 653 /* 654 * Allocate a new queue entry since the received segment did not 655 * collapse onto any other out-of-order block; thus we are allocating 656 * a new block. If it had collapsed, tiqe would not be NULL and 657 * we would be reusing it. 658 * XXX If we can't, just drop the packet. XXX 659 */ 660 if (tiqe == NULL) { 661 tiqe = tcpipqent_alloc(); 662 if (tiqe == NULL) { 663 tcpstat.tcps_rcvmemdrop++; 664 m_freem(m); 665 return (0); 666 } 667 } 668 669 /* 670 * Update the counters. 671 */ 672 tcpstat.tcps_rcvoopack++; 673 tcpstat.tcps_rcvoobyte += rcvoobyte; 674 if (rcvpartdupbyte) { 675 tcpstat.tcps_rcvpartduppack++; 676 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 677 } 678 679 /* 680 * Insert the new fragment queue entry into both queues. 681 */ 682 tiqe->ipqe_m = m; 683 tiqe->ipre_mlast = m; 684 tiqe->ipqe_seq = pkt_seq; 685 tiqe->ipqe_len = pkt_len; 686 tiqe->ipqe_flags = pkt_flags; 687 if (p == NULL) { 688 TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 689 #ifdef TCPREASS_DEBUG 690 if (tiqe->ipqe_seq != tp->rcv_nxt) 691 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 692 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 693 #endif 694 } else { 695 TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q); 696 #ifdef TCPREASS_DEBUG 697 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 698 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 699 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 700 #endif 701 } 702 tp->t_segqlen++; 703 704 skip_replacement: 705 706 TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 707 708 present: 709 /* 710 * Present data to user, advancing rcv_nxt through 711 * completed sequence space. 712 */ 713 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 714 return (0); 715 q = TAILQ_FIRST(&tp->segq); 716 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 717 return (0); 718 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 719 return (0); 720 721 tp->rcv_nxt += q->ipqe_len; 722 pkt_flags = q->ipqe_flags & TH_FIN; 723 ND6_HINT(tp); 724 725 TAILQ_REMOVE(&tp->segq, q, ipqe_q); 726 TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq); 727 tp->t_segqlen--; 728 KASSERT(tp->t_segqlen >= 0); 729 KASSERT(tp->t_segqlen != 0 || 730 (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq))); 731 if (so->so_state & SS_CANTRCVMORE) 732 m_freem(q->ipqe_m); 733 else 734 sbappendstream(&so->so_rcv, q->ipqe_m); 735 tcpipqent_free(q); 736 sorwakeup(so); 737 return (pkt_flags); 738 } 739 740 #ifdef INET6 741 int 742 tcp6_input(struct mbuf **mp, int *offp, int proto) 743 { 744 struct mbuf *m = *mp; 745 746 /* 747 * draft-itojun-ipv6-tcp-to-anycast 748 * better place to put this in? 749 */ 750 if (m->m_flags & M_ANYCAST6) { 751 struct ip6_hdr *ip6; 752 if (m->m_len < sizeof(struct ip6_hdr)) { 753 if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { 754 tcpstat.tcps_rcvshort++; 755 return IPPROTO_DONE; 756 } 757 } 758 ip6 = mtod(m, struct ip6_hdr *); 759 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 760 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 761 return IPPROTO_DONE; 762 } 763 764 tcp_input(m, *offp, proto); 765 return IPPROTO_DONE; 766 } 767 #endif 768 769 #ifdef INET 770 static void 771 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th) 772 { 773 char src[4*sizeof "123"]; 774 char dst[4*sizeof "123"]; 775 776 if (ip) { 777 strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src)); 778 strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst)); 779 } 780 else { 781 strlcpy(src, "(unknown)", sizeof(src)); 782 strlcpy(dst, "(unknown)", sizeof(dst)); 783 } 784 log(LOG_INFO, 785 "Connection attempt to TCP %s:%d from %s:%d\n", 786 dst, ntohs(th->th_dport), 787 src, ntohs(th->th_sport)); 788 } 789 #endif 790 791 #ifdef INET6 792 static void 793 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th) 794 { 795 char src[INET6_ADDRSTRLEN]; 796 char dst[INET6_ADDRSTRLEN]; 797 798 if (ip6) { 799 strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src)); 800 strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst)); 801 } 802 else { 803 strlcpy(src, "(unknown v6)", sizeof(src)); 804 strlcpy(dst, "(unknown v6)", sizeof(dst)); 805 } 806 log(LOG_INFO, 807 "Connection attempt to TCP [%s]:%d from [%s]:%d\n", 808 dst, ntohs(th->th_dport), 809 src, ntohs(th->th_sport)); 810 } 811 #endif 812 813 /* 814 * Checksum extended TCP header and data. 815 */ 816 int 817 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th, int toff, 818 int off, int tlen) 819 { 820 821 /* 822 * XXX it's better to record and check if this mbuf is 823 * already checked. 824 */ 825 826 switch (af) { 827 #ifdef INET 828 case AF_INET: 829 switch (m->m_pkthdr.csum_flags & 830 ((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) | 831 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) { 832 case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD: 833 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad); 834 goto badcsum; 835 836 case M_CSUM_TCPv4|M_CSUM_DATA: { 837 u_int32_t hw_csum = m->m_pkthdr.csum_data; 838 839 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data); 840 if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) { 841 const struct ip *ip = 842 mtod(m, const struct ip *); 843 844 hw_csum = in_cksum_phdr(ip->ip_src.s_addr, 845 ip->ip_dst.s_addr, 846 htons(hw_csum + tlen + off + IPPROTO_TCP)); 847 } 848 if ((hw_csum ^ 0xffff) != 0) 849 goto badcsum; 850 break; 851 } 852 853 case M_CSUM_TCPv4: 854 /* Checksum was okay. */ 855 TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok); 856 break; 857 858 default: 859 /* 860 * Must compute it ourselves. Maybe skip checksum 861 * on loopback interfaces. 862 */ 863 if (__predict_true(!(m->m_pkthdr.rcvif->if_flags & 864 IFF_LOOPBACK) || 865 tcp_do_loopback_cksum)) { 866 TCP_CSUM_COUNTER_INCR(&tcp_swcsum); 867 if (in4_cksum(m, IPPROTO_TCP, toff, 868 tlen + off) != 0) 869 goto badcsum; 870 } 871 break; 872 } 873 break; 874 #endif /* INET4 */ 875 876 #ifdef INET6 877 case AF_INET6: 878 if (__predict_true((m->m_flags & M_LOOP) == 0 || 879 tcp_do_loopback_cksum)) { 880 if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0) 881 goto badcsum; 882 } 883 break; 884 #endif /* INET6 */ 885 } 886 887 return 0; 888 889 badcsum: 890 tcpstat.tcps_rcvbadsum++; 891 return -1; 892 } 893 894 /* 895 * TCP input routine, follows pages 65-76 of the 896 * protocol specification dated September, 1981 very closely. 897 */ 898 void 899 tcp_input(struct mbuf *m, ...) 900 { 901 struct tcphdr *th; 902 struct ip *ip; 903 struct inpcb *inp; 904 #ifdef INET6 905 struct ip6_hdr *ip6; 906 struct in6pcb *in6p; 907 #endif 908 u_int8_t *optp = NULL; 909 int optlen = 0; 910 int len, tlen, toff, hdroptlen = 0; 911 struct tcpcb *tp = 0; 912 int tiflags; 913 struct socket *so = NULL; 914 int todrop, dupseg, acked, ourfinisacked, needoutput = 0; 915 #ifdef TCP_DEBUG 916 short ostate = 0; 917 #endif 918 int iss = 0; 919 u_long tiwin; 920 struct tcp_opt_info opti; 921 int off, iphlen; 922 va_list ap; 923 int af; /* af on the wire */ 924 struct mbuf *tcp_saveti = NULL; 925 926 MCLAIM(m, &tcp_rx_mowner); 927 va_start(ap, m); 928 toff = va_arg(ap, int); 929 (void)va_arg(ap, int); /* ignore value, advance ap */ 930 va_end(ap); 931 932 tcpstat.tcps_rcvtotal++; 933 934 bzero(&opti, sizeof(opti)); 935 opti.ts_present = 0; 936 opti.maxseg = 0; 937 938 /* 939 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 940 * 941 * TCP is, by definition, unicast, so we reject all 942 * multicast outright. 943 * 944 * Note, there are additional src/dst address checks in 945 * the AF-specific code below. 946 */ 947 if (m->m_flags & (M_BCAST|M_MCAST)) { 948 /* XXX stat */ 949 goto drop; 950 } 951 #ifdef INET6 952 if (m->m_flags & M_ANYCAST6) { 953 /* XXX stat */ 954 goto drop; 955 } 956 #endif 957 958 /* 959 * Get IP and TCP header. 960 * Note: IP leaves IP header in first mbuf. 961 */ 962 ip = mtod(m, struct ip *); 963 #ifdef INET6 964 ip6 = NULL; 965 #endif 966 switch (ip->ip_v) { 967 #ifdef INET 968 case 4: 969 af = AF_INET; 970 iphlen = sizeof(struct ip); 971 ip = mtod(m, struct ip *); 972 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 973 sizeof(struct tcphdr)); 974 if (th == NULL) { 975 tcpstat.tcps_rcvshort++; 976 return; 977 } 978 /* We do the checksum after PCB lookup... */ 979 len = ntohs(ip->ip_len); 980 tlen = len - toff; 981 break; 982 #endif 983 #ifdef INET6 984 case 6: 985 ip = NULL; 986 iphlen = sizeof(struct ip6_hdr); 987 af = AF_INET6; 988 ip6 = mtod(m, struct ip6_hdr *); 989 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 990 sizeof(struct tcphdr)); 991 if (th == NULL) { 992 tcpstat.tcps_rcvshort++; 993 return; 994 } 995 996 /* Be proactive about malicious use of IPv4 mapped address */ 997 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 998 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 999 /* XXX stat */ 1000 goto drop; 1001 } 1002 1003 /* 1004 * Be proactive about unspecified IPv6 address in source. 1005 * As we use all-zero to indicate unbounded/unconnected pcb, 1006 * unspecified IPv6 address can be used to confuse us. 1007 * 1008 * Note that packets with unspecified IPv6 destination is 1009 * already dropped in ip6_input. 1010 */ 1011 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1012 /* XXX stat */ 1013 goto drop; 1014 } 1015 1016 /* 1017 * Make sure destination address is not multicast. 1018 * Source address checked in ip6_input(). 1019 */ 1020 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1021 /* XXX stat */ 1022 goto drop; 1023 } 1024 1025 /* We do the checksum after PCB lookup... */ 1026 len = m->m_pkthdr.len; 1027 tlen = len - toff; 1028 break; 1029 #endif 1030 default: 1031 m_freem(m); 1032 return; 1033 } 1034 1035 KASSERT(TCP_HDR_ALIGNED_P(th)); 1036 1037 /* 1038 * Check that TCP offset makes sense, 1039 * pull out TCP options and adjust length. XXX 1040 */ 1041 off = th->th_off << 2; 1042 if (off < sizeof (struct tcphdr) || off > tlen) { 1043 tcpstat.tcps_rcvbadoff++; 1044 goto drop; 1045 } 1046 tlen -= off; 1047 1048 /* 1049 * tcp_input() has been modified to use tlen to mean the TCP data 1050 * length throughout the function. Other functions can use 1051 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1052 * rja 1053 */ 1054 1055 if (off > sizeof (struct tcphdr)) { 1056 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1057 if (th == NULL) { 1058 tcpstat.tcps_rcvshort++; 1059 return; 1060 } 1061 /* 1062 * NOTE: ip/ip6 will not be affected by m_pulldown() 1063 * (as they're before toff) and we don't need to update those. 1064 */ 1065 KASSERT(TCP_HDR_ALIGNED_P(th)); 1066 optlen = off - sizeof (struct tcphdr); 1067 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1068 /* 1069 * Do quick retrieval of timestamp options ("options 1070 * prediction?"). If timestamp is the only option and it's 1071 * formatted as recommended in RFC 1323 appendix A, we 1072 * quickly get the values now and not bother calling 1073 * tcp_dooptions(), etc. 1074 */ 1075 if ((optlen == TCPOLEN_TSTAMP_APPA || 1076 (optlen > TCPOLEN_TSTAMP_APPA && 1077 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1078 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1079 (th->th_flags & TH_SYN) == 0) { 1080 opti.ts_present = 1; 1081 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1082 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1083 optp = NULL; /* we've parsed the options */ 1084 } 1085 } 1086 tiflags = th->th_flags; 1087 1088 /* 1089 * Locate pcb for segment. 