1 /* $OpenBSD: tcp_input.c,v 1.292 2015/06/07 12:02:28 jsg Exp $ */ 2 /* $NetBSD: tcp_input.c,v 1.23 1996/02/13 23:43:44 christos Exp $ */ 3 4 /* 5 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994 6 * The Regents of the University of California. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 33 * 34 * NRL grants permission for redistribution and use in source and binary 35 * forms, with or without modification, of the software and documentation 36 * created at NRL provided that the following conditions are met: 37 * 38 * 1. Redistributions of source code must retain the above copyright 39 * notice, this list of conditions and the following disclaimer. 40 * 2. Redistributions in binary form must reproduce the above copyright 41 * notice, this list of conditions and the following disclaimer in the 42 * documentation and/or other materials provided with the distribution. 43 * 3. All advertising materials mentioning features or use of this software 44 * must display the following acknowledgements: 45 * This product includes software developed by the University of 46 * California, Berkeley and its contributors. 47 * This product includes software developed at the Information 48 * Technology Division, US Naval Research Laboratory. 49 * 4. Neither the name of the NRL nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 54 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 55 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 56 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 57 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 58 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 59 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 60 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 61 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 62 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 63 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 64 * 65 * The views and conclusions contained in the software and documentation 66 * are those of the authors and should not be interpreted as representing 67 * official policies, either expressed or implied, of the US Naval 68 * Research Laboratory (NRL). 69 */ 70 71 #include "pf.h" 72 73 #include <sys/param.h> 74 #include <sys/systm.h> 75 #include <sys/mbuf.h> 76 #include <sys/protosw.h> 77 #include <sys/socket.h> 78 #include <sys/socketvar.h> 79 #include <sys/timeout.h> 80 #include <sys/kernel.h> 81 #include <sys/pool.h> 82 83 #include <net/if.h> 84 #include <net/if_var.h> 85 #include <net/route.h> 86 87 #include <netinet/in.h> 88 #include <netinet/ip.h> 89 #include <netinet/in_pcb.h> 90 #include <netinet/ip_var.h> 91 #include <netinet/tcp.h> 92 #include <netinet/tcp_fsm.h> 93 #include <netinet/tcp_seq.h> 94 #include <netinet/tcp_timer.h> 95 #include <netinet/tcp_var.h> 96 #include <netinet/tcpip.h> 97 #include <netinet/tcp_debug.h> 98 99 #if NPF > 0 100 #include <net/pfvar.h> 101 #endif 102 103 struct tcpiphdr tcp_saveti; 104 105 int tcp_mss_adv(struct ifnet *, int); 106 int tcp_flush_queue(struct tcpcb *); 107 108 #ifdef INET6 109 #include <netinet6/in6_var.h> 110 #include <netinet6/nd6.h> 111 112 struct tcpipv6hdr tcp_saveti6; 113 114 /* for the packet header length in the mbuf */ 115 #define M_PH_LEN(m) (((struct mbuf *)(m))->m_pkthdr.len) 116 #define M_V6_LEN(m) (M_PH_LEN(m) - sizeof(struct ip6_hdr)) 117 #define M_V4_LEN(m) (M_PH_LEN(m) - sizeof(struct ip)) 118 #endif /* INET6 */ 119 120 int tcprexmtthresh = 3; 121 int tcptv_keep_init = TCPTV_KEEP_INIT; 122 123 int tcp_rst_ppslim = 100; /* 100pps */ 124 int tcp_rst_ppslim_count = 0; 125 struct timeval tcp_rst_ppslim_last; 126 127 int tcp_ackdrop_ppslim = 100; /* 100pps */ 128 int tcp_ackdrop_ppslim_count = 0; 129 struct timeval tcp_ackdrop_ppslim_last; 130 131 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 132 133 /* for modulo comparisons of timestamps */ 134 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 135 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 136 137 /* for TCP SACK comparisons */ 138 #define SEQ_MIN(a,b) (SEQ_LT(a,b) ? (a) : (b)) 139 #define SEQ_MAX(a,b) (SEQ_GT(a,b) ? (a) : (b)) 140 141 /* 142 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 143 */ 144 #ifdef INET6 145 #define ND6_HINT(tp) \ 146 do { \ 147 if (tp && tp->t_inpcb && (tp->t_inpcb->inp_flags & INP_IPV6) && \ 148 tp->t_inpcb->inp_route6.ro_rt) { \ 149 nd6_nud_hint(tp->t_inpcb->inp_route6.ro_rt, NULL, 0, \ 150 tp->t_inpcb->inp_rtableid); \ 151 } \ 152 } while (0) 153 #else 154 #define ND6_HINT(tp) 155 #endif 156 157 #ifdef TCP_ECN 158 /* 159 * ECN (Explicit Congestion Notification) support based on RFC3168 160 * implementation note: 161 * snd_last is used to track a recovery phase. 162 * when cwnd is reduced, snd_last is set to snd_max. 163 * while snd_last > snd_una, the sender is in a recovery phase and 164 * its cwnd should not be reduced again. 165 * snd_last follows snd_una when not in a recovery phase. 166 */ 167 #endif 168 169 /* 170 * Macro to compute ACK transmission behavior. Delay the ACK unless 171 * we have already delayed an ACK (must send an ACK every two segments). 172 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 173 * option is enabled or when the packet is coming from a loopback 174 * interface. 175 */ 176 #define TCP_SETUP_ACK(tp, tiflags, m) \ 177 do { \ 178 if ((tp)->t_flags & TF_DELACK || \ 179 (tcp_ack_on_push && (tiflags) & TH_PUSH) || \ 180 (m && (m->m_flags & M_PKTHDR) && m->m_pkthdr.rcvif && \ 181 (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK))) \ 182 tp->t_flags |= TF_ACKNOW; \ 183 else \ 184 TCP_SET_DELACK(tp); \ 185 } while (0) 186 187 void syn_cache_put(struct syn_cache *); 188 void syn_cache_rm(struct syn_cache *); 189 190 /* 191 * Insert segment ti into reassembly queue of tcp with 192 * control block tp. Return TH_FIN if reassembly now includes 193 * a segment with FIN. The macro form does the common case inline 194 * (segment is the next to be received on an established connection, 195 * and the queue is empty), avoiding linkage into and removal 196 * from the queue and repetition of various conversions. 197 * Set DELACK for segments received in order, but ack immediately 198 * when segments are out of order (so fast retransmit can work). 199 */ 200 201 int 202 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen) 203 { 204 struct tcpqent *p, *q, *nq, *tiqe; 205 206 /* 207 * Allocate a new queue entry, before we throw away any data. 208 * If we can't, just drop the packet. XXX 209 */ 210 tiqe = pool_get(&tcpqe_pool, PR_NOWAIT); 211 if (tiqe == NULL) { 212 tiqe = TAILQ_LAST(&tp->t_segq, tcpqehead); 213 if (tiqe != NULL && th->th_seq == tp->rcv_nxt) { 214 /* Reuse last entry since new segment fills a hole */ 215 m_freem(tiqe->tcpqe_m); 216 TAILQ_REMOVE(&tp->t_segq, tiqe, tcpqe_q); 217 } 218 if (tiqe == NULL || th->th_seq != tp->rcv_nxt) { 219 /* Flush segment queue for this connection */ 220 tcp_freeq(tp); 221 tcpstat.tcps_rcvmemdrop++; 222 m_freem(m); 223 return (0); 224 } 225 } 226 227 /* 228 * Find a segment which begins after this one does. 229 */ 230 for (p = NULL, q = TAILQ_FIRST(&tp->t_segq); q != NULL; 231 p = q, q = TAILQ_NEXT(q, tcpqe_q)) 232 if (SEQ_GT(q->tcpqe_tcp->th_seq, th->th_seq)) 233 break; 234 235 /* 236 * If there is a preceding segment, it may provide some of 237 * our data already. If so, drop the data from the incoming 238 * segment. If it provides all of our data, drop us. 239 */ 240 if (p != NULL) { 241 struct tcphdr *phdr = p->tcpqe_tcp; 242 int i; 243 244 /* conversion to int (in i) handles seq wraparound */ 245 i = phdr->th_seq + phdr->th_reseqlen - th->th_seq; 246 if (i > 0) { 247 if (i >= *tlen) { 248 tcpstat.tcps_rcvduppack++; 249 tcpstat.tcps_rcvdupbyte += *tlen; 250 m_freem(m); 251 pool_put(&tcpqe_pool, tiqe); 252 return (0); 253 } 254 m_adj(m, i); 255 *tlen -= i; 256 th->th_seq += i; 257 } 258 } 259 tcpstat.tcps_rcvoopack++; 260 tcpstat.tcps_rcvoobyte += *tlen; 261 262 /* 263 * While we overlap succeeding segments trim them or, 264 * if they are completely covered, dequeue them. 265 */ 266 for (; q != NULL; q = nq) { 267 struct tcphdr *qhdr = q->tcpqe_tcp; 268 int i = (th->th_seq + *tlen) - qhdr->th_seq; 269 270 if (i <= 0) 271 break; 272 if (i < qhdr->th_reseqlen) { 273 qhdr->th_seq += i; 274 qhdr->th_reseqlen -= i; 275 m_adj(q->tcpqe_m, i); 276 break; 277 } 278 nq = TAILQ_NEXT(q, tcpqe_q); 279 m_freem(q->tcpqe_m); 280 TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q); 281 pool_put(&tcpqe_pool, q); 282 } 283 284 /* Insert the new segment queue entry into place. */ 285 tiqe->tcpqe_m = m; 286 th->th_reseqlen = *tlen; 287 tiqe->tcpqe_tcp = th; 288 if (p == NULL) { 289 TAILQ_INSERT_HEAD(&tp->t_segq, tiqe, tcpqe_q); 290 } else { 291 TAILQ_INSERT_AFTER(&tp->t_segq, p, tiqe, tcpqe_q); 292 } 293 294 if (th->th_seq != tp->rcv_nxt) 295 return (0); 296 297 return (tcp_flush_queue(tp)); 298 } 299 300 int 301 tcp_flush_queue(struct tcpcb *tp) 302 { 303 struct socket *so = tp->t_inpcb->inp_socket; 304 struct tcpqent *q, *nq; 305 int flags; 306 307 /* 308 * Present data to user, advancing rcv_nxt through 309 * completed sequence space. 310 */ 311 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 312 return (0); 313 q = TAILQ_FIRST(&tp->t_segq); 314 if (q == NULL || q->tcpqe_tcp->th_seq != tp->rcv_nxt) 315 return (0); 316 if (tp->t_state == TCPS_SYN_RECEIVED && q->tcpqe_tcp->th_reseqlen) 317 return (0); 318 do { 319 tp->rcv_nxt += q->tcpqe_tcp->th_reseqlen; 320 flags = q->tcpqe_tcp->th_flags & TH_FIN; 321 322 nq = TAILQ_NEXT(q, tcpqe_q); 323 TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q); 324 ND6_HINT(tp); 325 if (so->so_state & SS_CANTRCVMORE) 326 m_freem(q->tcpqe_m); 327 else 328 sbappendstream(&so->so_rcv, q->tcpqe_m); 329 pool_put(&tcpqe_pool, q); 330 q = nq; 331 } while (q != NULL && q->tcpqe_tcp->th_seq == tp->rcv_nxt); 332 tp->t_flags |= TF_BLOCKOUTPUT; 333 sorwakeup(so); 334 tp->t_flags &= ~TF_BLOCKOUTPUT; 335 return (flags); 336 } 337 338 #ifdef INET6 339 int 340 tcp6_input(struct mbuf **mp, int *offp, int proto) 341 { 342 struct mbuf *m = *mp; 343 344 tcp_input(m, *offp, proto); 345 return IPPROTO_DONE; 346 } 347 #endif 348 349 /* 350 * TCP input routine, follows pages 65-76 of the 351 * protocol specification dated September, 1981 very closely. 352 */ 353 void 354 tcp_input(struct mbuf *m, ...) 355 { 356 struct ip *ip; 357 struct inpcb *inp = NULL; 358 u_int8_t *optp = NULL; 359 int optlen = 0; 360 int tlen, off; 361 struct tcpcb *tp = NULL; 362 int tiflags; 363 struct socket *so = NULL; 364 int todrop, acked, ourfinisacked; 365 int hdroptlen = 0; 366 short ostate = 0; 367 tcp_seq iss, *reuse = NULL; 368 u_long tiwin; 369 struct tcp_opt_info opti; 370 int iphlen; 371 va_list ap; 372 struct tcphdr *th; 373 #ifdef INET6 374 struct ip6_hdr *ip6 = NULL; 375 #endif /* INET6 */ 376 #ifdef IPSEC 377 struct m_tag *mtag; 378 struct tdb_ident *tdbi; 379 struct tdb *tdb; 380 int error; 381 #endif /* IPSEC */ 382 int af; 383 #ifdef TCP_ECN 384 u_char iptos; 385 #endif 386 387 va_start(ap, m); 388 iphlen = va_arg(ap, int); 389 va_end(ap); 390 391 tcpstat.tcps_rcvtotal++; 392 393 opti.ts_present = 0; 394 opti.maxseg = 0; 395 396 /* 397 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 398 */ 399 if (m->m_flags & (M_BCAST|M_MCAST)) 400 goto drop; 401 402 /* 403 * Before we do ANYTHING, we have to figure out if it's TCP/IPv6 or 404 * TCP/IPv4. 405 */ 406 switch (mtod(m, struct ip *)->ip_v) { 407 #ifdef INET6 408 case 6: 409 af = AF_INET6; 410 break; 411 #endif 412 case 4: 413 af = AF_INET; 414 break; 415 default: 416 m_freem(m); 417 return; /*EAFNOSUPPORT*/ 418 } 419 420 /* 421 * Get IP and TCP header together in first mbuf. 422 * Note: IP leaves IP header in first mbuf. 423 */ 424 switch (af) { 425 case AF_INET: 426 #ifdef DIAGNOSTIC 427 if (iphlen < sizeof(struct ip)) { 428 m_freem(m); 429 return; 430 } 431 #endif /* DIAGNOSTIC */ 432 break; 433 #ifdef INET6 434 case AF_INET6: 435 #ifdef DIAGNOSTIC 436 if (iphlen < sizeof(struct ip6_hdr)) { 437 m_freem(m); 438 return; 439 } 440 #endif /* DIAGNOSTIC */ 441 break; 442 #endif 443 default: 444 m_freem(m); 445 return; 446 } 447 448 IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, sizeof(*th)); 449 if (!th) { 450 tcpstat.tcps_rcvshort++; 451 return; 452 } 453 454 tlen = m->m_pkthdr.len - iphlen; 455 ip = NULL; 456 #ifdef INET6 457 ip6 = NULL; 458 #endif 459 switch (af) { 460 case AF_INET: 461 ip = mtod(m, struct ip *); 462 #ifdef TCP_ECN 463 /* save ip_tos before clearing it for checksum */ 464 iptos = ip->ip_tos; 465 #endif 466 break; 467 #ifdef INET6 468 case AF_INET6: 469 ip6 = mtod(m, struct ip6_hdr *); 470 #ifdef TCP_ECN 471 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 472 #endif 473 474 /* Be proactive about malicious use of IPv4 mapped address */ 475 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 476 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 477 /* XXX stat */ 478 goto drop; 479 } 480 481 /* 482 * Be proactive about unspecified IPv6 address in source. 483 * As we use all-zero to indicate unbounded/unconnected pcb, 484 * unspecified IPv6 address can be used to confuse us. 485 * 486 * Note that packets with unspecified IPv6 destination is 487 * already dropped in ip6_input. 488 */ 489 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 490 /* XXX stat */ 491 goto drop; 492 } 493 494 /* Discard packets to multicast */ 495 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 496 /* XXX stat */ 497 goto drop; 498 } 499 break; 500 #endif 501 } 502 503 /* 504 * Checksum extended TCP header and data. 505 */ 506 if ((m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_OK) == 0) { 507 int sum; 508 509 if (m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_BAD) { 510 tcpstat.tcps_rcvbadsum++; 511 goto drop; 512 } 513 tcpstat.tcps_inswcsum++; 514 switch (af) { 515 case AF_INET: 516 sum = in4_cksum(m, IPPROTO_TCP, iphlen, tlen); 517 break; 518 #ifdef INET6 519 case AF_INET6: 520 sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 521 tlen); 522 break; 523 #endif 524 } 525 if (sum != 0) { 526 tcpstat.tcps_rcvbadsum++; 527 goto drop; 528 } 529 } 530 531 /* 532 * Check that TCP offset makes sense, 533 * pull out TCP options and adjust length. XXX 534 */ 535 off = th->th_off << 2; 536 if (off < sizeof(struct tcphdr) || off > tlen) { 537 tcpstat.tcps_rcvbadoff++; 538 goto drop; 539 } 540 tlen -= off; 541 if (off > sizeof(struct tcphdr)) { 542 IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, off); 543 if (!th) { 544 tcpstat.tcps_rcvshort++; 545 return; 546 } 547 optlen = off - sizeof(struct tcphdr); 548 optp = (u_int8_t *)(th + 1); 549 /* 550 * Do quick retrieval of timestamp options ("options 551 * prediction?"). If timestamp is the only option and it's 552 * formatted as recommended in RFC 1323 appendix A, we 553 * quickly get the values now and not bother calling 554 * tcp_dooptions(), etc. 555 */ 556 if ((optlen == TCPOLEN_TSTAMP_APPA || 557 (optlen > TCPOLEN_TSTAMP_APPA && 558 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 559 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 560 (th->th_flags & TH_SYN) == 0) { 561 opti.ts_present = 1; 562 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 563 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 564 optp = NULL; /* we've parsed the options */ 565 } 566 } 567 tiflags = th->th_flags; 568 569 /* 570 * Convert TCP protocol specific fields to host format. 571 */ 572 NTOHL(th->th_seq); 573 NTOHL(th->th_ack); 574 NTOHS(th->th_win); 575 NTOHS(th->th_urp); 576 577 /* 578 * Locate pcb for segment. 