1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_ipfw.h" /* for ipfw_fwd */ 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #include "opt_ipsec.h" 37 #include "opt_mac.h" 38 #include "opt_tcpdebug.h" 39 #include "opt_tcp_input.h" 40 #include "opt_tcp_sack.h" 41 42 #include <sys/param.h> 43 #include <sys/kernel.h> 44 #include <sys/malloc.h> 45 #include <sys/mbuf.h> 46 #include <sys/proc.h> /* for proc0 declaration */ 47 #include <sys/protosw.h> 48 #include <sys/signalvar.h> 49 #include <sys/socket.h> 50 #include <sys/socketvar.h> 51 #include <sys/sysctl.h> 52 #include <sys/syslog.h> 53 #include <sys/systm.h> 54 55 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 56 57 #include <vm/uma.h> 58 59 #include <net/if.h> 60 #include <net/route.h> 61 62 #include <netinet/in.h> 63 #include <netinet/in_pcb.h> 64 #include <netinet/in_systm.h> 65 #include <netinet/in_var.h> 66 #include <netinet/ip.h> 67 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 68 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 69 #include <netinet/ip_var.h> 70 #include <netinet/ip_options.h> 71 #include <netinet/ip6.h> 72 #include <netinet/icmp6.h> 73 #include <netinet6/in6_pcb.h> 74 #include <netinet6/ip6_var.h> 75 #include <netinet6/nd6.h> 76 #include <netinet/tcp.h> 77 #include <netinet/tcp_fsm.h> 78 #include <netinet/tcp_seq.h> 79 #include <netinet/tcp_timer.h> 80 #include <netinet/tcp_var.h> 81 #include <netinet6/tcp6_var.h> 82 #include <netinet/tcpip.h> 83 #ifdef TCPDEBUG 84 #include <netinet/tcp_debug.h> 85 #endif /* TCPDEBUG */ 86 87 #ifdef FAST_IPSEC 88 #include <netipsec/ipsec.h> 89 #include <netipsec/ipsec6.h> 90 #endif /*FAST_IPSEC*/ 91 92 #ifdef IPSEC 93 #include <netinet6/ipsec.h> 94 #include <netinet6/ipsec6.h> 95 #include <netkey/key.h> 96 #endif /*IPSEC*/ 97 98 #include <machine/in_cksum.h> 99 100 #include <security/mac/mac_framework.h> 101 102 static const int tcprexmtthresh = 3; 103 104 struct tcpstat tcpstat; 105 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 106 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 107 108 static int tcp_log_in_vain = 0; 109 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 110 &tcp_log_in_vain, 0, "Log all incoming TCP connections"); 111 112 static int blackhole = 0; 113 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 114 &blackhole, 0, "Do not send RST when dropping refused connections"); 115 116 int tcp_delack_enabled = 1; 117 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 118 &tcp_delack_enabled, 0, 119 "Delay ACK to try and piggyback it onto a data packet"); 120 121 #ifdef TCP_DROP_SYNFIN 122 static int drop_synfin = 0; 123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 124 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 125 #endif 126 127 static int tcp_do_rfc3042 = 1; 128 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 129 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 130 131 static int tcp_do_rfc3390 = 1; 132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 133 &tcp_do_rfc3390, 0, 134 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 135 136 static int tcp_insecure_rst = 0; 137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 138 &tcp_insecure_rst, 0, 139 "Follow the old (insecure) criteria for accepting RST packets."); 140 141 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 142 "TCP Segment Reassembly Queue"); 143 144 static int tcp_reass_maxseg = 0; 145 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 146 &tcp_reass_maxseg, 0, 147 "Global maximum number of TCP Segments in Reassembly Queue"); 148 149 int tcp_reass_qsize = 0; 150 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 151 &tcp_reass_qsize, 0, 152 "Global number of TCP Segments currently in Reassembly Queue"); 153 154 static int tcp_reass_maxqlen = 48; 155 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 156 &tcp_reass_maxqlen, 0, 157 "Maximum number of TCP Segments per individual Reassembly Queue"); 158 159 static int tcp_reass_overflows = 0; 160 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 161 &tcp_reass_overflows, 0, 162 "Global number of TCP Segment Reassembly Queue Overflows"); 163 164 int tcp_do_autorcvbuf = 1; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 166 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing"); 167 168 int tcp_autorcvbuf_inc = 16*1024; 169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 170 &tcp_autorcvbuf_inc, 0, "Incrementor step size of automatic receive buffer"); 171 172 int tcp_autorcvbuf_max = 256*1024; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 174 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer"); 175 176 struct inpcbhead tcb; 177 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 178 struct inpcbinfo tcbinfo; 179 struct mtx *tcbinfo_mtx; 180 181 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 182 183 static void tcp_pulloutofband(struct socket *, 184 struct tcphdr *, struct mbuf *, int); 185 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 186 struct mbuf *); 187 static void tcp_xmit_timer(struct tcpcb *, int); 188 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 189 static int tcp_timewait(struct inpcb *, struct tcpopt *, 190 struct tcphdr *, struct mbuf *, int); 191 192 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 193 #ifdef INET6 194 #define ND6_HINT(tp) \ 195 do { \ 196 if ((tp) && (tp)->t_inpcb && \ 197 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 198 nd6_nud_hint(NULL, NULL, 0); \ 199 } while (0) 200 #else 201 #define ND6_HINT(tp) 202 #endif 203 204 /* 205 * Indicate whether this ack should be delayed. We can delay the ack if 206 * - there is no delayed ack timer in progress and 207 * - our last ack wasn't a 0-sized window. We never want to delay 208 * the ack that opens up a 0-sized window and 209 * - delayed acks are enabled or 210 * - this is a half-synchronized T/TCP connection. 211 */ 212 #define DELAY_ACK(tp) \ 213 ((!callout_active(tp->tt_delack) && \ 214 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 215 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 216 217 /* Initialize TCP reassembly queue */ 218 static void 219 tcp_reass_zone_change(void *tag) 220 { 221 222 tcp_reass_maxseg = nmbclusters / 16; 223 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 224 } 225 226 uma_zone_t tcp_reass_zone; 227 void 228 tcp_reass_init() 229 { 230 tcp_reass_maxseg = nmbclusters / 16; 231 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 232 &tcp_reass_maxseg); 233 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 234 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 235 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 236 EVENTHANDLER_REGISTER(nmbclusters_change, 237 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY); 238 } 239 240 static int 241 tcp_reass(tp, th, tlenp, m) 242 register struct tcpcb *tp; 243 register struct tcphdr *th; 244 int *tlenp; 245 struct mbuf *m; 246 { 247 struct tseg_qent *q; 248 struct tseg_qent *p = NULL; 249 struct tseg_qent *nq; 250 struct tseg_qent *te = NULL; 251 struct socket *so = tp->t_inpcb->inp_socket; 252 int flags; 253 254 INP_LOCK_ASSERT(tp->t_inpcb); 255 256 /* 257 * XXX: tcp_reass() is rather inefficient with its data structures 258 * and should be rewritten (see NetBSD for optimizations). While 259 * doing that it should move to its own file tcp_reass.c. 260 */ 261 262 /* 263 * Call with th==NULL after become established to 264 * force pre-ESTABLISHED data up to user socket. 265 */ 266 if (th == NULL) 267 goto present; 268 269 /* 270 * Limit the number of segments in the reassembly queue to prevent 271 * holding on to too many segments (and thus running out of mbufs). 272 * Make sure to let the missing segment through which caused this 273 * queue. Always keep one global queue entry spare to be able to 274 * process the missing segment. 275 */ 276 if (th->th_seq != tp->rcv_nxt && 277 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 278 tp->t_segqlen >= tcp_reass_maxqlen)) { 279 tcp_reass_overflows++; 280 tcpstat.tcps_rcvmemdrop++; 281 m_freem(m); 282 *tlenp = 0; 283 return (0); 284 } 285 286 /* 287 * Allocate a new queue entry. If we can't, or hit the zone limit 288 * just drop the pkt. 289 */ 290 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 291 if (te == NULL) { 292 tcpstat.tcps_rcvmemdrop++; 293 m_freem(m); 294 *tlenp = 0; 295 return (0); 296 } 297 tp->t_segqlen++; 298 tcp_reass_qsize++; 299 300 /* 301 * Find a segment which begins after this one does. 302 */ 303 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 304 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 305 break; 306 p = q; 307 } 308 309 /* 310 * If there is a preceding segment, it may provide some of 311 * our data already. If so, drop the data from the incoming 312 * segment. If it provides all of our data, drop us. 313 */ 314 if (p != NULL) { 315 register int i; 316 /* conversion to int (in i) handles seq wraparound */ 317 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 318 if (i > 0) { 319 if (i >= *tlenp) { 320 tcpstat.tcps_rcvduppack++; 321 tcpstat.tcps_rcvdupbyte += *tlenp; 322 m_freem(m); 323 uma_zfree(tcp_reass_zone, te); 324 tp->t_segqlen--; 325 tcp_reass_qsize--; 326 /* 327 * Try to present any queued data 328 * at the left window edge to the user. 329 * This is needed after the 3-WHS 330 * completes. 331 */ 332 goto present; /* ??? */ 333 } 334 m_adj(m, i); 335 *tlenp -= i; 336 th->th_seq += i; 337 } 338 } 339 tcpstat.tcps_rcvoopack++; 340 tcpstat.tcps_rcvoobyte += *tlenp; 341 342 /* 343 * While we overlap succeeding segments trim them or, 344 * if they are completely covered, dequeue them. 345 */ 346 while (q) { 347 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 348 if (i <= 0) 349 break; 350 if (i < q->tqe_len) { 351 q->tqe_th->th_seq += i; 352 q->tqe_len -= i; 353 m_adj(q->tqe_m, i); 354 break; 355 } 356 357 nq = LIST_NEXT(q, tqe_q); 358 LIST_REMOVE(q, tqe_q); 359 m_freem(q->tqe_m); 360 uma_zfree(tcp_reass_zone, q); 361 tp->t_segqlen--; 362 tcp_reass_qsize--; 363 q = nq; 364 } 365 366 /* Insert the new segment queue entry into place. */ 367 te->tqe_m = m; 368 te->tqe_th = th; 369 te->tqe_len = *tlenp; 370 371 if (p == NULL) { 372 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 373 } else { 374 LIST_INSERT_AFTER(p, te, tqe_q); 375 } 376 377 present: 378 /* 379 * Present data to user, advancing rcv_nxt through 380 * completed sequence space. 381 */ 382 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 383 return (0); 384 q = LIST_FIRST(&tp->t_segq); 385 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 386 return (0); 387 SOCKBUF_LOCK(&so->so_rcv); 388 do { 389 tp->rcv_nxt += q->tqe_len; 390 flags = q->tqe_th->th_flags & TH_FIN; 391 nq = LIST_NEXT(q, tqe_q); 392 LIST_REMOVE(q, tqe_q); 393 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 394 m_freem(q->tqe_m); 395 else 396 sbappendstream_locked(&so->so_rcv, q->tqe_m); 397 uma_zfree(tcp_reass_zone, q); 398 tp->t_segqlen--; 399 tcp_reass_qsize--; 400 q = nq; 401 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 402 ND6_HINT(tp); 403 sorwakeup_locked(so); 404 return (flags); 405 } 406 407 /* 408 * TCP input routine, follows pages 65-76 of the 409 * protocol specification dated September, 1981 very closely. 410 */ 411 #ifdef INET6 412 int 413 tcp6_input(mp, offp, proto) 414 struct mbuf **mp; 415 int *offp, proto; 416 { 417 register struct mbuf *m = *mp; 418 struct in6_ifaddr *ia6; 419 420 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 421 422 /* 423 * draft-itojun-ipv6-tcp-to-anycast 424 * better place to put this in? 425 */ 426 ia6 = ip6_getdstifaddr(m); 427 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 428 struct ip6_hdr *ip6; 429 430 ip6 = mtod(m, struct ip6_hdr *); 431 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 432 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 433 return IPPROTO_DONE; 434 } 435 436 tcp_input(m, *offp); 437 return IPPROTO_DONE; 438 } 439 #endif 440 441 void 442 tcp_input(m, off0) 443 register struct mbuf *m; 444 int off0; 445 { 446 register struct tcphdr *th; 447 register struct ip *ip = NULL; 448 register struct ipovly *ipov; 449 register struct inpcb *inp = NULL; 450 u_char *optp = NULL; 451 int optlen = 0; 452 int len, tlen, off; 453 int drop_hdrlen; 454 register struct tcpcb *tp = 0; 455 register int thflags; 456 struct socket *so = 0; 457 int todrop, acked, ourfinisacked, needoutput = 0; 458 u_long tiwin; 459 struct tcpopt to; /* options in this segment */ 460 int headlocked = 0; 461 #ifdef IPFIREWALL_FORWARD 462 struct m_tag *fwd_tag; 463 #endif 464 int rstreason; /* For badport_bandlim accounting purposes */ 465 466 struct ip6_hdr *ip6 = NULL; 467 #ifdef INET6 468 int isipv6; 469 char ip6buf[INET6_ADDRSTRLEN]; 470 #else 471 const int isipv6 = 0; 472 #endif 473 474 #ifdef TCPDEBUG 475 /* 476 * The size of tcp_saveipgen must be the size of the max ip header, 477 * now IPv6. 