tcp_input.c revision 169405
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: head/sys/netinet/tcp_input.c 169405 2007-05-09 06:09:40Z maxim $ 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 40#include <sys/param.h> 41#include <sys/kernel.h> 42#include <sys/malloc.h> 43#include <sys/mbuf.h> 44#include <sys/proc.h> /* for proc0 declaration */ 45#include <sys/protosw.h> 46#include <sys/signalvar.h> 47#include <sys/socket.h> 48#include <sys/socketvar.h> 49#include <sys/sysctl.h> 50#include <sys/syslog.h> 51#include <sys/systm.h> 52 53#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 54 55#include <vm/uma.h> 56 57#include <net/if.h> 58#include <net/route.h> 59 60#include <netinet/in.h> 61#include <netinet/in_pcb.h> 62#include <netinet/in_systm.h> 63#include <netinet/in_var.h> 64#include <netinet/ip.h> 65#include <netinet/ip_icmp.h> /* required for icmp_var.h */ 66#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 67#include <netinet/ip_var.h> 68#include <netinet/ip_options.h> 69#include <netinet/ip6.h> 70#include <netinet/icmp6.h> 71#include <netinet6/in6_pcb.h> 72#include <netinet6/ip6_var.h> 73#include <netinet6/nd6.h> 74#include <netinet/tcp.h> 75#include <netinet/tcp_fsm.h> 76#include <netinet/tcp_seq.h> 77#include <netinet/tcp_timer.h> 78#include <netinet/tcp_var.h> 79#include <netinet6/tcp6_var.h> 80#include <netinet/tcpip.h> 81#ifdef TCPDEBUG 82#include <netinet/tcp_debug.h> 83#endif /* TCPDEBUG */ 84 85#ifdef FAST_IPSEC 86#include <netipsec/ipsec.h> 87#include <netipsec/ipsec6.h> 88#endif /*FAST_IPSEC*/ 89 90#ifdef IPSEC 91#include <netinet6/ipsec.h> 92#include <netinet6/ipsec6.h> 93#include <netkey/key.h> 94#endif /*IPSEC*/ 95 96#include <machine/in_cksum.h> 97 98#include <security/mac/mac_framework.h> 99 100static const int tcprexmtthresh = 3; 101 102struct tcpstat tcpstat; 103SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 104 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 105 106static int tcp_log_in_vain = 0; 107SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 108 &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports"); 109 110static int blackhole = 0; 111SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 112 &blackhole, 0, "Do not send RST on segments to closed ports"); 113 114int tcp_delack_enabled = 1; 115SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 116 &tcp_delack_enabled, 0, 117 "Delay ACK to try and piggyback it onto a data packet"); 118 119static int drop_synfin = 0; 120SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 121 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 122 123static int tcp_do_rfc3042 = 1; 124SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 125 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 126 127static int tcp_do_rfc3390 = 1; 128SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 129 &tcp_do_rfc3390, 0, 130 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 131 132static int tcp_insecure_rst = 0; 133SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 134 &tcp_insecure_rst, 0, 135 "Follow the old (insecure) criteria for accepting RST packets"); 136 137SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 138 "TCP Segment Reassembly Queue"); 139 140static int tcp_reass_maxseg = 0; 141SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 142 &tcp_reass_maxseg, 0, 143 "Global maximum number of TCP Segments in Reassembly Queue"); 144 145int tcp_reass_qsize = 0; 146SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 147 &tcp_reass_qsize, 0, 148 "Global number of TCP Segments currently in Reassembly Queue"); 149 150static int tcp_reass_maxqlen = 48; 151SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 152 &tcp_reass_maxqlen, 0, 153 "Maximum number of TCP Segments per individual Reassembly Queue"); 154 155static int tcp_reass_overflows = 0; 156SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 157 &tcp_reass_overflows, 0, 158 "Global number of TCP Segment Reassembly Queue Overflows"); 159 160int tcp_do_autorcvbuf = 1; 161SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 162 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing"); 163 164int tcp_autorcvbuf_inc = 16*1024; 165SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 166 &tcp_autorcvbuf_inc, 0, 167 "Incrementor step size of automatic receive buffer"); 168 169int tcp_autorcvbuf_max = 256*1024; 170SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 171 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer"); 172 173struct inpcbhead tcb; 174#define tcb6 tcb /* for KAME src sync over BSD*'s */ 175struct inpcbinfo tcbinfo; 176 177static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 178static void tcp_do_segment(struct mbuf *, struct tcphdr *, 179 struct socket *, struct tcpcb *, int, int); 180static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 181 struct tcpcb *, int, int); 182static void tcp_pulloutofband(struct socket *, 183 struct tcphdr *, struct mbuf *, int); 184static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 185 struct mbuf *); 186static void tcp_xmit_timer(struct tcpcb *, int); 187static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 188static int tcp_timewait(struct inpcb *, struct tcpopt *, 189 struct tcphdr *, struct mbuf *, int); 190 191/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 192#ifdef INET6 193#define ND6_HINT(tp) \ 194do { \ 195 if ((tp) && (tp)->t_inpcb && \ 196 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 197 nd6_nud_hint(NULL, NULL, 0); \ 198} while (0) 199#else 200#define ND6_HINT(tp) 201#endif 202 203/* 204 * Indicate whether this ack should be delayed. We can delay the ack if 205 * - there is no delayed ack timer in progress and 206 * - our last ack wasn't a 0-sized window. We never want to delay 207 * the ack that opens up a 0-sized window and 208 * - delayed acks are enabled or 209 * - this is a half-synchronized T/TCP connection. 210 */ 211#define DELAY_ACK(tp) \ 212 ((!tcp_timer_active(tp, TT_DELACK) && \ 213 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 214 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 215 216/* Initialize TCP reassembly queue */ 217static void 218tcp_reass_zone_change(void *tag) 219{ 220 221 tcp_reass_maxseg = nmbclusters / 16; 222 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 223} 224 225uma_zone_t tcp_reass_zone; 226void 227tcp_reass_init() 228{ 229 tcp_reass_maxseg = nmbclusters / 16; 230 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 231 &tcp_reass_maxseg); 232 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 233 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 234 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 235 EVENTHANDLER_REGISTER(nmbclusters_change, 236 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY); 237} 238 239static int 240tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m) 241{ 242 struct tseg_qent *q; 243 struct tseg_qent *p = NULL; 244 struct tseg_qent *nq; 245 struct tseg_qent *te = NULL; 246 struct socket *so = tp->t_inpcb->inp_socket; 247 int flags; 248 249 INP_LOCK_ASSERT(tp->t_inpcb); 250 251 /* 252 * XXX: tcp_reass() is rather inefficient with its data structures 253 * and should be rewritten (see NetBSD for optimizations). While 254 * doing that it should move to its own file tcp_reass.c. 255 */ 256 257 /* 258 * Call with th==NULL after become established to 259 * force pre-ESTABLISHED data up to user socket. 260 */ 261 if (th == NULL) 262 goto present; 263 264 /* 265 * Limit the number of segments in the reassembly queue to prevent 266 * holding on to too many segments (and thus running out of mbufs). 267 * Make sure to let the missing segment through which caused this 268 * queue. Always keep one global queue entry spare to be able to 269 * process the missing segment. 270 */ 271 if (th->th_seq != tp->rcv_nxt && 272 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 273 tp->t_segqlen >= tcp_reass_maxqlen)) { 274 tcp_reass_overflows++; 275 tcpstat.tcps_rcvmemdrop++; 276 m_freem(m); 277 *tlenp = 0; 278 return (0); 279 } 280 281 /* 282 * Allocate a new queue entry. If we can't, or hit the zone limit 283 * just drop the pkt. 284 */ 285 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 286 if (te == NULL) { 287 tcpstat.tcps_rcvmemdrop++; 288 m_freem(m); 289 *tlenp = 0; 290 return (0); 291 } 292 tp->t_segqlen++; 293 tcp_reass_qsize++; 294 295 /* 296 * Find a segment which begins after this one does. 297 */ 298 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 299 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 300 break; 301 p = q; 302 } 303 304 /* 305 * If there is a preceding segment, it may provide some of 306 * our data already. If so, drop the data from the incoming 307 * segment. If it provides all of our data, drop us. 308 */ 309 if (p != NULL) { 310 int i; 311 /* conversion to int (in i) handles seq wraparound */ 312 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 313 if (i > 0) { 314 if (i >= *tlenp) { 315 tcpstat.tcps_rcvduppack++; 316 tcpstat.tcps_rcvdupbyte += *tlenp; 317 m_freem(m); 318 uma_zfree(tcp_reass_zone, te); 319 tp->t_segqlen--; 320 tcp_reass_qsize--; 321 /* 322 * Try to present any queued data 323 * at the left window edge to the user. 324 * This is needed after the 3-WHS 325 * completes. 326 */ 327 goto present; /* ??? */ 328 } 329 m_adj(m, i); 330 *tlenp -= i; 331 th->th_seq += i; 332 } 333 } 334 tcpstat.tcps_rcvoopack++; 335 tcpstat.tcps_rcvoobyte += *tlenp; 336 337 /* 338 * While we overlap succeeding segments trim them or, 339 * if they are completely covered, dequeue them. 340 */ 341 while (q) { 342 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 343 if (i <= 0) 344 break; 345 if (i < q->tqe_len) { 346 q->tqe_th->th_seq += i; 347 q->tqe_len -= i; 348 m_adj(q->tqe_m, i); 349 break; 350 } 351 352 nq = LIST_NEXT(q, tqe_q); 353 LIST_REMOVE(q, tqe_q); 354 m_freem(q->tqe_m); 355 uma_zfree(tcp_reass_zone, q); 356 tp->t_segqlen--; 357 tcp_reass_qsize--; 358 q = nq; 359 } 360 361 /* Insert the new segment queue entry into place. */ 362 te->tqe_m = m; 363 te->tqe_th = th; 364 te->tqe_len = *tlenp; 365 366 if (p == NULL) { 367 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 368 } else { 369 LIST_INSERT_AFTER(p, te, tqe_q); 370 } 371 372present: 373 /* 374 * Present data to user, advancing rcv_nxt through 375 * completed sequence space. 