579 */ 580 #if NPF > 0 581 if (m->m_pkthdr.pf.statekey) { 582 inp = m->m_pkthdr.pf.statekey->inp; 583 if (inp && inp->inp_pf_sk) 584 KASSERT(m->m_pkthdr.pf.statekey == inp->inp_pf_sk); 585 } 586 #endif 587 findpcb: 588 if (inp == NULL) { 589 switch (af) { 590 #ifdef INET6 591 case AF_INET6: 592 inp = in6_pcbhashlookup(&tcbtable, &ip6->ip6_src, 593 th->th_sport, &ip6->ip6_dst, th->th_dport, 594 m->m_pkthdr.ph_rtableid); 595 break; 596 #endif 597 case AF_INET: 598 inp = in_pcbhashlookup(&tcbtable, ip->ip_src, 599 th->th_sport, ip->ip_dst, th->th_dport, 600 m->m_pkthdr.ph_rtableid); 601 break; 602 } 603 #if NPF > 0 604 if (m->m_pkthdr.pf.statekey && inp) { 605 m->m_pkthdr.pf.statekey->inp = inp; 606 inp->inp_pf_sk = m->m_pkthdr.pf.statekey; 607 } 608 #endif 609 } 610 if (inp == NULL) { 611 int inpl_reverse = 0; 612 if (m->m_pkthdr.pf.flags & PF_TAG_TRANSLATE_LOCALHOST) 613 inpl_reverse = 1; 614 ++tcpstat.tcps_pcbhashmiss; 615 switch (af) { 616 #ifdef INET6 617 case AF_INET6: 618 inp = in6_pcblookup_listen(&tcbtable, 619 &ip6->ip6_dst, th->th_dport, inpl_reverse, m, 620 m->m_pkthdr.ph_rtableid); 621 break; 622 #endif /* INET6 */ 623 case AF_INET: 624 inp = in_pcblookup_listen(&tcbtable, 625 ip->ip_dst, th->th_dport, inpl_reverse, m, 626 m->m_pkthdr.ph_rtableid); 627 break; 628 } 629 /* 630 * If the state is CLOSED (i.e., TCB does not exist) then 631 * all data in the incoming segment is discarded. 632 * If the TCB exists but is in CLOSED state, it is embryonic, 633 * but should either do a listen or a connect soon. 634 */ 635 if (inp == NULL) { 636 ++tcpstat.tcps_noport; 637 goto dropwithreset_ratelim; 638 } 639 } 640 KASSERT(sotoinpcb(inp->inp_socket) == inp); 641 KASSERT(intotcpcb(inp)->t_inpcb == inp); 642 643 /* Check the minimum TTL for socket. */ 644 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) 645 goto drop; 646 647 tp = intotcpcb(inp); 648 if (tp == NULL) 649 goto dropwithreset_ratelim; 650 if (tp->t_state == TCPS_CLOSED) 651 goto drop; 652 653 /* Unscale the window into a 32-bit value. */ 654 if ((tiflags & TH_SYN) == 0) 655 tiwin = th->th_win << tp->snd_scale; 656 else 657 tiwin = th->th_win; 658 659 so = inp->inp_socket; 660 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 661 union syn_cache_sa src; 662 union syn_cache_sa dst; 663 664 bzero(&src, sizeof(src)); 665 bzero(&dst, sizeof(dst)); 666 switch (af) { 667 case AF_INET: 668 src.sin.sin_len = sizeof(struct sockaddr_in); 669 src.sin.sin_family = AF_INET; 670 src.sin.sin_addr = ip->ip_src; 671 src.sin.sin_port = th->th_sport; 672 673 dst.sin.sin_len = sizeof(struct sockaddr_in); 674 dst.sin.sin_family = AF_INET; 675 dst.sin.sin_addr = ip->ip_dst; 676 dst.sin.sin_port = th->th_dport; 677 break; 678 #ifdef INET6 679 case AF_INET6: 680 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 681 src.sin6.sin6_family = AF_INET6; 682 src.sin6.sin6_addr = ip6->ip6_src; 683 src.sin6.sin6_port = th->th_sport; 684 685 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 686 dst.sin6.sin6_family = AF_INET6; 687 dst.sin6.sin6_addr = ip6->ip6_dst; 688 dst.sin6.sin6_port = th->th_dport; 689 break; 690 #endif /* INET6 */ 691 default: 692 goto badsyn; /*sanity*/ 693 } 694 695 if (so->so_options & SO_DEBUG) { 696 ostate = tp->t_state; 697 switch (af) { 698 #ifdef INET6 699 case AF_INET6: 700 bcopy(ip6, &tcp_saveti6.ti6_i, sizeof(*ip6)); 701 bcopy(th, &tcp_saveti6.ti6_t, sizeof(*th)); 702 break; 703 #endif 704 case AF_INET: 705 bcopy(ip, &tcp_saveti.ti_i, sizeof(*ip)); 706 bcopy(th, &tcp_saveti.ti_t, sizeof(*th)); 707 break; 708 } 709 } 710 if (so->so_options & SO_ACCEPTCONN) { 711 switch (tiflags & (TH_RST|TH_SYN|TH_ACK)) { 712 713 case TH_SYN|TH_ACK|TH_RST: 714 case TH_SYN|TH_RST: 715 case TH_ACK|TH_RST: 716 case TH_RST: 717 syn_cache_reset(&src.sa, &dst.sa, th, 718 inp->inp_rtableid); 719 goto drop; 720 721 case TH_SYN|TH_ACK: 722 /* 723 * Received a SYN,ACK. This should 724 * never happen while we are in 725 * LISTEN. Send an RST. 726 */ 727 goto badsyn; 728 729 case TH_ACK: 730 so = syn_cache_get(&src.sa, &dst.sa, 731 th, iphlen, tlen, so, m); 732 if (so == NULL) { 733 /* 734 * We don't have a SYN for 735 * this ACK; send an RST. 736 */ 737 goto badsyn; 738 } else if (so == (struct socket *)(-1)) { 739 /* 740 * We were unable to create 741 * the connection. If the 742 * 3-way handshake was 743 * completed, and RST has 744 * been sent to the peer. 745 * Since the mbuf might be 746 * in use for the reply, 747 * do not free it. 748 */ 749 m = NULL; 750 goto drop; 751 } else { 752 /* 753 * We have created a 754 * full-blown connection. 755 */ 756 tp = NULL; 757 inp = sotoinpcb(so); 758 tp = intotcpcb(inp); 759 if (tp == NULL) 760 goto badsyn; /*XXX*/ 761 762 } 763 break; 764 765 default: 766 /* 767 * None of RST, SYN or ACK was set. 768 * This is an invalid packet for a 769 * TCB in LISTEN state. Send a RST. 770 */ 771 goto badsyn; 772 773 case TH_SYN: 774 /* 775 * Received a SYN. 776 */ 777 #ifdef INET6 778 /* 779 * If deprecated address is forbidden, we do 780 * not accept SYN to deprecated interface 781 * address to prevent any new inbound 782 * connection from getting established. 783 * When we do not accept SYN, we send a TCP 784 * RST, with deprecated source address (instead 785 * of dropping it). We compromise it as it is 786 * much better for peer to send a RST, and 787 * RST will be the final packet for the 788 * exchange. 789 * 790 * If we do not forbid deprecated addresses, we 791 * accept the SYN packet. RFC2462 does not 792 * suggest dropping SYN in this case. 793 * If we decipher RFC2462 5.5.4, it says like 794 * this: 795 * 1. use of deprecated addr with existing 796 * communication is okay - "SHOULD continue 797 * to be used" 798 * 2. use of it with new communication: 799 * (2a) "SHOULD NOT be used if alternate 800 * address with sufficient scope is 801 * available" 802 * (2b) nothing mentioned otherwise. 803 * Here we fall into (2b) case as we have no 804 * choice in our source address selection - we 805 * must obey the peer. 806 * 807 * The wording in RFC2462 is confusing, and 808 * there are multiple description text for 809 * deprecated address handling - worse, they 810 * are not exactly the same. I believe 5.5.4 811 * is the best one, so we follow 5.5.4. 812 */ 813 if (ip6 && !ip6_use_deprecated) { 814 struct in6_ifaddr *ia6; 815 816 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 817 &ip6->ip6_dst)) && 818 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 819 tp = NULL; 820 goto dropwithreset; 821 } 822 } 823 #endif 824 825 /* 826 * LISTEN socket received a SYN 827 * from itself? This can't possibly 828 * be valid; drop the packet. 829 */ 830 if (th->th_dport == th->th_sport) { 831 switch (af) { 832 #ifdef INET6 833 case AF_INET6: 834 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 835 &ip6->ip6_dst)) { 836 tcpstat.tcps_badsyn++; 837 goto drop; 838 } 839 break; 840 #endif /* INET6 */ 841 case AF_INET: 842 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) { 843 tcpstat.tcps_badsyn++; 844 goto drop; 845 } 846 break; 847 } 848 } 849 850 /* 851 * SYN looks ok; create compressed TCP 852 * state for it. 853 */ 854 if (so->so_qlen > so->so_qlimit || 855 syn_cache_add(&src.sa, &dst.sa, th, iphlen, 856 so, m, optp, optlen, &opti, reuse) == -1) { 857 tcpstat.tcps_dropsyn++; 858 goto drop; 859 } 860 return; 861 } 862 } 863 } 864 865 #ifdef DIAGNOSTIC 866 /* 867 * Should not happen now that all embryonic connections 868 * are handled with compressed state. 869 */ 870 if (tp->t_state == TCPS_LISTEN) 871 panic("tcp_input: TCPS_LISTEN"); 872 #endif 873 874 #if NPF > 0 875 if (m->m_pkthdr.pf.statekey && !m->m_pkthdr.pf.statekey->inp && 876 !inp->inp_pf_sk) { 877 m->m_pkthdr.pf.statekey->inp = inp; 878 inp->inp_pf_sk = m->m_pkthdr.pf.statekey; 879 } 880 /* The statekey has finished finding the inp, it is no longer needed. */ 881 m->m_pkthdr.pf.statekey = NULL; 882 #endif 883 884 #ifdef IPSEC 885 /* Find most recent IPsec tag */ 886 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL); 887 if (mtag != NULL) { 888 tdbi = (struct tdb_ident *)(mtag + 1); 889 tdb = gettdb(tdbi->rdomain, tdbi->spi, 890 &tdbi->dst, tdbi->proto); 891 } else 892 tdb = NULL; 893 ipsp_spd_lookup(m, af, iphlen, &error, IPSP_DIRECTION_IN, 894 tdb, inp, 0); 895 if (error) { 896 tcpstat.tcps_rcvnosec++; 897 goto drop; 898 } 899 #endif /* IPSEC */ 900 901 /* 902 * Segment received on connection. 903 * Reset idle time and keep-alive timer. 904 */ 905 tp->t_rcvtime = tcp_now; 906 if (TCPS_HAVEESTABLISHED(tp->t_state)) 907 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 908 909 #ifdef TCP_SACK 910 if (tp->sack_enable) 911 tcp_del_sackholes(tp, th); /* Delete stale SACK holes */ 912 #endif /* TCP_SACK */ 913 914 /* 915 * Process options. 916 */ 917 #ifdef TCP_SIGNATURE 918 if (optp || (tp->t_flags & TF_SIGNATURE)) 919 #else 920 if (optp) 921 #endif 922 if (tcp_dooptions(tp, optp, optlen, th, m, iphlen, &opti, 923 m->m_pkthdr.ph_rtableid)) 924 goto drop; 925 926 if (opti.ts_present && opti.ts_ecr) { 927 int rtt_test; 928 929 /* subtract out the tcp timestamp modulator */ 930 opti.ts_ecr -= tp->ts_modulate; 931 932 /* make sure ts_ecr is sensible */ 933 rtt_test = tcp_now - opti.ts_ecr; 934 if (rtt_test < 0 || rtt_test > TCP_RTT_MAX) 935 opti.ts_ecr = 0; 936 } 937 938 #ifdef TCP_ECN 939 /* if congestion experienced, set ECE bit in subsequent packets. */ 940 if ((iptos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) { 941 tp->t_flags |= TF_RCVD_CE; 942 tcpstat.tcps_ecn_rcvce++; 943 } 944 #endif 945 /* 946 * Header prediction: check for the two common cases 947 * of a uni-directional data xfer. If the packet has 948 * no control flags, is in-sequence, the window didn't 949 * change and we're not retransmitting, it's a 950 * candidate. If the length is zero and the ack moved 951 * forward, we're the sender side of the xfer. Just 952 * free the data acked & wake any higher level process 953 * that was blocked waiting for space. If the length 954 * is non-zero and the ack didn't move, we're the 955 * receiver side. If we're getting packets in-order 956 * (the reassembly queue is empty), add the data to 957 * the socket buffer and note that we need a delayed ack. 958 */ 959 if (tp->t_state == TCPS_ESTABLISHED && 960 #ifdef TCP_ECN 961 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) == TH_ACK && 962 #else 963 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 964 #endif 965 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 966 th->th_seq == tp->rcv_nxt && 967 tiwin && tiwin == tp->snd_wnd && 968 tp->snd_nxt == tp->snd_max) { 969 970 /* 971 * If last ACK falls within this segment's sequence numbers, 972 * record the timestamp. 973 * Fix from Braden, see Stevens p. 870 974 */ 975 if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 976 tp->ts_recent_age = tcp_now; 977 tp->ts_recent = opti.ts_val; 978 } 979 980 if (tlen == 0) { 981 if (SEQ_GT(th->th_ack, tp->snd_una) && 982 SEQ_LEQ(th->th_ack, tp->snd_max) && 983 tp->snd_cwnd >= tp->snd_wnd && 984 tp->t_dupacks == 0) { 985 /* 986 * this is a pure ack for outstanding data. 987 */ 988 ++tcpstat.tcps_predack; 989 if (opti.ts_present && opti.ts_ecr) 990 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr); 991 else if (tp->t_rtttime && 992 SEQ_GT(th->th_ack, tp->t_rtseq)) 993 tcp_xmit_timer(tp, 994 tcp_now - tp->t_rtttime); 995 acked = th->th_ack - tp->snd_una; 996 tcpstat.tcps_rcvackpack++; 997 tcpstat.tcps_rcvackbyte += acked; 998 ND6_HINT(tp); 999 sbdrop(&so->so_snd, acked); 1000 1001 /* 1002 * If we had a pending ICMP message that 1003 * refers to data that have just been 1004 * acknowledged, disregard the recorded ICMP 1005 * message. 1006 */ 1007 if ((tp->t_flags & TF_PMTUD_PEND) && 1008 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 1009 tp->t_flags &= ~TF_PMTUD_PEND; 1010 1011 /* 1012 * Keep track of the largest chunk of data 1013 * acknowledged since last PMTU update 1014 */ 1015 if (tp->t_pmtud_mss_acked < acked) 1016 tp->t_pmtud_mss_acked = acked; 1017 1018 tp->snd_una = th->th_ack; 1019 #if defined(TCP_SACK) || defined(TCP_ECN) 1020 /* 1021 * We want snd_last to track snd_una so 1022 * as to avoid sequence wraparound problems 1023 * for very large transfers. 1024 */ 1025 #ifdef TCP_ECN 1026 if (SEQ_GT(tp->snd_una, tp->snd_last)) 1027 #endif 1028 tp->snd_last = tp->snd_una; 1029 #endif /* TCP_SACK */ 1030 #if defined(TCP_SACK) && defined(TCP_FACK) 1031 tp->snd_fack = tp->snd_una; 1032 tp->retran_data = 0; 1033 #endif /* TCP_FACK */ 1034 m_freem(m); 1035 1036 /* 1037 * If all outstanding data are acked, stop 1038 * retransmit timer, otherwise restart timer 1039 * using current (possibly backed-off) value. 1040 * If process is waiting for space, 1041 * wakeup/selwakeup/signal. If data 1042 * are ready to send, let tcp_output 1043 * decide between more output or persist. 1044 */ 1045 if (tp->snd_una == tp->snd_max) 1046 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1047 else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1048 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1049 1050 tcp_update_sndspace(tp); 1051 if (sb_notify(&so->so_snd)) { 1052 tp->t_flags |= TF_BLOCKOUTPUT; 1053 sowwakeup(so); 1054 tp->t_flags &= ~TF_BLOCKOUTPUT; 1055 } 1056 if (so->so_snd.sb_cc || 1057 tp->t_flags & TF_NEEDOUTPUT) 1058 (void) tcp_output(tp); 1059 return; 1060 } 1061 } else if (th->th_ack == tp->snd_una && 1062 TAILQ_EMPTY(&tp->t_segq) && 1063 tlen <= sbspace(&so->so_rcv)) { 1064 /* 1065 * This is a pure, in-sequence data packet 1066 * with nothing on the reassembly queue and 1067 * we have enough buffer space to take it. 1068 */ 1069 #ifdef TCP_SACK 1070 /* Clean receiver SACK report if present */ 1071 if (tp->sack_enable && tp->rcv_numsacks) 1072 tcp_clean_sackreport(tp); 1073 #endif /* TCP_SACK */ 1074 ++tcpstat.tcps_preddat; 1075 tp->rcv_nxt += tlen; 1076 tcpstat.tcps_rcvpack++; 1077 tcpstat.tcps_rcvbyte += tlen; 1078 ND6_HINT(tp); 1079 1080 TCP_SETUP_ACK(tp, tiflags, m); 1081 /* 1082 * Drop TCP, IP headers and TCP options then add data 1083 * to socket buffer. 1084 */ 1085 if (so->so_state & SS_CANTRCVMORE) 1086 m_freem(m); 1087 else { 1088 if (opti.ts_present && opti.ts_ecr) { 1089 if (tp->rfbuf_ts < opti.ts_ecr && 1090 opti.ts_ecr - tp->rfbuf_ts < hz) { 1091 tcp_update_rcvspace(tp); 1092 /* Start over with next RTT. */ 1093 tp->rfbuf_cnt = 0; 1094 tp->rfbuf_ts = 0; 1095 } else 1096 tp->rfbuf_cnt += tlen; 1097 } 1098 m_adj(m, iphlen + off); 1099 sbappendstream(&so->so_rcv, m); 1100 } 1101 tp->t_flags |= TF_BLOCKOUTPUT; 1102 sorwakeup(so); 1103 tp->t_flags &= ~TF_BLOCKOUTPUT; 1104 if (tp->t_flags & (TF_ACKNOW|TF_NEEDOUTPUT)) 1105 (void) tcp_output(tp); 1106 return; 1107 } 1108 } 1109 1110 /* 1111 * Compute mbuf offset to TCP data segment. 1112 */ 1113 hdroptlen = iphlen + off; 1114 1115 /* 1116 * Calculate amount of space in receive window, 1117 * and then do TCP input processing. 1118 * Receive window is amount of space in rcv queue, 1119 * but not less than advertised window. 1120 */ 1121 { int win; 1122 1123 win = sbspace(&so->so_rcv); 1124 if (win < 0) 1125 win = 0; 1126 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1127 } 1128 1129 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1130 tp->rfbuf_cnt = 0; 1131 tp->rfbuf_ts = 0; 1132 1133 switch (tp->t_state) { 1134 1135 /* 1136 * If the state is SYN_RECEIVED: 1137 * if seg contains SYN/ACK, send an RST. 1138 * if seg contains an ACK, but not for our SYN/ACK, send an RST 1139 */ 1140 1141 case TCPS_SYN_RECEIVED: 1142 if (tiflags & TH_ACK) { 1143 if (tiflags & TH_SYN) { 1144 tcpstat.tcps_badsyn++; 1145 goto dropwithreset; 1146 } 1147 if (SEQ_LEQ(th->th_ack, tp->snd_una) || 1148 SEQ_GT(th->th_ack, tp->snd_max)) 1149 goto dropwithreset; 1150 } 1151 break; 1152 1153 /* 1154 * If the state is SYN_SENT: 1155 * if seg contains an ACK, but not for our SYN, drop the input. 