478 */ 479 u_char tcp_saveipgen[40]; 480 struct tcphdr tcp_savetcp; 481 short ostate = 0; 482 #endif 483 484 #ifdef INET6 485 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 486 #endif 487 bzero((char *)&to, sizeof(to)); 488 489 tcpstat.tcps_rcvtotal++; 490 491 if (isipv6) { 492 #ifdef INET6 493 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 494 ip6 = mtod(m, struct ip6_hdr *); 495 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 496 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 497 tcpstat.tcps_rcvbadsum++; 498 goto drop; 499 } 500 th = (struct tcphdr *)((caddr_t)ip6 + off0); 501 502 /* 503 * Be proactive about unspecified IPv6 address in source. 504 * As we use all-zero to indicate unbounded/unconnected pcb, 505 * unspecified IPv6 address can be used to confuse us. 506 * 507 * Note that packets with unspecified IPv6 destination is 508 * already dropped in ip6_input. 509 */ 510 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 511 /* XXX stat */ 512 goto drop; 513 } 514 #else 515 th = NULL; /* XXX: avoid compiler warning */ 516 #endif 517 } else { 518 /* 519 * Get IP and TCP header together in first mbuf. 520 * Note: IP leaves IP header in first mbuf. 521 */ 522 if (off0 > sizeof (struct ip)) { 523 ip_stripoptions(m, (struct mbuf *)0); 524 off0 = sizeof(struct ip); 525 } 526 if (m->m_len < sizeof (struct tcpiphdr)) { 527 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 528 tcpstat.tcps_rcvshort++; 529 return; 530 } 531 } 532 ip = mtod(m, struct ip *); 533 ipov = (struct ipovly *)ip; 534 th = (struct tcphdr *)((caddr_t)ip + off0); 535 tlen = ip->ip_len; 536 537 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 538 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 539 th->th_sum = m->m_pkthdr.csum_data; 540 else 541 th->th_sum = in_pseudo(ip->ip_src.s_addr, 542 ip->ip_dst.s_addr, 543 htonl(m->m_pkthdr.csum_data + 544 ip->ip_len + 545 IPPROTO_TCP)); 546 th->th_sum ^= 0xffff; 547 #ifdef TCPDEBUG 548 ipov->ih_len = (u_short)tlen; 549 ipov->ih_len = htons(ipov->ih_len); 550 #endif 551 } else { 552 /* 553 * Checksum extended TCP header and data. 554 */ 555 len = sizeof (struct ip) + tlen; 556 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 557 ipov->ih_len = (u_short)tlen; 558 ipov->ih_len = htons(ipov->ih_len); 559 th->th_sum = in_cksum(m, len); 560 } 561 if (th->th_sum) { 562 tcpstat.tcps_rcvbadsum++; 563 goto drop; 564 } 565 /* Re-initialization for later version check */ 566 ip->ip_v = IPVERSION; 567 } 568 569 /* 570 * Check that TCP offset makes sense, 571 * pull out TCP options and adjust length. XXX 572 */ 573 off = th->th_off << 2; 574 if (off < sizeof (struct tcphdr) || off > tlen) { 575 tcpstat.tcps_rcvbadoff++; 576 goto drop; 577 } 578 tlen -= off; /* tlen is used instead of ti->ti_len */ 579 if (off > sizeof (struct tcphdr)) { 580 if (isipv6) { 581 #ifdef INET6 582 IP6_EXTHDR_CHECK(m, off0, off, ); 583 ip6 = mtod(m, struct ip6_hdr *); 584 th = (struct tcphdr *)((caddr_t)ip6 + off0); 585 #endif 586 } else { 587 if (m->m_len < sizeof(struct ip) + off) { 588 if ((m = m_pullup(m, sizeof (struct ip) + off)) 589 == 0) { 590 tcpstat.tcps_rcvshort++; 591 return; 592 } 593 ip = mtod(m, struct ip *); 594 ipov = (struct ipovly *)ip; 595 th = (struct tcphdr *)((caddr_t)ip + off0); 596 } 597 } 598 optlen = off - sizeof (struct tcphdr); 599 optp = (u_char *)(th + 1); 600 } 601 thflags = th->th_flags; 602 603 #ifdef TCP_DROP_SYNFIN 604 /* 605 * If the drop_synfin option is enabled, drop all packets with 606 * both the SYN and FIN bits set. This prevents e.g. nmap from 607 * identifying the TCP/IP stack. 608 * 609 * This is a violation of the TCP specification. 610 */ 611 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 612 goto drop; 613 #endif 614 615 /* 616 * Convert TCP protocol specific fields to host format. 617 */ 618 th->th_seq = ntohl(th->th_seq); 619 th->th_ack = ntohl(th->th_ack); 620 th->th_win = ntohs(th->th_win); 621 th->th_urp = ntohs(th->th_urp); 622 623 /* 624 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 625 * until after ip6_savecontrol() is called and before other functions 626 * which don't want those proto headers. 627 * Because ip6_savecontrol() is going to parse the mbuf to 628 * search for data to be passed up to user-land, it wants mbuf 629 * parameters to be unchanged. 630 * XXX: the call of ip6_savecontrol() has been obsoleted based on 631 * latest version of the advanced API (20020110). 632 */ 633 drop_hdrlen = off0 + off; 634 635 /* 636 * Locate pcb for segment. 637 */ 638 INP_INFO_WLOCK(&tcbinfo); 639 headlocked = 1; 640 findpcb: 641 KASSERT(headlocked, ("tcp_input: findpcb: head not locked")); 642 #ifdef IPFIREWALL_FORWARD 643 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 644 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 645 646 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 647 struct sockaddr_in *next_hop; 648 649 next_hop = (struct sockaddr_in *)(fwd_tag+1); 650 /* 651 * Transparently forwarded. Pretend to be the destination. 652 * already got one like this? 653 */ 654 inp = in_pcblookup_hash(&tcbinfo, 655 ip->ip_src, th->th_sport, 656 ip->ip_dst, th->th_dport, 657 0, m->m_pkthdr.rcvif); 658 if (!inp) { 659 /* It's new. Try to find the ambushing socket. */ 660 inp = in_pcblookup_hash(&tcbinfo, 661 ip->ip_src, th->th_sport, 662 next_hop->sin_addr, 663 next_hop->sin_port ? 664 ntohs(next_hop->sin_port) : 665 th->th_dport, 666 INPLOOKUP_WILDCARD, 667 m->m_pkthdr.rcvif); 668 } 669 /* Remove the tag from the packet. We don't need it anymore. */ 670 m_tag_delete(m, fwd_tag); 671 } else { 672 #endif /* IPFIREWALL_FORWARD */ 673 if (isipv6) { 674 #ifdef INET6 675 inp = in6_pcblookup_hash(&tcbinfo, 676 &ip6->ip6_src, th->th_sport, 677 &ip6->ip6_dst, th->th_dport, 678 INPLOOKUP_WILDCARD, 679 m->m_pkthdr.rcvif); 680 #endif 681 } else 682 inp = in_pcblookup_hash(&tcbinfo, 683 ip->ip_src, th->th_sport, 684 ip->ip_dst, th->th_dport, 685 INPLOOKUP_WILDCARD, 686 m->m_pkthdr.rcvif); 687 #ifdef IPFIREWALL_FORWARD 688 } 689 #endif /* IPFIREWALL_FORWARD */ 690 691 #if defined(IPSEC) || defined(FAST_IPSEC) 692 #ifdef INET6 693 if (isipv6) { 694 if (inp != NULL && ipsec6_in_reject(m, inp)) { 695 #ifdef IPSEC 696 ipsec6stat.in_polvio++; 697 #endif 698 goto drop; 699 } 700 } else 701 #endif /* INET6 */ 702 if (inp != NULL && ipsec4_in_reject(m, inp)) { 703 #ifdef IPSEC 704 ipsecstat.in_polvio++; 705 #endif 706 goto drop; 707 } 708 #endif /*IPSEC || FAST_IPSEC*/ 709 710 /* 711 * If the state is CLOSED (i.e., TCB does not exist) then 712 * all data in the incoming segment is discarded. 713 * If the TCB exists but is in CLOSED state, it is embryonic, 714 * but should either do a listen or a connect soon. 715 */ 716 if (inp == NULL) { 717 if (tcp_log_in_vain) { 718 #ifdef INET6 719 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 720 #else 721 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 722 #endif 723 724 if (isipv6) { 725 #ifdef INET6 726 strcpy(dbuf, "["); 727 strcpy(sbuf, "["); 728 strcat(dbuf, 729 ip6_sprintf(ip6buf, &ip6->ip6_dst)); 730 strcat(sbuf, 731 ip6_sprintf(ip6buf, &ip6->ip6_src)); 732 strcat(dbuf, "]"); 733 strcat(sbuf, "]"); 734 #endif 735 } else { 736 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 737 strcpy(sbuf, inet_ntoa(ip->ip_src)); 738 } 739 switch (tcp_log_in_vain) { 740 case 1: 741 if ((thflags & TH_SYN) == 0) 742 break; 743 /* FALLTHROUGH */ 744 case 2: 745 log(LOG_INFO, 746 "Connection attempt to TCP %s:%d " 747 "from %s:%d flags:0x%02x\n", 748 dbuf, ntohs(th->th_dport), sbuf, 749 ntohs(th->th_sport), thflags); 750 break; 751 default: 752 break; 753 } 754 } 755 if (blackhole) { 756 switch (blackhole) { 757 case 1: 758 if (thflags & TH_SYN) 759 goto drop; 760 break; 761 case 2: 762 goto drop; 763 default: 764 goto drop; 765 } 766 } 767 rstreason = BANDLIM_RST_CLOSEDPORT; 768 goto dropwithreset; 769 } 770 INP_LOCK(inp); 771 772 /* Check the minimum TTL for socket. */ 773 if (inp->inp_ip_minttl != 0) { 774 #ifdef INET6 775 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 776 goto drop; 777 else 778 #endif 779 if (inp->inp_ip_minttl > ip->ip_ttl) 780 goto drop; 781 } 782 783 if (inp->inp_vflag & INP_TIMEWAIT) { 784 /* 785 * The only option of relevance is TOF_CC, and only if 786 * present in a SYN segment. See tcp_timewait(). 787 */ 788 if (thflags & TH_SYN) 789 tcp_dooptions(&to, optp, optlen, TO_SYN); 790 if (tcp_timewait(inp, &to, th, m, tlen)) 791 goto findpcb; 792 /* 793 * tcp_timewait unlocks inp. 794 */ 795 INP_INFO_WUNLOCK(&tcbinfo); 796 return; 797 } 798 tp = intotcpcb(inp); 799 if (tp == 0) { 800 INP_UNLOCK(inp); 801 rstreason = BANDLIM_RST_CLOSEDPORT; 802 goto dropwithreset; 803 } 804 if (tp->t_state == TCPS_CLOSED) 805 goto drop; 806 807 #ifdef MAC 808 INP_LOCK_ASSERT(inp); 809 if (mac_check_inpcb_deliver(inp, m)) 810 goto drop; 811 #endif 812 so = inp->inp_socket; 813 KASSERT(so != NULL, ("tcp_input: so == NULL")); 814 #ifdef TCPDEBUG 815 if (so->so_options & SO_DEBUG) { 816 ostate = tp->t_state; 817 if (isipv6) 818 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 819 else 820 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 821 tcp_savetcp = *th; 822 } 823 #endif 824 if (so->so_options & SO_ACCEPTCONN) { 825 struct in_conninfo inc; 826 827 bzero(&inc, sizeof(inc)); 828 #ifdef INET6 829 inc.inc_isipv6 = isipv6; 830 #endif 831 if (isipv6) { 832 inc.inc6_faddr = ip6->ip6_src; 833 inc.inc6_laddr = ip6->ip6_dst; 834 } else { 835 inc.inc_faddr = ip->ip_src; 836 inc.inc_laddr = ip->ip_dst; 837 } 838 inc.inc_fport = th->th_sport; 839 inc.inc_lport = th->th_dport; 840 841 /* 842 * If the state is LISTEN then ignore segment if it contains 843 * a RST. If the segment contains an ACK then it is bad and 844 * send a RST. If it does not contain a SYN then it is not 845 * interesting; drop it. 846 * 847 * If the state is SYN_RECEIVED (syncache) and seg contains 848 * an ACK, but not for our SYN/ACK, send a RST. If the seg 849 * contains a RST, check the sequence number to see if it 850 * is a valid reset segment. 851 */ 852 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 853 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 854 /* 855 * Parse the TCP options here because 856 * syncookies need access to the reflected 857 * timestamp. 858 */ 859 tcp_dooptions(&to, optp, optlen, 0); 860 if (!syncache_expand(&inc, &to, th, &so, m)) { 861 /* 862 * No syncache entry, or ACK was not 863 * for our SYN/ACK. Send a RST. 864 */ 865 tcpstat.tcps_badsyn++; 866 rstreason = BANDLIM_RST_OPENPORT; 867 goto dropwithreset; 868 } 869 if (so == NULL) { 870 /* 871 * Could not complete 3-way handshake, 872 * connection is being closed down, and 873 * syncache has free'd mbuf. 874 */ 875 INP_UNLOCK(inp); 876 INP_INFO_WUNLOCK(&tcbinfo); 877 return; 878 } 879 /* 880 * Socket is created in state SYN_RECEIVED. 881 * Continue processing segment. 882 */ 883 INP_UNLOCK(inp); 884 inp = sotoinpcb(so); 885 INP_LOCK(inp); 886 tp = intotcpcb(inp); 887 /* 888 * This is what would have happened in 889 * tcp_output() when the SYN,ACK was sent. 890 */ 891 tp->snd_up = tp->snd_una; 892 tp->snd_max = tp->snd_nxt = tp->iss + 1; 893 tp->last_ack_sent = tp->rcv_nxt; 894 goto after_listen; 895 } 896 if (thflags & TH_RST) { 897 syncache_chkrst(&inc, th); 898 goto drop; 899 } 900 if (thflags & TH_ACK) { 901 syncache_badack(&inc); 902 tcpstat.tcps_badsyn++; 903 rstreason = BANDLIM_RST_OPENPORT; 904 goto dropwithreset; 905 } 906 goto drop; 907 } 908 909 /* 910 * Segment's flags are (SYN) or (SYN|FIN). 911 */ 912 #ifdef INET6 913 /* 914 * If deprecated address is forbidden, 915 * we do not accept SYN to deprecated interface 916 * address to prevent any new inbound connection from 917 * getting established. 918 * When we do not accept SYN, we send a TCP RST, 919 * with deprecated source address (instead of dropping 920 * it). We compromise it as it is much better for peer 921 * to send a RST, and RST will be the final packet 922 * for the exchange. 923 * 924 * If we do not forbid deprecated addresses, we accept 925 * the SYN packet. RFC2462 does not suggest dropping 926 * SYN in this case. 927 * If we decipher RFC2462 5.5.4, it says like this: 928 * 1. use of deprecated addr with existing 929 * communication is okay - "SHOULD continue to be 930 * used" 931 * 2. use of it with new communication: 932 * (2a) "SHOULD NOT be used if alternate address 933 * with sufficient scope is available" 934 * (2b) nothing mentioned otherwise. 935 * Here we fall into (2b) case as we have no choice in 936 * our source address selection - we must obey the peer. 937 * 938 * The wording in RFC2462 is confusing, and there are 939 * multiple description text for deprecated address 940 * handling - worse, they are not exactly the same. 941 * I believe 5.5.4 is the best one, so we follow 5.5.4. 942 */ 943 if (isipv6 && !ip6_use_deprecated) { 944 struct in6_ifaddr *ia6; 945 946 if ((ia6 = ip6_getdstifaddr(m)) && 947 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 948 INP_UNLOCK(inp); 949 tp = NULL; 950 rstreason = BANDLIM_RST_OPENPORT; 951 goto dropwithreset; 952 } 953 } 954 #endif 955 /* 956 * If it is from this socket, drop it, it must be forged. 