376 */ 377 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 378 return (0); 379 q = LIST_FIRST(&tp->t_segq); 380 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 381 return (0); 382 SOCKBUF_LOCK(&so->so_rcv); 383 do { 384 tp->rcv_nxt += q->tqe_len; 385 flags = q->tqe_th->th_flags & TH_FIN; 386 nq = LIST_NEXT(q, tqe_q); 387 LIST_REMOVE(q, tqe_q); 388 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 389 m_freem(q->tqe_m); 390 else 391 sbappendstream_locked(&so->so_rcv, q->tqe_m); 392 uma_zfree(tcp_reass_zone, q); 393 tp->t_segqlen--; 394 tcp_reass_qsize--; 395 q = nq; 396 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 397 ND6_HINT(tp); 398 sorwakeup_locked(so); 399 return (flags); 400} 401 402/* 403 * TCP input routine, follows pages 65-76 of the 404 * protocol specification dated September, 1981 very closely. 405 */ 406#ifdef INET6 407int 408tcp6_input(struct mbuf **mp, int *offp, int proto) 409{ 410 struct mbuf *m = *mp; 411 struct in6_ifaddr *ia6; 412 413 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 414 415 /* 416 * draft-itojun-ipv6-tcp-to-anycast 417 * better place to put this in? 418 */ 419 ia6 = ip6_getdstifaddr(m); 420 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 421 struct ip6_hdr *ip6; 422 423 ip6 = mtod(m, struct ip6_hdr *); 424 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 425 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 426 return IPPROTO_DONE; 427 } 428 429 tcp_input(m, *offp); 430 return IPPROTO_DONE; 431} 432#endif 433 434void 435tcp_input(struct mbuf *m, int off0) 436{ 437 struct tcphdr *th; 438 struct ip *ip = NULL; 439 struct ipovly *ipov; 440 struct inpcb *inp = NULL; 441 struct tcpcb *tp = NULL; 442 struct socket *so = NULL; 443 u_char *optp = NULL; 444 int optlen = 0; 445 int len, tlen, off; 446 int drop_hdrlen; 447 int thflags; 448 int rstreason = 0; /* For badport_bandlim accounting purposes */ 449#ifdef IPFIREWALL_FORWARD 450 struct m_tag *fwd_tag; 451#endif 452#ifdef INET6 453 struct ip6_hdr *ip6 = NULL; 454 int isipv6; 455 char ip6buf[INET6_ADDRSTRLEN]; 456#else 457 const int isipv6 = 0; 458#endif 459 struct tcpopt to; /* options in this segment */ 460 461#ifdef TCPDEBUG 462 /* 463 * The size of tcp_saveipgen must be the size of the max ip header, 464 * now IPv6. 465 */ 466 u_char tcp_saveipgen[IP6_HDR_LEN]; 467 struct tcphdr tcp_savetcp; 468 short ostate = 0; 469#endif 470 471#ifdef INET6 472 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 473#endif 474 475 to.to_flags = 0; 476 tcpstat.tcps_rcvtotal++; 477 478 if (isipv6) { 479#ifdef INET6 480 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 481 ip6 = mtod(m, struct ip6_hdr *); 482 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 483 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 484 tcpstat.tcps_rcvbadsum++; 485 goto drop; 486 } 487 th = (struct tcphdr *)((caddr_t)ip6 + off0); 488 489 /* 490 * Be proactive about unspecified IPv6 address in source. 491 * As we use all-zero to indicate unbounded/unconnected pcb, 492 * unspecified IPv6 address can be used to confuse us. 493 * 494 * Note that packets with unspecified IPv6 destination is 495 * already dropped in ip6_input. 496 */ 497 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 498 /* XXX stat */ 499 goto drop; 500 } 501#else 502 th = NULL; /* XXX: avoid compiler warning */ 503#endif 504 } else { 505 /* 506 * Get IP and TCP header together in first mbuf. 507 * Note: IP leaves IP header in first mbuf. 508 */ 509 if (off0 > sizeof (struct ip)) { 510 ip_stripoptions(m, (struct mbuf *)0); 511 off0 = sizeof(struct ip); 512 } 513 if (m->m_len < sizeof (struct tcpiphdr)) { 514 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 515 == NULL) { 516 tcpstat.tcps_rcvshort++; 517 return; 518 } 519 } 520 ip = mtod(m, struct ip *); 521 ipov = (struct ipovly *)ip; 522 th = (struct tcphdr *)((caddr_t)ip + off0); 523 tlen = ip->ip_len; 524 525 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 526 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 527 th->th_sum = m->m_pkthdr.csum_data; 528 else 529 th->th_sum = in_pseudo(ip->ip_src.s_addr, 530 ip->ip_dst.s_addr, 531 htonl(m->m_pkthdr.csum_data + 532 ip->ip_len + 533 IPPROTO_TCP)); 534 th->th_sum ^= 0xffff; 535#ifdef TCPDEBUG 536 ipov->ih_len = (u_short)tlen; 537 ipov->ih_len = htons(ipov->ih_len); 538#endif 539 } else { 540 /* 541 * Checksum extended TCP header and data. 542 */ 543 len = sizeof (struct ip) + tlen; 544 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 545 ipov->ih_len = (u_short)tlen; 546 ipov->ih_len = htons(ipov->ih_len); 547 th->th_sum = in_cksum(m, len); 548 } 549 if (th->th_sum) { 550 tcpstat.tcps_rcvbadsum++; 551 goto drop; 552 } 553 /* Re-initialization for later version check */ 554 ip->ip_v = IPVERSION; 555 } 556 557 /* 558 * Check that TCP offset makes sense, 559 * pull out TCP options and adjust length. XXX 560 */ 561 off = th->th_off << 2; 562 if (off < sizeof (struct tcphdr) || off > tlen) { 563 tcpstat.tcps_rcvbadoff++; 564 goto drop; 565 } 566 tlen -= off; /* tlen is used instead of ti->ti_len */ 567 if (off > sizeof (struct tcphdr)) { 568 if (isipv6) { 569#ifdef INET6 570 IP6_EXTHDR_CHECK(m, off0, off, ); 571 ip6 = mtod(m, struct ip6_hdr *); 572 th = (struct tcphdr *)((caddr_t)ip6 + off0); 573#endif 574 } else { 575 if (m->m_len < sizeof(struct ip) + off) { 576 if ((m = m_pullup(m, sizeof (struct ip) + off)) 577 == NULL) { 578 tcpstat.tcps_rcvshort++; 579 return; 580 } 581 ip = mtod(m, struct ip *); 582 ipov = (struct ipovly *)ip; 583 th = (struct tcphdr *)((caddr_t)ip + off0); 584 } 585 } 586 optlen = off - sizeof (struct tcphdr); 587 optp = (u_char *)(th + 1); 588 } 589 thflags = th->th_flags; 590 591 /* 592 * If the drop_synfin option is enabled, drop all packets with 593 * both the SYN and FIN bits set. This prevents e.g. nmap from 594 * identifying the TCP/IP stack. 595 * 596 * This is a violation of the TCP specification. 597 */ 598 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 599 goto drop; 600 601 /* 602 * Convert TCP protocol specific fields to host format. 603 */ 604 th->th_seq = ntohl(th->th_seq); 605 th->th_ack = ntohl(th->th_ack); 606 th->th_win = ntohs(th->th_win); 607 th->th_urp = ntohs(th->th_urp); 608 609 /* 610 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 611 */ 612 drop_hdrlen = off0 + off; 613 614 /* 615 * Locate pcb for segment. 616 */ 617 INP_INFO_WLOCK(&tcbinfo); 618findpcb: 619 INP_INFO_WLOCK_ASSERT(&tcbinfo); 620#ifdef IPFIREWALL_FORWARD 621 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 622 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 623 624 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 625 struct sockaddr_in *next_hop; 626 627 next_hop = (struct sockaddr_in *)(fwd_tag+1); 628 /* 629 * Transparently forwarded. Pretend to be the destination. 630 * already got one like this? 631 */ 632 inp = in_pcblookup_hash(&tcbinfo, 633 ip->ip_src, th->th_sport, 634 ip->ip_dst, th->th_dport, 635 0, m->m_pkthdr.rcvif); 636 if (!inp) { 637 /* It's new. Try to find the ambushing socket. */ 638 inp = in_pcblookup_hash(&tcbinfo, 639 ip->ip_src, th->th_sport, 640 next_hop->sin_addr, 641 next_hop->sin_port ? 642 ntohs(next_hop->sin_port) : 643 th->th_dport, 644 INPLOOKUP_WILDCARD, 645 m->m_pkthdr.rcvif); 646 } 647 /* Remove the tag from the packet. We don't need it anymore. */ 648 m_tag_delete(m, fwd_tag); 649 } else 650#endif /* IPFIREWALL_FORWARD */ 651 { 652 if (isipv6) { 653#ifdef INET6 654 inp = in6_pcblookup_hash(&tcbinfo, 655 &ip6->ip6_src, th->th_sport, 656 &ip6->ip6_dst, th->th_dport, 657 INPLOOKUP_WILDCARD, 658 m->m_pkthdr.rcvif); 659#endif 660 } else 661 inp = in_pcblookup_hash(&tcbinfo, 662 ip->ip_src, th->th_sport, 663 ip->ip_dst, th->th_dport, 664 INPLOOKUP_WILDCARD, 665 m->m_pkthdr.rcvif); 666 } 667 668#if defined(IPSEC) || defined(FAST_IPSEC) 669#ifdef INET6 670 if (isipv6 && inp != NULL && ipsec6_in_reject(m, inp)) { 671#ifdef IPSEC 672 ipsec6stat.in_polvio++; 673#endif 674 goto dropunlock; 675 } else 676#endif /* INET6 */ 677 if (inp != NULL && ipsec4_in_reject(m, inp)) { 678#ifdef IPSEC 679 ipsecstat.in_polvio++; 680#endif 681 goto dropunlock; 682 } 683#endif /*IPSEC || FAST_IPSEC*/ 684 685 /* 686 * If the INPCB does not exist then all data in the incoming 687 * segment is discarded and an appropriate RST is sent back. 688 */ 689 if (inp == NULL) { 690 /* 691 * Log communication attempts to ports that are not 692 * in use. 693 */ 694 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 695 tcp_log_in_vain == 2) { 696#ifndef INET6 697 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 698#else 699 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 700 if (isipv6) { 701 strcpy(dbuf, "["); 702 strcat(dbuf, 703 ip6_sprintf(ip6buf, &ip6->ip6_dst)); 704 strcat(dbuf, "]"); 705 strcpy(sbuf, "["); 706 strcat(sbuf, 707 ip6_sprintf(ip6buf, &ip6->ip6_src)); 708 strcat(sbuf, "]"); 709 } else 710#endif /* INET6 */ 711 { 712 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 713 strcpy(sbuf, inet_ntoa(ip->ip_src)); 714 } 715 log(LOG_INFO, 716 "Connection attempt to TCP %s:%d " 717 "from %s:%d flags:0x%02x\n", 718 dbuf, ntohs(th->th_dport), sbuf, 719 ntohs(th->th_sport), thflags); 720 } 721 /* 722 * When blackholing do not respond with a RST but 723 * completely ignore the segment and drop it. 724 */ 725 if ((blackhole == 1 && (thflags & TH_SYN)) || 726 blackhole == 2) 727 goto dropunlock; 728 729 rstreason = BANDLIM_RST_CLOSEDPORT; 730 goto dropwithreset; 731 } 732 INP_LOCK(inp); 733 734 /* Check the minimum TTL for socket. */ 735 if (inp->inp_ip_minttl != 0) { 736#ifdef INET6 737 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 738 goto dropunlock; 739 else 740#endif 741 if (inp->inp_ip_minttl > ip->ip_ttl) 742 goto dropunlock; 743 } 744 745 /* 746 * A previous connection in TIMEWAIT state is supposed to catch 747 * stray or duplicate segments arriving late. If this segment 748 * was a legitimate new connection attempt the old INPCB gets 749 * removed and we can try again to find a listening socket. 