1 /* $OpenBSD: tcp_input.c,v 1.365 2020/06/19 22:47:22 procter 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/tcp_debug.h> 97 98 #if NPF > 0 99 #include <net/pfvar.h> 100 #endif 101 102 struct tcpiphdr tcp_saveti; 103 104 int tcp_mss_adv(struct mbuf *, int); 105 int tcp_flush_queue(struct tcpcb *); 106 107 #ifdef INET6 108 #include <netinet6/in6_var.h> 109 #include <netinet6/nd6.h> 110 111 struct tcpipv6hdr tcp_saveti6; 112 113 /* for the packet header length in the mbuf */ 114 #define M_PH_LEN(m) (((struct mbuf *)(m))->m_pkthdr.len) 115 #define M_V6_LEN(m) (M_PH_LEN(m) - sizeof(struct ip6_hdr)) 116 #define M_V4_LEN(m) (M_PH_LEN(m) - sizeof(struct ip)) 117 #endif /* INET6 */ 118 119 int tcprexmtthresh = 3; 120 int tcptv_keep_init = TCPTV_KEEP_INIT; 121 122 int tcp_rst_ppslim = 100; /* 100pps */ 123 int tcp_rst_ppslim_count = 0; 124 struct timeval tcp_rst_ppslim_last; 125 126 int tcp_ackdrop_ppslim = 100; /* 100pps */ 127 int tcp_ackdrop_ppslim_count = 0; 128 struct timeval tcp_ackdrop_ppslim_last; 129 130 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 131 132 /* for modulo comparisons of timestamps */ 133 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 134 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 135 136 /* for TCP SACK comparisons */ 137 #define SEQ_MIN(a,b) (SEQ_LT(a,b) ? (a) : (b)) 138 #define SEQ_MAX(a,b) (SEQ_GT(a,b) ? (a) : (b)) 139 140 /* 141 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 142 */ 143 #ifdef INET6 144 #define ND6_HINT(tp) \ 145 do { \ 146 if (tp && tp->t_inpcb && (tp->t_inpcb->inp_flags & INP_IPV6) && \ 147 rtisvalid(tp->t_inpcb->inp_route6.ro_rt)) { \ 148 nd6_nud_hint(tp->t_inpcb->inp_route6.ro_rt); \ 149 } \ 150 } while (0) 151 #else 152 #define ND6_HINT(tp) 153 #endif 154 155 #ifdef TCP_ECN 156 /* 157 * ECN (Explicit Congestion Notification) support based on RFC3168 158 * implementation note: 159 * snd_last is used to track a recovery phase. 160 * when cwnd is reduced, snd_last is set to snd_max. 161 * while snd_last > snd_una, the sender is in a recovery phase and 162 * its cwnd should not be reduced again. 163 * snd_last follows snd_una when not in a recovery phase. 164 */ 165 #endif 166 167 /* 168 * Macro to compute ACK transmission behavior. Delay the ACK unless 169 * we have already delayed an ACK (must send an ACK every two segments). 170 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 171 * option is enabled or when the packet is coming from a loopback 172 * interface. 173 */ 174 #define TCP_SETUP_ACK(tp, tiflags, m) \ 175 do { \ 176 struct ifnet *ifp = NULL; \ 177 if (m && (m->m_flags & M_PKTHDR)) \ 178 ifp = if_get(m->m_pkthdr.ph_ifidx); \ 179 if (TCP_TIMER_ISARMED(tp, TCPT_DELACK) || \ 180 (tcp_ack_on_push && (tiflags) & TH_PUSH) || \ 181 (ifp && (ifp->if_flags & IFF_LOOPBACK))) \ 182 tp->t_flags |= TF_ACKNOW; \ 183 else \ 184 TCP_TIMER_ARM_MSEC(tp, TCPT_DELACK, tcp_delack_msecs); \ 185 if_put(ifp); \ 186 } while (0) 187 188 void tcp_sack_partialack(struct tcpcb *, struct tcphdr *); 189 void tcp_newreno_partialack(struct tcpcb *, struct tcphdr *); 190 191 void syn_cache_put(struct syn_cache *); 192 void syn_cache_rm(struct syn_cache *); 193 int syn_cache_respond(struct syn_cache *, struct mbuf *); 194 void syn_cache_timer(void *); 195 void syn_cache_reaper(void *); 196 void syn_cache_insert(struct syn_cache *, struct tcpcb *); 197 void syn_cache_reset(struct sockaddr *, struct sockaddr *, 198 struct tcphdr *, u_int); 199 int syn_cache_add(struct sockaddr *, struct sockaddr *, struct tcphdr *, 200 unsigned int, struct socket *, struct mbuf *, u_char *, int, 201 struct tcp_opt_info *, tcp_seq *); 202 struct socket *syn_cache_get(struct sockaddr *, struct sockaddr *, 203 struct tcphdr *, unsigned int, unsigned int, struct socket *, 204 struct mbuf *); 205 struct syn_cache *syn_cache_lookup(struct sockaddr *, struct sockaddr *, 206 struct syn_cache_head **, u_int); 207 208 /* 209 * Insert segment ti into reassembly queue of tcp with 210 * control block tp. Return TH_FIN if reassembly now includes 211 * a segment with FIN. The macro form does the common case inline 212 * (segment is the next to be received on an established connection, 213 * and the queue is empty), avoiding linkage into and removal 214 * from the queue and repetition of various conversions. 215 * Set DELACK for segments received in order, but ack immediately 216 * when segments are out of order (so fast retransmit can work). 217 */ 218 219 int 220 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen) 221 { 222 struct tcpqent *p, *q, *nq, *tiqe; 223 224 /* 225 * Allocate a new queue entry, before we throw away any data. 226 * If we can't, just drop the packet. XXX 227 */ 228 tiqe = pool_get(&tcpqe_pool, PR_NOWAIT); 229 if (tiqe == NULL) { 230 tiqe = TAILQ_LAST(&tp->t_segq, tcpqehead); 231 if (tiqe != NULL && th->th_seq == tp->rcv_nxt) { 232 /* Reuse last entry since new segment fills a hole */ 233 m_freem(tiqe->tcpqe_m); 234 TAILQ_REMOVE(&tp->t_segq, tiqe, tcpqe_q); 235 } 236 if (tiqe == NULL || th->th_seq != tp->rcv_nxt) { 237 /* Flush segment queue for this connection */ 238 tcp_freeq(tp); 239 tcpstat_inc(tcps_rcvmemdrop); 240 m_freem(m); 241 return (0); 242 } 243 } 244 245 /* 246 * Find a segment which begins after this one does. 247 */ 248 for (p = NULL, q = TAILQ_FIRST(&tp->t_segq); q != NULL; 249 p = q, q = TAILQ_NEXT(q, tcpqe_q)) 250 if (SEQ_GT(q->tcpqe_tcp->th_seq, th->th_seq)) 251 break; 252 253 /* 254 * If there is a preceding segment, it may provide some of 255 * our data already. If so, drop the data from the incoming 256 * segment. If it provides all of our data, drop us. 257 */ 258 if (p != NULL) { 259 struct tcphdr *phdr = p->tcpqe_tcp; 260 int i; 261 262 /* conversion to int (in i) handles seq wraparound */ 263 i = phdr->th_seq + phdr->th_reseqlen - th->th_seq; 264 if (i > 0) { 265 if (i >= *tlen) { 266 tcpstat_pkt(tcps_rcvduppack, tcps_rcvdupbyte, 267 *tlen); 268 m_freem(m); 269 pool_put(&tcpqe_pool, tiqe); 270 return (0); 271 } 272 m_adj(m, i); 273 *tlen -= i; 274 th->th_seq += i; 275 } 276 } 277 tcpstat_pkt(tcps_rcvoopack, tcps_rcvoobyte, *tlen); 278 279 /* 280 * While we overlap succeeding segments trim them or, 281 * if they are completely covered, dequeue them. 282 */ 283 for (; q != NULL; q = nq) { 284 struct tcphdr *qhdr = q->tcpqe_tcp; 285 int i = (th->th_seq + *tlen) - qhdr->th_seq; 286 287 if (i <= 0) 288 break; 289 if (i < qhdr->th_reseqlen) { 290 qhdr->th_seq += i; 291 qhdr->th_reseqlen -= i; 292 m_adj(q->tcpqe_m, i); 293 break; 294 } 295 nq = TAILQ_NEXT(q, tcpqe_q); 296 m_freem(q->tcpqe_m); 297 TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q); 298 pool_put(&tcpqe_pool, q); 299 } 300 301 /* Insert the new segment queue entry into place. */ 302 tiqe->tcpqe_m = m; 303 th->th_reseqlen = *tlen; 304 tiqe->tcpqe_tcp = th; 305 if (p == NULL) { 306 TAILQ_INSERT_HEAD(&tp->t_segq, tiqe, tcpqe_q); 307 } else { 308 TAILQ_INSERT_AFTER(&tp->t_segq, p, tiqe, tcpqe_q); 309 } 310 311 if (th->th_seq != tp->rcv_nxt) 312 return (0); 313 314 return (tcp_flush_queue(tp)); 315 } 316 317 int 318 tcp_flush_queue(struct tcpcb *tp) 319 { 320 struct socket *so = tp->t_inpcb->inp_socket; 321 struct tcpqent *q, *nq; 322 int flags; 323 324 /* 325 * Present data to user, advancing rcv_nxt through 326 * completed sequence space. 327 */ 328 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 329 return (0); 330 q = TAILQ_FIRST(&tp->t_segq); 331 if (q == NULL || q->tcpqe_tcp->th_seq != tp->rcv_nxt) 332 return (0); 333 if (tp->t_state == TCPS_SYN_RECEIVED && q->tcpqe_tcp->th_reseqlen) 334 return (0); 335 do { 336 tp->rcv_nxt += q->tcpqe_tcp->th_reseqlen; 337 flags = q->tcpqe_tcp->th_flags & TH_FIN; 338 339 nq = TAILQ_NEXT(q, tcpqe_q); 340 TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q); 341 ND6_HINT(tp); 342 if (so->so_state & SS_CANTRCVMORE) 343 m_freem(q->tcpqe_m); 344 else 345 sbappendstream(so, &so->so_rcv, q->tcpqe_m); 346 pool_put(&tcpqe_pool, q); 347 q = nq; 348 } while (q != NULL && q->tcpqe_tcp->th_seq == tp->rcv_nxt); 349 tp->t_flags |= TF_BLOCKOUTPUT; 350 sorwakeup(so); 351 tp->t_flags &= ~TF_BLOCKOUTPUT; 352 return (flags); 353 } 354 355 /* 356 * TCP input routine, follows pages 65-76 of the 357 * protocol specification dated September, 1981 very closely. 358 */ 359 int 360 tcp_input(struct mbuf **mp, int *offp, int proto, int af) 361 { 362 struct mbuf *m = *mp; 363 int iphlen = *offp; 364 struct ip *ip = NULL; 365 struct inpcb *inp = NULL; 366 u_int8_t *optp = NULL; 367 int optlen = 0; 368 int tlen, off; 369 struct tcpcb *otp = NULL, *tp = NULL; 370 int tiflags; 371 struct socket *so = NULL; 372 int todrop, acked, ourfinisacked; 373 int hdroptlen = 0; 374 short ostate; 375 caddr_t saveti; 376 tcp_seq iss, *reuse = NULL; 377 u_long tiwin; 378 struct tcp_opt_info opti; 379 struct tcphdr *th; 380 #ifdef INET6 381 struct ip6_hdr *ip6 = NULL; 382 #endif /* INET6 */ 383 #ifdef IPSEC 384 struct m_tag *mtag; 385 struct tdb_ident *tdbi; 386 struct tdb *tdb; 387 int error; 388 #endif /* IPSEC */ 389 #ifdef TCP_ECN 390 u_char iptos; 391 #endif 392 393 tcpstat_inc(tcps_rcvtotal); 394 395 opti.ts_present = 0; 396 opti.maxseg = 0; 397 398 /* 399 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 400 */ 401 if (m->m_flags & (M_BCAST|M_MCAST)) 402 goto drop; 403 404 /* 405 * Get IP and TCP header together in first mbuf. 406 * Note: IP leaves IP header in first mbuf. 407 */ 408 IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, sizeof(*th)); 409 if (!th) { 410 tcpstat_inc(tcps_rcvshort); 411 return IPPROTO_DONE; 412 } 413 414 tlen = m->m_pkthdr.len - iphlen; 415 switch (af) { 416 case AF_INET: 417 ip = mtod(m, struct ip *); 418 #ifdef TCP_ECN 419 /* save ip_tos before clearing it for checksum */ 420 iptos = ip->ip_tos; 421 #endif 422 break; 423 #ifdef INET6 424 case AF_INET6: 425 ip6 = mtod(m, struct ip6_hdr *); 426 #ifdef TCP_ECN 427 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 428 #endif 429 430 /* 431 * Be proactive about unspecified IPv6 address in source. 432 * As we use all-zero to indicate unbounded/unconnected pcb, 433 * unspecified IPv6 address can be used to confuse us. 434 * 435 * Note that packets with unspecified IPv6 destination is 436 * already dropped in ip6_input. 437 */ 438 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 439 /* XXX stat */ 440 goto drop; 441 } 442 443 /* Discard packets to multicast */ 444 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 445 /* XXX stat */ 446 goto drop; 447 } 448 break; 449 #endif 450 default: 451 unhandled_af(af); 452 } 453 454 /* 455 * Checksum extended TCP header and data. 456 */ 457 if ((m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_OK) == 0) { 458 int sum; 459 460 if (m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_BAD) { 461 tcpstat_inc(tcps_rcvbadsum); 462 goto drop; 463 } 464 tcpstat_inc(tcps_inswcsum); 465 switch (af) { 466 case AF_INET: 467 sum = in4_cksum(m, IPPROTO_TCP, iphlen, tlen); 468 break; 469 #ifdef INET6 470 case AF_INET6: 471 sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 472 tlen); 473 break; 474 #endif 475 } 476 if (sum != 0) { 477 tcpstat_inc(tcps_rcvbadsum); 478 goto drop; 479 } 480 } 481 482 /* 483 * Check that TCP offset makes sense, 484 * pull out TCP options and adjust length. XXX 485 */ 486 off = th->th_off << 2; 487 if (off < sizeof(struct tcphdr) || off > tlen) { 488 tcpstat_inc(tcps_rcvbadoff); 489 goto drop; 490 } 491 tlen -= off; 492 if (off > sizeof(struct tcphdr)) { 493 IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, off); 494 if (!th) { 495 tcpstat_inc(tcps_rcvshort); 496 return IPPROTO_DONE; 497 } 498 optlen = off - sizeof(struct tcphdr); 499 optp = (u_int8_t *)(th + 1); 500 /* 501 * Do quick retrieval of timestamp options ("options 502 * prediction?"). If timestamp is the only option and it's 503 * formatted as recommended in RFC 1323 appendix A, we 504 * quickly get the values now and not bother calling 505 * tcp_dooptions(), etc. 506 */ 507 if ((optlen == TCPOLEN_TSTAMP_APPA || 508 (optlen > TCPOLEN_TSTAMP_APPA && 509 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 510 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 511 (th->th_flags & TH_SYN) == 0) { 512 opti.ts_present = 1; 513 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 514 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 515 optp = NULL; /* we've parsed the options */ 516 } 517 } 518 tiflags = th->th_flags; 519 520 /* 521 * Convert TCP protocol specific fields to host format. 522 */ 523 th->th_seq = ntohl(th->th_seq); 524 th->th_ack = ntohl(th->th_ack); 525 th->th_win = ntohs(th->th_win); 526 th->th_urp = ntohs(th->th_urp); 527 528 /* 529 * Locate pcb for segment. 530 */ 531 #if NPF > 0 532 inp = pf_inp_lookup(m); 533 #endif 534 findpcb: 535 if (inp == NULL) { 536 switch (af) { 537 #ifdef INET6 538 case AF_INET6: 539 inp = in6_pcbhashlookup(&tcbtable, &ip6->ip6_src, 540 th->th_sport, &ip6->ip6_dst, th->th_dport, 541 m->m_pkthdr.ph_rtableid); 542 break; 543 #endif 544 case AF_INET: 545 inp = in_pcbhashlookup(&tcbtable, ip->ip_src, 546 th->th_sport, ip->ip_dst, th->th_dport, 547 m->m_pkthdr.ph_rtableid); 548 break; 549 } 550 } 551 if (inp == NULL) { 552 tcpstat_inc(tcps_pcbhashmiss); 553 switch (af) { 554 #ifdef INET6 555 case AF_INET6: 556 inp = in6_pcblookup_listen(&tcbtable, &ip6->ip6_dst, 557 th->th_dport, m, m->m_pkthdr.ph_rtableid); 558 break; 559 #endif /* INET6 */ 560 case AF_INET: 561 inp = in_pcblookup_listen(&tcbtable, ip->ip_dst, 562 th->th_dport, m, m->m_pkthdr.ph_rtableid); 563 break; 564 } 565 /* 566 * If the state is CLOSED (i.e., TCB does not exist) then 567 * all data in the incoming segment is discarded. 