957 * Don't bother responding if the destination was a broadcast. 958 */ 959 if (th->th_dport == th->th_sport) { 960 if (isipv6) { 961 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 962 &ip6->ip6_src)) 963 goto drop; 964 } else { 965 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 966 goto drop; 967 } 968 } 969 /* 970 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 971 * 972 * Note that it is quite possible to receive unicast 973 * link-layer packets with a broadcast IP address. Use 974 * in_broadcast() to find them. 975 */ 976 if (m->m_flags & (M_BCAST|M_MCAST)) 977 goto drop; 978 if (isipv6) { 979 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 980 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 981 goto drop; 982 } else { 983 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 984 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 985 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 986 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 987 goto drop; 988 } 989 /* 990 * SYN appears to be valid; create compressed TCP state 991 * for syncache, or perform t/tcp connection. 992 */ 993 if (so->so_qlen <= so->so_qlimit) { 994 #ifdef TCPDEBUG 995 if (so->so_options & SO_DEBUG) 996 tcp_trace(TA_INPUT, ostate, tp, 997 (void *)tcp_saveipgen, &tcp_savetcp, 0); 998 #endif 999 tcp_dooptions(&to, optp, optlen, TO_SYN); 1000 if (!syncache_add(&inc, &to, th, inp, &so, m)) 1001 goto drop; /* XXX: does not happen */ 1002 if (so == NULL) { 1003 /* 1004 * Entry added to syncache, mbuf used to 1005 * send SYN,ACK packet. Everything unlocked 1006 * already. 1007 */ 1008 return; 1009 } 1010 panic("T/TCP not supported at the moment"); 1011 #if 0 /* T/TCP */ 1012 /* 1013 * Segment passed TAO tests. 1014 * XXX: Can't happen at the moment. 1015 */ 1016 INP_UNLOCK(inp); 1017 inp = sotoinpcb(so); 1018 INP_LOCK(inp); 1019 tp = intotcpcb(inp); 1020 tp->t_starttime = ticks; 1021 tp->t_state = TCPS_ESTABLISHED; 1022 1023 /* 1024 * T/TCP logic: 1025 * If there is a FIN or if there is data, then 1026 * delay SYN,ACK(SYN) in the hope of piggy-backing 1027 * it on a response segment. Otherwise must send 1028 * ACK now in case the other side is slow starting. 1029 */ 1030 if (thflags & TH_FIN || tlen != 0) 1031 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 1032 else 1033 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1034 tiwin = th->th_win << tp->snd_scale; 1035 tcpstat.tcps_connects++; 1036 soisconnected(so); 1037 goto trimthenstep6; 1038 #endif /* T/TCP */ 1039 } 1040 goto drop; 1041 } 1042 after_listen: 1043 KASSERT(headlocked, ("tcp_input: after_listen: head not locked")); 1044 INP_LOCK_ASSERT(inp); 1045 1046 /* Syncache takes care of sockets in the listen state. */ 1047 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN")); 1048 1049 /* 1050 * This is the second part of the MSS DoS prevention code (after 1051 * minmss on the sending side) and it deals with too many too small 1052 * tcp packets in a too short timeframe (1 second). 1053 * 1054 * For every full second we count the number of received packets 1055 * and bytes. If we get a lot of packets per second for this connection 1056 * (tcp_minmssoverload) we take a closer look at it and compute the 1057 * average packet size for the past second. If that is less than 1058 * tcp_minmss we get too many packets with very small payload which 1059 * is not good and burdens our system (and every packet generates 1060 * a wakeup to the process connected to our socket). We can reasonable 1061 * expect this to be small packet DoS attack to exhaust our CPU 1062 * cycles. 1063 * 1064 * Care has to be taken for the minimum packet overload value. This 1065 * value defines the minimum number of packets per second before we 1066 * start to worry. This must not be too low to avoid killing for 1067 * example interactive connections with many small packets like 1068 * telnet or SSH. 1069 * 1070 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1071 * this check. 1072 * 1073 * Account for packet if payload packet, skip over ACK, etc. 1074 */ 1075 if (tcp_minmss && tcp_minmssoverload && 1076 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1077 if ((unsigned int)(tp->rcv_second - ticks) < hz) { 1078 tp->rcv_pps++; 1079 tp->rcv_byps += tlen + off; 1080 if (tp->rcv_pps > tcp_minmssoverload) { 1081 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1082 printf("too many small tcp packets from " 1083 "%s:%u, av. %lubyte/packet, " 1084 "dropping connection\n", 1085 #ifdef INET6 1086 isipv6 ? 1087 ip6_sprintf(ip6buf, 1088 &inp->inp_inc.inc6_faddr) : 1089 #endif 1090 inet_ntoa(inp->inp_inc.inc_faddr), 1091 inp->inp_inc.inc_fport, 1092 tp->rcv_byps / tp->rcv_pps); 1093 KASSERT(headlocked, ("tcp_input: " 1094 "after_listen: tcp_drop: head " 1095 "not locked")); 1096 tp = tcp_drop(tp, ECONNRESET); 1097 tcpstat.tcps_minmssdrops++; 1098 goto drop; 1099 } 1100 } 1101 } else { 1102 tp->rcv_second = ticks + hz; 1103 tp->rcv_pps = 1; 1104 tp->rcv_byps = tlen + off; 1105 } 1106 } 1107 1108 /* 1109 * Segment received on connection. 1110 * Reset idle time and keep-alive timer. 1111 */ 1112 tp->t_rcvtime = ticks; 1113 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1114 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1115 1116 /* 1117 * Unscale the window into a 32-bit value. 1118 * This value is bogus for the TCPS_SYN_SENT state 1119 * and is overwritten later. 1120 */ 1121 tiwin = th->th_win << tp->snd_scale; 1122 1123 /* 1124 * Parse options on any incoming segment. 1125 */ 1126 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0); 1127 1128 /* 1129 * If echoed timestamp is later than the current time, 1130 * fall back to non RFC1323 RTT calculation. Normalize 1131 * timestamp if syncookies were used when this connection 1132 * was established. 1133 */ 1134 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1135 to.to_tsecr -= tp->ts_offset; 1136 if (TSTMP_GT(to.to_tsecr, ticks)) 1137 to.to_tsecr = 0; 1138 } 1139 1140 /* 1141 * Process options only when we get SYN/ACK back. The SYN case 1142 * for incoming connections is handled in tcp_syncache. 1143 * XXX this is traditional behavior, may need to be cleaned up. 1144 */ 1145 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1146 if ((to.to_flags & TOF_SCALE) && 1147 (tp->t_flags & TF_REQ_SCALE)) { 1148 tp->t_flags |= TF_RCVD_SCALE; 1149 tp->snd_scale = to.to_requested_s_scale; 1150 tp->snd_wnd = th->th_win << tp->snd_scale; 1151 tiwin = tp->snd_wnd; 1152 } 1153 if (to.to_flags & TOF_TS) { 1154 tp->t_flags |= TF_RCVD_TSTMP; 1155 tp->ts_recent = to.to_tsval; 1156 tp->ts_recent_age = ticks; 1157 } 1158 /* Initial send window, already scaled. */ 1159 tp->snd_wnd = th->th_win; 1160 if (to.to_flags & TOF_MSS) 1161 tcp_mss(tp, to.to_mss); 1162 if (tp->sack_enable) { 1163 if (!(to.to_flags & TOF_SACK)) 1164 tp->sack_enable = 0; 1165 else 1166 tp->t_flags |= TF_SACK_PERMIT; 1167 } 1168 1169 } 1170 1171 /* 1172 * Header prediction: check for the two common cases 1173 * of a uni-directional data xfer. If the packet has 1174 * no control flags, is in-sequence, the window didn't 1175 * change and we're not retransmitting, it's a 1176 * candidate. If the length is zero and the ack moved 1177 * forward, we're the sender side of the xfer. Just 1178 * free the data acked & wake any higher level process 1179 * that was blocked waiting for space. If the length 1180 * is non-zero and the ack didn't move, we're the 1181 * receiver side. If we're getting packets in-order 1182 * (the reassembly queue is empty), add the data to 1183 * the socket buffer and note that we need a delayed ack. 1184 * Make sure that the hidden state-flags are also off. 1185 * Since we check for TCPS_ESTABLISHED above, it can only 1186 * be TH_NEEDSYN. 1187 */ 1188 if (tp->t_state == TCPS_ESTABLISHED && 1189 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1190 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1191 ((to.to_flags & TOF_TS) == 0 || 1192 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1193 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1194 tp->snd_nxt == tp->snd_max) { 1195 1196 /* 1197 * If last ACK falls within this segment's sequence numbers, 1198 * record the timestamp. 1199 * NOTE that the test is modified according to the latest 1200 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1201 */ 1202 if ((to.to_flags & TOF_TS) != 0 && 1203 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1204 tp->ts_recent_age = ticks; 1205 tp->ts_recent = to.to_tsval; 1206 } 1207 1208 if (tlen == 0) { 1209 if (SEQ_GT(th->th_ack, tp->snd_una) && 1210 SEQ_LEQ(th->th_ack, tp->snd_max) && 1211 tp->snd_cwnd >= tp->snd_wnd && 1212 ((!tcp_do_newreno && !tp->sack_enable && 1213 tp->t_dupacks < tcprexmtthresh) || 1214 ((tcp_do_newreno || tp->sack_enable) && 1215 !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 && 1216 TAILQ_EMPTY(&tp->snd_holes)))) { 1217 KASSERT(headlocked, ("headlocked")); 1218 INP_INFO_WUNLOCK(&tcbinfo); 1219 headlocked = 0; 1220 /* 1221 * this is a pure ack for outstanding data. 1222 */ 1223 ++tcpstat.tcps_predack; 1224 /* 1225 * "bad retransmit" recovery 1226 */ 1227 if (tp->t_rxtshift == 1 && 1228 ticks < tp->t_badrxtwin) { 1229 ++tcpstat.tcps_sndrexmitbad; 1230 tp->snd_cwnd = tp->snd_cwnd_prev; 1231 tp->snd_ssthresh = 1232 tp->snd_ssthresh_prev; 1233 tp->snd_recover = tp->snd_recover_prev; 1234 if (tp->t_flags & TF_WASFRECOVERY) 1235 ENTER_FASTRECOVERY(tp); 1236 tp->snd_nxt = tp->snd_max; 1237 tp->t_badrxtwin = 0; 1238 } 1239 1240 /* 1241 * Recalculate the transmit timer / rtt. 1242 * 1243 * Some boxes send broken timestamp replies 1244 * during the SYN+ACK phase, ignore 1245 * timestamps of 0 or we could calculate a 1246 * huge RTT and blow up the retransmit timer. 1247 */ 1248 if ((to.to_flags & TOF_TS) != 0 && 1249 to.to_tsecr) { 1250 if (!tp->t_rttlow || 1251 tp->t_rttlow > ticks - to.to_tsecr) 1252 tp->t_rttlow = ticks - to.to_tsecr; 1253 tcp_xmit_timer(tp, 1254 ticks - to.to_tsecr + 1); 1255 } else if (tp->t_rtttime && 1256 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1257 if (!tp->t_rttlow || 1258 tp->t_rttlow > ticks - tp->t_rtttime) 1259 tp->t_rttlow = ticks - tp->t_rtttime; 1260 tcp_xmit_timer(tp, 1261 ticks - tp->t_rtttime); 1262 } 1263 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1264 acked = th->th_ack - tp->snd_una; 1265 tcpstat.tcps_rcvackpack++; 1266 tcpstat.tcps_rcvackbyte += acked; 1267 sbdrop(&so->so_snd, acked); 1268 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1269 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1270 tp->snd_recover = th->th_ack - 1; 1271 tp->snd_una = th->th_ack; 1272 /* 1273 * pull snd_wl2 up to prevent seq wrap relative 1274 * to th_ack. 1275 */ 1276 tp->snd_wl2 = th->th_ack; 1277 tp->t_dupacks = 0; 1278 m_freem(m); 1279 ND6_HINT(tp); /* some progress has been done */ 1280 1281 /* 1282 * If all outstanding data are acked, stop 1283 * retransmit timer, otherwise restart timer 1284 * using current (possibly backed-off) value. 1285 * If process is waiting for space, 1286 * wakeup/selwakeup/signal. If data 1287 * are ready to send, let tcp_output 1288 * decide between more output or persist. 1289 1290 #ifdef TCPDEBUG 1291 if (so->so_options & SO_DEBUG) 1292 tcp_trace(TA_INPUT, ostate, tp, 1293 (void *)tcp_saveipgen, 1294 &tcp_savetcp, 0); 1295 #endif 1296 */ 1297 if (tp->snd_una == tp->snd_max) 1298 callout_stop(tp->tt_rexmt); 1299 else if (!callout_active(tp->tt_persist)) 1300 callout_reset(tp->tt_rexmt, 1301 tp->t_rxtcur, 1302 tcp_timer_rexmt, tp); 1303 1304 sowwakeup(so); 1305 if (so->so_snd.sb_cc) 1306 (void) tcp_output(tp); 1307 goto check_delack; 1308 } 1309 } else if (th->th_ack == tp->snd_una && 1310 LIST_EMPTY(&tp->t_segq) && 1311 tlen <= sbspace(&so->so_rcv)) { 1312 int newsize = 0; /* automatic sockbuf scaling */ 1313 1314 KASSERT(headlocked, ("headlocked")); 1315 INP_INFO_WUNLOCK(&tcbinfo); 1316 headlocked = 0; 1317 /* 1318 * this is a pure, in-sequence data packet 1319 * with nothing on the reassembly queue and 1320 * we have enough buffer space to take it. 1321 */ 1322 /* Clean receiver SACK report if present */ 1323 if (tp->sack_enable && tp->rcv_numsacks) 1324 tcp_clean_sackreport(tp); 1325 ++tcpstat.tcps_preddat; 1326 tp->rcv_nxt += tlen; 1327 /* 1328 * Pull snd_wl1 up to prevent seq wrap relative to 1329 * th_seq. 1330 */ 1331 tp->snd_wl1 = th->th_seq; 1332 /* 1333 * Pull rcv_up up to prevent seq wrap relative to 1334 * rcv_nxt. 1335 */ 1336 tp->rcv_up = tp->rcv_nxt; 1337 tcpstat.tcps_rcvpack++; 1338 tcpstat.tcps_rcvbyte += tlen; 1339 ND6_HINT(tp); /* some progress has been done */ 1340 #ifdef TCPDEBUG 1341 if (so->so_options & SO_DEBUG) 1342 tcp_trace(TA_INPUT, ostate, tp, 1343 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1344 #endif 1345 /* 1346 * Automatic sizing of receive socket buffer. Often the send 1347 * buffer size is not optimally adjusted to the actual network 1348 * conditions at hand (delay bandwidth product). Setting the 1349 * buffer size too small limits throughput on links with high 1350 * bandwidth and high delay (eg. trans-continental/oceanic links). 