750 */ 751 if (inp->inp_vflag & INP_TIMEWAIT) { 752 if (thflags & TH_SYN) 753 tcp_dooptions(&to, optp, optlen, TO_SYN); 754 /* NB: tcp_timewait unlocks the INP and frees the mbuf. */ 755 if (tcp_timewait(inp, &to, th, m, tlen)) 756 goto findpcb; 757 INP_INFO_WUNLOCK(&tcbinfo); 758 return; 759 } 760 /* 761 * The TCPCB may no longer exist if the connection is winding 762 * down or it is in the CLOSED state. Either way we drop the 763 * segment and send an appropriate response. 764 */ 765 tp = intotcpcb(inp); 766 if (tp == NULL) { 767 rstreason = BANDLIM_RST_CLOSEDPORT; 768 goto dropwithreset; 769 } 770 if (tp->t_state == TCPS_CLOSED) 771 goto dropunlock; /* XXX: dropwithreset??? */ 772 773#ifdef MAC 774 INP_LOCK_ASSERT(inp); 775 if (mac_check_inpcb_deliver(inp, m)) 776 goto dropunlock; 777#endif 778 so = inp->inp_socket; 779 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 780#ifdef TCPDEBUG 781 if (so->so_options & SO_DEBUG) { 782 ostate = tp->t_state; 783#ifdef INET6 784 if (isipv6) 785 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 786 else 787#endif /* INET6 */ 788 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 789 tcp_savetcp = *th; 790 } 791#endif 792 /* 793 * When the socket is accepting connections (the INPCB is in LISTEN 794 * state) we look into the SYN cache if this is a new connection 795 * attempt or the completion of a previous one. 796 */ 797 if (so->so_options & SO_ACCEPTCONN) { 798 struct in_conninfo inc; 799 800 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 801 "tp not listening", __func__)); 802 803 bzero(&inc, sizeof(inc)); 804 inc.inc_isipv6 = isipv6; 805#ifdef INET6 806 if (isipv6) { 807 inc.inc6_faddr = ip6->ip6_src; 808 inc.inc6_laddr = ip6->ip6_dst; 809 } else 810#endif 811 { 812 inc.inc_faddr = ip->ip_src; 813 inc.inc_laddr = ip->ip_dst; 814 } 815 inc.inc_fport = th->th_sport; 816 inc.inc_lport = th->th_dport; 817 818 /* 819 * If the state is LISTEN then ignore segment if it contains 820 * a RST. If the segment contains an ACK then it is bad and 821 * send a RST. If it does not contain a SYN then it is not 822 * interesting; drop it. 823 * 824 * If the state is SYN_RECEIVED (syncache) and seg contains 825 * an ACK, but not for our SYN/ACK, send a RST. If the seg 826 * contains a RST, check the sequence number to see if it 827 * is a valid reset segment. 828 */ 829 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 830 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 831 /* 832 * Parse the TCP options here because 833 * syncookies need access to the reflected 834 * timestamp. 835 */ 836 tcp_dooptions(&to, optp, optlen, 0); 837 /* 838 * NB: syncache_expand() doesn't unlock 839 * inp and tcpinfo locks. 840 */ 841 if (!syncache_expand(&inc, &to, th, &so, m)) { 842 /* 843 * No syncache entry or ACK was not 844 * for our SYN/ACK. Send a RST. 845 */ 846 rstreason = BANDLIM_RST_OPENPORT; 847 goto dropwithreset; 848 } 849 if (so == NULL) { 850 /* 851 * We completed the 3-way handshake 852 * but could not allocate a socket 853 * either due to memory shortage, 854 * listen queue length limits or 855 * global socket limits. 856 */ 857 rstreason = BANDLIM_UNLIMITED; 858 goto dropwithreset; 859 } 860 /* 861 * Socket is created in state SYN_RECEIVED. 862 * Continue processing segment. 863 */ 864 INP_UNLOCK(inp); /* listen socket */ 865 inp = sotoinpcb(so); 866 INP_LOCK(inp); /* new connection */ 867 tp = intotcpcb(inp); 868 /* 869 * Process the segment and the data it 870 * contains. tcp_do_segment() consumes 871 * the mbuf chain and unlocks the inpcb. 872 */ 873 tcp_do_segment(m, th, so, tp, drop_hdrlen, 874 tlen); 875 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 876 return; 877 } 878 if (thflags & TH_RST) { 879 syncache_chkrst(&inc, th); 880 goto dropunlock; 881 } 882 if (thflags & TH_ACK) { 883 syncache_badack(&inc); 884 tcpstat.tcps_badsyn++; 885 rstreason = BANDLIM_RST_OPENPORT; 886 goto dropwithreset; 887 } 888 goto dropunlock; 889 } 890 891 /* 892 * Segment's flags are (SYN) or (SYN|FIN). 893 */ 894#ifdef INET6 895 /* 896 * If deprecated address is forbidden, 897 * we do not accept SYN to deprecated interface 898 * address to prevent any new inbound connection from 899 * getting established. 900 * When we do not accept SYN, we send a TCP RST, 901 * with deprecated source address (instead of dropping 902 * it). We compromise it as it is much better for peer 903 * to send a RST, and RST will be the final packet 904 * for the exchange. 905 * 906 * If we do not forbid deprecated addresses, we accept 907 * the SYN packet. RFC2462 does not suggest dropping 908 * SYN in this case. 909 * If we decipher RFC2462 5.5.4, it says like this: 910 * 1. use of deprecated addr with existing 911 * communication is okay - "SHOULD continue to be 912 * used" 913 * 2. use of it with new communication: 914 * (2a) "SHOULD NOT be used if alternate address 915 * with sufficient scope is available" 916 * (2b) nothing mentioned otherwise. 917 * Here we fall into (2b) case as we have no choice in 918 * our source address selection - we must obey the peer. 919 * 920 * The wording in RFC2462 is confusing, and there are 921 * multiple description text for deprecated address 922 * handling - worse, they are not exactly the same. 923 * I believe 5.5.4 is the best one, so we follow 5.5.4. 924 */ 925 if (isipv6 && !ip6_use_deprecated) { 926 struct in6_ifaddr *ia6; 927 928 if ((ia6 = ip6_getdstifaddr(m)) && 929 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 930 rstreason = BANDLIM_RST_OPENPORT; 931 goto dropwithreset; 932 } 933 } 934#endif 935 /* 936 * Basic sanity checks on incoming SYN requests: 937 * 938 * Don't bother responding if the destination was a 939 * broadcast according to RFC1122 4.2.3.10, p. 104. 940 * 941 * If it is from this socket, drop it, it must be forged. 942 * 943 * Note that it is quite possible to receive unicast 944 * link-layer packets with a broadcast IP address. Use 945 * in_broadcast() to find them. 946 */ 947 if (m->m_flags & (M_BCAST|M_MCAST)) 948 goto dropunlock; 949 if (isipv6) { 950#ifdef INET6 951 if (th->th_dport == th->th_sport && 952 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) 953 goto dropunlock; 954 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 955 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 956 goto dropunlock; 957#endif 958 } else { 959 if (th->th_dport == th->th_sport && 960 ip->ip_dst.s_addr == ip->ip_src.s_addr) 961 goto dropunlock; 962 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 963 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 964 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 965 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 966 goto dropunlock; 967 } 968 /* 969 * SYN appears to be valid. Create compressed TCP state 970 * for syncache. 971 */ 972#ifdef TCPDEBUG 973 if (so->so_options & SO_DEBUG) 974 tcp_trace(TA_INPUT, ostate, tp, 975 (void *)tcp_saveipgen, &tcp_savetcp, 0); 976#endif 977 tcp_dooptions(&to, optp, optlen, TO_SYN); 978 syncache_add(&inc, &to, th, inp, &so, m); 979 /* 980 * Entry added to syncache and mbuf consumed. 981 * Everything unlocked already by syncache_add(). 982 */ 983 return; 984 } 985 986 /* 987 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or late 988 * state. tcp_do_segment() always consumes the mbuf chain, unlocks the 989 * inpcb, and unlocks the pcbinfo. 990 */ 991 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen); 992 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 993 return; 994 995dropwithreset: 996 INP_INFO_WLOCK_ASSERT(&tcbinfo); 997 tcp_dropwithreset(m, th, tp, tlen, rstreason); 998 m = NULL; /* mbuf chain got consumed. */ 999dropunlock: 1000 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1001 if (inp != NULL) 1002 INP_UNLOCK(inp); 1003 INP_INFO_WUNLOCK(&tcbinfo); 1004drop: 1005 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 1006 if (m != NULL) 1007 m_freem(m); 1008 return; 1009} 1010 1011static void 1012tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1013 struct tcpcb *tp, int drop_hdrlen, int tlen) 1014{ 1015 int thflags, acked, ourfinisacked, needoutput = 0; 1016 int headlocked = 1; 1017 int rstreason, todrop, win; 1018 u_long tiwin; 1019 struct tcpopt to; 1020 1021#ifdef TCPDEBUG 1022 /* 1023 * The size of tcp_saveipgen must be the size of the max ip header, 1024 * now IPv6. 1025 */ 1026 u_char tcp_saveipgen[IP6_HDR_LEN]; 1027 struct tcphdr tcp_savetcp; 1028 short ostate = 0; 1029#endif 1030 thflags = th->th_flags; 1031 1032 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1033 INP_LOCK_ASSERT(tp->t_inpcb); 1034 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1035 __func__)); 1036 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1037 __func__)); 1038 1039 /* 1040 * Segment received on connection. 1041 * Reset idle time and keep-alive timer. 1042 */ 1043 tp->t_rcvtime = ticks; 1044 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1045 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1046 1047 /* 1048 * Unscale the window into a 32-bit value. 1049 * This value is bogus for the TCPS_SYN_SENT state 1050 * and is overwritten later. 1051 */ 1052 tiwin = th->th_win << tp->snd_scale; 1053 1054 /* 1055 * Parse options on any incoming segment. 1056 */ 1057 tcp_dooptions(&to, (u_char *)(th + 1), 1058 (th->th_off << 2) - sizeof(struct tcphdr), 1059 (thflags & TH_SYN) ? TO_SYN : 0); 1060 1061 /* 1062 * If echoed timestamp is later than the current time, 1063 * fall back to non RFC1323 RTT calculation. Normalize 1064 * timestamp if syncookies were used when this connection 1065 * was established. 