568 * If the TCB exists but is in CLOSED state, it is embryonic, 569 * but should either do a listen or a connect soon. 570 */ 571 } 572 #ifdef IPSEC 573 if (ipsec_in_use) { 574 /* Find most recent IPsec tag */ 575 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL); 576 if (mtag != NULL) { 577 tdbi = (struct tdb_ident *)(mtag + 1); 578 tdb = gettdb(tdbi->rdomain, tdbi->spi, 579 &tdbi->dst, tdbi->proto); 580 } else 581 tdb = NULL; 582 ipsp_spd_lookup(m, af, iphlen, &error, IPSP_DIRECTION_IN, 583 tdb, inp, 0); 584 if (error) { 585 tcpstat_inc(tcps_rcvnosec); 586 goto drop; 587 } 588 } 589 #endif /* IPSEC */ 590 591 if (inp == NULL) { 592 tcpstat_inc(tcps_noport); 593 goto dropwithreset_ratelim; 594 } 595 596 KASSERT(sotoinpcb(inp->inp_socket) == inp); 597 KASSERT(intotcpcb(inp) == NULL || intotcpcb(inp)->t_inpcb == inp); 598 soassertlocked(inp->inp_socket); 599 600 /* Check the minimum TTL for socket. */ 601 switch (af) { 602 case AF_INET: 603 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) 604 goto drop; 605 break; 606 #ifdef INET6 607 case AF_INET6: 608 if (inp->inp_ip6_minhlim && 609 inp->inp_ip6_minhlim > ip6->ip6_hlim) 610 goto drop; 611 break; 612 #endif 613 } 614 615 tp = intotcpcb(inp); 616 if (tp == NULL) 617 goto dropwithreset_ratelim; 618 if (tp->t_state == TCPS_CLOSED) 619 goto drop; 620 621 /* Unscale the window into a 32-bit value. */ 622 if ((tiflags & TH_SYN) == 0) 623 tiwin = th->th_win << tp->snd_scale; 624 else 625 tiwin = th->th_win; 626 627 so = inp->inp_socket; 628 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 629 union syn_cache_sa src; 630 union syn_cache_sa dst; 631 632 bzero(&src, sizeof(src)); 633 bzero(&dst, sizeof(dst)); 634 switch (af) { 635 case AF_INET: 636 src.sin.sin_len = sizeof(struct sockaddr_in); 637 src.sin.sin_family = AF_INET; 638 src.sin.sin_addr = ip->ip_src; 639 src.sin.sin_port = th->th_sport; 640 641 dst.sin.sin_len = sizeof(struct sockaddr_in); 642 dst.sin.sin_family = AF_INET; 643 dst.sin.sin_addr = ip->ip_dst; 644 dst.sin.sin_port = th->th_dport; 645 break; 646 #ifdef INET6 647 case AF_INET6: 648 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 649 src.sin6.sin6_family = AF_INET6; 650 src.sin6.sin6_addr = ip6->ip6_src; 651 src.sin6.sin6_port = th->th_sport; 652 653 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 654 dst.sin6.sin6_family = AF_INET6; 655 dst.sin6.sin6_addr = ip6->ip6_dst; 656 dst.sin6.sin6_port = th->th_dport; 657 break; 658 #endif /* INET6 */ 659 } 660 661 if (so->so_options & SO_DEBUG) { 662 otp = tp; 663 ostate = tp->t_state; 664 switch (af) { 665 #ifdef INET6 666 case AF_INET6: 667 saveti = (caddr_t) &tcp_saveti6; 668 memcpy(&tcp_saveti6.ti6_i, ip6, sizeof(*ip6)); 669 memcpy(&tcp_saveti6.ti6_t, th, sizeof(*th)); 670 break; 671 #endif 672 case AF_INET: 673 saveti = (caddr_t) &tcp_saveti; 674 memcpy(&tcp_saveti.ti_i, ip, sizeof(*ip)); 675 memcpy(&tcp_saveti.ti_t, th, sizeof(*th)); 676 break; 677 } 678 } 679 if (so->so_options & SO_ACCEPTCONN) { 680 switch (tiflags & (TH_RST|TH_SYN|TH_ACK)) { 681 682 case TH_SYN|TH_ACK|TH_RST: 683 case TH_SYN|TH_RST: 684 case TH_ACK|TH_RST: 685 case TH_RST: 686 syn_cache_reset(&src.sa, &dst.sa, th, 687 inp->inp_rtableid); 688 goto drop; 689 690 case TH_SYN|TH_ACK: 691 /* 692 * Received a SYN,ACK. This should 693 * never happen while we are in 694 * LISTEN. Send an RST. 695 */ 696 goto badsyn; 697 698 case TH_ACK: 699 so = syn_cache_get(&src.sa, &dst.sa, 700 th, iphlen, tlen, so, m); 701 if (so == NULL) { 702 /* 703 * We don't have a SYN for 704 * this ACK; send an RST. 705 */ 706 goto badsyn; 707 } else if (so == (struct socket *)(-1)) { 708 /* 709 * We were unable to create 710 * the connection. If the 711 * 3-way handshake was 712 * completed, and RST has 713 * been sent to the peer. 714 * Since the mbuf might be 715 * in use for the reply, 716 * do not free it. 717 */ 718 m = *mp = NULL; 719 goto drop; 720 } else { 721 /* 722 * We have created a 723 * full-blown connection. 724 */ 725 tp = NULL; 726 inp = sotoinpcb(so); 727 tp = intotcpcb(inp); 728 if (tp == NULL) 729 goto badsyn; /*XXX*/ 730 731 } 732 break; 733 734 default: 735 /* 736 * None of RST, SYN or ACK was set. 737 * This is an invalid packet for a 738 * TCB in LISTEN state. Send a RST. 739 */ 740 goto badsyn; 741 742 case TH_SYN: 743 /* 744 * Received a SYN. 745 */ 746 #ifdef INET6 747 /* 748 * If deprecated address is forbidden, we do 749 * not accept SYN to deprecated interface 750 * address to prevent any new inbound 751 * connection from getting established. 752 * When we do not accept SYN, we send a TCP 753 * RST, with deprecated source address (instead 754 * of dropping it). We compromise it as it is 755 * much better for peer to send a RST, and 756 * RST will be the final packet for the 757 * exchange. 758 * 759 * If we do not forbid deprecated addresses, we 760 * accept the SYN packet. RFC2462 does not 761 * suggest dropping SYN in this case. 762 * If we decipher RFC2462 5.5.4, it says like 763 * this: 764 * 1. use of deprecated addr with existing 765 * communication is okay - "SHOULD continue 766 * to be used" 767 * 2. use of it with new communication: 768 * (2a) "SHOULD NOT be used if alternate 769 * address with sufficient scope is 770 * available" 771 * (2b) nothing mentioned otherwise. 772 * Here we fall into (2b) case as we have no 773 * choice in our source address selection - we 774 * must obey the peer. 775 * 776 * The wording in RFC2462 is confusing, and 777 * there are multiple description text for 778 * deprecated address handling - worse, they 779 * are not exactly the same. I believe 5.5.4 780 * is the best one, so we follow 5.5.4. 781 */ 782 if (ip6 && !ip6_use_deprecated) { 783 struct in6_ifaddr *ia6; 784 struct ifnet *ifp = 785 if_get(m->m_pkthdr.ph_ifidx); 786 787 if (ifp && 788 (ia6 = in6ifa_ifpwithaddr(ifp, 789 &ip6->ip6_dst)) && 790 (ia6->ia6_flags & 791 IN6_IFF_DEPRECATED)) { 792 tp = NULL; 793 if_put(ifp); 794 goto dropwithreset; 795 } 796 if_put(ifp); 797 } 798 #endif 799 800 /* 801 * LISTEN socket received a SYN 802 * from itself? This can't possibly 803 * be valid; drop the packet. 804 */ 805 if (th->th_dport == th->th_sport) { 806 switch (af) { 807 #ifdef INET6 808 case AF_INET6: 809 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 810 &ip6->ip6_dst)) { 811 tcpstat_inc(tcps_badsyn); 812 goto drop; 813 } 814 break; 815 #endif /* INET6 */ 816 case AF_INET: 817 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) { 818 tcpstat_inc(tcps_badsyn); 819 goto drop; 820 } 821 break; 822 } 823 } 824 825 /* 826 * SYN looks ok; create compressed TCP 827 * state for it. 828 */ 829 if (so->so_qlen > so->so_qlimit || 830 syn_cache_add(&src.sa, &dst.sa, th, iphlen, 831 so, m, optp, optlen, &opti, reuse) == -1) { 832 tcpstat_inc(tcps_dropsyn); 833 goto drop; 834 } 835 return IPPROTO_DONE; 836 } 837 } 838 } 839 840 #ifdef DIAGNOSTIC 841 /* 842 * Should not happen now that all embryonic connections 843 * are handled with compressed state. 844 */ 845 if (tp->t_state == TCPS_LISTEN) 846 panic("tcp_input: TCPS_LISTEN"); 847 #endif 848 849 #if NPF > 0 850 pf_inp_link(m, inp); 851 #endif 852 853 /* 854 * Segment received on connection. 855 * Reset idle time and keep-alive timer. 856 */ 857 tp->t_rcvtime = tcp_now; 858 if (TCPS_HAVEESTABLISHED(tp->t_state)) 859 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 860 861 if (tp->sack_enable) 862 tcp_del_sackholes(tp, th); /* Delete stale SACK holes */ 863 864 /* 865 * Process options. 866 */ 867 #ifdef TCP_SIGNATURE 868 if (optp || (tp->t_flags & TF_SIGNATURE)) 869 #else 870 if (optp) 871 #endif 872 if (tcp_dooptions(tp, optp, optlen, th, m, iphlen, &opti, 873 m->m_pkthdr.ph_rtableid)) 874 goto drop; 875 876 if (opti.ts_present && opti.ts_ecr) { 877 int rtt_test; 878 879 /* subtract out the tcp timestamp modulator */ 880 opti.ts_ecr -= tp->ts_modulate; 881 882 /* make sure ts_ecr is sensible */ 883 rtt_test = tcp_now - opti.ts_ecr; 884 if (rtt_test < 0 || rtt_test > TCP_RTT_MAX) 885 opti.ts_ecr = 0; 886 } 887 888 #ifdef TCP_ECN 889 /* if congestion experienced, set ECE bit in subsequent packets. */ 890 if ((iptos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) { 891 tp->t_flags |= TF_RCVD_CE; 892 tcpstat_inc(tcps_ecn_rcvce); 893 } 894 #endif 895 /* 896 * Header prediction: check for the two common cases 897 * of a uni-directional data xfer. If the packet has 898 * no control flags, is in-sequence, the window didn't 899 * change and we're not retransmitting, it's a 900 * candidate. If the length is zero and the ack moved 901 * forward, we're the sender side of the xfer. Just 902 * free the data acked & wake any higher level process 903 * that was blocked waiting for space. If the length 904 * is non-zero and the ack didn't move, we're the 905 * receiver side. If we're getting packets in-order 906 * (the reassembly queue is empty), add the data to 907 * the socket buffer and note that we need a delayed ack. 908 */ 909 if (tp->t_state == TCPS_ESTABLISHED && 910 #ifdef TCP_ECN 911 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) == TH_ACK && 912 #else 913 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 914 #endif 915 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 916 th->th_seq == tp->rcv_nxt && 917 tiwin && tiwin == tp->snd_wnd && 918 tp->snd_nxt == tp->snd_max) { 919 920 /* 921 * If last ACK falls within this segment's sequence numbers, 922 * record the timestamp. 923 * Fix from Braden, see Stevens p. 870 924 */ 925 if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 926 tp->ts_recent_age = tcp_now; 927 tp->ts_recent = opti.ts_val; 928 } 929 930 if (tlen == 0) { 931 if (SEQ_GT(th->th_ack, tp->snd_una) && 932 SEQ_LEQ(th->th_ack, tp->snd_max) && 933 tp->snd_cwnd >= tp->snd_wnd && 934 tp->t_dupacks == 0) { 935 /* 936 * this is a pure ack for outstanding data. 937 */ 938 tcpstat_inc(tcps_predack); 939 if (opti.ts_present && opti.ts_ecr) 940 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr); 941 else if (tp->t_rtttime && 942 SEQ_GT(th->th_ack, tp->t_rtseq)) 943 tcp_xmit_timer(tp, 944 tcp_now - tp->t_rtttime); 945 acked = th->th_ack - tp->snd_una; 946 tcpstat_pkt(tcps_rcvackpack, tcps_rcvackbyte, 947 acked); 948 ND6_HINT(tp); 949 sbdrop(so, &so->so_snd, acked); 950 951 /* 952 * If we had a pending ICMP message that 953 * refers to data that have just been 954 * acknowledged, disregard the recorded ICMP 955 * message. 956 */ 957 if ((tp->t_flags & TF_PMTUD_PEND) && 958 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 959 tp->t_flags &= ~TF_PMTUD_PEND; 960 961 /* 962 * Keep track of the largest chunk of data 963 * acknowledged since last PMTU update 964 */ 965 if (tp->t_pmtud_mss_acked < acked) 966 tp->t_pmtud_mss_acked = acked; 967 968 tp->snd_una = th->th_ack; 969 /* 970 * We want snd_last to track snd_una so 971 * as to avoid sequence wraparound problems 972 * for very large transfers. 973 */ 974 #ifdef TCP_ECN 975 if (SEQ_GT(tp->snd_una, tp->snd_last)) 976 #endif 977 tp->snd_last = tp->snd_una; 978 m_freem(m); 979 980 /* 981 * If all outstanding data are acked, stop 982 * retransmit timer, otherwise restart timer 983 * using current (possibly backed-off) value. 984 * If process is waiting for space, 985 * wakeup/selwakeup/signal. If data 986 * are ready to send, let tcp_output 987 * decide between more output or persist. 988 */ 989 if (tp->snd_una == tp->snd_max) 990 TCP_TIMER_DISARM(tp, TCPT_REXMT); 991 else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 992 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 993 994 tcp_update_sndspace(tp); 995 if (sb_notify(so, &so->so_snd)) { 996 tp->t_flags |= TF_BLOCKOUTPUT; 997 sowwakeup(so); 998 tp->t_flags &= ~TF_BLOCKOUTPUT; 999 } 1000 if (so->so_snd.sb_cc || 1001 tp->t_flags & TF_NEEDOUTPUT) 1002 (void) tcp_output(tp); 1003 return IPPROTO_DONE; 1004 } 1005 } else if (th->th_ack == tp->snd_una && 1006 TAILQ_EMPTY(&tp->t_segq) && 1007 tlen <= sbspace(so, &so->so_rcv)) { 1008 /* 1009 * This is a pure, in-sequence data packet 1010 * with nothing on the reassembly queue and 1011 * we have enough buffer space to take it. 1012 */ 1013 /* Clean receiver SACK report if present */ 1014 if (tp->sack_enable && tp->rcv_numsacks) 1015 tcp_clean_sackreport(tp); 1016 tcpstat_inc(tcps_preddat); 1017 tp->rcv_nxt += tlen; 1018 tcpstat_pkt(tcps_rcvpack, tcps_rcvbyte, tlen); 1019 ND6_HINT(tp); 1020 1021 TCP_SETUP_ACK(tp, tiflags, m); 1022 /* 1023 * Drop TCP, IP headers and TCP options then add data 1024 * to socket buffer. 1025 */ 1026 if (so->so_state & SS_CANTRCVMORE) 1027 m_freem(m); 1028 else { 1029 if (opti.ts_present && opti.ts_ecr) { 1030 if (tp->rfbuf_ts < opti.ts_ecr && 1031 opti.ts_ecr - tp->rfbuf_ts < hz) { 1032 tcp_update_rcvspace(tp); 1033 /* Start over with next RTT. */ 1034 tp->rfbuf_cnt = 0; 1035 tp->rfbuf_ts = 0; 1036 } else 1037 tp->rfbuf_cnt += tlen; 1038 } 1039 m_adj(m, iphlen + off); 1040 sbappendstream(so, &so->so_rcv, m); 1041 } 1042 tp->t_flags |= TF_BLOCKOUTPUT; 1043 sorwakeup(so); 1044 tp->t_flags &= ~TF_BLOCKOUTPUT; 1045 if (tp->t_flags & (TF_ACKNOW|TF_NEEDOUTPUT)) 1046 (void) tcp_output(tp); 1047 return IPPROTO_DONE; 1048 } 1049 } 1050 1051 /* 1052 * Compute mbuf offset to TCP data segment. 1053 */ 1054 hdroptlen = iphlen + off; 1055 1056 /* 1057 * Calculate amount of space in receive window, 1058 * and then do TCP input processing. 1059 * Receive window is amount of space in rcv queue, 1060 * but not less than advertised window. 1061 */ 1062 { int win; 1063 1064 win = sbspace(so, &so->so_rcv); 1065 if (win < 0) 1066 win = 0; 1067 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1068 } 1069 1070 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1071 tp->rfbuf_cnt = 0; 1072 tp->rfbuf_ts = 0; 1073 1074 switch (tp->t_state) { 1075 1076 /* 1077 * If the state is SYN_RECEIVED: 1078 * if seg contains SYN/ACK, send an RST. 1079 * if seg contains an ACK, but not for our SYN/ACK, send an RST 1080 */ 1081 1082 case TCPS_SYN_RECEIVED: 1083 if (tiflags & TH_ACK) { 1084 if (tiflags & TH_SYN) { 1085 tcpstat_inc(tcps_badsyn); 1086 goto dropwithreset; 1087 } 1088 if (SEQ_LEQ(th->th_ack, tp->snd_una) || 1089 SEQ_GT(th->th_ack, tp->snd_max)) 1090 goto dropwithreset; 1091 } 1092 break; 1093 1094 /* 1095 * If the state is SYN_SENT: 1096 * if seg contains an ACK, but not for our SYN, drop the input. 1097 * if seg contains a RST, then drop the connection. 1098 * if seg does not contain SYN, then drop it. 1099 * Otherwise this is an acceptable SYN segment 1100 * initialize tp->rcv_nxt and tp->irs 1101 * if seg contains ack then advance tp->snd_una 1102 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1103 * arrange for segment to be acked (eventually) 1104 * continue processing rest of data/controls, beginning with URG 1105 */ 1106 case TCPS_SYN_SENT: 1107 if ((tiflags & TH_ACK) && 1108 (SEQ_LEQ(th->th_ack, tp->iss) || 1109 SEQ_GT(th->th_ack, tp->snd_max))) 1110 goto dropwithreset; 1111 if (tiflags & TH_RST) { 1112 #ifdef TCP_ECN 1113 /* if ECN is enabled, fall back to non-ecn at rexmit */ 1114 if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN)) 1115 goto drop; 1116 #endif 1117 if (tiflags & TH_ACK) 1118 tp = tcp_drop(tp, ECONNREFUSED); 1119 goto drop; 1120 } 1121 if ((tiflags & TH_SYN) == 0) 1122 goto drop; 1123 if (tiflags & TH_ACK) { 1124 tp->snd_una = th->th_ack; 1125 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1126 tp->snd_nxt = tp->snd_una; 1127 } 1128 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1129 tp->irs = th->th_seq; 1130 tcp_mss(tp, opti.maxseg); 1131 /* Reset initial window to 1 segment for retransmit */ 1132 if (tp->t_rxtshift > 0) 1133 tp->snd_cwnd = tp->t_maxseg; 1134 tcp_rcvseqinit(tp); 1135 tp->t_flags |= TF_ACKNOW; 1136 /* 1137 * If we've sent a SACK_PERMITTED option, and the peer 1138 * also replied with one, then TF_SACK_PERMIT should have 1139 * been set in tcp_dooptions(). If it was not, disable SACKs. 1140 */ 1141 if (tp->sack_enable) 1142 tp->sack_enable = tp->t_flags & TF_SACK_PERMIT; 1143 #ifdef TCP_ECN 1144 /* 1145 * if ECE is set but CWR is not set for SYN-ACK, or 1146 * both ECE and CWR are set for simultaneous open, 1147 * peer is ECN capable. 