tcp_input.c revision 1.229
1 /* $NetBSD: tcp_input.c,v 1.229 2005/06/06 12:10:09 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 1.229 2005/06/06 12:10:09 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 uint32_t ts_rtt; 926 927 MCLAIM(m, &tcp_rx_mowner); 928 va_start(ap, m); 929 toff = va_arg(ap, int); 930 (void)va_arg(ap, int); /* ignore value, advance ap */ 931 va_end(ap); 932 933 tcpstat.tcps_rcvtotal++; 934 935 bzero(&opti, sizeof(opti)); 936 opti.ts_present = 0; 937 opti.maxseg = 0; 938 939 /* 940 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN. 941 * 942 * TCP is, by definition, unicast, so we reject all 943 * multicast outright. 944 * 945 * Note, there are additional src/dst address checks in 946 * the AF-specific code below. 947 */ 948 if (m->m_flags & (M_BCAST|M_MCAST)) { 949 /* XXX stat */ 950 goto drop; 951 } 952 #ifdef INET6 953 if (m->m_flags & M_ANYCAST6) { 954 /* XXX stat */ 955 goto drop; 956 } 957 #endif 958 959 /* 960 * Get IP and TCP header. 961 * Note: IP leaves IP header in first mbuf. 962 */ 963 ip = mtod(m, struct ip *); 964 #ifdef INET6 965 ip6 = NULL; 966 #endif 967 switch (ip->ip_v) { 968 #ifdef INET 969 case 4: 970 af = AF_INET; 971 iphlen = sizeof(struct ip); 972 ip = mtod(m, struct ip *); 973 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 974 sizeof(struct tcphdr)); 975 if (th == NULL) { 976 tcpstat.tcps_rcvshort++; 977 return; 978 } 979 /* We do the checksum after PCB lookup... */ 980 len = ntohs(ip->ip_len); 981 tlen = len - toff; 982 break; 983 #endif 984 #ifdef INET6 985 case 6: 986 ip = NULL; 987 iphlen = sizeof(struct ip6_hdr); 988 af = AF_INET6; 989 ip6 = mtod(m, struct ip6_hdr *); 990 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, 991 sizeof(struct tcphdr)); 992 if (th == NULL) { 993 tcpstat.tcps_rcvshort++; 994 return; 995 } 996 997 /* Be proactive about malicious use of IPv4 mapped address */ 998 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 999 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 1000 /* XXX stat */ 1001 goto drop; 1002 } 1003 1004 /* 1005 * Be proactive about unspecified IPv6 address in source. 1006 * As we use all-zero to indicate unbounded/unconnected pcb, 1007 * unspecified IPv6 address can be used to confuse us. 1008 * 1009 * Note that packets with unspecified IPv6 destination is 1010 * already dropped in ip6_input. 1011 */ 1012 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 1013 /* XXX stat */ 1014 goto drop; 1015 } 1016 1017 /* 1018 * Make sure destination address is not multicast. 1019 * Source address checked in ip6_input(). 1020 */ 1021 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 1022 /* XXX stat */ 1023 goto drop; 1024 } 1025 1026 /* We do the checksum after PCB lookup... */ 1027 len = m->m_pkthdr.len; 1028 tlen = len - toff; 1029 break; 1030 #endif 1031 default: 1032 m_freem(m); 1033 return; 1034 } 1035 1036 KASSERT(TCP_HDR_ALIGNED_P(th)); 1037 1038 /* 1039 * Check that TCP offset makes sense, 1040 * pull out TCP options and adjust length. XXX 1041 */ 1042 off = th->th_off << 2; 1043 if (off < sizeof (struct tcphdr) || off > tlen) { 1044 tcpstat.tcps_rcvbadoff++; 1045 goto drop; 1046 } 1047 tlen -= off; 1048 1049 /* 1050 * tcp_input() has been modified to use tlen to mean the TCP data 1051 * length throughout the function. Other functions can use 1052 * m->m_pkthdr.len as the basis for calculating the TCP data length. 1053 * rja 1054 */ 1055 1056 if (off > sizeof (struct tcphdr)) { 1057 IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off); 1058 if (th == NULL) { 1059 tcpstat.tcps_rcvshort++; 1060 return; 1061 } 1062 /* 1063 * NOTE: ip/ip6 will not be affected by m_pulldown() 1064 * (as they're before toff) and we don't need to update those. 1065 */ 1066 KASSERT(TCP_HDR_ALIGNED_P(th)); 1067 optlen = off - sizeof (struct tcphdr); 1068 optp = ((u_int8_t *)th) + sizeof(struct tcphdr); 1069 /* 1070 * Do quick retrieval of timestamp options ("options 1071 * prediction?"). If timestamp is the only option and it's 1072 * formatted as recommended in RFC 1323 appendix A, we 1073 * quickly get the values now and not bother calling 1074 * tcp_dooptions(), etc. 1075 */ 1076 if ((optlen == TCPOLEN_TSTAMP_APPA || 1077 (optlen > TCPOLEN_TSTAMP_APPA && 1078 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 1079 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 1080 (th->th_flags & TH_SYN) == 0) { 1081 opti.ts_present = 1; 1082 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 1083 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 1084 optp = NULL; /* we've parsed the options */ 1085 } 1086 } 1087 tiflags = th->th_flags; 1088 1089 /* 1090 * Locate pcb for segment. 1091 */ 1092 findpcb: 1093 inp = NULL; 1094 #ifdef INET6 1095 in6p = NULL; 1096 #endif 1097 switch (af) { 1098 #ifdef INET 1099 case AF_INET: 1100 inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport, 1101 ip->ip_dst, th->th_dport); 1102 if (inp == 0) { 1103 ++tcpstat.tcps_pcbhashmiss; 1104 inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport); 1105 } 1106 #ifdef INET6 1107 if (inp == 0) { 1108 struct in6_addr s, d; 1109 1110 /* mapped addr case */ 1111 bzero(&s, sizeof(s)); 1112 s.s6_addr16[5] = htons(0xffff); 1113 bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src)); 1114 bzero(&d, sizeof(d)); 1115 d.s6_addr16[5] = htons(0xffff); 1116 bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst)); 1117 in6p = in6_pcblookup_connect(&tcbtable, &s, 1118 th->th_sport, &d, th->th_dport, 0); 1119 if (in6p == 0) { 1120 ++tcpstat.tcps_pcbhashmiss; 1121 in6p = in6_pcblookup_bind(&tcbtable, &d, 1122 th->th_dport, 0); 1123 } 1124 } 1125 #endif 1126 #ifndef INET6 1127 if (inp == 0) 1128 #else 1129 if (inp == 0 && in6p == 0) 1130 #endif 1131 { 1132 ++tcpstat.tcps_noport; 1133 if (tcp_log_refused && 1134 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1135 tcp4_log_refused(ip, th); 1136 } 1137 TCP_FIELDS_TO_HOST(th); 1138 goto dropwithreset_ratelim; 1139 } 1140 #if defined(IPSEC) || defined(FAST_IPSEC) 1141 if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 && 1142 ipsec4_in_reject(m, inp)) { 1143 ipsecstat.in_polvio++; 1144 goto drop; 1145 } 1146 #ifdef INET6 1147 else if (in6p && 1148 (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1149 ipsec4_in_reject_so(m, in6p->in6p_socket)) { 1150 ipsecstat.in_polvio++; 1151 goto drop; 1152 } 1153 #endif 1154 #endif /*IPSEC*/ 1155 break; 1156 #endif /*INET*/ 1157 #ifdef INET6 1158 case AF_INET6: 1159 { 1160 int faith; 1161 1162 #if defined(NFAITH) && NFAITH > 0 1163 faith = faithprefix(&ip6->ip6_dst); 1164 #else 1165 faith = 0; 1166 #endif 1167 in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src, 1168 th->th_sport, &ip6->ip6_dst, th->th_dport, faith); 1169 if (in6p == NULL) { 1170 ++tcpstat.tcps_pcbhashmiss; 1171 in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst, 1172 th->th_dport, faith); 1173 } 1174 if (in6p == NULL) { 1175 ++tcpstat.tcps_noport; 1176 if (tcp_log_refused && 1177 (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) { 1178 tcp6_log_refused(ip6, th); 1179 } 1180 TCP_FIELDS_TO_HOST(th); 1181 goto dropwithreset_ratelim; 1182 } 1183 #if defined(IPSEC) || defined(FAST_IPSEC) 1184 if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 && 1185 ipsec6_in_reject(m, in6p)) { 1186 ipsec6stat.in_polvio++; 1187 goto drop; 1188 } 1189 #endif /*IPSEC*/ 1190 break; 1191 } 1192 #endif 1193 } 1194 1195 /* 1196 * If the state is CLOSED (i.e., TCB does not exist) then 1197 * all data in the incoming segment is discarded. 1198 * If the TCB exists but is in CLOSED state, it is embryonic, 1199 * but should either do a listen or a connect soon. 1200 */ 1201 tp = NULL; 1202 so = NULL; 1203 if (inp) { 1204 tp = intotcpcb(inp); 1205 so = inp->inp_socket; 1206 } 1207 #ifdef INET6 1208 else if (in6p) { 1209 tp = in6totcpcb(in6p); 1210 so = in6p->in6p_socket; 1211 } 1212 #endif 1213 if (tp == 0) { 1214 TCP_FIELDS_TO_HOST(th); 1215 goto dropwithreset_ratelim; 1216 } 1217 if (tp->t_state == TCPS_CLOSED) 1218 goto drop; 1219 1220 /* 1221 * Checksum extended TCP header and data. 1222 */ 1223 if (tcp_input_checksum(af, m, th, toff, off, tlen)) 1224 goto badcsum; 1225 1226 TCP_FIELDS_TO_HOST(th); 1227 1228 /* Unscale the window into a 32-bit value. */ 1229 if ((tiflags & TH_SYN) == 0) 1230 tiwin = th->th_win << tp->snd_scale; 1231 else 1232 tiwin = th->th_win; 1233 1234 #ifdef INET6 1235 /* save packet options if user wanted */ 1236 if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) { 1237 if (in6p->in6p_options) { 1238 m_freem(in6p->in6p_options); 1239 in6p->in6p_options = 0; 1240 } 1241 ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m); 1242 } 1243 #endif 1244 1245 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 1246 union syn_cache_sa src; 1247 union syn_cache_sa dst; 1248 1249 bzero(&src, sizeof(src)); 1250 bzero(&dst, sizeof(dst)); 1251 switch (af) { 1252 #ifdef INET 1253 case AF_INET: 1254 src.sin.sin_len = sizeof(struct sockaddr_in); 1255 src.sin.sin_family = AF_INET; 1256 src.sin.sin_addr = ip->ip_src; 1257 src.sin.sin_port = th->th_sport; 1258 1259 dst.sin.sin_len = sizeof(struct sockaddr_in); 1260 dst.sin.sin_family = AF_INET; 1261 dst.sin.sin_addr = ip->ip_dst; 1262 dst.sin.sin_port = th->th_dport; 1263 break; 1264 #endif 1265 #ifdef INET6 1266 case AF_INET6: 1267 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 1268 src.sin6.sin6_family = AF_INET6; 1269 src.sin6.sin6_addr = ip6->ip6_src; 1270 src.sin6.sin6_port = th->th_sport; 1271 1272 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 1273 dst.sin6.sin6_family = AF_INET6; 1274 dst.sin6.sin6_addr = ip6->ip6_dst; 1275 dst.sin6.sin6_port = th->th_dport; 1276 break; 1277 #endif /* INET6 */ 1278 default: 1279 goto badsyn; /*sanity*/ 1280 } 1281 1282 if (so->so_options & SO_DEBUG) { 1283 #ifdef TCP_DEBUG 1284 ostate = tp->t_state; 1285 #endif 1286 1287 tcp_saveti = NULL; 1288 if (iphlen + sizeof(struct tcphdr) > MHLEN) 1289 goto nosave; 1290 1291 if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) { 1292 tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT); 1293 if (!tcp_saveti) 1294 goto nosave; 1295 } else { 1296 MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER); 1297 if (!tcp_saveti) 1298 goto nosave; 1299 MCLAIM(m, &tcp_mowner); 1300 tcp_saveti->m_len = iphlen; 1301 m_copydata(m, 0, iphlen, 1302 mtod(tcp_saveti, caddr_t)); 1303 } 1304 1305 if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) { 1306 m_freem(tcp_saveti); 1307 tcp_saveti = NULL; 1308 } else { 1309 tcp_saveti->m_len += sizeof(struct tcphdr); 1310 bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen, 1311 sizeof(struct tcphdr)); 1312 } 1313 nosave:; 1314 } 1315 if (so->so_options & SO_ACCEPTCONN) { 1316 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 1317 if (tiflags & TH_RST) { 1318 syn_cache_reset(&src.sa, &dst.sa, th); 1319 } else if ((tiflags & (TH_ACK|TH_SYN)) == 1320 (TH_ACK|TH_SYN)) { 1321 /* 1322 * Received a SYN,ACK. This should 1323 * never happen while we are in 1324 * LISTEN. Send an RST. 1325 */ 1326 goto badsyn; 1327 } else if (tiflags & TH_ACK) { 1328 so = syn_cache_get(&src.sa, &dst.sa, 1329 th, toff, tlen, so, m); 1330 if (so == NULL) { 1331 /* 1332 * We don't have a SYN for 1333 * this ACK; send an RST. 1334 */ 1335 goto badsyn; 1336 } else if (so == 1337 (struct socket *)(-1)) { 1338 /* 1339 * We were unable to create 1340 * the connection. If the 1341 * 3-way handshake was 1342 * completed, and RST has 1343 * been sent to the peer. 1344 * Since the mbuf might be 1345 * in use for the reply, 1346 * do not free it. 1347 */ 1348 m = NULL; 1349 } else { 1350 /* 1351 * We have created a 1352 * full-blown connection. 