1156 * if seg contains a RST, then drop the connection. 1157 * if seg does not contain SYN, then drop it. 1158 * Otherwise this is an acceptable SYN segment 1159 * initialize tp->rcv_nxt and tp->irs 1160 * if seg contains ack then advance tp->snd_una 1161 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1162 * arrange for segment to be acked (eventually) 1163 * continue processing rest of data/controls, beginning with URG 1164 */ 1165 case TCPS_SYN_SENT: 1166 if ((tiflags & TH_ACK) && 1167 (SEQ_LEQ(th->th_ack, tp->iss) || 1168 SEQ_GT(th->th_ack, tp->snd_max))) 1169 goto dropwithreset; 1170 if (tiflags & TH_RST) { 1171 #ifdef TCP_ECN 1172 /* if ECN is enabled, fall back to non-ecn at rexmit */ 1173 if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN)) 1174 goto drop; 1175 #endif 1176 if (tiflags & TH_ACK) 1177 tp = tcp_drop(tp, ECONNREFUSED); 1178 goto drop; 1179 } 1180 if ((tiflags & TH_SYN) == 0) 1181 goto drop; 1182 if (tiflags & TH_ACK) { 1183 tp->snd_una = th->th_ack; 1184 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1185 tp->snd_nxt = tp->snd_una; 1186 } 1187 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1188 tp->irs = th->th_seq; 1189 tcp_mss(tp, opti.maxseg); 1190 /* Reset initial window to 1 segment for retransmit */ 1191 if (tp->t_rxtshift > 0) 1192 tp->snd_cwnd = tp->t_maxseg; 1193 tcp_rcvseqinit(tp); 1194 tp->t_flags |= TF_ACKNOW; 1195 #ifdef TCP_SACK 1196 /* 1197 * If we've sent a SACK_PERMITTED option, and the peer 1198 * also replied with one, then TF_SACK_PERMIT should have 1199 * been set in tcp_dooptions(). If it was not, disable SACKs. 1200 */ 1201 if (tp->sack_enable) 1202 tp->sack_enable = tp->t_flags & TF_SACK_PERMIT; 1203 #endif 1204 #ifdef TCP_ECN 1205 /* 1206 * if ECE is set but CWR is not set for SYN-ACK, or 1207 * both ECE and CWR are set for simultaneous open, 1208 * peer is ECN capable. 1209 */ 1210 if (tcp_do_ecn) { 1211 switch (tiflags & (TH_ACK|TH_ECE|TH_CWR)) { 1212 case TH_ACK|TH_ECE: 1213 case TH_ECE|TH_CWR: 1214 tp->t_flags |= TF_ECN_PERMIT; 1215 tiflags &= ~(TH_ECE|TH_CWR); 1216 tcpstat.tcps_ecn_accepts++; 1217 } 1218 } 1219 #endif 1220 1221 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 1222 tcpstat.tcps_connects++; 1223 soisconnected(so); 1224 tp->t_state = TCPS_ESTABLISHED; 1225 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1226 /* Do window scaling on this connection? */ 1227 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1228 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1229 tp->snd_scale = tp->requested_s_scale; 1230 tp->rcv_scale = tp->request_r_scale; 1231 } 1232 tcp_flush_queue(tp); 1233 1234 /* 1235 * if we didn't have to retransmit the SYN, 1236 * use its rtt as our initial srtt & rtt var. 1237 */ 1238 if (tp->t_rtttime) 1239 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1240 /* 1241 * Since new data was acked (the SYN), open the 1242 * congestion window by one MSS. We do this 1243 * here, because we won't go through the normal 1244 * ACK processing below. And since this is the 1245 * start of the connection, we know we are in 1246 * the exponential phase of slow-start. 1247 */ 1248 tp->snd_cwnd += tp->t_maxseg; 1249 } else 1250 tp->t_state = TCPS_SYN_RECEIVED; 1251 1252 #if 0 1253 trimthenstep6: 1254 #endif 1255 /* 1256 * Advance th->th_seq to correspond to first data byte. 1257 * If data, trim to stay within window, 1258 * dropping FIN if necessary. 1259 */ 1260 th->th_seq++; 1261 if (tlen > tp->rcv_wnd) { 1262 todrop = tlen - tp->rcv_wnd; 1263 m_adj(m, -todrop); 1264 tlen = tp->rcv_wnd; 1265 tiflags &= ~TH_FIN; 1266 tcpstat.tcps_rcvpackafterwin++; 1267 tcpstat.tcps_rcvbyteafterwin += todrop; 1268 } 1269 tp->snd_wl1 = th->th_seq - 1; 1270 tp->rcv_up = th->th_seq; 1271 goto step6; 1272 /* 1273 * If a new connection request is received while in TIME_WAIT, 1274 * drop the old connection and start over if the if the 1275 * timestamp or the sequence numbers are above the previous 1276 * ones. 1277 */ 1278 case TCPS_TIME_WAIT: 1279 if (((tiflags & (TH_SYN|TH_ACK)) == TH_SYN) && 1280 ((opti.ts_present && 1281 TSTMP_LT(tp->ts_recent, opti.ts_val)) || 1282 SEQ_GT(th->th_seq, tp->rcv_nxt))) { 1283 #if NPF > 0 1284 /* 1285 * The socket will be recreated but the new state 1286 * has already been linked to the socket. Remove the 1287 * link between old socket and new state. 1288 */ 1289 if (inp->inp_pf_sk) { 1290 inp->inp_pf_sk->inp = NULL; 1291 inp->inp_pf_sk = NULL; 1292 } 1293 #endif 1294 /* 1295 * Advance the iss by at least 32768, but 1296 * clear the msb in order to make sure 1297 * that SEG_LT(snd_nxt, iss). 1298 */ 1299 iss = tp->snd_nxt + 1300 ((arc4random() & 0x7fffffff) | 0x8000); 1301 reuse = &iss; 1302 tp = tcp_close(tp); 1303 inp = NULL; 1304 goto findpcb; 1305 } 1306 } 1307 1308 /* 1309 * States other than LISTEN or SYN_SENT. 1310 * First check timestamp, if present. 1311 * Then check that at least some bytes of segment are within 1312 * receive window. If segment begins before rcv_nxt, 1313 * drop leading data (and SYN); if nothing left, just ack. 1314 * 1315 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1316 * and it's less than opti.ts_recent, drop it. 1317 */ 1318 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1319 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1320 1321 /* Check to see if ts_recent is over 24 days old. */ 1322 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1323 /* 1324 * Invalidate ts_recent. If this segment updates 1325 * ts_recent, the age will be reset later and ts_recent 1326 * will get a valid value. If it does not, setting 1327 * ts_recent to zero will at least satisfy the 1328 * requirement that zero be placed in the timestamp 1329 * echo reply when ts_recent isn't valid. The 1330 * age isn't reset until we get a valid ts_recent 1331 * because we don't want out-of-order segments to be 1332 * dropped when ts_recent is old. 1333 */ 1334 tp->ts_recent = 0; 1335 } else { 1336 tcpstat.tcps_rcvduppack++; 1337 tcpstat.tcps_rcvdupbyte += tlen; 1338 tcpstat.tcps_pawsdrop++; 1339 goto dropafterack; 1340 } 1341 } 1342 1343 todrop = tp->rcv_nxt - th->th_seq; 1344 if (todrop > 0) { 1345 if (tiflags & TH_SYN) { 1346 tiflags &= ~TH_SYN; 1347 th->th_seq++; 1348 if (th->th_urp > 1) 1349 th->th_urp--; 1350 else 1351 tiflags &= ~TH_URG; 1352 todrop--; 1353 } 1354 if (todrop > tlen || 1355 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1356 /* 1357 * Any valid FIN must be to the left of the 1358 * window. At this point, FIN must be a 1359 * duplicate or out-of-sequence, so drop it. 1360 */ 1361 tiflags &= ~TH_FIN; 1362 /* 1363 * Send ACK to resynchronize, and drop any data, 1364 * but keep on processing for RST or ACK. 1365 */ 1366 tp->t_flags |= TF_ACKNOW; 1367 tcpstat.tcps_rcvdupbyte += todrop = tlen; 1368 tcpstat.tcps_rcvduppack++; 1369 } else { 1370 tcpstat.tcps_rcvpartduppack++; 1371 tcpstat.tcps_rcvpartdupbyte += todrop; 1372 } 1373 hdroptlen += todrop; /* drop from head afterwards */ 1374 th->th_seq += todrop; 1375 tlen -= todrop; 1376 if (th->th_urp > todrop) 1377 th->th_urp -= todrop; 1378 else { 1379 tiflags &= ~TH_URG; 1380 th->th_urp = 0; 1381 } 1382 } 1383 1384 /* 1385 * If new data are received on a connection after the 1386 * user processes are gone, then RST the other end. 1387 */ 1388 if ((so->so_state & SS_NOFDREF) && 1389 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1390 tp = tcp_close(tp); 1391 tcpstat.tcps_rcvafterclose++; 1392 goto dropwithreset; 1393 } 1394 1395 /* 1396 * If segment ends after window, drop trailing data 1397 * (and PUSH and FIN); if nothing left, just ACK. 1398 */ 1399 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1400 if (todrop > 0) { 1401 tcpstat.tcps_rcvpackafterwin++; 1402 if (todrop >= tlen) { 1403 tcpstat.tcps_rcvbyteafterwin += tlen; 1404 /* 1405 * If window is closed can only take segments at 1406 * window edge, and have to drop data and PUSH from 1407 * incoming segments. Continue processing, but 1408 * remember to ack. Otherwise, drop segment 1409 * and ack. 1410 */ 1411 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1412 tp->t_flags |= TF_ACKNOW; 1413 tcpstat.tcps_rcvwinprobe++; 1414 } else 1415 goto dropafterack; 1416 } else 1417 tcpstat.tcps_rcvbyteafterwin += todrop; 1418 m_adj(m, -todrop); 1419 tlen -= todrop; 1420 tiflags &= ~(TH_PUSH|TH_FIN); 1421 } 1422 1423 /* 1424 * If last ACK falls within this segment's sequence numbers, 1425 * record its timestamp if it's more recent. 1426 * Cf fix from Braden, see Stevens p. 870 1427 */ 1428 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1429 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1430 if (SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1431 ((tiflags & (TH_SYN|TH_FIN)) != 0))) 1432 tp->ts_recent = opti.ts_val; 1433 else 1434 tp->ts_recent = 0; 1435 tp->ts_recent_age = tcp_now; 1436 } 1437 1438 /* 1439 * If the RST bit is set examine the state: 1440 * SYN_RECEIVED STATE: 1441 * If passive open, return to LISTEN state. 1442 * If active open, inform user that connection was refused. 1443 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1444 * Inform user that connection was reset, and close tcb. 1445 * CLOSING, LAST_ACK, TIME_WAIT STATES 1446 * Close the tcb. 1447 */ 1448 if (tiflags & TH_RST) { 1449 if (th->th_seq != tp->last_ack_sent && 1450 th->th_seq != tp->rcv_nxt && 1451 th->th_seq != (tp->rcv_nxt + 1)) 1452 goto drop; 1453 1454 switch (tp->t_state) { 1455 case TCPS_SYN_RECEIVED: 1456 #ifdef TCP_ECN 1457 /* if ECN is enabled, fall back to non-ecn at rexmit */ 1458 if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN)) 1459 goto drop; 1460 #endif 1461 so->so_error = ECONNREFUSED; 1462 goto close; 1463 1464 case TCPS_ESTABLISHED: 1465 case TCPS_FIN_WAIT_1: 1466 case TCPS_FIN_WAIT_2: 1467 case TCPS_CLOSE_WAIT: 1468 so->so_error = ECONNRESET; 1469 close: 1470 tp->t_state = TCPS_CLOSED; 1471 tcpstat.tcps_drops++; 1472 tp = tcp_close(tp); 1473 goto drop; 1474 case TCPS_CLOSING: 1475 case TCPS_LAST_ACK: 1476 case TCPS_TIME_WAIT: 1477 tp = tcp_close(tp); 1478 goto drop; 1479 } 1480 } 1481 1482 /* 1483 * If a SYN is in the window, then this is an 1484 * error and we ACK and drop the packet. 1485 */ 1486 if (tiflags & TH_SYN) 1487 goto dropafterack_ratelim; 1488 1489 /* 1490 * If the ACK bit is off we drop the segment and return. 1491 */ 1492 if ((tiflags & TH_ACK) == 0) { 1493 if (tp->t_flags & TF_ACKNOW) 1494 goto dropafterack; 1495 else 1496 goto drop; 1497 } 1498 1499 /* 1500 * Ack processing. 1501 */ 1502 switch (tp->t_state) { 1503 1504 /* 1505 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1506 * ESTABLISHED state and continue processing. 1507 * The ACK was checked above. 1508 */ 1509 case TCPS_SYN_RECEIVED: 1510 tcpstat.tcps_connects++; 1511 soisconnected(so); 1512 tp->t_state = TCPS_ESTABLISHED; 1513 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1514 /* Do window scaling? */ 1515 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1516 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1517 tp->snd_scale = tp->requested_s_scale; 1518 tp->rcv_scale = tp->request_r_scale; 1519 tiwin = th->th_win << tp->snd_scale; 1520 } 1521 tcp_flush_queue(tp); 1522 tp->snd_wl1 = th->th_seq - 1; 1523 /* fall into ... */ 1524 1525 /* 1526 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1527 * ACKs. If the ack is in the range 1528 * tp->snd_una < th->th_ack <= tp->snd_max 1529 * then advance tp->snd_una to th->th_ack and drop 1530 * data from the retransmission queue. If this ACK reflects 1531 * more up to date window information we update our window information. 1532 */ 1533 case TCPS_ESTABLISHED: 1534 case TCPS_FIN_WAIT_1: 1535 case TCPS_FIN_WAIT_2: 1536 case TCPS_CLOSE_WAIT: 1537 case TCPS_CLOSING: 1538 case TCPS_LAST_ACK: 1539 case TCPS_TIME_WAIT: 1540 #ifdef TCP_ECN 1541 /* 1542 * if we receive ECE and are not already in recovery phase, 1543 * reduce cwnd by half but don't slow-start. 1544 * advance snd_last to snd_max not to reduce cwnd again 1545 * until all outstanding packets are acked. 1546 */ 1547 if (tcp_do_ecn && (tiflags & TH_ECE)) { 1548 if ((tp->t_flags & TF_ECN_PERMIT) && 1549 SEQ_GEQ(tp->snd_una, tp->snd_last)) { 1550 u_int win; 1551 1552 win = min(tp->snd_wnd, tp->snd_cwnd) / tp->t_maxseg; 1553 if (win > 1) { 1554 tp->snd_ssthresh = win / 2 * tp->t_maxseg; 1555 tp->snd_cwnd = tp->snd_ssthresh; 1556 tp->snd_last = tp->snd_max; 1557 tp->t_flags |= TF_SEND_CWR; 1558 tcpstat.tcps_cwr_ecn++; 1559 } 1560 } 1561 tcpstat.tcps_ecn_rcvece++; 1562 } 1563 /* 1564 * if we receive CWR, we know that the peer has reduced 1565 * its congestion window. stop sending ecn-echo. 1566 */ 1567 if ((tiflags & TH_CWR)) { 1568 tp->t_flags &= ~TF_RCVD_CE; 1569 tcpstat.tcps_ecn_rcvcwr++; 1570 } 1571 #endif /* TCP_ECN */ 1572 1573 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1574 /* 1575 * Duplicate/old ACK processing. 1576 * Increments t_dupacks: 1577 * Pure duplicate (same seq/ack/window, no data) 1578 * Doesn't affect t_dupacks: 1579 * Data packets. 1580 * Normal window updates (window opens) 1581 * Resets t_dupacks: 1582 * New data ACKed. 1583 * Window shrinks 1584 * Old ACK 1585 */ 1586 if (tlen) { 1587 /* Drop very old ACKs unless th_seq matches */ 1588 if (th->th_seq != tp->rcv_nxt && 1589 SEQ_LT(th->th_ack, 1590 tp->snd_una - tp->max_sndwnd)) { 1591 tcpstat.tcps_rcvacktooold++; 1592 goto drop; 1593 } 1594 break; 1595 } 1596 /* 1597 * If we get an old ACK, there is probably packet 1598 * reordering going on. Be conservative and reset 1599 * t_dupacks so that we are less aggressive in 1600 * doing a fast retransmit. 1601 */ 1602 if (th->th_ack != tp->snd_una) { 1603 tp->t_dupacks = 0; 1604 break; 1605 } 1606 if (tiwin == tp->snd_wnd) { 1607 tcpstat.tcps_rcvdupack++; 1608 /* 1609 * If we have outstanding data (other than 1610 * a window probe), this is a completely 1611 * duplicate ack (ie, window info didn't 1612 * change), the ack is the biggest we've 1613 * seen and we've seen exactly our rexmt 1614 * threshold of them, assume a packet 1615 * has been dropped and retransmit it. 1616 * Kludge snd_nxt & the congestion 1617 * window so we send only this one 1618 * packet. 1619 * 1620 * We know we're losing at the current 1621 * window size so do congestion avoidance 1622 * (set ssthresh to half the current window 1623 * and pull our congestion window back to 1624 * the new ssthresh). 1625 * 1626 * Dup acks mean that packets have left the 1627 * network (they're now cached at the receiver) 1628 * so bump cwnd by the amount in the receiver 1629 * to keep a constant cwnd packets in the 1630 * network. 1631 */ 1632 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0) 1633 tp->t_dupacks = 0; 1634 #if defined(TCP_SACK) && defined(TCP_FACK) 1635 /* 1636 * In FACK, can enter fast rec. if the receiver 1637 * reports a reass. queue longer than 3 segs. 1638 */ 1639 else if (++tp->t_dupacks == tcprexmtthresh || 1640 ((SEQ_GT(tp->snd_fack, tcprexmtthresh * 1641 tp->t_maxseg + tp->snd_una)) && 1642 SEQ_GT(tp->snd_una, tp->snd_last))) { 1643 #else 1644 else if (++tp->t_dupacks == tcprexmtthresh) { 1645 #endif /* TCP_FACK */ 1646 tcp_seq onxt = tp->snd_nxt; 1647 u_long win = 1648 ulmin(tp->snd_wnd, tp->snd_cwnd) / 1649 2 / tp->t_maxseg; 1650 1651 #if defined(TCP_SACK) || defined(TCP_ECN) 1652 if (SEQ_LT(th->th_ack, tp->snd_last)){ 1653 /* 1654 * False fast retx after 1655 * timeout. Do not cut window. 