1148 */ 1149 if (tcp_do_ecn) { 1150 switch (tiflags & (TH_ACK|TH_ECE|TH_CWR)) { 1151 case TH_ACK|TH_ECE: 1152 case TH_ECE|TH_CWR: 1153 tp->t_flags |= TF_ECN_PERMIT; 1154 tiflags &= ~(TH_ECE|TH_CWR); 1155 tcpstat_inc(tcps_ecn_accepts); 1156 } 1157 } 1158 #endif 1159 1160 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 1161 tcpstat_inc(tcps_connects); 1162 tp->t_flags |= TF_BLOCKOUTPUT; 1163 soisconnected(so); 1164 tp->t_flags &= ~TF_BLOCKOUTPUT; 1165 tp->t_state = TCPS_ESTABLISHED; 1166 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1167 /* Do window scaling on this connection? */ 1168 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1169 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1170 tp->snd_scale = tp->requested_s_scale; 1171 tp->rcv_scale = tp->request_r_scale; 1172 } 1173 tcp_flush_queue(tp); 1174 1175 /* 1176 * if we didn't have to retransmit the SYN, 1177 * use its rtt as our initial srtt & rtt var. 1178 */ 1179 if (tp->t_rtttime) 1180 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1181 /* 1182 * Since new data was acked (the SYN), open the 1183 * congestion window by one MSS. We do this 1184 * here, because we won't go through the normal 1185 * ACK processing below. And since this is the 1186 * start of the connection, we know we are in 1187 * the exponential phase of slow-start. 1188 */ 1189 tp->snd_cwnd += tp->t_maxseg; 1190 } else 1191 tp->t_state = TCPS_SYN_RECEIVED; 1192 1193 #if 0 1194 trimthenstep6: 1195 #endif 1196 /* 1197 * Advance th->th_seq to correspond to first data byte. 1198 * If data, trim to stay within window, 1199 * dropping FIN if necessary. 1200 */ 1201 th->th_seq++; 1202 if (tlen > tp->rcv_wnd) { 1203 todrop = tlen - tp->rcv_wnd; 1204 m_adj(m, -todrop); 1205 tlen = tp->rcv_wnd; 1206 tiflags &= ~TH_FIN; 1207 tcpstat_pkt(tcps_rcvpackafterwin, tcps_rcvbyteafterwin, 1208 todrop); 1209 } 1210 tp->snd_wl1 = th->th_seq - 1; 1211 tp->rcv_up = th->th_seq; 1212 goto step6; 1213 /* 1214 * If a new connection request is received while in TIME_WAIT, 1215 * drop the old connection and start over if the if the 1216 * timestamp or the sequence numbers are above the previous 1217 * ones. 1218 */ 1219 case TCPS_TIME_WAIT: 1220 if (((tiflags & (TH_SYN|TH_ACK)) == TH_SYN) && 1221 ((opti.ts_present && 1222 TSTMP_LT(tp->ts_recent, opti.ts_val)) || 1223 SEQ_GT(th->th_seq, tp->rcv_nxt))) { 1224 #if NPF > 0 1225 /* 1226 * The socket will be recreated but the new state 1227 * has already been linked to the socket. Remove the 1228 * link between old socket and new state. 1229 */ 1230 pf_inp_unlink(inp); 1231 #endif 1232 /* 1233 * Advance the iss by at least 32768, but 1234 * clear the msb in order to make sure 1235 * that SEG_LT(snd_nxt, iss). 1236 */ 1237 iss = tp->snd_nxt + 1238 ((arc4random() & 0x7fffffff) | 0x8000); 1239 reuse = &iss; 1240 tp = tcp_close(tp); 1241 inp = NULL; 1242 goto findpcb; 1243 } 1244 } 1245 1246 /* 1247 * States other than LISTEN or SYN_SENT. 1248 * First check timestamp, if present. 1249 * Then check that at least some bytes of segment are within 1250 * receive window. If segment begins before rcv_nxt, 1251 * drop leading data (and SYN); if nothing left, just ack. 1252 * 1253 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1254 * and it's less than opti.ts_recent, drop it. 1255 */ 1256 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1257 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1258 1259 /* Check to see if ts_recent is over 24 days old. */ 1260 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1261 /* 1262 * Invalidate ts_recent. If this segment updates 1263 * ts_recent, the age will be reset later and ts_recent 1264 * will get a valid value. If it does not, setting 1265 * ts_recent to zero will at least satisfy the 1266 * requirement that zero be placed in the timestamp 1267 * echo reply when ts_recent isn't valid. The 1268 * age isn't reset until we get a valid ts_recent 1269 * because we don't want out-of-order segments to be 1270 * dropped when ts_recent is old. 1271 */ 1272 tp->ts_recent = 0; 1273 } else { 1274 tcpstat_pkt(tcps_rcvduppack, tcps_rcvdupbyte, tlen); 1275 tcpstat_inc(tcps_pawsdrop); 1276 if (tlen) 1277 goto dropafterack; 1278 goto drop; 1279 } 1280 } 1281 1282 todrop = tp->rcv_nxt - th->th_seq; 1283 if (todrop > 0) { 1284 if (tiflags & TH_SYN) { 1285 tiflags &= ~TH_SYN; 1286 th->th_seq++; 1287 if (th->th_urp > 1) 1288 th->th_urp--; 1289 else 1290 tiflags &= ~TH_URG; 1291 todrop--; 1292 } 1293 if (todrop > tlen || 1294 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1295 /* 1296 * Any valid FIN must be to the left of the 1297 * window. At this point, FIN must be a 1298 * duplicate or out-of-sequence, so drop it. 1299 */ 1300 tiflags &= ~TH_FIN; 1301 /* 1302 * Send ACK to resynchronize, and drop any data, 1303 * but keep on processing for RST or ACK. 1304 */ 1305 tp->t_flags |= TF_ACKNOW; 1306 todrop = tlen; 1307 tcpstat_pkt(tcps_rcvduppack, tcps_rcvdupbyte, todrop); 1308 } else { 1309 tcpstat_pkt(tcps_rcvpartduppack, tcps_rcvpartdupbyte, 1310 todrop); 1311 } 1312 hdroptlen += todrop; /* drop from head afterwards */ 1313 th->th_seq += todrop; 1314 tlen -= todrop; 1315 if (th->th_urp > todrop) 1316 th->th_urp -= todrop; 1317 else { 1318 tiflags &= ~TH_URG; 1319 th->th_urp = 0; 1320 } 1321 } 1322 1323 /* 1324 * If new data are received on a connection after the 1325 * user processes are gone, then RST the other end. 1326 */ 1327 if ((so->so_state & SS_NOFDREF) && 1328 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1329 tp = tcp_close(tp); 1330 tcpstat_inc(tcps_rcvafterclose); 1331 goto dropwithreset; 1332 } 1333 1334 /* 1335 * If segment ends after window, drop trailing data 1336 * (and PUSH and FIN); if nothing left, just ACK. 1337 */ 1338 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1339 if (todrop > 0) { 1340 tcpstat_inc(tcps_rcvpackafterwin); 1341 if (todrop >= tlen) { 1342 tcpstat_add(tcps_rcvbyteafterwin, tlen); 1343 /* 1344 * If window is closed can only take segments at 1345 * window edge, and have to drop data and PUSH from 1346 * incoming segments. Continue processing, but 1347 * remember to ack. Otherwise, drop segment 1348 * and ack. 1349 */ 1350 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1351 tp->t_flags |= TF_ACKNOW; 1352 tcpstat_inc(tcps_rcvwinprobe); 1353 } else 1354 goto dropafterack; 1355 } else 1356 tcpstat_add(tcps_rcvbyteafterwin, todrop); 1357 m_adj(m, -todrop); 1358 tlen -= todrop; 1359 tiflags &= ~(TH_PUSH|TH_FIN); 1360 } 1361 1362 /* 1363 * If last ACK falls within this segment's sequence numbers, 1364 * record its timestamp if it's more recent. 1365 * NOTE that the test is modified according to the latest 1366 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1367 */ 1368 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1369 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1370 tp->ts_recent_age = tcp_now; 1371 tp->ts_recent = opti.ts_val; 1372 } 1373 1374 /* 1375 * If the RST bit is set examine the state: 1376 * SYN_RECEIVED STATE: 1377 * If passive open, return to LISTEN state. 1378 * If active open, inform user that connection was refused. 1379 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1380 * Inform user that connection was reset, and close tcb. 1381 * CLOSING, LAST_ACK, TIME_WAIT STATES 1382 * Close the tcb. 1383 */ 1384 if (tiflags & TH_RST) { 1385 if (th->th_seq != tp->last_ack_sent && 1386 th->th_seq != tp->rcv_nxt && 1387 th->th_seq != (tp->rcv_nxt + 1)) 1388 goto drop; 1389 1390 switch (tp->t_state) { 1391 case TCPS_SYN_RECEIVED: 1392 #ifdef TCP_ECN 1393 /* if ECN is enabled, fall back to non-ecn at rexmit */ 1394 if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN)) 1395 goto drop; 1396 #endif 1397 so->so_error = ECONNREFUSED; 1398 goto close; 1399 1400 case TCPS_ESTABLISHED: 1401 case TCPS_FIN_WAIT_1: 1402 case TCPS_FIN_WAIT_2: 1403 case TCPS_CLOSE_WAIT: 1404 so->so_error = ECONNRESET; 1405 close: 1406 tp->t_state = TCPS_CLOSED; 1407 tcpstat_inc(tcps_drops); 1408 tp = tcp_close(tp); 1409 goto drop; 1410 case TCPS_CLOSING: 1411 case TCPS_LAST_ACK: 1412 case TCPS_TIME_WAIT: 1413 tp = tcp_close(tp); 1414 goto drop; 1415 } 1416 } 1417 1418 /* 1419 * If a SYN is in the window, then this is an 1420 * error and we ACK and drop the packet. 1421 */ 1422 if (tiflags & TH_SYN) 1423 goto dropafterack_ratelim; 1424 1425 /* 1426 * If the ACK bit is off we drop the segment and return. 1427 */ 1428 if ((tiflags & TH_ACK) == 0) { 1429 if (tp->t_flags & TF_ACKNOW) 1430 goto dropafterack; 1431 else 1432 goto drop; 1433 } 1434 1435 /* 1436 * Ack processing. 1437 */ 1438 switch (tp->t_state) { 1439 1440 /* 1441 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1442 * ESTABLISHED state and continue processing. 1443 * The ACK was checked above. 1444 */ 1445 case TCPS_SYN_RECEIVED: 1446 tcpstat_inc(tcps_connects); 1447 tp->t_flags |= TF_BLOCKOUTPUT; 1448 soisconnected(so); 1449 tp->t_flags &= ~TF_BLOCKOUTPUT; 1450 tp->t_state = TCPS_ESTABLISHED; 1451 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1452 /* Do window scaling? */ 1453 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1454 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1455 tp->snd_scale = tp->requested_s_scale; 1456 tp->rcv_scale = tp->request_r_scale; 1457 tiwin = th->th_win << tp->snd_scale; 1458 } 1459 tcp_flush_queue(tp); 1460 tp->snd_wl1 = th->th_seq - 1; 1461 /* fall into ... */ 1462 1463 /* 1464 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1465 * ACKs. If the ack is in the range 1466 * tp->snd_una < th->th_ack <= tp->snd_max 1467 * then advance tp->snd_una to th->th_ack and drop 1468 * data from the retransmission queue. If this ACK reflects 1469 * more up to date window information we update our window information. 1470 */ 1471 case TCPS_ESTABLISHED: 1472 case TCPS_FIN_WAIT_1: 1473 case TCPS_FIN_WAIT_2: 1474 case TCPS_CLOSE_WAIT: 1475 case TCPS_CLOSING: 1476 case TCPS_LAST_ACK: 1477 case TCPS_TIME_WAIT: 1478 #ifdef TCP_ECN 1479 /* 1480 * if we receive ECE and are not already in recovery phase, 1481 * reduce cwnd by half but don't slow-start. 1482 * advance snd_last to snd_max not to reduce cwnd again 1483 * until all outstanding packets are acked. 1484 */ 1485 if (tcp_do_ecn && (tiflags & TH_ECE)) { 1486 if ((tp->t_flags & TF_ECN_PERMIT) && 1487 SEQ_GEQ(tp->snd_una, tp->snd_last)) { 1488 u_int win; 1489 1490 win = min(tp->snd_wnd, tp->snd_cwnd) / tp->t_maxseg; 1491 if (win > 1) { 1492 tp->snd_ssthresh = win / 2 * tp->t_maxseg; 1493 tp->snd_cwnd = tp->snd_ssthresh; 1494 tp->snd_last = tp->snd_max; 1495 tp->t_flags |= TF_SEND_CWR; 1496 tcpstat_inc(tcps_cwr_ecn); 1497 } 1498 } 1499 tcpstat_inc(tcps_ecn_rcvece); 1500 } 1501 /* 1502 * if we receive CWR, we know that the peer has reduced 1503 * its congestion window. stop sending ecn-echo. 1504 */ 1505 if ((tiflags & TH_CWR)) { 1506 tp->t_flags &= ~TF_RCVD_CE; 1507 tcpstat_inc(tcps_ecn_rcvcwr); 1508 } 1509 #endif /* TCP_ECN */ 1510 1511 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1512 /* 1513 * Duplicate/old ACK processing. 1514 * Increments t_dupacks: 1515 * Pure duplicate (same seq/ack/window, no data) 1516 * Doesn't affect t_dupacks: 1517 * Data packets. 1518 * Normal window updates (window opens) 1519 * Resets t_dupacks: 1520 * New data ACKed. 1521 * Window shrinks 1522 * Old ACK 1523 */ 1524 if (tlen) { 1525 /* Drop very old ACKs unless th_seq matches */ 1526 if (th->th_seq != tp->rcv_nxt && 1527 SEQ_LT(th->th_ack, 1528 tp->snd_una - tp->max_sndwnd)) { 1529 tcpstat_inc(tcps_rcvacktooold); 1530 goto drop; 1531 } 1532 break; 1533 } 1534 /* 1535 * If we get an old ACK, there is probably packet 1536 * reordering going on. Be conservative and reset 1537 * t_dupacks so that we are less aggressive in 1538 * doing a fast retransmit. 1539 */ 1540 if (th->th_ack != tp->snd_una) { 1541 tp->t_dupacks = 0; 1542 break; 1543 } 1544 if (tiwin == tp->snd_wnd) { 1545 tcpstat_inc(tcps_rcvdupack); 1546 /* 1547 * If we have outstanding data (other than 1548 * a window probe), this is a completely 1549 * duplicate ack (ie, window info didn't 1550 * change), the ack is the biggest we've 1551 * seen and we've seen exactly our rexmt 1552 * threshold of them, assume a packet 1553 * has been dropped and retransmit it. 1554 * Kludge snd_nxt & the congestion 1555 * window so we send only this one 1556 * packet. 1557 * 1558 * We know we're losing at the current 1559 * window size so do congestion avoidance 1560 * (set ssthresh to half the current window 1561 * and pull our congestion window back to 1562 * the new ssthresh). 1563 * 1564 * Dup acks mean that packets have left the 1565 * network (they're now cached at the receiver) 1566 * so bump cwnd by the amount in the receiver 1567 * to keep a constant cwnd packets in the 1568 * network. 1569 */ 1570 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0) 1571 tp->t_dupacks = 0; 1572 else if (++tp->t_dupacks == tcprexmtthresh) { 1573 tcp_seq onxt = tp->snd_nxt; 1574 u_long win = 1575 ulmin(tp->snd_wnd, tp->snd_cwnd) / 1576 2 / tp->t_maxseg; 1577 1578 if (SEQ_LT(th->th_ack, tp->snd_last)){ 1579 /* 1580 * False fast retx after 1581 * timeout. Do not cut window. 1582 */ 1583 tp->t_dupacks = 0; 1584 goto drop; 1585 } 1586 if (win < 2) 1587 win = 2; 1588 tp->snd_ssthresh = win * tp->t_maxseg; 1589 tp->snd_last = tp->snd_max; 1590 if (tp->sack_enable) { 1591 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1592 tp->t_rtttime = 0; 1593 #ifdef TCP_ECN 1594 tp->t_flags |= TF_SEND_CWR; 1595 #endif 1596 tcpstat_inc(tcps_cwr_frecovery); 1597 tcpstat_inc(tcps_sack_recovery_episode); 1598 /* 1599 * tcp_output() will send 1600 * oldest SACK-eligible rtx. 1601 */ 1602 (void) tcp_output(tp); 1603 tp->snd_cwnd = tp->snd_ssthresh+ 1604 tp->t_maxseg * tp->t_dupacks; 1605 goto drop; 1606 } 1607 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1608 tp->t_rtttime = 0; 1609 tp->snd_nxt = th->th_ack; 1610 tp->snd_cwnd = tp->t_maxseg; 1611 #ifdef TCP_ECN 1612 tp->t_flags |= TF_SEND_CWR; 1613 #endif 1614 tcpstat_inc(tcps_cwr_frecovery); 1615 tcpstat_inc(tcps_sndrexmitfast); 1616 (void) tcp_output(tp); 1617 1618 tp->snd_cwnd = tp->snd_ssthresh + 1619 tp->t_maxseg * tp->t_dupacks; 1620 if (SEQ_GT(onxt, tp->snd_nxt)) 1621 tp->snd_nxt = onxt; 1622 goto drop; 1623 } else if (tp->t_dupacks > tcprexmtthresh) { 1624 tp->snd_cwnd += tp->t_maxseg; 1625 (void) tcp_output(tp); 1626 goto drop; 1627 } 1628 } else if (tiwin < tp->snd_wnd) { 1629 /* 1630 * The window was retracted! Previous dup 1631 * ACKs may have been due to packets arriving 1632 * after the shrunken window, not a missing 1633 * packet, so play it safe and reset t_dupacks 1634 */ 1635 tp->t_dupacks = 0; 1636 } 1637 break; 1638 } 1639 /* 1640 * If the congestion window was inflated to account 1641 * for the other side's cached packets, retract it. 1642 */ 1643 if (tp->t_dupacks >= tcprexmtthresh) { 1644 /* Check for a partial ACK */ 1645 if (SEQ_LT(th->th_ack, tp->snd_last)) { 1646 if (tp->sack_enable) 1647 tcp_sack_partialack(tp, th); 1648 else 1649 tcp_newreno_partialack(tp, th); 1650 } else { 1651 /* Out of fast recovery */ 1652 tp->snd_cwnd = tp->snd_ssthresh; 1653 if (tcp_seq_subtract(tp->snd_max, th->th_ack) < 1654 tp->snd_ssthresh) 1655 tp->snd_cwnd = 1656 tcp_seq_subtract(tp->snd_max, 1657 th->th_ack); 1658 tp->t_dupacks = 0; 1659 } 1660 } else { 1661 /* 1662 * Reset the duplicate ACK counter if we 1663 * were not in fast recovery. 