1066 */ 1067 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1068 to.to_tsecr -= tp->ts_offset; 1069 if (TSTMP_GT(to.to_tsecr, ticks)) 1070 to.to_tsecr = 0; 1071 } 1072 1073 /* 1074 * Process options only when we get SYN/ACK back. The SYN case 1075 * for incoming connections is handled in tcp_syncache. 1076 * XXX this is traditional behavior, may need to be cleaned up. 1077 */ 1078 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1079 if ((to.to_flags & TOF_SCALE) && 1080 (tp->t_flags & TF_REQ_SCALE)) { 1081 tp->t_flags |= TF_RCVD_SCALE; 1082 tp->snd_scale = to.to_wscale; 1083 tp->snd_wnd = th->th_win << tp->snd_scale; 1084 tiwin = tp->snd_wnd; 1085 } 1086 if (to.to_flags & TOF_TS) { 1087 tp->t_flags |= TF_RCVD_TSTMP; 1088 tp->ts_recent = to.to_tsval; 1089 tp->ts_recent_age = ticks; 1090 } 1091 /* Initial send window, already scaled. */ 1092 tp->snd_wnd = th->th_win; 1093 if (to.to_flags & TOF_MSS) 1094 tcp_mss(tp, to.to_mss); 1095 if ((tp->t_flags & TF_SACK_PERMIT) && 1096 (to.to_flags & TOF_SACKPERM) == 0) 1097 tp->t_flags &= ~TF_SACK_PERMIT; 1098 } 1099 1100 /* 1101 * Header prediction: check for the two common cases 1102 * of a uni-directional data xfer. If the packet has 1103 * no control flags, is in-sequence, the window didn't 1104 * change and we're not retransmitting, it's a 1105 * candidate. If the length is zero and the ack moved 1106 * forward, we're the sender side of the xfer. Just 1107 * free the data acked & wake any higher level process 1108 * that was blocked waiting for space. If the length 1109 * is non-zero and the ack didn't move, we're the 1110 * receiver side. If we're getting packets in-order 1111 * (the reassembly queue is empty), add the data to 1112 * the socket buffer and note that we need a delayed ack. 1113 * Make sure that the hidden state-flags are also off. 1114 * Since we check for TCPS_ESTABLISHED first, it can only 1115 * be TH_NEEDSYN. 1116 */ 1117 if (tp->t_state == TCPS_ESTABLISHED && 1118 th->th_seq == tp->rcv_nxt && 1119 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1120 tp->snd_nxt == tp->snd_max && 1121 tiwin && tiwin == tp->snd_wnd && 1122 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1123 LIST_EMPTY(&tp->t_segq) && 1124 ((to.to_flags & TOF_TS) == 0 || 1125 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1126 1127 /* 1128 * If last ACK falls within this segment's sequence numbers, 1129 * record the timestamp. 1130 * NOTE that the test is modified according to the latest 1131 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1132 */ 1133 if ((to.to_flags & TOF_TS) != 0 && 1134 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1135 tp->ts_recent_age = ticks; 1136 tp->ts_recent = to.to_tsval; 1137 } 1138 1139 if (tlen == 0) { 1140 if (SEQ_GT(th->th_ack, tp->snd_una) && 1141 SEQ_LEQ(th->th_ack, tp->snd_max) && 1142 tp->snd_cwnd >= tp->snd_wnd && 1143 ((!tcp_do_newreno && 1144 !(tp->t_flags & TF_SACK_PERMIT) && 1145 tp->t_dupacks < tcprexmtthresh) || 1146 ((tcp_do_newreno || 1147 (tp->t_flags & TF_SACK_PERMIT)) && 1148 !IN_FASTRECOVERY(tp) && 1149 (to.to_flags & TOF_SACK) == 0 && 1150 TAILQ_EMPTY(&tp->snd_holes)))) { 1151 KASSERT(headlocked, 1152 ("%s: headlocked", __func__)); 1153 INP_INFO_WUNLOCK(&tcbinfo); 1154 headlocked = 0; 1155 /* 1156 * this is a pure ack for outstanding data. 1157 */ 1158 ++tcpstat.tcps_predack; 1159 /* 1160 * "bad retransmit" recovery 1161 */ 1162 if (tp->t_rxtshift == 1 && 1163 ticks < tp->t_badrxtwin) { 1164 ++tcpstat.tcps_sndrexmitbad; 1165 tp->snd_cwnd = tp->snd_cwnd_prev; 1166 tp->snd_ssthresh = 1167 tp->snd_ssthresh_prev; 1168 tp->snd_recover = tp->snd_recover_prev; 1169 if (tp->t_flags & TF_WASFRECOVERY) 1170 ENTER_FASTRECOVERY(tp); 1171 tp->snd_nxt = tp->snd_max; 1172 tp->t_badrxtwin = 0; 1173 } 1174 1175 /* 1176 * Recalculate the transmit timer / rtt. 1177 * 1178 * Some boxes send broken timestamp replies 1179 * during the SYN+ACK phase, ignore 1180 * timestamps of 0 or we could calculate a 1181 * huge RTT and blow up the retransmit timer. 1182 */ 1183 if ((to.to_flags & TOF_TS) != 0 && 1184 to.to_tsecr) { 1185 if (!tp->t_rttlow || 1186 tp->t_rttlow > ticks - to.to_tsecr) 1187 tp->t_rttlow = ticks - to.to_tsecr; 1188 tcp_xmit_timer(tp, 1189 ticks - to.to_tsecr + 1); 1190 } else if (tp->t_rtttime && 1191 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1192 if (!tp->t_rttlow || 1193 tp->t_rttlow > ticks - tp->t_rtttime) 1194 tp->t_rttlow = ticks - tp->t_rtttime; 1195 tcp_xmit_timer(tp, 1196 ticks - tp->t_rtttime); 1197 } 1198 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1199 acked = th->th_ack - tp->snd_una; 1200 tcpstat.tcps_rcvackpack++; 1201 tcpstat.tcps_rcvackbyte += acked; 1202 sbdrop(&so->so_snd, acked); 1203 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1204 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1205 tp->snd_recover = th->th_ack - 1; 1206 tp->snd_una = th->th_ack; 1207 /* 1208 * pull snd_wl2 up to prevent seq wrap relative 1209 * to th_ack. 1210 */ 1211 tp->snd_wl2 = th->th_ack; 1212 tp->t_dupacks = 0; 1213 m_freem(m); 1214 ND6_HINT(tp); /* some progress has been done */ 1215 1216 /* 1217 * If all outstanding data are acked, stop 1218 * retransmit timer, otherwise restart timer 1219 * using current (possibly backed-off) value. 1220 * If process is waiting for space, 1221 * wakeup/selwakeup/signal. If data 1222 * are ready to send, let tcp_output 1223 * decide between more output or persist. 1224 1225#ifdef TCPDEBUG 1226 if (so->so_options & SO_DEBUG) 1227 tcp_trace(TA_INPUT, ostate, tp, 1228 (void *)tcp_saveipgen, 1229 &tcp_savetcp, 0); 1230#endif 1231 */ 1232 if (tp->snd_una == tp->snd_max) 1233 tcp_timer_activate(tp, TT_REXMT, 0); 1234 else if (!tcp_timer_active(tp, TT_PERSIST)) 1235 tcp_timer_activate(tp, TT_REXMT, 1236 tp->t_rxtcur); 1237 1238 sowwakeup(so); 1239 if (so->so_snd.sb_cc) 1240 (void) tcp_output(tp); 1241 goto check_delack; 1242 } 1243 } else if (th->th_ack == tp->snd_una && 1244 tlen <= sbspace(&so->so_rcv)) { 1245 int newsize = 0; /* automatic sockbuf scaling */ 1246 1247 KASSERT(headlocked, ("%s: headlocked", __func__)); 1248 INP_INFO_WUNLOCK(&tcbinfo); 1249 headlocked = 0; 1250 /* 1251 * this is a pure, in-sequence data packet 1252 * with nothing on the reassembly queue and 1253 * we have enough buffer space to take it. 1254 */ 1255 /* Clean receiver SACK report if present */ 1256 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1257 tcp_clean_sackreport(tp); 1258 ++tcpstat.tcps_preddat; 1259 tp->rcv_nxt += tlen; 1260 /* 1261 * Pull snd_wl1 up to prevent seq wrap relative to 1262 * th_seq. 1263 */ 1264 tp->snd_wl1 = th->th_seq; 1265 /* 1266 * Pull rcv_up up to prevent seq wrap relative to 1267 * rcv_nxt. 1268 */ 1269 tp->rcv_up = tp->rcv_nxt; 1270 tcpstat.tcps_rcvpack++; 1271 tcpstat.tcps_rcvbyte += tlen; 1272 ND6_HINT(tp); /* some progress has been done */ 1273#ifdef TCPDEBUG 1274 if (so->so_options & SO_DEBUG) 1275 tcp_trace(TA_INPUT, ostate, tp, 1276 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1277#endif 1278 /* 1279 * Automatic sizing of receive socket buffer. Often the send 1280 * buffer size is not optimally adjusted to the actual network 1281 * conditions at hand (delay bandwidth product). Setting the 1282 * buffer size too small limits throughput on links with high 1283 * bandwidth and high delay (eg. trans-continental/oceanic links). 1284 * 1285 * On the receive side the socket buffer memory is only rarely 1286 * used to any significant extent. This allows us to be much 1287 * more aggressive in scaling the receive socket buffer. For 1288 * the case that the buffer space is actually used to a large 1289 * extent and we run out of kernel memory we can simply drop 1290 * the new segments; TCP on the sender will just retransmit it 1291 * later. Setting the buffer size too big may only consume too 1292 * much kernel memory if the application doesn't read() from 1293 * the socket or packet loss or reordering makes use of the 1294 * reassembly queue. 1295 * 1296 * The criteria to step up the receive buffer one notch are: 1297 * 1. the number of bytes received during the time it takes 1298 * one timestamp to be reflected back to us (the RTT); 1299 * 2. received bytes per RTT is within seven eighth of the 1300 * current socket buffer size; 1301 * 3. receive buffer size has not hit maximal automatic size; 1302 * 1303 * This algorithm does one step per RTT at most and only if 1304 * we receive a bulk stream w/o packet losses or reorderings. 1305 * Shrinking the buffer during idle times is not necessary as 1306 * it doesn't consume any memory when idle. 1307 * 1308 * TODO: Only step up if the application is actually serving 1309 * the buffer to better manage the socket buffer resources. 1310 */ 1311 if (tcp_do_autorcvbuf && 1312 to.to_tsecr && 1313 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1314 if (to.to_tsecr > tp->rfbuf_ts && 1315 to.to_tsecr - tp->rfbuf_ts < hz) { 1316 if (tp->rfbuf_cnt > 1317 (so->so_rcv.sb_hiwat / 8 * 7) && 1318 so->so_rcv.sb_hiwat < 1319 tcp_autorcvbuf_max) { 1320 newsize = 1321 min(so->so_rcv.sb_hiwat + 1322 tcp_autorcvbuf_inc, 1323 tcp_autorcvbuf_max); 1324 } 1325 /* Start over with next RTT. */ 1326 tp->rfbuf_ts = 0; 1327 tp->rfbuf_cnt = 0; 1328 } else 1329 tp->rfbuf_cnt += tlen; /* add up */ 1330 } 1331 1332 /* Add data to socket buffer. */ 1333 SOCKBUF_LOCK(&so->so_rcv); 1334 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1335 m_freem(m); 1336 } else { 1337 /* 1338 * Set new socket buffer size. 1339 * Give up when limit is reached. 1340 */ 1341 if (newsize) 1342 if (!sbreserve_locked(&so->so_rcv, 1343 newsize, so, curthread)) 1344 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1345 m_adj(m, drop_hdrlen); /* delayed header drop */ 1346 sbappendstream_locked(&so->so_rcv, m); 1347 } 1348 sorwakeup_locked(so); 1349 if (DELAY_ACK(tp)) { 1350 tp->t_flags |= TF_DELACK; 1351 } else { 1352 tp->t_flags |= TF_ACKNOW; 1353 tcp_output(tp); 1354 } 1355 goto check_delack; 1356 } 1357 } 1358 1359 /* 1360 * Calculate amount of space in receive window, 1361 * and then do TCP input processing. 