1090 */ 1091 findpcb: 1092 inp = NULL; 1093 #ifdef INET6 1094 in6p = NULL; 1095 #endif 1096 switch (af) { 1097 #ifdef INET 1098 case AF_INET: 1099 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1100 ip->ip_dst, th->th_dport); 1101 if (inp == 0) { 1102 ++tcpstat.tcps_pcbhashmiss; 1103 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1104 } 1105 #ifdef INET6 1106 if (inp == 0) { 1107 struct in6_addr s, d; 1108 1109 /* mapped addr case */ 1110 bzero(&s, sizeof(s)); 1111 s.s6_addr16[5] = htons(0xffff); 1112 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1113 bzero(&d, sizeof(d)); 1114 d.s6_addr16[5] = htons(0xffff); 1115 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1116 in6p = in6_pcblookup_connect(&tcbtable, &s, 1117 th->th_sport, &d, th->th_dport, 0); 1118 if (in6p == 0) { 1119 ++tcpstat.tcps_pcbhashmiss; 1120 in6p = in6_pcblookup_bind(&tcbtable, &d, 1121 th->th_dport, 0); 1122 } 1123 } 1124 #endif 1125 #ifndef INET6 1126 if (inp == 0) 1127 #else 1128 if (inp == 0 && in6p == 0) 1129 #endif 1130 { 1131 ++tcpstat.tcps_noport; 1132 if (tcp_log_refused && 1133 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1134 tcp4_log_refused(ip, th); 1135 } 1136 TCP_FIELDS_TO_HOST(th); 1137 goto dropwithreset_ratelim; 1138 } 1139 #if defined(IPSEC) || defined(FAST_IPSEC) 1140 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1141 ipsec4_in_reject(m, inp)) { 1142 ipsecstat.in_polvio++; 1143 goto drop; 1144 } 1145 #ifdef INET6 1146 else if (in6p && 1147 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1148 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1149 ipsecstat.in_polvio++; 1150 goto drop; 1151 } 1152 #endif 1153 #endif /*IPSEC*/ 1154 break; 1155 #endif /*INET*/ 1156 #ifdef INET6 1157 case AF_INET6: 1158 { 1159 int faith; 1160 1161 #if defined(NFAITH) && NFAITH > 0 1162 faith = faithprefix(&ip6->ip6_dst); 1163 #else 1164 faith = 0; 1165 #endif 1166 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1167 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1168 if (in6p == NULL) { 1169 ++tcpstat.tcps_pcbhashmiss; 1170 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1171 th->th_dport, faith); 1172 } 1173 if (in6p == NULL) { 1174 ++tcpstat.tcps_noport; 1175 if (tcp_log_refused && 1176 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1177 tcp6_log_refused(ip6, th); 1178 } 1179 TCP_FIELDS_TO_HOST(th); 1180 goto dropwithreset_ratelim; 1181 } 1182 #if defined(IPSEC) || defined(FAST_IPSEC) 1183 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1184 ipsec6_in_reject(m, in6p)) { 1185 ipsec6stat.in_polvio++; 1186 goto drop; 1187 } 1188 #endif /*IPSEC*/ 1189 break; 1190 } 1191 #endif 1192 } 1193 1194 /* 1195 * If the state is CLOSED (i.e., TCB does not exist) then 1196 * all data in the incoming segment is discarded. 1197 * If the TCB exists but is in CLOSED state, it is embryonic, 1198 * but should either do a listen or a connect soon. 1199 */ 1200 tp = NULL; 1201 so = NULL; 1202 if (inp) { 1203 tp = intotcpcb(inp); 1204 so = inp->inp_socket; 1205 } 1206 #ifdef INET6 1207 else if (in6p) { 1208 tp = in6totcpcb(in6p); 1209 so = in6p->in6p_socket; 1210 } 1211 #endif 1212 if (tp == 0) { 1213 TCP_FIELDS_TO_HOST(th); 1214 goto dropwithreset_ratelim; 1215 } 1216 if (tp->t_state == TCPS_CLOSED) 1217 goto drop; 1218 1219 /* 1220 * Checksum extended TCP header and data. 1221 */ 1222 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1223 goto badcsum; 1224 1225 TCP_FIELDS_TO_HOST(th); 1226 1227 /* Unscale the window into a 32-bit value. */ 1228 if ((tiflags & TH_SYN) == 0) 1229 tiwin = th->th_win << tp->snd_scale; 1230 else 1231 tiwin = th->th_win; 1232 1233 #ifdef INET6 1234 /* save packet options if user wanted */ 1235 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1236 if (in6p->in6p_options) { 1237 m_freem(in6p->in6p_options); 1238 in6p->in6p_options = 0; 1239 } 1240 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1241 } 1242 #endif 1243 1244 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1245 union syn_cache_sa src; 1246 union syn_cache_sa dst; 1247 1248 bzero(&src, sizeof(src)); 1249 bzero(&dst, sizeof(dst)); 1250 switch (af) { 1251 #ifdef INET 1252 case AF_INET: 1253 src.sin.sin_len = sizeof(struct sockaddr_in); 1254 src.sin.sin_family = AF_INET; 1255 src.sin.sin_addr = ip->ip_src; 1256 src.sin.sin_port = th->th_sport; 1257 1258 dst.sin.sin_len = sizeof(struct sockaddr_in); 1259 dst.sin.sin_family = AF_INET; 1260 dst.sin.sin_addr = ip->ip_dst; 1261 dst.sin.sin_port = th->th_dport; 1262 break; 1263 #endif 1264 #ifdef INET6 1265 case AF_INET6: 1266 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1267 src.sin6.sin6_family = AF_INET6; 1268 src.sin6.sin6_addr = ip6->ip6_src; 1269 src.sin6.sin6_port = th->th_sport; 1270 1271 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1272 dst.sin6.sin6_family = AF_INET6; 1273 dst.sin6.sin6_addr = ip6->ip6_dst; 1274 dst.sin6.sin6_port = th->th_dport; 1275 break; 1276 #endif /* INET6 */ 1277 default: 1278 goto badsyn; /*sanity*/ 1279 } 1280 1281 if (so->so_options & SO_DEBUG) { 1282 #ifdef TCP_DEBUG 1283 ostate = tp->t_state; 1284 #endif 1285 1286 tcp_saveti = NULL; 1287 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1288 goto nosave; 1289 1290 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1291 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1292 if (!tcp_saveti) 1293 goto nosave; 1294 } else { 1295 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1296 if (!tcp_saveti) 1297 goto nosave; 1298 MCLAIM(m, &tcp_mowner); 1299 tcp_saveti->m_len = iphlen; 1300 m_copydata(m, 0, iphlen, 1301 mtod(tcp_saveti, caddr_t)); 1302 } 1303 1304 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1305 m_freem(tcp_saveti); 1306 tcp_saveti = NULL; 1307 } else { 1308 tcp_saveti->m_len += sizeof(struct tcphdr); 1309 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1310 sizeof(struct tcphdr)); 1311 } 1312 nosave:; 1313 } 1314 if (so->so_options & SO_ACCEPTCONN) { 1315 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1316 if (tiflags & TH_RST) { 1317 syn_cache_reset(&src.sa, &dst.sa, th); 1318 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1319 (TH_ACK|TH_SYN)) { 1320 /* 1321 * Received a SYN,ACK. This should 1322 * never happen while we are in 1323 * LISTEN. Send an RST. 1324 */ 1325 goto badsyn; 1326 } else if (tiflags & TH_ACK) { 1327 so = syn_cache_get(&src.sa, &dst.sa, 1328 th, toff, tlen, so, m); 1329 if (so == NULL) { 1330 /* 1331 * We don't have a SYN for 1332 * this ACK; send an RST. 1333 */ 1334 goto badsyn; 1335 } else if (so == 1336 (struct socket *)(-1)) { 1337 /* 1338 * We were unable to create 1339 * the connection. If the 1340 * 3-way handshake was 1341 * completed, and RST has 1342 * been sent to the peer. 1343 * Since the mbuf might be 1344 * in use for the reply, 1345 * do not free it. 1346 */ 1347 m = NULL; 1348 } else { 1349 /* 1350 * We have created a 1351 * full-blown connection. 1352 */ 1353 tp = NULL; 1354 inp = NULL; 1355 #ifdef INET6 1356 in6p = NULL; 1357 #endif 1358 switch (so->so_proto->pr_domain->dom_family) { 1359 #ifdef INET 1360 case AF_INET: 1361 inp = sotoinpcb(so); 1362 tp = intotcpcb(inp); 1363 break; 1364 #endif 1365 #ifdef INET6 1366 case AF_INET6: 1367 in6p = sotoin6pcb(so); 1368 tp = in6totcpcb(in6p); 1369 break; 1370 #endif 1371 } 1372 if (tp == NULL) 1373 goto badsyn; /*XXX*/ 1374 tiwin <<= tp->snd_scale; 1375 goto after_listen; 1376 } 1377 } else { 1378 /* 1379 * None of RST, SYN or ACK was set. 1380 * This is an invalid packet for a 1381 * TCB in LISTEN state. Send a RST. 1382 */ 1383 goto badsyn; 1384 } 1385 } else { 1386 /* 1387 * Received a SYN. 1388 */ 1389 1390 #ifdef INET6 1391 /* 1392 * If deprecated address is forbidden, we do 1393 * not accept SYN to deprecated interface 1394 * address to prevent any new inbound 1395 * connection from getting established. 1396 * When we do not accept SYN, we send a TCP 1397 * RST, with deprecated source address (instead 1398 * of dropping it). We compromise it as it is 1399 * much better for peer to send a RST, and 1400 * RST will be the final packet for the 1401 * exchange. 1402 * 1403 * If we do not forbid deprecated addresses, we 1404 * accept the SYN packet. RFC2462 does not 1405 * suggest dropping SYN in this case. 1406 * If we decipher RFC2462 5.5.4, it says like 1407 * this: 1408 * 1. use of deprecated addr with existing 1409 * communication is okay - "SHOULD continue 1410 * to be used" 1411 * 2. use of it with new communication: 1412 * (2a) "SHOULD NOT be used if alternate 1413 * address with sufficient scope is 1414 * available" 1415 * (2b) nothing mentioned otherwise. 1416 * Here we fall into (2b) case as we have no 1417 * choice in our source address selection - we 1418 * must obey the peer. 1419 * 1420 * The wording in RFC2462 is confusing, and 1421 * there are multiple description text for 1422 * deprecated address handling - worse, they 1423 * are not exactly the same. I believe 5.5.4 1424 * is the best one, so we follow 5.5.4. 1425 */ 1426 if (af == AF_INET6 && !ip6_use_deprecated) { 1427 struct in6_ifaddr *ia6; 1428 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1429 &ip6->ip6_dst)) && 1430 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1431 tp = NULL; 1432 goto dropwithreset; 1433 } 1434 } 1435 #endif 1436 1437 #ifdef IPSEC 1438 switch (af) { 1439 #ifdef INET 1440 case AF_INET: 1441 if (ipsec4_in_reject_so(m, so)) { 1442 ipsecstat.in_polvio++; 1443 tp = NULL; 1444 goto dropwithreset; 1445 } 1446 break; 1447 #endif 1448 #ifdef INET6 1449 case AF_INET6: 1450 if (ipsec6_in_reject_so(m, so)) { 1451 ipsec6stat.in_polvio++; 1452 tp = NULL; 1453 goto dropwithreset; 1454 } 1455 break; 1456 #endif 1457 } 1458 #endif 1459 1460 /* 1461 * LISTEN socket received a SYN 1462 * from itself? This can't possibly 1463 * be valid; drop the packet. 1464 */ 1465 if (th->th_sport == th->th_dport) { 1466 int i; 1467 1468 switch (af) { 1469 #ifdef INET 1470 case AF_INET: 1471 i = in_hosteq(ip->ip_src, ip->ip_dst); 1472 break; 1473 #endif 1474 #ifdef INET6 1475 case AF_INET6: 1476 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1477 break; 1478 #endif 1479 default: 1480 i = 1; 1481 } 1482 if (i) { 1483 tcpstat.tcps_badsyn++; 1484 goto drop; 1485 } 1486 } 1487 1488 /* 1489 * SYN looks ok; create compressed TCP 1490 * state for it. 1491 */ 1492 if (so->so_qlen <= so->so_qlimit && 1493 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1494 so, m, optp, optlen, &opti)) 1495 m = NULL; 1496 } 1497 goto drop; 1498 } 1499 } 1500 1501 after_listen: 1502 #ifdef DIAGNOSTIC 1503 /* 1504 * Should not happen now that all embryonic connections 1505 * are handled with compressed state. 1506 */ 1507 if (tp->t_state == TCPS_LISTEN) 1508 panic("tcp_input: TCPS_LISTEN"); 1509 #endif 1510 1511 /* 1512 * Segment received on connection. 1513 * Reset idle time and keep-alive timer. 1514 */ 1515 tp->t_rcvtime = tcp_now; 1516 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1517 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1518 1519 /* 1520 * Process options. 1521 */ 1522 #ifdef TCP_SIGNATURE 1523 if (optp || (tp->t_flags & TF_SIGNATURE)) 1524 #else 1525 if (optp) 1526 #endif 1527 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1528 goto drop; 1529 1530 if (TCP_SACK_ENABLED(tp)) { 1531 tcp_del_sackholes(tp, th); 1532 } 1533 1534 if (opti.ts_present && opti.ts_ecr) { 1535 /* 1536 * Calculate the RTT from the returned time stamp and the 1537 * connection's time base. If the time stamp is later than 1538 * the current time, or is extremely old, fall back to non-1323 1539 * RTT calculation. Since ts_ecr is unsigned, we can test both 1540 * at the same time. 1541 */ 1542 opti.ts_ecr = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1543 if (opti.ts_ecr > TCP_PAWS_IDLE) 1544 opti.ts_ecr = 0; 1545 } 1546 1547 /* 1548 * Header prediction: check for the two common cases 1549 * of a uni-directional data xfer. If the packet has 1550 * no control flags, is in-sequence, the window didn't 1551 * change and we're not retransmitting, it's a 1552 * candidate. If the length is zero and the ack moved 1553 * forward, we're the sender side of the xfer. Just 1554 * free the data acked & wake any higher level process 1555 * that was blocked waiting for space. If the length 1556 * is non-zero and the ack didn't move, we're the 1557 * receiver side. If we're getting packets in-order 1558 * (the reassembly queue is empty), add the data to 1559 * the socket buffer and note that we need a delayed ack. 1560 */ 1561 if (tp->t_state == TCPS_ESTABLISHED && 1562 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1563 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1564 th->th_seq == tp->rcv_nxt && 1565 tiwin && tiwin == tp->snd_wnd && 1566 tp->snd_nxt == tp->snd_max) { 1567 1568 /* 1569 * If last ACK falls within this segment's sequence numbers, 1570 * record the timestamp. 1571 */ 1572 if (opti.ts_present && 1573 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1574 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1575 tp->ts_recent_age = tcp_now; 1576 tp->ts_recent = opti.ts_val; 1577 } 1578 1579 if (tlen == 0) { 1580 /* Ack prediction. */ 1581 if (SEQ_GT(th->th_ack, tp->snd_una) && 1582 SEQ_LEQ(th->th_ack, tp->snd_max) && 1583 tp->snd_cwnd >= tp->snd_wnd && 1584 tp->t_partialacks < 0) { 1585 /* 1586 * this is a pure ack for outstanding data. 1587 */ 1588 ++tcpstat.tcps_predack; 1589 if (opti.ts_present && opti.ts_ecr) 1590 tcp_xmit_timer(tp, opti.ts_ecr); 1591 else if (tp->t_rtttime && 1592 SEQ_GT(th->th_ack, tp->t_rtseq)) 1593 tcp_xmit_timer(tp, 1594 tcp_now - tp->t_rtttime); 1595 acked = th->th_ack - tp->snd_una; 1596 tcpstat.tcps_rcvackpack++; 1597 tcpstat.tcps_rcvackbyte += acked; 1598 ND6_HINT(tp); 1599 1600 if (acked > (tp->t_lastoff - tp->t_inoff)) 1601 tp->t_lastm = NULL; 1602 sbdrop(&so->so_snd, acked); 1603 tp->t_lastoff -= acked; 1604 1605 tp->snd_una = th->th_ack; 1606 tp->snd_fack = tp->snd_una; 1607 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1608 tp->snd_high = tp->snd_una; 1609 m_freem(m); 1610 1611 /* 1612 * If all outstanding data are acked, stop 1613 * retransmit timer, otherwise restart timer 1614 * using current (possibly backed-off) value. 