1656 */ 1657 tp->t_dupacks = 0; 1658 goto drop; 1659 } 1660 #endif 1661 if (win < 2) 1662 win = 2; 1663 tp->snd_ssthresh = win * tp->t_maxseg; 1664 #ifdef TCP_SACK 1665 tp->snd_last = tp->snd_max; 1666 if (tp->sack_enable) { 1667 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1668 tp->t_rtttime = 0; 1669 #ifdef TCP_ECN 1670 tp->t_flags |= TF_SEND_CWR; 1671 #endif 1672 tcpstat.tcps_cwr_frecovery++; 1673 tcpstat.tcps_sack_recovery_episode++; 1674 #if defined(TCP_SACK) && defined(TCP_FACK) 1675 tp->t_dupacks = tcprexmtthresh; 1676 (void) tcp_output(tp); 1677 /* 1678 * During FR, snd_cwnd is held 1679 * constant for FACK. 1680 */ 1681 tp->snd_cwnd = tp->snd_ssthresh; 1682 #else 1683 /* 1684 * tcp_output() will send 1685 * oldest SACK-eligible rtx. 1686 */ 1687 (void) tcp_output(tp); 1688 tp->snd_cwnd = tp->snd_ssthresh+ 1689 tp->t_maxseg * tp->t_dupacks; 1690 #endif /* TCP_FACK */ 1691 goto drop; 1692 } 1693 #endif /* TCP_SACK */ 1694 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1695 tp->t_rtttime = 0; 1696 tp->snd_nxt = th->th_ack; 1697 tp->snd_cwnd = tp->t_maxseg; 1698 #ifdef TCP_ECN 1699 tp->t_flags |= TF_SEND_CWR; 1700 #endif 1701 tcpstat.tcps_cwr_frecovery++; 1702 tcpstat.tcps_sndrexmitfast++; 1703 (void) tcp_output(tp); 1704 1705 tp->snd_cwnd = tp->snd_ssthresh + 1706 tp->t_maxseg * tp->t_dupacks; 1707 if (SEQ_GT(onxt, tp->snd_nxt)) 1708 tp->snd_nxt = onxt; 1709 goto drop; 1710 } else if (tp->t_dupacks > tcprexmtthresh) { 1711 #if defined(TCP_SACK) && defined(TCP_FACK) 1712 /* 1713 * while (awnd < cwnd) 1714 * sendsomething(); 1715 */ 1716 if (tp->sack_enable) { 1717 if (tp->snd_awnd < tp->snd_cwnd) 1718 tcp_output(tp); 1719 goto drop; 1720 } 1721 #endif /* TCP_FACK */ 1722 tp->snd_cwnd += tp->t_maxseg; 1723 (void) tcp_output(tp); 1724 goto drop; 1725 } 1726 } else if (tiwin < tp->snd_wnd) { 1727 /* 1728 * The window was retracted! Previous dup 1729 * ACKs may have been due to packets arriving 1730 * after the shrunken window, not a missing 1731 * packet, so play it safe and reset t_dupacks 1732 */ 1733 tp->t_dupacks = 0; 1734 } 1735 break; 1736 } 1737 /* 1738 * If the congestion window was inflated to account 1739 * for the other side's cached packets, retract it. 1740 */ 1741 #if defined(TCP_SACK) 1742 if (tp->sack_enable) { 1743 if (tp->t_dupacks >= tcprexmtthresh) { 1744 /* Check for a partial ACK */ 1745 if (tcp_sack_partialack(tp, th)) { 1746 #if defined(TCP_SACK) && defined(TCP_FACK) 1747 /* Force call to tcp_output */ 1748 if (tp->snd_awnd < tp->snd_cwnd) 1749 tp->t_flags |= TF_NEEDOUTPUT; 1750 #else 1751 tp->snd_cwnd += tp->t_maxseg; 1752 tp->t_flags |= TF_NEEDOUTPUT; 1753 #endif /* TCP_FACK */ 1754 } else { 1755 /* Out of fast recovery */ 1756 tp->snd_cwnd = tp->snd_ssthresh; 1757 if (tcp_seq_subtract(tp->snd_max, 1758 th->th_ack) < tp->snd_ssthresh) 1759 tp->snd_cwnd = 1760 tcp_seq_subtract(tp->snd_max, 1761 th->th_ack); 1762 tp->t_dupacks = 0; 1763 #if defined(TCP_SACK) && defined(TCP_FACK) 1764 if (SEQ_GT(th->th_ack, tp->snd_fack)) 1765 tp->snd_fack = th->th_ack; 1766 #endif /* TCP_FACK */ 1767 } 1768 } 1769 } else { 1770 if (tp->t_dupacks >= tcprexmtthresh && 1771 !tcp_newreno(tp, th)) { 1772 /* Out of fast recovery */ 1773 tp->snd_cwnd = tp->snd_ssthresh; 1774 if (tcp_seq_subtract(tp->snd_max, th->th_ack) < 1775 tp->snd_ssthresh) 1776 tp->snd_cwnd = 1777 tcp_seq_subtract(tp->snd_max, 1778 th->th_ack); 1779 tp->t_dupacks = 0; 1780 } 1781 } 1782 if (tp->t_dupacks < tcprexmtthresh) 1783 tp->t_dupacks = 0; 1784 #else /* else no TCP_SACK */ 1785 if (tp->t_dupacks >= tcprexmtthresh && 1786 tp->snd_cwnd > tp->snd_ssthresh) 1787 tp->snd_cwnd = tp->snd_ssthresh; 1788 tp->t_dupacks = 0; 1789 #endif 1790 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1791 tcpstat.tcps_rcvacktoomuch++; 1792 goto dropafterack_ratelim; 1793 } 1794 acked = th->th_ack - tp->snd_una; 1795 tcpstat.tcps_rcvackpack++; 1796 tcpstat.tcps_rcvackbyte += acked; 1797 1798 /* 1799 * If we have a timestamp reply, update smoothed 1800 * round trip time. If no timestamp is present but 1801 * transmit timer is running and timed sequence 1802 * number was acked, update smoothed round trip time. 1803 * Since we now have an rtt measurement, cancel the 1804 * timer backoff (cf., Phil Karn's retransmit alg.). 1805 * Recompute the initial retransmit timer. 1806 */ 1807 if (opti.ts_present && opti.ts_ecr) 1808 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr); 1809 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1810 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1811 1812 /* 1813 * If all outstanding data is acked, stop retransmit 1814 * timer and remember to restart (more output or persist). 1815 * If there is more data to be acked, restart retransmit 1816 * timer, using current (possibly backed-off) value. 1817 */ 1818 if (th->th_ack == tp->snd_max) { 1819 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1820 tp->t_flags |= TF_NEEDOUTPUT; 1821 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1822 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1823 /* 1824 * When new data is acked, open the congestion window. 1825 * If the window gives us less than ssthresh packets 1826 * in flight, open exponentially (maxseg per packet). 1827 * Otherwise open linearly: maxseg per window 1828 * (maxseg^2 / cwnd per packet). 1829 */ 1830 { 1831 u_int cw = tp->snd_cwnd; 1832 u_int incr = tp->t_maxseg; 1833 1834 if (cw > tp->snd_ssthresh) 1835 incr = incr * incr / cw; 1836 #if defined (TCP_SACK) 1837 if (tp->t_dupacks < tcprexmtthresh) 1838 #endif 1839 tp->snd_cwnd = ulmin(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1840 } 1841 ND6_HINT(tp); 1842 if (acked > so->so_snd.sb_cc) { 1843 tp->snd_wnd -= so->so_snd.sb_cc; 1844 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1845 ourfinisacked = 1; 1846 } else { 1847 sbdrop(&so->so_snd, acked); 1848 tp->snd_wnd -= acked; 1849 ourfinisacked = 0; 1850 } 1851 1852 tcp_update_sndspace(tp); 1853 if (sb_notify(&so->so_snd)) { 1854 tp->t_flags |= TF_BLOCKOUTPUT; 1855 sowwakeup(so); 1856 tp->t_flags &= ~TF_BLOCKOUTPUT; 1857 } 1858 1859 /* 1860 * If we had a pending ICMP message that referred to data 1861 * that have just been acknowledged, disregard the recorded 1862 * ICMP message. 1863 */ 1864 if ((tp->t_flags & TF_PMTUD_PEND) && 1865 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 1866 tp->t_flags &= ~TF_PMTUD_PEND; 1867 1868 /* 1869 * Keep track of the largest chunk of data acknowledged 1870 * since last PMTU update 1871 */ 1872 if (tp->t_pmtud_mss_acked < acked) 1873 tp->t_pmtud_mss_acked = acked; 1874 1875 tp->snd_una = th->th_ack; 1876 #ifdef TCP_ECN 1877 /* sync snd_last with snd_una */ 1878 if (SEQ_GT(tp->snd_una, tp->snd_last)) 1879 tp->snd_last = tp->snd_una; 1880 #endif 1881 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1882 tp->snd_nxt = tp->snd_una; 1883 #if defined (TCP_SACK) && defined (TCP_FACK) 1884 if (SEQ_GT(tp->snd_una, tp->snd_fack)) { 1885 tp->snd_fack = tp->snd_una; 1886 /* Update snd_awnd for partial ACK 1887 * without any SACK blocks. 1888 */ 1889 tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt, 1890 tp->snd_fack) + tp->retran_data; 1891 } 1892 #endif 1893 1894 switch (tp->t_state) { 1895 1896 /* 1897 * In FIN_WAIT_1 STATE in addition to the processing 1898 * for the ESTABLISHED state if our FIN is now acknowledged 1899 * then enter FIN_WAIT_2. 1900 */ 1901 case TCPS_FIN_WAIT_1: 1902 if (ourfinisacked) { 1903 /* 1904 * If we can't receive any more 1905 * data, then closing user can proceed. 1906 * Starting the timer is contrary to the 1907 * specification, but if we don't get a FIN 1908 * we'll hang forever. 1909 */ 1910 if (so->so_state & SS_CANTRCVMORE) { 1911 soisdisconnected(so); 1912 TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle); 1913 } 1914 tp->t_state = TCPS_FIN_WAIT_2; 1915 } 1916 break; 1917 1918 /* 1919 * In CLOSING STATE in addition to the processing for 1920 * the ESTABLISHED state if the ACK acknowledges our FIN 1921 * then enter the TIME-WAIT state, otherwise ignore 1922 * the segment. 1923 */ 1924 case TCPS_CLOSING: 1925 if (ourfinisacked) { 1926 tp->t_state = TCPS_TIME_WAIT; 1927 tcp_canceltimers(tp); 1928 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1929 soisdisconnected(so); 1930 } 1931 break; 1932 1933 /* 1934 * In LAST_ACK, we may still be waiting for data to drain 1935 * and/or to be acked, as well as for the ack of our FIN. 1936 * If our FIN is now acknowledged, delete the TCB, 1937 * enter the closed state and return. 1938 */ 1939 case TCPS_LAST_ACK: 1940 if (ourfinisacked) { 1941 tp = tcp_close(tp); 1942 goto drop; 1943 } 1944 break; 1945 1946 /* 1947 * In TIME_WAIT state the only thing that should arrive 1948 * is a retransmission of the remote FIN. Acknowledge 1949 * it and restart the finack timer. 1950 */ 1951 case TCPS_TIME_WAIT: 1952 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1953 goto dropafterack; 1954 } 1955 } 1956 1957 step6: 1958 /* 1959 * Update window information. 1960 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1961 */ 1962 if ((tiflags & TH_ACK) && 1963 (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq && 1964 (SEQ_LT(tp->snd_wl2, th->th_ack) || 1965 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 1966 /* keep track of pure window updates */ 1967 if (tlen == 0 && 1968 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1969 tcpstat.tcps_rcvwinupd++; 1970 tp->snd_wnd = tiwin; 1971 tp->snd_wl1 = th->th_seq; 1972 tp->snd_wl2 = th->th_ack; 1973 if (tp->snd_wnd > tp->max_sndwnd) 1974 tp->max_sndwnd = tp->snd_wnd; 1975 tp->t_flags |= TF_NEEDOUTPUT; 1976 } 1977 1978 /* 1979 * Process segments with URG. 1980 */ 1981 if ((tiflags & TH_URG) && th->th_urp && 1982 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1983 /* 1984 * This is a kludge, but if we receive and accept 1985 * random urgent pointers, we'll crash in 1986 * soreceive. It's hard to imagine someone 1987 * actually wanting to send this much urgent data. 1988 */ 1989 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1990 th->th_urp = 0; /* XXX */ 1991 tiflags &= ~TH_URG; /* XXX */ 1992 goto dodata; /* XXX */ 1993 } 1994 /* 1995 * If this segment advances the known urgent pointer, 1996 * then mark the data stream. This should not happen 1997 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1998 * a FIN has been received from the remote side. 1999 * In these states we ignore the URG. 2000 * 2001 * According to RFC961 (Assigned Protocols), 2002 * the urgent pointer points to the last octet 2003 * of urgent data. We continue, however, 2004 * to consider it to indicate the first octet 2005 * of data past the urgent section as the original 2006 * spec states (in one of two places). 2007 */ 2008 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2009 tp->rcv_up = th->th_seq + th->th_urp; 2010 so->so_oobmark = so->so_rcv.sb_cc + 2011 (tp->rcv_up - tp->rcv_nxt) - 1; 2012 if (so->so_oobmark == 0) 2013 so->so_state |= SS_RCVATMARK; 2014 sohasoutofband(so); 2015 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2016 } 2017 /* 2018 * Remove out of band data so doesn't get presented to user. 2019 * This can happen independent of advancing the URG pointer, 2020 * but if two URG's are pending at once, some out-of-band 2021 * data may creep in... ick. 2022 */ 2023 if (th->th_urp <= (u_int16_t) tlen && 2024 (so->so_options & SO_OOBINLINE) == 0) 2025 tcp_pulloutofband(so, th->th_urp, m, hdroptlen); 2026 } else 2027 /* 2028 * If no out of band data is expected, 2029 * pull receive urgent pointer along 2030 * with the receive window. 2031 */ 2032 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2033 tp->rcv_up = tp->rcv_nxt; 2034 dodata: /* XXX */ 2035 2036 /* 2037 * Process the segment text, merging it into the TCP sequencing queue, 2038 * and arranging for acknowledgment of receipt if necessary. 2039 * This process logically involves adjusting tp->rcv_wnd as data 2040 * is presented to the user (this happens in tcp_usrreq.c, 2041 * case PRU_RCVD). If a FIN has already been received on this 2042 * connection then we just ignore the text. 2043 */ 2044 if ((tlen || (tiflags & TH_FIN)) && 2045 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2046 #ifdef TCP_SACK 2047 tcp_seq laststart = th->th_seq; 2048 tcp_seq lastend = th->th_seq + tlen; 2049 #endif 2050 if (th->th_seq == tp->rcv_nxt && TAILQ_EMPTY(&tp->t_segq) && 2051 tp->t_state == TCPS_ESTABLISHED) { 2052 TCP_SETUP_ACK(tp, tiflags, m); 2053 tp->rcv_nxt += tlen; 2054 tiflags = th->th_flags & TH_FIN; 2055 tcpstat.tcps_rcvpack++; 2056 tcpstat.tcps_rcvbyte += tlen; 2057 ND6_HINT(tp); 2058 if (so->so_state & SS_CANTRCVMORE) 2059 m_freem(m); 2060 else { 2061 m_adj(m, hdroptlen); 2062 sbappendstream(&so->so_rcv, m); 2063 } 2064 tp->t_flags |= TF_BLOCKOUTPUT; 2065 sorwakeup(so); 2066 tp->t_flags &= ~TF_BLOCKOUTPUT; 2067 } else { 2068 m_adj(m, hdroptlen); 2069 tiflags = tcp_reass(tp, th, m, &tlen); 2070 tp->t_flags |= TF_ACKNOW; 2071 } 2072 #ifdef TCP_SACK 2073 if (tp->sack_enable) 2074 tcp_update_sack_list(tp, laststart, lastend); 2075 #endif 2076 2077 /* 2078 * variable len never referenced again in modern BSD, 2079 * so why bother computing it ?? 2080 */ 2081 #if 0 2082 /* 2083 * Note the amount of data that peer has sent into 2084 * our window, in order to estimate the sender's 2085 * buffer size. 2086 */ 2087 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2088 #endif /* 0 */ 2089 } else { 2090 m_freem(m); 2091 tiflags &= ~TH_FIN; 2092 } 2093 2094 /* 2095 * If FIN is received ACK the FIN and let the user know 2096 * that the connection is closing. Ignore a FIN received before 2097 * the connection is fully established. 2098 */ 2099 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2100 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2101 socantrcvmore(so); 2102 tp->t_flags |= TF_ACKNOW; 2103 tp->rcv_nxt++; 2104 } 2105 switch (tp->t_state) { 2106 2107 /* 2108 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2109 */ 2110 case TCPS_ESTABLISHED: 2111 tp->t_state = TCPS_CLOSE_WAIT; 2112 break; 2113 2114 /* 2115 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2116 * enter the CLOSING state. 2117 */ 2118 case TCPS_FIN_WAIT_1: 2119 tp->t_state = TCPS_CLOSING; 2120 break; 2121 2122 /* 2123 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2124 * starting the time-wait timer, turning off the other 2125 * standard timers. 2126 */ 2127 case TCPS_FIN_WAIT_2: 2128 tp->t_state = TCPS_TIME_WAIT; 2129 tcp_canceltimers(tp); 2130 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2131 soisdisconnected(so); 2132 break; 2133 2134 /* 2135 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2136 */ 2137 case TCPS_TIME_WAIT: 2138 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2139 break; 2140 } 2141 } 2142 if (so->so_options & SO_DEBUG) { 2143 switch (tp->pf) { 2144 #ifdef INET6 2145 case PF_INET6: 2146 tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti6, 2147 0, tlen); 2148 break; 2149 #endif /* INET6 */ 2150 case PF_INET: 2151 tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti, 2152 0, tlen); 2153 break; 2154 } 2155 } 2156 2157 /* 2158 * Return any desired output. 2159 */ 2160 if (tp->t_flags & (TF_ACKNOW|TF_NEEDOUTPUT)) 2161 (void) tcp_output(tp); 2162 return; 2163 2164 badsyn: 2165 /* 2166 * Received a bad SYN. Increment counters and dropwithreset. 2167 */ 2168 tcpstat.tcps_badsyn++; 2169 tp = NULL; 2170 goto dropwithreset; 2171 2172 dropafterack_ratelim: 2173 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2174 tcp_ackdrop_ppslim) == 0) { 2175 /* XXX stat */ 2176 goto drop; 2177 } 2178 /* ...fall into dropafterack... */ 2179 2180 dropafterack: 2181 /* 2182 * Generate an ACK dropping incoming segment if it occupies 2183 * sequence space, where the ACK reflects our state. 2184 */ 2185 if (tiflags & TH_RST) 2186 goto drop; 2187 m_freem(m); 2188 tp->t_flags |= TF_ACKNOW; 2189 (void) tcp_output(tp); 2190 return; 2191 2192 dropwithreset_ratelim: 2193 /* 2194 * We may want to rate-limit RSTs in certain situations, 2195 * particularly if we are sending an RST in response to 2196 * an attempt to connect to or otherwise communicate with 2197 * a port for which we have no socket. 