1362 * Receive window is amount of space in rcv queue, 1363 * but not less than advertised window. 1364 */ 1365 win = sbspace(&so->so_rcv); 1366 if (win < 0) 1367 win = 0; 1368 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1369 1370 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1371 tp->rfbuf_ts = 0; 1372 tp->rfbuf_cnt = 0; 1373 1374 switch (tp->t_state) { 1375 1376 /* 1377 * If the state is SYN_RECEIVED: 1378 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1379 */ 1380 case TCPS_SYN_RECEIVED: 1381 if ((thflags & TH_ACK) && 1382 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1383 SEQ_GT(th->th_ack, tp->snd_max))) { 1384 rstreason = BANDLIM_RST_OPENPORT; 1385 goto dropwithreset; 1386 } 1387 break; 1388 1389 /* 1390 * If the state is SYN_SENT: 1391 * if seg contains an ACK, but not for our SYN, drop the input. 1392 * if seg contains a RST, then drop the connection. 1393 * if seg does not contain SYN, then drop it. 1394 * Otherwise this is an acceptable SYN segment 1395 * initialize tp->rcv_nxt and tp->irs 1396 * if seg contains ack then advance tp->snd_una 1397 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1398 * arrange for segment to be acked (eventually) 1399 * continue processing rest of data/controls, beginning with URG 1400 */ 1401 case TCPS_SYN_SENT: 1402 if ((thflags & TH_ACK) && 1403 (SEQ_LEQ(th->th_ack, tp->iss) || 1404 SEQ_GT(th->th_ack, tp->snd_max))) { 1405 rstreason = BANDLIM_UNLIMITED; 1406 goto dropwithreset; 1407 } 1408 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1409 tp = tcp_drop(tp, ECONNREFUSED); 1410 if (thflags & TH_RST) 1411 goto drop; 1412 if (!(thflags & TH_SYN)) 1413 goto drop; 1414 1415 tp->irs = th->th_seq; 1416 tcp_rcvseqinit(tp); 1417 if (thflags & TH_ACK) { 1418 tcpstat.tcps_connects++; 1419 soisconnected(so); 1420#ifdef MAC 1421 SOCK_LOCK(so); 1422 mac_set_socket_peer_from_mbuf(m, so); 1423 SOCK_UNLOCK(so); 1424#endif 1425 /* Do window scaling on this connection? */ 1426 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1427 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1428 tp->rcv_scale = tp->request_r_scale; 1429 } 1430 tp->rcv_adv += tp->rcv_wnd; 1431 tp->snd_una++; /* SYN is acked */ 1432 /* 1433 * If there's data, delay ACK; if there's also a FIN 1434 * ACKNOW will be turned on later. 1435 */ 1436 if (DELAY_ACK(tp) && tlen != 0) 1437 tcp_timer_activate(tp, TT_DELACK, 1438 tcp_delacktime); 1439 else 1440 tp->t_flags |= TF_ACKNOW; 1441 /* 1442 * Received <SYN,ACK> in SYN_SENT[*] state. 1443 * Transitions: 1444 * SYN_SENT --> ESTABLISHED 1445 * SYN_SENT* --> FIN_WAIT_1 1446 */ 1447 tp->t_starttime = ticks; 1448 if (tp->t_flags & TF_NEEDFIN) { 1449 tp->t_state = TCPS_FIN_WAIT_1; 1450 tp->t_flags &= ~TF_NEEDFIN; 1451 thflags &= ~TH_SYN; 1452 } else { 1453 tp->t_state = TCPS_ESTABLISHED; 1454 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1455 } 1456 } else { 1457 /* 1458 * Received initial SYN in SYN-SENT[*] state => 1459 * simultaneous open. If segment contains CC option 1460 * and there is a cached CC, apply TAO test. 1461 * If it succeeds, connection is * half-synchronized. 1462 * Otherwise, do 3-way handshake: 1463 * SYN-SENT -> SYN-RECEIVED 1464 * SYN-SENT* -> SYN-RECEIVED* 1465 * If there was no CC option, clear cached CC value. 1466 */ 1467 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1468 tcp_timer_activate(tp, TT_REXMT, 0); 1469 tp->t_state = TCPS_SYN_RECEIVED; 1470 } 1471 1472 KASSERT(headlocked, ("%s: trimthenstep6: head not locked", 1473 __func__)); 1474 INP_LOCK_ASSERT(tp->t_inpcb); 1475 1476 /* 1477 * Advance th->th_seq to correspond to first data byte. 1478 * If data, trim to stay within window, 1479 * dropping FIN if necessary. 1480 */ 1481 th->th_seq++; 1482 if (tlen > tp->rcv_wnd) { 1483 todrop = tlen - tp->rcv_wnd; 1484 m_adj(m, -todrop); 1485 tlen = tp->rcv_wnd; 1486 thflags &= ~TH_FIN; 1487 tcpstat.tcps_rcvpackafterwin++; 1488 tcpstat.tcps_rcvbyteafterwin += todrop; 1489 } 1490 tp->snd_wl1 = th->th_seq - 1; 1491 tp->rcv_up = th->th_seq; 1492 /* 1493 * Client side of transaction: already sent SYN and data. 1494 * If the remote host used T/TCP to validate the SYN, 1495 * our data will be ACK'd; if so, enter normal data segment 1496 * processing in the middle of step 5, ack processing. 1497 * Otherwise, goto step 6. 1498 */ 1499 if (thflags & TH_ACK) 1500 goto process_ACK; 1501 1502 goto step6; 1503 1504 /* 1505 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1506 * do normal processing. 1507 * 1508 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1509 */ 1510 case TCPS_LAST_ACK: 1511 case TCPS_CLOSING: 1512 break; /* continue normal processing */ 1513 } 1514 1515 /* 1516 * States other than LISTEN or SYN_SENT. 1517 * First check the RST flag and sequence number since reset segments 1518 * are exempt from the timestamp and connection count tests. This 1519 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1520 * below which allowed reset segments in half the sequence space 1521 * to fall though and be processed (which gives forged reset 1522 * segments with a random sequence number a 50 percent chance of 1523 * killing a connection). 1524 * Then check timestamp, if present. 1525 * Then check the connection count, if present. 1526 * Then check that at least some bytes of segment are within 1527 * receive window. If segment begins before rcv_nxt, 1528 * drop leading data (and SYN); if nothing left, just ack. 1529 * 1530 * 1531 * If the RST bit is set, check the sequence number to see 1532 * if this is a valid reset segment. 1533 * RFC 793 page 37: 1534 * In all states except SYN-SENT, all reset (RST) segments 1535 * are validated by checking their SEQ-fields. A reset is 1536 * valid if its sequence number is in the window. 1537 * Note: this does not take into account delayed ACKs, so 1538 * we should test against last_ack_sent instead of rcv_nxt. 1539 * The sequence number in the reset segment is normally an 1540 * echo of our outgoing acknowlegement numbers, but some hosts 1541 * send a reset with the sequence number at the rightmost edge 1542 * of our receive window, and we have to handle this case. 1543 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1544 * that brute force RST attacks are possible. To combat this, 1545 * we use a much stricter check while in the ESTABLISHED state, 1546 * only accepting RSTs where the sequence number is equal to 1547 * last_ack_sent. In all other states (the states in which a 1548 * RST is more likely), the more permissive check is used. 1549 * If we have multiple segments in flight, the intial reset 1550 * segment sequence numbers will be to the left of last_ack_sent, 1551 * but they will eventually catch up. 1552 * In any case, it never made sense to trim reset segments to 1553 * fit the receive window since RFC 1122 says: 1554 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1555 * 1556 * A TCP SHOULD allow a received RST segment to include data. 1557 * 1558 * DISCUSSION 1559 * It has been suggested that a RST segment could contain 1560 * ASCII text that encoded and explained the cause of the 1561 * RST. No standard has yet been established for such 1562 * data. 1563 * 1564 * If the reset segment passes the sequence number test examine 1565 * the state: 1566 * SYN_RECEIVED STATE: 1567 * If passive open, return to LISTEN state. 1568 * If active open, inform user that connection was refused. 1569 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1570 * Inform user that connection was reset, and close tcb. 1571 * CLOSING, LAST_ACK STATES: 1572 * Close the tcb. 1573 * TIME_WAIT STATE: 1574 * Drop the segment - see Stevens, vol. 2, p. 964 and 1575 * RFC 1337. 1576 */ 1577 if (thflags & TH_RST) { 1578 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1579 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1580 switch (tp->t_state) { 1581 1582 case TCPS_SYN_RECEIVED: 1583 so->so_error = ECONNREFUSED; 1584 goto close; 1585 1586 case TCPS_ESTABLISHED: 1587 if (tcp_insecure_rst == 0 && 1588 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 1589 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 1590 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1591 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 1592 tcpstat.tcps_badrst++; 1593 goto drop; 1594 } 1595 case TCPS_FIN_WAIT_1: 1596 case TCPS_FIN_WAIT_2: 1597 case TCPS_CLOSE_WAIT: 1598 so->so_error = ECONNRESET; 1599 close: 1600 tp->t_state = TCPS_CLOSED; 1601 tcpstat.tcps_drops++; 1602 KASSERT(headlocked, ("%s: trimthenstep6: " 1603 "tcp_close: head not locked", __func__)); 1604 tp = tcp_close(tp); 1605 break; 1606 1607 case TCPS_CLOSING: 1608 case TCPS_LAST_ACK: 1609 KASSERT(headlocked, ("%s: trimthenstep6: " 1610 "tcp_close.2: head not locked", __func__)); 1611 tp = tcp_close(tp); 1612 break; 1613 } 1614 } 1615 goto drop; 1616 } 1617 1618 /* 1619 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1620 * and it's less than ts_recent, drop it. 1621 */ 1622 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1623 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1624 1625 /* Check to see if ts_recent is over 24 days old. */ 1626 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1627 /* 1628 * Invalidate ts_recent. If this segment updates 1629 * ts_recent, the age will be reset later and ts_recent 1630 * will get a valid value. If it does not, setting 1631 * ts_recent to zero will at least satisfy the 1632 * requirement that zero be placed in the timestamp 1633 * echo reply when ts_recent isn't valid. The 1634 * age isn't reset until we get a valid ts_recent 1635 * because we don't want out-of-order segments to be 1636 * dropped when ts_recent is old. 1637 */ 1638 tp->ts_recent = 0; 1639 } else { 1640 tcpstat.tcps_rcvduppack++; 1641 tcpstat.tcps_rcvdupbyte += tlen; 1642 tcpstat.tcps_pawsdrop++; 1643 if (tlen) 1644 goto dropafterack; 1645 goto drop; 1646 } 1647 } 1648 1649 /* 1650 * In the SYN-RECEIVED state, validate that the packet belongs to 1651 * this connection before trimming the data to fit the receive 1652 * window. Check the sequence number versus IRS since we know 1653 * the sequence numbers haven't wrapped. This is a partial fix 1654 * for the "LAND" DoS attack. 