1351 * 1352 * On the receive side the socket buffer memory is only rarely 1353 * used to any significant extent. This allows us to be much 1354 * more aggressive in scaling the receive socket buffer. For 1355 * the case that the buffer space is actually used to a large 1356 * extent and we run out of kernel memory we can simply drop 1357 * the new segments; TCP on the sender will just retransmit it 1358 * later. Setting the buffer size too big may only consume too 1359 * much kernel memory if the application doesn't read() from 1360 * the socket or packet loss or reordering makes use of the 1361 * reassembly queue. 1362 * 1363 * The criteria to step up the receive buffer one notch are: 1364 * 1. the number of bytes received during the time it takes 1365 * one timestamp to be reflected back to us (the RTT); 1366 * 2. received bytes per RTT is within seven eighth of the 1367 * current socket buffer size; 1368 * 3. receive buffer size has not hit maximal automatic size; 1369 * 1370 * This algorithm does one step per RTT at most and only if 1371 * we receive a bulk stream w/o packet losses or reorderings. 1372 * Shrinking the buffer during idle times is not necessary as 1373 * it doesn't consume any memory when idle. 1374 * 1375 * TODO: Only step up if the application is actually serving 1376 * the buffer to better manage the socket buffer resources. 1377 */ 1378 if (tcp_do_autorcvbuf && 1379 to.to_tsecr && 1380 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1381 if (to.to_tsecr > tp->rfbuf_ts && 1382 to.to_tsecr - tp->rfbuf_ts < hz) { 1383 if (tp->rfbuf_cnt > 1384 (so->so_rcv.sb_hiwat / 8 * 7) && 1385 so->so_rcv.sb_hiwat < 1386 tcp_autorcvbuf_max) { 1387 newsize = 1388 min(so->so_rcv.sb_hiwat + 1389 tcp_autorcvbuf_inc, 1390 tcp_autorcvbuf_max); 1391 } 1392 /* Start over with next RTT. */ 1393 tp->rfbuf_ts = 0; 1394 tp->rfbuf_cnt = 0; 1395 } else 1396 tp->rfbuf_cnt += tlen; /* add up */ 1397 } 1398 1399 /* Add data to socket buffer. */ 1400 SOCKBUF_LOCK(&so->so_rcv); 1401 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1402 m_freem(m); 1403 } else { 1404 /* 1405 * Set new socket buffer size. 1406 * Give up when limit is reached. 1407 */ 1408 if (newsize) 1409 if (!sbreserve_locked(&so->so_rcv, 1410 newsize, so, curthread)) 1411 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1412 m_adj(m, drop_hdrlen); /* delayed header drop */ 1413 sbappendstream_locked(&so->so_rcv, m); 1414 } 1415 sorwakeup_locked(so); 1416 if (DELAY_ACK(tp)) { 1417 tp->t_flags |= TF_DELACK; 1418 } else { 1419 tp->t_flags |= TF_ACKNOW; 1420 tcp_output(tp); 1421 } 1422 goto check_delack; 1423 } 1424 } 1425 1426 /* 1427 * Calculate amount of space in receive window, 1428 * and then do TCP input processing. 1429 * Receive window is amount of space in rcv queue, 1430 * but not less than advertised window. 1431 */ 1432 { int win; 1433 1434 win = sbspace(&so->so_rcv); 1435 if (win < 0) 1436 win = 0; 1437 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1438 } 1439 1440 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1441 tp->rfbuf_ts = 0; 1442 tp->rfbuf_cnt = 0; 1443 1444 switch (tp->t_state) { 1445 1446 /* 1447 * If the state is SYN_RECEIVED: 1448 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1449 */ 1450 case TCPS_SYN_RECEIVED: 1451 if ((thflags & TH_ACK) && 1452 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1453 SEQ_GT(th->th_ack, tp->snd_max))) { 1454 rstreason = BANDLIM_RST_OPENPORT; 1455 goto dropwithreset; 1456 } 1457 break; 1458 1459 /* 1460 * If the state is SYN_SENT: 1461 * if seg contains an ACK, but not for our SYN, drop the input. 1462 * if seg contains a RST, then drop the connection. 1463 * if seg does not contain SYN, then drop it. 1464 * Otherwise this is an acceptable SYN segment 1465 * initialize tp->rcv_nxt and tp->irs 1466 * if seg contains ack then advance tp->snd_una 1467 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1468 * arrange for segment to be acked (eventually) 1469 * continue processing rest of data/controls, beginning with URG 1470 */ 1471 case TCPS_SYN_SENT: 1472 if ((thflags & TH_ACK) && 1473 (SEQ_LEQ(th->th_ack, tp->iss) || 1474 SEQ_GT(th->th_ack, tp->snd_max))) { 1475 rstreason = BANDLIM_UNLIMITED; 1476 goto dropwithreset; 1477 } 1478 if (thflags & TH_RST) { 1479 if (thflags & TH_ACK) { 1480 KASSERT(headlocked, ("tcp_input: after_listen" 1481 ": tcp_drop.2: head not locked")); 1482 tp = tcp_drop(tp, ECONNREFUSED); 1483 } 1484 goto drop; 1485 } 1486 if ((thflags & TH_SYN) == 0) 1487 goto drop; 1488 1489 tp->irs = th->th_seq; 1490 tcp_rcvseqinit(tp); 1491 if (thflags & TH_ACK) { 1492 tcpstat.tcps_connects++; 1493 soisconnected(so); 1494 #ifdef MAC 1495 SOCK_LOCK(so); 1496 mac_set_socket_peer_from_mbuf(m, so); 1497 SOCK_UNLOCK(so); 1498 #endif 1499 /* Do window scaling on this connection? */ 1500 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1501 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1502 tp->rcv_scale = tp->request_r_scale; 1503 } 1504 tp->rcv_adv += tp->rcv_wnd; 1505 tp->snd_una++; /* SYN is acked */ 1506 /* 1507 * If there's data, delay ACK; if there's also a FIN 1508 * ACKNOW will be turned on later. 1509 */ 1510 if (DELAY_ACK(tp) && tlen != 0) 1511 callout_reset(tp->tt_delack, tcp_delacktime, 1512 tcp_timer_delack, tp); 1513 else 1514 tp->t_flags |= TF_ACKNOW; 1515 /* 1516 * Received <SYN,ACK> in SYN_SENT[*] state. 1517 * Transitions: 1518 * SYN_SENT --> ESTABLISHED 1519 * SYN_SENT* --> FIN_WAIT_1 1520 */ 1521 tp->t_starttime = ticks; 1522 if (tp->t_flags & TF_NEEDFIN) { 1523 tp->t_state = TCPS_FIN_WAIT_1; 1524 tp->t_flags &= ~TF_NEEDFIN; 1525 thflags &= ~TH_SYN; 1526 } else { 1527 tp->t_state = TCPS_ESTABLISHED; 1528 callout_reset(tp->tt_keep, tcp_keepidle, 1529 tcp_timer_keep, tp); 1530 } 1531 } else { 1532 /* 1533 * Received initial SYN in SYN-SENT[*] state => 1534 * simultaneous open. If segment contains CC option 1535 * and there is a cached CC, apply TAO test. 1536 * If it succeeds, connection is * half-synchronized. 1537 * Otherwise, do 3-way handshake: 1538 * SYN-SENT -> SYN-RECEIVED 1539 * SYN-SENT* -> SYN-RECEIVED* 1540 * If there was no CC option, clear cached CC value. 1541 */ 1542 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1543 callout_stop(tp->tt_rexmt); 1544 tp->t_state = TCPS_SYN_RECEIVED; 1545 } 1546 1547 #if 0 /* T/TCP */ 1548 trimthenstep6: 1549 #endif 1550 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not " 1551 "locked")); 1552 INP_LOCK_ASSERT(inp); 1553 1554 /* 1555 * Advance th->th_seq to correspond to first data byte. 1556 * If data, trim to stay within window, 1557 * dropping FIN if necessary. 1558 */ 1559 th->th_seq++; 1560 if (tlen > tp->rcv_wnd) { 1561 todrop = tlen - tp->rcv_wnd; 1562 m_adj(m, -todrop); 1563 tlen = tp->rcv_wnd; 1564 thflags &= ~TH_FIN; 1565 tcpstat.tcps_rcvpackafterwin++; 1566 tcpstat.tcps_rcvbyteafterwin += todrop; 1567 } 1568 tp->snd_wl1 = th->th_seq - 1; 1569 tp->rcv_up = th->th_seq; 1570 /* 1571 * Client side of transaction: already sent SYN and data. 1572 * If the remote host used T/TCP to validate the SYN, 1573 * our data will be ACK'd; if so, enter normal data segment 1574 * processing in the middle of step 5, ack processing. 1575 * Otherwise, goto step 6. 1576 */ 1577 if (thflags & TH_ACK) 1578 goto process_ACK; 1579 1580 goto step6; 1581 1582 /* 1583 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1584 * do normal processing. 1585 * 1586 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1587 */ 1588 case TCPS_LAST_ACK: 1589 case TCPS_CLOSING: 1590 case TCPS_TIME_WAIT: 1591 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1592 break; /* continue normal processing */ 1593 } 1594 1595 /* 1596 * States other than LISTEN or SYN_SENT. 1597 * First check the RST flag and sequence number since reset segments 1598 * are exempt from the timestamp and connection count tests. This 1599 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1600 * below which allowed reset segments in half the sequence space 1601 * to fall though and be processed (which gives forged reset 1602 * segments with a random sequence number a 50 percent chance of 1603 * killing a connection). 1604 * Then check timestamp, if present. 1605 * Then check the connection count, if present. 1606 * Then check that at least some bytes of segment are within 1607 * receive window. If segment begins before rcv_nxt, 1608 * drop leading data (and SYN); if nothing left, just ack. 1609 * 1610 * 1611 * If the RST bit is set, check the sequence number to see 1612 * if this is a valid reset segment. 1613 * RFC 793 page 37: 1614 * In all states except SYN-SENT, all reset (RST) segments 1615 * are validated by checking their SEQ-fields. A reset is 1616 * valid if its sequence number is in the window. 1617 * Note: this does not take into account delayed ACKs, so 1618 * we should test against last_ack_sent instead of rcv_nxt. 1619 * The sequence number in the reset segment is normally an 1620 * echo of our outgoing acknowlegement numbers, but some hosts 1621 * send a reset with the sequence number at the rightmost edge 1622 * of our receive window, and we have to handle this case. 1623 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1624 * that brute force RST attacks are possible. To combat this, 1625 * we use a much stricter check while in the ESTABLISHED state, 1626 * only accepting RSTs where the sequence number is equal to 1627 * last_ack_sent. In all other states (the states in which a 1628 * RST is more likely), the more permissive check is used. 1629 * If we have multiple segments in flight, the intial reset 1630 * segment sequence numbers will be to the left of last_ack_sent, 1631 * but they will eventually catch up. 1632 * In any case, it never made sense to trim reset segments to 1633 * fit the receive window since RFC 1122 says: 1634 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1635 * 1636 * A TCP SHOULD allow a received RST segment to include data. 1637 * 1638 * DISCUSSION 1639 * It has been suggested that a RST segment could contain 1640 * ASCII text that encoded and explained the cause of the 1641 * RST. No standard has yet been established for such 1642 * data. 1643 * 1644 * If the reset segment passes the sequence number test examine 1645 * the state: 1646 * SYN_RECEIVED STATE: 1647 * If passive open, return to LISTEN state. 1648 * If active open, inform user that connection was refused. 1649 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1650 * Inform user that connection was reset, and close tcb. 1651 * CLOSING, LAST_ACK STATES: 1652 * Close the tcb. 1653 * TIME_WAIT STATE: 1654 * Drop the segment - see Stevens, vol. 2, p. 964 and 1655 * RFC 1337. 1656 */ 1657 if (thflags & TH_RST) { 1658 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1659 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 1660 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 1661 switch (tp->t_state) { 1662 1663 case TCPS_SYN_RECEIVED: 1664 so->so_error = ECONNREFUSED; 1665 goto close; 1666 1667 case TCPS_ESTABLISHED: 1668 if (tp->last_ack_sent != th->th_seq && 1669 tcp_insecure_rst == 0) { 1670 tcpstat.tcps_badrst++; 1671 goto drop; 1672 } 1673 case TCPS_FIN_WAIT_1: 1674 case TCPS_FIN_WAIT_2: 1675 case TCPS_CLOSE_WAIT: 1676 so->so_error = ECONNRESET; 1677 close: 1678 tp->t_state = TCPS_CLOSED; 1679 tcpstat.tcps_drops++; 1680 KASSERT(headlocked, ("tcp_input: " 1681 "trimthenstep6: tcp_close: head not " 1682 "locked")); 1683 tp = tcp_close(tp); 1684 break; 1685 1686 case TCPS_CLOSING: 1687 case TCPS_LAST_ACK: 1688 KASSERT(headlocked, ("trimthenstep6: " 1689 "tcp_close.2: head not locked")); 1690 tp = tcp_close(tp); 1691 break; 1692 1693 case TCPS_TIME_WAIT: 1694 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1695 ("timewait")); 1696 break; 1697 } 1698 } 1699 goto drop; 1700 } 1701 1702 /* 1703 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1704 * and it's less than ts_recent, drop it. 1705 */ 1706 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1707 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1708 1709 /* Check to see if ts_recent is over 24 days old. */ 1710 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1711 /* 1712 * Invalidate ts_recent. If this segment updates 1713 * ts_recent, the age will be reset later and ts_recent 1714 * will get a valid value. If it does not, setting 1715 * ts_recent to zero will at least satisfy the 1716 * requirement that zero be placed in the timestamp 1717 * echo reply when ts_recent isn't valid. The 1718 * age isn't reset until we get a valid ts_recent 1719 * because we don't want out-of-order segments to be 1720 * dropped when ts_recent is old. 1721 */ 1722 tp->ts_recent = 0; 1723 } else { 1724 tcpstat.tcps_rcvduppack++; 1725 tcpstat.tcps_rcvdupbyte += tlen; 1726 tcpstat.tcps_pawsdrop++; 1727 if (tlen) 1728 goto dropafterack; 1729 goto drop; 1730 } 1731 } 1732 1733 /* 1734 * In the SYN-RECEIVED state, validate that the packet belongs to 1735 * this connection before trimming the data to fit the receive 1736 * window. Check the sequence number versus IRS since we know 1737 * the sequence numbers haven't wrapped. This is a partial fix 1738 * for the "LAND" DoS attack. 