1664 */ 1665 tp->t_dupacks = 0; 1666 } 1667 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1668 tcpstat_inc(tcps_rcvacktoomuch); 1669 goto dropafterack_ratelim; 1670 } 1671 acked = th->th_ack - tp->snd_una; 1672 tcpstat_pkt(tcps_rcvackpack, tcps_rcvackbyte, acked); 1673 1674 /* 1675 * If we have a timestamp reply, update smoothed 1676 * round trip time. If no timestamp is present but 1677 * transmit timer is running and timed sequence 1678 * number was acked, update smoothed round trip time. 1679 * Since we now have an rtt measurement, cancel the 1680 * timer backoff (cf., Phil Karn's retransmit alg.). 1681 * Recompute the initial retransmit timer. 1682 */ 1683 if (opti.ts_present && opti.ts_ecr) 1684 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr); 1685 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1686 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1687 1688 /* 1689 * If all outstanding data is acked, stop retransmit 1690 * timer and remember to restart (more output or persist). 1691 * If there is more data to be acked, restart retransmit 1692 * timer, using current (possibly backed-off) value. 1693 */ 1694 if (th->th_ack == tp->snd_max) { 1695 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1696 tp->t_flags |= TF_NEEDOUTPUT; 1697 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1698 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1699 /* 1700 * When new data is acked, open the congestion window. 1701 * If the window gives us less than ssthresh packets 1702 * in flight, open exponentially (maxseg per packet). 1703 * Otherwise open linearly: maxseg per window 1704 * (maxseg^2 / cwnd per packet). 1705 */ 1706 { 1707 u_int cw = tp->snd_cwnd; 1708 u_int incr = tp->t_maxseg; 1709 1710 if (cw > tp->snd_ssthresh) 1711 incr = max(incr * incr / cw, 1); 1712 if (tp->t_dupacks < tcprexmtthresh) 1713 tp->snd_cwnd = ulmin(cw + incr, 1714 TCP_MAXWIN << tp->snd_scale); 1715 } 1716 ND6_HINT(tp); 1717 if (acked > so->so_snd.sb_cc) { 1718 if (tp->snd_wnd > so->so_snd.sb_cc) 1719 tp->snd_wnd -= so->so_snd.sb_cc; 1720 else 1721 tp->snd_wnd = 0; 1722 sbdrop(so, &so->so_snd, (int)so->so_snd.sb_cc); 1723 ourfinisacked = 1; 1724 } else { 1725 sbdrop(so, &so->so_snd, acked); 1726 if (tp->snd_wnd > acked) 1727 tp->snd_wnd -= acked; 1728 else 1729 tp->snd_wnd = 0; 1730 ourfinisacked = 0; 1731 } 1732 1733 tcp_update_sndspace(tp); 1734 if (sb_notify(so, &so->so_snd)) { 1735 tp->t_flags |= TF_BLOCKOUTPUT; 1736 sowwakeup(so); 1737 tp->t_flags &= ~TF_BLOCKOUTPUT; 1738 } 1739 1740 /* 1741 * If we had a pending ICMP message that referred to data 1742 * that have just been acknowledged, disregard the recorded 1743 * ICMP message. 1744 */ 1745 if ((tp->t_flags & TF_PMTUD_PEND) && 1746 SEQ_GT(th->th_ack, tp->t_pmtud_th_seq)) 1747 tp->t_flags &= ~TF_PMTUD_PEND; 1748 1749 /* 1750 * Keep track of the largest chunk of data acknowledged 1751 * since last PMTU update 1752 */ 1753 if (tp->t_pmtud_mss_acked < acked) 1754 tp->t_pmtud_mss_acked = acked; 1755 1756 tp->snd_una = th->th_ack; 1757 #ifdef TCP_ECN 1758 /* sync snd_last with snd_una */ 1759 if (SEQ_GT(tp->snd_una, tp->snd_last)) 1760 tp->snd_last = tp->snd_una; 1761 #endif 1762 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1763 tp->snd_nxt = tp->snd_una; 1764 1765 switch (tp->t_state) { 1766 1767 /* 1768 * In FIN_WAIT_1 STATE in addition to the processing 1769 * for the ESTABLISHED state if our FIN is now acknowledged 1770 * then enter FIN_WAIT_2. 1771 */ 1772 case TCPS_FIN_WAIT_1: 1773 if (ourfinisacked) { 1774 /* 1775 * If we can't receive any more 1776 * data, then closing user can proceed. 1777 * Starting the timer is contrary to the 1778 * specification, but if we don't get a FIN 1779 * we'll hang forever. 1780 */ 1781 if (so->so_state & SS_CANTRCVMORE) { 1782 tp->t_flags |= TF_BLOCKOUTPUT; 1783 soisdisconnected(so); 1784 tp->t_flags &= ~TF_BLOCKOUTPUT; 1785 TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle); 1786 } 1787 tp->t_state = TCPS_FIN_WAIT_2; 1788 } 1789 break; 1790 1791 /* 1792 * In CLOSING STATE in addition to the processing for 1793 * the ESTABLISHED state if the ACK acknowledges our FIN 1794 * then enter the TIME-WAIT state, otherwise ignore 1795 * the segment. 1796 */ 1797 case TCPS_CLOSING: 1798 if (ourfinisacked) { 1799 tp->t_state = TCPS_TIME_WAIT; 1800 tcp_canceltimers(tp); 1801 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1802 tp->t_flags |= TF_BLOCKOUTPUT; 1803 soisdisconnected(so); 1804 tp->t_flags &= ~TF_BLOCKOUTPUT; 1805 } 1806 break; 1807 1808 /* 1809 * In LAST_ACK, we may still be waiting for data to drain 1810 * and/or to be acked, as well as for the ack of our FIN. 1811 * If our FIN is now acknowledged, delete the TCB, 1812 * enter the closed state and return. 1813 */ 1814 case TCPS_LAST_ACK: 1815 if (ourfinisacked) { 1816 tp = tcp_close(tp); 1817 goto drop; 1818 } 1819 break; 1820 1821 /* 1822 * In TIME_WAIT state the only thing that should arrive 1823 * is a retransmission of the remote FIN. Acknowledge 1824 * it and restart the finack timer. 1825 */ 1826 case TCPS_TIME_WAIT: 1827 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1828 goto dropafterack; 1829 } 1830 } 1831 1832 step6: 1833 /* 1834 * Update window information. 1835 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1836 */ 1837 if ((tiflags & TH_ACK) && 1838 (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq && 1839 (SEQ_LT(tp->snd_wl2, th->th_ack) || 1840 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 1841 /* keep track of pure window updates */ 1842 if (tlen == 0 && 1843 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1844 tcpstat_inc(tcps_rcvwinupd); 1845 tp->snd_wnd = tiwin; 1846 tp->snd_wl1 = th->th_seq; 1847 tp->snd_wl2 = th->th_ack; 1848 if (tp->snd_wnd > tp->max_sndwnd) 1849 tp->max_sndwnd = tp->snd_wnd; 1850 tp->t_flags |= TF_NEEDOUTPUT; 1851 } 1852 1853 /* 1854 * Process segments with URG. 1855 */ 1856 if ((tiflags & TH_URG) && th->th_urp && 1857 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1858 /* 1859 * This is a kludge, but if we receive and accept 1860 * random urgent pointers, we'll crash in 1861 * soreceive. It's hard to imagine someone 1862 * actually wanting to send this much urgent data. 1863 */ 1864 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1865 th->th_urp = 0; /* XXX */ 1866 tiflags &= ~TH_URG; /* XXX */ 1867 goto dodata; /* XXX */ 1868 } 1869 /* 1870 * If this segment advances the known urgent pointer, 1871 * then mark the data stream. This should not happen 1872 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1873 * a FIN has been received from the remote side. 1874 * In these states we ignore the URG. 1875 * 1876 * According to RFC961 (Assigned Protocols), 1877 * the urgent pointer points to the last octet 1878 * of urgent data. We continue, however, 1879 * to consider it to indicate the first octet 1880 * of data past the urgent section as the original 1881 * spec states (in one of two places). 1882 */ 1883 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1884 tp->rcv_up = th->th_seq + th->th_urp; 1885 so->so_oobmark = so->so_rcv.sb_cc + 1886 (tp->rcv_up - tp->rcv_nxt) - 1; 1887 if (so->so_oobmark == 0) 1888 so->so_state |= SS_RCVATMARK; 1889 sohasoutofband(so); 1890 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1891 } 1892 /* 1893 * Remove out of band data so doesn't get presented to user. 1894 * This can happen independent of advancing the URG pointer, 1895 * but if two URG's are pending at once, some out-of-band 1896 * data may creep in... ick. 1897 */ 1898 if (th->th_urp <= (u_int16_t) tlen && 1899 (so->so_options & SO_OOBINLINE) == 0) 1900 tcp_pulloutofband(so, th->th_urp, m, hdroptlen); 1901 } else 1902 /* 1903 * If no out of band data is expected, 1904 * pull receive urgent pointer along 1905 * with the receive window. 1906 */ 1907 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1908 tp->rcv_up = tp->rcv_nxt; 1909 dodata: /* XXX */ 1910 1911 /* 1912 * Process the segment text, merging it into the TCP sequencing queue, 1913 * and arranging for acknowledgment of receipt if necessary. 1914 * This process logically involves adjusting tp->rcv_wnd as data 1915 * is presented to the user (this happens in tcp_usrreq.c, 1916 * case PRU_RCVD). If a FIN has already been received on this 1917 * connection then we just ignore the text. 1918 */ 1919 if ((tlen || (tiflags & TH_FIN)) && 1920 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1921 tcp_seq laststart = th->th_seq; 1922 tcp_seq lastend = th->th_seq + tlen; 1923 1924 if (th->th_seq == tp->rcv_nxt && TAILQ_EMPTY(&tp->t_segq) && 1925 tp->t_state == TCPS_ESTABLISHED) { 1926 TCP_SETUP_ACK(tp, tiflags, m); 1927 tp->rcv_nxt += tlen; 1928 tiflags = th->th_flags & TH_FIN; 1929 tcpstat_pkt(tcps_rcvpack, tcps_rcvbyte, tlen); 1930 ND6_HINT(tp); 1931 if (so->so_state & SS_CANTRCVMORE) 1932 m_freem(m); 1933 else { 1934 m_adj(m, hdroptlen); 1935 sbappendstream(so, &so->so_rcv, m); 1936 } 1937 tp->t_flags |= TF_BLOCKOUTPUT; 1938 sorwakeup(so); 1939 tp->t_flags &= ~TF_BLOCKOUTPUT; 1940 } else { 1941 m_adj(m, hdroptlen); 1942 tiflags = tcp_reass(tp, th, m, &tlen); 1943 tp->t_flags |= TF_ACKNOW; 1944 } 1945 if (tp->sack_enable) 1946 tcp_update_sack_list(tp, laststart, lastend); 1947 1948 /* 1949 * variable len never referenced again in modern BSD, 1950 * so why bother computing it ?? 1951 */ 1952 #if 0 1953 /* 1954 * Note the amount of data that peer has sent into 1955 * our window, in order to estimate the sender's 1956 * buffer size. 1957 */ 1958 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1959 #endif /* 0 */ 1960 } else { 1961 m_freem(m); 1962 tiflags &= ~TH_FIN; 1963 } 1964 1965 /* 1966 * If FIN is received ACK the FIN and let the user know 1967 * that the connection is closing. Ignore a FIN received before 1968 * the connection is fully established. 1969 */ 1970 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 1971 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1972 tp->t_flags |= TF_BLOCKOUTPUT; 1973 socantrcvmore(so); 1974 tp->t_flags &= ~TF_BLOCKOUTPUT; 1975 tp->t_flags |= TF_ACKNOW; 1976 tp->rcv_nxt++; 1977 } 1978 switch (tp->t_state) { 1979 1980 /* 1981 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 1982 */ 1983 case TCPS_ESTABLISHED: 1984 tp->t_state = TCPS_CLOSE_WAIT; 1985 break; 1986 1987 /* 1988 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1989 * enter the CLOSING state. 1990 */ 1991 case TCPS_FIN_WAIT_1: 1992 tp->t_state = TCPS_CLOSING; 1993 break; 1994 1995 /* 1996 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1997 * starting the time-wait timer, turning off the other 1998 * standard timers. 1999 */ 2000 case TCPS_FIN_WAIT_2: 2001 tp->t_state = TCPS_TIME_WAIT; 2002 tcp_canceltimers(tp); 2003 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2004 tp->t_flags |= TF_BLOCKOUTPUT; 2005 soisdisconnected(so); 2006 tp->t_flags &= ~TF_BLOCKOUTPUT; 2007 break; 2008 2009 /* 2010 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2011 */ 2012 case TCPS_TIME_WAIT: 2013 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2014 break; 2015 } 2016 } 2017 if (otp) 2018 tcp_trace(TA_INPUT, ostate, tp, otp, saveti, 0, tlen); 2019 2020 /* 2021 * Return any desired output. 2022 */ 2023 if (tp->t_flags & (TF_ACKNOW|TF_NEEDOUTPUT)) 2024 (void) tcp_output(tp); 2025 return IPPROTO_DONE; 2026 2027 badsyn: 2028 /* 2029 * Received a bad SYN. Increment counters and dropwithreset. 2030 */ 2031 tcpstat_inc(tcps_badsyn); 2032 tp = NULL; 2033 goto dropwithreset; 2034 2035 dropafterack_ratelim: 2036 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2037 tcp_ackdrop_ppslim) == 0) { 2038 /* XXX stat */ 2039 goto drop; 2040 } 2041 /* ...fall into dropafterack... */ 2042 2043 dropafterack: 2044 /* 2045 * Generate an ACK dropping incoming segment if it occupies 2046 * sequence space, where the ACK reflects our state. 2047 */ 2048 if (tiflags & TH_RST) 2049 goto drop; 2050 m_freem(m); 2051 tp->t_flags |= TF_ACKNOW; 2052 (void) tcp_output(tp); 2053 return IPPROTO_DONE; 2054 2055 dropwithreset_ratelim: 2056 /* 2057 * We may want to rate-limit RSTs in certain situations, 2058 * particularly if we are sending an RST in response to 2059 * an attempt to connect to or otherwise communicate with 2060 * a port for which we have no socket. 2061 */ 2062 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2063 tcp_rst_ppslim) == 0) { 2064 /* XXX stat */ 2065 goto drop; 2066 } 2067 /* ...fall into dropwithreset... */ 2068 2069 dropwithreset: 2070 /* 2071 * Generate a RST, dropping incoming segment. 2072 * Make ACK acceptable to originator of segment. 2073 * Don't bother to respond to RST. 2074 */ 2075 if (tiflags & TH_RST) 2076 goto drop; 2077 if (tiflags & TH_ACK) { 2078 tcp_respond(tp, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack, 2079 TH_RST, m->m_pkthdr.ph_rtableid); 2080 } else { 2081 if (tiflags & TH_SYN) 2082 tlen++; 2083 tcp_respond(tp, mtod(m, caddr_t), th, th->th_seq + tlen, 2084 (tcp_seq)0, TH_RST|TH_ACK, m->m_pkthdr.ph_rtableid); 2085 } 2086 m_freem(m); 2087 return IPPROTO_DONE; 2088 2089 drop: 2090 /* 2091 * Drop space held by incoming segment and return. 2092 */ 2093 if (otp) 2094 tcp_trace(TA_DROP, ostate, tp, otp, saveti, 0, tlen); 2095 2096 m_freem(m); 2097 return IPPROTO_DONE; 2098 } 2099 2100 int 2101 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th, 2102 struct mbuf *m, int iphlen, struct tcp_opt_info *oi, 2103 u_int rtableid) 2104 { 2105 u_int16_t mss = 0; 2106 int opt, optlen; 2107 #ifdef TCP_SIGNATURE 2108 caddr_t sigp = NULL; 2109 struct tdb *tdb = NULL; 2110 #endif /* TCP_SIGNATURE */ 2111 2112 for (; cp && cnt > 0; cnt -= optlen, cp += optlen) { 2113 opt = cp[0]; 2114 if (opt == TCPOPT_EOL) 2115 break; 2116 if (opt == TCPOPT_NOP) 2117 optlen = 1; 2118 else { 2119 if (cnt < 2) 2120 break; 2121 optlen = cp[1]; 2122 if (optlen < 2 || optlen > cnt) 2123 break; 2124 } 2125 switch (opt) { 2126 2127 default: 2128 continue; 2129 2130 case TCPOPT_MAXSEG: 2131 if (optlen != TCPOLEN_MAXSEG) 2132 continue; 2133 if (!(th->th_flags & TH_SYN)) 2134 continue; 2135 if (TCPS_HAVERCVDSYN(tp->t_state)) 2136 continue; 2137 memcpy(&mss, cp + 2, sizeof(mss)); 2138 mss = ntohs(mss); 2139 oi->maxseg = mss; 2140 break; 2141 2142 case TCPOPT_WINDOW: 2143 if (optlen != TCPOLEN_WINDOW) 2144 continue; 2145 if (!