1353 */ 1354 tp = NULL; 1355 inp = NULL; 1356 #ifdef INET6 1357 in6p = NULL; 1358 #endif 1359 switch (so->so_proto->pr_domain->dom_family) { 1360 #ifdef INET 1361 case AF_INET: 1362 inp = sotoinpcb(so); 1363 tp = intotcpcb(inp); 1364 break; 1365 #endif 1366 #ifdef INET6 1367 case AF_INET6: 1368 in6p = sotoin6pcb(so); 1369 tp = in6totcpcb(in6p); 1370 break; 1371 #endif 1372 } 1373 if (tp == NULL) 1374 goto badsyn; /*XXX*/ 1375 tiwin <<= tp->snd_scale; 1376 goto after_listen; 1377 } 1378 } else { 1379 /* 1380 * None of RST, SYN or ACK was set. 1381 * This is an invalid packet for a 1382 * TCB in LISTEN state. Send a RST. 1383 */ 1384 goto badsyn; 1385 } 1386 } else { 1387 /* 1388 * Received a SYN. 1389 */ 1390 1391 #ifdef INET6 1392 /* 1393 * If deprecated address is forbidden, we do 1394 * not accept SYN to deprecated interface 1395 * address to prevent any new inbound 1396 * connection from getting established. 1397 * When we do not accept SYN, we send a TCP 1398 * RST, with deprecated source address (instead 1399 * of dropping it). We compromise it as it is 1400 * much better for peer to send a RST, and 1401 * RST will be the final packet for the 1402 * exchange. 1403 * 1404 * If we do not forbid deprecated addresses, we 1405 * accept the SYN packet. RFC2462 does not 1406 * suggest dropping SYN in this case. 1407 * If we decipher RFC2462 5.5.4, it says like 1408 * this: 1409 * 1. use of deprecated addr with existing 1410 * communication is okay - "SHOULD continue 1411 * to be used" 1412 * 2. use of it with new communication: 1413 * (2a) "SHOULD NOT be used if alternate 1414 * address with sufficient scope is 1415 * available" 1416 * (2b) nothing mentioned otherwise. 1417 * Here we fall into (2b) case as we have no 1418 * choice in our source address selection - we 1419 * must obey the peer. 1420 * 1421 * The wording in RFC2462 is confusing, and 1422 * there are multiple description text for 1423 * deprecated address handling - worse, they 1424 * are not exactly the same. I believe 5.5.4 1425 * is the best one, so we follow 5.5.4. 1426 */ 1427 if (af == AF_INET6 && !ip6_use_deprecated) { 1428 struct in6_ifaddr *ia6; 1429 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 1430 &ip6->ip6_dst)) && 1431 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1432 tp = NULL; 1433 goto dropwithreset; 1434 } 1435 } 1436 #endif 1437 1438 #ifdef IPSEC 1439 switch (af) { 1440 #ifdef INET 1441 case AF_INET: 1442 if (ipsec4_in_reject_so(m, so)) { 1443 ipsecstat.in_polvio++; 1444 tp = NULL; 1445 goto dropwithreset; 1446 } 1447 break; 1448 #endif 1449 #ifdef INET6 1450 case AF_INET6: 1451 if (ipsec6_in_reject_so(m, so)) { 1452 ipsec6stat.in_polvio++; 1453 tp = NULL; 1454 goto dropwithreset; 1455 } 1456 break; 1457 #endif 1458 } 1459 #endif 1460 1461 /* 1462 * LISTEN socket received a SYN 1463 * from itself? This can't possibly 1464 * be valid; drop the packet. 1465 */ 1466 if (th->th_sport == th->th_dport) { 1467 int i; 1468 1469 switch (af) { 1470 #ifdef INET 1471 case AF_INET: 1472 i = in_hosteq(ip->ip_src, ip->ip_dst); 1473 break; 1474 #endif 1475 #ifdef INET6 1476 case AF_INET6: 1477 i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst); 1478 break; 1479 #endif 1480 default: 1481 i = 1; 1482 } 1483 if (i) { 1484 tcpstat.tcps_badsyn++; 1485 goto drop; 1486 } 1487 } 1488 1489 /* 1490 * SYN looks ok; create compressed TCP 1491 * state for it. 1492 */ 1493 if (so->so_qlen <= so->so_qlimit && 1494 syn_cache_add(&src.sa, &dst.sa, th, tlen, 1495 so, m, optp, optlen, &opti)) 1496 m = NULL; 1497 } 1498 goto drop; 1499 } 1500 } 1501 1502 after_listen: 1503 #ifdef DIAGNOSTIC 1504 /* 1505 * Should not happen now that all embryonic connections 1506 * are handled with compressed state. 1507 */ 1508 if (tp->t_state == TCPS_LISTEN) 1509 panic("tcp_input: TCPS_LISTEN"); 1510 #endif 1511 1512 /* 1513 * Segment received on connection. 1514 * Reset idle time and keep-alive timer. 1515 */ 1516 tp->t_rcvtime = tcp_now; 1517 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1518 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1519 1520 /* 1521 * Process options. 1522 */ 1523 #ifdef TCP_SIGNATURE 1524 if (optp || (tp->t_flags & TF_SIGNATURE)) 1525 #else 1526 if (optp) 1527 #endif 1528 if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0) 1529 goto drop; 1530 1531 if (TCP_SACK_ENABLED(tp)) { 1532 tcp_del_sackholes(tp, th); 1533 } 1534 1535 if (opti.ts_present && opti.ts_ecr) { 1536 /* 1537 * Calculate the RTT from the returned time stamp and the 1538 * connection's time base. If the time stamp is later than 1539 * the current time, or is extremely old, fall back to non-1323 1540 * RTT calculation. Since ts_ecr is unsigned, we can test both 1541 * at the same time. 1542 */ 1543 ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1; 1544 if (ts_rtt > TCP_PAWS_IDLE) 1545 ts_rtt = 0; 1546 } else { 1547 ts_rtt = 0; 1548 } 1549 1550 /* 1551 * Header prediction: check for the two common cases 1552 * of a uni-directional data xfer. If the packet has 1553 * no control flags, is in-sequence, the window didn't 1554 * change and we're not retransmitting, it's a 1555 * candidate. If the length is zero and the ack moved 1556 * forward, we're the sender side of the xfer. Just 1557 * free the data acked & wake any higher level process 1558 * that was blocked waiting for space. If the length 1559 * is non-zero and the ack didn't move, we're the 1560 * receiver side. If we're getting packets in-order 1561 * (the reassembly queue is empty), add the data to 1562 * the socket buffer and note that we need a delayed ack. 1563 */ 1564 if (tp->t_state == TCPS_ESTABLISHED && 1565 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1566 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 1567 th->th_seq == tp->rcv_nxt && 1568 tiwin && tiwin == tp->snd_wnd && 1569 tp->snd_nxt == tp->snd_max) { 1570 1571 /* 1572 * If last ACK falls within this segment's sequence numbers, 1573 * record the timestamp. 1574 */ 1575 if (opti.ts_present && 1576 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1577 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) { 1578 tp->ts_recent_age = tcp_now; 1579 tp->ts_recent = opti.ts_val; 1580 } 1581 1582 if (tlen == 0) { 1583 /* Ack prediction. */ 1584 if (SEQ_GT(th->th_ack, tp->snd_una) && 1585 SEQ_LEQ(th->th_ack, tp->snd_max) && 1586 tp->snd_cwnd >= tp->snd_wnd && 1587 tp->t_partialacks < 0) { 1588 /* 1589 * this is a pure ack for outstanding data. 1590 */ 1591 ++tcpstat.tcps_predack; 1592 if (ts_rtt) 1593 tcp_xmit_timer(tp, ts_rtt); 1594 else if (tp->t_rtttime && 1595 SEQ_GT(th->th_ack, tp->t_rtseq)) 1596 tcp_xmit_timer(tp, 1597 tcp_now - tp->t_rtttime); 1598 acked = th->th_ack - tp->snd_una; 1599 tcpstat.tcps_rcvackpack++; 1600 tcpstat.tcps_rcvackbyte += acked; 1601 ND6_HINT(tp); 1602 1603 if (acked > (tp->t_lastoff - tp->t_inoff)) 1604 tp->t_lastm = NULL; 1605 sbdrop(&so->so_snd, acked); 1606 tp->t_lastoff -= acked; 1607 1608 tp->snd_una = th->th_ack; 1609 tp->snd_fack = tp->snd_una; 1610 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1611 tp->snd_high = tp->snd_una; 1612 m_freem(m); 1613 1614 /* 1615 * If all outstanding data are acked, stop 1616 * retransmit timer, otherwise restart timer 1617 * using current (possibly backed-off) value. 1618 * If process is waiting for space, 1619 * wakeup/selwakeup/signal. If data 1620 * are ready to send, let tcp_output 1621 * decide between more output or persist. 1622 */ 1623 if (tp->snd_una == tp->snd_max) 1624 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1625 else if (TCP_TIMER_ISARMED(tp, 1626 TCPT_PERSIST) == 0) 1627 TCP_TIMER_ARM(tp, TCPT_REXMT, 1628 tp->t_rxtcur); 1629 1630 sowwakeup(so); 1631 if (so->so_snd.sb_cc) 1632 (void) tcp_output(tp); 1633 if (tcp_saveti) 1634 m_freem(tcp_saveti); 1635 return; 1636 } 1637 } else if (th->th_ack == tp->snd_una && 1638 TAILQ_FIRST(&tp->segq) == NULL && 1639 tlen <= sbspace(&so->so_rcv)) { 1640 /* 1641 * this is a pure, in-sequence data packet 1642 * with nothing on the reassembly queue and 1643 * we have enough buffer space to take it. 1644 */ 1645 ++tcpstat.tcps_preddat; 1646 tp->rcv_nxt += tlen; 1647 tcpstat.tcps_rcvpack++; 1648 tcpstat.tcps_rcvbyte += tlen; 1649 ND6_HINT(tp); 1650 /* 1651 * Drop TCP, IP headers and TCP options then add data 1652 * to socket buffer. 1653 */ 1654 if (so->so_state & SS_CANTRCVMORE) 1655 m_freem(m); 1656 else { 1657 m_adj(m, toff + off); 1658 sbappendstream(&so->so_rcv, m); 1659 } 1660 sorwakeup(so); 1661 TCP_SETUP_ACK(tp, th); 1662 if (tp->t_flags & TF_ACKNOW) 1663 (void) tcp_output(tp); 1664 if (tcp_saveti) 1665 m_freem(tcp_saveti); 1666 return; 1667 } 1668 } 1669 1670 /* 1671 * Compute mbuf offset to TCP data segment. 1672 */ 1673 hdroptlen = toff + off; 1674 1675 /* 1676 * Calculate amount of space in receive window, 1677 * and then do TCP input processing. 1678 * Receive window is amount of space in rcv queue, 1679 * but not less than advertised window. 1680 */ 1681 { int win; 1682 1683 win = sbspace(&so->so_rcv); 1684 if (win < 0) 1685 win = 0; 1686 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1687 } 1688 1689 switch (tp->t_state) { 1690 case TCPS_LISTEN: 1691 /* 1692 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1693 */ 1694 if (m->m_flags & (M_BCAST|M_MCAST)) 1695 goto drop; 1696 switch (af) { 1697 #ifdef INET6 1698 case AF_INET6: 1699 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 1700 goto drop; 1701 break; 1702 #endif /* INET6 */ 1703 case AF_INET: 1704 if (IN_MULTICAST(ip->ip_dst.s_addr) || 1705 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1706 goto drop; 1707 break; 1708 } 1709 break; 1710 1711 /* 1712 * If the state is SYN_SENT: 1713 * if seg contains an ACK, but not for our SYN, drop the input. 1714 * if seg contains a RST, then drop the connection. 1715 * if seg does not contain SYN, then drop it. 1716 * Otherwise this is an acceptable SYN segment 1717 * initialize tp->rcv_nxt and tp->irs 1718 * if seg contains ack then advance tp->snd_una 1719 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1720 * arrange for segment to be acked (eventually) 1721 * continue processing rest of data/controls, beginning with URG 1722 */ 1723 case TCPS_SYN_SENT: 1724 if ((tiflags & TH_ACK) && 1725 (SEQ_LEQ(th->th_ack, tp->iss) || 1726 SEQ_GT(th->th_ack, tp->snd_max))) 1727 goto dropwithreset; 1728 if (tiflags & TH_RST) { 1729 if (tiflags & TH_ACK) 1730 tp = tcp_drop(tp, ECONNREFUSED); 1731 goto drop; 1732 } 1733 if ((tiflags & TH_SYN) == 0) 1734 goto drop; 1735 if (tiflags & TH_ACK) { 1736 tp->snd_una = th->th_ack; 1737 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1738 tp->snd_nxt = tp->snd_una; 1739 if (SEQ_LT(tp->snd_high, tp->snd_una)) 1740 tp->snd_high = tp->snd_una; 1741 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1742 } 1743 tp->irs = th->th_seq; 1744 tcp_rcvseqinit(tp); 1745 tp->t_flags |= TF_ACKNOW; 1746 tcp_mss_from_peer(tp, opti.maxseg); 1747 1748 /* 1749 * Initialize the initial congestion window. If we 1750 * had to retransmit the SYN, we must initialize cwnd 1751 * to 1 segment (i.e. the Loss Window). 1752 */ 1753 if (tp->t_flags & TF_SYN_REXMT) 1754 tp->snd_cwnd = tp->t_peermss; 1755 else { 1756 int ss = tcp_init_win; 1757 #ifdef INET 1758 if (inp != NULL && in_localaddr(inp->inp_faddr)) 1759 ss = tcp_init_win_local; 1760 #endif 1761 #ifdef INET6 1762 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 1763 ss = tcp_init_win_local; 1764 #endif 1765 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 1766 } 1767 1768 tcp_rmx_rtt(tp); 1769 if (tiflags & TH_ACK) { 1770 tcpstat.tcps_connects++; 1771 soisconnected(so); 1772 tcp_established(tp); 1773 /* Do window scaling on this connection? */ 1774 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1775 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1776 tp->snd_scale = tp->requested_s_scale; 1777 tp->rcv_scale = tp->request_r_scale; 1778 } 1779 TCP_REASS_LOCK(tp); 1780 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 1781 TCP_REASS_UNLOCK(tp); 1782 /* 1783 * if we didn't have to retransmit the SYN, 1784 * use its rtt as our initial srtt & rtt var. 1785 */ 1786 if (tp->t_rtttime) 1787 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1788 } else 1789 tp->t_state = TCPS_SYN_RECEIVED; 1790 1791 /* 1792 * Advance th->th_seq to correspond to first data byte. 1793 * If data, trim to stay within window, 1794 * dropping FIN if necessary. 