1615 * If process is waiting for space, 1616 * wakeup/selwakeup/signal. If data 1617 * are ready to send, let tcp_output 1618 * decide between more output or persist. 1619 */ 1620 if (tp->snd_una == tp->snd_max) 1621 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1622 else if (TCP_TIMER_ISARMED(tp, 1623 TCPT_PERSIST) == 0) 1624 TCP_TIMER_ARM(tp, TCPT_REXMT, 1625 tp->t_rxtcur); 1626 1627 sowwakeup(so); 1628 if (so->so_snd.sb_cc) 1629 (void) tcp_output(tp); 1630 if (tcp_saveti) 1631 m_freem(tcp_saveti); 1632 return; 1633 } 1634 } else if (th->th_ack == tp->snd_una && 1635 TAILQ_FIRST(&tp->segq) == NULL && 1636 tlen <= sbspace(&so->so_rcv)) { 1637 /* 1638 * this is a pure, in-sequence data packet 1639 * with nothing on the reassembly queue and 1640 * we have enough buffer space to take it. 1641 */ 1642 ++tcpstat.tcps_preddat; 1643 tp->rcv_nxt += tlen; 1644 tcpstat.tcps_rcvpack++; 1645 tcpstat.tcps_rcvbyte += tlen; 1646 ND6_HINT(tp); 1647 /* 1648 * Drop TCP, IP headers and TCP options then add data 1649 * to socket buffer. 1650 */ 1651 if (so->so_state & SS_CANTRCVMORE) 1652 m_freem(m); 1653 else { 1654 m_adj(m, toff + off); 1655 sbappendstream(&so->so_rcv, m); 1656 } 1657 sorwakeup(so); 1658 TCP_SETUP_ACK(tp, th); 1659 if (tp->t_flags & TF_ACKNOW) 1660 (void) tcp_output(tp); 1661 if (tcp_saveti) 1662 m_freem(tcp_saveti); 1663 return; 1664 } 1665 } 1666 1667 /* 1668 * Compute mbuf offset to TCP data segment. 1669 */ 1670 hdroptlen = toff + off; 1671 1672 /* 1673 * Calculate amount of space in receive window, 1674 * and then do TCP input processing. 1675 * Receive window is amount of space in rcv queue, 1676 * but not less than advertised window. 1677 */ 1678 { int win; 1679 1680 win = sbspace(&so->so_rcv); 1681 if (win < 0) 1682 win = 0; 1683 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1684 } 1685 1686 switch (tp->t_state) { 1687 case TCPS_LISTEN: 1688 /* 1689 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1690 */ 1691 if (m->m_flags & (M_BCAST|M_MCAST)) 1692 goto drop; 1693 switch (af) { 1694 #ifdef INET6 1695 case AF_INET6: 1696 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1697 goto drop; 1698 break; 1699 #endif /* INET6 */ 1700 case AF_INET: 1701 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1702 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1703 goto drop; 1704 break; 1705 } 1706 break; 1707 1708 /* 1709 * If the state is SYN_SENT: 1710 * if seg contains an ACK, but not for our SYN, drop the input. 1711 * if seg contains a RST, then drop the connection. 1712 * if seg does not contain SYN, then drop it. 1713 * Otherwise this is an acceptable SYN segment 1714 * initialize tp->rcv_nxt and tp->irs 1715 * if seg contains ack then advance tp->snd_una 1716 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1717 * arrange for segment to be acked (eventually) 1718 * continue processing rest of data/controls, beginning with URG 1719 */ 1720 case TCPS_SYN_SENT: 1721 if ((tiflags & TH_ACK) && 1722 (SEQ_LEQ(th->th_ack, tp->iss) || 1723 SEQ_GT(th->th_ack, tp->snd_max))) 1724 goto dropwithreset; 1725 if (tiflags & TH_RST) { 1726 if (tiflags & TH_ACK) 1727 tp = tcp_drop(tp, ECONNREFUSED); 1728 goto drop; 1729 } 1730 if ((tiflags & TH_SYN) == 0) 1731 goto drop; 1732 if (tiflags & TH_ACK) { 1733 tp->snd_una = th->th_ack; 1734 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1735 tp->snd_nxt = tp->snd_una; 1736 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1737 tp->snd_high = tp->snd_una; 1738 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1739 } 1740 tp->irs = th->th_seq; 1741 tcp_rcvseqinit(tp); 1742 tp->t_flags |= TF_ACKNOW; 1743 tcp_mss_from_peer(tp, opti.maxseg); 1744 1745 /* 1746 * Initialize the initial congestion window. If we 1747 * had to retransmit the SYN, we must initialize cwnd 1748 * to 1 segment (i.e. the Loss Window). 1749 */ 1750 if (tp->t_flags & TF_SYN_REXMT) 1751 tp->snd_cwnd = tp->t_peermss; 1752 else { 1753 int ss = tcp_init_win; 1754 #ifdef INET 1755 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1756 ss = tcp_init_win_local; 1757 #endif 1758 #ifdef INET6 1759 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1760 ss = tcp_init_win_local; 1761 #endif 1762 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1763 } 1764 1765 tcp_rmx_rtt(tp); 1766 if (tiflags & TH_ACK) { 1767 tcpstat.tcps_connects++; 1768 soisconnected(so); 1769 tcp_established(tp); 1770 /* Do window scaling on this connection? */ 1771 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1772 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1773 tp->snd_scale = tp->requested_s_scale; 1774 tp->rcv_scale = tp->request_r_scale; 1775 } 1776 TCP_REASS_LOCK(tp); 1777 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1778 TCP_REASS_UNLOCK(tp); 1779 /* 1780 * if we didn't have to retransmit the SYN, 1781 * use its rtt as our initial srtt & rtt var. 1782 */ 1783 if (tp->t_rtttime) 1784 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1785 } else 1786 tp->t_state = TCPS_SYN_RECEIVED; 1787 1788 /* 1789 * Advance th->th_seq to correspond to first data byte. 1790 * If data, trim to stay within window, 1791 * dropping FIN if necessary. 1792 */ 1793 th->th_seq++; 1794 if (tlen > tp->rcv_wnd) { 1795 todrop = tlen - tp->rcv_wnd; 1796 m_adj(m, -todrop); 1797 tlen = tp->rcv_wnd; 1798 tiflags &= ~TH_FIN; 1799 tcpstat.tcps_rcvpackafterwin++; 1800 tcpstat.tcps_rcvbyteafterwin += todrop; 1801 } 1802 tp->snd_wl1 = th->th_seq - 1; 1803 tp->rcv_up = th->th_seq; 1804 goto step6; 1805 1806 /* 1807 * If the state is SYN_RECEIVED: 1808 * If seg contains an ACK, but not for our SYN, drop the input 1809 * and generate an RST. See page 36, rfc793 1810 */ 1811 case TCPS_SYN_RECEIVED: 1812 if ((tiflags & TH_ACK) && 1813 (SEQ_LEQ(th->th_ack, tp->iss) || 1814 SEQ_GT(th->th_ack, tp->snd_max))) 1815 goto dropwithreset; 1816 break; 1817 } 1818 1819 /* 1820 * States other than LISTEN or SYN_SENT. 1821 * First check timestamp, if present. 1822 * Then check that at least some bytes of segment are within 1823 * receive window. If segment begins before rcv_nxt, 1824 * drop leading data (and SYN); if nothing left, just ack. 1825 * 1826 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1827 * and it's less than ts_recent, drop it. 1828 */ 1829 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1830 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1831 1832 /* Check to see if ts_recent is over 24 days old. */ 1833 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 1834 /* 1835 * Invalidate ts_recent. If this segment updates 1836 * ts_recent, the age will be reset later and ts_recent 1837 * will get a valid value. If it does not, setting 1838 * ts_recent to zero will at least satisfy the 1839 * requirement that zero be placed in the timestamp 1840 * echo reply when ts_recent isn't valid. The 1841 * age isn't reset until we get a valid ts_recent 1842 * because we don't want out-of-order segments to be 1843 * dropped when ts_recent is old. 1844 */ 1845 tp->ts_recent = 0; 1846 } else { 1847 tcpstat.tcps_rcvduppack++; 1848 tcpstat.tcps_rcvdupbyte += tlen; 1849 tcpstat.tcps_pawsdrop++; 1850 tcp_new_dsack(tp, th->th_seq, tlen); 1851 goto dropafterack; 1852 } 1853 } 1854 1855 todrop = tp->rcv_nxt - th->th_seq; 1856 dupseg = FALSE; 1857 if (todrop > 0) { 1858 if (tiflags & TH_SYN) { 1859 tiflags &= ~TH_SYN; 1860 th->th_seq++; 1861 if (th->th_urp > 1) 1862 th->th_urp--; 1863 else { 1864 tiflags &= ~TH_URG; 1865 th->th_urp = 0; 1866 } 1867 todrop--; 1868 } 1869 if (todrop > tlen || 1870 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1871 /* 1872 * Any valid FIN or RST must be to the left of the 1873 * window. At this point the FIN or RST must be a 1874 * duplicate or out of sequence; drop it. 1875 */ 1876 if (tiflags & TH_RST) 1877 goto drop; 1878 tiflags &= ~(TH_FIN|TH_RST); 1879 /* 1880 * Send an ACK to resynchronize and drop any data. 1881 * But keep on processing for RST or ACK. 1882 */ 1883 tp->t_flags |= TF_ACKNOW; 1884 todrop = tlen; 1885 dupseg = TRUE; 1886 tcpstat.tcps_rcvdupbyte += todrop; 1887 tcpstat.tcps_rcvduppack++; 1888 } else if ((tiflags & TH_RST) && 1889 th->th_seq != tp->last_ack_sent) { 1890 /* 1891 * Test for reset before adjusting the sequence 1892 * number for overlapping data. 1893 */ 1894 goto dropafterack_ratelim; 1895 } else { 1896 tcpstat.tcps_rcvpartduppack++; 1897 tcpstat.tcps_rcvpartdupbyte += todrop; 1898 } 1899 tcp_new_dsack(tp, th->th_seq, todrop); 1900 hdroptlen += todrop; /*drop from head afterwards*/ 1901 th->th_seq += todrop; 1902 tlen -= todrop; 1903 if (th->th_urp > todrop) 1904 th->th_urp -= todrop; 1905 else { 1906 tiflags &= ~TH_URG; 1907 th->th_urp = 0; 1908 } 1909 } 1910 1911 /* 1912 * If new data are received on a connection after the 1913 * user processes are gone, then RST the other end. 1914 */ 1915 if ((so->so_state & SS_NOFDREF) && 1916 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1917 tp = tcp_close(tp); 1918 tcpstat.tcps_rcvafterclose++; 1919 goto dropwithreset; 1920 } 1921 1922 /* 1923 * If segment ends after window, drop trailing data 1924 * (and PUSH and FIN); if nothing left, just ACK. 1925 */ 1926 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1927 if (todrop > 0) { 1928 tcpstat.tcps_rcvpackafterwin++; 1929 if (todrop >= tlen) { 1930 /* 1931 * The segment actually starts after the window. 1932 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 1933 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 1934 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 1935 */ 1936 tcpstat.tcps_rcvbyteafterwin += tlen; 1937 /* 1938 * If a new connection request is received 1939 * while in TIME_WAIT, drop the old connection 1940 * and start over if the sequence numbers 1941 * are above the previous ones. 1942 * 1943 * NOTE: We will checksum the packet again, and 1944 * so we need to put the header fields back into 1945 * network order! 1946 * XXX This kind of sucks, but we don't expect 1947 * XXX this to happen very often, so maybe it 1948 * XXX doesn't matter so much. 1949 */ 1950 if (tiflags & TH_SYN && 1951 tp->t_state == TCPS_TIME_WAIT && 1952 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1953 iss = tcp_new_iss(tp, tp->snd_nxt); 1954 tp = tcp_close(tp); 1955 TCP_FIELDS_TO_NET(th); 1956 goto findpcb; 1957 } 1958 /* 1959 * If window is closed can only take segments at 1960 * window edge, and have to drop data and PUSH from 1961 * incoming segments. Continue processing, but 1962 * remember to ack. Otherwise, drop segment 1963 * and (if not RST) ack. 1964 */ 1965 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1966 tp->t_flags |= TF_ACKNOW; 1967 tcpstat.tcps_rcvwinprobe++; 1968 } else 1969 goto dropafterack; 1970 } else 1971 tcpstat.tcps_rcvbyteafterwin += todrop; 1972 m_adj(m, -todrop); 1973 tlen -= todrop; 1974 tiflags &= ~(TH_PUSH|TH_FIN); 1975 } 1976 1977 /* 1978 * If last ACK falls within this segment's sequence numbers, 1979 * and the timestamp is newer, record it. 1980 */ 1981 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1982 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1983 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1984 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1985 tp->ts_recent_age = tcp_now; 1986 tp->ts_recent = opti.ts_val; 1987 } 1988 1989 /* 1990 * If the RST bit is set examine the state: 1991 * SYN_RECEIVED STATE: 1992 * If passive open, return to LISTEN state. 1993 * If active open, inform user that connection was refused. 1994 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1995 * Inform user that connection was reset, and close tcb. 1996 * CLOSING, LAST_ACK, TIME_WAIT STATES 1997 * Close the tcb. 1998 */ 1999 if (tiflags & TH_RST) { 2000 if (th->th_seq != tp->last_ack_sent) 2001 goto dropafterack_ratelim; 2002 2003 switch (tp->t_state) { 2004 case TCPS_SYN_RECEIVED: 2005 so->so_error = ECONNREFUSED; 2006 goto close; 2007 2008 case TCPS_ESTABLISHED: 2009 case TCPS_FIN_WAIT_1: 2010 case TCPS_FIN_WAIT_2: 2011 case TCPS_CLOSE_WAIT: 2012 so->so_error = ECONNRESET; 2013 close: 2014 tp->t_state = TCPS_CLOSED; 2015 tcpstat.tcps_drops++; 2016 tp = tcp_close(tp); 2017 goto drop; 2018 2019 case TCPS_CLOSING: 2020 case TCPS_LAST_ACK: 2021 case TCPS_TIME_WAIT: 2022 tp = tcp_close(tp); 2023 goto drop; 2024 } 2025 } 2026 2027 /* 2028 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2029 * we must be in a synchronized state. RFC791 states (under RST 2030 * generation) that any unacceptable segment (an out-of-order SYN 2031 * qualifies) received in a synchronized state must elicit only an 2032 * empty acknowledgment segment ... and the connection remains in 2033 * the same state. 