(th->th_flags & TH_SYN)) 2146 continue; 2147 if (TCPS_HAVERCVDSYN(tp->t_state)) 2148 continue; 2149 tp->t_flags |= TF_RCVD_SCALE; 2150 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2151 break; 2152 2153 case TCPOPT_TIMESTAMP: 2154 if (optlen != TCPOLEN_TIMESTAMP) 2155 continue; 2156 oi->ts_present = 1; 2157 memcpy(&oi->ts_val, cp + 2, sizeof(oi->ts_val)); 2158 oi->ts_val = ntohl(oi->ts_val); 2159 memcpy(&oi->ts_ecr, cp + 6, sizeof(oi->ts_ecr)); 2160 oi->ts_ecr = ntohl(oi->ts_ecr); 2161 2162 if (!(th->th_flags & TH_SYN)) 2163 continue; 2164 if (TCPS_HAVERCVDSYN(tp->t_state)) 2165 continue; 2166 /* 2167 * A timestamp received in a SYN makes 2168 * it ok to send timestamp requests and replies. 2169 */ 2170 tp->t_flags |= TF_RCVD_TSTMP; 2171 tp->ts_recent = oi->ts_val; 2172 tp->ts_recent_age = tcp_now; 2173 break; 2174 2175 case TCPOPT_SACK_PERMITTED: 2176 if (!tp->sack_enable || optlen!=TCPOLEN_SACK_PERMITTED) 2177 continue; 2178 if (!(th->th_flags & TH_SYN)) 2179 continue; 2180 if (TCPS_HAVERCVDSYN(tp->t_state)) 2181 continue; 2182 /* MUST only be set on SYN */ 2183 tp->t_flags |= TF_SACK_PERMIT; 2184 break; 2185 case TCPOPT_SACK: 2186 tcp_sack_option(tp, th, cp, optlen); 2187 break; 2188 #ifdef TCP_SIGNATURE 2189 case TCPOPT_SIGNATURE: 2190 if (optlen != TCPOLEN_SIGNATURE) 2191 continue; 2192 2193 if (sigp && timingsafe_bcmp(sigp, cp + 2, 16)) 2194 return (-1); 2195 2196 sigp = cp + 2; 2197 break; 2198 #endif /* TCP_SIGNATURE */ 2199 } 2200 } 2201 2202 #ifdef TCP_SIGNATURE 2203 if (tp->t_flags & TF_SIGNATURE) { 2204 union sockaddr_union src, dst; 2205 2206 memset(&src, 0, sizeof(union sockaddr_union)); 2207 memset(&dst, 0, sizeof(union sockaddr_union)); 2208 2209 switch (tp->pf) { 2210 case 0: 2211 case AF_INET: 2212 src.sa.sa_len = sizeof(struct sockaddr_in); 2213 src.sa.sa_family = AF_INET; 2214 src.sin.sin_addr = mtod(m, struct ip *)->ip_src; 2215 dst.sa.sa_len = sizeof(struct sockaddr_in); 2216 dst.sa.sa_family = AF_INET; 2217 dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst; 2218 break; 2219 #ifdef INET6 2220 case AF_INET6: 2221 src.sa.sa_len = sizeof(struct sockaddr_in6); 2222 src.sa.sa_family = AF_INET6; 2223 src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src; 2224 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2225 dst.sa.sa_family = AF_INET6; 2226 dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst; 2227 break; 2228 #endif /* INET6 */ 2229 } 2230 2231 tdb = gettdbbysrcdst(rtable_l2(rtableid), 2232 0, &src, &dst, IPPROTO_TCP); 2233 2234 /* 2235 * We don't have an SA for this peer, so we turn off 2236 * TF_SIGNATURE on the listen socket 2237 */ 2238 if (tdb == NULL && tp->t_state == TCPS_LISTEN) 2239 tp->t_flags &= ~TF_SIGNATURE; 2240 2241 } 2242 2243 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2244 tcpstat_inc(tcps_rcvbadsig); 2245 return (-1); 2246 } 2247 2248 if (sigp) { 2249 char sig[16]; 2250 2251 if (tdb == NULL) { 2252 tcpstat_inc(tcps_rcvbadsig); 2253 return (-1); 2254 } 2255 2256 if (tcp_signature(tdb, tp->pf, m, th, iphlen, 1, sig) < 0) 2257 return (-1); 2258 2259 if (timingsafe_bcmp(sig, sigp, 16)) { 2260 tcpstat_inc(tcps_rcvbadsig); 2261 return (-1); 2262 } 2263 2264 tcpstat_inc(tcps_rcvgoodsig); 2265 } 2266 #endif /* TCP_SIGNATURE */ 2267 2268 return (0); 2269 } 2270 2271 u_long 2272 tcp_seq_subtract(u_long a, u_long b) 2273 { 2274 return ((long)(a - b)); 2275 } 2276 2277 /* 2278 * This function is called upon receipt of new valid data (while not in header 2279 * prediction mode), and it updates the ordered list of sacks. 2280 */ 2281 void 2282 tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_laststart, 2283 tcp_seq rcv_lastend) 2284 { 2285 /* 2286 * First reported block MUST be the most recent one. Subsequent 2287 * blocks SHOULD be in the order in which they arrived at the 2288 * receiver. These two conditions make the implementation fully 2289 * compliant with RFC 2018. 2290 */ 2291 int i, j = 0, count = 0, lastpos = -1; 2292 struct sackblk sack, firstsack, temp[MAX_SACK_BLKS]; 2293 2294 /* First clean up current list of sacks */ 2295 for (i = 0; i < tp->rcv_numsacks; i++) { 2296 sack = tp->sackblks[i]; 2297 if (sack.start == 0 && sack.end == 0) { 2298 count++; /* count = number of blocks to be discarded */ 2299 continue; 2300 } 2301 if (SEQ_LEQ(sack.end, tp->rcv_nxt)) { 2302 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2303 count++; 2304 } else { 2305 temp[j].start = tp->sackblks[i].start; 2306 temp[j++].end = tp->sackblks[i].end; 2307 } 2308 } 2309 tp->rcv_numsacks -= count; 2310 if (tp->rcv_numsacks == 0) { /* no sack blocks currently (fast path) */ 2311 tcp_clean_sackreport(tp); 2312 if (SEQ_LT(tp->rcv_nxt, rcv_laststart)) { 2313 /* ==> need first sack block */ 2314 tp->sackblks[0].start = rcv_laststart; 2315 tp->sackblks[0].end = rcv_lastend; 2316 tp->rcv_numsacks = 1; 2317 } 2318 return; 2319 } 2320 /* Otherwise, sack blocks are already present. */ 2321 for (i = 0; i < tp->rcv_numsacks; i++) 2322 tp->sackblks[i] = temp[i]; /* first copy back sack list */ 2323 if (SEQ_GEQ(tp->rcv_nxt, rcv_lastend)) 2324 return; /* sack list remains unchanged */ 2325 /* 2326 * From here, segment just received should be (part of) the 1st sack. 2327 * Go through list, possibly coalescing sack block entries. 2328 */ 2329 firstsack.start = rcv_laststart; 2330 firstsack.end = rcv_lastend; 2331 for (i = 0; i < tp->rcv_numsacks; i++) { 2332 sack = tp->sackblks[i]; 2333 if (SEQ_LT(sack.end, firstsack.start) || 2334 SEQ_GT(sack.start, firstsack.end)) 2335 continue; /* no overlap */ 2336 if (sack.start == firstsack.start && sack.end == firstsack.end){ 2337 /* 2338 * identical block; delete it here since we will 2339 * move it to the front of the list. 2340 */ 2341 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2342 lastpos = i; /* last posn with a zero entry */ 2343 continue; 2344 } 2345 if (SEQ_LEQ(sack.start, firstsack.start)) 2346 firstsack.start = sack.start; /* merge blocks */ 2347 if (SEQ_GEQ(sack.end, firstsack.end)) 2348 firstsack.end = sack.end; /* merge blocks */ 2349 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2350 lastpos = i; /* last posn with a zero entry */ 2351 } 2352 if (lastpos != -1) { /* at least one merge */ 2353 for (i = 0, j = 1; i < tp->rcv_numsacks; i++) { 2354 sack = tp->sackblks[i]; 2355 if (sack.start == 0 && sack.end == 0) 2356 continue; 2357 temp[j++] = sack; 2358 } 2359 tp->rcv_numsacks = j; /* including first blk (added later) */ 2360 for (i = 1; i < tp->rcv_numsacks; i++) /* now copy back */ 2361 tp->sackblks[i] = temp[i]; 2362 } else { /* no merges -- shift sacks by 1 */ 2363 if (tp->rcv_numsacks < MAX_SACK_BLKS) 2364 tp->rcv_numsacks++; 2365 for (i = tp->rcv_numsacks-1; i > 0; i--) 2366 tp->sackblks[i] = tp->sackblks[i-1]; 2367 } 2368 tp->sackblks[0] = firstsack; 2369 return; 2370 } 2371 2372 /* 2373 * Process the TCP SACK option. tp->snd_holes is an ordered list 2374 * of holes (oldest to newest, in terms of the sequence space). 2375 */ 2376 void 2377 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen) 2378 { 2379 int tmp_olen; 2380 u_char *tmp_cp; 2381 struct sackhole *cur, *p, *temp; 2382 2383 if (!tp->sack_enable) 2384 return; 2385 /* SACK without ACK doesn't make sense. */ 2386 if ((th->th_flags & TH_ACK) == 0) 2387 return; 2388 /* Make sure the ACK on this segment is in [snd_una, snd_max]. */ 2389 if (SEQ_LT(th->th_ack, tp->snd_una) || 2390 SEQ_GT(th->th_ack, tp->snd_max)) 2391 return; 2392 /* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */ 2393 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2394 return; 2395 /* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */ 2396 tmp_cp = cp + 2; 2397 tmp_olen = optlen - 2; 2398 tcpstat_inc(tcps_sack_rcv_opts); 2399 if (tp->snd_numholes < 0) 2400 tp->snd_numholes = 0; 2401 if (tp->t_maxseg == 0) 2402 panic("tcp_sack_option"); /* Should never happen */ 2403 while (tmp_olen > 0) { 2404 struct sackblk sack; 2405 2406 memcpy(&sack.start, tmp_cp, sizeof(tcp_seq)); 2407 sack.start = ntohl(sack.start); 2408 memcpy(&sack.end, tmp_cp + sizeof(tcp_seq), sizeof(tcp_seq)); 2409 sack.end = ntohl(sack.end); 2410 tmp_olen -= TCPOLEN_SACK; 2411 tmp_cp += TCPOLEN_SACK; 2412 if (SEQ_LEQ(sack.end, sack.start)) 2413 continue; /* bad SACK fields */ 2414 if (SEQ_LEQ(sack.end, tp->snd_una)) 2415 continue; /* old block */ 2416 if (SEQ_GT(th->th_ack, tp->snd_una)) { 2417 if (SEQ_LT(sack.start, th->th_ack)) 2418 continue; 2419 } 2420 if (SEQ_GT(sack.end, tp->snd_max)) 2421 continue; 2422 if (tp->snd_holes == NULL) { /* first hole */ 2423 tp->snd_holes = (struct sackhole *) 2424 pool_get(&sackhl_pool, PR_NOWAIT); 2425 if (tp->snd_holes == NULL) { 2426 /* ENOBUFS, so ignore SACKed block for now */ 2427 goto dropped; 2428 } 2429 cur = tp->snd_holes; 2430 cur->start = th->th_ack; 2431 cur->end = sack.start; 2432 cur->rxmit = cur->start; 2433 cur->next = NULL; 2434 tp->snd_numholes = 1; 2435 tp->rcv_lastsack = sack.end; 2436 /* 2437 * dups is at least one. If more data has been 2438 * SACKed, it can be greater than one. 2439 */ 2440 cur->dups = min(tcprexmtthresh, 2441 ((sack.end - cur->end)/tp->t_maxseg)); 2442 if (cur->dups < 1) 2443 cur->dups = 1; 2444 continue; /* with next sack block */ 2445 } 2446 /* Go thru list of holes: p = previous, cur = current */ 2447 p = cur = tp->snd_holes; 2448 while (cur) { 2449 if (SEQ_LEQ(sack.end, cur->start)) 2450 /* SACKs data before the current hole */ 2451 break; /* no use going through more holes */ 2452 if (SEQ_GEQ(sack.start, cur->end)) { 2453 /* SACKs data beyond the current hole */ 2454 cur->dups++; 2455 if (((sack.end - cur->end)/tp->t_maxseg) >= 2456 tcprexmtthresh) 2457 cur->dups = tcprexmtthresh; 2458 p = cur; 2459 cur = cur->next; 2460 continue; 2461 } 2462 if (SEQ_LEQ(sack.start, cur->start)) { 2463 /* Data acks at least the beginning of hole */ 2464 if (SEQ_GEQ(sack.end, cur->end)) { 2465 /* Acks entire hole, so delete hole */ 2466 if (p != cur) { 2467 p->next = cur->next; 2468 pool_put(&sackhl_pool, cur); 2469 cur = p->next; 2470 } else { 2471 cur = cur->next; 2472 pool_put(&sackhl_pool, p); 2473 p = cur; 2474 tp->snd_holes = p; 2475 } 2476 tp->snd_numholes--; 2477 continue; 2478 } 2479 /* otherwise, move start of hole forward */ 2480 cur->start = sack.end; 2481 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); 2482 p = cur; 2483 cur = cur->next; 2484 continue; 2485 } 2486 /* move end of hole backward */ 2487 if (SEQ_GEQ(sack.end, cur->end)) { 2488 cur->end = sack.start; 2489 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 2490 cur->dups++; 2491 if (((sack.end - cur->end)/tp->t_maxseg) >= 2492 tcprexmtthresh) 2493 cur->dups = tcprexmtthresh; 2494 p = cur; 2495 cur = cur->next; 2496 continue; 2497 } 2498 if (SEQ_LT(cur->start, sack.start) && 2499 SEQ_GT(cur->end, sack.end)) { 2500 /* 2501 * ACKs some data in middle of a hole; need to 2502 * split current hole 2503 */ 2504 if (tp->snd_numholes >= TCP_SACKHOLE_LIMIT) 2505 goto dropped; 2506 temp = (struct sackhole *) 2507 pool_get(&sackhl_pool, PR_NOWAIT); 2508 if (temp == NULL) 2509 goto dropped; /* ENOBUFS */ 2510 temp->next = cur->next; 2511 temp->start = sack.end; 2512 temp->end = cur->end; 2513 temp->dups = cur->dups; 2514 temp->rxmit = SEQ_MAX(cur->rxmit, temp->start); 2515 cur->end = sack.start; 2516 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); 2517 cur->dups++; 2518 if (((sack.end - cur->end)/tp->t_maxseg) >= 2519 tcprexmtthresh) 2520 cur->dups = tcprexmtthresh; 2521 cur->next = temp; 2522 p = temp; 2523 cur = p->next; 2524 tp->snd_numholes++; 2525 } 2526 } 2527 /* At this point, p points to the last hole on the list */ 2528 if (SEQ_LT(tp->rcv_lastsack, sack.start)) { 2529 /* 2530 * Need to append new hole at end. 2531 * Last hole is p (and it's not NULL). 2532 */ 2533 if (tp->snd_numholes >= TCP_SACKHOLE_LIMIT) 2534 goto dropped; 2535 temp = (struct sackhole *) 2536 pool_get(&sackhl_pool, PR_NOWAIT); 2537 if (temp == NULL) 2538 goto dropped; /* ENOBUFS */ 2539 temp->start = tp->rcv_lastsack; 2540 temp->end = sack.start; 2541 temp->dups = min(tcprexmtthresh, 2542 ((sack.end - sack.start)/tp->t_maxseg)); 2543 if (temp->dups < 1) 2544 temp->dups = 1; 2545 temp->rxmit = temp->start; 2546 temp->next = 0; 2547 p->next = temp; 2548 tp->rcv_lastsack = sack.end; 2549 tp->snd_numholes++; 2550 } 2551 } 2552 return; 2553 dropped: 2554 tcpstat_inc(tcps_sack_drop_opts); 2555 } 2556 2557 /* 2558 * Delete stale (i.e, cumulatively ack'd) holes. Hole is deleted only if 2559 * it is completely acked; otherwise, tcp_sack_option(), called from 2560 * tcp_dooptions(), will fix up the hole. 2561 */ 2562 void 2563 tcp_del_sackholes(struct tcpcb *tp, struct tcphdr *th) 2564 { 2565 if (tp->sack_enable && tp->t_state != TCPS_LISTEN) { 2566 /* max because this could be an older ack just arrived */ 2567 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 2568 th->th_ack : tp->snd_una; 2569 struct sackhole *cur = tp->snd_holes; 2570 struct sackhole *prev; 2571 while (cur) 2572 if (SEQ_LEQ(cur->end, lastack)) { 2573 prev = cur; 2574 cur = cur->next; 2575 pool_put(&sackhl_pool, prev); 2576 tp->snd_numholes--; 2577 } else if (SEQ_LT(cur->start, lastack)) { 2578 cur->start = lastack; 2579 if (SEQ_LT(cur->rxmit, cur->start)) 2580 cur->rxmit = cur->start; 2581 break; 2582 } else 2583 break; 2584 tp->snd_holes = cur; 2585 } 2586 } 2587 2588 /* 2589 * Delete all receiver-side SACK information. 2590 */ 2591 void 2592 tcp_clean_sackreport(struct tcpcb *tp) 2593 { 2594 int i; 2595 2596 tp->rcv_numsacks = 0; 2597 for (i = 0; i < MAX_SACK_BLKS; i++) 2598 tp->sackblks[i].start = tp->sackblks[i].end=0; 2599 2600 } 2601 2602 /* 2603 * Partial ack handling within a sack recovery episode. When a partial ack 2604 * arrives, turn off retransmission timer, deflate the window, do not clear 2605 * tp->t_dupacks. 2606 */ 2607 void 2608 tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th) 2609 { 2610 /* Turn off retx. timer (will start again next segment) */ 2611 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2612 tp->t_rtttime = 0; 2613 /* 2614 * Partial window deflation. This statement relies on the 2615 * fact that tp->snd_una has not been updated yet. 2616 */ 2617 if (tp->snd_cwnd > (th->th_ack - tp->snd_una)) { 2618 tp->snd_cwnd -= th->th_ack - tp->snd_una; 2619 tp->snd_cwnd += tp->t_maxseg; 2620 } else 2621 tp->snd_cwnd = tp->t_maxseg; 2622 tp->snd_cwnd += tp->t_maxseg; 2623 tp->t_flags |= TF_NEEDOUTPUT; 2624 } 2625 2626 /* 2627 * Pull out of band byte out of a segment so 2628 * it doesn't appear in the user's data queue. 2629 * It is still reflected in the segment length for 2630 * sequencing purposes. 