1795 */ 1796 th->th_seq++; 1797 if (tlen > tp->rcv_wnd) { 1798 todrop = tlen - tp->rcv_wnd; 1799 m_adj(m, -todrop); 1800 tlen = tp->rcv_wnd; 1801 tiflags &= ~TH_FIN; 1802 tcpstat.tcps_rcvpackafterwin++; 1803 tcpstat.tcps_rcvbyteafterwin += todrop; 1804 } 1805 tp->snd_wl1 = th->th_seq - 1; 1806 tp->rcv_up = th->th_seq; 1807 goto step6; 1808 1809 /* 1810 * If the state is SYN_RECEIVED: 1811 * If seg contains an ACK, but not for our SYN, drop the input 1812 * and generate an RST. See page 36, rfc793 1813 */ 1814 case TCPS_SYN_RECEIVED: 1815 if ((tiflags & TH_ACK) && 1816 (SEQ_LEQ(th->th_ack, tp->iss) || 1817 SEQ_GT(th->th_ack, tp->snd_max))) 1818 goto dropwithreset; 1819 break; 1820 } 1821 1822 /* 1823 * States other than LISTEN or SYN_SENT. 1824 * First check timestamp, if present. 1825 * Then check that at least some bytes of segment are within 1826 * receive window. If segment begins before rcv_nxt, 1827 * drop leading data (and SYN); if nothing left, just ack. 1828 * 1829 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1830 * and it's less than ts_recent, drop it. 1831 */ 1832 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1833 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1834 1835 /* Check to see if ts_recent is over 24 days old. */ 1836 if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) { 1837 /* 1838 * Invalidate ts_recent. If this segment updates 1839 * ts_recent, the age will be reset later and ts_recent 1840 * will get a valid value. If it does not, setting 1841 * ts_recent to zero will at least satisfy the 1842 * requirement that zero be placed in the timestamp 1843 * echo reply when ts_recent isn't valid. The 1844 * age isn't reset until we get a valid ts_recent 1845 * because we don't want out-of-order segments to be 1846 * dropped when ts_recent is old. 1847 */ 1848 tp->ts_recent = 0; 1849 } else { 1850 tcpstat.tcps_rcvduppack++; 1851 tcpstat.tcps_rcvdupbyte += tlen; 1852 tcpstat.tcps_pawsdrop++; 1853 tcp_new_dsack(tp, th->th_seq, tlen); 1854 goto dropafterack; 1855 } 1856 } 1857 1858 todrop = tp->rcv_nxt - th->th_seq; 1859 dupseg = FALSE; 1860 if (todrop > 0) { 1861 if (tiflags & TH_SYN) { 1862 tiflags &= ~TH_SYN; 1863 th->th_seq++; 1864 if (th->th_urp > 1) 1865 th->th_urp--; 1866 else { 1867 tiflags &= ~TH_URG; 1868 th->th_urp = 0; 1869 } 1870 todrop--; 1871 } 1872 if (todrop > tlen || 1873 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1874 /* 1875 * Any valid FIN or RST must be to the left of the 1876 * window. At this point the FIN or RST must be a 1877 * duplicate or out of sequence; drop it. 1878 */ 1879 if (tiflags & TH_RST) 1880 goto drop; 1881 tiflags &= ~(TH_FIN|TH_RST); 1882 /* 1883 * Send an ACK to resynchronize and drop any data. 1884 * But keep on processing for RST or ACK. 1885 */ 1886 tp->t_flags |= TF_ACKNOW; 1887 todrop = tlen; 1888 dupseg = TRUE; 1889 tcpstat.tcps_rcvdupbyte += todrop; 1890 tcpstat.tcps_rcvduppack++; 1891 } else if ((tiflags & TH_RST) && 1892 th->th_seq != tp->last_ack_sent) { 1893 /* 1894 * Test for reset before adjusting the sequence 1895 * number for overlapping data. 1896 */ 1897 goto dropafterack_ratelim; 1898 } else { 1899 tcpstat.tcps_rcvpartduppack++; 1900 tcpstat.tcps_rcvpartdupbyte += todrop; 1901 } 1902 tcp_new_dsack(tp, th->th_seq, todrop); 1903 hdroptlen += todrop; /*drop from head afterwards*/ 1904 th->th_seq += todrop; 1905 tlen -= todrop; 1906 if (th->th_urp > todrop) 1907 th->th_urp -= todrop; 1908 else { 1909 tiflags &= ~TH_URG; 1910 th->th_urp = 0; 1911 } 1912 } 1913 1914 /* 1915 * If new data are received on a connection after the 1916 * user processes are gone, then RST the other end. 1917 */ 1918 if ((so->so_state & SS_NOFDREF) && 1919 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1920 tp = tcp_close(tp); 1921 tcpstat.tcps_rcvafterclose++; 1922 goto dropwithreset; 1923 } 1924 1925 /* 1926 * If segment ends after window, drop trailing data 1927 * (and PUSH and FIN); if nothing left, just ACK. 1928 */ 1929 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1930 if (todrop > 0) { 1931 tcpstat.tcps_rcvpackafterwin++; 1932 if (todrop >= tlen) { 1933 /* 1934 * The segment actually starts after the window. 1935 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen 1936 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0 1937 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd 1938 */ 1939 tcpstat.tcps_rcvbyteafterwin += tlen; 1940 /* 1941 * If a new connection request is received 1942 * while in TIME_WAIT, drop the old connection 1943 * and start over if the sequence numbers 1944 * are above the previous ones. 1945 * 1946 * NOTE: We will checksum the packet again, and 1947 * so we need to put the header fields back into 1948 * network order! 1949 * XXX This kind of sucks, but we don't expect 1950 * XXX this to happen very often, so maybe it 1951 * XXX doesn't matter so much. 1952 */ 1953 if (tiflags & TH_SYN && 1954 tp->t_state == TCPS_TIME_WAIT && 1955 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1956 iss = tcp_new_iss(tp, tp->snd_nxt); 1957 tp = tcp_close(tp); 1958 TCP_FIELDS_TO_NET(th); 1959 goto findpcb; 1960 } 1961 /* 1962 * If window is closed can only take segments at 1963 * window edge, and have to drop data and PUSH from 1964 * incoming segments. Continue processing, but 1965 * remember to ack. Otherwise, drop segment 1966 * and (if not RST) ack. 1967 */ 1968 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1969 tp->t_flags |= TF_ACKNOW; 1970 tcpstat.tcps_rcvwinprobe++; 1971 } else 1972 goto dropafterack; 1973 } else 1974 tcpstat.tcps_rcvbyteafterwin += todrop; 1975 m_adj(m, -todrop); 1976 tlen -= todrop; 1977 tiflags &= ~(TH_PUSH|TH_FIN); 1978 } 1979 1980 /* 1981 * If last ACK falls within this segment's sequence numbers, 1982 * and the timestamp is newer, record it. 1983 */ 1984 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1985 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1986 SEQ_LT(tp->last_ack_sent, th->th_seq + tlen + 1987 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1988 tp->ts_recent_age = tcp_now; 1989 tp->ts_recent = opti.ts_val; 1990 } 1991 1992 /* 1993 * If the RST bit is set examine the state: 1994 * SYN_RECEIVED STATE: 1995 * If passive open, return to LISTEN state. 1996 * If active open, inform user that connection was refused. 1997 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1998 * Inform user that connection was reset, and close tcb. 1999 * CLOSING, LAST_ACK, TIME_WAIT STATES 2000 * Close the tcb. 2001 */ 2002 if (tiflags & TH_RST) { 2003 if (th->th_seq != tp->last_ack_sent) 2004 goto dropafterack_ratelim; 2005 2006 switch (tp->t_state) { 2007 case TCPS_SYN_RECEIVED: 2008 so->so_error = ECONNREFUSED; 2009 goto close; 2010 2011 case TCPS_ESTABLISHED: 2012 case TCPS_FIN_WAIT_1: 2013 case TCPS_FIN_WAIT_2: 2014 case TCPS_CLOSE_WAIT: 2015 so->so_error = ECONNRESET; 2016 close: 2017 tp->t_state = TCPS_CLOSED; 2018 tcpstat.tcps_drops++; 2019 tp = tcp_close(tp); 2020 goto drop; 2021 2022 case TCPS_CLOSING: 2023 case TCPS_LAST_ACK: 2024 case TCPS_TIME_WAIT: 2025 tp = tcp_close(tp); 2026 goto drop; 2027 } 2028 } 2029 2030 /* 2031 * Since we've covered the SYN-SENT and SYN-RECEIVED states above 2032 * we must be in a synchronized state. RFC791 states (under RST 2033 * generation) that any unacceptable segment (an out-of-order SYN 2034 * qualifies) received in a synchronized state must elicit only an 2035 * empty acknowledgment segment ... and the connection remains in 2036 * the same state. 2037 */ 2038 if (tiflags & TH_SYN) { 2039 if (tp->rcv_nxt == th->th_seq) { 2040 tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1, 2041 TH_ACK); 2042 if (tcp_saveti) 2043 m_freem(tcp_saveti); 2044 return; 2045 } 2046 2047 goto dropafterack_ratelim; 2048 } 2049 2050 /* 2051 * If the ACK bit is off we drop the segment and return. 2052 */ 2053 if ((tiflags & TH_ACK) == 0) { 2054 if (tp->t_flags & TF_ACKNOW) 2055 goto dropafterack; 2056 else 2057 goto drop; 2058 } 2059 2060 /* 2061 * Ack processing. 2062 */ 2063 switch (tp->t_state) { 2064 2065 /* 2066 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 2067 * ESTABLISHED state and continue processing, otherwise 2068 * send an RST. 2069 */ 2070 case TCPS_SYN_RECEIVED: 2071 if (SEQ_GT(tp->snd_una, th->th_ack) || 2072 SEQ_GT(th->th_ack, tp->snd_max)) 2073 goto dropwithreset; 2074 tcpstat.tcps_connects++; 2075 soisconnected(so); 2076 tcp_established(tp); 2077 /* Do window scaling? */ 2078 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2079 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2080 tp->snd_scale = tp->requested_s_scale; 2081 tp->rcv_scale = tp->request_r_scale; 2082 } 2083 TCP_REASS_LOCK(tp); 2084 (void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen); 2085 TCP_REASS_UNLOCK(tp); 2086 tp->snd_wl1 = th->th_seq - 1; 2087 /* fall into ... */ 2088 2089 /* 2090 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2091 * ACKs. If the ack is in the range 2092 * tp->snd_una < th->th_ack <= tp->snd_max 2093 * then advance tp->snd_una to th->th_ack and drop 2094 * data from the retransmission queue. If this ACK reflects 2095 * more up to date window information we update our window information. 2096 */ 2097 case TCPS_ESTABLISHED: 2098 case TCPS_FIN_WAIT_1: 2099 case TCPS_FIN_WAIT_2: 2100 case TCPS_CLOSE_WAIT: 2101 case TCPS_CLOSING: 2102 case TCPS_LAST_ACK: 2103 case TCPS_TIME_WAIT: 2104 2105 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2106 if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) { 2107 tcpstat.tcps_rcvdupack++; 2108 /* 2109 * If we have outstanding data (other than 2110 * a window probe), this is a completely 2111 * duplicate ack (ie, window info didn't 2112 * change), the ack is the biggest we've 2113 * seen and we've seen exactly our rexmt 2114 * threshhold of them, assume a packet 2115 * has been dropped and retransmit it. 2116 * Kludge snd_nxt & the congestion 2117 * window so we send only this one 2118 * packet. 2119 * 2120 * We know we're losing at the current 2121 * window size so do congestion avoidance 2122 * (set ssthresh to half the current window 2123 * and pull our congestion window back to 2124 * the new ssthresh). 2125 * 2126 * Dup acks mean that packets have left the 2127 * network (they're now cached at the receiver) 2128 * so bump cwnd by the amount in the receiver 2129 * to keep a constant cwnd packets in the 2130 * network. 2131 * 2132 * If we are using TCP/SACK, then enter 2133 * Fast Recovery if the receiver SACKs 2134 * data that is tcprexmtthresh * MSS 2135 * bytes past the last ACKed segment, 2136 * irrespective of the number of DupAcks. 2137 */ 2138 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 2139 th->th_ack != tp->snd_una) 2140 tp->t_dupacks = 0; 2141 else if (tp->t_partialacks < 0 && 2142 (++tp->t_dupacks == tcprexmtthresh || 2143 TCP_FACK_FASTRECOV(tp))) { 2144 tcp_seq onxt; 2145 u_int win; 2146 2147 if (tcp_do_newreno && 2148 SEQ_LT(th->th_ack, tp->snd_high)) { 2149 /* 2150 * False fast retransmit after 2151 * timeout. Do not enter fast 2152 * recovery. 2153 */ 2154 tp->t_dupacks = 0; 2155 break; 2156 } 2157 2158 onxt = tp->snd_nxt; 2159 win = min(tp->snd_wnd, tp->snd_cwnd) / 2160 2 / tp->t_segsz; 2161 if (win < 2) 2162 win = 2; 2163 tp->snd_ssthresh = win * tp->t_segsz; 2164 tp->snd_recover = tp->snd_max; 2165 tp->t_partialacks = 0; 2166 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2167 tp->t_rtttime = 0; 2168 if (TCP_SACK_ENABLED(tp)) { 2169 tp->t_dupacks = tcprexmtthresh; 2170 tp->sack_newdata = tp->snd_nxt; 2171 tp->snd_cwnd = tp->t_segsz; 2172 (void) tcp_output(tp); 2173 goto drop; 2174 } 2175 tp->snd_nxt = th->th_ack; 2176 tp->snd_cwnd = tp->t_segsz; 2177 (void) tcp_output(tp); 2178 tp->snd_cwnd = tp->snd_ssthresh + 2179 tp->t_segsz * tp->t_dupacks; 2180 if (SEQ_GT(onxt, tp->snd_nxt)) 2181 tp->snd_nxt = onxt; 2182 goto drop; 2183 } else if (tp->t_dupacks > tcprexmtthresh) { 2184 tp->snd_cwnd += tp->t_segsz; 2185 (void) tcp_output(tp); 2186 goto drop; 2187 } 2188 } else { 2189 /* 2190 * If the ack appears to be very old, only 2191 * allow data that is in-sequence. This 2192 * makes it somewhat more difficult to insert 2193 * forged data by guessing sequence numbers. 