2034 */ 2035 if (tiflags & TH_SYN) { 2036 if (tp->rcv_nxt == th->th_seq) { 2037 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2038 TH_ACK); 2039 if (tcp_saveti) 2040 m_freem(tcp_saveti); 2041 return; 2042 } 2043 2044 goto dropafterack_ratelim; 2045 } 2046 2047 /* 2048 * If the ACK bit is off we drop the segment and return. 2049 */ 2050 if ((tiflags & TH_ACK) == 0) { 2051 if (tp->t_flags & TF_ACKNOW) 2052 goto dropafterack; 2053 else 2054 goto drop; 2055 } 2056 2057 /* 2058 * Ack processing. 2059 */ 2060 switch (tp->t_state) { 2061 2062 /* 2063 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2064 * ESTABLISHED state and continue processing, otherwise 2065 * send an RST. 2066 */ 2067 case TCPS_SYN_RECEIVED: 2068 if (SEQ_GT(tp->snd_una, th->th_ack) || 2069 SEQ_GT(th->th_ack, tp->snd_max)) 2070 goto dropwithreset; 2071 tcpstat.tcps_connects++; 2072 soisconnected(so); 2073 tcp_established(tp); 2074 /* Do window scaling? */ 2075 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2076 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2077 tp->snd_scale = tp->requested_s_scale; 2078 tp->rcv_scale = tp->request_r_scale; 2079 } 2080 TCP_REASS_LOCK(tp); 2081 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2082 TCP_REASS_UNLOCK(tp); 2083 tp->snd_wl1 = th->th_seq - 1; 2084 /* fall into ... */ 2085 2086 /* 2087 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2088 * ACKs. If the ack is in the range 2089 * tp->snd_una < th->th_ack <= tp->snd_max 2090 * then advance tp->snd_una to th->th_ack and drop 2091 * data from the retransmission queue. If this ACK reflects 2092 * more up to date window information we update our window information. 2093 */ 2094 case TCPS_ESTABLISHED: 2095 case TCPS_FIN_WAIT_1: 2096 case TCPS_FIN_WAIT_2: 2097 case TCPS_CLOSE_WAIT: 2098 case TCPS_CLOSING: 2099 case TCPS_LAST_ACK: 2100 case TCPS_TIME_WAIT: 2101 2102 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2103 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2104 tcpstat.tcps_rcvdupack++; 2105 /* 2106 * If we have outstanding data (other than 2107 * a window probe), this is a completely 2108 * duplicate ack (ie, window info didn't 2109 * change), the ack is the biggest we've 2110 * seen and we've seen exactly our rexmt 2111 * threshhold of them, assume a packet 2112 * has been dropped and retransmit it. 2113 * Kludge snd_nxt & the congestion 2114 * window so we send only this one 2115 * packet. 2116 * 2117 * We know we're losing at the current 2118 * window size so do congestion avoidance 2119 * (set ssthresh to half the current window 2120 * and pull our congestion window back to 2121 * the new ssthresh). 2122 * 2123 * Dup acks mean that packets have left the 2124 * network (they're now cached at the receiver) 2125 * so bump cwnd by the amount in the receiver 2126 * to keep a constant cwnd packets in the 2127 * network. 2128 * 2129 * If we are using TCP/SACK, then enter 2130 * Fast Recovery if the receiver SACKs 2131 * data that is tcprexmtthresh * MSS 2132 * bytes past the last ACKed segment, 2133 * irrespective of the number of DupAcks. 2134 */ 2135 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2136 th->th_ack != tp->snd_una) 2137 tp->t_dupacks = 0; 2138 else if (tp->t_partialacks < 0 && 2139 (++tp->t_dupacks == tcprexmtthresh || 2140 TCP_FACK_FASTRECOV(tp))) { 2141 tcp_seq onxt; 2142 u_int win; 2143 2144 if (tcp_do_newreno && 2145 SEQ_LT(th->th_ack, tp->snd_high)) { 2146 /* 2147 * False fast retransmit after 2148 * timeout. Do not enter fast 2149 * recovery. 2150 */ 2151 tp->t_dupacks = 0; 2152 break; 2153 } 2154 2155 onxt = tp->snd_nxt; 2156 win = min(tp->snd_wnd, tp->snd_cwnd) / 2157 2 / tp->t_segsz; 2158 if (win < 2) 2159 win = 2; 2160 tp->snd_ssthresh = win * tp->t_segsz; 2161 tp->snd_recover = tp->snd_max; 2162 tp->t_partialacks = 0; 2163 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2164 tp->t_rtttime = 0; 2165 if (TCP_SACK_ENABLED(tp)) { 2166 tp->t_dupacks = tcprexmtthresh; 2167 tp->sack_newdata = tp->snd_nxt; 2168 tp->snd_cwnd = tp->t_segsz; 2169 (void) tcp_output(tp); 2170 goto drop; 2171 } 2172 tp->snd_nxt = th->th_ack; 2173 tp->snd_cwnd = tp->t_segsz; 2174 (void) tcp_output(tp); 2175 tp->snd_cwnd = tp->snd_ssthresh + 2176 tp->t_segsz * tp->t_dupacks; 2177 if (SEQ_GT(onxt, tp->snd_nxt)) 2178 tp->snd_nxt = onxt; 2179 goto drop; 2180 } else if (tp->t_dupacks > tcprexmtthresh) { 2181 tp->snd_cwnd += tp->t_segsz; 2182 (void) tcp_output(tp); 2183 goto drop; 2184 } 2185 } else { 2186 /* 2187 * If the ack appears to be very old, only 2188 * allow data that is in-sequence. This 2189 * makes it somewhat more difficult to insert 2190 * forged data by guessing sequence numbers. 2191 * Sent an ack to try to update the send 2192 * sequence number on the other side. 2193 */ 2194 if (tlen && th->th_seq != tp->rcv_nxt && 2195 SEQ_LT(th->th_ack, 2196 tp->snd_una - tp->max_sndwnd)) 2197 goto dropafterack; 2198 } 2199 break; 2200 } 2201 /* 2202 * If the congestion window was inflated to account 2203 * for the other side's cached packets, retract it. 2204 */ 2205 if (TCP_SACK_ENABLED(tp)) 2206 tcp_sack_newack(tp, th); 2207 else if (tcp_do_newreno) 2208 tcp_newreno_newack(tp, th); 2209 else 2210 tcp_reno_newack(tp, th); 2211 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2212 tcpstat.tcps_rcvacktoomuch++; 2213 goto dropafterack; 2214 } 2215 acked = th->th_ack - tp->snd_una; 2216 tcpstat.tcps_rcvackpack++; 2217 tcpstat.tcps_rcvackbyte += acked; 2218 2219 /* 2220 * If we have a timestamp reply, update smoothed 2221 * round trip time. If no timestamp is present but 2222 * transmit timer is running and timed sequence 2223 * number was acked, update smoothed round trip time. 2224 * Since we now have an rtt measurement, cancel the 2225 * timer backoff (cf., Phil Karn's retransmit alg.). 2226 * Recompute the initial retransmit timer. 2227 */ 2228 if (opti.ts_present && opti.ts_ecr) 2229 tcp_xmit_timer(tp, opti.ts_ecr); 2230 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2231 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2232 2233 /* 2234 * If all outstanding data is acked, stop retransmit 2235 * timer and remember to restart (more output or persist). 2236 * If there is more data to be acked, restart retransmit 2237 * timer, using current (possibly backed-off) value. 2238 */ 2239 if (th->th_ack == tp->snd_max) { 2240 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2241 needoutput = 1; 2242 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2243 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2244 /* 2245 * When new data is acked, open the congestion window. 2246 * If the window gives us less than ssthresh packets 2247 * in flight, open exponentially (segsz per packet). 2248 * Otherwise open linearly: segsz per window 2249 * (segsz^2 / cwnd per packet), plus a constant 2250 * fraction of a packet (segsz/8) to help larger windows 2251 * open quickly enough. 2252 * 2253 * If we are still in fast recovery (meaning we are using 2254 * NewReno and we have only received partial acks), do not 2255 * inflate the window yet. 2256 */ 2257 if (tp->t_partialacks < 0) { 2258 u_int cw = tp->snd_cwnd; 2259 u_int incr = tp->t_segsz; 2260 2261 if (cw >= tp->snd_ssthresh) 2262 incr = incr * incr / cw; 2263 tp->snd_cwnd = min(cw + incr, 2264 TCP_MAXWIN << tp->snd_scale); 2265 } 2266 ND6_HINT(tp); 2267 if (acked > so->so_snd.sb_cc) { 2268 tp->snd_wnd -= so->so_snd.sb_cc; 2269 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2270 ourfinisacked = 1; 2271 } else { 2272 if (acked > (tp->t_lastoff - tp->t_inoff)) 2273 tp->t_lastm = NULL; 2274 sbdrop(&so->so_snd, acked); 2275 tp->t_lastoff -= acked; 2276 tp->snd_wnd -= acked; 2277 ourfinisacked = 0; 2278 } 2279 sowwakeup(so); 2280 tp->snd_una = th->th_ack; 2281 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2282 tp->snd_fack = tp->snd_una; 2283 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2284 tp->snd_nxt = tp->snd_una; 2285 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2286 tp->snd_high = tp->snd_una; 2287 2288 switch (tp->t_state) { 2289 2290 /* 2291 * In FIN_WAIT_1 STATE in addition to the processing 2292 * for the ESTABLISHED state if our FIN is now acknowledged 2293 * then enter FIN_WAIT_2. 2294 */ 2295 case TCPS_FIN_WAIT_1: 2296 if (ourfinisacked) { 2297 /* 2298 * If we can't receive any more 2299 * data, then closing user can proceed. 2300 * Starting the timer is contrary to the 2301 * specification, but if we don't get a FIN 2302 * we'll hang forever. 2303 */ 2304 if (so->so_state & SS_CANTRCVMORE) { 2305 soisdisconnected(so); 2306 if (tcp_maxidle > 0) 2307 TCP_TIMER_ARM(tp, TCPT_2MSL, 2308 tcp_maxidle); 2309 } 2310 tp->t_state = TCPS_FIN_WAIT_2; 2311 } 2312 break; 2313 2314 /* 2315 * In CLOSING STATE in addition to the processing for 2316 * the ESTABLISHED state if the ACK acknowledges our FIN 2317 * then enter the TIME-WAIT state, otherwise ignore 2318 * the segment. 2319 */ 2320 case TCPS_CLOSING: 2321 if (ourfinisacked) { 2322 tp->t_state = TCPS_TIME_WAIT; 2323 tcp_canceltimers(tp); 2324 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2325 soisdisconnected(so); 2326 } 2327 break; 2328 2329 /* 2330 * In LAST_ACK, we may still be waiting for data to drain 2331 * and/or to be acked, as well as for the ack of our FIN. 2332 * If our FIN is now acknowledged, delete the TCB, 2333 * enter the closed state and return. 2334 */ 2335 case TCPS_LAST_ACK: 2336 if (ourfinisacked) { 2337 tp = tcp_close(tp); 2338 goto drop; 2339 } 2340 break; 2341 2342 /* 2343 * In TIME_WAIT state the only thing that should arrive 2344 * is a retransmission of the remote FIN. Acknowledge 2345 * it and restart the finack timer. 2346 */ 2347 case TCPS_TIME_WAIT: 2348 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2349 goto dropafterack; 2350 } 2351 } 2352 2353 step6: 2354 /* 2355 * Update window information. 2356 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2357 */ 2358 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2359 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2360 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2361 /* keep track of pure window updates */ 2362 if (tlen == 0 && 2363 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2364 tcpstat.tcps_rcvwinupd++; 2365 tp->snd_wnd = tiwin; 2366 tp->snd_wl1 = th->th_seq; 2367 tp->snd_wl2 = th->th_ack; 2368 if (tp->snd_wnd > tp->max_sndwnd) 2369 tp->max_sndwnd = tp->snd_wnd; 2370 needoutput = 1; 2371 } 2372 2373 /* 2374 * Process segments with URG. 2375 */ 2376 if ((tiflags & TH_URG) && th->th_urp && 2377 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2378 /* 2379 * This is a kludge, but if we receive and accept 2380 * random urgent pointers, we'll crash in 2381 * soreceive. It's hard to imagine someone 2382 * actually wanting to send this much urgent data. 2383 */ 2384 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2385 th->th_urp = 0; /* XXX */ 2386 tiflags &= ~TH_URG; /* XXX */ 2387 goto dodata; /* XXX */ 2388 } 2389 /* 2390 * If this segment advances the known urgent pointer, 2391 * then mark the data stream. This should not happen 2392 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2393 * a FIN has been received from the remote side. 2394 * In these states we ignore the URG. 2395 * 2396 * According to RFC961 (Assigned Protocols), 2397 * the urgent pointer points to the last octet 2398 * of urgent data. We continue, however, 2399 * to consider it to indicate the first octet 2400 * of data past the urgent section as the original 2401 * spec states (in one of two places). 2402 */ 2403 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2404 tp->rcv_up = th->th_seq + th->th_urp; 2405 so->so_oobmark = so->so_rcv.sb_cc + 2406 (tp->rcv_up - tp->rcv_nxt) - 1; 2407 if (so->so_oobmark == 0) 2408 so->so_state |= SS_RCVATMARK; 2409 sohasoutofband(so); 2410 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2411 } 2412 /* 2413 * Remove out of band data so doesn't get presented to user. 2414 * This can happen independent of advancing the URG pointer, 2415 * but if two URG's are pending at once, some out-of-band 2416 * data may creep in... ick. 2417 */ 2418 if (th->th_urp <= (u_int16_t) tlen 2419 #ifdef SO_OOBINLINE 2420 && (so->so_options & SO_OOBINLINE) == 0 2421 #endif 2422 ) 2423 tcp_pulloutofband(so, th, m, hdroptlen); 2424 } else 2425 /* 2426 * If no out of band data is expected, 2427 * pull receive urgent pointer along 2428 * with the receive window. 2429 */ 2430 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2431 tp->rcv_up = tp->rcv_nxt; 2432 dodata: /* XXX */ 2433 2434 /* 2435 * Process the segment text, merging it into the TCP sequencing queue, 2436 * and arranging for acknowledgement of receipt if necessary. 