2631 */ 2632 void 2633 tcp_pulloutofband(struct socket *so, u_int urgent, struct mbuf *m, int off) 2634 { 2635 int cnt = off + urgent - 1; 2636 2637 while (cnt >= 0) { 2638 if (m->m_len > cnt) { 2639 char *cp = mtod(m, caddr_t) + cnt; 2640 struct tcpcb *tp = sototcpcb(so); 2641 2642 tp->t_iobc = *cp; 2643 tp->t_oobflags |= TCPOOB_HAVEDATA; 2644 memmove(cp, cp + 1, m->m_len - cnt - 1); 2645 m->m_len--; 2646 return; 2647 } 2648 cnt -= m->m_len; 2649 m = m->m_next; 2650 if (m == NULL) 2651 break; 2652 } 2653 panic("tcp_pulloutofband"); 2654 } 2655 2656 /* 2657 * Collect new round-trip time estimate 2658 * and update averages and current timeout. 2659 */ 2660 void 2661 tcp_xmit_timer(struct tcpcb *tp, int rtt) 2662 { 2663 short delta; 2664 short rttmin; 2665 2666 if (rtt < 0) 2667 rtt = 0; 2668 else if (rtt > TCP_RTT_MAX) 2669 rtt = TCP_RTT_MAX; 2670 2671 tcpstat_inc(tcps_rttupdated); 2672 if (tp->t_srtt != 0) { 2673 /* 2674 * delta is fixed point with 2 (TCP_RTT_BASE_SHIFT) bits 2675 * after the binary point (scaled by 4), whereas 2676 * srtt is stored as fixed point with 5 bits after the 2677 * binary point (i.e., scaled by 32). The following magic 2678 * is equivalent to the smoothing algorithm in rfc793 with 2679 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2680 * point). 2681 */ 2682 delta = (rtt << TCP_RTT_BASE_SHIFT) - 2683 (tp->t_srtt >> TCP_RTT_SHIFT); 2684 if ((tp->t_srtt += delta) <= 0) 2685 tp->t_srtt = 1 << TCP_RTT_BASE_SHIFT; 2686 /* 2687 * We accumulate a smoothed rtt variance (actually, a 2688 * smoothed mean difference), then set the retransmit 2689 * timer to smoothed rtt + 4 times the smoothed variance. 2690 * rttvar is stored as fixed point with 4 bits after the 2691 * binary point (scaled by 16). The following is 2692 * equivalent to rfc793 smoothing with an alpha of .75 2693 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2694 * rfc793's wired-in beta. 2695 */ 2696 if (delta < 0) 2697 delta = -delta; 2698 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2699 if ((tp->t_rttvar += delta) <= 0) 2700 tp->t_rttvar = 1 << TCP_RTT_BASE_SHIFT; 2701 } else { 2702 /* 2703 * No rtt measurement yet - use the unsmoothed rtt. 2704 * Set the variance to half the rtt (so our first 2705 * retransmit happens at 3*rtt). 2706 */ 2707 tp->t_srtt = (rtt + 1) << (TCP_RTT_SHIFT + TCP_RTT_BASE_SHIFT); 2708 tp->t_rttvar = (rtt + 1) << 2709 (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT - 1); 2710 } 2711 tp->t_rtttime = 0; 2712 tp->t_rxtshift = 0; 2713 2714 /* 2715 * the retransmit should happen at rtt + 4 * rttvar. 2716 * Because of the way we do the smoothing, srtt and rttvar 2717 * will each average +1/2 tick of bias. When we compute 2718 * the retransmit timer, we want 1/2 tick of rounding and 2719 * 1 extra tick because of +-1/2 tick uncertainty in the 2720 * firing of the timer. The bias will give us exactly the 2721 * 1.5 tick we need. But, because the bias is 2722 * statistical, we have to test that we don't drop below 2723 * the minimum feasible timer (which is 2 ticks). 2724 */ 2725 rttmin = min(max(rtt + 2, tp->t_rttmin), TCPTV_REXMTMAX); 2726 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 2727 2728 /* 2729 * We received an ack for a packet that wasn't retransmitted; 2730 * it is probably safe to discard any error indications we've 2731 * received recently. This isn't quite right, but close enough 2732 * for now (a route might have failed after we sent a segment, 2733 * and the return path might not be symmetrical). 2734 */ 2735 tp->t_softerror = 0; 2736 } 2737 2738 /* 2739 * Determine a reasonable value for maxseg size. 2740 * If the route is known, check route for mtu. 2741 * If none, use an mss that can be handled on the outgoing 2742 * interface without forcing IP to fragment; if bigger than 2743 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2744 * to utilize large mbufs. If no route is found, route has no mtu, 2745 * or the destination isn't local, use a default, hopefully conservative 2746 * size (usually 512 or the default IP max size, but no more than the mtu 2747 * of the interface), as we can't discover anything about intervening 2748 * gateways or networks. We also initialize the congestion/slow start 2749 * window to be a single segment if the destination isn't local. 2750 * While looking at the routing entry, we also initialize other path-dependent 2751 * parameters from pre-set or cached values in the routing entry. 2752 * 2753 * Also take into account the space needed for options that we 2754 * send regularly. Make maxseg shorter by that amount to assure 2755 * that we can send maxseg amount of data even when the options 2756 * are present. Store the upper limit of the length of options plus 2757 * data in maxopd. 2758 * 2759 * NOTE: offer == -1 indicates that the maxseg size changed due to 2760 * Path MTU discovery. 2761 */ 2762 int 2763 tcp_mss(struct tcpcb *tp, int offer) 2764 { 2765 struct rtentry *rt; 2766 struct ifnet *ifp = NULL; 2767 int mss, mssopt; 2768 int iphlen; 2769 struct inpcb *inp; 2770 2771 inp = tp->t_inpcb; 2772 2773 mssopt = mss = tcp_mssdflt; 2774 2775 rt = in_pcbrtentry(inp); 2776 2777 if (rt == NULL) 2778 goto out; 2779 2780 ifp = if_get(rt->rt_ifidx); 2781 if (ifp == NULL) 2782 goto out; 2783 2784 switch (tp->pf) { 2785 #ifdef INET6 2786 case AF_INET6: 2787 iphlen = sizeof(struct ip6_hdr); 2788 break; 2789 #endif 2790 case AF_INET: 2791 iphlen = sizeof(struct ip); 2792 break; 2793 default: 2794 /* the family does not support path MTU discovery */ 2795 goto out; 2796 } 2797 2798 /* 2799 * if there's an mtu associated with the route and we support 2800 * path MTU discovery for the underlying protocol family, use it. 2801 */ 2802 if (rt->rt_mtu) { 2803 /* 2804 * One may wish to lower MSS to take into account options, 2805 * especially security-related options. 2806 */ 2807 if (tp->pf == AF_INET6 && rt->rt_mtu < IPV6_MMTU) { 2808 /* 2809 * RFC2460 section 5, last paragraph: if path MTU is 2810 * smaller than 1280, use 1280 as packet size and 2811 * attach fragment header. 2812 */ 2813 mss = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) - 2814 sizeof(struct tcphdr); 2815 } else { 2816 mss = rt->rt_mtu - iphlen - 2817 sizeof(struct tcphdr); 2818 } 2819 } else if (ifp->if_flags & IFF_LOOPBACK) { 2820 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 2821 } else if (tp->pf == AF_INET) { 2822 if (ip_mtudisc) 2823 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 2824 } 2825 #ifdef INET6 2826 else if (tp->pf == AF_INET6) { 2827 /* 2828 * for IPv6, path MTU discovery is always turned on, 2829 * or the node must use packet size <= 1280. 2830 */ 2831 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 2832 } 2833 #endif /* INET6 */ 2834 2835 /* Calculate the value that we offer in TCPOPT_MAXSEG */ 2836 if (offer != -1) { 2837 mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 2838 mssopt = max(tcp_mssdflt, mssopt); 2839 } 2840 out: 2841 if_put(ifp); 2842 /* 2843 * The current mss, t_maxseg, is initialized to the default value. 2844 * If we compute a smaller value, reduce the current mss. 2845 * If we compute a larger value, return it for use in sending 2846 * a max seg size option, but don't store it for use 2847 * unless we received an offer at least that large from peer. 2848 * 2849 * However, do not accept offers lower than the minimum of 2850 * the interface MTU and 216. 2851 */ 2852 if (offer > 0) 2853 tp->t_peermss = offer; 2854 if (tp->t_peermss) 2855 mss = min(mss, max(tp->t_peermss, 216)); 2856 2857 /* sanity - at least max opt. space */ 2858 mss = max(mss, 64); 2859 2860 /* 2861 * maxopd stores the maximum length of data AND options 2862 * in a segment; maxseg is the amount of data in a normal 2863 * segment. We need to store this value (maxopd) apart 2864 * from maxseg, because now every segment carries options 2865 * and thus we normally have somewhat less data in segments. 2866 */ 2867 tp->t_maxopd = mss; 2868 2869 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2870 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2871 mss -= TCPOLEN_TSTAMP_APPA; 2872 #ifdef TCP_SIGNATURE 2873 if (tp->t_flags & TF_SIGNATURE) 2874 mss -= TCPOLEN_SIGLEN; 2875 #endif 2876 2877 if (offer == -1) { 2878 /* mss changed due to Path MTU discovery */ 2879 tp->t_flags &= ~TF_PMTUD_PEND; 2880 tp->t_pmtud_mtu_sent = 0; 2881 tp->t_pmtud_mss_acked = 0; 2882 if (mss < tp->t_maxseg) { 2883 /* 2884 * Follow suggestion in RFC 2414 to reduce the 2885 * congestion window by the ratio of the old 2886 * segment size to the new segment size. 2887 */ 2888 tp->snd_cwnd = ulmax((tp->snd_cwnd / tp->t_maxseg) * 2889 mss, mss); 2890 } 2891 } else if (tcp_do_rfc3390 == 2) { 2892 /* increase initial window */ 2893 tp->snd_cwnd = ulmin(10 * mss, ulmax(2 * mss, 14600)); 2894 } else if (tcp_do_rfc3390) { 2895 /* increase initial window */ 2896 tp->snd_cwnd = ulmin(4 * mss, ulmax(2 * mss, 4380)); 2897 } else 2898 tp->snd_cwnd = mss; 2899 2900 tp->t_maxseg = mss; 2901 2902 return (offer != -1 ? mssopt : mss); 2903 } 2904 2905 u_int 2906 tcp_hdrsz(struct tcpcb *tp) 2907 { 2908 u_int hlen; 2909 2910 switch (tp->pf) { 2911 #ifdef INET6 2912 case AF_INET6: 2913 hlen = sizeof(struct ip6_hdr); 2914 break; 2915 #endif 2916 case AF_INET: 2917 hlen = sizeof(struct ip); 2918 break; 2919 default: 2920 hlen = 0; 2921 break; 2922 } 2923 hlen += sizeof(struct tcphdr); 2924 2925 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2926 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2927 hlen += TCPOLEN_TSTAMP_APPA; 2928 #ifdef TCP_SIGNATURE 2929 if (tp->t_flags & TF_SIGNATURE) 2930 hlen += TCPOLEN_SIGLEN; 2931 #endif 2932 return (hlen); 2933 } 2934 2935 /* 2936 * Set connection variables based on the effective MSS. 2937 * We are passed the TCPCB for the actual connection. If we 2938 * are the server, we are called by the compressed state engine 2939 * when the 3-way handshake is complete. If we are the client, 2940 * we are called when we receive the SYN,ACK from the server. 2941 * 2942 * NOTE: The t_maxseg value must be initialized in the TCPCB 2943 * before this routine is called! 2944 */ 2945 void 2946 tcp_mss_update(struct tcpcb *tp) 2947 { 2948 int mss; 2949 u_long bufsize; 2950 struct rtentry *rt; 2951 struct socket *so; 2952 2953 so = tp->t_inpcb->inp_socket; 2954 mss = tp->t_maxseg; 2955 2956 rt = in_pcbrtentry(tp->t_inpcb); 2957 2958 if (rt == NULL) 2959 return; 2960 2961 bufsize = so->so_snd.sb_hiwat; 2962 if (bufsize < mss) { 2963 mss = bufsize; 2964 /* Update t_maxseg and t_maxopd */ 2965 tcp_mss(tp, mss); 2966 } else { 2967 bufsize = roundup(bufsize, mss); 2968 if (bufsize > sb_max) 2969 bufsize = sb_max; 2970 (void)sbreserve(so, &so->so_snd, bufsize); 2971 } 2972 2973 bufsize = so->so_rcv.sb_hiwat; 2974 if (bufsize > mss) { 2975 bufsize = roundup(bufsize, mss); 2976 if (bufsize > sb_max) 2977 bufsize = sb_max; 2978 (void)sbreserve(so, &so->so_rcv, bufsize); 2979 } 2980 2981 } 2982 2983 /* 2984 * When a partial ack arrives, force the retransmission of the 2985 * next unacknowledged segment. Do not clear tp->t_dupacks. 2986 * By setting snd_nxt to ti_ack, this forces retransmission timer 2987 * to be started again. 2988 */ 2989 void 2990 tcp_newreno_partialack(struct tcpcb *tp, struct tcphdr *th) 2991 { 2992 /* 2993 * snd_una has not been updated and the socket send buffer 2994 * not yet drained of the acked data, so we have to leave 2995 * snd_una as it was to get the correct data offset in 2996 * tcp_output(). 2997 */ 2998 tcp_seq onxt = tp->snd_nxt; 2999 u_long ocwnd = tp->snd_cwnd; 3000 3001 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3002 tp->t_rtttime = 0; 3003 tp->snd_nxt = th->th_ack; 3004 /* 3005 * Set snd_cwnd to one segment beyond acknowledged offset 3006 * (tp->snd_una not yet updated when this function is called) 3007 */ 3008 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3009 (void)tcp_output(tp); 3010 tp->snd_cwnd = ocwnd; 3011 if (SEQ_GT(onxt, tp->snd_nxt)) 3012 tp->snd_nxt = onxt; 3013 /* 3014 * Partial window deflation. Relies on fact that tp->snd_una 3015 * not updated yet. 3016 */ 3017 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3018 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3019 else 3020 tp->snd_cwnd = 0; 3021 tp->snd_cwnd += tp->t_maxseg; 3022 } 3023 3024 int 3025 tcp_mss_adv(struct mbuf *m, int af) 3026 { 3027 int mss = 0; 3028 int iphlen; 3029 struct ifnet *ifp = NULL; 3030 3031 if (m && (m->m_flags & M_PKTHDR)) 3032 ifp = if_get(m->m_pkthdr.ph_ifidx); 3033 3034 switch (af) { 3035 case AF_INET: 3036 if (ifp != NULL) 3037 mss = ifp->if_mtu; 3038 iphlen = sizeof(struct ip); 3039 break; 3040 #ifdef INET6 3041 case AF_INET6: 3042 if (ifp != NULL) 3043 mss = ifp->if_mtu; 3044 iphlen = sizeof(struct ip6_hdr); 3045 break; 3046 #endif 3047 default: 3048 unhandled_af(af); 3049 } 3050 if_put(ifp); 3051 mss = mss - iphlen - sizeof(struct tcphdr); 3052 return (max(mss, tcp_mssdflt)); 3053 } 3054 3055 /* 3056 * TCP compressed state engine. Currently used to hold compressed 3057 * state for SYN_RECEIVED. 3058 */ 3059 3060 /* syn hash parameters */ 3061 int tcp_syn_hash_size = TCP_SYN_HASH_SIZE; 3062 int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE; 3063 int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE; 3064 int tcp_syn_use_limit = 100000; 3065 3066 struct syn_cache_set tcp_syn_cache[2]; 3067 int tcp_syn_cache_active; 3068 3069 #define SYN_HASH(sa, sp, dp, rand) \ 3070 (((sa)->s_addr ^ (rand)[0]) * \ 3071 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp))) ^ (rand)[4])) 3072 #ifndef INET6 3073 #define SYN_HASHALL(hash, src, dst, rand) \ 3074 do { \ 3075 hash = SYN_HASH(&satosin(src)->sin_addr, \ 3076 satosin(src)->sin_port, \ 3077 satosin(dst)->sin_port, (rand)); \ 3078 } while (/*CONSTCOND*/ 0) 3079 #else 3080 #define SYN_HASH6(sa, sp, dp, rand) \ 3081 (((sa)->s6_addr32[0] ^ (rand)[0]) * \ 3082 ((sa)->s6_addr32[1] ^ (rand)[1]) * \ 3083 ((sa)->s6_addr32[2] ^ (rand)[2]) * \ 3084 ((sa)->s6_addr32[3] ^ (rand)[3]) * \ 3085 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp))) ^ (rand)[4])) 3086 3087 #define SYN_HASHALL(hash, src, dst, rand) \ 3088 do { \ 3089 switch ((src)->sa_family) { \ 3090 case AF_INET: \ 3091 hash = SYN_HASH(&satosin(src)->sin_addr, \ 3092 satosin(src)->sin_port, \ 3093 satosin(dst)->sin_port, (rand)); \ 3094 break; \ 3095 case AF_INET6: \ 3096 hash = SYN_HASH6(&satosin6(src)->sin6_addr, \ 3097 satosin6(src)->sin6_port, \ 3098 satosin6(dst)->sin6_port, (rand)); \ 3099 break; \ 3100 default: \ 3101 hash = 0; \ 3102 } \ 3103 } while (/*CONSTCOND*/0) 3104 #endif /* INET6 */ 3105 3106 void 3107 syn_cache_rm(struct syn_cache *sc) 3108 { 3109 sc->sc_flags |= SCF_DEAD; 3110 TAILQ_REMOVE(&sc->sc_buckethead->sch_bucket, sc, sc_bucketq); 3111 sc->sc_tp = NULL; 3112 LIST_REMOVE(sc, sc_tpq); 3113 sc->sc_buckethead->sch_length--; 3114 timeout_del(&sc->sc_timer); 3115 sc->sc_set->scs_count--; 3116 } 3117 3118 void 3119 syn_cache_put(struct syn_cache *sc) 3120 { 3121 m_free(sc->sc_ipopts); 3122 if (sc->sc_route4.ro_rt != NULL) { 3123 rtfree(sc->sc_route4.ro_rt); 3124 sc->sc_route4.ro_rt = NULL; 3125 } 3126 timeout_set(&sc->sc_timer, syn_cache_reaper, sc); 3127 timeout_add(&sc->sc_timer, 0); 3128 } 3129 3130 struct pool syn_cache_pool; 3131 3132 /* 3133 * We don't estimate RTT with SYNs, so each packet starts with the default 3134 * RTT and each timer step has a fixed timeout value. 