2194 * Sent an ack to try to update the send 2195 * sequence number on the other side. 2196 */ 2197 if (tlen && th->th_seq != tp->rcv_nxt && 2198 SEQ_LT(th->th_ack, 2199 tp->snd_una - tp->max_sndwnd)) 2200 goto dropafterack; 2201 } 2202 break; 2203 } 2204 /* 2205 * If the congestion window was inflated to account 2206 * for the other side's cached packets, retract it. 2207 */ 2208 if (TCP_SACK_ENABLED(tp)) 2209 tcp_sack_newack(tp, th); 2210 else if (tcp_do_newreno) 2211 tcp_newreno_newack(tp, th); 2212 else 2213 tcp_reno_newack(tp, th); 2214 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2215 tcpstat.tcps_rcvacktoomuch++; 2216 goto dropafterack; 2217 } 2218 acked = th->th_ack - tp->snd_una; 2219 tcpstat.tcps_rcvackpack++; 2220 tcpstat.tcps_rcvackbyte += acked; 2221 2222 /* 2223 * If we have a timestamp reply, update smoothed 2224 * round trip time. If no timestamp is present but 2225 * transmit timer is running and timed sequence 2226 * number was acked, update smoothed round trip time. 2227 * Since we now have an rtt measurement, cancel the 2228 * timer backoff (cf., Phil Karn's retransmit alg.). 2229 * Recompute the initial retransmit timer. 2230 */ 2231 if (ts_rtt) 2232 tcp_xmit_timer(tp, ts_rtt); 2233 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2234 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 2235 2236 /* 2237 * If all outstanding data is acked, stop retransmit 2238 * timer and remember to restart (more output or persist). 2239 * If there is more data to be acked, restart retransmit 2240 * timer, using current (possibly backed-off) value. 2241 */ 2242 if (th->th_ack == tp->snd_max) { 2243 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2244 needoutput = 1; 2245 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 2246 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2247 /* 2248 * When new data is acked, open the congestion window. 2249 * If the window gives us less than ssthresh packets 2250 * in flight, open exponentially (segsz per packet). 2251 * Otherwise open linearly: segsz per window 2252 * (segsz^2 / cwnd per packet). 2253 * 2254 * If we are still in fast recovery (meaning we are using 2255 * NewReno and we have only received partial acks), do not 2256 * inflate the window yet. 2257 */ 2258 if (tp->t_partialacks < 0) { 2259 u_int cw = tp->snd_cwnd; 2260 u_int incr = tp->t_segsz; 2261 2262 if (cw >= tp->snd_ssthresh) 2263 incr = incr * incr / cw; 2264 tp->snd_cwnd = min(cw + incr, 2265 TCP_MAXWIN << tp->snd_scale); 2266 } 2267 ND6_HINT(tp); 2268 if (acked > so->so_snd.sb_cc) { 2269 tp->snd_wnd -= so->so_snd.sb_cc; 2270 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2271 ourfinisacked = 1; 2272 } else { 2273 if (acked > (tp->t_lastoff - tp->t_inoff)) 2274 tp->t_lastm = NULL; 2275 sbdrop(&so->so_snd, acked); 2276 tp->t_lastoff -= acked; 2277 tp->snd_wnd -= acked; 2278 ourfinisacked = 0; 2279 } 2280 sowwakeup(so); 2281 tp->snd_una = th->th_ack; 2282 if (SEQ_GT(tp->snd_una, tp->snd_fack)) 2283 tp->snd_fack = tp->snd_una; 2284 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2285 tp->snd_nxt = tp->snd_una; 2286 if (SEQ_LT(tp->snd_high, tp->snd_una)) 2287 tp->snd_high = tp->snd_una; 2288 2289 switch (tp->t_state) { 2290 2291 /* 2292 * In FIN_WAIT_1 STATE in addition to the processing 2293 * for the ESTABLISHED state if our FIN is now acknowledged 2294 * then enter FIN_WAIT_2. 2295 */ 2296 case TCPS_FIN_WAIT_1: 2297 if (ourfinisacked) { 2298 /* 2299 * If we can't receive any more 2300 * data, then closing user can proceed. 2301 * Starting the timer is contrary to the 2302 * specification, but if we don't get a FIN 2303 * we'll hang forever. 2304 */ 2305 if (so->so_state & SS_CANTRCVMORE) { 2306 soisdisconnected(so); 2307 if (tcp_maxidle > 0) 2308 TCP_TIMER_ARM(tp, TCPT_2MSL, 2309 tcp_maxidle); 2310 } 2311 tp->t_state = TCPS_FIN_WAIT_2; 2312 } 2313 break; 2314 2315 /* 2316 * In CLOSING STATE in addition to the processing for 2317 * the ESTABLISHED state if the ACK acknowledges our FIN 2318 * then enter the TIME-WAIT state, otherwise ignore 2319 * the segment. 2320 */ 2321 case TCPS_CLOSING: 2322 if (ourfinisacked) { 2323 tp->t_state = TCPS_TIME_WAIT; 2324 tcp_canceltimers(tp); 2325 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2326 soisdisconnected(so); 2327 } 2328 break; 2329 2330 /* 2331 * In LAST_ACK, we may still be waiting for data to drain 2332 * and/or to be acked, as well as for the ack of our FIN. 2333 * If our FIN is now acknowledged, delete the TCB, 2334 * enter the closed state and return. 2335 */ 2336 case TCPS_LAST_ACK: 2337 if (ourfinisacked) { 2338 tp = tcp_close(tp); 2339 goto drop; 2340 } 2341 break; 2342 2343 /* 2344 * In TIME_WAIT state the only thing that should arrive 2345 * is a retransmission of the remote FIN. Acknowledge 2346 * it and restart the finack timer. 2347 */ 2348 case TCPS_TIME_WAIT: 2349 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2350 goto dropafterack; 2351 } 2352 } 2353 2354 step6: 2355 /* 2356 * Update window information. 2357 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2358 */ 2359 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 2360 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 2361 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 2362 /* keep track of pure window updates */ 2363 if (tlen == 0 && 2364 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2365 tcpstat.tcps_rcvwinupd++; 2366 tp->snd_wnd = tiwin; 2367 tp->snd_wl1 = th->th_seq; 2368 tp->snd_wl2 = th->th_ack; 2369 if (tp->snd_wnd > tp->max_sndwnd) 2370 tp->max_sndwnd = tp->snd_wnd; 2371 needoutput = 1; 2372 } 2373 2374 /* 2375 * Process segments with URG. 2376 */ 2377 if ((tiflags & TH_URG) && th->th_urp && 2378 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2379 /* 2380 * This is a kludge, but if we receive and accept 2381 * random urgent pointers, we'll crash in 2382 * soreceive. It's hard to imagine someone 2383 * actually wanting to send this much urgent data. 2384 */ 2385 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2386 th->th_urp = 0; /* XXX */ 2387 tiflags &= ~TH_URG; /* XXX */ 2388 goto dodata; /* XXX */ 2389 } 2390 /* 2391 * If this segment advances the known urgent pointer, 2392 * then mark the data stream. This should not happen 2393 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2394 * a FIN has been received from the remote side. 2395 * In these states we ignore the URG. 2396 * 2397 * According to RFC961 (Assigned Protocols), 2398 * the urgent pointer points to the last octet 2399 * of urgent data. We continue, however, 2400 * to consider it to indicate the first octet 2401 * of data past the urgent section as the original 2402 * spec states (in one of two places). 2403 */ 2404 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2405 tp->rcv_up = th->th_seq + th->th_urp; 2406 so->so_oobmark = so->so_rcv.sb_cc + 2407 (tp->rcv_up - tp->rcv_nxt) - 1; 2408 if (so->so_oobmark == 0) 2409 so->so_state |= SS_RCVATMARK; 2410 sohasoutofband(so); 2411 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2412 } 2413 /* 2414 * Remove out of band data so doesn't get presented to user. 2415 * This can happen independent of advancing the URG pointer, 2416 * but if two URG's are pending at once, some out-of-band 2417 * data may creep in... ick. 2418 */ 2419 if (th->th_urp <= (u_int16_t) tlen 2420 #ifdef SO_OOBINLINE 2421 && (so->so_options & SO_OOBINLINE) == 0 2422 #endif 2423 ) 2424 tcp_pulloutofband(so, th, m, hdroptlen); 2425 } else 2426 /* 2427 * If no out of band data is expected, 2428 * pull receive urgent pointer along 2429 * with the receive window. 2430 */ 2431 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2432 tp->rcv_up = tp->rcv_nxt; 2433 dodata: /* XXX */ 2434 2435 /* 2436 * Process the segment text, merging it into the TCP sequencing queue, 2437 * and arranging for acknowledgement of receipt if necessary. 2438 * This process logically involves adjusting tp->rcv_wnd as data 2439 * is presented to the user (this happens in tcp_usrreq.c, 2440 * case PRU_RCVD). If a FIN has already been received on this 2441 * connection then we just ignore the text. 2442 */ 2443 if ((tlen || (tiflags & TH_FIN)) && 2444 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2445 /* 2446 * Insert segment ti into reassembly queue of tcp with 2447 * control block tp. Return TH_FIN if reassembly now includes 2448 * a segment with FIN. The macro form does the common case 2449 * inline (segment is the next to be received on an 2450 * established connection, and the queue is empty), 2451 * avoiding linkage into and removal from the queue and 2452 * repetition of various conversions. 2453 * Set DELACK for segments received in order, but ack 2454 * immediately when segments are out of order 2455 * (so fast retransmit can work). 2456 */ 2457 /* NOTE: this was TCP_REASS() macro, but used only once */ 2458 TCP_REASS_LOCK(tp); 2459 if (th->th_seq == tp->rcv_nxt && 2460 TAILQ_FIRST(&tp->segq) == NULL && 2461 tp->t_state == TCPS_ESTABLISHED) { 2462 TCP_SETUP_ACK(tp, th); 2463 tp->rcv_nxt += tlen; 2464 tiflags = th->th_flags & TH_FIN; 2465 tcpstat.tcps_rcvpack++; 2466 tcpstat.tcps_rcvbyte += tlen; 2467 ND6_HINT(tp); 2468 if (so->so_state & SS_CANTRCVMORE) 2469 m_freem(m); 2470 else { 2471 m_adj(m, hdroptlen); 2472 sbappendstream(&(so)->so_rcv, m); 2473 } 2474 sorwakeup(so); 2475 } else { 2476 m_adj(m, hdroptlen); 2477 tiflags = tcp_reass(tp, th, m, &tlen); 2478 tp->t_flags |= TF_ACKNOW; 2479 } 2480 TCP_REASS_UNLOCK(tp); 2481 2482 /* 2483 * Note the amount of data that peer has sent into 2484 * our window, in order to estimate the sender's 2485 * buffer size. 2486 */ 2487 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2488 } else { 2489 m_freem(m); 2490 m = NULL; 2491 tiflags &= ~TH_FIN; 2492 } 2493 2494 /* 2495 * If FIN is received ACK the FIN and let the user know 2496 * that the connection is closing. Ignore a FIN received before 2497 * the connection is fully established. 2498 */ 2499 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2500 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2501 socantrcvmore(so); 2502 tp->t_flags |= TF_ACKNOW; 2503 tp->rcv_nxt++; 2504 } 2505 switch (tp->t_state) { 2506 2507 /* 2508 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2509 */ 2510 case TCPS_ESTABLISHED: 2511 tp->t_state = TCPS_CLOSE_WAIT; 2512 break; 2513 2514 /* 2515 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2516 * enter the CLOSING state. 2517 */ 2518 case TCPS_FIN_WAIT_1: 2519 tp->t_state = TCPS_CLOSING; 2520 break; 2521 2522 /* 2523 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2524 * starting the time-wait timer, turning off the other 2525 * standard timers. 2526 */ 2527 case TCPS_FIN_WAIT_2: 2528 tp->t_state = TCPS_TIME_WAIT; 2529 tcp_canceltimers(tp); 2530 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2531 soisdisconnected(so); 2532 break; 2533 2534 /* 2535 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2536 */ 2537 case TCPS_TIME_WAIT: 2538 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2539 break; 2540 } 2541 } 2542 #ifdef TCP_DEBUG 2543 if (so->so_options & SO_DEBUG) 2544 tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0); 2545 #endif 2546 2547 /* 2548 * Return any desired output. 2549 */ 2550 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2551 (void) tcp_output(tp); 2552 } 2553 if (tcp_saveti) 2554 m_freem(tcp_saveti); 2555 return; 2556 2557 badsyn: 2558 /* 2559 * Received a bad SYN. Increment counters and dropwithreset. 