1739 */ 1740 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1741 rstreason = BANDLIM_RST_OPENPORT; 1742 goto dropwithreset; 1743 } 1744 1745 todrop = tp->rcv_nxt - th->th_seq; 1746 if (todrop > 0) { 1747 if (thflags & TH_SYN) { 1748 thflags &= ~TH_SYN; 1749 th->th_seq++; 1750 if (th->th_urp > 1) 1751 th->th_urp--; 1752 else 1753 thflags &= ~TH_URG; 1754 todrop--; 1755 } 1756 /* 1757 * Following if statement from Stevens, vol. 2, p. 960. 1758 */ 1759 if (todrop > tlen 1760 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1761 /* 1762 * Any valid FIN must be to the left of the window. 1763 * At this point the FIN must be a duplicate or out 1764 * of sequence; drop it. 1765 */ 1766 thflags &= ~TH_FIN; 1767 1768 /* 1769 * Send an ACK to resynchronize and drop any data. 1770 * But keep on processing for RST or ACK. 1771 */ 1772 tp->t_flags |= TF_ACKNOW; 1773 todrop = tlen; 1774 tcpstat.tcps_rcvduppack++; 1775 tcpstat.tcps_rcvdupbyte += todrop; 1776 } else { 1777 tcpstat.tcps_rcvpartduppack++; 1778 tcpstat.tcps_rcvpartdupbyte += todrop; 1779 } 1780 drop_hdrlen += todrop; /* drop from the top afterwards */ 1781 th->th_seq += todrop; 1782 tlen -= todrop; 1783 if (th->th_urp > todrop) 1784 th->th_urp -= todrop; 1785 else { 1786 thflags &= ~TH_URG; 1787 th->th_urp = 0; 1788 } 1789 } 1790 1791 /* 1792 * If new data are received on a connection after the 1793 * user processes are gone, then RST the other end. 1794 */ 1795 if ((so->so_state & SS_NOFDREF) && 1796 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1797 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not " 1798 "locked")); 1799 tp = tcp_close(tp); 1800 tcpstat.tcps_rcvafterclose++; 1801 rstreason = BANDLIM_UNLIMITED; 1802 goto dropwithreset; 1803 } 1804 1805 /* 1806 * If segment ends after window, drop trailing data 1807 * (and PUSH and FIN); if nothing left, just ACK. 1808 */ 1809 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1810 if (todrop > 0) { 1811 tcpstat.tcps_rcvpackafterwin++; 1812 if (todrop >= tlen) { 1813 tcpstat.tcps_rcvbyteafterwin += tlen; 1814 /* 1815 * If a new connection request is received 1816 * while in TIME_WAIT, drop the old connection 1817 * and start over if the sequence numbers 1818 * are above the previous ones. 1819 */ 1820 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1821 if (thflags & TH_SYN && 1822 tp->t_state == TCPS_TIME_WAIT && 1823 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1824 KASSERT(headlocked, ("trimthenstep6: " 1825 "tcp_close.4: head not locked")); 1826 tp = tcp_close(tp); 1827 goto findpcb; 1828 } 1829 /* 1830 * If window is closed can only take segments at 1831 * window edge, and have to drop data and PUSH from 1832 * incoming segments. Continue processing, but 1833 * remember to ack. Otherwise, drop segment 1834 * and ack. 1835 */ 1836 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1837 tp->t_flags |= TF_ACKNOW; 1838 tcpstat.tcps_rcvwinprobe++; 1839 } else 1840 goto dropafterack; 1841 } else 1842 tcpstat.tcps_rcvbyteafterwin += todrop; 1843 m_adj(m, -todrop); 1844 tlen -= todrop; 1845 thflags &= ~(TH_PUSH|TH_FIN); 1846 } 1847 1848 /* 1849 * If last ACK falls within this segment's sequence numbers, 1850 * record its timestamp. 1851 * NOTE: 1852 * 1) That the test incorporates suggestions from the latest 1853 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1854 * 2) That updating only on newer timestamps interferes with 1855 * our earlier PAWS tests, so this check should be solely 1856 * predicated on the sequence space of this segment. 1857 * 3) That we modify the segment boundary check to be 1858 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1859 * instead of RFC1323's 1860 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1861 * This modified check allows us to overcome RFC1323's 1862 * limitations as described in Stevens TCP/IP Illustrated 1863 * Vol. 2 p.869. In such cases, we can still calculate the 1864 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1865 */ 1866 if ((to.to_flags & TOF_TS) != 0 && 1867 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1868 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1869 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1870 tp->ts_recent_age = ticks; 1871 tp->ts_recent = to.to_tsval; 1872 } 1873 1874 /* 1875 * If a SYN is in the window, then this is an 1876 * error and we send an RST and drop the connection. 1877 */ 1878 if (thflags & TH_SYN) { 1879 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: " 1880 "head not locked")); 1881 tp = tcp_drop(tp, ECONNRESET); 1882 rstreason = BANDLIM_UNLIMITED; 1883 goto drop; 1884 } 1885 1886 /* 1887 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1888 * flag is on (half-synchronized state), then queue data for 1889 * later processing; else drop segment and return. 1890 */ 1891 if ((thflags & TH_ACK) == 0) { 1892 if (tp->t_state == TCPS_SYN_RECEIVED || 1893 (tp->t_flags & TF_NEEDSYN)) 1894 goto step6; 1895 else if (tp->t_flags & TF_ACKNOW) 1896 goto dropafterack; 1897 else 1898 goto drop; 1899 } 1900 1901 /* 1902 * Ack processing. 1903 */ 1904 switch (tp->t_state) { 1905 1906 /* 1907 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1908 * ESTABLISHED state and continue processing. 1909 * The ACK was checked above. 1910 */ 1911 case TCPS_SYN_RECEIVED: 1912 1913 tcpstat.tcps_connects++; 1914 soisconnected(so); 1915 /* Do window scaling? */ 1916 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1917 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1918 tp->rcv_scale = tp->request_r_scale; 1919 tp->snd_wnd = tiwin; 1920 } 1921 /* 1922 * Make transitions: 1923 * SYN-RECEIVED -> ESTABLISHED 1924 * SYN-RECEIVED* -> FIN-WAIT-1 1925 */ 1926 tp->t_starttime = ticks; 1927 if (tp->t_flags & TF_NEEDFIN) { 1928 tp->t_state = TCPS_FIN_WAIT_1; 1929 tp->t_flags &= ~TF_NEEDFIN; 1930 } else { 1931 tp->t_state = TCPS_ESTABLISHED; 1932 callout_reset(tp->tt_keep, tcp_keepidle, 1933 tcp_timer_keep, tp); 1934 } 1935 /* 1936 * If segment contains data or ACK, will call tcp_reass() 1937 * later; if not, do so now to pass queued data to user. 1938 */ 1939 if (tlen == 0 && (thflags & TH_FIN) == 0) 1940 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1941 (struct mbuf *)0); 1942 tp->snd_wl1 = th->th_seq - 1; 1943 /* FALLTHROUGH */ 1944 1945 /* 1946 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1947 * ACKs. If the ack is in the range 1948 * tp->snd_una < th->th_ack <= tp->snd_max 1949 * then advance tp->snd_una to th->th_ack and drop 1950 * data from the retransmission queue. If this ACK reflects 1951 * more up to date window information we update our window information. 1952 */ 1953 case TCPS_ESTABLISHED: 1954 case TCPS_FIN_WAIT_1: 1955 case TCPS_FIN_WAIT_2: 1956 case TCPS_CLOSE_WAIT: 1957 case TCPS_CLOSING: 1958 case TCPS_LAST_ACK: 1959 case TCPS_TIME_WAIT: 1960 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1961 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1962 tcpstat.tcps_rcvacktoomuch++; 1963 goto dropafterack; 1964 } 1965 if (tp->sack_enable && 1966 (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes))) 1967 tcp_sack_doack(tp, &to, th->th_ack); 1968 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1969 if (tlen == 0 && tiwin == tp->snd_wnd) { 1970 tcpstat.tcps_rcvdupack++; 1971 /* 1972 * If we have outstanding data (other than 1973 * a window probe), this is a completely 1974 * duplicate ack (ie, window info didn't 1975 * change), the ack is the biggest we've 1976 * seen and we've seen exactly our rexmt 1977 * threshhold of them, assume a packet 1978 * has been dropped and retransmit it. 1979 * Kludge snd_nxt & the congestion 1980 * window so we send only this one 1981 * packet. 1982 * 1983 * We know we're losing at the current 1984 * window size so do congestion avoidance 1985 * (set ssthresh to half the current window 1986 * and pull our congestion window back to 1987 * the new ssthresh). 1988 * 1989 * Dup acks mean that packets have left the 1990 * network (they're now cached at the receiver) 1991 * so bump cwnd by the amount in the receiver 1992 * to keep a constant cwnd packets in the 1993 * network. 1994 */ 1995 if (!callout_active(tp->tt_rexmt) || 1996 th->th_ack != tp->snd_una) 1997 tp->t_dupacks = 0; 1998 else if (++tp->t_dupacks > tcprexmtthresh || 1999 ((tcp_do_newreno || tp->sack_enable) && 2000 IN_FASTRECOVERY(tp))) { 2001 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 2002 int awnd; 2003 2004 /* 2005 * Compute the amount of data in flight first. 2006 * We can inject new data into the pipe iff 2007 * we have less than 1/2 the original window's 2008 * worth of data in flight. 2009 */ 2010 awnd = (tp->snd_nxt - tp->snd_fack) + 2011 tp->sackhint.sack_bytes_rexmit; 2012 if (awnd < tp->snd_ssthresh) { 2013 tp->snd_cwnd += tp->t_maxseg; 2014 if (tp->snd_cwnd > tp->snd_ssthresh) 2015 tp->snd_cwnd = tp->snd_ssthresh; 2016 } 2017 } else 2018 tp->snd_cwnd += tp->t_maxseg; 2019 (void) tcp_output(tp); 2020 goto drop; 2021 } else if (tp->t_dupacks == tcprexmtthresh) { 2022 tcp_seq onxt = tp->snd_nxt; 2023 u_int win; 2024 2025 /* 2026 * If we're doing sack, check to 2027 * see if we're already in sack 2028 * recovery. If we're not doing sack, 2029 * check to see if we're in newreno 2030 * recovery. 2031 */ 2032 if (tp->sack_enable) { 2033 if (IN_FASTRECOVERY(tp)) { 2034 tp->t_dupacks = 0; 2035 break; 2036 } 2037 } else if (tcp_do_newreno) { 2038 if (SEQ_LEQ(th->th_ack, 2039 tp->snd_recover)) { 2040 tp->t_dupacks = 0; 2041 break; 2042 } 2043 } 2044 win = min(tp->snd_wnd, tp->snd_cwnd) / 2045 2 / tp->t_maxseg; 2046 if (win < 2) 2047 win = 2; 2048 tp->snd_ssthresh = win * tp->t_maxseg; 2049 ENTER_FASTRECOVERY(tp); 2050 tp->snd_recover = tp->snd_max; 2051 callout_stop(tp->tt_rexmt); 2052 tp->t_rtttime = 0; 2053 if (tp->sack_enable) { 2054 tcpstat.tcps_sack_recovery_episode++; 2055 tp->sack_newdata = tp->snd_nxt; 2056 tp->snd_cwnd = tp->t_maxseg; 2057 (void) tcp_output(tp); 2058 goto drop; 2059 } 2060 tp->snd_nxt = th->th_ack; 2061 tp->snd_cwnd = tp->t_maxseg; 2062 (void) tcp_output(tp); 2063 KASSERT(tp->snd_limited <= 2, 2064 ("tp->snd_limited too big")); 2065 tp->snd_cwnd = tp->snd_ssthresh + 2066 tp->t_maxseg * 2067 (tp->t_dupacks - tp->snd_limited); 2068 if (SEQ_GT(onxt, tp->snd_nxt)) 2069 tp->snd_nxt = onxt; 2070 goto drop; 2071 } else if (tcp_do_rfc3042) { 2072 u_long oldcwnd = tp->snd_cwnd; 2073 tcp_seq oldsndmax = tp->snd_max; 2074 u_int sent; 2075 2076 KASSERT(tp->t_dupacks == 1 || 2077 tp->t_dupacks == 2, 2078 ("dupacks not 1 or 2")); 2079 if (tp->t_dupacks == 1) 2080 tp->snd_limited = 0; 2081 tp->snd_cwnd = 2082 (tp->snd_nxt - tp->snd_una) + 2083 (tp->t_dupacks - tp->snd_limited) * 2084 tp->t_maxseg; 2085 (void) tcp_output(tp); 2086 sent = tp->snd_max - oldsndmax; 2087 if (sent > tp->t_maxseg) { 2088 KASSERT((tp->t_dupacks == 2 && 2089 tp->snd_limited == 0) || 2090 (sent == tp->t_maxseg + 1 && 2091 tp->t_flags & TF_SENTFIN), 2092 ("sent too much")); 2093 tp->snd_limited = 2; 2094 } else if (sent > 0) 2095 ++tp->snd_limited; 2096 tp->snd_cwnd = oldcwnd; 2097 goto drop; 2098 } 2099 } else 2100 tp->t_dupacks = 0; 2101 break; 2102 } 2103 2104 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 2105 2106 /* 2107 * If the congestion window was inflated to account 2108 * for the other side's cached packets, retract it. 2109 */ 2110 if (tcp_do_newreno || tp->sack_enable) { 2111 if (IN_FASTRECOVERY(tp)) { 2112 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2113 if (tp->sack_enable) 2114 tcp_sack_partialack(tp, th); 2115 else 2116 tcp_newreno_partial_ack(tp, th); 2117 } else { 2118 /* 2119 * Out of fast recovery. 2120 * Window inflation should have left us 2121 * with approximately snd_ssthresh 2122 * outstanding data. 2123 * But in case we would be inclined to 2124 * send a burst, better to do it via 2125 * the slow start mechanism. 2126 */ 2127 if (SEQ_GT(th->th_ack + 2128 tp->snd_ssthresh, 2129 tp->snd_max)) 2130 tp->snd_cwnd = tp->snd_max - 2131 th->th_ack + 2132 tp->t_maxseg; 2133 else 2134 tp->snd_cwnd = tp->snd_ssthresh; 2135 } 2136 } 2137 } else { 2138 if (tp->t_dupacks >= tcprexmtthresh && 2139 tp->snd_cwnd > tp->snd_ssthresh) 2140 tp->snd_cwnd = tp->snd_ssthresh; 2141 } 2142 tp->t_dupacks = 0; 2143 /* 2144 * If we reach this point, ACK is not a duplicate, 2145 * i.e., it ACKs something we sent. 2146 */ 2147 if (tp->t_flags & TF_NEEDSYN) { 2148 /* 2149 * T/TCP: Connection was half-synchronized, and our 2150 * SYN has been ACK'd (so connection is now fully 2151 * synchronized). Go to non-starred state, 2152 * increment snd_una for ACK of SYN, and check if 2153 * we can do window scaling. 