2198 */ 2199 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2200 tcp_rst_ppslim) == 0) { 2201 /* XXX stat */ 2202 goto drop; 2203 } 2204 /* ...fall into dropwithreset... */ 2205 2206 dropwithreset: 2207 /* 2208 * Generate a RST, dropping incoming segment. 2209 * Make ACK acceptable to originator of segment. 2210 * Don't bother to respond to RST. 2211 */ 2212 if (tiflags & TH_RST) 2213 goto drop; 2214 if (tiflags & TH_ACK) { 2215 tcp_respond(tp, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack, 2216 TH_RST, m->m_pkthdr.ph_rtableid); 2217 } else { 2218 if (tiflags & TH_SYN) 2219 tlen++; 2220 tcp_respond(tp, mtod(m, caddr_t), th, th->th_seq + tlen, 2221 (tcp_seq)0, TH_RST|TH_ACK, m->m_pkthdr.ph_rtableid); 2222 } 2223 m_freem(m); 2224 return; 2225 2226 drop: 2227 /* 2228 * Drop space held by incoming segment and return. 2229 */ 2230 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) { 2231 switch (tp->pf) { 2232 #ifdef INET6 2233 case PF_INET6: 2234 tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti6, 2235 0, tlen); 2236 break; 2237 #endif /* INET6 */ 2238 case PF_INET: 2239 tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti, 2240 0, tlen); 2241 break; 2242 } 2243 } 2244 2245 m_freem(m); 2246 return; 2247 } 2248 2249 int 2250 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, 2251 struct mbuf *m, int iphlen, struct tcp_opt_info *oi, 2252 u_int rtableid) 2253 { 2254 u_int16_t mss = 0; 2255 int opt, optlen; 2256 #ifdef TCP_SIGNATURE 2257 caddr_t sigp = NULL; 2258 struct tdb *tdb = NULL; 2259 #endif /* TCP_SIGNATURE */ 2260 2261 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2262 opt = cp[0]; 2263 if (opt == TCPOPT_EOL) 2264 break; 2265 if (opt == TCPOPT_NOP) 2266 optlen = 1; 2267 else { 2268 if (cnt < 2) 2269 break; 2270 optlen = cp[1]; 2271 if (optlen < 2 || optlen > cnt) 2272 break; 2273 } 2274 switch (opt) { 2275 2276 default: 2277 continue; 2278 2279 case TCPOPT_MAXSEG: 2280 if (optlen != TCPOLEN_MAXSEG) 2281 continue; 2282 if (!(th->th_flags & TH_SYN)) 2283 continue; 2284 if (TCPS_HAVERCVDSYN(tp->t_state)) 2285 continue; 2286 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 2287 NTOHS(mss); 2288 oi->maxseg = mss; 2289 break; 2290 2291 case TCPOPT_WINDOW: 2292 if (optlen != TCPOLEN_WINDOW) 2293 continue; 2294 if (!(th->th_flags & TH_SYN)) 2295 continue; 2296 if (TCPS_HAVERCVDSYN(tp->t_state)) 2297 continue; 2298 tp->t_flags |= TF_RCVD_SCALE; 2299 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2300 break; 2301 2302 case TCPOPT_TIMESTAMP: 2303 if (optlen != TCPOLEN_TIMESTAMP) 2304 continue; 2305 oi->ts_present = 1; 2306 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2307 NTOHL(oi->ts_val); 2308 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2309 NTOHL(oi->ts_ecr); 2310 2311 if (!(th->th_flags & TH_SYN)) 2312 continue; 2313 if (TCPS_HAVERCVDSYN(tp->t_state)) 2314 continue; 2315 /* 2316 * A timestamp received in a SYN makes 2317 * it ok to send timestamp requests and replies. 2318 */ 2319 tp->t_flags |= TF_RCVD_TSTMP; 2320 tp->ts_recent = oi->ts_val; 2321 tp->ts_recent_age = tcp_now; 2322 break; 2323 2324 #ifdef TCP_SACK 2325 case TCPOPT_SACK_PERMITTED: 2326 if (!tp->sack_enable || optlen!=TCPOLEN_SACK_PERMITTED) 2327 continue; 2328 if (!(th->th_flags & TH_SYN)) 2329 continue; 2330 if (TCPS_HAVERCVDSYN(tp->t_state)) 2331 continue; 2332 /* MUST only be set on SYN */ 2333 tp->t_flags |= TF_SACK_PERMIT; 2334 break; 2335 case TCPOPT_SACK: 2336 tcp_sack_option(tp, th, cp, optlen); 2337 break; 2338 #endif 2339 #ifdef TCP_SIGNATURE 2340 case TCPOPT_SIGNATURE: 2341 if (optlen != TCPOLEN_SIGNATURE) 2342 continue; 2343 2344 if (sigp && timingsafe_bcmp(sigp, cp + 2, 16)) 2345 return (-1); 2346 2347 sigp = cp + 2; 2348 break; 2349 #endif /* TCP_SIGNATURE */ 2350 } 2351 } 2352 2353 #ifdef TCP_SIGNATURE 2354 if (tp->t_flags & TF_SIGNATURE) { 2355 union sockaddr_union src, dst; 2356 2357 memset(&src, 0, sizeof(union sockaddr_union)); 2358 memset(&dst, 0, sizeof(union sockaddr_union)); 2359 2360 switch (tp->pf) { 2361 case 0: 2362 case AF_INET: 2363 src.sa.sa_len = sizeof(struct sockaddr_in); 2364 src.sa.sa_family = AF_INET; 2365 src.sin.sin_addr = mtod(m, struct ip *)->ip_src; 2366 dst.sa.sa_len = sizeof(struct sockaddr_in); 2367 dst.sa.sa_family = AF_INET; 2368 dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst; 2369 break; 2370 #ifdef INET6 2371 case AF_INET6: 2372 src.sa.sa_len = sizeof(struct sockaddr_in6); 2373 src.sa.sa_family = AF_INET6; 2374 src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src; 2375 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2376 dst.sa.sa_family = AF_INET6; 2377 dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst; 2378 break; 2379 #endif /* INET6 */ 2380 } 2381 2382 tdb = gettdbbysrcdst(rtable_l2(rtableid), 2383 0, &src, &dst, IPPROTO_TCP); 2384 2385 /* 2386 * We don't have an SA for this peer, so we turn off 2387 * TF_SIGNATURE on the listen socket 2388 */ 2389 if (tdb == NULL && tp->t_state == TCPS_LISTEN) 2390 tp->t_flags &= ~TF_SIGNATURE; 2391 2392 } 2393 2394 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2395 tcpstat.tcps_rcvbadsig++; 2396 return (-1); 2397 } 2398 2399 if (sigp) { 2400 char sig[16]; 2401 2402 if (tdb == NULL) { 2403 tcpstat.tcps_rcvbadsig++; 2404 return (-1); 2405 } 2406 2407 if (tcp_signature(tdb, tp->pf, m, th, iphlen, 1, sig) < 0) 2408 return (-1); 2409 2410 if (timingsafe_bcmp(sig, sigp, 16)) { 2411 tcpstat.tcps_rcvbadsig++; 2412 return (-1); 2413 } 2414 2415 tcpstat.tcps_rcvgoodsig++; 2416 } 2417 #endif /* TCP_SIGNATURE */ 2418 2419 return (0); 2420 } 2421 2422 #if defined(TCP_SACK) 2423 u_long 2424 tcp_seq_subtract(u_long a, u_long b) 2425 { 2426 return ((long)(a - b)); 2427 } 2428 #endif 2429 2430 2431 #ifdef TCP_SACK 2432 /* 2433 * This function is called upon receipt of new valid data (while not in header 2434 * prediction mode), and it updates the ordered list of sacks. 2435 */ 2436 void 2437 tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_laststart, 2438 tcp_seq rcv_lastend) 2439 { 2440 /* 2441 * First reported block MUST be the most recent one. Subsequent 2442 * blocks SHOULD be in the order in which they arrived at the 2443 * receiver. These two conditions make the implementation fully 2444 * compliant with RFC 2018. 2445 */ 2446 int i, j = 0, count = 0, lastpos = -1; 2447 struct sackblk sack, firstsack, temp[MAX_SACK_BLKS]; 2448 2449 /* First clean up current list of sacks */ 2450 for (i = 0; i < tp->rcv_numsacks; i++) { 2451 sack = tp->sackblks[i]; 2452 if (sack.start == 0 && sack.end == 0) { 2453 count++; /* count = number of blocks to be discarded */ 2454 continue; 2455 } 2456 if (SEQ_LEQ(sack.end, tp->rcv_nxt)) { 2457 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2458 count++; 2459 } else { 2460 temp[j].start = tp->sackblks[i].start; 2461 temp[j++].end = tp->sackblks[i].end; 2462 } 2463 } 2464 tp->rcv_numsacks -= count; 2465 if (tp->rcv_numsacks == 0) { /* no sack blocks currently (fast path) */ 2466 tcp_clean_sackreport(tp); 2467 if (SEQ_LT(tp->rcv_nxt, rcv_laststart)) { 2468 /* ==> need first sack block */ 2469 tp->sackblks[0].start = rcv_laststart; 2470 tp->sackblks[0].end = rcv_lastend; 2471 tp->rcv_numsacks = 1; 2472 } 2473 return; 2474 } 2475 /* Otherwise, sack blocks are already present. */ 2476 for (i = 0; i < tp->rcv_numsacks; i++) 2477 tp->sackblks[i] = temp[i]; /* first copy back sack list */ 2478 if (SEQ_GEQ(tp->rcv_nxt, rcv_lastend)) 2479 return; /* sack list remains unchanged */ 2480 /* 2481 * From here, segment just received should be (part of) the 1st sack. 2482 * Go through list, possibly coalescing sack block entries. 2483 */ 2484 firstsack.start = rcv_laststart; 2485 firstsack.end = rcv_lastend; 2486 for (i = 0; i < tp->rcv_numsacks; i++) { 2487 sack = tp->sackblks[i]; 2488 if (SEQ_LT(sack.end, firstsack.start) || 2489 SEQ_GT(sack.start, firstsack.end)) 2490 continue; /* no overlap */ 2491 if (sack.start == firstsack.start && sack.end == firstsack.end){ 2492 /* 2493 * identical block; delete it here since we will 2494 * move it to the front of the list. 2495 */ 2496 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2497 lastpos = i; /* last posn with a zero entry */ 2498 continue; 2499 } 2500 if (SEQ_LEQ(sack.start, firstsack.start)) 2501 firstsack.start = sack.start; /* merge blocks */ 2502 if (SEQ_GEQ(sack.end, firstsack.end)) 2503 firstsack.end = sack.end; /* merge blocks */ 2504 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2505 lastpos = i; /* last posn with a zero entry */ 2506 } 2507 if (lastpos != -1) { /* at least one merge */ 2508 for (i = 0, j = 1; i < tp->rcv_numsacks; i++) { 2509 sack = tp->sackblks[i]; 2510 if (sack.start == 0 && sack.end == 0) 2511 continue; 2512 temp[j++] = sack; 2513 } 2514 tp->rcv_numsacks = j; /* including first blk (added later) */ 2515 for (i = 1; i < tp->rcv_numsacks; i++) /* now copy back */ 2516 tp->sackblks[i] = temp[i]; 2517 } else { /* no merges -- shift sacks by 1 */ 2518 if (tp->rcv_numsacks < MAX_SACK_BLKS) 2519 tp->rcv_numsacks++; 2520 for (i = tp->rcv_numsacks-1; i > 0; i--) 2521 tp->sackblks[i] = tp->sackblks[i-1]; 2522 } 2523 tp->sackblks[0] = firstsack; 2524 return; 2525 } 2526 2527 /* 2528 * Process the TCP SACK option. tp->snd_holes is an ordered list 2529 * of holes (oldest to newest, in terms of the sequence space). 2530 */ 2531 void 2532 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen) 2533 { 2534 int tmp_olen; 2535 u_char *tmp_cp; 2536 struct sackhole *cur, *p, *temp; 2537 2538 if (!tp->sack_enable) 2539 return; 2540 /* SACK without ACK doesn't make sense. */ 2541 if ((th->th_flags & TH_ACK) == 0) 2542 return; 2543 /* Make sure the ACK on this segment is in [snd_una, snd_max]. */ 2544 if (SEQ_LT(th->th_ack, tp->snd_una) || 2545 SEQ_GT(th->th_ack, tp->snd_max)) 2546 return; 2547 /* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */ 2548 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2549 return; 2550 /* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */ 2551 tmp_cp = cp + 2; 2552 tmp_olen = optlen - 2; 2553 tcpstat.tcps_sack_rcv_opts++; 2554 if (tp->snd_numholes < 0) 2555 tp->snd_numholes = 0; 2556 if (tp->t_maxseg == 0) 2557 panic("tcp_sack_option"); /* Should never happen */ 2558 while (tmp_olen > 0) { 2559 struct sackblk sack; 2560 2561 bcopy(tmp_cp, (char *) &(sack.start), sizeof(tcp_seq)); 2562 NTOHL(sack.start); 2563 bcopy(tmp_cp + sizeof(tcp_seq), 2564 (char *) &(sack.end), sizeof(tcp_seq)); 2565 NTOHL(sack.end); 2566 tmp_olen -= TCPOLEN_SACK; 2567 tmp_cp += TCPOLEN_SACK; 2568 if (SEQ_LEQ(sack.end, sack.start)) 2569 continue; /* bad SACK fields */ 2570 if (SEQ_LEQ(sack.end, tp->snd_una)) 2571 continue; /* old block */ 2572 #if defined(TCP_SACK) && defined(TCP_FACK) 2573 /* Updates snd_fack. */ 2574 if (SEQ_GT(sack.end, tp->snd_fack)) 2575 tp->snd_fack = sack.end; 2576 #endif /* TCP_FACK */ 2577 if (SEQ_GT(th->th_ack, tp->snd_una)) { 2578 if (SEQ_LT(sack.start, th->th_ack)) 2579 continue; 2580 } 2581 if (SEQ_GT(sack.end, tp->snd_max)) 2582 continue; 2583 if (tp->snd_holes == NULL) { /* first hole */ 2584 tp->snd_holes = (struct sackhole *) 2585 pool_get(&sackhl_pool, PR_NOWAIT); 2586 if (tp->snd_holes == NULL) { 2587 /* ENOBUFS, so ignore SACKed block for now*/ 2588 goto done; 2589 } 2590 cur = tp->snd_holes; 2591 cur->start = th->th_ack; 2592 cur->end = sack.start; 2593 cur->rxmit = cur->start; 2594 cur->next = NULL; 2595 tp->snd_numholes = 1; 2596 tp->rcv_lastsack = sack.end; 2597 /* 2598 * dups is at least one. If more data has been 2599 * SACKed, it can be greater than one. 2600 */ 2601 cur->dups = min(tcprexmtthresh, 2602 ((sack.end - cur->end)/tp->t_maxseg)); 2603 if (cur->dups < 1) 2604 cur->dups = 1; 2605 continue; /* with next sack block */ 2606 } 2607 /* Go thru list of holes: p = previous, cur = current */ 2608 p = cur = tp->snd_holes; 2609 while (cur) { 2610 if (SEQ_LEQ(sack.end, cur->start)) 2611 /* SACKs data before the current hole */ 2612 break; /* no use going through more holes */ 2613 if (SEQ_GEQ(sack.start, cur->end)) { 2614 /* SACKs data beyond the current hole */ 2615 cur->dups++; 2616 if (((sack.end - cur->end)/tp->t_maxseg) >= 2617 tcprexmtthresh) 2618 cur->dups = tcprexmtthresh; 2619 p = cur; 2620 cur = cur->next; 2621 continue; 2622 } 2623 if (SEQ_LEQ(sack.start, cur->start)) { 2624 /* Data acks at least the beginning of hole */ 2625 #if defined(TCP_SACK) && defined(TCP_FACK) 2626 if (SEQ_GT(sack.end, cur->rxmit)) 2627 tp->retran_data -= 2628 tcp_seq_subtract(cur->rxmit, 2629 cur->start); 2630 else 2631 tp->retran_data -= 2632 tcp_seq_subtract(sack.end, 2633 cur->start); 2634 #endif /* TCP_FACK */ 2635 if (SEQ_GEQ(sack.end, cur->end)) { 2636 /* Acks entire hole, so delete hole */ 2637 if (p != cur) { 2638 p->next = cur->next; 2639 pool_put(&sackhl_pool, cur); 2640 cur = p->next; 2641 } else { 2642 cur = cur->next; 2643 pool_put(&sackhl_pool, p); 2644 p = cur; 2645 tp->snd_holes = p; 2646 } 2647 tp->snd_numholes--; 2648 continue; 2649 } 2650 /* otherwise, move start of hole forward */ 2651 cur->start = sack.end; 2652 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 2653 p = cur; 2654 cur = cur->next; 2655 continue; 2656 } 2657 /* move end of hole backward */ 2658 if (SEQ_GEQ(sack.end, cur->end)) { 2659 #if defined(TCP_SACK) && defined(TCP_FACK) 2660 if (SEQ_GT(cur->rxmit, sack.start)) 2661 tp->retran_data -= 2662 tcp_seq_subtract(cur->rxmit, 2663 sack.start); 2664 #endif /* TCP_FACK */ 2665 cur->end = sack.start; 2666 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 2667 cur->dups++; 2668 if (((sack.end - cur->end)/tp->t_maxseg) >= 2669 tcprexmtthresh) 2670 cur->dups = tcprexmtthresh; 2671 p = cur; 2672 cur = cur->next; 2673 continue; 2674 } 2675 if (SEQ_LT(cur->start, sack.start) && 2676 SEQ_GT(cur->end, sack.end)) { 2677 /* 2678 * ACKs some data in middle of a hole; need to 2679 * split current hole 2680 */ 2681 temp = (struct sackhole *) 2682 pool_get(&sackhl_pool, PR_NOWAIT); 2683 if (temp == NULL) 2684 goto done; /* ENOBUFS */ 2685 #if defined(TCP_SACK) && defined(TCP_FACK) 2686 if (SEQ_GT(cur->rxmit, sack.end)) 2687 tp->retran_data -= 2688 tcp_seq_subtract(sack.end, 2689 sack.start); 2690 else if (SEQ_GT(cur->rxmit, sack.start)) 2691 tp->retran_data -= 2692 tcp_seq_subtract(cur->rxmit, 2693 sack.start); 2694 #endif /* TCP_FACK */ 2695 temp->next = cur->next; 2696 temp->start = sack.end; 2697 temp->end = cur->end; 2698 temp->dups = cur->dups; 2699 temp->rxmit = SEQ_MAX(cur->rxmit, temp->start); 2700 cur->end = sack.start; 2701 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 2702 cur->dups++; 2703 if (((sack.end - cur->end)/tp->t_maxseg) >= 2704 tcprexmtthresh) 2705 cur->dups = tcprexmtthresh; 2706 cur->next = temp; 2707 p = temp; 2708 cur = p->next; 2709 tp->snd_numholes++; 2710 } 2711 } 2712 /* At this point, p points to the last hole on the list */ 2713 if (SEQ_LT(tp->rcv_lastsack, sack.start)) { 2714 /* 2715 * Need to append new hole at end. 2716 * Last hole is p (and it's not NULL). 2717 */ 2718 temp = (struct sackhole *) 2719 pool_get(&sackhl_pool, PR_NOWAIT); 2720 if (temp == NULL) 2721 goto done; /* ENOBUFS */ 2722 temp->start = tp->rcv_lastsack; 2723 temp->end = sack.start; 2724 temp->dups = min(tcprexmtthresh, 2725 ((sack.end - sack.start)/tp->t_maxseg)); 2726 if (temp->dups < 1) 2727 temp->dups = 1; 2728 temp->rxmit = temp->start; 2729 temp->next = 0; 2730 p->next = temp; 2731 tp->rcv_lastsack = sack.end; 2732 tp->snd_numholes++; 2733 } 2734 } 2735 done: 2736 #if defined(TCP_SACK) && defined(TCP_FACK) 2737 /* 2738 * Update retran_data and snd_awnd. Go through the list of 2739 * holes. Increment retran_data by (hole->rxmit - hole->start). 