2560 */ 2561 tcpstat.tcps_badsyn++; 2562 tp = NULL; 2563 goto dropwithreset; 2564 2565 dropafterack: 2566 /* 2567 * Generate an ACK dropping incoming segment if it occupies 2568 * sequence space, where the ACK reflects our state. 2569 */ 2570 if (tiflags & TH_RST) 2571 goto drop; 2572 goto dropafterack2; 2573 2574 dropafterack_ratelim: 2575 /* 2576 * We may want to rate-limit ACKs against SYN/RST attack. 2577 */ 2578 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2579 tcp_ackdrop_ppslim) == 0) { 2580 /* XXX stat */ 2581 goto drop; 2582 } 2583 /* ...fall into dropafterack2... */ 2584 2585 dropafterack2: 2586 m_freem(m); 2587 tp->t_flags |= TF_ACKNOW; 2588 (void) tcp_output(tp); 2589 if (tcp_saveti) 2590 m_freem(tcp_saveti); 2591 return; 2592 2593 dropwithreset_ratelim: 2594 /* 2595 * We may want to rate-limit RSTs in certain situations, 2596 * particularly if we are sending an RST in response to 2597 * an attempt to connect to or otherwise communicate with 2598 * a port for which we have no socket. 2599 */ 2600 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2601 tcp_rst_ppslim) == 0) { 2602 /* XXX stat */ 2603 goto drop; 2604 } 2605 /* ...fall into dropwithreset... */ 2606 2607 dropwithreset: 2608 /* 2609 * Generate a RST, dropping incoming segment. 2610 * Make ACK acceptable to originator of segment. 2611 */ 2612 if (tiflags & TH_RST) 2613 goto drop; 2614 2615 switch (af) { 2616 #ifdef INET6 2617 case AF_INET6: 2618 /* For following calls to tcp_respond */ 2619 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) 2620 goto drop; 2621 break; 2622 #endif /* INET6 */ 2623 case AF_INET: 2624 if (IN_MULTICAST(ip->ip_dst.s_addr) || 2625 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2626 goto drop; 2627 } 2628 2629 if (tiflags & TH_ACK) 2630 (void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 2631 else { 2632 if (tiflags & TH_SYN) 2633 tlen++; 2634 (void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0, 2635 TH_RST|TH_ACK); 2636 } 2637 if (tcp_saveti) 2638 m_freem(tcp_saveti); 2639 return; 2640 2641 badcsum: 2642 drop: 2643 /* 2644 * Drop space held by incoming segment and return. 2645 */ 2646 if (tp) { 2647 if (tp->t_inpcb) 2648 so = tp->t_inpcb->inp_socket; 2649 #ifdef INET6 2650 else if (tp->t_in6pcb) 2651 so = tp->t_in6pcb->in6p_socket; 2652 #endif 2653 else 2654 so = NULL; 2655 #ifdef TCP_DEBUG 2656 if (so && (so->so_options & SO_DEBUG) != 0) 2657 tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0); 2658 #endif 2659 } 2660 if (tcp_saveti) 2661 m_freem(tcp_saveti); 2662 m_freem(m); 2663 return; 2664 } 2665 2666 #ifdef TCP_SIGNATURE 2667 int 2668 tcp_signature_apply(void *fstate, caddr_t data, u_int len) 2669 { 2670 2671 MD5Update(fstate, (u_char *)data, len); 2672 return (0); 2673 } 2674 2675 struct secasvar * 2676 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th) 2677 { 2678 struct secasvar *sav; 2679 #ifdef FAST_IPSEC 2680 union sockaddr_union dst; 2681 #endif 2682 struct ip *ip; 2683 struct ip6_hdr *ip6; 2684 2685 ip = mtod(m, struct ip *); 2686 switch (ip->ip_v) { 2687 case 4: 2688 ip = mtod(m, struct ip *); 2689 ip6 = NULL; 2690 break; 2691 case 6: 2692 ip = NULL; 2693 ip6 = mtod(m, struct ip6_hdr *); 2694 break; 2695 default: 2696 return (NULL); 2697 } 2698 2699 #ifdef FAST_IPSEC 2700 /* Extract the destination from the IP header in the mbuf. */ 2701 bzero(&dst, sizeof(union sockaddr_union)); 2702 dst.sa.sa_len = sizeof(struct sockaddr_in); 2703 dst.sa.sa_family = AF_INET; 2704 dst.sin.sin_addr = ip->ip_dst; 2705 2706 /* 2707 * Look up an SADB entry which matches the address of the peer. 2708 */ 2709 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2710 #else 2711 if (ip) 2712 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, 2713 (caddr_t)&ip->ip_dst, IPPROTO_TCP, 2714 htonl(TCP_SIG_SPI), 0, 0); 2715 else 2716 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, 2717 (caddr_t)&ip6->ip6_dst, IPPROTO_TCP, 2718 htonl(TCP_SIG_SPI), 0, 0); 2719 #endif 2720 2721 return (sav); /* freesav must be performed by caller */ 2722 } 2723 2724 int 2725 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff, 2726 struct secasvar *sav, char *sig) 2727 { 2728 MD5_CTX ctx; 2729 struct ip *ip; 2730 struct ipovly *ipovly; 2731 struct ip6_hdr *ip6; 2732 struct ippseudo ippseudo; 2733 struct ip6_hdr_pseudo ip6pseudo; 2734 struct tcphdr th0; 2735 int l, tcphdrlen; 2736 2737 if (sav == NULL) 2738 return (-1); 2739 2740 tcphdrlen = th->th_off * 4; 2741 2742 switch (mtod(m, struct ip *)->ip_v) { 2743 case 4: 2744 ip = mtod(m, struct ip *); 2745 ip6 = NULL; 2746 break; 2747 case 6: 2748 ip = NULL; 2749 ip6 = mtod(m, struct ip6_hdr *); 2750 break; 2751 default: 2752 return (-1); 2753 } 2754 2755 MD5Init(&ctx); 2756 2757 if (ip) { 2758 memset(&ippseudo, 0, sizeof(ippseudo)); 2759 ipovly = (struct ipovly *)ip; 2760 ippseudo.ippseudo_src = ipovly->ih_src; 2761 ippseudo.ippseudo_dst = ipovly->ih_dst; 2762 ippseudo.ippseudo_pad = 0; 2763 ippseudo.ippseudo_p = IPPROTO_TCP; 2764 ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff); 2765 MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo)); 2766 } else { 2767 memset(&ip6pseudo, 0, sizeof(ip6pseudo)); 2768 ip6pseudo.ip6ph_src = ip6->ip6_src; 2769 in6_clearscope(&ip6pseudo.ip6ph_src); 2770 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 2771 in6_clearscope(&ip6pseudo.ip6ph_dst); 2772 ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff); 2773 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2774 MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo)); 2775 } 2776 2777 th0 = *th; 2778 th0.th_sum = 0; 2779 MD5Update(&ctx, (char *)&th0, sizeof(th0)); 2780 2781 l = m->m_pkthdr.len - thoff - tcphdrlen; 2782 if (l > 0) 2783 m_apply(m, thoff + tcphdrlen, 2784 m->m_pkthdr.len - thoff - tcphdrlen, 2785 tcp_signature_apply, &ctx); 2786 2787 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2788 MD5Final(sig, &ctx); 2789 2790 return (0); 2791 } 2792 #endif 2793 2794 int 2795 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, 2796 struct mbuf *m, int toff, struct tcp_opt_info *oi) 2797 { 2798 u_int16_t mss; 2799 int opt, optlen = 0; 2800 #ifdef TCP_SIGNATURE 2801 caddr_t sigp = NULL; 2802 char sigbuf[TCP_SIGLEN]; 2803 struct secasvar *sav = NULL; 2804 #endif 2805 2806 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2807 opt = cp[0]; 2808 if (opt == TCPOPT_EOL) 2809 break; 2810 if (opt == TCPOPT_NOP) 2811 optlen = 1; 2812 else { 2813 if (cnt < 2) 2814 break; 2815 optlen = cp[1]; 2816 if (optlen < 2 || optlen > cnt) 2817 break; 2818 } 2819 switch (opt) { 2820 2821 default: 2822 continue; 2823 2824 case TCPOPT_MAXSEG: 2825 if (optlen != TCPOLEN_MAXSEG) 2826 continue; 2827 if (!(th->th_flags & TH_SYN)) 2828 continue; 2829 bcopy(cp + 2, &mss, sizeof(mss)); 2830 oi->maxseg = ntohs(mss); 2831 break; 2832 2833 case TCPOPT_WINDOW: 2834 if (optlen != TCPOLEN_WINDOW) 2835 continue; 2836 if (!(th->th_flags & TH_SYN)) 2837 continue; 2838 tp->t_flags |= TF_RCVD_SCALE; 2839 tp->requested_s_scale = cp[2]; 2840 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 2841 #if 0 /*XXX*/ 2842 char *p; 2843 2844 if (ip) 2845 p = ntohl(ip->ip_src); 2846 #ifdef INET6 2847 else if (ip6) 2848 p = ip6_sprintf(&ip6->ip6_src); 2849 #endif 2850 else 2851 p = "(unknown)"; 2852 log(LOG_ERR, "TCP: invalid wscale %d from %s, " 2853 "assuming %d\n", 2854 tp->requested_s_scale, p, 2855 TCP_MAX_WINSHIFT); 2856 #else 2857 log(LOG_ERR, "TCP: invalid wscale %d, " 2858 "assuming %d\n", 2859 tp->requested_s_scale, 2860 TCP_MAX_WINSHIFT); 2861 #endif 2862 tp->requested_s_scale = TCP_MAX_WINSHIFT; 2863 } 2864 break; 2865 2866 case TCPOPT_TIMESTAMP: 2867 if (optlen != TCPOLEN_TIMESTAMP) 2868 continue; 2869 oi->ts_present = 1; 2870 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2871 NTOHL(oi->ts_val); 2872 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2873 NTOHL(oi->ts_ecr); 2874 2875 /* 2876 * A timestamp received in a SYN makes 2877 * it ok to send timestamp requests and replies. 2878 */ 2879 if (th->th_flags & TH_SYN) { 2880 tp->t_flags |= TF_RCVD_TSTMP; 2881 tp->ts_recent = oi->ts_val; 2882 tp->ts_recent_age = tcp_now; 2883 } 2884 break; 2885 case TCPOPT_SACK_PERMITTED: 2886 if (optlen != TCPOLEN_SACK_PERMITTED) 2887 continue; 2888 if (!(th->th_flags & TH_SYN)) 2889 continue; 2890 if (tcp_do_sack) { 2891 tp->t_flags |= TF_SACK_PERMIT; 2892 tp->t_flags |= TF_WILL_SACK; 2893 } 2894 break; 2895 2896 case TCPOPT_SACK: 2897 tcp_sack_option(tp, th, cp, optlen); 2898 break; 2899 #ifdef TCP_SIGNATURE 2900 case TCPOPT_SIGNATURE: 2901 if (optlen != TCPOLEN_SIGNATURE) 2902 continue; 2903 if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN)) 2904 return (-1); 2905 2906 sigp = sigbuf; 2907 memcpy(sigbuf, cp + 2, TCP_SIGLEN); 2908 memset(cp + 2, 0, TCP_SIGLEN); 2909 tp->t_flags |= TF_SIGNATURE; 2910 break; 2911 #endif 2912 } 2913 } 2914 2915 #ifdef TCP_SIGNATURE 2916 if (tp->t_flags & TF_SIGNATURE) { 2917 2918 sav = tcp_signature_getsav(m, th); 2919 2920 if (sav == NULL && tp->t_state == TCPS_LISTEN) 2921 return (-1); 2922 } 2923 2924 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2925 if (sav == NULL) 2926 return (-1); 2927 #ifdef FAST_IPSEC 2928 KEY_FREESAV(&sav); 2929 #else 2930 key_freesav(sav); 2931 #endif 2932 return (-1); 2933 } 2934 2935 if (sigp) { 2936 char sig[TCP_SIGLEN]; 2937 2938 TCP_FIELDS_TO_NET(th); 2939 if (tcp_signature(m, th, toff, sav, sig) < 0) { 2940 TCP_FIELDS_TO_HOST(th); 2941 if (sav == NULL) 2942 return (-1); 2943 #ifdef FAST_IPSEC 2944 KEY_FREESAV(&sav); 2945 #else 2946 key_freesav(sav); 2947 #endif 2948 return (-1); 2949 } 2950 TCP_FIELDS_TO_HOST(th); 2951 2952 if (bcmp(sig, sigp, TCP_SIGLEN)) { 2953 tcpstat.tcps_badsig++; 2954 if (sav == NULL) 2955 return (-1); 2956 #ifdef FAST_IPSEC 2957 KEY_FREESAV(&sav); 2958 #else 2959 key_freesav(sav); 2960 #endif 2961 return (-1); 2962 } else 2963 tcpstat.tcps_goodsig++; 2964 2965 key_sa_recordxfer(sav, m); 2966 #ifdef FAST_IPSEC 2967 KEY_FREESAV(&sav); 2968 #else 2969 key_freesav(sav); 2970 #endif 2971 } 2972 #endif 2973 2974 return (0); 2975 } 2976 2977 /* 2978 * Pull out of band byte out of a segment so 2979 * it doesn't appear in the user's data queue. 2980 * It is still reflected in the segment length for 2981 * sequencing purposes. 2982 */ 2983 void 2984 tcp_pulloutofband(struct socket *so, struct tcphdr *th, 2985 struct mbuf *m, int off) 2986 { 2987 int cnt = off + th->th_urp - 1; 2988 2989 while (cnt >= 0) { 2990 if (m->m_len > cnt) { 2991 char *cp = mtod(m, caddr_t) + cnt; 2992 struct tcpcb *tp = sototcpcb(so); 2993 2994 tp->t_iobc = *cp; 2995 tp->t_oobflags |= TCPOOB_HAVEDATA; 2996 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2997 m->m_len--; 2998 return; 2999 } 3000 cnt -= m->m_len; 3001 m = m->m_next; 3002 if (m == 0) 3003 break; 3004 } 3005 panic("tcp_pulloutofband"); 3006 } 3007 3008 /* 3009 * Collect new round-trip time estimate 3010 * and update averages and current timeout. 3011 */ 3012 void 3013 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt) 3014 { 3015 int32_t delta; 3016 3017 tcpstat.tcps_rttupdated++; 3018 if (tp->t_srtt != 0) { 3019 /* 3020 * srtt is stored as fixed point with 3 bits after the 3021 * binary point (i.e., scaled by 8). The following magic 3022 * is equivalent to the smoothing algorithm in rfc793 with 3023 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3024 * point). Adjust rtt to origin 0. 3025 */ 3026 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 3027 if ((tp->t_srtt += delta) <= 0) 3028 tp->t_srtt = 1 << 2; 3029 /* 3030 * We accumulate a smoothed rtt variance (actually, a 3031 * smoothed mean difference), then set the retransmit 3032 * timer to smoothed rtt + 4 times the smoothed variance. 