3135 */ 3136 #define SYN_CACHE_TIMER_ARM(sc) \ 3137 do { \ 3138 TCPT_RANGESET((sc)->sc_rxtcur, \ 3139 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3140 TCPTV_REXMTMAX); \ 3141 if (!timeout_initialized(&(sc)->sc_timer)) \ 3142 timeout_set_proc(&(sc)->sc_timer, syn_cache_timer, (sc)); \ 3143 timeout_add(&(sc)->sc_timer, (sc)->sc_rxtcur * (hz / PR_SLOWHZ)); \ 3144 } while (/*CONSTCOND*/0) 3145 3146 #define SYN_CACHE_TIMESTAMP(sc) tcp_now + (sc)->sc_modulate 3147 3148 void 3149 syn_cache_init(void) 3150 { 3151 int i; 3152 3153 /* Initialize the hash buckets. */ 3154 tcp_syn_cache[0].scs_buckethead = mallocarray(tcp_syn_hash_size, 3155 sizeof(struct syn_cache_head), M_SYNCACHE, M_WAITOK|M_ZERO); 3156 tcp_syn_cache[1].scs_buckethead = mallocarray(tcp_syn_hash_size, 3157 sizeof(struct syn_cache_head), M_SYNCACHE, M_WAITOK|M_ZERO); 3158 tcp_syn_cache[0].scs_size = tcp_syn_hash_size; 3159 tcp_syn_cache[1].scs_size = tcp_syn_hash_size; 3160 for (i = 0; i < tcp_syn_hash_size; i++) { 3161 TAILQ_INIT(&tcp_syn_cache[0].scs_buckethead[i].sch_bucket); 3162 TAILQ_INIT(&tcp_syn_cache[1].scs_buckethead[i].sch_bucket); 3163 } 3164 3165 /* Initialize the syn cache pool. */ 3166 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, IPL_SOFTNET, 3167 0, "syncache", NULL); 3168 } 3169 3170 void 3171 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp) 3172 { 3173 struct syn_cache_set *set = &tcp_syn_cache[tcp_syn_cache_active]; 3174 struct syn_cache_head *scp; 3175 struct syn_cache *sc2; 3176 int i; 3177 3178 NET_ASSERT_LOCKED(); 3179 3180 /* 3181 * If there are no entries in the hash table, reinitialize 3182 * the hash secrets. To avoid useless cache swaps and 3183 * reinitialization, use it until the limit is reached. 3184 * An emtpy cache is also the oportunity to resize the hash. 3185 */ 3186 if (set->scs_count == 0 && set->scs_use <= 0) { 3187 set->scs_use = tcp_syn_use_limit; 3188 if (set->scs_size != tcp_syn_hash_size) { 3189 scp = mallocarray(tcp_syn_hash_size, sizeof(struct 3190 syn_cache_head), M_SYNCACHE, M_NOWAIT|M_ZERO); 3191 if (scp == NULL) { 3192 /* Try again next time. */ 3193 set->scs_use = 0; 3194 } else { 3195 free(set->scs_buckethead, M_SYNCACHE, 3196 set->scs_size * 3197 sizeof(struct syn_cache_head)); 3198 set->scs_buckethead = scp; 3199 set->scs_size = tcp_syn_hash_size; 3200 for (i = 0; i < tcp_syn_hash_size; i++) 3201 TAILQ_INIT(&scp[i].sch_bucket); 3202 } 3203 } 3204 arc4random_buf(set->scs_random, sizeof(set->scs_random)); 3205 tcpstat_inc(tcps_sc_seedrandom); 3206 } 3207 3208 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa, 3209 set->scs_random); 3210 scp = &set->scs_buckethead[sc->sc_hash % set->scs_size]; 3211 sc->sc_buckethead = scp; 3212 3213 /* 3214 * Make sure that we don't overflow the per-bucket 3215 * limit or the total cache size limit. 3216 */ 3217 if (scp->sch_length >= tcp_syn_bucket_limit) { 3218 tcpstat_inc(tcps_sc_bucketoverflow); 3219 /* 3220 * Someone might attack our bucket hash function. Reseed 3221 * with random as soon as the passive syn cache gets empty. 3222 */ 3223 set->scs_use = 0; 3224 /* 3225 * The bucket is full. Toss the oldest element in the 3226 * bucket. This will be the first entry in the bucket. 3227 */ 3228 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3229 #ifdef DIAGNOSTIC 3230 /* 3231 * This should never happen; we should always find an 3232 * entry in our bucket. 3233 */ 3234 if (sc2 == NULL) 3235 panic("%s: bucketoverflow: impossible", __func__); 3236 #endif 3237 syn_cache_rm(sc2); 3238 syn_cache_put(sc2); 3239 } else if (set->scs_count >= tcp_syn_cache_limit) { 3240 struct syn_cache_head *scp2, *sce; 3241 3242 tcpstat_inc(tcps_sc_overflowed); 3243 /* 3244 * The cache is full. Toss the oldest entry in the 3245 * first non-empty bucket we can find. 3246 * 3247 * XXX We would really like to toss the oldest 3248 * entry in the cache, but we hope that this 3249 * condition doesn't happen very often. 3250 */ 3251 scp2 = scp; 3252 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3253 sce = &set->scs_buckethead[set->scs_size]; 3254 for (++scp2; scp2 != scp; scp2++) { 3255 if (scp2 >= sce) 3256 scp2 = &set->scs_buckethead[0]; 3257 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3258 break; 3259 } 3260 #ifdef DIAGNOSTIC 3261 /* 3262 * This should never happen; we should always find a 3263 * non-empty bucket. 3264 */ 3265 if (scp2 == scp) 3266 panic("%s: cacheoverflow: impossible", 3267 __func__); 3268 #endif 3269 } 3270 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3271 syn_cache_rm(sc2); 3272 syn_cache_put(sc2); 3273 } 3274 3275 /* 3276 * Initialize the entry's timer. 3277 */ 3278 sc->sc_rxttot = 0; 3279 sc->sc_rxtshift = 0; 3280 SYN_CACHE_TIMER_ARM(sc); 3281 3282 /* Link it from tcpcb entry */ 3283 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3284 3285 /* Put it into the bucket. */ 3286 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3287 scp->sch_length++; 3288 sc->sc_set = set; 3289 set->scs_count++; 3290 set->scs_use--; 3291 3292 tcpstat_inc(tcps_sc_added); 3293 3294 /* 3295 * If the active cache has exceeded its use limit and 3296 * the passive syn cache is empty, exchange their roles. 3297 */ 3298 if (set->scs_use <= 0 && 3299 tcp_syn_cache[!tcp_syn_cache_active].scs_count == 0) 3300 tcp_syn_cache_active = !tcp_syn_cache_active; 3301 } 3302 3303 /* 3304 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3305 * If we have retransmitted an entry the maximum number of times, expire 3306 * that entry. 3307 */ 3308 void 3309 syn_cache_timer(void *arg) 3310 { 3311 struct syn_cache *sc = arg; 3312 3313 NET_LOCK(); 3314 if (sc->sc_flags & SCF_DEAD) 3315 goto out; 3316 3317 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3318 /* Drop it -- too many retransmissions. */ 3319 goto dropit; 3320 } 3321 3322 /* 3323 * Compute the total amount of time this entry has 3324 * been on a queue. If this entry has been on longer 3325 * than the keep alive timer would allow, expire it. 3326 */ 3327 sc->sc_rxttot += sc->sc_rxtcur; 3328 if (sc->sc_rxttot >= tcptv_keep_init) 3329 goto dropit; 3330 3331 tcpstat_inc(tcps_sc_retransmitted); 3332 (void) syn_cache_respond(sc, NULL); 3333 3334 /* Advance the timer back-off. */ 3335 sc->sc_rxtshift++; 3336 SYN_CACHE_TIMER_ARM(sc); 3337 3338 out: 3339 NET_UNLOCK(); 3340 return; 3341 3342 dropit: 3343 tcpstat_inc(tcps_sc_timed_out); 3344 syn_cache_rm(sc); 3345 syn_cache_put(sc); 3346 NET_UNLOCK(); 3347 } 3348 3349 void 3350 syn_cache_reaper(void *arg) 3351 { 3352 struct syn_cache *sc = arg; 3353 3354 pool_put(&syn_cache_pool, (sc)); 3355 return; 3356 } 3357 3358 /* 3359 * Remove syn cache created by the specified tcb entry, 3360 * because this does not make sense to keep them 3361 * (if there's no tcb entry, syn cache entry will never be used) 3362 */ 3363 void 3364 syn_cache_cleanup(struct tcpcb *tp) 3365 { 3366 struct syn_cache *sc, *nsc; 3367 3368 NET_ASSERT_LOCKED(); 3369 3370 LIST_FOREACH_SAFE(sc, &tp->t_sc, sc_tpq, nsc) { 3371 #ifdef DIAGNOSTIC 3372 if (sc->sc_tp != tp) 3373 panic("invalid sc_tp in syn_cache_cleanup"); 3374 #endif 3375 syn_cache_rm(sc); 3376 syn_cache_put(sc); 3377 } 3378 /* just for safety */ 3379 LIST_INIT(&tp->t_sc); 3380 } 3381 3382 /* 3383 * Find an entry in the syn cache. 3384 */ 3385 struct syn_cache * 3386 syn_cache_lookup(struct sockaddr *src, struct sockaddr *dst, 3387 struct syn_cache_head **headp, u_int rtableid) 3388 { 3389 struct syn_cache_set *sets[2]; 3390 struct syn_cache *sc; 3391 struct syn_cache_head *scp; 3392 u_int32_t hash; 3393 int i; 3394 3395 NET_ASSERT_LOCKED(); 3396 3397 /* Check the active cache first, the passive cache is likely emtpy. */ 3398 sets[0] = &tcp_syn_cache[tcp_syn_cache_active]; 3399 sets[1] = &tcp_syn_cache[!tcp_syn_cache_active]; 3400 for (i = 0; i < 2; i++) { 3401 if (sets[i]->scs_count == 0) 3402 continue; 3403 SYN_HASHALL(hash, src, dst, sets[i]->scs_random); 3404 scp = &sets[i]->scs_buckethead[hash % sets[i]->scs_size]; 3405 *headp = scp; 3406 TAILQ_FOREACH(sc, &scp->sch_bucket, sc_bucketq) { 3407 if (sc->sc_hash != hash) 3408 continue; 3409 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3410 !bcmp(&sc->sc_dst, dst, dst->sa_len) && 3411 rtable_l2(rtableid) == rtable_l2(sc->sc_rtableid)) 3412 return (sc); 3413 } 3414 } 3415 return (NULL); 3416 } 3417 3418 /* 3419 * This function gets called when we receive an ACK for a 3420 * socket in the LISTEN state. We look up the connection 3421 * in the syn cache, and if its there, we pull it out of 3422 * the cache and turn it into a full-blown connection in 3423 * the SYN-RECEIVED state. 3424 * 3425 * The return values may not be immediately obvious, and their effects 3426 * can be subtle, so here they are: 3427 * 3428 * NULL SYN was not found in cache; caller should drop the 3429 * packet and send an RST. 3430 * 3431 * -1 We were unable to create the new connection, and are 3432 * aborting it. An ACK,RST is being sent to the peer 3433 * (unless we got screwey sequence numbners; see below), 3434 * because the 3-way handshake has been completed. Caller 3435 * should not free the mbuf, since we may be using it. If 3436 * we are not, we will free it. 3437 * 3438 * Otherwise, the return value is a pointer to the new socket 3439 * associated with the connection. 3440 */ 3441 struct socket * 3442 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3443 u_int hlen, u_int tlen, struct socket *so, struct mbuf *m) 3444 { 3445 struct syn_cache *sc; 3446 struct syn_cache_head *scp; 3447 struct inpcb *inp, *oldinp; 3448 struct tcpcb *tp = NULL; 3449 struct mbuf *am; 3450 struct socket *oso; 3451 3452 NET_ASSERT_LOCKED(); 3453 3454 sc = syn_cache_lookup(src, dst, &scp, sotoinpcb(so)->inp_rtableid); 3455 if (sc == NULL) 3456 return (NULL); 3457 3458 /* 3459 * Verify the sequence and ack numbers. Try getting the correct 3460 * response again. 3461 */ 3462 if ((th->th_ack != sc->sc_iss + 1) || 3463 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3464 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3465 (void) syn_cache_respond(sc, m); 3466 return ((struct socket *)(-1)); 3467 } 3468 3469 /* Remove this cache entry */ 3470 syn_cache_rm(sc); 3471 3472 /* 3473 * Ok, create the full blown connection, and set things up 3474 * as they would have been set up if we had created the 3475 * connection when the SYN arrived. If we can't create 3476 * the connection, abort it. 3477 */ 3478 oso = so; 3479 so = sonewconn(so, SS_ISCONNECTED); 3480 if (so == NULL) 3481 goto resetandabort; 3482 3483 oldinp = sotoinpcb(oso); 3484 inp = sotoinpcb(so); 3485 3486 #ifdef IPSEC 3487 /* 3488 * We need to copy the required security levels 3489 * from the old pcb. Ditto for any other 3490 * IPsec-related information. 3491 */ 3492 memcpy(inp->inp_seclevel, oldinp->inp_seclevel, 3493 sizeof(oldinp->inp_seclevel)); 3494 #endif /* IPSEC */ 3495 #ifdef INET6 3496 /* 3497 * inp still has the OLD in_pcb stuff, set the 3498 * v6-related flags on the new guy, too. 3499 */ 3500 inp->inp_flags |= (oldinp->inp_flags & INP_IPV6); 3501 if (inp->inp_flags & INP_IPV6) { 3502 inp->inp_ipv6.ip6_hlim = oldinp->inp_ipv6.ip6_hlim; 3503 inp->inp_hops = oldinp->inp_hops; 3504 } else 3505 #endif /* INET6 */ 3506 { 3507 inp->inp_ip.ip_ttl = oldinp->inp_ip.ip_ttl; 3508 } 3509 3510 #if NPF > 0 3511 if (m->m_pkthdr.pf.flags & PF_TAG_DIVERTED) { 3512 struct pf_divert *divert; 3513 3514 divert = pf_find_divert(m); 3515 KASSERT(divert != NULL); 3516 inp->inp_rtableid = divert->rdomain; 3517 } else 3518 #endif 3519 /* inherit rtable from listening socket */ 3520 inp->inp_rtableid = sc->sc_rtableid; 3521 3522 inp->inp_lport = th->th_dport; 3523 switch (src->sa_family) { 3524 #ifdef INET6 3525 case AF_INET6: 3526 inp->inp_laddr6 = satosin6(dst)->sin6_addr; 3527 break; 3528 #endif /* INET6 */ 3529 case AF_INET: 3530 inp->inp_laddr = satosin(dst)->sin_addr; 3531 inp->inp_options = ip_srcroute(m); 3532 if (inp->inp_options == NULL) { 3533 inp->inp_options = sc->sc_ipopts; 3534 sc->sc_ipopts = NULL; 3535 } 3536 break; 3537 } 3538 in_pcbrehash(inp); 3539 3540 /* 3541 * Give the new socket our cached route reference. 3542 */ 3543 if (src->sa_family == AF_INET) 3544 inp->inp_route = sc->sc_route4; /* struct assignment */ 3545 #ifdef INET6 3546 else 3547 inp->inp_route6 = sc->sc_route6; 3548 #endif 3549 sc->sc_route4.ro_rt = NULL; 3550 3551 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3552 if (am == NULL) 3553 goto resetandabort; 3554 am->m_len = src->sa_len; 3555 memcpy(mtod(am, caddr_t), src, src->sa_len); 3556 if (in_pcbconnect(inp, am)) { 3557 (void) m_free(am); 3558 goto resetandabort; 3559 } 3560 (void) m_free(am); 3561 3562 tp = intotcpcb(inp); 3563 tp->t_flags = sototcpcb(oso)->t_flags & (TF_NOPUSH|TF_NODELAY); 3564 if (sc->sc_request_r_scale != 15) { 3565 tp->requested_s_scale = sc->sc_requested_s_scale; 3566 tp->request_r_scale = sc->sc_request_r_scale; 3567 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3568 } 3569 if (sc->sc_flags & SCF_TIMESTAMP) 3570 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3571 3572 tp->t_template = tcp_template(tp); 3573 if (tp->t_template == 0) { 3574 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3575 so = NULL; 3576 goto abort; 3577 } 3578 tp->sack_enable = sc->sc_flags & SCF_SACK_PERMIT; 3579 tp->ts_modulate = sc->sc_modulate; 3580 tp->ts_recent = sc->sc_timestamp; 3581 tp->iss = sc->sc_iss; 3582 tp->irs = sc->sc_irs; 3583 tcp_sendseqinit(tp); 3584 tp->snd_last = tp->snd_una; 3585 #ifdef TCP_ECN 3586 if (sc->sc_flags & SCF_ECN_PERMIT) { 3587 tp->t_flags |= TF_ECN_PERMIT; 3588 tcpstat_inc(tcps_ecn_accepts); 3589 } 3590 #endif 3591 if (sc->sc_flags & SCF_SACK_PERMIT) 3592 tp->t_flags |= TF_SACK_PERMIT; 3593 #ifdef TCP_SIGNATURE 3594 if (sc->sc_flags & SCF_SIGNATURE) 3595 tp->t_flags |= TF_SIGNATURE; 3596 #endif 3597 tcp_rcvseqinit(tp); 3598 tp->t_state = TCPS_SYN_RECEIVED; 3599 tp->t_rcvtime = tcp_now; 3600 TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init); 3601 tcpstat_inc(tcps_accepts); 3602 3603 tcp_mss(tp, sc->sc_peermaxseg); /* sets t_maxseg */ 3604 if (sc->sc_peermaxseg) 3605 tcp_mss_update(tp); 3606 /* Reset initial window to 1 segment for retransmit */ 3607 if (sc->sc_rxtshift > 0) 3608 tp->snd_cwnd = tp->t_maxseg; 3609 tp->snd_wl1 = sc->sc_irs; 3610 tp->rcv_up = sc->sc_irs + 1; 3611 3612 /* 3613 * This is what whould have happened in tcp_output() when 3614 * the SYN,ACK was sent. 3615 */ 3616 tp->snd_up = tp->snd_una; 3617 tp->snd_max = tp->snd_nxt = tp->iss+1; 3618 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3619 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3620 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3621 tp->last_ack_sent = tp->rcv_nxt; 3622 3623 tcpstat_inc(tcps_sc_completed); 3624 syn_cache_put(sc); 3625 return (so); 3626 3627 resetandabort: 3628 tcp_respond(NULL, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack, TH_RST, 3629 m->m_pkthdr.ph_rtableid); 3630 abort: 3631 m_freem(m); 3632 if (so != NULL) 3633 (void) soabort(so); 3634 syn_cache_put(sc); 3635 tcpstat_inc(tcps_sc_aborted); 3636 return ((struct socket *)(-1)); 3637 } 3638 3639 /* 3640 * This function is called when we get a RST for a 3641 * non-existent connection, so that we can see if the 3642 * connection is in the syn cache. If it is, zap it. 3643 */ 3644 3645 void 3646 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3647 u_int rtableid) 3648 { 3649 struct syn_cache *sc; 3650 struct syn_cache_head *scp; 3651 3652 NET_ASSERT_LOCKED(); 3653 3654 if ((sc = syn_cache_lookup(src, dst, &scp, rtableid)) == NULL) 3655 return; 3656 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3657 SEQ_GT(th->th_seq, sc->sc_irs + 1)) 3658 return; 3659 syn_cache_rm(sc); 3660 tcpstat_inc(tcps_sc_reset); 3661 syn_cache_put(sc); 3662 } 3663 3664 void 3665 syn_cache_unreach(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3666 u_int rtableid) 3667 { 3668 struct syn_cache *sc; 3669 struct syn_cache_head *scp; 3670 3671 NET_ASSERT_LOCKED(); 3672 3673 if ((sc = syn_cache_lookup(src, dst, &scp, rtableid)) == NULL) 3674 return; 3675 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3676 if (ntohl (th->th_seq) != sc->sc_iss) { 3677 return; 3678 } 3679 3680 /* 3681 * If we've retransmitted 3 times and this is our second error, 3682 * we remove the entry. Otherwise, we allow it to continue on. 3683 * This prevents us from incorrectly nuking an entry during a 3684 * spurious network outage. 