2740 */ 2741 tp->retran_data = 0; 2742 cur = tp->snd_holes; 2743 while (cur) { 2744 tp->retran_data += cur->rxmit - cur->start; 2745 cur = cur->next; 2746 } 2747 tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt, tp->snd_fack) + 2748 tp->retran_data; 2749 #endif /* TCP_FACK */ 2750 2751 return; 2752 } 2753 2754 /* 2755 * Delete stale (i.e, cumulatively ack'd) holes. Hole is deleted only if 2756 * it is completely acked; otherwise, tcp_sack_option(), called from 2757 * tcp_dooptions(), will fix up the hole. 2758 */ 2759 void 2760 tcp_del_sackholes(struct tcpcb *tp, struct tcphdr *th) 2761 { 2762 if (tp->sack_enable && tp->t_state != TCPS_LISTEN) { 2763 /* max because this could be an older ack just arrived */ 2764 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 2765 th->th_ack : tp->snd_una; 2766 struct sackhole *cur = tp->snd_holes; 2767 struct sackhole *prev; 2768 while (cur) 2769 if (SEQ_LEQ(cur->end, lastack)) { 2770 prev = cur; 2771 cur = cur->next; 2772 pool_put(&sackhl_pool, prev); 2773 tp->snd_numholes--; 2774 } else if (SEQ_LT(cur->start, lastack)) { 2775 cur->start = lastack; 2776 if (SEQ_LT(cur->rxmit, cur->start)) 2777 cur->rxmit = cur->start; 2778 break; 2779 } else 2780 break; 2781 tp->snd_holes = cur; 2782 } 2783 } 2784 2785 /* 2786 * Delete all receiver-side SACK information. 2787 */ 2788 void 2789 tcp_clean_sackreport(struct tcpcb *tp) 2790 { 2791 int i; 2792 2793 tp->rcv_numsacks = 0; 2794 for (i = 0; i < MAX_SACK_BLKS; i++) 2795 tp->sackblks[i].start = tp->sackblks[i].end=0; 2796 2797 } 2798 2799 /* 2800 * Checks for partial ack. If partial ack arrives, turn off retransmission 2801 * timer, deflate the window, do not clear tp->t_dupacks, and return 1. 2802 * If the ack advances at least to tp->snd_last, return 0. 2803 */ 2804 int 2805 tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th) 2806 { 2807 if (SEQ_LT(th->th_ack, tp->snd_last)) { 2808 /* Turn off retx. timer (will start again next segment) */ 2809 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2810 tp->t_rtttime = 0; 2811 #ifndef TCP_FACK 2812 /* 2813 * Partial window deflation. This statement relies on the 2814 * fact that tp->snd_una has not been updated yet. In FACK 2815 * hold snd_cwnd constant during fast recovery. 2816 */ 2817 if (tp->snd_cwnd > (th->th_ack - tp->snd_una)) { 2818 tp->snd_cwnd -= th->th_ack - tp->snd_una; 2819 tp->snd_cwnd += tp->t_maxseg; 2820 } else 2821 tp->snd_cwnd = tp->t_maxseg; 2822 #endif 2823 return (1); 2824 } 2825 return (0); 2826 } 2827 #endif /* TCP_SACK */ 2828 2829 /* 2830 * Pull out of band byte out of a segment so 2831 * it doesn't appear in the user's data queue. 2832 * It is still reflected in the segment length for 2833 * sequencing purposes. 2834 */ 2835 void 2836 tcp_pulloutofband(struct socket *so, u_int urgent, struct mbuf *m, int off) 2837 { 2838 int cnt = off + urgent - 1; 2839 2840 while (cnt >= 0) { 2841 if (m->m_len > cnt) { 2842 char *cp = mtod(m, caddr_t) + cnt; 2843 struct tcpcb *tp = sototcpcb(so); 2844 2845 tp->t_iobc = *cp; 2846 tp->t_oobflags |= TCPOOB_HAVEDATA; 2847 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2848 m->m_len--; 2849 return; 2850 } 2851 cnt -= m->m_len; 2852 m = m->m_next; 2853 if (m == NULL) 2854 break; 2855 } 2856 panic("tcp_pulloutofband"); 2857 } 2858 2859 /* 2860 * Collect new round-trip time estimate 2861 * and update averages and current timeout. 2862 */ 2863 void 2864 tcp_xmit_timer(struct tcpcb *tp, int rtt) 2865 { 2866 short delta; 2867 short rttmin; 2868 2869 if (rtt < 0) 2870 rtt = 0; 2871 else if (rtt > TCP_RTT_MAX) 2872 rtt = TCP_RTT_MAX; 2873 2874 tcpstat.tcps_rttupdated++; 2875 if (tp->t_srtt != 0) { 2876 /* 2877 * delta is fixed point with 2 (TCP_RTT_BASE_SHIFT) bits 2878 * after the binary point (scaled by 4), whereas 2879 * srtt is stored as fixed point with 5 bits after the 2880 * binary point (i.e., scaled by 32). The following magic 2881 * is equivalent to the smoothing algorithm in rfc793 with 2882 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2883 * point). 2884 */ 2885 delta = (rtt << TCP_RTT_BASE_SHIFT) - 2886 (tp->t_srtt >> TCP_RTT_SHIFT); 2887 if ((tp->t_srtt += delta) <= 0) 2888 tp->t_srtt = 1 << TCP_RTT_BASE_SHIFT; 2889 /* 2890 * We accumulate a smoothed rtt variance (actually, a 2891 * smoothed mean difference), then set the retransmit 2892 * timer to smoothed rtt + 4 times the smoothed variance. 2893 * rttvar is stored as fixed point with 4 bits after the 2894 * binary point (scaled by 16). The following is 2895 * equivalent to rfc793 smoothing with an alpha of .75 2896 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2897 * rfc793's wired-in beta. 2898 */ 2899 if (delta < 0) 2900 delta = -delta; 2901 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2902 if ((tp->t_rttvar += delta) <= 0) 2903 tp->t_rttvar = 1 << TCP_RTT_BASE_SHIFT; 2904 } else { 2905 /* 2906 * No rtt measurement yet - use the unsmoothed rtt. 2907 * Set the variance to half the rtt (so our first 2908 * retransmit happens at 3*rtt). 2909 */ 2910 tp->t_srtt = (rtt + 1) << (TCP_RTT_SHIFT + TCP_RTT_BASE_SHIFT); 2911 tp->t_rttvar = (rtt + 1) << 2912 (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT - 1); 2913 } 2914 tp->t_rtttime = 0; 2915 tp->t_rxtshift = 0; 2916 2917 /* 2918 * the retransmit should happen at rtt + 4 * rttvar. 2919 * Because of the way we do the smoothing, srtt and rttvar 2920 * will each average +1/2 tick of bias. When we compute 2921 * the retransmit timer, we want 1/2 tick of rounding and 2922 * 1 extra tick because of +-1/2 tick uncertainty in the 2923 * firing of the timer. The bias will give us exactly the 2924 * 1.5 tick we need. But, because the bias is 2925 * statistical, we have to test that we don't drop below 2926 * the minimum feasible timer (which is 2 ticks). 2927 */ 2928 rttmin = min(max(rtt + 2, tp->t_rttmin), TCPTV_REXMTMAX); 2929 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 2930 2931 /* 2932 * We received an ack for a packet that wasn't retransmitted; 2933 * it is probably safe to discard any error indications we've 2934 * received recently. This isn't quite right, but close enough 2935 * for now (a route might have failed after we sent a segment, 2936 * and the return path might not be symmetrical). 2937 */ 2938 tp->t_softerror = 0; 2939 } 2940 2941 /* 2942 * Determine a reasonable value for maxseg size. 2943 * If the route is known, check route for mtu. 2944 * If none, use an mss that can be handled on the outgoing 2945 * interface without forcing IP to fragment; if bigger than 2946 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2947 * to utilize large mbufs. If no route is found, route has no mtu, 2948 * or the destination isn't local, use a default, hopefully conservative 2949 * size (usually 512 or the default IP max size, but no more than the mtu 2950 * of the interface), as we can't discover anything about intervening 2951 * gateways or networks. We also initialize the congestion/slow start 2952 * window to be a single segment if the destination isn't local. 2953 * While looking at the routing entry, we also initialize other path-dependent 2954 * parameters from pre-set or cached values in the routing entry. 2955 * 2956 * Also take into account the space needed for options that we 2957 * send regularly. Make maxseg shorter by that amount to assure 2958 * that we can send maxseg amount of data even when the options 2959 * are present. Store the upper limit of the length of options plus 2960 * data in maxopd. 2961 * 2962 * NOTE: offer == -1 indicates that the maxseg size changed due to 2963 * Path MTU discovery. 2964 */ 2965 int 2966 tcp_mss(struct tcpcb *tp, int offer) 2967 { 2968 struct rtentry *rt; 2969 struct ifnet *ifp; 2970 int mss, mssopt; 2971 int iphlen; 2972 struct inpcb *inp; 2973 2974 inp = tp->t_inpcb; 2975 2976 mssopt = mss = tcp_mssdflt; 2977 2978 rt = in_pcbrtentry(inp); 2979 2980 if (rt == NULL) 2981 goto out; 2982 2983 ifp = rt->rt_ifp; 2984 2985 switch (tp->pf) { 2986 #ifdef INET6 2987 case AF_INET6: 2988 iphlen = sizeof(struct ip6_hdr); 2989 break; 2990 #endif 2991 case AF_INET: 2992 iphlen = sizeof(struct ip); 2993 break; 2994 default: 2995 /* the family does not support path MTU discovery */ 2996 goto out; 2997 } 2998 2999 /* 3000 * if there's an mtu associated with the route and we support 3001 * path MTU discovery for the underlying protocol family, use it. 3002 */ 3003 if (rt->rt_rmx.rmx_mtu) { 3004 /* 3005 * One may wish to lower MSS to take into account options, 3006 * especially security-related options. 3007 */ 3008 if (tp->pf == AF_INET6 && rt->rt_rmx.rmx_mtu < IPV6_MMTU) { 3009 /* 3010 * RFC2460 section 5, last paragraph: if path MTU is 3011 * smaller than 1280, use 1280 as packet size and 3012 * attach fragment header. 3013 */ 3014 mss = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) - 3015 sizeof(struct tcphdr); 3016 } else { 3017 mss = rt->rt_rmx.rmx_mtu - iphlen - 3018 sizeof(struct tcphdr); 3019 } 3020 } else if (!ifp) { 3021 /* 3022 * ifp may be null and rmx_mtu may be zero in certain 3023 * v6 cases (e.g., if ND wasn't able to resolve the 3024 * destination host. 3025 */ 3026 goto out; 3027 } else if (ifp->if_flags & IFF_LOOPBACK) { 3028 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 3029 } else if (tp->pf == AF_INET) { 3030 if (ip_mtudisc) 3031 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 3032 } 3033 #ifdef INET6 3034 else if (tp->pf == AF_INET6) { 3035 /* 3036 * for IPv6, path MTU discovery is always turned on, 3037 * or the node must use packet size <= 1280. 3038 */ 3039 mss = IN6_LINKMTU(ifp) - iphlen - sizeof(struct tcphdr); 3040 } 3041 #endif /* INET6 */ 3042 3043 /* Calculate the value that we offer in TCPOPT_MAXSEG */ 3044 if (offer != -1) { 3045 #ifndef INET6 3046 mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 3047 #else 3048 if (tp->pf == AF_INET6) 3049 mssopt = IN6_LINKMTU(ifp) - iphlen - 3050 sizeof(struct tcphdr); 3051 else 3052 mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 3053 #endif 3054 3055 mssopt = max(tcp_mssdflt, mssopt); 3056 } 3057 3058 out: 3059 /* 3060 * The current mss, t_maxseg, is initialized to the default value. 3061 * If we compute a smaller value, reduce the current mss. 3062 * If we compute a larger value, return it for use in sending 3063 * a max seg size option, but don't store it for use 3064 * unless we received an offer at least that large from peer. 3065 * 3066 * However, do not accept offers lower than the minimum of 3067 * the interface MTU and 216. 3068 */ 3069 if (offer > 0) 3070 tp->t_peermss = offer; 3071 if (tp->t_peermss) 3072 mss = min(mss, max(tp->t_peermss, 216)); 3073 3074 /* sanity - at least max opt. space */ 3075 mss = max(mss, 64); 3076 3077 /* 3078 * maxopd stores the maximum length of data AND options 3079 * in a segment; maxseg is the amount of data in a normal 3080 * segment. We need to store this value (maxopd) apart 3081 * from maxseg, because now every segment carries options 3082 * and thus we normally have somewhat less data in segments. 3083 */ 3084 tp->t_maxopd = mss; 3085 3086 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3087 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 3088 mss -= TCPOLEN_TSTAMP_APPA; 3089 #ifdef TCP_SIGNATURE 3090 if (tp->t_flags & TF_SIGNATURE) 3091 mss -= TCPOLEN_SIGLEN; 3092 #endif 3093 3094 if (offer == -1) { 3095 /* mss changed due to Path MTU discovery */ 3096 tp->t_flags &= ~TF_PMTUD_PEND; 3097 tp->t_pmtud_mtu_sent = 0; 3098 tp->t_pmtud_mss_acked = 0; 3099 if (mss < tp->t_maxseg) { 3100 /* 3101 * Follow suggestion in RFC 2414 to reduce the 3102 * congestion window by the ratio of the old 3103 * segment size to the new segment size. 3104 */ 3105 tp->snd_cwnd = ulmax((tp->snd_cwnd / tp->t_maxseg) * 3106 mss, mss); 3107 } 3108 } else if (tcp_do_rfc3390 == 2) { 3109 /* increase initial window */ 3110 tp->snd_cwnd = ulmin(10 * mss, ulmax(2 * mss, 14600)); 3111 } else if (tcp_do_rfc3390) { 3112 /* increase initial window */ 3113 tp->snd_cwnd = ulmin(4 * mss, ulmax(2 * mss, 4380)); 3114 } else 3115 tp->snd_cwnd = mss; 3116 3117 tp->t_maxseg = mss; 3118 3119 return (offer != -1 ? mssopt : mss); 3120 } 3121 3122 u_int 3123 tcp_hdrsz(struct tcpcb *tp) 3124 { 3125 u_int hlen; 3126 3127 switch (tp->pf) { 3128 #ifdef INET6 3129 case AF_INET6: 3130 hlen = sizeof(struct ip6_hdr); 3131 break; 3132 #endif 3133 case AF_INET: 3134 hlen = sizeof(struct ip); 3135 break; 3136 default: 3137 hlen = 0; 3138 break; 3139 } 3140 hlen += sizeof(struct tcphdr); 3141 3142 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3143 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 3144 hlen += TCPOLEN_TSTAMP_APPA; 3145 #ifdef TCP_SIGNATURE 3146 if (tp->t_flags & TF_SIGNATURE) 3147 hlen += TCPOLEN_SIGLEN; 3148 #endif 3149 return (hlen); 3150 } 3151 3152 /* 3153 * Set connection variables based on the effective MSS. 3154 * We are passed the TCPCB for the actual connection. If we 3155 * are the server, we are called by the compressed state engine 3156 * when the 3-way handshake is complete. If we are the client, 3157 * we are called when we receive the SYN,ACK from the server. 3158 * 3159 * NOTE: The t_maxseg value must be initialized in the TCPCB 3160 * before this routine is called! 3161 */ 3162 void 3163 tcp_mss_update(struct tcpcb *tp) 3164 { 3165 int mss; 3166 u_long bufsize; 3167 struct rtentry *rt; 3168 struct socket *so; 3169 3170 so = tp->t_inpcb->inp_socket; 3171 mss = tp->t_maxseg; 3172 3173 rt = in_pcbrtentry(tp->t_inpcb); 3174 3175 if (rt == NULL) 3176 return; 3177 3178 bufsize = so->so_snd.sb_hiwat; 3179 if (bufsize < mss) { 3180 mss = bufsize; 3181 /* Update t_maxseg and t_maxopd */ 3182 tcp_mss(tp, mss); 3183 } else { 3184 bufsize = roundup(bufsize, mss); 3185 if (bufsize > sb_max) 3186 bufsize = sb_max; 3187 (void)sbreserve(&so->so_snd, bufsize); 3188 } 3189 3190 bufsize = so->so_rcv.sb_hiwat; 3191 if (bufsize > mss) { 3192 bufsize = roundup(bufsize, mss); 3193 if (bufsize > sb_max) 3194 bufsize = sb_max; 3195 (void)sbreserve(&so->so_rcv, bufsize); 3196 } 3197 3198 } 3199 3200 #if defined (TCP_SACK) 3201 /* 3202 * Checks for partial ack. If partial ack arrives, force the retransmission 3203 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 3204 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 3205 * be started again. If the ack advances at least to tp->snd_last, return 0. 3206 */ 3207 int 3208 tcp_newreno(struct tcpcb *tp, struct tcphdr *th) 3209 { 3210 if (SEQ_LT(th->th_ack, tp->snd_last)) { 3211 /* 3212 * snd_una has not been updated and the socket send buffer 3213 * not yet drained of the acked data, so we have to leave 3214 * snd_una as it was to get the correct data offset in 3215 * tcp_output(). 3216 */ 3217 tcp_seq onxt = tp->snd_nxt; 3218 u_long ocwnd = tp->snd_cwnd; 3219 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3220 tp->t_rtttime = 0; 3221 tp->snd_nxt = th->th_ack; 3222 /* 3223 * Set snd_cwnd to one segment beyond acknowledged offset 3224 * (tp->snd_una not yet updated when this function is called) 3225 */ 3226 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3227 (void) tcp_output(tp); 3228 tp->snd_cwnd = ocwnd; 3229 if (SEQ_GT(onxt, tp->snd_nxt)) 3230 tp->snd_nxt = onxt; 3231 /* 3232 * Partial window deflation. Relies on fact that tp->snd_una 3233 * not updated yet. 3234 */ 3235 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3236 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3237 else 3238 tp->snd_cwnd = 0; 3239 tp->snd_cwnd += tp->t_maxseg; 3240 3241 return 1; 3242 } 3243 return 0; 3244 } 3245 #endif /* TCP_SACK */ 3246 3247 int 3248 tcp_mss_adv(struct ifnet *ifp, int af) 3249 { 3250 int mss = 0; 3251 int iphlen; 3252 3253 switch (af) { 3254 case AF_INET: 3255 if (ifp != NULL) 3256 mss = ifp->if_mtu; 3257 iphlen = sizeof(struct ip); 3258 break; 3259 #ifdef INET6 3260 case AF_INET6: 3261 if (ifp != NULL) 3262 mss = IN6_LINKMTU(ifp); 3263 iphlen = sizeof(struct ip6_hdr); 3264 break; 3265 #endif 3266 default: 3267 unhandled_af(af); 3268 } 3269 mss = mss - iphlen - sizeof(struct tcphdr); 3270 return (max(mss, tcp_mssdflt)); 3271 } 3272 3273 /* 3274 * TCP compressed state engine. Currently used to hold compressed 3275 * state for SYN_RECEIVED. 