3033 * rttvar is stored as fixed point with 2 bits after the 3034 * binary point (scaled by 4). The following is 3035 * equivalent to rfc793 smoothing with an alpha of .75 3036 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3037 * rfc793's wired-in beta. 3038 */ 3039 if (delta < 0) 3040 delta = -delta; 3041 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 3042 if ((tp->t_rttvar += delta) <= 0) 3043 tp->t_rttvar = 1 << 2; 3044 } else { 3045 /* 3046 * No rtt measurement yet - use the unsmoothed rtt. 3047 * Set the variance to half the rtt (so our first 3048 * retransmit happens at 3*rtt). 3049 */ 3050 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 3051 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 3052 } 3053 tp->t_rtttime = 0; 3054 tp->t_rxtshift = 0; 3055 3056 /* 3057 * the retransmit should happen at rtt + 4 * rttvar. 3058 * Because of the way we do the smoothing, srtt and rttvar 3059 * will each average +1/2 tick of bias. When we compute 3060 * the retransmit timer, we want 1/2 tick of rounding and 3061 * 1 extra tick because of +-1/2 tick uncertainty in the 3062 * firing of the timer. The bias will give us exactly the 3063 * 1.5 tick we need. But, because the bias is 3064 * statistical, we have to test that we don't drop below 3065 * the minimum feasible timer (which is 2 ticks). 3066 */ 3067 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3068 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3069 3070 /* 3071 * We received an ack for a packet that wasn't retransmitted; 3072 * it is probably safe to discard any error indications we've 3073 * received recently. This isn't quite right, but close enough 3074 * for now (a route might have failed after we sent a segment, 3075 * and the return path might not be symmetrical). 3076 */ 3077 tp->t_softerror = 0; 3078 } 3079 3080 void 3081 tcp_reno_newack(struct tcpcb *tp, struct tcphdr *th) 3082 { 3083 if (tp->t_partialacks < 0) { 3084 /* 3085 * We were not in fast recovery. Reset the duplicate ack 3086 * counter. 3087 */ 3088 tp->t_dupacks = 0; 3089 } else { 3090 /* 3091 * Clamp the congestion window to the crossover point and 3092 * exit fast recovery. 3093 */ 3094 if (tp->snd_cwnd > tp->snd_ssthresh) 3095 tp->snd_cwnd = tp->snd_ssthresh; 3096 tp->t_partialacks = -1; 3097 tp->t_dupacks = 0; 3098 } 3099 } 3100 3101 /* 3102 * Implement the NewReno response to a new ack, checking for partial acks in 3103 * fast recovery. 3104 */ 3105 void 3106 tcp_newreno_newack(struct tcpcb *tp, struct tcphdr *th) 3107 { 3108 if (tp->t_partialacks < 0) { 3109 /* 3110 * We were not in fast recovery. Reset the duplicate ack 3111 * counter. 3112 */ 3113 tp->t_dupacks = 0; 3114 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 3115 /* 3116 * This is a partial ack. Retransmit the first unacknowledged 3117 * segment and deflate the congestion window by the amount of 3118 * acknowledged data. Do not exit fast recovery. 3119 */ 3120 tcp_seq onxt = tp->snd_nxt; 3121 u_long ocwnd = tp->snd_cwnd; 3122 3123 /* 3124 * snd_una has not yet been updated and the socket's send 3125 * buffer has not yet drained off the ACK'd data, so we 3126 * have to leave snd_una as it was to get the correct data 3127 * offset in tcp_output(). 3128 */ 3129 if (++tp->t_partialacks == 1) 3130 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3131 tp->t_rtttime = 0; 3132 tp->snd_nxt = th->th_ack; 3133 /* 3134 * Set snd_cwnd to one segment beyond ACK'd offset. snd_una 3135 * is not yet updated when we're called. 3136 */ 3137 tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una); 3138 (void) tcp_output(tp); 3139 tp->snd_cwnd = ocwnd; 3140 if (SEQ_GT(onxt, tp->snd_nxt)) 3141 tp->snd_nxt = onxt; 3142 /* 3143 * Partial window deflation. Relies on fact that tp->snd_una 3144 * not updated yet. 3145 */ 3146 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz); 3147 } else { 3148 /* 3149 * Complete ack. Inflate the congestion window to ssthresh 3150 * and exit fast recovery. 3151 * 3152 * Window inflation should have left us with approx. 3153 * snd_ssthresh outstanding data. But in case we 3154 * would be inclined to send a burst, better to do 3155 * it via the slow start mechanism. 3156 */ 3157 if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh) 3158 tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack) 3159 + tp->t_segsz; 3160 else 3161 tp->snd_cwnd = tp->snd_ssthresh; 3162 tp->t_partialacks = -1; 3163 tp->t_dupacks = 0; 3164 } 3165 } 3166 3167 3168 /* 3169 * TCP compressed state engine. Currently used to hold compressed 3170 * state for SYN_RECEIVED. 3171 */ 3172 3173 u_long syn_cache_count; 3174 u_int32_t syn_hash1, syn_hash2; 3175 3176 #define SYN_HASH(sa, sp, dp) \ 3177 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3178 ((u_int32_t)(sp)))^syn_hash2))) 3179 #ifndef INET6 3180 #define SYN_HASHALL(hash, src, dst) \ 3181 do { \ 3182 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3183 ((const struct sockaddr_in *)(src))->sin_port, \ 3184 ((const struct sockaddr_in *)(dst))->sin_port); \ 3185 } while (/*CONSTCOND*/ 0) 3186 #else 3187 #define SYN_HASH6(sa, sp, dp) \ 3188 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3189 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3190 & 0x7fffffff) 3191 3192 #define SYN_HASHALL(hash, src, dst) \ 3193 do { \ 3194 switch ((src)->sa_family) { \ 3195 case AF_INET: \ 3196 hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \ 3197 ((const struct sockaddr_in *)(src))->sin_port, \ 3198 ((const struct sockaddr_in *)(dst))->sin_port); \ 3199 break; \ 3200 case AF_INET6: \ 3201 hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \ 3202 ((const struct sockaddr_in6 *)(src))->sin6_port, \ 3203 ((const struct sockaddr_in6 *)(dst))->sin6_port); \ 3204 break; \ 3205 default: \ 3206 hash = 0; \ 3207 } \ 3208 } while (/*CONSTCOND*/0) 3209 #endif /* INET6 */ 3210 3211 #define SYN_CACHE_RM(sc) \ 3212 do { \ 3213 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3214 (sc), sc_bucketq); \ 3215 (sc)->sc_tp = NULL; \ 3216 LIST_REMOVE((sc), sc_tpq); \ 3217 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3218 callout_stop(&(sc)->sc_timer); \ 3219 syn_cache_count--; \ 3220 } while (/*CONSTCOND*/0) 3221 3222 #define SYN_CACHE_PUT(sc) \ 3223 do { \ 3224 if ((sc)->sc_ipopts) \ 3225 (void) m_free((sc)->sc_ipopts); \ 3226 if ((sc)->sc_route4.ro_rt != NULL) \ 3227 RTFREE((sc)->sc_route4.ro_rt); \ 3228 if (callout_invoking(&(sc)->sc_timer)) \ 3229 (sc)->sc_flags |= SCF_DEAD; \ 3230 else \ 3231 pool_put(&syn_cache_pool, (sc)); \ 3232 } while (/*CONSTCOND*/0) 3233 3234 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL); 3235 3236 /* 3237 * We don't estimate RTT with SYNs, so each packet starts with the default 3238 * RTT and each timer step has a fixed timeout value. 3239 */ 3240 #define SYN_CACHE_TIMER_ARM(sc) \ 3241 do { \ 3242 TCPT_RANGESET((sc)->sc_rxtcur, \ 3243 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3244 TCPTV_REXMTMAX); \ 3245 callout_reset(&(sc)->sc_timer, \ 3246 (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc)); \ 3247 } while (/*CONSTCOND*/0) 3248 3249 #define SYN_CACHE_TIMESTAMP(sc) (tcp_now - (sc)->sc_timebase) 3250 3251 void 3252 syn_cache_init(void) 3253 { 3254 int i; 3255 3256 /* Initialize the hash buckets. */ 3257 for (i = 0; i < tcp_syn_cache_size; i++) 3258 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3259 } 3260 3261 void 3262 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3263 { 3264 struct syn_cache_head *scp; 3265 struct syn_cache *sc2; 3266 int s; 3267 3268 /* 3269 * If there are no entries in the hash table, reinitialize 3270 * the hash secrets. 3271 */ 3272 if (syn_cache_count == 0) { 3273 syn_hash1 = arc4random(); 3274 syn_hash2 = arc4random(); 3275 } 3276 3277 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3278 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3279 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3280 3281 /* 3282 * Make sure that we don't overflow the per-bucket 3283 * limit or the total cache size limit. 3284 */ 3285 s = splsoftnet(); 3286 if (scp->sch_length >= tcp_syn_bucket_limit) { 3287 tcpstat.tcps_sc_bucketoverflow++; 3288 /* 3289 * The bucket is full. Toss the oldest element in the 3290 * bucket. This will be the first entry in the bucket. 3291 */ 3292 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3293 #ifdef DIAGNOSTIC 3294 /* 3295 * This should never happen; we should always find an 3296 * entry in our bucket. 3297 */ 3298 if (sc2 == NULL) 3299 panic("syn_cache_insert: bucketoverflow: impossible"); 3300 #endif 3301 SYN_CACHE_RM(sc2); 3302 SYN_CACHE_PUT(sc2); 3303 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3304 struct syn_cache_head *scp2, *sce; 3305 3306 tcpstat.tcps_sc_overflowed++; 3307 /* 3308 * The cache is full. Toss the oldest entry in the 3309 * first non-empty bucket we can find. 3310 * 3311 * XXX We would really like to toss the oldest 3312 * entry in the cache, but we hope that this 3313 * condition doesn't happen very often. 3314 */ 3315 scp2 = scp; 3316 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3317 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3318 for (++scp2; scp2 != scp; scp2++) { 3319 if (scp2 >= sce) 3320 scp2 = &tcp_syn_cache[0]; 3321 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3322 break; 3323 } 3324 #ifdef DIAGNOSTIC 3325 /* 3326 * This should never happen; we should always find a 3327 * non-empty bucket. 3328 */ 3329 if (scp2 == scp) 3330 panic("syn_cache_insert: cacheoverflow: " 3331 "impossible"); 3332 #endif 3333 } 3334 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3335 SYN_CACHE_RM(sc2); 3336 SYN_CACHE_PUT(sc2); 3337 } 3338 3339 /* 3340 * Initialize the entry's timer. 3341 */ 3342 sc->sc_rxttot = 0; 3343 sc->sc_rxtshift = 0; 3344 SYN_CACHE_TIMER_ARM(sc); 3345 3346 /* Link it from tcpcb entry */ 3347 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3348 3349 /* Put it into the bucket. */ 3350 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3351 scp->sch_length++; 3352 syn_cache_count++; 3353 3354 tcpstat.tcps_sc_added++; 3355 splx(s); 3356 } 3357 3358 /* 3359 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3360 * If we have retransmitted an entry the maximum number of times, expire 3361 * that entry. 3362 */ 3363 void 3364 syn_cache_timer(void *arg) 3365 { 3366 struct syn_cache *sc = arg; 3367 int s; 3368 3369 s = splsoftnet(); 3370 callout_ack(&sc->sc_timer); 3371 3372 if (__predict_false(sc->sc_flags & SCF_DEAD)) { 3373 tcpstat.tcps_sc_delayed_free++; 3374 pool_put(&syn_cache_pool, sc); 3375 splx(s); 3376 return; 3377 } 3378 3379 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3380 /* Drop it -- too many retransmissions. */ 3381 goto dropit; 3382 } 3383 3384 /* 3385 * Compute the total amount of time this entry has 3386 * been on a queue. If this entry has been on longer 3387 * than the keep alive timer would allow, expire it. 3388 */ 3389 sc->sc_rxttot += sc->sc_rxtcur; 3390 if (sc->sc_rxttot >= TCPTV_KEEP_INIT) 3391 goto dropit; 3392 3393 tcpstat.tcps_sc_retransmitted++; 3394 (void) syn_cache_respond(sc, NULL); 3395 3396 /* Advance the timer back-off. */ 3397 sc->sc_rxtshift++; 3398 SYN_CACHE_TIMER_ARM(sc); 3399 3400 splx(s); 3401 return; 3402 3403 dropit: 3404 tcpstat.tcps_sc_timed_out++; 3405 SYN_CACHE_RM(sc); 3406 SYN_CACHE_PUT(sc); 3407 splx(s); 3408 } 3409 3410 /* 3411 * Remove syn cache created by the specified tcb entry, 3412 * because this does not make sense to keep them 3413 * (if there's no tcb entry, syn cache entry will never be used) 3414 */ 3415 void 3416 syn_cache_cleanup(struct tcpcb *tp) 3417 { 3418 struct syn_cache *sc, *nsc; 3419 int s; 3420 3421 s = splsoftnet(); 3422 3423 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3424 nsc = LIST_NEXT(sc, sc_tpq); 3425 3426 #ifdef DIAGNOSTIC 3427 if (sc->sc_tp != tp) 3428 panic("invalid sc_tp in syn_cache_cleanup"); 3429 #endif 3430 SYN_CACHE_RM(sc); 3431 SYN_CACHE_PUT(sc); 3432 } 3433 /* just for safety */ 3434 LIST_INIT(&tp->t_sc); 3435 3436 splx(s); 3437 } 3438 3439 /* 3440 * Find an entry in the syn cache. 