1655 */ 1656 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1657 rstreason = BANDLIM_RST_OPENPORT; 1658 goto dropwithreset; 1659 } 1660 1661 todrop = tp->rcv_nxt - th->th_seq; 1662 if (todrop > 0) { 1663 if (thflags & TH_SYN) { 1664 thflags &= ~TH_SYN; 1665 th->th_seq++; 1666 if (th->th_urp > 1) 1667 th->th_urp--; 1668 else 1669 thflags &= ~TH_URG; 1670 todrop--; 1671 } 1672 /* 1673 * Following if statement from Stevens, vol. 2, p. 960. 1674 */ 1675 if (todrop > tlen 1676 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1677 /* 1678 * Any valid FIN must be to the left of the window. 1679 * At this point the FIN must be a duplicate or out 1680 * of sequence; drop it. 1681 */ 1682 thflags &= ~TH_FIN; 1683 1684 /* 1685 * Send an ACK to resynchronize and drop any data. 1686 * But keep on processing for RST or ACK. 1687 */ 1688 tp->t_flags |= TF_ACKNOW; 1689 todrop = tlen; 1690 tcpstat.tcps_rcvduppack++; 1691 tcpstat.tcps_rcvdupbyte += todrop; 1692 } else { 1693 tcpstat.tcps_rcvpartduppack++; 1694 tcpstat.tcps_rcvpartdupbyte += todrop; 1695 } 1696 drop_hdrlen += todrop; /* drop from the top afterwards */ 1697 th->th_seq += todrop; 1698 tlen -= todrop; 1699 if (th->th_urp > todrop) 1700 th->th_urp -= todrop; 1701 else { 1702 thflags &= ~TH_URG; 1703 th->th_urp = 0; 1704 } 1705 } 1706 1707 /* 1708 * If new data are received on a connection after the 1709 * user processes are gone, then RST the other end. 1710 */ 1711 if ((so->so_state & SS_NOFDREF) && 1712 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1713 KASSERT(headlocked, ("%s: trimthenstep6: tcp_close.3: head " 1714 "not locked", __func__)); 1715 tp = tcp_close(tp); 1716 tcpstat.tcps_rcvafterclose++; 1717 rstreason = BANDLIM_UNLIMITED; 1718 goto dropwithreset; 1719 } 1720 1721 /* 1722 * If segment ends after window, drop trailing data 1723 * (and PUSH and FIN); if nothing left, just ACK. 1724 */ 1725 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 1726 if (todrop > 0) { 1727 tcpstat.tcps_rcvpackafterwin++; 1728 if (todrop >= tlen) { 1729 tcpstat.tcps_rcvbyteafterwin += tlen; 1730 /* 1731 * If window is closed can only take segments at 1732 * window edge, and have to drop data and PUSH from 1733 * incoming segments. Continue processing, but 1734 * remember to ack. Otherwise, drop segment 1735 * and ack. 1736 */ 1737 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1738 tp->t_flags |= TF_ACKNOW; 1739 tcpstat.tcps_rcvwinprobe++; 1740 } else 1741 goto dropafterack; 1742 } else 1743 tcpstat.tcps_rcvbyteafterwin += todrop; 1744 m_adj(m, -todrop); 1745 tlen -= todrop; 1746 thflags &= ~(TH_PUSH|TH_FIN); 1747 } 1748 1749 /* 1750 * If last ACK falls within this segment's sequence numbers, 1751 * record its timestamp. 1752 * NOTE: 1753 * 1) That the test incorporates suggestions from the latest 1754 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1755 * 2) That updating only on newer timestamps interferes with 1756 * our earlier PAWS tests, so this check should be solely 1757 * predicated on the sequence space of this segment. 1758 * 3) That we modify the segment boundary check to be 1759 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1760 * instead of RFC1323's 1761 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1762 * This modified check allows us to overcome RFC1323's 1763 * limitations as described in Stevens TCP/IP Illustrated 1764 * Vol. 2 p.869. In such cases, we can still calculate the 1765 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1766 */ 1767 if ((to.to_flags & TOF_TS) != 0 && 1768 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1769 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1770 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1771 tp->ts_recent_age = ticks; 1772 tp->ts_recent = to.to_tsval; 1773 } 1774 1775 /* 1776 * If a SYN is in the window, then this is an 1777 * error and we send an RST and drop the connection. 1778 */ 1779 if (thflags & TH_SYN) { 1780 KASSERT(headlocked, ("%s: tcp_drop: trimthenstep6: " 1781 "head not locked", __func__)); 1782 tp = tcp_drop(tp, ECONNRESET); 1783 rstreason = BANDLIM_UNLIMITED; 1784 goto drop; 1785 } 1786 1787 /* 1788 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1789 * flag is on (half-synchronized state), then queue data for 1790 * later processing; else drop segment and return. 1791 */ 1792 if ((thflags & TH_ACK) == 0) { 1793 if (tp->t_state == TCPS_SYN_RECEIVED || 1794 (tp->t_flags & TF_NEEDSYN)) 1795 goto step6; 1796 else if (tp->t_flags & TF_ACKNOW) 1797 goto dropafterack; 1798 else 1799 goto drop; 1800 } 1801 1802 /* 1803 * Ack processing. 1804 */ 1805 switch (tp->t_state) { 1806 1807 /* 1808 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1809 * ESTABLISHED state and continue processing. 1810 * The ACK was checked above. 1811 */ 1812 case TCPS_SYN_RECEIVED: 1813 1814 tcpstat.tcps_connects++; 1815 soisconnected(so); 1816 /* Do window scaling? */ 1817 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1818 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1819 tp->rcv_scale = tp->request_r_scale; 1820 tp->snd_wnd = tiwin; 1821 } 1822 /* 1823 * Make transitions: 1824 * SYN-RECEIVED -> ESTABLISHED 1825 * SYN-RECEIVED* -> FIN-WAIT-1 1826 */ 1827 tp->t_starttime = ticks; 1828 if (tp->t_flags & TF_NEEDFIN) { 1829 tp->t_state = TCPS_FIN_WAIT_1; 1830 tp->t_flags &= ~TF_NEEDFIN; 1831 } else { 1832 tp->t_state = TCPS_ESTABLISHED; 1833 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1834 } 1835 /* 1836 * If segment contains data or ACK, will call tcp_reass() 1837 * later; if not, do so now to pass queued data to user. 1838 */ 1839 if (tlen == 0 && (thflags & TH_FIN) == 0) 1840 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1841 (struct mbuf *)0); 1842 tp->snd_wl1 = th->th_seq - 1; 1843 /* FALLTHROUGH */ 1844 1845 /* 1846 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1847 * ACKs. If the ack is in the range 1848 * tp->snd_una < th->th_ack <= tp->snd_max 1849 * then advance tp->snd_una to th->th_ack and drop 1850 * data from the retransmission queue. If this ACK reflects 1851 * more up to date window information we update our window information. 1852 */ 1853 case TCPS_ESTABLISHED: 1854 case TCPS_FIN_WAIT_1: 1855 case TCPS_FIN_WAIT_2: 1856 case TCPS_CLOSE_WAIT: 1857 case TCPS_CLOSING: 1858 case TCPS_LAST_ACK: 1859 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1860 tcpstat.tcps_rcvacktoomuch++; 1861 goto dropafterack; 1862 } 1863 if ((tp->t_flags & TF_SACK_PERMIT) && 1864 ((to.to_flags & TOF_SACK) || 1865 !TAILQ_EMPTY(&tp->snd_holes))) 1866 tcp_sack_doack(tp, &to, th->th_ack); 1867 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1868 if (tlen == 0 && tiwin == tp->snd_wnd) { 1869 tcpstat.tcps_rcvdupack++; 1870 /* 1871 * If we have outstanding data (other than 1872 * a window probe), this is a completely 1873 * duplicate ack (ie, window info didn't 1874 * change), the ack is the biggest we've 1875 * seen and we've seen exactly our rexmt 1876 * threshhold of them, assume a packet 1877 * has been dropped and retransmit it. 1878 * Kludge snd_nxt & the congestion 1879 * window so we send only this one 1880 * packet. 1881 * 1882 * We know we're losing at the current 1883 * window size so do congestion avoidance 1884 * (set ssthresh to half the current window 1885 * and pull our congestion window back to 1886 * the new ssthresh). 1887 * 1888 * Dup acks mean that packets have left the 1889 * network (they're now cached at the receiver) 1890 * so bump cwnd by the amount in the receiver 1891 * to keep a constant cwnd packets in the 1892 * network. 1893 */ 1894 if (!tcp_timer_active(tp, TT_REXMT) || 1895 th->th_ack != tp->snd_una) 1896 tp->t_dupacks = 0; 1897 else if (++tp->t_dupacks > tcprexmtthresh || 1898 ((tcp_do_newreno || 1899 (tp->t_flags & TF_SACK_PERMIT)) && 1900 IN_FASTRECOVERY(tp))) { 1901 if ((tp->t_flags & TF_SACK_PERMIT) && 1902 IN_FASTRECOVERY(tp)) { 1903 int awnd; 1904 1905 /* 1906 * Compute the amount of data in flight first. 1907 * We can inject new data into the pipe iff 1908 * we have less than 1/2 the original window's 1909 * worth of data in flight. 1910 */ 1911 awnd = (tp->snd_nxt - tp->snd_fack) + 1912 tp->sackhint.sack_bytes_rexmit; 1913 if (awnd < tp->snd_ssthresh) { 1914 tp->snd_cwnd += tp->t_maxseg; 1915 if (tp->snd_cwnd > tp->snd_ssthresh) 1916 tp->snd_cwnd = tp->snd_ssthresh; 1917 } 1918 } else 1919 tp->snd_cwnd += tp->t_maxseg; 1920 (void) tcp_output(tp); 1921 goto drop; 1922 } else if (tp->t_dupacks == tcprexmtthresh) { 1923 tcp_seq onxt = tp->snd_nxt; 1924 u_int win; 1925 1926 /* 1927 * If we're doing sack, check to 1928 * see if we're already in sack 1929 * recovery. If we're not doing sack, 1930 * check to see if we're in newreno 1931 * recovery. 