2437 * This process logically involves adjusting tp->rcv_wnd as data 2438 * is presented to the user (this happens in tcp_usrreq.c, 2439 * case PRU_RCVD). If a FIN has already been received on this 2440 * connection then we just ignore the text. 2441 */ 2442 if ((tlen || (tiflags & TH_FIN)) && 2443 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2444 /* 2445 * Insert segment ti into reassembly queue of tcp with 2446 * control block tp. Return TH_FIN if reassembly now includes 2447 * a segment with FIN. The macro form does the common case 2448 * inline (segment is the next to be received on an 2449 * established connection, and the queue is empty), 2450 * avoiding linkage into and removal from the queue and 2451 * repetition of various conversions. 2452 * Set DELACK for segments received in order, but ack 2453 * immediately when segments are out of order 2454 * (so fast retransmit can work). 2455 */ 2456 /* NOTE: this was TCP_REASS() macro, but used only once */ 2457 TCP_REASS_LOCK(tp); 2458 if (th->th_seq == tp->rcv_nxt && 2459 TAILQ_FIRST(&tp->segq) == NULL && 2460 tp->t_state == TCPS_ESTABLISHED) { 2461 TCP_SETUP_ACK(tp, th); 2462 tp->rcv_nxt += tlen; 2463 tiflags = th->th_flags & TH_FIN; 2464 tcpstat.tcps_rcvpack++; 2465 tcpstat.tcps_rcvbyte += tlen; 2466 ND6_HINT(tp); 2467 if (so->so_state & SS_CANTRCVMORE) 2468 m_freem(m); 2469 else { 2470 m_adj(m, hdroptlen); 2471 sbappendstream(&(so)->so_rcv, m); 2472 } 2473 sorwakeup(so); 2474 } else { 2475 m_adj(m, hdroptlen); 2476 tiflags = tcp_reass(tp, th, m, &tlen); 2477 tp->t_flags |= TF_ACKNOW; 2478 } 2479 TCP_REASS_UNLOCK(tp); 2480 2481 /* 2482 * Note the amount of data that peer has sent into 2483 * our window, in order to estimate the sender's 2484 * buffer size. 2485 */ 2486 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2487 } else { 2488 m_freem(m); 2489 m = NULL; 2490 tiflags &= ~TH_FIN; 2491 } 2492 2493 /* 2494 * If FIN is received ACK the FIN and let the user know 2495 * that the connection is closing. Ignore a FIN received before 2496 * the connection is fully established. 2497 */ 2498 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2499 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2500 socantrcvmore(so); 2501 tp->t_flags |= TF_ACKNOW; 2502 tp->rcv_nxt++; 2503 } 2504 switch (tp->t_state) { 2505 2506 /* 2507 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2508 */ 2509 case TCPS_ESTABLISHED: 2510 tp->t_state = TCPS_CLOSE_WAIT; 2511 break; 2512 2513 /* 2514 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2515 * enter the CLOSING state. 2516 */ 2517 case TCPS_FIN_WAIT_1: 2518 tp->t_state = TCPS_CLOSING; 2519 break; 2520 2521 /* 2522 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2523 * starting the time-wait timer, turning off the other 2524 * standard timers. 2525 */ 2526 case TCPS_FIN_WAIT_2: 2527 tp->t_state = TCPS_TIME_WAIT; 2528 tcp_canceltimers(tp); 2529 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2530 soisdisconnected(so); 2531 break; 2532 2533 /* 2534 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2535 */ 2536 case TCPS_TIME_WAIT: 2537 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2538 break; 2539 } 2540 } 2541 #ifdef TCP_DEBUG 2542 if (so->so_options & SO_DEBUG) 2543 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2544 #endif 2545 2546 /* 2547 * Return any desired output. 2548 */ 2549 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2550 (void) tcp_output(tp); 2551 } 2552 if (tcp_saveti) 2553 m_freem(tcp_saveti); 2554 return; 2555 2556 badsyn: 2557 /* 2558 * Received a bad SYN. Increment counters and dropwithreset. 2559 */ 2560 tcpstat.tcps_badsyn++; 2561 tp = NULL; 2562 goto dropwithreset; 2563 2564 dropafterack: 2565 /* 2566 * Generate an ACK dropping incoming segment if it occupies 2567 * sequence space, where the ACK reflects our state. 2568 */ 2569 if (tiflags & TH_RST) 2570 goto drop; 2571 goto dropafterack2; 2572 2573 dropafterack_ratelim: 2574 /* 2575 * We may want to rate-limit ACKs against SYN/RST attack. 2576 */ 2577 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2578 tcp_ackdrop_ppslim) == 0) { 2579 /* XXX stat */ 2580 goto drop; 2581 } 2582 /* ...fall into dropafterack2... */ 2583 2584 dropafterack2: 2585 m_freem(m); 2586 tp->t_flags |= TF_ACKNOW; 2587 (void) tcp_output(tp); 2588 if (tcp_saveti) 2589 m_freem(tcp_saveti); 2590 return; 2591 2592 dropwithreset_ratelim: 2593 /* 2594 * We may want to rate-limit RSTs in certain situations, 2595 * particularly if we are sending an RST in response to 2596 * an attempt to connect to or otherwise communicate with 2597 * a port for which we have no socket. 2598 */ 2599 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2600 tcp_rst_ppslim) == 0) { 2601 /* XXX stat */ 2602 goto drop; 2603 } 2604 /* ...fall into dropwithreset... */ 2605 2606 dropwithreset: 2607 /* 2608 * Generate a RST, dropping incoming segment. 2609 * Make ACK acceptable to originator of segment. 2610 */ 2611 if (tiflags & TH_RST) 2612 goto drop; 2613 2614 switch (af) { 2615 #ifdef INET6 2616 case AF_INET6: 2617 /* For following calls to tcp_respond */ 2618 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2619 goto drop; 2620 break; 2621 #endif /* INET6 */ 2622 case AF_INET: 2623 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2624 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2625 goto drop; 2626 } 2627 2628 if (tiflags & TH_ACK) 2629 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2630 else { 2631 if (tiflags & TH_SYN) 2632 tlen++; 2633 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2634 TH_RST|TH_ACK); 2635 } 2636 if (tcp_saveti) 2637 m_freem(tcp_saveti); 2638 return; 2639 2640 badcsum: 2641 drop: 2642 /* 2643 * Drop space held by incoming segment and return. 2644 */ 2645 if (tp) { 2646 if (tp->t_inpcb) 2647 so = tp->t_inpcb->inp_socket; 2648 #ifdef INET6 2649 else if (tp->t_in6pcb) 2650 so = tp->t_in6pcb->in6p_socket; 2651 #endif 2652 else 2653 so = NULL; 2654 #ifdef TCP_DEBUG 2655 if (so && (so->so_options & SO_DEBUG) != 0) 2656 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2657 #endif 2658 } 2659 if (tcp_saveti) 2660 m_freem(tcp_saveti); 2661 m_freem(m); 2662 return; 2663 } 2664 2665 #ifdef TCP_SIGNATURE 2666 int 2667 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2668 { 2669 2670 MD5Update(fstate, (u_char *)data, len); 2671 return (0); 2672 } 2673 2674 struct secasvar * 2675 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2676 { 2677 struct secasvar *sav; 2678 #ifdef FAST_IPSEC 2679 union sockaddr_union dst; 2680 #endif 2681 struct ip *ip; 2682 struct ip6_hdr *ip6; 2683 2684 ip = mtod(m, struct ip *); 2685 switch (ip->ip_v) { 2686 case 4: 2687 ip = mtod(m, struct ip *); 2688 ip6 = NULL; 2689 break; 2690 case 6: 2691 ip = NULL; 2692 ip6 = mtod(m, struct ip6_hdr *); 2693 break; 2694 default: 2695 return (NULL); 2696 } 2697 2698 #ifdef FAST_IPSEC 2699 /* Extract the destination from the IP header in the mbuf. */ 2700 bzero(&dst, sizeof(union sockaddr_union)); 2701 dst.sa.sa_len = sizeof(struct sockaddr_in); 2702 dst.sa.sa_family = AF_INET; 2703 dst.sin.sin_addr = ip->ip_dst; 2704 2705 /* 2706 * Look up an SADB entry which matches the address of the peer. 2707 */ 2708 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2709 #else 2710 if (ip) 2711 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2712 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2713 htonl(TCP_SIG_SPI)); 2714 else 2715 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2716 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2717 htonl(TCP_SIG_SPI)); 2718 #endif 2719 2720 return (sav); /* freesav must be performed by caller */ 2721 } 2722 2723 int 2724 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2725 struct secasvar *sav, char *sig) 2726 { 2727 MD5_CTX ctx; 2728 struct ip *ip; 2729 struct ipovly *ipovly; 2730 struct ip6_hdr *ip6; 2731 struct ippseudo ippseudo; 2732 struct ip6_hdr_pseudo ip6pseudo; 2733 struct tcphdr th0; 2734 int l, tcphdrlen; 2735 2736 if (sav == NULL) 2737 return (-1); 2738 2739 tcphdrlen = th->th_off * 4; 2740 2741 switch (mtod(m, struct ip *)->ip_v) { 2742 case 4: 2743 ip = mtod(m, struct ip *); 2744 ip6 = NULL; 2745 break; 2746 case 6: 2747 ip = NULL; 2748 ip6 = mtod(m, struct ip6_hdr *); 2749 break; 2750 default: 2751 return (-1); 2752 } 2753 2754 MD5Init(&ctx); 2755 2756 if (ip) { 2757 memset(&ippseudo, 0, sizeof(ippseudo)); 2758 ipovly = (struct ipovly *)ip; 2759 ippseudo.ippseudo_src = ipovly->ih_src; 2760 ippseudo.ippseudo_dst = ipovly->ih_dst; 2761 ippseudo.ippseudo_pad = 0; 2762 ippseudo.ippseudo_p = IPPROTO_TCP; 2763 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2764 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2765 } else { 2766 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2767 ip6pseudo.ip6ph_src = ip6->ip6_src; 2768 in6_clearscope(&ip6pseudo.ip6ph_src); 2769 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2770 in6_clearscope(&ip6pseudo.ip6ph_dst); 2771 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2772 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2773 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2774 } 2775 2776 th0 = *th; 2777 th0.th_sum = 0; 2778 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2779 2780 l = m->m_pkthdr.len - thoff - tcphdrlen; 2781 if (l > 0) 2782 m_apply(m, thoff + tcphdrlen, 2783 m->m_pkthdr.len - thoff - tcphdrlen, 2784 tcp_signature_apply, &ctx); 2785 2786 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2787 MD5Final(sig, &ctx); 2788 2789 return (0); 2790 } 2791 #endif 2792 2793 int 2794 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, 2795 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2796 { 2797 u_int16_t mss; 2798 int opt, optlen = 0; 2799 #ifdef TCP_SIGNATURE 2800 caddr_t sigp = NULL; 2801 char sigbuf[TCP_SIGLEN]; 2802 struct secasvar *sav = NULL; 2803 #endif 2804 2805 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2806 opt = cp[0]; 2807 if (opt == TCPOPT_EOL) 2808 break; 2809 if (opt == TCPOPT_NOP) 2810 optlen = 1; 2811 else { 2812 if (cnt < 2) 2813 break; 2814 optlen = cp[1]; 2815 if (optlen < 2 || optlen > cnt) 2816 break; 2817 } 2818 switch (opt) { 2819 2820 default: 2821 continue; 2822 2823 case TCPOPT_MAXSEG: 2824 if (optlen != TCPOLEN_MAXSEG) 2825 continue; 2826 if (!(th->th_flags & TH_SYN)) 2827 continue; 2828 bcopy(cp + 2, &mss, sizeof(mss)); 2829 oi->maxseg = ntohs(mss); 2830 break; 2831 2832 case TCPOPT_WINDOW: 2833 if (optlen != TCPOLEN_WINDOW) 2834 continue; 2835 if (!(th->th_flags & TH_SYN)) 2836 continue; 2837 tp->t_flags |= TF_RCVD_SCALE; 2838 tp->requested_s_scale = cp[2]; 2839 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2840 #if 0 /*XXX*/ 2841 char *p; 2842 2843 if (ip) 2844 p = ntohl(ip->ip_src); 2845 #ifdef INET6 2846 else if (ip6) 2847 p = ip6_sprintf(&ip6->ip6_src); 2848 #endif 2849 else 2850 p = "(unknown)"; 2851 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2852 "assuming %d\n", 2853 tp->requested_s_scale, p, 2854 TCP_MAX_WINSHIFT); 2855 #else 2856 log(LOG_ERR, "TCP: invalid wscale %d, " 2857 "assuming %d\n", 2858 tp->requested_s_scale, 2859 TCP_MAX_WINSHIFT); 2860 #endif 2861 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2862 } 2863 break; 2864 2865 case TCPOPT_TIMESTAMP: 2866 if (optlen != TCPOLEN_TIMESTAMP) 2867 continue; 2868 oi->ts_present = 1; 2869 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2870 NTOHL(oi->ts_val); 2871 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2872 NTOHL(oi->ts_ecr); 2873 2874 /* 2875 * A timestamp received in a SYN makes 2876 * it ok to send timestamp requests and replies. 2877 */ 2878 if (th->th_flags & TH_SYN) { 2879 tp->t_flags |= TF_RCVD_TSTMP; 2880 tp->ts_recent = oi->ts_val; 2881 tp->ts_recent_age = tcp_now; 2882 } 2883 break; 2884 case TCPOPT_SACK_PERMITTED: 2885 if (optlen != TCPOLEN_SACK_PERMITTED) 2886 continue; 2887 if (!