3441 */ 3442 struct syn_cache * 3443 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst, 3444 struct syn_cache_head **headp) 3445 { 3446 struct syn_cache *sc; 3447 struct syn_cache_head *scp; 3448 u_int32_t hash; 3449 int s; 3450 3451 SYN_HASHALL(hash, src, dst); 3452 3453 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3454 *headp = scp; 3455 s = splsoftnet(); 3456 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3457 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3458 if (sc->sc_hash != hash) 3459 continue; 3460 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3461 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3462 splx(s); 3463 return (sc); 3464 } 3465 } 3466 splx(s); 3467 return (NULL); 3468 } 3469 3470 /* 3471 * This function gets called when we receive an ACK for a 3472 * socket in the LISTEN state. We look up the connection 3473 * in the syn cache, and if its there, we pull it out of 3474 * the cache and turn it into a full-blown connection in 3475 * the SYN-RECEIVED state. 3476 * 3477 * The return values may not be immediately obvious, and their effects 3478 * can be subtle, so here they are: 3479 * 3480 * NULL SYN was not found in cache; caller should drop the 3481 * packet and send an RST. 3482 * 3483 * -1 We were unable to create the new connection, and are 3484 * aborting it. An ACK,RST is being sent to the peer 3485 * (unless we got screwey sequence numbners; see below), 3486 * because the 3-way handshake has been completed. Caller 3487 * should not free the mbuf, since we may be using it. If 3488 * we are not, we will free it. 3489 * 3490 * Otherwise, the return value is a pointer to the new socket 3491 * associated with the connection. 3492 */ 3493 struct socket * 3494 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, 3495 struct tcphdr *th, unsigned int hlen, unsigned int tlen, 3496 struct socket *so, struct mbuf *m) 3497 { 3498 struct syn_cache *sc; 3499 struct syn_cache_head *scp; 3500 struct inpcb *inp = NULL; 3501 #ifdef INET6 3502 struct in6pcb *in6p = NULL; 3503 #endif 3504 struct tcpcb *tp = 0; 3505 struct mbuf *am; 3506 int s; 3507 struct socket *oso; 3508 3509 s = splsoftnet(); 3510 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3511 splx(s); 3512 return (NULL); 3513 } 3514 3515 /* 3516 * Verify the sequence and ack numbers. Try getting the correct 3517 * response again. 3518 */ 3519 if ((th->th_ack != sc->sc_iss + 1) || 3520 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3521 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3522 (void) syn_cache_respond(sc, m); 3523 splx(s); 3524 return ((struct socket *)(-1)); 3525 } 3526 3527 /* Remove this cache entry */ 3528 SYN_CACHE_RM(sc); 3529 splx(s); 3530 3531 /* 3532 * Ok, create the full blown connection, and set things up 3533 * as they would have been set up if we had created the 3534 * connection when the SYN arrived. If we can't create 3535 * the connection, abort it. 3536 */ 3537 /* 3538 * inp still has the OLD in_pcb stuff, set the 3539 * v6-related flags on the new guy, too. This is 3540 * done particularly for the case where an AF_INET6 3541 * socket is bound only to a port, and a v4 connection 3542 * comes in on that port. 3543 * we also copy the flowinfo from the original pcb 3544 * to the new one. 3545 */ 3546 oso = so; 3547 so = sonewconn(so, SS_ISCONNECTED); 3548 if (so == NULL) 3549 goto resetandabort; 3550 3551 switch (so->so_proto->pr_domain->dom_family) { 3552 #ifdef INET 3553 case AF_INET: 3554 inp = sotoinpcb(so); 3555 break; 3556 #endif 3557 #ifdef INET6 3558 case AF_INET6: 3559 in6p = sotoin6pcb(so); 3560 break; 3561 #endif 3562 } 3563 switch (src->sa_family) { 3564 #ifdef INET 3565 case AF_INET: 3566 if (inp) { 3567 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3568 inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port; 3569 inp->inp_options = ip_srcroute(); 3570 in_pcbstate(inp, INP_BOUND); 3571 if (inp->inp_options == NULL) { 3572 inp->inp_options = sc->sc_ipopts; 3573 sc->sc_ipopts = NULL; 3574 } 3575 } 3576 #ifdef INET6 3577 else if (in6p) { 3578 /* IPv4 packet to AF_INET6 socket */ 3579 bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr)); 3580 in6p->in6p_laddr.s6_addr16[5] = htons(0xffff); 3581 bcopy(&((struct sockaddr_in *)dst)->sin_addr, 3582 &in6p->in6p_laddr.s6_addr32[3], 3583 sizeof(((struct sockaddr_in *)dst)->sin_addr)); 3584 in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port; 3585 in6totcpcb(in6p)->t_family = AF_INET; 3586 if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY) 3587 in6p->in6p_flags |= IN6P_IPV6_V6ONLY; 3588 else 3589 in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY; 3590 in6_pcbstate(in6p, IN6P_BOUND); 3591 } 3592 #endif 3593 break; 3594 #endif 3595 #ifdef INET6 3596 case AF_INET6: 3597 if (in6p) { 3598 in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr; 3599 in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port; 3600 in6_pcbstate(in6p, IN6P_BOUND); 3601 } 3602 break; 3603 #endif 3604 } 3605 #ifdef INET6 3606 if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) { 3607 struct in6pcb *oin6p = sotoin6pcb(oso); 3608 /* inherit socket options from the listening socket */ 3609 in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS); 3610 if (in6p->in6p_flags & IN6P_CONTROLOPTS) { 3611 m_freem(in6p->in6p_options); 3612 in6p->in6p_options = 0; 3613 } 3614 ip6_savecontrol(in6p, &in6p->in6p_options, 3615 mtod(m, struct ip6_hdr *), m); 3616 } 3617 #endif 3618 3619 #if defined(IPSEC) || defined(FAST_IPSEC) 3620 /* 3621 * we make a copy of policy, instead of sharing the policy, 3622 * for better behavior in terms of SA lookup and dead SA removal. 3623 */ 3624 if (inp) { 3625 /* copy old policy into new socket's */ 3626 if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp)) 3627 printf("tcp_input: could not copy policy\n"); 3628 } 3629 #ifdef INET6 3630 else if (in6p) { 3631 /* copy old policy into new socket's */ 3632 if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp, 3633 in6p->in6p_sp)) 3634 printf("tcp_input: could not copy policy\n"); 3635 } 3636 #endif 3637 #endif 3638 3639 /* 3640 * Give the new socket our cached route reference. 3641 */ 3642 if (inp) 3643 inp->inp_route = sc->sc_route4; /* struct assignment */ 3644 #ifdef INET6 3645 else 3646 in6p->in6p_route = sc->sc_route6; 3647 #endif 3648 sc->sc_route4.ro_rt = NULL; 3649 3650 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3651 if (am == NULL) 3652 goto resetandabort; 3653 MCLAIM(am, &tcp_mowner); 3654 am->m_len = src->sa_len; 3655 bcopy(src, mtod(am, caddr_t), src->sa_len); 3656 if (inp) { 3657 if (in_pcbconnect(inp, am)) { 3658 (void) m_free(am); 3659 goto resetandabort; 3660 } 3661 } 3662 #ifdef INET6 3663 else if (in6p) { 3664 if (src->sa_family == AF_INET) { 3665 /* IPv4 packet to AF_INET6 socket */ 3666 struct sockaddr_in6 *sin6; 3667 sin6 = mtod(am, struct sockaddr_in6 *); 3668 am->m_len = sizeof(*sin6); 3669 bzero(sin6, sizeof(*sin6)); 3670 sin6->sin6_family = AF_INET6; 3671 sin6->sin6_len = sizeof(*sin6); 3672 sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port; 3673 sin6->sin6_addr.s6_addr16[5] = htons(0xffff); 3674 bcopy(&((struct sockaddr_in *)src)->sin_addr, 3675 &sin6->sin6_addr.s6_addr32[3], 3676 sizeof(sin6->sin6_addr.s6_addr32[3])); 3677 } 3678 if (in6_pcbconnect(in6p, am)) { 3679 (void) m_free(am); 3680 goto resetandabort; 3681 } 3682 } 3683 #endif 3684 else { 3685 (void) m_free(am); 3686 goto resetandabort; 3687 } 3688 (void) m_free(am); 3689 3690 if (inp) 3691 tp = intotcpcb(inp); 3692 #ifdef INET6 3693 else if (in6p) 3694 tp = in6totcpcb(in6p); 3695 #endif 3696 else 3697 tp = NULL; 3698 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3699 if (sc->sc_request_r_scale != 15) { 3700 tp->requested_s_scale = sc->sc_requested_s_scale; 3701 tp->request_r_scale = sc->sc_request_r_scale; 3702 tp->snd_scale = sc->sc_requested_s_scale; 3703 tp->rcv_scale = sc->sc_request_r_scale; 3704 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3705 } 3706 if (sc->sc_flags & SCF_TIMESTAMP) 3707 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3708 tp->ts_timebase = sc->sc_timebase; 3709 3710 tp->t_template = tcp_template(tp); 3711 if (tp->t_template == 0) { 3712 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3713 so = NULL; 3714 m_freem(m); 3715 goto abort; 3716 } 3717 3718 tp->iss = sc->sc_iss; 3719 tp->irs = sc->sc_irs; 3720 tcp_sendseqinit(tp); 3721 tcp_rcvseqinit(tp); 3722 tp->t_state = TCPS_SYN_RECEIVED; 3723 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 3724 tcpstat.tcps_accepts++; 3725 3726 if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack) 3727 tp->t_flags |= TF_WILL_SACK; 3728 3729 #ifdef TCP_SIGNATURE 3730 if (sc->sc_flags & SCF_SIGNATURE) 3731 tp->t_flags |= TF_SIGNATURE; 3732 #endif 3733 3734 /* Initialize tp->t_ourmss before we deal with the peer's! */ 3735 tp->t_ourmss = sc->sc_ourmaxseg; 3736 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 3737 3738 /* 3739 * Initialize the initial congestion window. If we 3740 * had to retransmit the SYN,ACK, we must initialize cwnd 3741 * to 1 segment (i.e. the Loss Window). 3742 */ 3743 if (sc->sc_rxtshift) 3744 tp->snd_cwnd = tp->t_peermss; 3745 else { 3746 int ss = tcp_init_win; 3747 #ifdef INET 3748 if (inp != NULL && in_localaddr(inp->inp_faddr)) 3749 ss = tcp_init_win_local; 3750 #endif 3751 #ifdef INET6 3752 if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr)) 3753 ss = tcp_init_win_local; 3754 #endif 3755 tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss); 3756 } 3757 3758 tcp_rmx_rtt(tp); 3759 tp->snd_wl1 = sc->sc_irs; 3760 tp->rcv_up = sc->sc_irs + 1; 3761 3762 /* 3763 * This is what whould have happened in tcp_output() when 3764 * the SYN,ACK was sent. 3765 */ 3766 tp->snd_up = tp->snd_una; 3767 tp->snd_max = tp->snd_nxt = tp->iss+1; 3768 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3769 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3770 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3771 tp->last_ack_sent = tp->rcv_nxt; 3772 tp->t_partialacks = -1; 3773 tp->t_dupacks = 0; 3774 3775 tcpstat.tcps_sc_completed++; 3776 SYN_CACHE_PUT(sc); 3777 return (so); 3778 3779 resetandabort: 3780 (void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST); 3781 abort: 3782 if (so != NULL) 3783 (void) soabort(so); 3784 SYN_CACHE_PUT(sc); 3785 tcpstat.tcps_sc_aborted++; 3786 return ((struct socket *)(-1)); 3787 } 3788 3789 /* 3790 * This function is called when we get a RST for a 3791 * non-existent connection, so that we can see if the 3792 * connection is in the syn cache. If it is, zap it. 3793 */ 3794 3795 void 3796 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th) 3797 { 3798 struct syn_cache *sc; 3799 struct syn_cache_head *scp; 3800 int s = splsoftnet(); 3801 3802 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3803 splx(s); 3804 return; 3805 } 3806 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3807 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3808 splx(s); 3809 return; 3810 } 3811 SYN_CACHE_RM(sc); 3812 splx(s); 3813 tcpstat.tcps_sc_reset++; 3814 SYN_CACHE_PUT(sc); 3815 } 3816 3817 void 3818 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst, 3819 struct tcphdr *th) 3820 { 3821 struct syn_cache *sc; 3822 struct syn_cache_head *scp; 3823 int s; 3824 3825 s = splsoftnet(); 3826 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3827 splx(s); 3828 return; 3829 } 3830 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3831 if (ntohl (th->th_seq) != sc->sc_iss) { 3832 splx(s); 3833 return; 3834 } 3835 3836 /* 3837 * If we've retransmitted 3 times and this is our second error, 3838 * we remove the entry. Otherwise, we allow it to continue on. 3839 * This prevents us from incorrectly nuking an entry during a 3840 * spurious network outage. 3841 * 3842 * See tcp_notify(). 