3685 * 3686 * See tcp_notify(). 3687 */ 3688 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3689 sc->sc_flags |= SCF_UNREACH; 3690 return; 3691 } 3692 3693 syn_cache_rm(sc); 3694 tcpstat_inc(tcps_sc_unreach); 3695 syn_cache_put(sc); 3696 } 3697 3698 /* 3699 * Given a LISTEN socket and an inbound SYN request, add 3700 * this to the syn cache, and send back a segment: 3701 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3702 * to the source. 3703 * 3704 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3705 * Doing so would require that we hold onto the data and deliver it 3706 * to the application. However, if we are the target of a SYN-flood 3707 * DoS attack, an attacker could send data which would eventually 3708 * consume all available buffer space if it were ACKed. By not ACKing 3709 * the data, we avoid this DoS scenario. 3710 */ 3711 3712 int 3713 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th, 3714 u_int iphlen, struct socket *so, struct mbuf *m, u_char *optp, int optlen, 3715 struct tcp_opt_info *oi, tcp_seq *issp) 3716 { 3717 struct tcpcb tb, *tp; 3718 long win; 3719 struct syn_cache *sc; 3720 struct syn_cache_head *scp; 3721 struct mbuf *ipopts; 3722 3723 tp = sototcpcb(so); 3724 3725 /* 3726 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3727 * 3728 * Note this check is performed in tcp_input() very early on. 3729 */ 3730 3731 /* 3732 * Initialize some local state. 3733 */ 3734 win = sbspace(so, &so->so_rcv); 3735 if (win > TCP_MAXWIN) 3736 win = TCP_MAXWIN; 3737 3738 bzero(&tb, sizeof(tb)); 3739 #ifdef TCP_SIGNATURE 3740 if (optp || (tp->t_flags & TF_SIGNATURE)) { 3741 #else 3742 if (optp) { 3743 #endif 3744 tb.pf = tp->pf; 3745 tb.sack_enable = tp->sack_enable; 3746 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3747 #ifdef TCP_SIGNATURE 3748 if (tp->t_flags & TF_SIGNATURE) 3749 tb.t_flags |= TF_SIGNATURE; 3750 #endif 3751 tb.t_state = TCPS_LISTEN; 3752 if (tcp_dooptions(&tb, optp, optlen, th, m, iphlen, oi, 3753 sotoinpcb(so)->inp_rtableid)) 3754 return (-1); 3755 } 3756 3757 switch (src->sa_family) { 3758 case AF_INET: 3759 /* 3760 * Remember the IP options, if any. 3761 */ 3762 ipopts = ip_srcroute(m); 3763 break; 3764 default: 3765 ipopts = NULL; 3766 } 3767 3768 /* 3769 * See if we already have an entry for this connection. 3770 * If we do, resend the SYN,ACK. We do not count this 3771 * as a retransmission (XXX though maybe we should). 3772 */ 3773 sc = syn_cache_lookup(src, dst, &scp, sotoinpcb(so)->inp_rtableid); 3774 if (sc != NULL) { 3775 tcpstat_inc(tcps_sc_dupesyn); 3776 if (ipopts) { 3777 /* 3778 * If we were remembering a previous source route, 3779 * forget it and use the new one we've been given. 3780 */ 3781 m_free(sc->sc_ipopts); 3782 sc->sc_ipopts = ipopts; 3783 } 3784 sc->sc_timestamp = tb.ts_recent; 3785 if (syn_cache_respond(sc, m) == 0) { 3786 tcpstat_inc(tcps_sndacks); 3787 tcpstat_inc(tcps_sndtotal); 3788 } 3789 return (0); 3790 } 3791 3792 sc = pool_get(&syn_cache_pool, PR_NOWAIT|PR_ZERO); 3793 if (sc == NULL) { 3794 m_free(ipopts); 3795 return (-1); 3796 } 3797 3798 /* 3799 * Fill in the cache, and put the necessary IP and TCP 3800 * options into the reply. 3801 */ 3802 memcpy(&sc->sc_src, src, src->sa_len); 3803 memcpy(&sc->sc_dst, dst, dst->sa_len); 3804 sc->sc_rtableid = sotoinpcb(so)->inp_rtableid; 3805 sc->sc_flags = 0; 3806 sc->sc_ipopts = ipopts; 3807 sc->sc_irs = th->th_seq; 3808 3809 sc->sc_iss = issp ? *issp : arc4random(); 3810 sc->sc_peermaxseg = oi->maxseg; 3811 sc->sc_ourmaxseg = tcp_mss_adv(m, sc->sc_src.sa.sa_family); 3812 sc->sc_win = win; 3813 sc->sc_timestamp = tb.ts_recent; 3814 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 3815 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) { 3816 sc->sc_flags |= SCF_TIMESTAMP; 3817 sc->sc_modulate = arc4random(); 3818 } 3819 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 3820 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 3821 sc->sc_requested_s_scale = tb.requested_s_scale; 3822 sc->sc_request_r_scale = 0; 3823 /* 3824 * Pick the smallest possible scaling factor that 3825 * will still allow us to scale up to sb_max. 3826 * 3827 * We do this because there are broken firewalls that 3828 * will corrupt the window scale option, leading to 3829 * the other endpoint believing that our advertised 3830 * window is unscaled. At scale factors larger than 3831 * 5 the unscaled window will drop below 1500 bytes, 3832 * leading to serious problems when traversing these 3833 * broken firewalls. 3834 * 3835 * With the default sbmax of 256K, a scale factor 3836 * of 3 will be chosen by this algorithm. Those who 3837 * choose a larger sbmax should watch out 3838 * for the compatiblity problems mentioned above. 3839 * 3840 * RFC1323: The Window field in a SYN (i.e., a <SYN> 3841 * or <SYN,ACK>) segment itself is never scaled. 3842 */ 3843 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 3844 (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max) 3845 sc->sc_request_r_scale++; 3846 } else { 3847 sc->sc_requested_s_scale = 15; 3848 sc->sc_request_r_scale = 15; 3849 } 3850 #ifdef TCP_ECN 3851 /* 3852 * if both ECE and CWR flag bits are set, peer is ECN capable. 3853 */ 3854 if (tcp_do_ecn && 3855 (th->th_flags & (TH_ECE|TH_CWR)) == (TH_ECE|TH_CWR)) 3856 sc->sc_flags |= SCF_ECN_PERMIT; 3857 #endif 3858 /* 3859 * Set SCF_SACK_PERMIT if peer did send a SACK_PERMITTED option 3860 * (i.e., if tcp_dooptions() did set TF_SACK_PERMIT). 3861 */ 3862 if (tb.sack_enable && (tb.t_flags & TF_SACK_PERMIT)) 3863 sc->sc_flags |= SCF_SACK_PERMIT; 3864 #ifdef TCP_SIGNATURE 3865 if (tb.t_flags & TF_SIGNATURE) 3866 sc->sc_flags |= SCF_SIGNATURE; 3867 #endif 3868 sc->sc_tp = tp; 3869 if (syn_cache_respond(sc, m) == 0) { 3870 syn_cache_insert(sc, tp); 3871 tcpstat_inc(tcps_sndacks); 3872 tcpstat_inc(tcps_sndtotal); 3873 } else { 3874 syn_cache_put(sc); 3875 tcpstat_inc(tcps_sc_dropped); 3876 } 3877 3878 return (0); 3879 } 3880 3881 int 3882 syn_cache_respond(struct syn_cache *sc, struct mbuf *m) 3883 { 3884 u_int8_t *optp; 3885 int optlen, error; 3886 u_int16_t tlen; 3887 struct ip *ip = NULL; 3888 #ifdef INET6 3889 struct ip6_hdr *ip6 = NULL; 3890 #endif 3891 struct tcphdr *th; 3892 u_int hlen; 3893 struct inpcb *inp; 3894 3895 switch (sc->sc_src.sa.sa_family) { 3896 case AF_INET: 3897 hlen = sizeof(struct ip); 3898 break; 3899 #ifdef INET6 3900 case AF_INET6: 3901 hlen = sizeof(struct ip6_hdr); 3902 break; 3903 #endif 3904 default: 3905 m_freem(m); 3906 return (EAFNOSUPPORT); 3907 } 3908 3909 /* Compute the size of the TCP options. */ 3910 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 3911 ((sc->sc_flags & SCF_SACK_PERMIT) ? 4 : 0) + 3912 #ifdef TCP_SIGNATURE 3913 ((sc->sc_flags & SCF_SIGNATURE) ? TCPOLEN_SIGLEN : 0) + 3914 #endif 3915 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 3916 3917 tlen = hlen + sizeof(struct tcphdr) + optlen; 3918 3919 /* 3920 * Create the IP+TCP header from scratch. 3921 */ 3922 m_freem(m); 3923 #ifdef DIAGNOSTIC 3924 if (max_linkhdr + tlen > MCLBYTES) 3925 return (ENOBUFS); 3926 #endif 3927 MGETHDR(m, M_DONTWAIT, MT_DATA); 3928 if (m && max_linkhdr + tlen > MHLEN) { 3929 MCLGET(m, M_DONTWAIT); 3930 if ((m->m_flags & M_EXT) == 0) { 3931 m_freem(m); 3932 m = NULL; 3933 } 3934 } 3935 if (m == NULL) 3936 return (ENOBUFS); 3937 3938 /* Fixup the mbuf. */ 3939 m->m_data += max_linkhdr; 3940 m->m_len = m->m_pkthdr.len = tlen; 3941 m->m_pkthdr.ph_ifidx = 0; 3942 m->m_pkthdr.ph_rtableid = sc->sc_rtableid; 3943 memset(mtod(m, u_char *), 0, tlen); 3944 3945 switch (sc->sc_src.sa.sa_family) { 3946 case AF_INET: 3947 ip = mtod(m, struct ip *); 3948 ip->ip_dst = sc->sc_src.sin.sin_addr; 3949 ip->ip_src = sc->sc_dst.sin.sin_addr; 3950 ip->ip_p = IPPROTO_TCP; 3951 th = (struct tcphdr *)(ip + 1); 3952 th->th_dport = sc->sc_src.sin.sin_port; 3953 th->th_sport = sc->sc_dst.sin.sin_port; 3954 break; 3955 #ifdef INET6 3956 case AF_INET6: 3957 ip6 = mtod(m, struct ip6_hdr *); 3958 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 3959 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 3960 ip6->ip6_nxt = IPPROTO_TCP; 3961 /* ip6_plen will be updated in ip6_output() */ 3962 th = (struct tcphdr *)(ip6 + 1); 3963 th->th_dport = sc->sc_src.sin6.sin6_port; 3964 th->th_sport = sc->sc_dst.sin6.sin6_port; 3965 break; 3966 #endif 3967 default: 3968 unhandled_af(sc->sc_src.sa.sa_family); 3969 } 3970 3971 th->th_seq = htonl(sc->sc_iss); 3972 th->th_ack = htonl(sc->sc_irs + 1); 3973 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 3974 th->th_flags = TH_SYN|TH_ACK; 3975 #ifdef TCP_ECN 3976 /* Set ECE for SYN-ACK if peer supports ECN. */ 3977 if (tcp_do_ecn && (sc->sc_flags & SCF_ECN_PERMIT)) 3978 th->th_flags |= TH_ECE; 3979 #endif 3980 th->th_win = htons(sc->sc_win); 3981 /* th_sum already 0 */ 3982 /* th_urp already 0 */ 3983 3984 /* Tack on the TCP options. */ 3985 optp = (u_int8_t *)(th + 1); 3986 *optp++ = TCPOPT_MAXSEG; 3987 *optp++ = 4; 3988 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 3989 *optp++ = sc->sc_ourmaxseg & 0xff; 3990 3991 /* Include SACK_PERMIT_HDR option if peer has already done so. */ 3992 if (sc->sc_flags & SCF_SACK_PERMIT) { 3993 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMIT_HDR); 3994 optp += 4; 3995 } 3996 3997 if (sc->sc_request_r_scale != 15) { 3998 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 3999 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4000 sc->sc_request_r_scale); 4001 optp += 4; 4002 } 4003 4004 if (sc->sc_flags & SCF_TIMESTAMP) { 4005 u_int32_t *lp = (u_int32_t *)(optp); 4006 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4007 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4008 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4009 *lp = htonl(sc->sc_timestamp); 4010 optp += TCPOLEN_TSTAMP_APPA; 4011 } 4012 4013 #ifdef TCP_SIGNATURE 4014 if (sc->sc_flags & SCF_SIGNATURE) { 4015 union sockaddr_union src, dst; 4016 struct tdb *tdb; 4017 4018 bzero(&src, sizeof(union sockaddr_union)); 4019 bzero(&dst, sizeof(union sockaddr_union)); 4020 src.sa.sa_len = sc->sc_src.sa.sa_len; 4021 src.sa.sa_family = sc->sc_src.sa.sa_family; 4022 dst.sa.sa_len = sc->sc_dst.sa.sa_len; 4023 dst.sa.sa_family = sc->sc_dst.sa.sa_family; 4024 4025 switch (sc->sc_src.sa.sa_family) { 4026 case 0: /*default to PF_INET*/ 4027 case AF_INET: 4028 src.sin.sin_addr = mtod(m, struct ip *)->ip_src; 4029 dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst; 4030 break; 4031 #ifdef INET6 4032 case AF_INET6: 4033 src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src; 4034 dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst; 4035 break; 4036 #endif /* INET6 */ 4037 } 4038 4039 tdb = gettdbbysrcdst(rtable_l2(sc->sc_rtableid), 4040 0, &src, &dst, IPPROTO_TCP); 4041 if (tdb == NULL) { 4042 m_freem(m); 4043 return (EPERM); 4044 } 4045 4046 /* Send signature option */ 4047 *(optp++) = TCPOPT_SIGNATURE; 4048 *(optp++) = TCPOLEN_SIGNATURE; 4049 4050 if (tcp_signature(tdb, sc->sc_src.sa.sa_family, m, th, 4051 hlen, 0, optp) < 0) { 4052 m_freem(m); 4053 return (EINVAL); 4054 } 4055 optp += 16; 4056 4057 /* Pad options list to the next 32 bit boundary and 4058 * terminate it. 4059 */ 4060 *optp++ = TCPOPT_NOP; 4061 *optp++ = TCPOPT_EOL; 4062 } 4063 #endif /* TCP_SIGNATURE */ 4064 4065 /* Compute the packet's checksum. */ 4066 switch (sc->sc_src.sa.sa_family) { 4067 case AF_INET: 4068 ip->ip_len = htons(tlen - hlen); 4069 th->th_sum = 0; 4070 th->th_sum = in_cksum(m, tlen); 4071 break; 4072 #ifdef INET6 4073 case AF_INET6: 4074 ip6->ip6_plen = htons(tlen - hlen); 4075 th->th_sum = 0; 4076 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4077 break; 4078 #endif 4079 } 4080 4081 /* use IPsec policy and ttl from listening socket, on SYN ACK */ 4082 inp = sc->sc_tp ? sc->sc_tp->t_inpcb : NULL; 4083 4084 /* 4085 * Fill in some straggling IP bits. Note the stack expects 4086 * ip_len to be in host order, for convenience. 4087 */ 4088 switch (sc->sc_src.sa.sa_family) { 4089 case AF_INET: 4090 ip->ip_len = htons(tlen); 4091 ip->ip_ttl = inp ? inp->inp_ip.ip_ttl : ip_defttl; 4092 if (inp != NULL) 4093 ip->ip_tos = inp->inp_ip.ip_tos; 4094 break; 4095 #ifdef INET6 4096 case AF_INET6: 4097 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4098 ip6->ip6_vfc |= IPV6_VERSION; 4099 ip6->ip6_plen = htons(tlen - hlen); 4100 /* ip6_hlim will be initialized afterwards */ 4101 /* leave flowlabel = 0, it is legal and require no state mgmt */ 4102 break; 4103 #endif 4104 } 4105 4106 switch (sc->sc_src.sa.sa_family) { 4107 case AF_INET: 4108 error = ip_output(m, sc->sc_ipopts, &sc->sc_route4, 4109 (ip_mtudisc ? IP_MTUDISC : 0), NULL, inp, 0); 4110 break; 4111 #ifdef INET6 4112 case AF_INET6: 4113 ip6->ip6_hlim = in6_selecthlim(inp); 4114 4115 error = ip6_output(m, NULL /*XXX*/, &sc->sc_route6, 0, 4116 NULL, NULL); 4117 break; 4118 #endif 4119 default: 4120 error = EAFNOSUPPORT; 4121 break; 4122 } 4123 return (error); 4124 } 4125