2154 */ 2155 tp->t_flags &= ~TF_NEEDSYN; 2156 tp->snd_una++; 2157 /* Do window scaling? */ 2158 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2159 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2160 tp->rcv_scale = tp->request_r_scale; 2161 /* Send window already scaled. */ 2162 } 2163 } 2164 2165 process_ACK: 2166 KASSERT(headlocked, ("tcp_input: process_ACK: head not " 2167 "locked")); 2168 INP_LOCK_ASSERT(inp); 2169 2170 acked = th->th_ack - tp->snd_una; 2171 tcpstat.tcps_rcvackpack++; 2172 tcpstat.tcps_rcvackbyte += acked; 2173 2174 /* 2175 * If we just performed our first retransmit, and the ACK 2176 * arrives within our recovery window, then it was a mistake 2177 * to do the retransmit in the first place. Recover our 2178 * original cwnd and ssthresh, and proceed to transmit where 2179 * we left off. 2180 */ 2181 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2182 ++tcpstat.tcps_sndrexmitbad; 2183 tp->snd_cwnd = tp->snd_cwnd_prev; 2184 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2185 tp->snd_recover = tp->snd_recover_prev; 2186 if (tp->t_flags & TF_WASFRECOVERY) 2187 ENTER_FASTRECOVERY(tp); 2188 tp->snd_nxt = tp->snd_max; 2189 tp->t_badrxtwin = 0; /* XXX probably not required */ 2190 } 2191 2192 /* 2193 * If we have a timestamp reply, update smoothed 2194 * round trip time. If no timestamp is present but 2195 * transmit timer is running and timed sequence 2196 * number was acked, update smoothed round trip time. 2197 * Since we now have an rtt measurement, cancel the 2198 * timer backoff (cf., Phil Karn's retransmit alg.). 2199 * Recompute the initial retransmit timer. 2200 * 2201 * Some boxes send broken timestamp replies 2202 * during the SYN+ACK phase, ignore 2203 * timestamps of 0 or we could calculate a 2204 * huge RTT and blow up the retransmit timer. 2205 */ 2206 if ((to.to_flags & TOF_TS) != 0 && 2207 to.to_tsecr) { 2208 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2209 tp->t_rttlow = ticks - to.to_tsecr; 2210 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2211 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2212 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2213 tp->t_rttlow = ticks - tp->t_rtttime; 2214 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2215 } 2216 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2217 2218 /* 2219 * If all outstanding data is acked, stop retransmit 2220 * timer and remember to restart (more output or persist). 2221 * If there is more data to be acked, restart retransmit 2222 * timer, using current (possibly backed-off) value. 2223 */ 2224 if (th->th_ack == tp->snd_max) { 2225 callout_stop(tp->tt_rexmt); 2226 needoutput = 1; 2227 } else if (!callout_active(tp->tt_persist)) 2228 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2229 tcp_timer_rexmt, tp); 2230 2231 /* 2232 * If no data (only SYN) was ACK'd, 2233 * skip rest of ACK processing. 2234 */ 2235 if (acked == 0) 2236 goto step6; 2237 2238 /* 2239 * When new data is acked, open the congestion window. 2240 * If the window gives us less than ssthresh packets 2241 * in flight, open exponentially (maxseg per packet). 2242 * Otherwise open linearly: maxseg per window 2243 * (maxseg^2 / cwnd per packet). 2244 */ 2245 if ((!tcp_do_newreno && !tp->sack_enable) || 2246 !IN_FASTRECOVERY(tp)) { 2247 register u_int cw = tp->snd_cwnd; 2248 register u_int incr = tp->t_maxseg; 2249 if (cw > tp->snd_ssthresh) 2250 incr = incr * incr / cw; 2251 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2252 } 2253 SOCKBUF_LOCK(&so->so_snd); 2254 if (acked > so->so_snd.sb_cc) { 2255 tp->snd_wnd -= so->so_snd.sb_cc; 2256 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2257 ourfinisacked = 1; 2258 } else { 2259 sbdrop_locked(&so->so_snd, acked); 2260 tp->snd_wnd -= acked; 2261 ourfinisacked = 0; 2262 } 2263 sowwakeup_locked(so); 2264 /* detect una wraparound */ 2265 if ((tcp_do_newreno || tp->sack_enable) && 2266 !IN_FASTRECOVERY(tp) && 2267 SEQ_GT(tp->snd_una, tp->snd_recover) && 2268 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2269 tp->snd_recover = th->th_ack - 1; 2270 if ((tcp_do_newreno || tp->sack_enable) && 2271 IN_FASTRECOVERY(tp) && 2272 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2273 EXIT_FASTRECOVERY(tp); 2274 tp->snd_una = th->th_ack; 2275 if (tp->sack_enable) { 2276 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2277 tp->snd_recover = tp->snd_una; 2278 } 2279 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2280 tp->snd_nxt = tp->snd_una; 2281 2282 switch (tp->t_state) { 2283 2284 /* 2285 * In FIN_WAIT_1 STATE in addition to the processing 2286 * for the ESTABLISHED state if our FIN is now acknowledged 2287 * then enter FIN_WAIT_2. 2288 */ 2289 case TCPS_FIN_WAIT_1: 2290 if (ourfinisacked) { 2291 /* 2292 * If we can't receive any more 2293 * data, then closing user can proceed. 2294 * Starting the timer is contrary to the 2295 * specification, but if we don't get a FIN 2296 * we'll hang forever. 2297 */ 2298 /* XXXjl 2299 * we should release the tp also, and use a 2300 * compressed state. 2301 */ 2302 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2303 int timeout; 2304 2305 soisdisconnected(so); 2306 timeout = (tcp_fast_finwait2_recycle) ? 2307 tcp_finwait2_timeout : tcp_maxidle; 2308 callout_reset(tp->tt_2msl, timeout, 2309 tcp_timer_2msl, tp); 2310 } 2311 tp->t_state = TCPS_FIN_WAIT_2; 2312 } 2313 break; 2314 2315 /* 2316 * In CLOSING STATE in addition to the processing for 2317 * the ESTABLISHED state if the ACK acknowledges our FIN 2318 * then enter the TIME-WAIT state, otherwise ignore 2319 * the segment. 2320 */ 2321 case TCPS_CLOSING: 2322 if (ourfinisacked) { 2323 KASSERT(headlocked, ("tcp_input: process_ACK: " 2324 "head not locked")); 2325 tcp_twstart(tp); 2326 INP_INFO_WUNLOCK(&tcbinfo); 2327 m_freem(m); 2328 return; 2329 } 2330 break; 2331 2332 /* 2333 * In LAST_ACK, we may still be waiting for data to drain 2334 * and/or to be acked, as well as for the ack of our FIN. 2335 * If our FIN is now acknowledged, delete the TCB, 2336 * enter the closed state and return. 2337 */ 2338 case TCPS_LAST_ACK: 2339 if (ourfinisacked) { 2340 KASSERT(headlocked, ("tcp_input: process_ACK:" 2341 " tcp_close: head not locked")); 2342 tp = tcp_close(tp); 2343 goto drop; 2344 } 2345 break; 2346 2347 /* 2348 * In TIME_WAIT state the only thing that should arrive 2349 * is a retransmission of the remote FIN. Acknowledge 2350 * it and restart the finack timer. 2351 */ 2352 case TCPS_TIME_WAIT: 2353 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2354 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2355 tcp_timer_2msl, tp); 2356 goto dropafterack; 2357 } 2358 } 2359 2360 step6: 2361 KASSERT(headlocked, ("tcp_input: step6: head not locked")); 2362 INP_LOCK_ASSERT(inp); 2363 2364 /* 2365 * Update window information. 2366 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2367 */ 2368 if ((thflags & TH_ACK) && 2369 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2370 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2371 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2372 /* keep track of pure window updates */ 2373 if (tlen == 0 && 2374 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2375 tcpstat.tcps_rcvwinupd++; 2376 tp->snd_wnd = tiwin; 2377 tp->snd_wl1 = th->th_seq; 2378 tp->snd_wl2 = th->th_ack; 2379 if (tp->snd_wnd > tp->max_sndwnd) 2380 tp->max_sndwnd = tp->snd_wnd; 2381 needoutput = 1; 2382 } 2383 2384 /* 2385 * Process segments with URG. 2386 */ 2387 if ((thflags & TH_URG) && th->th_urp && 2388 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2389 /* 2390 * This is a kludge, but if we receive and accept 2391 * random urgent pointers, we'll crash in 2392 * soreceive. It's hard to imagine someone 2393 * actually wanting to send this much urgent data. 2394 */ 2395 SOCKBUF_LOCK(&so->so_rcv); 2396 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2397 th->th_urp = 0; /* XXX */ 2398 thflags &= ~TH_URG; /* XXX */ 2399 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2400 goto dodata; /* XXX */ 2401 } 2402 /* 2403 * If this segment advances the known urgent pointer, 2404 * then mark the data stream. This should not happen 2405 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2406 * a FIN has been received from the remote side. 2407 * In these states we ignore the URG. 2408 * 2409 * According to RFC961 (Assigned Protocols), 2410 * the urgent pointer points to the last octet 2411 * of urgent data. We continue, however, 2412 * to consider it to indicate the first octet 2413 * of data past the urgent section as the original 2414 * spec states (in one of two places). 2415 */ 2416 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2417 tp->rcv_up = th->th_seq + th->th_urp; 2418 so->so_oobmark = so->so_rcv.sb_cc + 2419 (tp->rcv_up - tp->rcv_nxt) - 1; 2420 if (so->so_oobmark == 0) 2421 so->so_rcv.sb_state |= SBS_RCVATMARK; 2422 sohasoutofband(so); 2423 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2424 } 2425 SOCKBUF_UNLOCK(&so->so_rcv); 2426 /* 2427 * Remove out of band data so doesn't get presented to user. 2428 * This can happen independent of advancing the URG pointer, 2429 * but if two URG's are pending at once, some out-of-band 2430 * data may creep in... ick. 2431 */ 2432 if (th->th_urp <= (u_long)tlen && 2433 !(so->so_options & SO_OOBINLINE)) { 2434 /* hdr drop is delayed */ 2435 tcp_pulloutofband(so, th, m, drop_hdrlen); 2436 } 2437 } else { 2438 /* 2439 * If no out of band data is expected, 2440 * pull receive urgent pointer along 2441 * with the receive window. 2442 */ 2443 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2444 tp->rcv_up = tp->rcv_nxt; 2445 } 2446 dodata: /* XXX */ 2447 KASSERT(headlocked, ("tcp_input: dodata: head not locked")); 2448 INP_LOCK_ASSERT(inp); 2449 2450 /* 2451 * Process the segment text, merging it into the TCP sequencing queue, 2452 * and arranging for acknowledgment of receipt if necessary. 2453 * This process logically involves adjusting tp->rcv_wnd as data 2454 * is presented to the user (this happens in tcp_usrreq.c, 2455 * case PRU_RCVD). If a FIN has already been received on this 2456 * connection then we just ignore the text. 2457 */ 2458 if ((tlen || (thflags & TH_FIN)) && 2459 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2460 tcp_seq save_start = th->th_seq; 2461 tcp_seq save_end = th->th_seq + tlen; 2462 m_adj(m, drop_hdrlen); /* delayed header drop */ 2463 /* 2464 * Insert segment which includes th into TCP reassembly queue 2465 * with control block tp. Set thflags to whether reassembly now 2466 * includes a segment with FIN. This handles the common case 2467 * inline (segment is the next to be received on an established 2468 * connection, and the queue is empty), avoiding linkage into 2469 * and removal from the queue and repetition of various 2470 * conversions. 2471 * Set DELACK for segments received in order, but ack 2472 * immediately when segments are out of order (so 2473 * fast retransmit can work). 2474 */ 2475 if (th->th_seq == tp->rcv_nxt && 2476 LIST_EMPTY(&tp->t_segq) && 2477 TCPS_HAVEESTABLISHED(tp->t_state)) { 2478 if (DELAY_ACK(tp)) 2479 tp->t_flags |= TF_DELACK; 2480 else 2481 tp->t_flags |= TF_ACKNOW; 2482 tp->rcv_nxt += tlen; 2483 thflags = th->th_flags & TH_FIN; 2484 tcpstat.tcps_rcvpack++; 2485 tcpstat.tcps_rcvbyte += tlen; 2486 ND6_HINT(tp); 2487 SOCKBUF_LOCK(&so->so_rcv); 2488 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2489 m_freem(m); 2490 else 2491 sbappendstream_locked(&so->so_rcv, m); 2492 sorwakeup_locked(so); 2493 } else { 2494 thflags = tcp_reass(tp, th, &tlen, m); 2495 tp->t_flags |= TF_ACKNOW; 2496 } 2497 if (tlen > 0 && tp->sack_enable) 2498 tcp_update_sack_list(tp, save_start, save_end); 2499 /* 2500 * Note the amount of data that peer has sent into 2501 * our window, in order to estimate the sender's 2502 * buffer size. 2503 */ 2504 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2505 } else { 2506 m_freem(m); 2507 thflags &= ~TH_FIN; 2508 } 2509 2510 /* 2511 * If FIN is received ACK the FIN and let the user know 2512 * that the connection is closing. 2513 */ 2514 if (thflags & TH_FIN) { 2515 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2516 socantrcvmore(so); 2517 /* 2518 * If connection is half-synchronized 2519 * (ie NEEDSYN flag on) then delay ACK, 2520 * so it may be piggybacked when SYN is sent. 2521 * Otherwise, since we received a FIN then no 2522 * more input can be expected, send ACK now. 2523 */ 2524 if (tp->t_flags & TF_NEEDSYN) 2525 tp->t_flags |= TF_DELACK; 2526 else 2527 tp->t_flags |= TF_ACKNOW; 2528 tp->rcv_nxt++; 2529 } 2530 switch (tp->t_state) { 2531 2532 /* 2533 * In SYN_RECEIVED and ESTABLISHED STATES 2534 * enter the CLOSE_WAIT state. 