1932 */ 1933 if (tp->t_flags & TF_SACK_PERMIT) { 1934 if (IN_FASTRECOVERY(tp)) { 1935 tp->t_dupacks = 0; 1936 break; 1937 } 1938 } else if (tcp_do_newreno) { 1939 if (SEQ_LEQ(th->th_ack, 1940 tp->snd_recover)) { 1941 tp->t_dupacks = 0; 1942 break; 1943 } 1944 } 1945 win = min(tp->snd_wnd, tp->snd_cwnd) / 1946 2 / tp->t_maxseg; 1947 if (win < 2) 1948 win = 2; 1949 tp->snd_ssthresh = win * tp->t_maxseg; 1950 ENTER_FASTRECOVERY(tp); 1951 tp->snd_recover = tp->snd_max; 1952 tcp_timer_activate(tp, TT_REXMT, 0); 1953 tp->t_rtttime = 0; 1954 if (tp->t_flags & TF_SACK_PERMIT) { 1955 tcpstat.tcps_sack_recovery_episode++; 1956 tp->sack_newdata = tp->snd_nxt; 1957 tp->snd_cwnd = tp->t_maxseg; 1958 (void) tcp_output(tp); 1959 goto drop; 1960 } 1961 tp->snd_nxt = th->th_ack; 1962 tp->snd_cwnd = tp->t_maxseg; 1963 (void) tcp_output(tp); 1964 KASSERT(tp->snd_limited <= 2, 1965 ("%s: tp->snd_limited too big", 1966 __func__)); 1967 tp->snd_cwnd = tp->snd_ssthresh + 1968 tp->t_maxseg * 1969 (tp->t_dupacks - tp->snd_limited); 1970 if (SEQ_GT(onxt, tp->snd_nxt)) 1971 tp->snd_nxt = onxt; 1972 goto drop; 1973 } else if (tcp_do_rfc3042) { 1974 u_long oldcwnd = tp->snd_cwnd; 1975 tcp_seq oldsndmax = tp->snd_max; 1976 u_int sent; 1977 1978 KASSERT(tp->t_dupacks == 1 || 1979 tp->t_dupacks == 2, 1980 ("%s: dupacks not 1 or 2", 1981 __func__)); 1982 if (tp->t_dupacks == 1) 1983 tp->snd_limited = 0; 1984 tp->snd_cwnd = 1985 (tp->snd_nxt - tp->snd_una) + 1986 (tp->t_dupacks - tp->snd_limited) * 1987 tp->t_maxseg; 1988 (void) tcp_output(tp); 1989 sent = tp->snd_max - oldsndmax; 1990 if (sent > tp->t_maxseg) { 1991 KASSERT((tp->t_dupacks == 2 && 1992 tp->snd_limited == 0) || 1993 (sent == tp->t_maxseg + 1 && 1994 tp->t_flags & TF_SENTFIN), 1995 ("%s: sent too much", 1996 __func__)); 1997 tp->snd_limited = 2; 1998 } else if (sent > 0) 1999 ++tp->snd_limited; 2000 tp->snd_cwnd = oldcwnd; 2001 goto drop; 2002 } 2003 } else 2004 tp->t_dupacks = 0; 2005 break; 2006 } 2007 2008 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2009 ("%s: th_ack <= snd_una", __func__)); 2010 2011 /* 2012 * If the congestion window was inflated to account 2013 * for the other side's cached packets, retract it. 2014 */ 2015 if (tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) { 2016 if (IN_FASTRECOVERY(tp)) { 2017 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2018 if (tp->t_flags & TF_SACK_PERMIT) 2019 tcp_sack_partialack(tp, th); 2020 else 2021 tcp_newreno_partial_ack(tp, th); 2022 } else { 2023 /* 2024 * Out of fast recovery. 2025 * Window inflation should have left us 2026 * with approximately snd_ssthresh 2027 * outstanding data. 2028 * But in case we would be inclined to 2029 * send a burst, better to do it via 2030 * the slow start mechanism. 2031 */ 2032 if (SEQ_GT(th->th_ack + 2033 tp->snd_ssthresh, 2034 tp->snd_max)) 2035 tp->snd_cwnd = tp->snd_max - 2036 th->th_ack + 2037 tp->t_maxseg; 2038 else 2039 tp->snd_cwnd = tp->snd_ssthresh; 2040 } 2041 } 2042 } else { 2043 if (tp->t_dupacks >= tcprexmtthresh && 2044 tp->snd_cwnd > tp->snd_ssthresh) 2045 tp->snd_cwnd = tp->snd_ssthresh; 2046 } 2047 tp->t_dupacks = 0; 2048 /* 2049 * If we reach this point, ACK is not a duplicate, 2050 * i.e., it ACKs something we sent. 2051 */ 2052 if (tp->t_flags & TF_NEEDSYN) { 2053 /* 2054 * T/TCP: Connection was half-synchronized, and our 2055 * SYN has been ACK'd (so connection is now fully 2056 * synchronized). Go to non-starred state, 2057 * increment snd_una for ACK of SYN, and check if 2058 * we can do window scaling. 2059 */ 2060 tp->t_flags &= ~TF_NEEDSYN; 2061 tp->snd_una++; 2062 /* Do window scaling? */ 2063 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2064 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2065 tp->rcv_scale = tp->request_r_scale; 2066 /* Send window already scaled. */ 2067 } 2068 } 2069 2070process_ACK: 2071 KASSERT(headlocked, ("%s: process_ACK: head not locked", 2072 __func__)); 2073 INP_LOCK_ASSERT(tp->t_inpcb); 2074 2075 acked = th->th_ack - tp->snd_una; 2076 tcpstat.tcps_rcvackpack++; 2077 tcpstat.tcps_rcvackbyte += acked; 2078 2079 /* 2080 * If we just performed our first retransmit, and the ACK 2081 * arrives within our recovery window, then it was a mistake 2082 * to do the retransmit in the first place. Recover our 2083 * original cwnd and ssthresh, and proceed to transmit where 2084 * we left off. 2085 */ 2086 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2087 ++tcpstat.tcps_sndrexmitbad; 2088 tp->snd_cwnd = tp->snd_cwnd_prev; 2089 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2090 tp->snd_recover = tp->snd_recover_prev; 2091 if (tp->t_flags & TF_WASFRECOVERY) 2092 ENTER_FASTRECOVERY(tp); 2093 tp->snd_nxt = tp->snd_max; 2094 tp->t_badrxtwin = 0; /* XXX probably not required */ 2095 } 2096 2097 /* 2098 * If we have a timestamp reply, update smoothed 2099 * round trip time. If no timestamp is present but 2100 * transmit timer is running and timed sequence 2101 * number was acked, update smoothed round trip time. 2102 * Since we now have an rtt measurement, cancel the 2103 * timer backoff (cf., Phil Karn's retransmit alg.). 2104 * Recompute the initial retransmit timer. 2105 * 2106 * Some boxes send broken timestamp replies 2107 * during the SYN+ACK phase, ignore 2108 * timestamps of 0 or we could calculate a 2109 * huge RTT and blow up the retransmit timer. 2110 */ 2111 if ((to.to_flags & TOF_TS) != 0 && 2112 to.to_tsecr) { 2113 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2114 tp->t_rttlow = ticks - to.to_tsecr; 2115 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2116 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2117 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2118 tp->t_rttlow = ticks - tp->t_rtttime; 2119 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2120 } 2121 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2122 2123 /* 2124 * If all outstanding data is acked, stop retransmit 2125 * timer and remember to restart (more output or persist). 2126 * If there is more data to be acked, restart retransmit 2127 * timer, using current (possibly backed-off) value. 2128 */ 2129 if (th->th_ack == tp->snd_max) { 2130 tcp_timer_activate(tp, TT_REXMT, 0); 2131 needoutput = 1; 2132 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2133 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2134 2135 /* 2136 * If no data (only SYN) was ACK'd, 2137 * skip rest of ACK processing. 2138 */ 2139 if (acked == 0) 2140 goto step6; 2141 2142 /* 2143 * When new data is acked, open the congestion window. 2144 * If the window gives us less than ssthresh packets 2145 * in flight, open exponentially (maxseg per packet). 2146 * Otherwise open linearly: maxseg per window 2147 * (maxseg^2 / cwnd per packet). 2148 */ 2149 if ((!tcp_do_newreno && !(tp->t_flags & TF_SACK_PERMIT)) || 2150 !IN_FASTRECOVERY(tp)) { 2151 u_int cw = tp->snd_cwnd; 2152 u_int incr = tp->t_maxseg; 2153 if (cw > tp->snd_ssthresh) 2154 incr = incr * incr / cw; 2155 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2156 } 2157 SOCKBUF_LOCK(&so->so_snd); 2158 if (acked > so->so_snd.sb_cc) { 2159 tp->snd_wnd -= so->so_snd.sb_cc; 2160 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2161 ourfinisacked = 1; 2162 } else { 2163 sbdrop_locked(&so->so_snd, acked); 2164 tp->snd_wnd -= acked; 2165 ourfinisacked = 0; 2166 } 2167 sowwakeup_locked(so); 2168 /* detect una wraparound */ 2169 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2170 !IN_FASTRECOVERY(tp) && 2171 SEQ_GT(tp->snd_una, tp->snd_recover) && 2172 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2173 tp->snd_recover = th->th_ack - 1; 2174 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2175 IN_FASTRECOVERY(tp) && 2176 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2177 EXIT_FASTRECOVERY(tp); 2178 tp->snd_una = th->th_ack; 2179 if (tp->t_flags & TF_SACK_PERMIT) { 2180 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2181 tp->snd_recover = tp->snd_una; 2182 } 2183 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2184 tp->snd_nxt = tp->snd_una; 2185 2186 switch (tp->t_state) { 2187 2188 /* 2189 * In FIN_WAIT_1 STATE in addition to the processing 2190 * for the ESTABLISHED state if our FIN is now acknowledged 2191 * then enter FIN_WAIT_2. 2192 */ 2193 case TCPS_FIN_WAIT_1: 2194 if (ourfinisacked) { 2195 /* 2196 * If we can't receive any more 2197 * data, then closing user can proceed. 2198 * Starting the timer is contrary to the 2199 * specification, but if we don't get a FIN 2200 * we'll hang forever. 2201 */ 2202 /* XXXjl 2203 * we should release the tp also, and use a 2204 * compressed state. 2205 */ 2206 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2207 int timeout; 2208 2209 soisdisconnected(so); 2210 timeout = (tcp_fast_finwait2_recycle) ? 2211 tcp_finwait2_timeout : tcp_maxidle; 2212 tcp_timer_activate(tp, TT_2MSL, timeout); 2213 } 2214 tp->t_state = TCPS_FIN_WAIT_2; 2215 } 2216 break; 2217 2218 /* 2219 * In CLOSING STATE in addition to the processing for 2220 * the ESTABLISHED state if the ACK acknowledges our FIN 2221 * then enter the TIME-WAIT state, otherwise ignore 2222 * the segment. 2223 */ 2224 case TCPS_CLOSING: 2225 if (ourfinisacked) { 2226 KASSERT(headlocked, ("%s: process_ACK: " 2227 "head not locked", __func__)); 2228 tcp_twstart(tp); 2229 INP_INFO_WUNLOCK(&tcbinfo); 2230 headlocked = 0; 2231 m_freem(m); 2232 return; 2233 } 2234 break; 2235 2236 /* 2237 * In LAST_ACK, we may still be waiting for data to drain 2238 * and/or to be acked, as well as for the ack of our FIN. 2239 * If our FIN is now acknowledged, delete the TCB, 2240 * enter the closed state and return. 2241 */ 2242 case TCPS_LAST_ACK: 2243 if (ourfinisacked) { 2244 KASSERT(headlocked, ("%s: process_ACK: " 2245 "tcp_close: head not locked", __func__)); 2246 tp = tcp_close(tp); 2247 goto drop; 2248 } 2249 break; 2250 } 2251 } 2252 2253step6: 2254 KASSERT(headlocked, ("%s: step6: head not locked", __func__)); 2255 INP_LOCK_ASSERT(tp->t_inpcb); 2256 2257 /* 2258 * Update window information. 