(th->th_flags & TH_SYN)) 2888 continue; 2889 if (tcp_do_sack) { 2890 tp->t_flags |= TF_SACK_PERMIT; 2891 tp->t_flags |= TF_WILL_SACK; 2892 } 2893 break; 2894 2895 case TCPOPT_SACK: 2896 tcp_sack_option(tp, th, cp, optlen); 2897 break; 2898 #ifdef TCP_SIGNATURE 2899 case TCPOPT_SIGNATURE: 2900 if (optlen != TCPOLEN_SIGNATURE) 2901 continue; 2902 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2903 return (-1); 2904 2905 sigp = sigbuf; 2906 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2907 memset(cp + 2, 0, TCP_SIGLEN); 2908 tp->t_flags |= TF_SIGNATURE; 2909 break; 2910 #endif 2911 } 2912 } 2913 2914 #ifdef TCP_SIGNATURE 2915 if (tp->t_flags & TF_SIGNATURE) { 2916 2917 sav = tcp_signature_getsav(m, th); 2918 2919 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2920 return (-1); 2921 } 2922 2923 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2924 if (sav == NULL) 2925 return (-1); 2926 #ifdef FAST_IPSEC 2927 KEY_FREESAV(&sav); 2928 #else 2929 key_freesav(sav); 2930 #endif 2931 return (-1); 2932 } 2933 2934 if (sigp) { 2935 char sig[TCP_SIGLEN]; 2936 2937 TCP_FIELDS_TO_NET(th); 2938 if (tcp_signature(m, th, toff, sav, sig) < 0) { 2939 TCP_FIELDS_TO_HOST(th); 2940 if (sav == NULL) 2941 return (-1); 2942 #ifdef FAST_IPSEC 2943 KEY_FREESAV(&sav); 2944 #else 2945 key_freesav(sav); 2946 #endif 2947 return (-1); 2948 } 2949 TCP_FIELDS_TO_HOST(th); 2950 2951 if (bcmp(sig, sigp, TCP_SIGLEN)) { 2952 tcpstat.tcps_badsig++; 2953 if (sav == NULL) 2954 return (-1); 2955 #ifdef FAST_IPSEC 2956 KEY_FREESAV(&sav); 2957 #else 2958 key_freesav(sav); 2959 #endif 2960 return (-1); 2961 } else 2962 tcpstat.tcps_goodsig++; 2963 2964 key_sa_recordxfer(sav, m); 2965 #ifdef FAST_IPSEC 2966 KEY_FREESAV(&sav); 2967 #else 2968 key_freesav(sav); 2969 #endif 2970 } 2971 #endif 2972 2973 return (0); 2974 } 2975 2976 /* 2977 * Pull out of band byte out of a segment so 2978 * it doesn't appear in the user's data queue. 2979 * It is still reflected in the segment length for 2980 * sequencing purposes. 2981 */ 2982 void 2983 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 2984 struct mbuf *m, int off) 2985 { 2986 int cnt = off + th->th_urp - 1; 2987 2988 while (cnt >= 0) { 2989 if (m->m_len > cnt) { 2990 char *cp = mtod(m, caddr_t) + cnt; 2991 struct tcpcb *tp = sototcpcb(so); 2992 2993 tp->t_iobc = *cp; 2994 tp->t_oobflags |= TCPOOB_HAVEDATA; 2995 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2996 m->m_len--; 2997 return; 2998 } 2999 cnt -= m->m_len; 3000 m = m->m_next; 3001 if (m == 0) 3002 break; 3003 } 3004 panic("tcp_pulloutofband"); 3005 } 3006 3007 /* 3008 * Collect new round-trip time estimate 3009 * and update averages and current timeout. 3010 */ 3011 void 3012 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3013 { 3014 int32_t delta; 3015 3016 tcpstat.tcps_rttupdated++; 3017 if (tp->t_srtt != 0) { 3018 /* 3019 * srtt is stored as fixed point with 3 bits after the 3020 * binary point (i.e., scaled by 8). The following magic 3021 * is equivalent to the smoothing algorithm in rfc793 with 3022 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3023 * point). Adjust rtt to origin 0. 3024 */ 3025 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3026 if ((tp->t_srtt += delta) <= 0) 3027 tp->t_srtt = 1 << 2; 3028 /* 3029 * We accumulate a smoothed rtt variance (actually, a 3030 * smoothed mean difference), then set the retransmit 3031 * timer to smoothed rtt + 4 times the smoothed variance. 3032 * rttvar is stored as fixed point with 2 bits after the 3033 * binary point (scaled by 4). The following is 3034 * equivalent to rfc793 smoothing with an alpha of .75 3035 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3036 * rfc793's wired-in beta. 3037 */ 3038 if (delta < 0) 3039 delta = -delta; 3040 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3041 if ((tp->t_rttvar += delta) <= 0) 3042 tp->t_rttvar = 1 << 2; 3043 } else { 3044 /* 3045 * No rtt measurement yet - use the unsmoothed rtt. 3046 * Set the variance to half the rtt (so our first 3047 * retransmit happens at 3*rtt). 3048 */ 3049 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3050 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3051 } 3052 tp->t_rtttime = 0; 3053 tp->t_rxtshift = 0; 3054 3055 /* 3056 * the retransmit should happen at rtt + 4 * rttvar. 3057 * Because of the way we do the smoothing, srtt and rttvar 3058 * will each average +1/2 tick of bias. When we compute 3059 * the retransmit timer, we want 1/2 tick of rounding and 3060 * 1 extra tick because of +-1/2 tick uncertainty in the 3061 * firing of the timer. The bias will give us exactly the 3062 * 1.5 tick we need. But, because the bias is 3063 * statistical, we have to test that we don't drop below 3064 * the minimum feasible timer (which is 2 ticks). 3065 */ 3066 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3067 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3068 3069 /* 3070 * We received an ack for a packet that wasn't retransmitted; 3071 * it is probably safe to discard any error indications we've 3072 * received recently. This isn't quite right, but close enough 3073 * for now (a route might have failed after we sent a segment, 3074 * and the return path might not be symmetrical). 3075 */ 3076 tp->t_softerror = 0; 3077 } 3078 3079 void 3080 tcp_reno_newack(struct tcpcb *tp, struct tcphdr *th) 3081 { 3082 if (tp->t_partialacks < 0) { 3083 /* 3084 * We were not in fast recovery. Reset the duplicate ack 3085 * counter. 3086 */ 3087 tp->t_dupacks = 0; 3088 } else { 3089 /* 3090 * Clamp the congestion window to the crossover point and 3091 * exit fast recovery. 3092 */ 3093 if (tp->snd_cwnd > tp->snd_ssthresh) 3094 tp->snd_cwnd = tp->snd_ssthresh; 3095 tp->t_partialacks = -1; 3096 tp->t_dupacks = 0; 3097 } 3098 } 3099 3100 /* 3101 * Implement the NewReno response to a new ack, checking for partial acks in 3102 * fast recovery. 3103 */ 3104 void 3105 tcp_newreno_newack(struct tcpcb *tp, struct tcphdr *th) 3106 { 3107 if (tp->t_partialacks < 0) { 3108 /* 3109 * We were not in fast recovery. Reset the duplicate ack 3110 * counter. 3111 */ 3112 tp->t_dupacks = 0; 3113 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3114 /* 3115 * This is a partial ack. Retransmit the first unacknowledged 3116 * segment and deflate the congestion window by the amount of 3117 * acknowledged data. Do not exit fast recovery. 3118 */ 3119 tcp_seq onxt = tp->snd_nxt; 3120 u_long ocwnd = tp->snd_cwnd; 3121 3122 /* 3123 * snd_una has not yet been updated and the socket's send 3124 * buffer has not yet drained off the ACK'd data, so we 3125 * have to leave snd_una as it was to get the correct data 3126 * offset in tcp_output(). 3127 */ 3128 if (++tp->t_partialacks == 1) 3129 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3130 tp->t_rtttime = 0; 3131 tp->snd_nxt = th->th_ack; 3132 /* 3133 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3134 * is not yet updated when we're called. 3135 */ 3136 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3137 (void) tcp_output(tp); 3138 tp->snd_cwnd = ocwnd; 3139 if (SEQ_GT(onxt, tp->snd_nxt)) 3140 tp->snd_nxt = onxt; 3141 /* 3142 * Partial window deflation. Relies on fact that tp->snd_una 3143 * not updated yet. 3144 */ 3145 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3146 } else { 3147 /* 3148 * Complete ack. Inflate the congestion window to ssthresh 3149 * and exit fast recovery. 3150 * 3151 * Window inflation should have left us with approx. 3152 * snd_ssthresh outstanding data. But in case we 3153 * would be inclined to send a burst, better to do 3154 * it via the slow start mechanism. 3155 */ 3156 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 3157 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 3158 + tp->t_segsz; 3159 else 3160 tp->snd_cwnd = tp->snd_ssthresh; 3161 tp->t_partialacks = -1; 3162 tp->t_dupacks = 0; 3163 } 3164 } 3165 3166 3167 /* 3168 * TCP compressed state engine. Currently used to hold compressed 3169 * state for SYN_RECEIVED. 3170 */ 3171 3172 u_long syn_cache_count; 3173 u_int32_t syn_hash1, syn_hash2; 3174 3175 #define SYN_HASH(sa, sp, dp) \ 3176 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3177 ((u_int32_t)(sp)))^syn_hash2))) 3178 #ifndef INET6 3179 #define SYN_HASHALL(hash, src, dst) \ 3180 do { \ 3181 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3182 ((struct sockaddr_in *)(src))->sin_port, \ 3183 ((struct sockaddr_in *)(dst))->sin_port); \ 3184 } while (/*CONSTCOND*/ 0) 3185 #else 3186 #define SYN_HASH6(sa, sp, dp) \ 3187 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3188 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3189 & 0x7fffffff) 3190 3191 #define SYN_HASHALL(hash, src, dst) \ 3192 do { \ 3193 switch ((src)->sa_family) { \ 3194 case AF_INET: \ 3195 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3196 ((struct sockaddr_in *)(src))->sin_port, \ 3197 ((struct sockaddr_in *)(dst))->sin_port); \ 3198 break; \ 3199 case AF_INET6: \ 3200 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 3201 ((struct sockaddr_in6 *)(src))->sin6_port, \ 3202 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 3203 break; \ 3204 default: \ 3205 hash = 0; \ 3206 } \ 3207 } while (/*CONSTCOND*/0) 3208 #endif /* INET6 */ 3209 3210 #define SYN_CACHE_RM(sc) \ 3211 do { \ 3212 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3213 (sc), sc_bucketq); \ 3214 (sc)->sc_tp = NULL; \ 3215 LIST_REMOVE((sc), sc_tpq); \ 3216 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3217 callout_stop(&(sc)->sc_timer); \ 3218 syn_cache_count--; \ 3219 } while (/*CONSTCOND*/0) 3220 3221 #define SYN_CACHE_PUT(sc) \ 3222 do { \ 3223 if ((sc)->sc_ipopts) \ 3224 (void) m_free((sc)->sc_ipopts); \ 3225 if ((sc)->sc_route4.ro_rt != NULL) \ 3226 RTFREE((sc)->sc_route4.ro_rt); \ 3227 if (callout_invoking(&(sc)->sc_timer)) \ 3228 (sc)->sc_flags |= SCF_DEAD; \ 3229 else \ 3230 pool_put(&syn_cache_pool, (sc)); \ 3231 } while (/*CONSTCOND*/0) 3232 3233 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3234 3235 /* 3236 * We don't estimate RTT with SYNs, so each packet starts with the default 3237 * RTT and each timer step has a fixed timeout value. 3238 */ 3239 #define SYN_CACHE_TIMER_ARM(sc) \ 3240 do { \ 3241 TCPT_RANGESET((sc)->sc_rxtcur, \ 3242 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3243 TCPTV_REXMTMAX); \ 3244 callout_reset(&(sc)->sc_timer, \ 3245 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3246 } while (/*CONSTCOND*/0) 3247 3248 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3249 3250 void 3251 syn_cache_init(void) 3252 { 3253 int i; 3254 3255 /* Initialize the hash buckets. */ 3256 for (i = 0; i < tcp_syn_cache_size; i++) 3257 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3258 } 3259 3260 void 3261 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3262 { 3263 struct syn_cache_head *scp; 3264 struct syn_cache *sc2; 3265 int s; 3266 3267 /* 3268 * If there are no entries in the hash table, reinitialize 3269 * the hash secrets. 3270 */ 3271 if (syn_cache_count == 0) { 3272 syn_hash1 = arc4random(); 3273 syn_hash2 = arc4random(); 3274 } 3275 3276 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3277 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3278 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3279 3280 /* 3281 * Make sure that we don't overflow the per-bucket 3282 * limit or the total cache size limit. 3283 */ 3284 s = splsoftnet(); 3285 if (scp->sch_length >= tcp_syn_bucket_limit) { 3286 tcpstat.tcps_sc_bucketoverflow++; 3287 /* 3288 * The bucket is full. Toss the oldest element in the 3289 * bucket. This will be the first entry in the bucket. 3290 */ 3291 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3292 #ifdef DIAGNOSTIC 3293 /* 3294 * This should never happen; we should always find an 3295 * entry in our bucket. 3296 */ 3297 if (sc2 == NULL) 3298 panic("syn_cache_insert: bucketoverflow: impossible"); 3299 #endif 3300 SYN_CACHE_RM(sc2); 3301 SYN_CACHE_PUT(sc2); 3302 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3303 struct syn_cache_head *scp2, *sce; 3304 3305 tcpstat.tcps_sc_overflowed++; 3306 /* 3307 * The cache is full. Toss the oldest entry in the 3308 * first non-empty bucket we can find. 3309 * 3310 * XXX We would really like to toss the oldest 3311 * entry in the cache, but we hope that this 3312 * condition doesn't happen very often. 3313 */ 3314 scp2 = scp; 3315 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3316 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3317 for (++scp2; scp2 != scp; scp2++) { 3318 if (scp2 >= sce) 3319 scp2 = &tcp_syn_cache[0]; 3320 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3321 break; 3322 } 3323 #ifdef DIAGNOSTIC 3324 /* 3325 * This should never happen; we should always find a 3326 * non-empty bucket. 3327 */ 3328 if (scp2 == scp) 3329 panic("syn_cache_insert: cacheoverflow: " 3330 "impossible"); 3331 #endif 3332 } 3333 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3334 SYN_CACHE_RM(sc2); 3335 SYN_CACHE_PUT(sc2); 3336 } 3337 3338 /* 3339 * Initialize the entry's timer. 3340 */ 3341 sc->sc_rxttot = 0; 3342 sc->sc_rxtshift = 0; 3343 SYN_CACHE_TIMER_ARM(sc); 3344 3345 /* Link it from tcpcb entry */ 3346 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3347 3348 /* Put it into the bucket. */ 3349 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3350 scp->sch_length++; 3351 syn_cache_count++; 3352 3353 tcpstat.tcps_sc_added++; 3354 splx(s); 3355 } 3356 3357 /* 3358 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3359 * If we have retransmitted an entry the maximum number of times, expire 3360 * that entry. 