2535 */ 2536 case TCPS_SYN_RECEIVED: 2537 tp->t_starttime = ticks; 2538 /*FALLTHROUGH*/ 2539 case TCPS_ESTABLISHED: 2540 tp->t_state = TCPS_CLOSE_WAIT; 2541 break; 2542 2543 /* 2544 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2545 * enter the CLOSING state. 2546 */ 2547 case TCPS_FIN_WAIT_1: 2548 tp->t_state = TCPS_CLOSING; 2549 break; 2550 2551 /* 2552 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2553 * starting the time-wait timer, turning off the other 2554 * standard timers. 2555 */ 2556 case TCPS_FIN_WAIT_2: 2557 KASSERT(headlocked == 1, ("tcp_input: dodata: " 2558 "TCP_FIN_WAIT_2: head not locked")); 2559 tcp_twstart(tp); 2560 INP_INFO_WUNLOCK(&tcbinfo); 2561 return; 2562 2563 /* 2564 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2565 */ 2566 case TCPS_TIME_WAIT: 2567 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2568 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2569 tcp_timer_2msl, tp); 2570 break; 2571 } 2572 } 2573 INP_INFO_WUNLOCK(&tcbinfo); 2574 headlocked = 0; 2575 #ifdef TCPDEBUG 2576 if (so->so_options & SO_DEBUG) 2577 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2578 &tcp_savetcp, 0); 2579 #endif 2580 2581 /* 2582 * Return any desired output. 2583 */ 2584 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2585 (void) tcp_output(tp); 2586 2587 check_delack: 2588 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked")); 2589 INP_LOCK_ASSERT(inp); 2590 if (tp->t_flags & TF_DELACK) { 2591 tp->t_flags &= ~TF_DELACK; 2592 callout_reset(tp->tt_delack, tcp_delacktime, 2593 tcp_timer_delack, tp); 2594 } 2595 INP_UNLOCK(inp); 2596 return; 2597 2598 dropafterack: 2599 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked")); 2600 /* 2601 * Generate an ACK dropping incoming segment if it occupies 2602 * sequence space, where the ACK reflects our state. 2603 * 2604 * We can now skip the test for the RST flag since all 2605 * paths to this code happen after packets containing 2606 * RST have been dropped. 2607 * 2608 * In the SYN-RECEIVED state, don't send an ACK unless the 2609 * segment we received passes the SYN-RECEIVED ACK test. 2610 * If it fails send a RST. This breaks the loop in the 2611 * "LAND" DoS attack, and also prevents an ACK storm 2612 * between two listening ports that have been sent forged 2613 * SYN segments, each with the source address of the other. 2614 */ 2615 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2616 (SEQ_GT(tp->snd_una, th->th_ack) || 2617 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2618 rstreason = BANDLIM_RST_OPENPORT; 2619 goto dropwithreset; 2620 } 2621 #ifdef TCPDEBUG 2622 if (so->so_options & SO_DEBUG) 2623 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2624 &tcp_savetcp, 0); 2625 #endif 2626 KASSERT(headlocked, ("headlocked should be 1")); 2627 INP_INFO_WUNLOCK(&tcbinfo); 2628 tp->t_flags |= TF_ACKNOW; 2629 (void) tcp_output(tp); 2630 INP_UNLOCK(inp); 2631 m_freem(m); 2632 return; 2633 2634 dropwithreset: 2635 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked")); 2636 /* 2637 * Generate a RST, dropping incoming segment. 2638 * Make ACK acceptable to originator of segment. 2639 * Don't bother to respond if destination was broadcast/multicast. 2640 */ 2641 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2642 goto drop; 2643 if (isipv6) { 2644 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2645 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2646 goto drop; 2647 } else { 2648 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2649 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2650 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2651 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2652 goto drop; 2653 } 2654 /* IPv6 anycast check is done at tcp6_input() */ 2655 2656 /* 2657 * Perform bandwidth limiting. 2658 */ 2659 if (badport_bandlim(rstreason) < 0) 2660 goto drop; 2661 2662 #ifdef TCPDEBUG 2663 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2664 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2665 &tcp_savetcp, 0); 2666 #endif 2667 2668 if (thflags & TH_ACK) 2669 /* mtod() below is safe as long as hdr dropping is delayed */ 2670 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2671 TH_RST); 2672 else { 2673 if (thflags & TH_SYN) 2674 tlen++; 2675 /* mtod() below is safe as long as hdr dropping is delayed */ 2676 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2677 (tcp_seq)0, TH_RST|TH_ACK); 2678 } 2679 2680 if (tp != NULL) 2681 INP_UNLOCK(inp); 2682 if (headlocked) 2683 INP_INFO_WUNLOCK(&tcbinfo); 2684 return; 2685 2686 drop: 2687 /* 2688 * Drop space held by incoming segment and return. 2689 */ 2690 #ifdef TCPDEBUG 2691 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2692 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2693 &tcp_savetcp, 0); 2694 #endif 2695 if (tp != NULL) 2696 INP_UNLOCK(inp); 2697 if (headlocked) 2698 INP_INFO_WUNLOCK(&tcbinfo); 2699 m_freem(m); 2700 return; 2701 } 2702 2703 /* 2704 * Parse TCP options and place in tcpopt. 2705 */ 2706 static void 2707 tcp_dooptions(to, cp, cnt, flags) 2708 struct tcpopt *to; 2709 u_char *cp; 2710 int cnt; 2711 int flags; 2712 { 2713 int opt, optlen; 2714 2715 to->to_flags = 0; 2716 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2717 opt = cp[0]; 2718 if (opt == TCPOPT_EOL) 2719 break; 2720 if (opt == TCPOPT_NOP) 2721 optlen = 1; 2722 else { 2723 if (cnt < 2) 2724 break; 2725 optlen = cp[1]; 2726 if (optlen < 2 || optlen > cnt) 2727 break; 2728 } 2729 switch (opt) { 2730 case TCPOPT_MAXSEG: 2731 if (optlen != TCPOLEN_MAXSEG) 2732 continue; 2733 if (!(flags & TO_SYN)) 2734 continue; 2735 to->to_flags |= TOF_MSS; 2736 bcopy((char *)cp + 2, 2737 (char *)&to->to_mss, sizeof(to->to_mss)); 2738 to->to_mss = ntohs(to->to_mss); 2739 break; 2740 case TCPOPT_WINDOW: 2741 if (optlen != TCPOLEN_WINDOW) 2742 continue; 2743 if (!(flags & TO_SYN)) 2744 continue; 2745 to->to_flags |= TOF_SCALE; 2746 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2747 break; 2748 case TCPOPT_TIMESTAMP: 2749 if (optlen != TCPOLEN_TIMESTAMP) 2750 continue; 2751 to->to_flags |= TOF_TS; 2752 bcopy((char *)cp + 2, 2753 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2754 to->to_tsval = ntohl(to->to_tsval); 2755 bcopy((char *)cp + 6, 2756 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2757 to->to_tsecr = ntohl(to->to_tsecr); 2758 break; 2759 #ifdef TCP_SIGNATURE 2760 /* 2761 * XXX In order to reply to a host which has set the 2762 * TCP_SIGNATURE option in its initial SYN, we have to 2763 * record the fact that the option was observed here 2764 * for the syncache code to perform the correct response. 2765 */ 2766 case TCPOPT_SIGNATURE: 2767 if (optlen != TCPOLEN_SIGNATURE) 2768 continue; 2769 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2770 break; 2771 #endif 2772 case TCPOPT_SACK_PERMITTED: 2773 if (optlen != TCPOLEN_SACK_PERMITTED) 2774 continue; 2775 if (!(flags & TO_SYN)) 2776 continue; 2777 if (!tcp_do_sack) 2778 continue; 2779 to->to_flags |= TOF_SACK; 2780 break; 2781 case TCPOPT_SACK: 2782 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2783 continue; 2784 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2785 to->to_sacks = cp + 2; 2786 tcpstat.tcps_sack_rcv_blocks++; 2787 break; 2788 default: 2789 continue; 2790 } 2791 } 2792 } 2793 2794 /* 2795 * Pull out of band byte out of a segment so 2796 * it doesn't appear in the user's data queue. 2797 * It is still reflected in the segment length for 2798 * sequencing purposes. 2799 */ 2800 static void 2801 tcp_pulloutofband(so, th, m, off) 2802 struct socket *so; 2803 struct tcphdr *th; 2804 register struct mbuf *m; 2805 int off; /* delayed to be droped hdrlen */ 2806 { 2807 int cnt = off + th->th_urp - 1; 2808 2809 while (cnt >= 0) { 2810 if (m->m_len > cnt) { 2811 char *cp = mtod(m, caddr_t) + cnt; 2812 struct tcpcb *tp = sototcpcb(so); 2813 2814 tp->t_iobc = *cp; 2815 tp->t_oobflags |= TCPOOB_HAVEDATA; 2816 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2817 m->m_len--; 2818 if (m->m_flags & M_PKTHDR) 2819 m->m_pkthdr.len--; 2820 return; 2821 } 2822 cnt -= m->m_len; 2823 m = m->m_next; 2824 if (m == 0) 2825 break; 2826 } 2827 panic("tcp_pulloutofband"); 2828 } 2829 2830 /* 2831 * Collect new round-trip time estimate 2832 * and update averages and current timeout. 2833 */ 2834 static void 2835 tcp_xmit_timer(tp, rtt) 2836 register struct tcpcb *tp; 2837 int rtt; 2838 { 2839 register int delta; 2840 2841 INP_LOCK_ASSERT(tp->t_inpcb); 2842 2843 tcpstat.tcps_rttupdated++; 2844 tp->t_rttupdated++; 2845 if (tp->t_srtt != 0) { 2846 /* 2847 * srtt is stored as fixed point with 5 bits after the 2848 * binary point (i.e., scaled by 8). The following magic 2849 * is equivalent to the smoothing algorithm in rfc793 with 2850 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2851 * point). Adjust rtt to origin 0. 2852 */ 2853 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2854 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2855 2856 if ((tp->t_srtt += delta) <= 0) 2857 tp->t_srtt = 1; 2858 2859 /* 2860 * We accumulate a smoothed rtt variance (actually, a 2861 * smoothed mean difference), then set the retransmit 2862 * timer to smoothed rtt + 4 times the smoothed variance. 2863 * rttvar is stored as fixed point with 4 bits after the 2864 * binary point (scaled by 16). The following is 2865 * equivalent to rfc793 smoothing with an alpha of .75 2866 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2867 * rfc793's wired-in beta. 2868 */ 2869 if (delta < 0) 2870 delta = -delta; 2871 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2872 if ((tp->t_rttvar += delta) <= 0) 2873 tp->t_rttvar = 1; 2874 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2875 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2876 } else { 2877 /* 2878 * No rtt measurement yet - use the unsmoothed rtt. 2879 * Set the variance to half the rtt (so our first 2880 * retransmit happens at 3*rtt). 2881 */ 2882 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2883 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2884 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2885 } 2886 tp->t_rtttime = 0; 2887 tp->t_rxtshift = 0; 2888 2889 /* 2890 * the retransmit should happen at rtt + 4 * rttvar. 2891 * Because of the way we do the smoothing, srtt and rttvar 2892 * will each average +1/2 tick of bias. When we compute 2893 * the retransmit timer, we want 1/2 tick of rounding and 2894 * 1 extra tick because of +-1/2 tick uncertainty in the 2895 * firing of the timer. The bias will give us exactly the 2896 * 1.5 tick we need. But, because the bias is 2897 * statistical, we have to test that we don't drop below 2898 * the minimum feasible timer (which is 2 ticks). 2899 */ 2900 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2901 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2902 2903 /* 2904 * We received an ack for a packet that wasn't retransmitted; 2905 * it is probably safe to discard any error indications we've 2906 * received recently. This isn't quite right, but close enough 2907 * for now (a route might have failed after we sent a segment, 2908 * and the return path might not be symmetrical). 2909 */ 2910 tp->t_softerror = 0; 2911 } 2912 2913 /* 2914 * Determine a reasonable value for maxseg size. 2915 * If the route is known, check route for mtu. 2916 * If none, use an mss that can be handled on the outgoing 2917 * interface without forcing IP to fragment; if bigger than 2918 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2919 * to utilize large mbufs. If no route is found, route has no mtu, 2920 * or the destination isn't local, use a default, hopefully conservative 2921 * size (usually 512 or the default IP max size, but no more than the mtu 2922 * of the interface), as we can't discover anything about intervening 2923 * gateways or networks. We also initialize the congestion/slow start 2924 * window to be a single segment if the destination isn't local. 2925 * While looking at the routing entry, we also initialize other path-dependent 2926 * parameters from pre-set or cached values in the routing entry. 2927 * 2928 * Also take into account the space needed for options that we 2929 * send regularly. Make maxseg shorter by that amount to assure 2930 * that we can send maxseg amount of data even when the options 2931 * are present. Store the upper limit of the length of options plus 2932 * data in maxopd. 2933 * 2934 * 2935 * In case of T/TCP, we call this routine during implicit connection 2936 * setup as well (offer = -1), to initialize maxseg from the cached 2937 * MSS of our peer. 2938 * 2939 * NOTE that this routine is only called when we process an incoming 2940 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2941 */ 2942 void 2943 tcp_mss(tp, offer) 2944 struct tcpcb *tp; 2945 int offer; 2946 { 2947 int rtt, mss; 2948 u_long bufsize; 2949 u_long maxmtu; 2950 struct inpcb *inp = tp->t_inpcb; 2951 struct socket *so; 2952 struct hc_metrics_lite metrics; 2953 int origoffer = offer; 2954 int mtuflags = 0; 2955 #ifdef INET6 2956 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2957 size_t min_protoh = isipv6 ? 