3843 */ 3844 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3845 sc->sc_flags |= SCF_UNREACH; 3846 splx(s); 3847 return; 3848 } 3849 3850 SYN_CACHE_RM(sc); 3851 splx(s); 3852 tcpstat.tcps_sc_unreach++; 3853 SYN_CACHE_PUT(sc); 3854 } 3855 3856 /* 3857 * Given a LISTEN socket and an inbound SYN request, add 3858 * this to the syn cache, and send back a segment: 3859 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3860 * to the source. 3861 * 3862 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3863 * Doing so would require that we hold onto the data and deliver it 3864 * to the application. However, if we are the target of a SYN-flood 3865 * DoS attack, an attacker could send data which would eventually 3866 * consume all available buffer space if it were ACKed. By not ACKing 3867 * the data, we avoid this DoS scenario. 3868 */ 3869 3870 int 3871 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3872 unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp, 3873 int optlen, struct tcp_opt_info *oi) 3874 { 3875 struct tcpcb tb, *tp; 3876 long win; 3877 struct syn_cache *sc; 3878 struct syn_cache_head *scp; 3879 struct mbuf *ipopts; 3880 struct tcp_opt_info opti; 3881 3882 tp = sototcpcb(so); 3883 3884 bzero(&opti, sizeof(opti)); 3885 3886 /* 3887 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3888 * 3889 * Note this check is performed in tcp_input() very early on. 3890 */ 3891 3892 /* 3893 * Initialize some local state. 3894 */ 3895 win = sbspace(&so->so_rcv); 3896 if (win > TCP_MAXWIN) 3897 win = TCP_MAXWIN; 3898 3899 switch (src->sa_family) { 3900 #ifdef INET 3901 case AF_INET: 3902 /* 3903 * Remember the IP options, if any. 3904 */ 3905 ipopts = ip_srcroute(); 3906 break; 3907 #endif 3908 default: 3909 ipopts = NULL; 3910 } 3911 3912 #ifdef TCP_SIGNATURE 3913 if (optp || (tp->t_flags & TF_SIGNATURE)) 3914 #else 3915 if (optp) 3916 #endif 3917 { 3918 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3919 #ifdef TCP_SIGNATURE 3920 tb.t_flags |= (tp->t_flags & TF_SIGNATURE); 3921 #endif 3922 if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len - 3923 sizeof(struct tcphdr) - optlen - hlen, oi) < 0) 3924 return (0); 3925 } else 3926 tb.t_flags = 0; 3927 3928 /* 3929 * See if we already have an entry for this connection. 3930 * If we do, resend the SYN,ACK. We do not count this 3931 * as a retransmission (XXX though maybe we should). 3932 */ 3933 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3934 tcpstat.tcps_sc_dupesyn++; 3935 if (ipopts) { 3936 /* 3937 * If we were remembering a previous source route, 3938 * forget it and use the new one we've been given. 3939 */ 3940 if (sc->sc_ipopts) 3941 (void) m_free(sc->sc_ipopts); 3942 sc->sc_ipopts = ipopts; 3943 } 3944 sc->sc_timestamp = tb.ts_recent; 3945 if (syn_cache_respond(sc, m) == 0) { 3946 tcpstat.tcps_sndacks++; 3947 tcpstat.tcps_sndtotal++; 3948 } 3949 return (1); 3950 } 3951 3952 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3953 if (sc == NULL) { 3954 if (ipopts) 3955 (void) m_free(ipopts); 3956 return (0); 3957 } 3958 3959 /* 3960 * Fill in the cache, and put the necessary IP and TCP 3961 * options into the reply. 3962 */ 3963 bzero(sc, sizeof(struct syn_cache)); 3964 callout_init(&sc->sc_timer); 3965 bcopy(src, &sc->sc_src, src->sa_len); 3966 bcopy(dst, &sc->sc_dst, dst->sa_len); 3967 sc->sc_flags = 0; 3968 sc->sc_ipopts = ipopts; 3969 sc->sc_irs = th->th_seq; 3970 switch (src->sa_family) { 3971 #ifdef INET 3972 case AF_INET: 3973 { 3974 struct sockaddr_in *srcin = (void *) src; 3975 struct sockaddr_in *dstin = (void *) dst; 3976 3977 sc->sc_iss = tcp_new_iss1(&dstin->sin_addr, 3978 &srcin->sin_addr, dstin->sin_port, 3979 srcin->sin_port, sizeof(dstin->sin_addr), 0); 3980 break; 3981 } 3982 #endif /* INET */ 3983 #ifdef INET6 3984 case AF_INET6: 3985 { 3986 struct sockaddr_in6 *srcin6 = (void *) src; 3987 struct sockaddr_in6 *dstin6 = (void *) dst; 3988 3989 sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr, 3990 &srcin6->sin6_addr, dstin6->sin6_port, 3991 srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0); 3992 break; 3993 } 3994 #endif /* INET6 */ 3995 } 3996 sc->sc_peermaxseg = oi->maxseg; 3997 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 3998 m->m_pkthdr.rcvif : NULL, 3999 sc->sc_src.sa.sa_family); 4000 sc->sc_win = win; 4001 sc->sc_timebase = tcp_now; /* see tcp_newtcpcb() */ 4002 sc->sc_timestamp = tb.ts_recent; 4003 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4004 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4005 sc->sc_flags |= SCF_TIMESTAMP; 4006 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4007 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4008 sc->sc_requested_s_scale = tb.requested_s_scale; 4009 sc->sc_request_r_scale = 0; 4010 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4011 TCP_MAXWIN << sc->sc_request_r_scale < 4012 so->so_rcv.sb_hiwat) 4013 sc->sc_request_r_scale++; 4014 } else { 4015 sc->sc_requested_s_scale = 15; 4016 sc->sc_request_r_scale = 15; 4017 } 4018 if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack) 4019 sc->sc_flags |= SCF_SACK_PERMIT; 4020 #ifdef TCP_SIGNATURE 4021 if (tb.t_flags & TF_SIGNATURE) 4022 sc->sc_flags |= SCF_SIGNATURE; 4023 #endif 4024 sc->sc_tp = tp; 4025 if (syn_cache_respond(sc, m) == 0) { 4026 syn_cache_insert(sc, tp); 4027 tcpstat.tcps_sndacks++; 4028 tcpstat.tcps_sndtotal++; 4029 } else { 4030 SYN_CACHE_PUT(sc); 4031 tcpstat.tcps_sc_dropped++; 4032 } 4033 return (1); 4034 } 4035 4036 int 4037 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4038 { 4039 struct route *ro; 4040 u_int8_t *optp; 4041 int optlen, error; 4042 u_int16_t tlen; 4043 struct ip *ip = NULL; 4044 #ifdef INET6 4045 struct ip6_hdr *ip6 = NULL; 4046 #endif 4047 struct tcpcb *tp; 4048 struct tcphdr *th; 4049 u_int hlen; 4050 struct socket *so; 4051 4052 switch (sc->sc_src.sa.sa_family) { 4053 case AF_INET: 4054 hlen = sizeof(struct ip); 4055 ro = &sc->sc_route4; 4056 break; 4057 #ifdef INET6 4058 case AF_INET6: 4059 hlen = sizeof(struct ip6_hdr); 4060 ro = (struct route *)&sc->sc_route6; 4061 break; 4062 #endif 4063 default: 4064 if (m) 4065 m_freem(m); 4066 return (EAFNOSUPPORT); 4067 } 4068 4069 /* Compute the size of the TCP options. */ 4070 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4071 ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) + 4072 #ifdef TCP_SIGNATURE 4073 ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) + 4074 #endif 4075 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4076 4077 tlen = hlen + sizeof(struct tcphdr) + optlen; 4078 4079 /* 4080 * Create the IP+TCP header from scratch. 4081 */ 4082 if (m) 4083 m_freem(m); 4084 #ifdef DIAGNOSTIC 4085 if (max_linkhdr + tlen > MCLBYTES) 4086 return (ENOBUFS); 4087 #endif 4088 MGETHDR(m, M_DONTWAIT, MT_DATA); 4089 if (m && tlen > MHLEN) { 4090 MCLGET(m, M_DONTWAIT); 4091 if ((m->m_flags & M_EXT) == 0) { 4092 m_freem(m); 4093 m = NULL; 4094 } 4095 } 4096 if (m == NULL) 4097 return (ENOBUFS); 4098 MCLAIM(m, &tcp_tx_mowner); 4099 4100 /* Fixup the mbuf. */ 4101 m->m_data += max_linkhdr; 4102 m->m_len = m->m_pkthdr.len = tlen; 4103 if (sc->sc_tp) { 4104 tp = sc->sc_tp; 4105 if (tp->t_inpcb) 4106 so = tp->t_inpcb->inp_socket; 4107 #ifdef INET6 4108 else if (tp->t_in6pcb) 4109 so = tp->t_in6pcb->in6p_socket; 4110 #endif 4111 else 4112 so = NULL; 4113 } else 4114 so = NULL; 4115 m->m_pkthdr.rcvif = NULL; 4116 memset(mtod(m, u_char *), 0, tlen); 4117 4118 switch (sc->sc_src.sa.sa_family) { 4119 case AF_INET: 4120 ip = mtod(m, struct ip *); 4121 ip->ip_v = 4; 4122 ip->ip_dst = sc->sc_src.sin.sin_addr; 4123 ip->ip_src = sc->sc_dst.sin.sin_addr; 4124 ip->ip_p = IPPROTO_TCP; 4125 th = (struct tcphdr *)(ip + 1); 4126 th->th_dport = sc->sc_src.sin.sin_port; 4127 th->th_sport = sc->sc_dst.sin.sin_port; 4128 break; 4129 #ifdef INET6 4130 case AF_INET6: 4131 ip6 = mtod(m, struct ip6_hdr *); 4132 ip6->ip6_vfc = IPV6_VERSION; 4133 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4134 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4135 ip6->ip6_nxt = IPPROTO_TCP; 4136 /* ip6_plen will be updated in ip6_output() */ 4137 th = (struct tcphdr *)(ip6 + 1); 4138 th->th_dport = sc->sc_src.sin6.sin6_port; 4139 th->th_sport = sc->sc_dst.sin6.sin6_port; 4140 break; 4141 #endif 4142 default: 4143 th = NULL; 4144 } 4145 4146 th->th_seq = htonl(sc->sc_iss); 4147 th->th_ack = htonl(sc->sc_irs + 1); 4148 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4149 th->th_flags = TH_SYN|TH_ACK; 4150 th->th_win = htons(sc->sc_win); 4151 /* th_sum already 0 */ 4152 /* th_urp already 0 */ 4153 4154 /* Tack on the TCP options. */ 4155 optp = (u_int8_t *)(th + 1); 4156 *optp++ = TCPOPT_MAXSEG; 4157 *optp++ = 4; 4158 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4159 *optp++ = sc->sc_ourmaxseg & 0xff; 4160 4161 if (sc->sc_request_r_scale != 15) { 4162 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4163 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4164 sc->sc_request_r_scale); 4165 optp += 4; 4166 } 4167 4168 if (sc->sc_flags & SCF_TIMESTAMP) { 4169 u_int32_t *lp = (u_int32_t *)(optp); 4170 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4171 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4172 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4173 *lp = htonl(sc->sc_timestamp); 4174 optp += TCPOLEN_TSTAMP_APPA; 4175 } 4176 4177 if (sc->sc_flags & SCF_SACK_PERMIT) { 4178 u_int8_t *p = optp; 4179 4180 /* Let the peer know that we will SACK. */ 4181 p[0] = TCPOPT_SACK_PERMITTED; 4182 p[1] = 2; 4183 p[2] = TCPOPT_NOP; 4184 p[3] = TCPOPT_NOP; 4185 optp += 4; 4186 } 4187 4188 #ifdef TCP_SIGNATURE 4189 if (sc->sc_flags & SCF_SIGNATURE) { 4190 struct secasvar *sav; 4191 u_int8_t *sigp; 4192 4193 sav = tcp_signature_getsav(m, th); 4194 4195 if (sav == NULL) { 4196 if (m) 4197 m_freem(m); 4198 return (EPERM); 4199 } 4200 4201 *optp++ = TCPOPT_SIGNATURE; 4202 *optp++ = TCPOLEN_SIGNATURE; 4203 sigp = optp; 4204 bzero(optp, TCP_SIGLEN); 4205 optp += TCP_SIGLEN; 4206 *optp++ = TCPOPT_NOP; 4207 *optp++ = TCPOPT_EOL; 4208 4209 (void)tcp_signature(m, th, hlen, sav, sigp); 4210 4211 key_sa_recordxfer(sav, m); 4212 #ifdef FAST_IPSEC 4213 KEY_FREESAV(&sav); 4214 #else 4215 key_freesav(sav); 4216 #endif 4217 } 4218 #endif 4219 4220 /* Compute the packet's checksum. */ 4221 switch (sc->sc_src.sa.sa_family) { 4222 case AF_INET: 4223 ip->ip_len = htons(tlen - hlen); 4224 th->th_sum = 0; 4225 th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4226 break; 4227 #ifdef INET6 4228 case AF_INET6: 4229 ip6->ip6_plen = htons(tlen - hlen); 4230 th->th_sum = 0; 4231 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4232 break; 4233 #endif 4234 } 4235 4236 /* 4237 * Fill in some straggling IP bits. Note the stack expects 4238 * ip_len to be in host order, for convenience. 4239 */ 4240 switch (sc->sc_src.sa.sa_family) { 4241 #ifdef INET 4242 case AF_INET: 4243 ip->ip_len = htons(tlen); 4244 ip->ip_ttl = ip_defttl; 4245 /* XXX tos? */ 4246 break; 4247 #endif 4248 #ifdef INET6 4249 case AF_INET6: 4250 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4251 ip6->ip6_vfc |= IPV6_VERSION; 4252 ip6->ip6_plen = htons(tlen - hlen); 4253 /* ip6_hlim will be initialized afterwards */ 4254 /* XXX flowlabel? */ 4255 break; 4256 #endif 4257 } 4258 4259 /* XXX use IPsec policy on listening socket, on SYN ACK */ 4260 tp = sc->sc_tp; 4261 4262 switch (sc->sc_src.sa.sa_family) { 4263 #ifdef INET 4264 case AF_INET: 4265 error = ip_output(m, sc->sc_ipopts, ro, 4266 (ip_mtudisc ? IP_MTUDISC : 0), 4267 (struct ip_moptions *)NULL, so); 4268 break; 4269 #endif 4270 #ifdef INET6 4271 case AF_INET6: 4272 ip6->ip6_hlim = in6_selecthlim(NULL, 4273 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4274 4275 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4276 (struct ip6_moptions *)0, so, NULL); 4277 break; 4278 #endif 4279 default: 4280 error = EAFNOSUPPORT; 4281 break; 4282 } 4283 return (error); 4284 } 4285
Indexes created Tue Sep 23 01:09:51 GMT 2025