3276 */ 3277 3278 u_long syn_cache_count; 3279 u_int32_t syn_hash1, syn_hash2; 3280 3281 #define SYN_HASH(sa, sp, dp) \ 3282 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3283 ((u_int32_t)(sp)))^syn_hash2))) 3284 #ifndef INET6 3285 #define SYN_HASHALL(hash, src, dst) \ 3286 do { \ 3287 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3288 ((struct sockaddr_in *)(src))->sin_port, \ 3289 ((struct sockaddr_in *)(dst))->sin_port); \ 3290 } while (/*CONSTCOND*/ 0) 3291 #else 3292 #define SYN_HASH6(sa, sp, dp) \ 3293 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3294 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3295 & 0x7fffffff) 3296 3297 #define SYN_HASHALL(hash, src, dst) \ 3298 do { \ 3299 switch ((src)->sa_family) { \ 3300 case AF_INET: \ 3301 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3302 ((struct sockaddr_in *)(src))->sin_port, \ 3303 ((struct sockaddr_in *)(dst))->sin_port); \ 3304 break; \ 3305 case AF_INET6: \ 3306 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 3307 ((struct sockaddr_in6 *)(src))->sin6_port, \ 3308 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 3309 break; \ 3310 default: \ 3311 hash = 0; \ 3312 } \ 3313 } while (/*CONSTCOND*/0) 3314 #endif /* INET6 */ 3315 3316 void 3317 syn_cache_rm(struct syn_cache *sc) 3318 { 3319 sc->sc_flags |= SCF_DEAD; 3320 TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket, 3321 sc, sc_bucketq); 3322 sc->sc_tp = NULL; 3323 LIST_REMOVE(sc, sc_tpq); 3324 tcp_syn_cache[sc->sc_bucketidx].sch_length--; 3325 timeout_del(&sc->sc_timer); 3326 syn_cache_count--; 3327 } 3328 3329 void 3330 syn_cache_put(struct syn_cache *sc) 3331 { 3332 if (sc->sc_ipopts) 3333 (void) m_free(sc->sc_ipopts); 3334 if (sc->sc_route4.ro_rt != NULL) { 3335 rtfree(sc->sc_route4.ro_rt); 3336 sc->sc_route4.ro_rt = NULL; 3337 } 3338 timeout_set(&sc->sc_timer, syn_cache_reaper, sc); 3339 timeout_add(&sc->sc_timer, 0); 3340 } 3341 3342 struct pool syn_cache_pool; 3343 3344 /* 3345 * We don't estimate RTT with SYNs, so each packet starts with the default 3346 * RTT and each timer step has a fixed timeout value. 3347 */ 3348 #define SYN_CACHE_TIMER_ARM(sc) \ 3349 do { \ 3350 TCPT_RANGESET((sc)->sc_rxtcur, \ 3351 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3352 TCPTV_REXMTMAX); \ 3353 if (!timeout_initialized(&(sc)->sc_timer)) \ 3354 timeout_set(&(sc)->sc_timer, syn_cache_timer, (sc)); \ 3355 timeout_add(&(sc)->sc_timer, (sc)->sc_rxtcur * (hz / PR_SLOWHZ)); \ 3356 } while (/*CONSTCOND*/0) 3357 3358 #define SYN_CACHE_TIMESTAMP(sc) tcp_now + (sc)->sc_modulate 3359 3360 void 3361 syn_cache_init() 3362 { 3363 int i; 3364 3365 /* Initialize the hash buckets. */ 3366 for (i = 0; i < tcp_syn_cache_size; i++) 3367 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3368 3369 /* Initialize the syn cache pool. */ 3370 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3371 "syncache", NULL); 3372 } 3373 3374 void 3375 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3376 { 3377 struct syn_cache_head *scp; 3378 struct syn_cache *sc2; 3379 int s; 3380 3381 /* 3382 * If there are no entries in the hash table, reinitialize 3383 * the hash secrets. 3384 */ 3385 if (syn_cache_count == 0) { 3386 syn_hash1 = arc4random(); 3387 syn_hash2 = arc4random(); 3388 } 3389 3390 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3391 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3392 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3393 3394 /* 3395 * Make sure that we don't overflow the per-bucket 3396 * limit or the total cache size limit. 3397 */ 3398 s = splsoftnet(); 3399 if (scp->sch_length >= tcp_syn_bucket_limit) { 3400 tcpstat.tcps_sc_bucketoverflow++; 3401 /* 3402 * The bucket is full. Toss the oldest element in the 3403 * bucket. This will be the first entry in the bucket. 3404 */ 3405 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3406 #ifdef DIAGNOSTIC 3407 /* 3408 * This should never happen; we should always find an 3409 * entry in our bucket. 3410 */ 3411 if (sc2 == NULL) 3412 panic("syn_cache_insert: bucketoverflow: impossible"); 3413 #endif 3414 syn_cache_rm(sc2); 3415 syn_cache_put(sc2); 3416 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3417 struct syn_cache_head *scp2, *sce; 3418 3419 tcpstat.tcps_sc_overflowed++; 3420 /* 3421 * The cache is full. Toss the oldest entry in the 3422 * first non-empty bucket we can find. 3423 * 3424 * XXX We would really like to toss the oldest 3425 * entry in the cache, but we hope that this 3426 * condition doesn't happen very often. 3427 */ 3428 scp2 = scp; 3429 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3430 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3431 for (++scp2; scp2 != scp; scp2++) { 3432 if (scp2 >= sce) 3433 scp2 = &tcp_syn_cache[0]; 3434 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3435 break; 3436 } 3437 #ifdef DIAGNOSTIC 3438 /* 3439 * This should never happen; we should always find a 3440 * non-empty bucket. 3441 */ 3442 if (scp2 == scp) 3443 panic("syn_cache_insert: cacheoverflow: " 3444 "impossible"); 3445 #endif 3446 } 3447 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3448 syn_cache_rm(sc2); 3449 syn_cache_put(sc2); 3450 } 3451 3452 /* 3453 * Initialize the entry's timer. 3454 */ 3455 sc->sc_rxttot = 0; 3456 sc->sc_rxtshift = 0; 3457 SYN_CACHE_TIMER_ARM(sc); 3458 3459 /* Link it from tcpcb entry */ 3460 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3461 3462 /* Put it into the bucket. */ 3463 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3464 scp->sch_length++; 3465 syn_cache_count++; 3466 3467 tcpstat.tcps_sc_added++; 3468 splx(s); 3469 } 3470 3471 /* 3472 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3473 * If we have retransmitted an entry the maximum number of times, expire 3474 * that entry. 3475 */ 3476 void 3477 syn_cache_timer(void *arg) 3478 { 3479 struct syn_cache *sc = arg; 3480 int s; 3481 3482 s = splsoftnet(); 3483 if (sc->sc_flags & SCF_DEAD) { 3484 splx(s); 3485 return; 3486 } 3487 3488 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3489 /* Drop it -- too many retransmissions. */ 3490 goto dropit; 3491 } 3492 3493 /* 3494 * Compute the total amount of time this entry has 3495 * been on a queue. If this entry has been on longer 3496 * than the keep alive timer would allow, expire it. 3497 */ 3498 sc->sc_rxttot += sc->sc_rxtcur; 3499 if (sc->sc_rxttot >= tcptv_keep_init) 3500 goto dropit; 3501 3502 tcpstat.tcps_sc_retransmitted++; 3503 (void) syn_cache_respond(sc, NULL); 3504 3505 /* Advance the timer back-off. */ 3506 sc->sc_rxtshift++; 3507 SYN_CACHE_TIMER_ARM(sc); 3508 3509 splx(s); 3510 return; 3511 3512 dropit: 3513 tcpstat.tcps_sc_timed_out++; 3514 syn_cache_rm(sc); 3515 syn_cache_put(sc); 3516 splx(s); 3517 } 3518 3519 void 3520 syn_cache_reaper(void *arg) 3521 { 3522 struct syn_cache *sc = arg; 3523 int s; 3524 3525 s = splsoftnet(); 3526 pool_put(&syn_cache_pool, (sc)); 3527 splx(s); 3528 return; 3529 } 3530 3531 /* 3532 * Remove syn cache created by the specified tcb entry, 3533 * because this does not make sense to keep them 3534 * (if there's no tcb entry, syn cache entry will never be used) 3535 */ 3536 void 3537 syn_cache_cleanup(struct tcpcb *tp) 3538 { 3539 struct syn_cache *sc, *nsc; 3540 int s; 3541 3542 s = splsoftnet(); 3543 3544 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3545 nsc = LIST_NEXT(sc, sc_tpq); 3546 3547 #ifdef DIAGNOSTIC 3548 if (sc->sc_tp != tp) 3549 panic("invalid sc_tp in syn_cache_cleanup"); 3550 #endif 3551 syn_cache_rm(sc); 3552 syn_cache_put(sc); 3553 } 3554 /* just for safety */ 3555 LIST_INIT(&tp->t_sc); 3556 3557 splx(s); 3558 } 3559 3560 /* 3561 * Find an entry in the syn cache. 3562 */ 3563 struct syn_cache * 3564 syn_cache_lookup(struct sockaddr *src, struct sockaddr *dst, 3565 struct syn_cache_head **headp, u_int rtableid) 3566 { 3567 struct syn_cache *sc; 3568 struct syn_cache_head *scp; 3569 u_int32_t hash; 3570 int s; 3571 3572 SYN_HASHALL(hash, src, dst); 3573 3574 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3575 *headp = scp; 3576 s = splsoftnet(); 3577 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3578 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3579 if (sc->sc_hash != hash) 3580 continue; 3581 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3582 !bcmp(&sc->sc_dst, dst, dst->sa_len) && 3583 rtable_l2(rtableid) == rtable_l2(sc->sc_rtableid)) { 3584 splx(s); 3585 return (sc); 3586 } 3587 } 3588 splx(s); 3589 return (NULL); 3590 } 3591 3592 /* 3593 * This function gets called when we receive an ACK for a 3594 * socket in the LISTEN state. We look up the connection 3595 * in the syn cache, and if its there, we pull it out of 3596 * the cache and turn it into a full-blown connection in 3597 * the SYN-RECEIVED state. 3598 * 3599 * The return values may not be immediately obvious, and their effects 3600 * can be subtle, so here they are: 3601 * 3602 * NULL SYN was not found in cache; caller should drop the 3603 * packet and send an RST. 3604 * 3605 * -1 We were unable to create the new connection, and are 3606 * aborting it. An ACK,RST is being sent to the peer 3607 * (unless we got screwey sequence numbners; see below), 3608 * because the 3-way handshake has been completed. Caller 3609 * should not free the mbuf, since we may be using it. If 3610 * we are not, we will free it. 3611 * 3612 * Otherwise, the return value is a pointer to the new socket 3613 * associated with the connection. 3614 */ 3615 struct socket * 3616 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3617 u_int hlen, u_int tlen, struct socket *so, struct mbuf *m) 3618 { 3619 struct syn_cache *sc; 3620 struct syn_cache_head *scp; 3621 struct inpcb *inp = NULL; 3622 struct tcpcb *tp = NULL; 3623 struct mbuf *am; 3624 int s; 3625 struct socket *oso; 3626 #if NPF > 0 3627 struct pf_divert *divert = NULL; 3628 #endif 3629 3630 s = splsoftnet(); 3631 if ((sc = syn_cache_lookup(src, dst, &scp, 3632 sotoinpcb(so)->inp_rtableid)) == NULL) { 3633 splx(s); 3634 return (NULL); 3635 } 3636 3637 /* 3638 * Verify the sequence and ack numbers. Try getting the correct 3639 * response again. 3640 */ 3641 if ((th->th_ack != sc->sc_iss + 1) || 3642 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3643 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3644 (void) syn_cache_respond(sc, m); 3645 splx(s); 3646 return ((struct socket *)(-1)); 3647 } 3648 3649 /* Remove this cache entry */ 3650 syn_cache_rm(sc); 3651 splx(s); 3652 3653 /* 3654 * Ok, create the full blown connection, and set things up 3655 * as they would have been set up if we had created the 3656 * connection when the SYN arrived. If we can't create 3657 * the connection, abort it. 3658 */ 3659 oso = so; 3660 so = sonewconn(so, SS_ISCONNECTED); 3661 if (so == NULL) 3662 goto resetandabort; 3663 3664 inp = sotoinpcb(oso); 3665 3666 #ifdef IPSEC 3667 /* 3668 * We need to copy the required security levels 3669 * from the old pcb. Ditto for any other 3670 * IPsec-related information. 3671 */ 3672 { 3673 struct inpcb *newinp = sotoinpcb(so); 3674 bcopy(inp->inp_seclevel, newinp->inp_seclevel, 3675 sizeof(inp->inp_seclevel)); 3676 } 3677 #endif /* IPSEC */ 3678 #ifdef INET6 3679 /* 3680 * inp still has the OLD in_pcb stuff, set the 3681 * v6-related flags on the new guy, too. 3682 */ 3683 { 3684 int flags = inp->inp_flags; 3685 struct inpcb *oldinpcb = inp; 3686 3687 inp = sotoinpcb(so); 3688 inp->inp_flags |= (flags & INP_IPV6); 3689 if ((inp->inp_flags & INP_IPV6) != 0) { 3690 inp->inp_ipv6.ip6_hlim = 3691 oldinpcb->inp_ipv6.ip6_hlim; 3692 } 3693 } 3694 #else /* INET6 */ 3695 inp = sotoinpcb(so); 3696 #endif /* INET6 */ 3697 3698 #if NPF > 0 3699 if (m && m->m_pkthdr.pf.flags & PF_TAG_DIVERTED && 3700 (divert = pf_find_divert(m)) != NULL) 3701 inp->inp_rtableid = divert->rdomain; 3702 else 3703 #endif 3704 /* inherit rtable from listening socket */ 3705 inp->inp_rtableid = sc->sc_rtableid; 3706 3707 inp->inp_lport = th->th_dport; 3708 switch (src->sa_family) { 3709 #ifdef INET6 3710 case AF_INET6: 3711 inp->inp_laddr6 = ((struct sockaddr_in6 *)dst)->sin6_addr; 3712 break; 3713 #endif /* INET6 */ 3714 case AF_INET: 3715 3716 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3717 inp->inp_options = ip_srcroute(m); 3718 if (inp->inp_options == NULL) { 3719 inp->inp_options = sc->sc_ipopts; 3720 sc->sc_ipopts = NULL; 3721 } 3722 break; 3723 } 3724 in_pcbrehash(inp); 3725 3726 /* 3727 * Give the new socket our cached route reference. 3728 */ 3729 if (src->sa_family == AF_INET) 3730 inp->inp_route = sc->sc_route4; /* struct assignment */ 3731 #ifdef INET6 3732 else 3733 inp->inp_route6 = sc->sc_route6; 3734 #endif 3735 sc->sc_route4.ro_rt = NULL; 3736 3737 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3738 if (am == NULL) 3739 goto resetandabort; 3740 am->m_len = src->sa_len; 3741 bcopy(src, mtod(am, caddr_t), src->sa_len); 3742 3743 switch (src->sa_family) { 3744 case AF_INET: 3745 /* drop IPv4 packet to AF_INET6 socket */ 3746 if (inp->inp_flags & INP_IPV6) { 3747 (void) m_free(am); 3748 goto resetandabort; 3749 } 3750 if (in_pcbconnect(inp, am)) { 3751 (void) m_free(am); 3752 goto resetandabort; 3753 } 3754 break; 3755 #ifdef INET6 3756 case AF_INET6: 3757 if (in6_pcbconnect(inp, am)) { 3758 (void) m_free(am); 3759 goto resetandabort; 3760 } 3761 break; 3762 #endif 3763 } 3764 (void) m_free(am); 3765 3766 tp = intotcpcb(inp); 3767 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3768 if (sc->sc_request_r_scale != 15) { 3769 tp->requested_s_scale = sc->sc_requested_s_scale; 3770 tp->request_r_scale = sc->sc_request_r_scale; 3771 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3772 } 3773 if (sc->sc_flags & SCF_TIMESTAMP) 3774 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3775 3776 tp->t_template = tcp_template(tp); 3777 if (tp->t_template == 0) { 3778 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3779 so = NULL; 3780 m_freem(m); 3781 goto abort; 3782 } 3783 #ifdef TCP_SACK 3784 tp->sack_enable = sc->sc_flags & SCF_SACK_PERMIT; 3785 #endif 3786 3787 tp->ts_modulate = sc->sc_modulate; 3788 tp->ts_recent = sc->sc_timestamp; 3789 tp->iss = sc->sc_iss; 3790 tp->irs = sc->sc_irs; 3791 tcp_sendseqinit(tp); 3792 #if defined (TCP_SACK) || defined(TCP_ECN) 3793 tp->snd_last = tp->snd_una; 3794 #endif /* TCP_SACK */ 3795 #if defined(TCP_SACK) && defined(TCP_FACK) 3796 tp->snd_fack = tp->snd_una; 3797 tp->retran_data = 0; 3798 tp->snd_awnd = 0; 3799 #endif /* TCP_FACK */ 3800 #ifdef TCP_ECN 3801 if (sc->sc_flags & SCF_ECN_PERMIT) { 3802 tp->t_flags |= TF_ECN_PERMIT; 3803 tcpstat.tcps_ecn_accepts++; 3804 } 3805 #endif 3806 #ifdef TCP_SACK 3807 if (sc->sc_flags & SCF_SACK_PERMIT) 3808 tp->t_flags |= TF_SACK_PERMIT; 3809 #endif 3810 #ifdef TCP_SIGNATURE 3811 if (sc->sc_flags & SCF_SIGNATURE) 3812 tp->t_flags |= TF_SIGNATURE; 3813 #endif 3814 tcp_rcvseqinit(tp); 3815 tp->t_state = TCPS_SYN_RECEIVED; 3816 tp->t_rcvtime = tcp_now; 3817 TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init); 3818 tcpstat.tcps_accepts++; 3819 3820 tcp_mss(tp, sc->sc_peermaxseg); /* sets t_maxseg */ 3821 if (sc->sc_peermaxseg) 3822 tcp_mss_update(tp); 3823 /* Reset initial window to 1 segment for retransmit */ 3824 