2958 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2959 sizeof (struct tcpiphdr); 2960 #else 2961 const size_t min_protoh = sizeof(struct tcpiphdr); 2962 #endif 2963 2964 /* initialize */ 2965 #ifdef INET6 2966 if (isipv6) { 2967 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags); 2968 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2969 } else 2970 #endif 2971 { 2972 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags); 2973 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2974 } 2975 so = inp->inp_socket; 2976 2977 /* 2978 * no route to sender, stay with default mss and return 2979 */ 2980 if (maxmtu == 0) 2981 return; 2982 2983 /* what have we got? */ 2984 switch (offer) { 2985 case 0: 2986 /* 2987 * Offer == 0 means that there was no MSS on the SYN 2988 * segment, in this case we use tcp_mssdflt. 2989 */ 2990 offer = 2991 #ifdef INET6 2992 isipv6 ? tcp_v6mssdflt : 2993 #endif 2994 tcp_mssdflt; 2995 break; 2996 2997 case -1: 2998 /* 2999 * Offer == -1 means that we didn't receive SYN yet. 3000 */ 3001 /* FALLTHROUGH */ 3002 3003 default: 3004 /* 3005 * Prevent DoS attack with too small MSS. Round up 3006 * to at least minmss. 3007 */ 3008 offer = max(offer, tcp_minmss); 3009 /* 3010 * Sanity check: make sure that maxopd will be large 3011 * enough to allow some data on segments even if the 3012 * all the option space is used (40bytes). Otherwise 3013 * funny things may happen in tcp_output. 3014 */ 3015 offer = max(offer, 64); 3016 } 3017 3018 /* 3019 * rmx information is now retrieved from tcp_hostcache 3020 */ 3021 tcp_hc_get(&inp->inp_inc, &metrics); 3022 3023 /* 3024 * if there's a discovered mtu int tcp hostcache, use it 3025 * else, use the link mtu. 3026 */ 3027 if (metrics.rmx_mtu) 3028 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3029 else { 3030 #ifdef INET6 3031 if (isipv6) { 3032 mss = maxmtu - min_protoh; 3033 if (!path_mtu_discovery && 3034 !in6_localaddr(&inp->in6p_faddr)) 3035 mss = min(mss, tcp_v6mssdflt); 3036 } else 3037 #endif 3038 { 3039 mss = maxmtu - min_protoh; 3040 if (!path_mtu_discovery && 3041 !in_localaddr(inp->inp_faddr)) 3042 mss = min(mss, tcp_mssdflt); 3043 } 3044 } 3045 mss = min(mss, offer); 3046 3047 /* 3048 * maxopd stores the maximum length of data AND options 3049 * in a segment; maxseg is the amount of data in a normal 3050 * segment. We need to store this value (maxopd) apart 3051 * from maxseg, because now every segment carries options 3052 * and thus we normally have somewhat less data in segments. 3053 */ 3054 tp->t_maxopd = mss; 3055 3056 /* 3057 * origoffer==-1 indicates, that no segments were received yet. 3058 * In this case we just guess. 3059 */ 3060 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3061 (origoffer == -1 || 3062 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3063 mss -= TCPOLEN_TSTAMP_APPA; 3064 tp->t_maxseg = mss; 3065 3066 #if (MCLBYTES & (MCLBYTES - 1)) == 0 3067 if (mss > MCLBYTES) 3068 mss &= ~(MCLBYTES-1); 3069 #else 3070 if (mss > MCLBYTES) 3071 mss = mss / MCLBYTES * MCLBYTES; 3072 #endif 3073 tp->t_maxseg = mss; 3074 3075 /* 3076 * If there's a pipesize, change the socket buffer to that size, 3077 * don't change if sb_hiwat is different than default (then it 3078 * has been changed on purpose with setsockopt). 3079 * Make the socket buffers an integral number of mss units; 3080 * if the mss is larger than the socket buffer, decrease the mss. 3081 */ 3082 SOCKBUF_LOCK(&so->so_snd); 3083 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 3084 bufsize = metrics.rmx_sendpipe; 3085 else 3086 bufsize = so->so_snd.sb_hiwat; 3087 if (bufsize < mss) 3088 mss = bufsize; 3089 else { 3090 bufsize = roundup(bufsize, mss); 3091 if (bufsize > sb_max) 3092 bufsize = sb_max; 3093 if (bufsize > so->so_snd.sb_hiwat) 3094 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3095 } 3096 SOCKBUF_UNLOCK(&so->so_snd); 3097 tp->t_maxseg = mss; 3098 3099 SOCKBUF_LOCK(&so->so_rcv); 3100 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3101 bufsize = metrics.rmx_recvpipe; 3102 else 3103 bufsize = so->so_rcv.sb_hiwat; 3104 if (bufsize > mss) { 3105 bufsize = roundup(bufsize, mss); 3106 if (bufsize > sb_max) 3107 bufsize = sb_max; 3108 if (bufsize > so->so_rcv.sb_hiwat) 3109 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3110 } 3111 SOCKBUF_UNLOCK(&so->so_rcv); 3112 /* 3113 * While we're here, check the others too 3114 */ 3115 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 3116 tp->t_srtt = rtt; 3117 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3118 tcpstat.tcps_usedrtt++; 3119 if (metrics.rmx_rttvar) { 3120 tp->t_rttvar = metrics.rmx_rttvar; 3121 tcpstat.tcps_usedrttvar++; 3122 } else { 3123 /* default variation is +- 1 rtt */ 3124 tp->t_rttvar = 3125 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3126 } 3127 TCPT_RANGESET(tp->t_rxtcur, 3128 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3129 tp->t_rttmin, TCPTV_REXMTMAX); 3130 } 3131 if (metrics.rmx_ssthresh) { 3132 /* 3133 * There's some sort of gateway or interface 3134 * buffer limit on the path. Use this to set 3135 * the slow start threshhold, but set the 3136 * threshold to no less than 2*mss. 3137 */ 3138 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3139 tcpstat.tcps_usedssthresh++; 3140 } 3141 if (metrics.rmx_bandwidth) 3142 tp->snd_bandwidth = metrics.rmx_bandwidth; 3143 3144 /* 3145 * Set the slow-start flight size depending on whether this 3146 * is a local network or not. 3147 * 3148 * Extend this so we cache the cwnd too and retrieve it here. 3149 * Make cwnd even bigger than RFC3390 suggests but only if we 3150 * have previous experience with the remote host. Be careful 3151 * not make cwnd bigger than remote receive window or our own 3152 * send socket buffer. Maybe put some additional upper bound 3153 * on the retrieved cwnd. Should do incremental updates to 3154 * hostcache when cwnd collapses so next connection doesn't 3155 * overloads the path again. 3156 * 3157 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3158 * We currently check only in syncache_socket for that. 3159 */ 3160 #define TCP_METRICS_CWND 3161 #ifdef TCP_METRICS_CWND 3162 if (metrics.rmx_cwnd) 3163 tp->snd_cwnd = max(mss, 3164 min(metrics.rmx_cwnd / 2, 3165 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3166 else 3167 #endif 3168 if (tcp_do_rfc3390) 3169 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3170 #ifdef INET6 3171 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3172 (!isipv6 && in_localaddr(inp->inp_faddr))) 3173 #else 3174 else if (in_localaddr(inp->inp_faddr)) 3175 #endif 3176 tp->snd_cwnd = mss * ss_fltsz_local; 3177 else 3178 tp->snd_cwnd = mss * ss_fltsz; 3179 3180 /* Check the interface for TSO capabilities. */ 3181 if (mtuflags & CSUM_TSO) 3182 tp->t_flags |= TF_TSO; 3183 } 3184 3185 /* 3186 * Determine the MSS option to send on an outgoing SYN. 3187 */ 3188 int 3189 tcp_mssopt(inc) 3190 struct in_conninfo *inc; 3191 { 3192 int mss = 0; 3193 u_long maxmtu = 0; 3194 u_long thcmtu = 0; 3195 size_t min_protoh; 3196 #ifdef INET6 3197 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3198 #endif 3199 3200 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3201 3202 #ifdef INET6 3203 if (isipv6) { 3204 mss = tcp_v6mssdflt; 3205 maxmtu = tcp_maxmtu6(inc, NULL); 3206 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3207 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3208 } else 3209 #endif 3210 { 3211 mss = tcp_mssdflt; 3212 maxmtu = tcp_maxmtu(inc, NULL); 3213 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3214 min_protoh = sizeof(struct tcpiphdr); 3215 } 3216 if (maxmtu && thcmtu) 3217 mss = min(maxmtu, thcmtu) - min_protoh; 3218 else if (maxmtu || thcmtu) 3219 mss = max(maxmtu, thcmtu) - min_protoh; 3220 3221 return (mss); 3222 } 3223 3224 3225 /* 3226 * On a partial ack arrives, force the retransmission of the 3227 * next unacknowledged segment. Do not clear tp->t_dupacks. 3228 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3229 * be started again. 3230 */ 3231 static void 3232 tcp_newreno_partial_ack(tp, th) 3233 struct tcpcb *tp; 3234 struct tcphdr *th; 3235 { 3236 tcp_seq onxt = tp->snd_nxt; 3237 u_long ocwnd = tp->snd_cwnd; 3238 3239 callout_stop(tp->tt_rexmt); 3240 tp->t_rtttime = 0; 3241 tp->snd_nxt = th->th_ack; 3242 /* 3243 * Set snd_cwnd to one segment beyond acknowledged offset. 3244 * (tp->snd_una has not yet been updated when this function is called.) 3245 */ 3246 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3247 tp->t_flags |= TF_ACKNOW; 3248 (void) tcp_output(tp); 3249 tp->snd_cwnd = ocwnd; 3250 if (SEQ_GT(onxt, tp->snd_nxt)) 3251 tp->snd_nxt = onxt; 3252 /* 3253 * Partial window deflation. Relies on fact that tp->snd_una 3254 * not updated yet. 3255 */ 3256 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3257 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3258 else 3259 tp->snd_cwnd = 0; 3260 tp->snd_cwnd += tp->t_maxseg; 3261 } 3262 3263 /* 3264 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3265 * looking for a pcb in the listen state. Returns 0 otherwise. 3266 */ 3267 static int 3268 tcp_timewait(inp, to, th, m, tlen) 3269 struct inpcb *inp; 3270 struct tcpopt *to; 3271 struct tcphdr *th; 3272 struct mbuf *m; 3273 int tlen; 3274 { 3275 struct tcptw *tw; 3276 int thflags; 3277 tcp_seq seq; 3278 #ifdef INET6 3279 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3280 #else 3281 const int isipv6 = 0; 3282 #endif 3283 3284 /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */ 3285 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3286 INP_LOCK_ASSERT(inp); 3287 3288 /* 3289 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 3290 * still present. This is undesirable, but temporarily necessary 3291 * until we work out how to handle inpcb's who's timewait state has 3292 * been removed. 3293 */ 3294 tw = intotw(inp); 3295 if (tw == NULL) 3296 goto drop; 3297 3298 thflags = th->th_flags; 3299 3300 /* 3301 * NOTE: for FIN_WAIT_2 (to be added later), 3302 * must validate sequence number before accepting RST 3303 */ 3304 3305 /* 3306 * If the segment contains RST: 3307 * Drop the segment - see Stevens, vol. 2, p. 964 and 3308 * RFC 1337. 3309 */ 3310 if (thflags & TH_RST) 3311 goto drop; 3312 3313 #if 0 3314 /* PAWS not needed at the moment */ 3315 /* 3316 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3317 * and it's less than ts_recent, drop it. 3318 */ 3319 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3320 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3321 if ((thflags & TH_ACK) == 0) 3322 goto drop; 3323 goto ack; 3324 } 3325 /* 3326 * ts_recent is never updated because we never accept new segments. 3327 */ 3328 #endif 3329 3330 /* 3331 * If a new connection request is received 3332 * while in TIME_WAIT, drop the old connection 3333 * and start over if the sequence numbers 3334 * are above the previous ones. 3335 */ 3336 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3337 tcp_twclose(tw, 0); 3338 return (1); 3339 } 3340 3341 /* 3342 * Drop the the segment if it does not contain an ACK. 3343 */ 3344 if ((thflags & TH_ACK) == 0) 3345 goto drop; 3346 3347 /* 3348 * Reset the 2MSL timer if this is a duplicate FIN. 3349 */ 3350 if (thflags & TH_FIN) { 3351 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3352 if (seq + 1 == tw->rcv_nxt) 3353 tcp_timer_2msl_reset(tw, 1); 3354 } 3355 3356 /* 3357 * Acknowledge the segment if it has data or is not a duplicate ACK. 3358 */ 3359 if (thflags != TH_ACK || tlen != 0 || 3360 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3361 tcp_twrespond(tw, TH_ACK); 3362 goto drop; 3363 3364 /* 3365 * Generate a RST, dropping incoming segment. 3366 * Make ACK acceptable to originator of segment. 3367 * Don't bother to respond if destination was broadcast/multicast. 3368 */ 3369 if (m->m_flags & (M_BCAST|M_MCAST)) 3370 goto drop; 3371 if (isipv6) { 3372 struct ip6_hdr *ip6; 3373 3374 /* IPv6 anycast check is done at tcp6_input() */ 3375 ip6 = mtod(m, struct ip6_hdr *); 3376 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3377 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3378 goto drop; 3379 } else { 3380 struct ip *ip; 3381 3382 ip = mtod(m, struct ip *); 3383 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3384 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3385 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3386 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3387 goto drop; 3388 } 3389 if (thflags & TH_ACK) { 3390 tcp_respond(NULL, 3391 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3392 } else { 3393 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3394 tcp_respond(NULL, 3395 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3396 } 3397 INP_UNLOCK(inp); 3398 return (0); 3399 3400 drop: 3401 INP_UNLOCK(inp); 3402 m_freem(m); 3403 return (0); 3404 } 3405