3361 */ 3362 void 3363 syn_cache_timer(void *arg) 3364 { 3365 struct syn_cache *sc = arg; 3366 int s; 3367 3368 s = splsoftnet(); 3369 callout_ack(&sc->sc_timer); 3370 3371 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3372 tcpstat.tcps_sc_delayed_free++; 3373 pool_put(&syn_cache_pool, sc); 3374 splx(s); 3375 return; 3376 } 3377 3378 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3379 /* Drop it -- too many retransmissions. */ 3380 goto dropit; 3381 } 3382 3383 /* 3384 * Compute the total amount of time this entry has 3385 * been on a queue. If this entry has been on longer 3386 * than the keep alive timer would allow, expire it. 3387 */ 3388 sc->sc_rxttot += sc->sc_rxtcur; 3389 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3390 goto dropit; 3391 3392 tcpstat.tcps_sc_retransmitted++; 3393 (void) syn_cache_respond(sc, NULL); 3394 3395 /* Advance the timer back-off. */ 3396 sc->sc_rxtshift++; 3397 SYN_CACHE_TIMER_ARM(sc); 3398 3399 splx(s); 3400 return; 3401 3402 dropit: 3403 tcpstat.tcps_sc_timed_out++; 3404 SYN_CACHE_RM(sc); 3405 SYN_CACHE_PUT(sc); 3406 splx(s); 3407 } 3408 3409 /* 3410 * Remove syn cache created by the specified tcb entry, 3411 * because this does not make sense to keep them 3412 * (if there's no tcb entry, syn cache entry will never be used) 3413 */ 3414 void 3415 syn_cache_cleanup(struct tcpcb *tp) 3416 { 3417 struct syn_cache *sc, *nsc; 3418 int s; 3419 3420 s = splsoftnet(); 3421 3422 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3423 nsc = LIST_NEXT(sc, sc_tpq); 3424 3425 #ifdef DIAGNOSTIC 3426 if (sc->sc_tp != tp) 3427 panic("invalid sc_tp in syn_cache_cleanup"); 3428 #endif 3429 SYN_CACHE_RM(sc); 3430 SYN_CACHE_PUT(sc); 3431 } 3432 /* just for safety */ 3433 LIST_INIT(&tp->t_sc); 3434 3435 splx(s); 3436 } 3437 3438 /* 3439 * Find an entry in the syn cache. 3440 */ 3441 struct syn_cache * 3442 syn_cache_lookup(struct sockaddr *src, struct sockaddr *dst, 3443 struct syn_cache_head **headp) 3444 { 3445 struct syn_cache *sc; 3446 struct syn_cache_head *scp; 3447 u_int32_t hash; 3448 int s; 3449 3450 SYN_HASHALL(hash, src, dst); 3451 3452 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3453 *headp = scp; 3454 s = splsoftnet(); 3455 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3456 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3457 if (sc->sc_hash != hash) 3458 continue; 3459 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3460 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3461 splx(s); 3462 return (sc); 3463 } 3464 } 3465 splx(s); 3466 return (NULL); 3467 } 3468 3469 /* 3470 * This function gets called when we receive an ACK for a 3471 * socket in the LISTEN state. We look up the connection 3472 * in the syn cache, and if its there, we pull it out of 3473 * the cache and turn it into a full-blown connection in 3474 * the SYN-RECEIVED state. 3475 * 3476 * The return values may not be immediately obvious, and their effects 3477 * can be subtle, so here they are: 3478 * 3479 * NULL SYN was not found in cache; caller should drop the 3480 * packet and send an RST. 3481 * 3482 * -1 We were unable to create the new connection, and are 3483 * aborting it. An ACK,RST is being sent to the peer 3484 * (unless we got screwey sequence numbners; see below), 3485 * because the 3-way handshake has been completed. Caller 3486 * should not free the mbuf, since we may be using it. If 3487 * we are not, we will free it. 3488 * 3489 * Otherwise, the return value is a pointer to the new socket 3490 * associated with the connection. 3491 */ 3492 struct socket * 3493 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3494 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3495 struct socket *so, struct mbuf *m) 3496 { 3497 struct syn_cache *sc; 3498 struct syn_cache_head *scp; 3499 struct inpcb *inp = NULL; 3500 #ifdef INET6 3501 struct in6pcb *in6p = NULL; 3502 #endif 3503 struct tcpcb *tp = 0; 3504 struct mbuf *am; 3505 int s; 3506 struct socket *oso; 3507 3508 s = splsoftnet(); 3509 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3510 splx(s); 3511 return (NULL); 3512 } 3513 3514 /* 3515 * Verify the sequence and ack numbers. Try getting the correct 3516 * response again. 3517 */ 3518 if ((th->th_ack != sc->sc_iss + 1) || 3519 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3520 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3521 (void) syn_cache_respond(sc, m); 3522 splx(s); 3523 return ((struct socket *)(-1)); 3524 } 3525 3526 /* Remove this cache entry */ 3527 SYN_CACHE_RM(sc); 3528 splx(s); 3529 3530 /* 3531 * Ok, create the full blown connection, and set things up 3532 * as they would have been set up if we had created the 3533 * connection when the SYN arrived. If we can't create 3534 * the connection, abort it. 3535 */ 3536 /* 3537 * inp still has the OLD in_pcb stuff, set the 3538 * v6-related flags on the new guy, too. This is 3539 * done particularly for the case where an AF_INET6 3540 * socket is bound only to a port, and a v4 connection 3541 * comes in on that port. 3542 * we also copy the flowinfo from the original pcb 3543 * to the new one. 3544 */ 3545 oso = so; 3546 so = sonewconn(so, SS_ISCONNECTED); 3547 if (so == NULL) 3548 goto resetandabort; 3549 3550 switch (so->so_proto->pr_domain->dom_family) { 3551 #ifdef INET 3552 case AF_INET: 3553 inp = sotoinpcb(so); 3554 break; 3555 #endif 3556 #ifdef INET6 3557 case AF_INET6: 3558 in6p = sotoin6pcb(so); 3559 break; 3560 #endif 3561 } 3562 switch (src->sa_family) { 3563 #ifdef INET 3564 case AF_INET: 3565 if (inp) { 3566 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3567 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3568 inp->inp_options = ip_srcroute(); 3569 in_pcbstate(inp, INP_BOUND); 3570 if (inp->inp_options == NULL) { 3571 inp->inp_options = sc->sc_ipopts; 3572 sc->sc_ipopts = NULL; 3573 } 3574 } 3575 #ifdef INET6 3576 else if (in6p) { 3577 /* IPv4 packet to AF_INET6 socket */ 3578 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3579 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3580 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3581 &in6p->in6p_laddr.s6_addr32[3], 3582 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3583 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3584 in6totcpcb(in6p)->t_family = AF_INET; 3585 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3586 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3587 else 3588 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3589 in6_pcbstate(in6p, IN6P_BOUND); 3590 } 3591 #endif 3592 break; 3593 #endif 3594 #ifdef INET6 3595 case AF_INET6: 3596 if (in6p) { 3597 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3598 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3599 in6_pcbstate(in6p, IN6P_BOUND); 3600 } 3601 break; 3602 #endif 3603 } 3604 #ifdef INET6 3605 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3606 struct in6pcb *oin6p = sotoin6pcb(oso); 3607 /* inherit socket options from the listening socket */ 3608 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3609 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3610 m_freem(in6p->in6p_options); 3611 in6p->in6p_options = 0; 3612 } 3613 ip6_savecontrol(in6p, &in6p->in6p_options, 3614 mtod(m, struct ip6_hdr *), m); 3615 } 3616 #endif 3617 3618 #if defined(IPSEC) || defined(FAST_IPSEC) 3619 /* 3620 * we make a copy of policy, instead of sharing the policy, 3621 * for better behavior in terms of SA lookup and dead SA removal. 3622 */ 3623 if (inp) { 3624 /* copy old policy into new socket's */ 3625 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3626 printf("tcp_input: could not copy policy\n"); 3627 } 3628 #ifdef INET6 3629 else if (in6p) { 3630 /* copy old policy into new socket's */ 3631 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3632 in6p->in6p_sp)) 3633 printf("tcp_input: could not copy policy\n"); 3634 } 3635 #endif 3636 #endif 3637 3638 /* 3639 * Give the new socket our cached route reference. 3640 */ 3641 if (inp) 3642 inp->inp_route = sc->sc_route4; /* struct assignment */ 3643 #ifdef INET6 3644 else 3645 in6p->in6p_route = sc->sc_route6; 3646 #endif 3647 sc->sc_route4.ro_rt = NULL; 3648 3649 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3650 if (am == NULL) 3651 goto resetandabort; 3652 MCLAIM(am, &tcp_mowner); 3653 am->m_len = src->sa_len; 3654 bcopy(src, mtod(am, caddr_t), src->sa_len); 3655 if (inp) { 3656 if (in_pcbconnect(inp, am)) { 3657 (void) m_free(am); 3658 goto resetandabort; 3659 } 3660 } 3661 #ifdef INET6 3662 else if (in6p) { 3663 if (src->sa_family == AF_INET) { 3664 /* IPv4 packet to AF_INET6 socket */ 3665 struct sockaddr_in6 *sin6; 3666 sin6 = mtod(am, struct sockaddr_in6 *); 3667 am->m_len = sizeof(*sin6); 3668 bzero(sin6, sizeof(*sin6)); 3669 sin6->sin6_family = AF_INET6; 3670 sin6->sin6_len = sizeof(*sin6); 3671 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3672 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3673 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3674 &sin6->sin6_addr.s6_addr32[3], 3675 sizeof(sin6->sin6_addr.s6_addr32[3])); 3676 } 3677 if (in6_pcbconnect(in6p, am)) { 3678 (void) m_free(am); 3679 goto resetandabort; 3680 } 3681 } 3682 #endif 3683 else { 3684 (void) m_free(am); 3685 goto resetandabort; 3686 } 3687 (void) m_free(am); 3688 3689 if (inp) 3690 tp = intotcpcb(inp); 3691 #ifdef INET6 3692 else if (in6p) 3693 tp = in6totcpcb(in6p); 3694 #endif 3695 else 3696 tp = NULL; 3697 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3698 if (sc->sc_request_r_scale != 15) { 3699 tp->requested_s_scale = sc->sc_requested_s_scale; 3700 tp->request_r_scale = sc->sc_request_r_scale; 3701 tp->snd_scale = sc->sc_requested_s_scale; 3702 tp->rcv_scale = sc->sc_request_r_scale; 3703 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3704 } 3705 if (sc->sc_flags & SCF_TIMESTAMP) 3706 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3707 tp->ts_timebase = sc->sc_timebase; 3708 3709 tp->t_template = tcp_template(tp); 3710 if (tp->t_template == 0) { 3711 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3712 so = NULL; 3713 m_freem(m); 3714 goto abort; 3715 } 3716 3717 tp->iss = sc->sc_iss; 3718 tp->irs = sc->sc_irs; 3719 tcp_sendseqinit(tp); 3720 tcp_rcvseqinit(tp); 3721 tp->t_state = TCPS_SYN_RECEIVED; 3722 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3723 tcpstat.tcps_accepts++; 3724 3725 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3726 tp->t_flags |= TF_WILL_SACK; 3727 3728 #ifdef TCP_SIGNATURE 3729 if (sc->sc_flags & SCF_SIGNATURE) 3730 tp->t_flags |= TF_SIGNATURE; 3731 #endif 3732 3733 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3734 tp->t_ourmss = sc->sc_ourmaxseg; 3735 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3736 3737 /* 3738 * Initialize the initial congestion window. If we 3739 * had to retransmit the SYN,ACK, we must initialize cwnd 3740 * to 1 segment (i.e. the Loss Window). 3741 */ 3742 if (sc->sc_rxtshift) 3743 tp->snd_cwnd = tp->t_peermss; 3744 else { 3745 int ss = tcp_init_win; 3746 #ifdef INET 3747 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3748 ss = tcp_init_win_local; 3749 #endif 3750 #ifdef INET6 3751 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3752 ss = tcp_init_win_local; 3753 #endif 3754 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3755 } 3756 3757 tcp_rmx_rtt(tp); 3758 tp->snd_wl1 = sc->sc_irs; 3759 tp->rcv_up = sc->sc_irs + 1; 3760 3761 /* 3762 * This is what whould have happened in tcp_output() when 3763 * the SYN,ACK was sent. 3764 */ 3765 tp->snd_up = tp->snd_una; 3766 tp->snd_max = tp->snd_nxt = tp->iss+1; 3767 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3768 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3769 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3770 tp->last_ack_sent = tp->rcv_nxt; 3771 tp->t_partialacks = -1; 3772 tp->t_dupacks = 0; 3773 3774 tcpstat.tcps_sc_completed++; 3775 SYN_CACHE_PUT(sc); 3776 return (so); 3777 3778 resetandabort: 3779 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3780 abort: 3781 if (so != NULL) 3782 (void) soabort(so); 3783 SYN_CACHE_PUT(sc); 3784 tcpstat.tcps_sc_aborted++; 3785 return ((struct socket *)(-1)); 3786 } 3787 3788 /* 3789 * This function is called when we get a RST for a 3790 * non-existent connection, so that we can see if the 3791 * connection is in the syn cache. If it is, zap it. 3792 */ 3793 3794 void 3795 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3796 { 3797 struct syn_cache *sc; 3798 struct syn_cache_head *scp; 3799 int s = splsoftnet(); 3800 3801 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3802 splx(s); 3803 return; 3804 } 3805 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3806 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3807 splx(s); 3808 return; 3809 } 3810 SYN_CACHE_RM(sc); 3811 splx(s); 3812 tcpstat.tcps_sc_reset++; 3813 SYN_CACHE_PUT(sc); 3814 } 3815 3816 void 3817 syn_cache_unreach(struct sockaddr *src, struct sockaddr *dst, 3818 struct tcphdr *th) 3819 { 3820 struct syn_cache *sc; 3821 struct syn_cache_head *scp; 3822 int s; 3823 3824 s = splsoftnet(); 3825 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3826 splx(s); 3827 return; 3828 } 3829 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3830 if (ntohl (th->th_seq) != sc->sc_iss) { 3831 splx(s); 3832 return; 3833 } 3834 3835 /* 3836 * If we've retransmitted 3 times and this is our second error, 3837 * we remove the entry. Otherwise, we allow it to continue on. 3838 * This prevents us from incorrectly nuking an entry during a 3839 * spurious network outage. 