if (sc->sc_rxtshift > 0) 3825 tp->snd_cwnd = tp->t_maxseg; 3826 tp->snd_wl1 = sc->sc_irs; 3827 tp->rcv_up = sc->sc_irs + 1; 3828 3829 /* 3830 * This is what whould have happened in tcp_output() when 3831 * the SYN,ACK was sent. 3832 */ 3833 tp->snd_up = tp->snd_una; 3834 tp->snd_max = tp->snd_nxt = tp->iss+1; 3835 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3836 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3837 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3838 tp->last_ack_sent = tp->rcv_nxt; 3839 3840 tcpstat.tcps_sc_completed++; 3841 syn_cache_put(sc); 3842 return (so); 3843 3844 resetandabort: 3845 tcp_respond(NULL, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack, TH_RST, 3846 m->m_pkthdr.ph_rtableid); 3847 m_freem(m); 3848 abort: 3849 if (so != NULL) 3850 (void) soabort(so); 3851 syn_cache_put(sc); 3852 tcpstat.tcps_sc_aborted++; 3853 return ((struct socket *)(-1)); 3854 } 3855 3856 /* 3857 * This function is called when we get a RST for a 3858 * non-existent connection, so that we can see if the 3859 * connection is in the syn cache. If it is, zap it. 3860 */ 3861 3862 void 3863 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3864 u_int rtableid) 3865 { 3866 struct syn_cache *sc; 3867 struct syn_cache_head *scp; 3868 int s = splsoftnet(); 3869 3870 if ((sc = syn_cache_lookup(src, dst, &scp, rtableid)) == NULL) { 3871 splx(s); 3872 return; 3873 } 3874 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3875 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3876 splx(s); 3877 return; 3878 } 3879 syn_cache_rm(sc); 3880 splx(s); 3881 tcpstat.tcps_sc_reset++; 3882 syn_cache_put(sc); 3883 } 3884 3885 void 3886 syn_cache_unreach(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3887 u_int rtableid) 3888 { 3889 struct syn_cache *sc; 3890 struct syn_cache_head *scp; 3891 int s; 3892 3893 s = splsoftnet(); 3894 if ((sc = syn_cache_lookup(src, dst, &scp, rtableid)) == NULL) { 3895 splx(s); 3896 return; 3897 } 3898 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3899 if (ntohl (th->th_seq) != sc->sc_iss) { 3900 splx(s); 3901 return; 3902 } 3903 3904 /* 3905 * If we've retransmitted 3 times and this is our second error, 3906 * we remove the entry. Otherwise, we allow it to continue on. 3907 * This prevents us from incorrectly nuking an entry during a 3908 * spurious network outage. 3909 * 3910 * See tcp_notify(). 3911 */ 3912 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3913 sc->sc_flags |= SCF_UNREACH; 3914 splx(s); 3915 return; 3916 } 3917 3918 syn_cache_rm(sc); 3919 splx(s); 3920 tcpstat.tcps_sc_unreach++; 3921 syn_cache_put(sc); 3922 } 3923 3924 /* 3925 * Given a LISTEN socket and an inbound SYN request, add 3926 * this to the syn cache, and send back a segment: 3927 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3928 * to the source. 3929 * 3930 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3931 * Doing so would require that we hold onto the data and deliver it 3932 * to the application. However, if we are the target of a SYN-flood 3933 * DoS attack, an attacker could send data which would eventually 3934 * consume all available buffer space if it were ACKed. By not ACKing 3935 * the data, we avoid this DoS scenario. 3936 */ 3937 3938 int 3939 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3940 u_int iphlen, struct socket *so, struct mbuf *m, u_char *optp, int optlen, 3941 struct tcp_opt_info *oi, tcp_seq *issp) 3942 { 3943 struct tcpcb tb, *tp; 3944 long win; 3945 struct syn_cache *sc; 3946 struct syn_cache_head *scp; 3947 struct mbuf *ipopts; 3948 3949 tp = sototcpcb(so); 3950 3951 /* 3952 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3953 * 3954 * Note this check is performed in tcp_input() very early on. 3955 */ 3956 3957 /* 3958 * Initialize some local state. 3959 */ 3960 win = sbspace(&so->so_rcv); 3961 if (win > TCP_MAXWIN) 3962 win = TCP_MAXWIN; 3963 3964 bzero(&tb, sizeof(tb)); 3965 #ifdef TCP_SIGNATURE 3966 if (optp || (tp->t_flags & TF_SIGNATURE)) { 3967 #else 3968 if (optp) { 3969 #endif 3970 tb.pf = tp->pf; 3971 #ifdef TCP_SACK 3972 tb.sack_enable = tp->sack_enable; 3973 #endif 3974 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3975 #ifdef TCP_SIGNATURE 3976 if (tp->t_flags & TF_SIGNATURE) 3977 tb.t_flags |= TF_SIGNATURE; 3978 #endif 3979 tb.t_state = TCPS_LISTEN; 3980 if (tcp_dooptions(&tb, optp, optlen, th, m, iphlen, oi, 3981 sotoinpcb(so)->inp_rtableid)) 3982 return (-1); 3983 } 3984 3985 switch (src->sa_family) { 3986 case AF_INET: 3987 /* 3988 * Remember the IP options, if any. 3989 */ 3990 ipopts = ip_srcroute(m); 3991 break; 3992 default: 3993 ipopts = NULL; 3994 } 3995 3996 /* 3997 * See if we already have an entry for this connection. 3998 * If we do, resend the SYN,ACK. We do not count this 3999 * as a retransmission (XXX though maybe we should). 4000 */ 4001 if ((sc = syn_cache_lookup(src, dst, &scp, sotoinpcb(so)->inp_rtableid)) 4002 != NULL) { 4003 tcpstat.tcps_sc_dupesyn++; 4004 if (ipopts) { 4005 /* 4006 * If we were remembering a previous source route, 4007 * forget it and use the new one we've been given. 4008 */ 4009 if (sc->sc_ipopts) 4010 (void) m_free(sc->sc_ipopts); 4011 sc->sc_ipopts = ipopts; 4012 } 4013 sc->sc_timestamp = tb.ts_recent; 4014 if (syn_cache_respond(sc, m) == 0) { 4015 tcpstat.tcps_sndacks++; 4016 tcpstat.tcps_sndtotal++; 4017 } 4018 return (0); 4019 } 4020 4021 sc = pool_get(&syn_cache_pool, PR_NOWAIT|PR_ZERO); 4022 if (sc == NULL) { 4023 if (ipopts) 4024 (void) m_free(ipopts); 4025 return (-1); 4026 } 4027 4028 /* 4029 * Fill in the cache, and put the necessary IP and TCP 4030 * options into the reply. 4031 */ 4032 bcopy(src, &sc->sc_src, src->sa_len); 4033 bcopy(dst, &sc->sc_dst, dst->sa_len); 4034 sc->sc_rtableid = sotoinpcb(so)->inp_rtableid; 4035 sc->sc_flags = 0; 4036 sc->sc_ipopts = ipopts; 4037 sc->sc_irs = th->th_seq; 4038 4039 sc->sc_iss = issp ? *issp : arc4random(); 4040 sc->sc_peermaxseg = oi->maxseg; 4041 sc->sc_ourmaxseg = tcp_mss_adv(m->m_flags & M_PKTHDR ? 4042 m->m_pkthdr.rcvif : NULL, sc->sc_src.sa.sa_family); 4043 sc->sc_win = win; 4044 sc->sc_timestamp = tb.ts_recent; 4045 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4046 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) { 4047 sc->sc_flags |= SCF_TIMESTAMP; 4048 sc->sc_modulate = arc4random(); 4049 } 4050 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4051 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4052 sc->sc_requested_s_scale = tb.requested_s_scale; 4053 sc->sc_request_r_scale = 0; 4054 /* 4055 * Pick the smallest possible scaling factor that 4056 * will still allow us to scale up to sb_max. 4057 * 4058 * We do this because there are broken firewalls that 4059 * will corrupt the window scale option, leading to 4060 * the other endpoint believing that our advertised 4061 * window is unscaled. At scale factors larger than 4062 * 5 the unscaled window will drop below 1500 bytes, 4063 * leading to serious problems when traversing these 4064 * broken firewalls. 4065 * 4066 * With the default sbmax of 256K, a scale factor 4067 * of 3 will be chosen by this algorithm. Those who 4068 * choose a larger sbmax should watch out 4069 * for the compatiblity problems mentioned above. 4070 * 4071 * RFC1323: The Window field in a SYN (i.e., a <SYN> 4072 * or <SYN,ACK>) segment itself is never scaled. 4073 */ 4074 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4075 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 4076 sc->sc_request_r_scale++; 4077 } else { 4078 sc->sc_requested_s_scale = 15; 4079 sc->sc_request_r_scale = 15; 4080 } 4081 #ifdef TCP_ECN 4082 /* 4083 * if both ECE and CWR flag bits are set, peer is ECN capable. 4084 */ 4085 if (tcp_do_ecn && 4086 (th->th_flags & (TH_ECE|TH_CWR)) == (TH_ECE|TH_CWR)) 4087 sc->sc_flags |= SCF_ECN_PERMIT; 4088 #endif 4089 #ifdef TCP_SACK 4090 /* 4091 * Set SCF_SACK_PERMIT if peer did send a SACK_PERMITTED option 4092 * (i.e., if tcp_dooptions() did set TF_SACK_PERMIT). 4093 */ 4094 if (tb.sack_enable && (tb.t_flags & TF_SACK_PERMIT)) 4095 sc->sc_flags |= SCF_SACK_PERMIT; 4096 #endif 4097 #ifdef TCP_SIGNATURE 4098 if (tb.t_flags & TF_SIGNATURE) 4099 sc->sc_flags |= SCF_SIGNATURE; 4100 #endif 4101 sc->sc_tp = tp; 4102 if (syn_cache_respond(sc, m) == 0) { 4103 syn_cache_insert(sc, tp); 4104 tcpstat.tcps_sndacks++; 4105 tcpstat.tcps_sndtotal++; 4106 } else { 4107 syn_cache_put(sc); 4108 tcpstat.tcps_sc_dropped++; 4109 } 4110 4111 return (0); 4112 } 4113 4114 int 4115 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 4116 { 4117 struct route *ro; 4118 u_int8_t *optp; 4119 int optlen, error; 4120 u_int16_t tlen; 4121 struct ip *ip = NULL; 4122 #ifdef INET6 4123 struct ip6_hdr *ip6 = NULL; 4124 #endif 4125 struct tcphdr *th; 4126 u_int hlen; 4127 struct inpcb *inp; 4128 4129 switch (sc->sc_src.sa.sa_family) { 4130 case AF_INET: 4131 hlen = sizeof(struct ip); 4132 ro = &sc->sc_route4; 4133 break; 4134 #ifdef INET6 4135 case AF_INET6: 4136 hlen = sizeof(struct ip6_hdr); 4137 ro = (struct route *)&sc->sc_route6; 4138 break; 4139 #endif 4140 default: 4141 if (m) 4142 m_freem(m); 4143 return (EAFNOSUPPORT); 4144 } 4145 4146 /* Compute the size of the TCP options. */ 4147 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4148 #ifdef TCP_SACK 4149 ((sc->sc_flags & SCF_SACK_PERMIT) ? 4 : 0) + 4150 #endif 4151 #ifdef TCP_SIGNATURE 4152 ((sc->sc_flags & SCF_SIGNATURE) ? TCPOLEN_SIGLEN : 0) + 4153 #endif 4154 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4155 4156 tlen = hlen + sizeof(struct tcphdr) + optlen; 4157 4158 /* 4159 * Create the IP+TCP header from scratch. 4160 */ 4161 if (m) 4162 m_freem(m); 4163 #ifdef DIAGNOSTIC 4164 if (max_linkhdr + tlen > MCLBYTES) 4165 return (ENOBUFS); 4166 #endif 4167 MGETHDR(m, M_DONTWAIT, MT_DATA); 4168 if (m && max_linkhdr + tlen > MHLEN) { 4169 MCLGET(m, M_DONTWAIT); 4170 if ((m->m_flags & M_EXT) == 0) { 4171 m_freem(m); 4172 m = NULL; 4173 } 4174 } 4175 if (m == NULL) 4176 return (ENOBUFS); 4177 4178 /* Fixup the mbuf. */ 4179 m->m_data += max_linkhdr; 4180 m->m_len = m->m_pkthdr.len = tlen; 4181 m->m_pkthdr.rcvif = NULL; 4182 m->m_pkthdr.ph_rtableid = sc->sc_rtableid; 4183 memset(mtod(m, u_char *), 0, tlen); 4184 4185 switch (sc->sc_src.sa.sa_family) { 4186 case AF_INET: 4187 ip = mtod(m, struct ip *); 4188 ip->ip_dst = sc->sc_src.sin.sin_addr; 4189 ip->ip_src = sc->sc_dst.sin.sin_addr; 4190 ip->ip_p = IPPROTO_TCP; 4191 th = (struct tcphdr *)(ip + 1); 4192 th->th_dport = sc->sc_src.sin.sin_port; 4193 th->th_sport = sc->sc_dst.sin.sin_port; 4194 break; 4195 #ifdef INET6 4196 case AF_INET6: 4197 ip6 = mtod(m, struct ip6_hdr *); 4198 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4199 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4200 ip6->ip6_nxt = IPPROTO_TCP; 4201 /* ip6_plen will be updated in ip6_output() */ 4202 th = (struct tcphdr *)(ip6 + 1); 4203 th->th_dport = sc->sc_src.sin6.sin6_port; 4204 th->th_sport = sc->sc_dst.sin6.sin6_port; 4205 break; 4206 #endif 4207 default: 4208 unhandled_af(sc->sc_src.sa.sa_family); 4209 } 4210 4211 th->th_seq = htonl(sc->sc_iss); 4212 th->th_ack = htonl(sc->sc_irs + 1); 4213 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4214 th->th_flags = TH_SYN|TH_ACK; 4215 #ifdef TCP_ECN 4216 /* Set ECE for SYN-ACK if peer supports ECN. */ 4217 if (tcp_do_ecn && (sc->sc_flags & SCF_ECN_PERMIT)) 4218 th->th_flags |= TH_ECE; 4219 #endif 4220 th->th_win = htons(sc->sc_win); 4221 /* th_sum already 0 */ 4222 /* th_urp already 0 */ 4223 4224 /* Tack on the TCP options. */ 4225 optp = (u_int8_t *)(th + 1); 4226 *optp++ = TCPOPT_MAXSEG; 4227 *optp++ = 4; 4228 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4229 *optp++ = sc->sc_ourmaxseg & 0xff; 4230 4231 #ifdef TCP_SACK 4232 /* Include SACK_PERMIT_HDR option if peer has already done so. */ 4233 if (sc->sc_flags & SCF_SACK_PERMIT) { 4234 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMIT_HDR); 4235 optp += 4; 4236 } 4237 #endif 4238 4239 if (sc->sc_request_r_scale != 15) { 4240 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4241 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4242 sc->sc_request_r_scale); 4243 optp += 4; 4244 } 4245 4246 if (sc->sc_flags & SCF_TIMESTAMP) { 4247 u_int32_t *lp = (u_int32_t *)(optp); 4248 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4249 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4250 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4251 *lp = htonl(sc->sc_timestamp); 4252 optp += TCPOLEN_TSTAMP_APPA; 4253 } 4254 4255 #ifdef TCP_SIGNATURE 4256 if (sc->sc_flags & SCF_SIGNATURE) { 4257 union sockaddr_union src, dst; 4258 struct tdb *tdb; 4259 4260 bzero(&src, sizeof(union sockaddr_union)); 4261 bzero(&dst, sizeof(union sockaddr_union)); 4262 src.sa.sa_len = sc->sc_src.sa.sa_len; 4263 src.sa.sa_family = sc->sc_src.sa.sa_family; 4264 dst.sa.sa_len = sc->sc_dst.sa.sa_len; 4265 dst.sa.sa_family = sc->sc_dst.sa.sa_family; 4266 4267 switch (sc->sc_src.sa.sa_family) { 4268 case 0: /*default to PF_INET*/ 4269 case AF_INET: 4270 src.sin.sin_addr = mtod(m, struct ip *)->ip_src; 4271 dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst; 4272 break; 4273 #ifdef INET6 4274 case AF_INET6: 4275 src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src; 4276 dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst; 4277 break; 4278 #endif /* INET6 */ 4279 } 4280 4281 tdb = gettdbbysrcdst(rtable_l2(sc->sc_rtableid), 4282 0, &src, &dst, IPPROTO_TCP); 4283 if (tdb == NULL) { 4284 if (m) 4285 m_freem(m); 4286 return (EPERM); 4287 } 4288 4289 /* Send signature option */ 4290 *(optp++) = TCPOPT_SIGNATURE; 4291 *(optp++) = TCPOLEN_SIGNATURE; 4292 4293 if (tcp_signature(tdb, sc->sc_src.sa.sa_family, m, th, 4294 hlen, 0, optp) < 0) { 4295 if (m) 4296 m_freem(m); 4297 return (EINVAL); 4298 } 4299 optp += 16; 4300 4301 /* Pad options list to the next 32 bit boundary and 4302 * terminate it. 4303 */ 4304 *optp++ = TCPOPT_NOP; 4305 *optp++ = TCPOPT_EOL; 4306 } 4307 #endif /* TCP_SIGNATURE */ 4308 4309 /* Compute the packet's checksum. */ 4310 switch (sc->sc_src.sa.sa_family) { 4311 case AF_INET: 4312 ip->ip_len = htons(tlen - hlen); 4313 th->th_sum = 0; 4314 th->th_sum = in_cksum(m, tlen); 4315 break; 4316 #ifdef INET6 4317 case AF_INET6: 4318 ip6->ip6_plen = htons(tlen - hlen); 4319 th->th_sum = 0; 4320 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4321 break; 4322 #endif 4323 } 4324 4325 /* use IPsec policy and ttl from listening socket, on SYN ACK */ 4326 inp = sc->sc_tp ? sc->sc_tp->t_inpcb : NULL; 4327 4328 /* 4329 * Fill in some straggling IP bits. Note the stack expects 4330 * ip_len to be in host order, for convenience. 4331 */ 4332 switch (sc->sc_src.sa.sa_family) { 4333 case AF_INET: 4334 ip->ip_len = htons(tlen); 4335 ip->ip_ttl = inp ? inp->inp_ip.ip_ttl : ip_defttl; 4336 if (inp != NULL) 4337 ip->ip_tos = inp->inp_ip.ip_tos; 4338 break; 4339 #ifdef INET6 4340 case AF_INET6: 4341 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4342 ip6->ip6_vfc |= IPV6_VERSION; 4343 ip6->ip6_plen = htons(tlen - hlen); 4344 /* ip6_hlim will be initialized afterwards */ 4345 /* leave flowlabel = 0, it is legal and require no state mgmt */ 4346 break; 4347 #endif 4348 } 4349 4350 switch (sc->sc_src.sa.sa_family) { 4351 case AF_INET: 4352 error = ip_output(m, sc->sc_ipopts, ro, 4353 (ip_mtudisc ? IP_MTUDISC : 0), NULL, inp, 0); 4354 break; 4355 #ifdef INET6 4356 case AF_INET6: 4357 ip6->ip6_hlim = in6_selecthlim(NULL, 4358 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4359 4360 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4361 NULL, NULL, NULL); 4362 break; 4363 #endif 4364 default: 4365 error = EAFNOSUPPORT; 4366 break; 4367 } 4368 return (error); 4369 } 4370