3840 * 3841 * See tcp_notify(). 3842 */ 3843 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3844 sc->sc_flags |= SCF_UNREACH; 3845 splx(s); 3846 return; 3847 } 3848 3849 SYN_CACHE_RM(sc); 3850 splx(s); 3851 tcpstat.tcps_sc_unreach++; 3852 SYN_CACHE_PUT(sc); 3853 } 3854 3855 /* 3856 * Given a LISTEN socket and an inbound SYN request, add 3857 * this to the syn cache, and send back a segment: 3858 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3859 * to the source. 3860 * 3861 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3862 * Doing so would require that we hold onto the data and deliver it 3863 * to the application. However, if we are the target of a SYN-flood 3864 * DoS attack, an attacker could send data which would eventually 3865 * consume all available buffer space if it were ACKed. By not ACKing 3866 * the data, we avoid this DoS scenario. 3867 */ 3868 3869 int 3870 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3871 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3872 int optlen, struct tcp_opt_info *oi) 3873 { 3874 struct tcpcb tb, *tp; 3875 long win; 3876 struct syn_cache *sc; 3877 struct syn_cache_head *scp; 3878 struct mbuf *ipopts; 3879 struct tcp_opt_info opti; 3880 3881 tp = sototcpcb(so); 3882 3883 bzero(&opti, sizeof(opti)); 3884 3885 /* 3886 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3887 * 3888 * Note this check is performed in tcp_input() very early on. 3889 */ 3890 3891 /* 3892 * Initialize some local state. 3893 */ 3894 win = sbspace(&so->so_rcv); 3895 if (win > TCP_MAXWIN) 3896 win = TCP_MAXWIN; 3897 3898 switch (src->sa_family) { 3899 #ifdef INET 3900 case AF_INET: 3901 /* 3902 * Remember the IP options, if any. 3903 */ 3904 ipopts = ip_srcroute(); 3905 break; 3906 #endif 3907 default: 3908 ipopts = NULL; 3909 } 3910 3911 #ifdef TCP_SIGNATURE 3912 if (optp || (tp->t_flags & TF_SIGNATURE)) 3913 #else 3914 if (optp) 3915 #endif 3916 { 3917 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3918 #ifdef TCP_SIGNATURE 3919 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3920 #endif 3921 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3922 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3923 return (0); 3924 } else 3925 tb.t_flags = 0; 3926 3927 /* 3928 * See if we already have an entry for this connection. 3929 * If we do, resend the SYN,ACK. We do not count this 3930 * as a retransmission (XXX though maybe we should). 3931 */ 3932 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3933 tcpstat.tcps_sc_dupesyn++; 3934 if (ipopts) { 3935 /* 3936 * If we were remembering a previous source route, 3937 * forget it and use the new one we've been given. 3938 */ 3939 if (sc->sc_ipopts) 3940 (void) m_free(sc->sc_ipopts); 3941 sc->sc_ipopts = ipopts; 3942 } 3943 sc->sc_timestamp = tb.ts_recent; 3944 if (syn_cache_respond(sc, m) == 0) { 3945 tcpstat.tcps_sndacks++; 3946 tcpstat.tcps_sndtotal++; 3947 } 3948 return (1); 3949 } 3950 3951 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3952 if (sc == NULL) { 3953 if (ipopts) 3954 (void) m_free(ipopts); 3955 return (0); 3956 } 3957 3958 /* 3959 * Fill in the cache, and put the necessary IP and TCP 3960 * options into the reply. 3961 */ 3962 bzero(sc, sizeof(struct syn_cache)); 3963 callout_init(&sc->sc_timer); 3964 bcopy(src, &sc->sc_src, src->sa_len); 3965 bcopy(dst, &sc->sc_dst, dst->sa_len); 3966 sc->sc_flags = 0; 3967 sc->sc_ipopts = ipopts; 3968 sc->sc_irs = th->th_seq; 3969 switch (src->sa_family) { 3970 #ifdef INET 3971 case AF_INET: 3972 { 3973 struct sockaddr_in *srcin = (void *) src; 3974 struct sockaddr_in *dstin = (void *) dst; 3975 3976 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3977 &srcin->sin_addr, dstin->sin_port, 3978 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3979 break; 3980 } 3981 #endif /* INET */ 3982 #ifdef INET6 3983 case AF_INET6: 3984 { 3985 struct sockaddr_in6 *srcin6 = (void *) src; 3986 struct sockaddr_in6 *dstin6 = (void *) dst; 3987 3988 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3989 &srcin6->sin6_addr, dstin6->sin6_port, 3990 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3991 break; 3992 } 3993 #endif /* INET6 */ 3994 } 3995 sc->sc_peermaxseg = oi->maxseg; 3996 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3997 m->m_pkthdr.rcvif : NULL, 3998 sc->sc_src.sa.sa_family); 3999 sc->sc_win = win; 4000 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 4001 sc->sc_timestamp = tb.ts_recent; 4002 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4003 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4004 sc->sc_flags |= SCF_TIMESTAMP; 4005 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4006 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4007 sc->sc_requested_s_scale = tb.requested_s_scale; 4008 sc->sc_request_r_scale = 0; 4009 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4010 TCP_MAXWIN << sc->sc_request_r_scale < 4011 so->so_rcv.sb_hiwat) 4012 sc->sc_request_r_scale++; 4013 } else { 4014 sc->sc_requested_s_scale = 15; 4015 sc->sc_request_r_scale = 15; 4016 } 4017 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4018 sc->sc_flags |= SCF_SACK_PERMIT; 4019 #ifdef TCP_SIGNATURE 4020 if (tb.t_flags & TF_SIGNATURE) 4021 sc->sc_flags |= SCF_SIGNATURE; 4022 #endif 4023 sc->sc_tp = tp; 4024 if (syn_cache_respond(sc, m) == 0) { 4025 syn_cache_insert(sc, tp); 4026 tcpstat.tcps_sndacks++; 4027 tcpstat.tcps_sndtotal++; 4028 } else { 4029 SYN_CACHE_PUT(sc); 4030 tcpstat.tcps_sc_dropped++; 4031 } 4032 return (1); 4033 } 4034 4035 int 4036 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4037 { 4038 struct route *ro; 4039 u_int8_t *optp; 4040 int optlen, error; 4041 u_int16_t tlen; 4042 struct ip *ip = NULL; 4043 #ifdef INET6 4044 struct ip6_hdr *ip6 = NULL; 4045 #endif 4046 struct tcpcb *tp; 4047 struct tcphdr *th; 4048 u_int hlen; 4049 struct socket *so; 4050 4051 switch (sc->sc_src.sa.sa_family) { 4052 case AF_INET: 4053 hlen = sizeof(struct ip); 4054 ro = &sc->sc_route4; 4055 break; 4056 #ifdef INET6 4057 case AF_INET6: 4058 hlen = sizeof(struct ip6_hdr); 4059 ro = (struct route *)&sc->sc_route6; 4060 break; 4061 #endif 4062 default: 4063 if (m) 4064 m_freem(m); 4065 return (EAFNOSUPPORT); 4066 } 4067 4068 /* Compute the size of the TCP options. */ 4069 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4070 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4071 #ifdef TCP_SIGNATURE 4072 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4073 #endif 4074 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4075 4076 tlen = hlen + sizeof(struct tcphdr) + optlen; 4077 4078 /* 4079 * Create the IP+TCP header from scratch. 4080 */ 4081 if (m) 4082 m_freem(m); 4083 #ifdef DIAGNOSTIC 4084 if (max_linkhdr + tlen > MCLBYTES) 4085 return (ENOBUFS); 4086 #endif 4087 MGETHDR(m, M_DONTWAIT, MT_DATA); 4088 if (m && tlen > MHLEN) { 4089 MCLGET(m, M_DONTWAIT); 4090 if ((m->m_flags & M_EXT) == 0) { 4091 m_freem(m); 4092 m = NULL; 4093 } 4094 } 4095 if (m == NULL) 4096 return (ENOBUFS); 4097 MCLAIM(m, &tcp_tx_mowner); 4098 4099 /* Fixup the mbuf. */ 4100 m->m_data += max_linkhdr; 4101 m->m_len = m->m_pkthdr.len = tlen; 4102 if (sc->sc_tp) { 4103 tp = sc->sc_tp; 4104 if (tp->t_inpcb) 4105 so = tp->t_inpcb->inp_socket; 4106 #ifdef INET6 4107 else if (tp->t_in6pcb) 4108 so = tp->t_in6pcb->in6p_socket; 4109 #endif 4110 else 4111 so = NULL; 4112 } else 4113 so = NULL; 4114 m->m_pkthdr.rcvif = NULL; 4115 memset(mtod(m, u_char *), 0, tlen); 4116 4117 switch (sc->sc_src.sa.sa_family) { 4118 case AF_INET: 4119 ip = mtod(m, struct ip *); 4120 ip->ip_v = 4; 4121 ip->ip_dst = sc->sc_src.sin.sin_addr; 4122 ip->ip_src = sc->sc_dst.sin.sin_addr; 4123 ip->ip_p = IPPROTO_TCP; 4124 th = (struct tcphdr *)(ip + 1); 4125 th->th_dport = sc->sc_src.sin.sin_port; 4126 th->th_sport = sc->sc_dst.sin.sin_port; 4127 break; 4128 #ifdef INET6 4129 case AF_INET6: 4130 ip6 = mtod(m, struct ip6_hdr *); 4131 ip6->ip6_vfc = IPV6_VERSION; 4132 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4133 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4134 ip6->ip6_nxt = IPPROTO_TCP; 4135 /* ip6_plen will be updated in ip6_output() */ 4136 th = (struct tcphdr *)(ip6 + 1); 4137 th->th_dport = sc->sc_src.sin6.sin6_port; 4138 th->th_sport = sc->sc_dst.sin6.sin6_port; 4139 break; 4140 #endif 4141 default: 4142 th = NULL; 4143 } 4144 4145 th->th_seq = htonl(sc->sc_iss); 4146 th->th_ack = htonl(sc->sc_irs + 1); 4147 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4148 th->th_flags = TH_SYN|TH_ACK; 4149 th->th_win = htons(sc->sc_win); 4150 /* th_sum already 0 */ 4151 /* th_urp already 0 */ 4152 4153 /* Tack on the TCP options. */ 4154 optp = (u_int8_t *)(th + 1); 4155 *optp++ = TCPOPT_MAXSEG; 4156 *optp++ = 4; 4157 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4158 *optp++ = sc->sc_ourmaxseg & 0xff; 4159 4160 if (sc->sc_request_r_scale != 15) { 4161 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4162 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4163 sc->sc_request_r_scale); 4164 optp += 4; 4165 } 4166 4167 if (sc->sc_flags & SCF_TIMESTAMP) { 4168 u_int32_t *lp = (u_int32_t *)(optp); 4169 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4170 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4171 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4172 *lp = htonl(sc->sc_timestamp); 4173 optp += TCPOLEN_TSTAMP_APPA; 4174 } 4175 4176 if (sc->sc_flags & SCF_SACK_PERMIT) { 4177 u_int8_t *p = optp; 4178 4179 /* Let the peer know that we will SACK. */ 4180 p[0] = TCPOPT_SACK_PERMITTED; 4181 p[1] = 2; 4182 p[2] = TCPOPT_NOP; 4183 p[3] = TCPOPT_NOP; 4184 optp += 4; 4185 } 4186 4187 #ifdef TCP_SIGNATURE 4188 if (sc->sc_flags & SCF_SIGNATURE) { 4189 struct secasvar *sav; 4190 u_int8_t *sigp; 4191 4192 sav = tcp_signature_getsav(m, th); 4193 4194 if (sav == NULL) { 4195 if (m) 4196 m_freem(m); 4197 return (EPERM); 4198 } 4199 4200 *optp++ = TCPOPT_SIGNATURE; 4201 *optp++ = TCPOLEN_SIGNATURE; 4202 sigp = optp; 4203 bzero(optp, TCP_SIGLEN); 4204 optp += TCP_SIGLEN; 4205 *optp++ = TCPOPT_NOP; 4206 *optp++ = TCPOPT_EOL; 4207 4208 (void)tcp_signature(m, th, hlen, sav, sigp); 4209 4210 key_sa_recordxfer(sav, m); 4211 #ifdef FAST_IPSEC 4212 KEY_FREESAV(&sav); 4213 #else 4214 key_freesav(sav); 4215 #endif 4216 } 4217 #endif 4218 4219 /* Compute the packet's checksum. */ 4220 switch (sc->sc_src.sa.sa_family) { 4221 case AF_INET: 4222 ip->ip_len = htons(tlen - hlen); 4223 th->th_sum = 0; 4224 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4225 break; 4226 #ifdef INET6 4227 case AF_INET6: 4228 ip6->ip6_plen = htons(tlen - hlen); 4229 th->th_sum = 0; 4230 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4231 break; 4232 #endif 4233 } 4234 4235 /* 4236 * Fill in some straggling IP bits. Note the stack expects 4237 * ip_len to be in host order, for convenience. 4238 */ 4239 switch (sc->sc_src.sa.sa_family) { 4240 #ifdef INET 4241 case AF_INET: 4242 ip->ip_len = htons(tlen); 4243 ip->ip_ttl = ip_defttl; 4244 /* XXX tos? */ 4245 break; 4246 #endif 4247 #ifdef INET6 4248 case AF_INET6: 4249 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4250 ip6->ip6_vfc |= IPV6_VERSION; 4251 ip6->ip6_plen = htons(tlen - hlen); 4252 /* ip6_hlim will be initialized afterwards */ 4253 /* XXX flowlabel? */ 4254 break; 4255 #endif 4256 } 4257 4258 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4259 tp = sc->sc_tp; 4260 4261 switch (sc->sc_src.sa.sa_family) { 4262 #ifdef INET 4263 case AF_INET: 4264 error = ip_output(m, sc->sc_ipopts, ro, 4265 (ip_mtudisc ? IP_MTUDISC : 0), 4266 (struct ip_moptions *)NULL, so); 4267 break; 4268 #endif 4269 #ifdef INET6 4270 case AF_INET6: 4271 ip6->ip6_hlim = in6_selecthlim(NULL, 4272 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4273 4274 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4275 (struct ip6_moptions *)0, so, NULL); 4276 break; 4277 #endif 4278 default: 4279 error = EAFNOSUPPORT; 4280 break; 4281 } 4282 return (error); 4283 } 4284
Indexes created Mon Sep 22 10:09:38 GMT 2025