2259 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2260 */ 2261 if ((thflags & TH_ACK) && 2262 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2263 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2264 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2265 /* keep track of pure window updates */ 2266 if (tlen == 0 && 2267 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2268 tcpstat.tcps_rcvwinupd++; 2269 tp->snd_wnd = tiwin; 2270 tp->snd_wl1 = th->th_seq; 2271 tp->snd_wl2 = th->th_ack; 2272 if (tp->snd_wnd > tp->max_sndwnd) 2273 tp->max_sndwnd = tp->snd_wnd; 2274 needoutput = 1; 2275 } 2276 2277 /* 2278 * Process segments with URG. 2279 */ 2280 if ((thflags & TH_URG) && th->th_urp && 2281 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2282 /* 2283 * This is a kludge, but if we receive and accept 2284 * random urgent pointers, we'll crash in 2285 * soreceive. It's hard to imagine someone 2286 * actually wanting to send this much urgent data. 2287 */ 2288 SOCKBUF_LOCK(&so->so_rcv); 2289 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2290 th->th_urp = 0; /* XXX */ 2291 thflags &= ~TH_URG; /* XXX */ 2292 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2293 goto dodata; /* XXX */ 2294 } 2295 /* 2296 * If this segment advances the known urgent pointer, 2297 * then mark the data stream. This should not happen 2298 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2299 * a FIN has been received from the remote side. 2300 * In these states we ignore the URG. 2301 * 2302 * According to RFC961 (Assigned Protocols), 2303 * the urgent pointer points to the last octet 2304 * of urgent data. We continue, however, 2305 * to consider it to indicate the first octet 2306 * of data past the urgent section as the original 2307 * spec states (in one of two places). 2308 */ 2309 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2310 tp->rcv_up = th->th_seq + th->th_urp; 2311 so->so_oobmark = so->so_rcv.sb_cc + 2312 (tp->rcv_up - tp->rcv_nxt) - 1; 2313 if (so->so_oobmark == 0) 2314 so->so_rcv.sb_state |= SBS_RCVATMARK; 2315 sohasoutofband(so); 2316 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2317 } 2318 SOCKBUF_UNLOCK(&so->so_rcv); 2319 /* 2320 * Remove out of band data so doesn't get presented to user. 2321 * This can happen independent of advancing the URG pointer, 2322 * but if two URG's are pending at once, some out-of-band 2323 * data may creep in... ick. 2324 */ 2325 if (th->th_urp <= (u_long)tlen && 2326 !(so->so_options & SO_OOBINLINE)) { 2327 /* hdr drop is delayed */ 2328 tcp_pulloutofband(so, th, m, drop_hdrlen); 2329 } 2330 } else { 2331 /* 2332 * If no out of band data is expected, 2333 * pull receive urgent pointer along 2334 * with the receive window. 2335 */ 2336 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2337 tp->rcv_up = tp->rcv_nxt; 2338 } 2339dodata: /* XXX */ 2340 KASSERT(headlocked, ("%s: dodata: head not locked", __func__)); 2341 INP_LOCK_ASSERT(tp->t_inpcb); 2342 2343 /* 2344 * Process the segment text, merging it into the TCP sequencing queue, 2345 * and arranging for acknowledgment of receipt if necessary. 2346 * This process logically involves adjusting tp->rcv_wnd as data 2347 * is presented to the user (this happens in tcp_usrreq.c, 2348 * case PRU_RCVD). If a FIN has already been received on this 2349 * connection then we just ignore the text. 2350 */ 2351 if ((tlen || (thflags & TH_FIN)) && 2352 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2353 tcp_seq save_start = th->th_seq; 2354 tcp_seq save_end = th->th_seq + tlen; 2355 m_adj(m, drop_hdrlen); /* delayed header drop */ 2356 /* 2357 * Insert segment which includes th into TCP reassembly queue 2358 * with control block tp. Set thflags to whether reassembly now 2359 * includes a segment with FIN. This handles the common case 2360 * inline (segment is the next to be received on an established 2361 * connection, and the queue is empty), avoiding linkage into 2362 * and removal from the queue and repetition of various 2363 * conversions. 2364 * Set DELACK for segments received in order, but ack 2365 * immediately when segments are out of order (so 2366 * fast retransmit can work). 2367 */ 2368 if (th->th_seq == tp->rcv_nxt && 2369 LIST_EMPTY(&tp->t_segq) && 2370 TCPS_HAVEESTABLISHED(tp->t_state)) { 2371 if (DELAY_ACK(tp)) 2372 tp->t_flags |= TF_DELACK; 2373 else 2374 tp->t_flags |= TF_ACKNOW; 2375 tp->rcv_nxt += tlen; 2376 thflags = th->th_flags & TH_FIN; 2377 tcpstat.tcps_rcvpack++; 2378 tcpstat.tcps_rcvbyte += tlen; 2379 ND6_HINT(tp); 2380 SOCKBUF_LOCK(&so->so_rcv); 2381 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2382 m_freem(m); 2383 else 2384 sbappendstream_locked(&so->so_rcv, m); 2385 sorwakeup_locked(so); 2386 } else { 2387 thflags = tcp_reass(tp, th, &tlen, m); 2388 tp->t_flags |= TF_ACKNOW; 2389 } 2390 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2391 tcp_update_sack_list(tp, save_start, save_end); 2392#if 0 2393 /* 2394 * Note the amount of data that peer has sent into 2395 * our window, in order to estimate the sender's 2396 * buffer size. 2397 * XXX: Unused. 2398 */ 2399 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2400#endif 2401 } else { 2402 m_freem(m); 2403 thflags &= ~TH_FIN; 2404 } 2405 2406 /* 2407 * If FIN is received ACK the FIN and let the user know 2408 * that the connection is closing. 2409 */ 2410 if (thflags & TH_FIN) { 2411 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2412 socantrcvmore(so); 2413 /* 2414 * If connection is half-synchronized 2415 * (ie NEEDSYN flag on) then delay ACK, 2416 * so it may be piggybacked when SYN is sent. 2417 * Otherwise, since we received a FIN then no 2418 * more input can be expected, send ACK now. 2419 */ 2420 if (tp->t_flags & TF_NEEDSYN) 2421 tp->t_flags |= TF_DELACK; 2422 else 2423 tp->t_flags |= TF_ACKNOW; 2424 tp->rcv_nxt++; 2425 } 2426 switch (tp->t_state) { 2427 2428 /* 2429 * In SYN_RECEIVED and ESTABLISHED STATES 2430 * enter the CLOSE_WAIT state. 2431 */ 2432 case TCPS_SYN_RECEIVED: 2433 tp->t_starttime = ticks; 2434 /*FALLTHROUGH*/ 2435 case TCPS_ESTABLISHED: 2436 tp->t_state = TCPS_CLOSE_WAIT; 2437 break; 2438 2439 /* 2440 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2441 * enter the CLOSING state. 2442 */ 2443 case TCPS_FIN_WAIT_1: 2444 tp->t_state = TCPS_CLOSING; 2445 break; 2446 2447 /* 2448 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2449 * starting the time-wait timer, turning off the other 2450 * standard timers. 2451 */ 2452 case TCPS_FIN_WAIT_2: 2453 KASSERT(headlocked == 1, ("%s: dodata: " 2454 "TCP_FIN_WAIT_2: head not locked", __func__)); 2455 tcp_twstart(tp); 2456 INP_INFO_WUNLOCK(&tcbinfo); 2457 return; 2458 } 2459 } 2460 INP_INFO_WUNLOCK(&tcbinfo); 2461 headlocked = 0; 2462#ifdef TCPDEBUG 2463 if (so->so_options & SO_DEBUG) 2464 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2465 &tcp_savetcp, 0); 2466#endif 2467 2468 /* 2469 * Return any desired output. 2470 */ 2471 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2472 (void) tcp_output(tp); 2473 2474check_delack: 2475 KASSERT(headlocked == 0, ("%s: check_delack: head locked", 2476 __func__)); 2477 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 2478 INP_LOCK_ASSERT(tp->t_inpcb); 2479 if (tp->t_flags & TF_DELACK) { 2480 tp->t_flags &= ~TF_DELACK; 2481 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2482 } 2483 INP_UNLOCK(tp->t_inpcb); 2484 return; 2485 2486dropafterack: 2487 KASSERT(headlocked, ("%s: dropafterack: head not locked", __func__)); 2488 /* 2489 * Generate an ACK dropping incoming segment if it occupies 2490 * sequence space, where the ACK reflects our state. 2491 * 2492 * We can now skip the test for the RST flag since all 2493 * paths to this code happen after packets containing 2494 * RST have been dropped. 2495 * 2496 * In the SYN-RECEIVED state, don't send an ACK unless the 2497 * segment we received passes the SYN-RECEIVED ACK test. 2498 * If it fails send a RST. This breaks the loop in the 2499 * "LAND" DoS attack, and also prevents an ACK storm 2500 * between two listening ports that have been sent forged 2501 * SYN segments, each with the source address of the other. 2502 */ 2503 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2504 (SEQ_GT(tp->snd_una, th->th_ack) || 2505 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2506 rstreason = BANDLIM_RST_OPENPORT; 2507 goto dropwithreset; 2508 } 2509#ifdef TCPDEBUG 2510 if (so->so_options & SO_DEBUG) 2511 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2512 &tcp_savetcp, 0); 2513#endif 2514 KASSERT(headlocked, ("%s: headlocked should be 1", __func__)); 2515 INP_INFO_WUNLOCK(&tcbinfo); 2516 tp->t_flags |= TF_ACKNOW; 2517 (void) tcp_output(tp); 2518 INP_UNLOCK(tp->t_inpcb); 2519 m_freem(m); 2520 return; 2521 2522dropwithreset: 2523 KASSERT(headlocked, ("%s: dropwithreset: head not locked", __func__)); 2524 2525 tcp_dropwithreset(m, th, tp, tlen, rstreason); 2526 2527 if (tp != NULL) 2528 INP_UNLOCK(tp->t_inpcb); 2529 if (headlocked) 2530 INP_INFO_WUNLOCK(&tcbinfo); 2531 return; 2532 2533drop: 2534 /* 2535 * Drop space held by incoming segment and return. 2536 */ 2537#ifdef TCPDEBUG 2538 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2539 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2540 &tcp_savetcp, 0); 2541#endif 2542 if (tp != NULL) 2543 INP_UNLOCK(tp->t_inpcb); 2544 if (headlocked) 2545 INP_INFO_WUNLOCK(&tcbinfo); 2546 m_freem(m); 2547 return; 2548} 2549 2550/* 2551 * Issue RST and make ACK acceptable to originator of segment. 2552 * The mbuf must still include the original packet header. 2553 * tp may be NULL. 2554 */ 2555static void 2556tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 2557 int tlen, int rstreason) 2558{ 2559 struct ip *ip; 2560#ifdef INET6 2561 struct ip6_hdr *ip6; 2562#endif 2563 /* Don't bother if destination was broadcast/multicast. */ 2564 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2565 goto drop; 2566#ifdef INET6 2567 if (mtod(m, struct ip *)->ip_v == 6) { 2568 ip6 = mtod(m, struct ip6_hdr *); 2569 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2570 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2571 goto drop; 2572 /* IPv6 anycast check is done at tcp6_input() */ 2573 } else 2574#endif 2575 { 2576 ip = mtod(m, struct ip *); 2577 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2578 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2579 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2580 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2581 goto drop; 2582 } 2583 2584 /* Perform bandwidth limiting. */ 2585 if (badport_bandlim(rstreason) < 0) 2586 goto drop; 2587 2588 /* tcp_respond consumes the mbuf chain. */ 2589 if (th->th_flags & TH_ACK) { 2590 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 2591 th->th_ack, TH_RST); 2592 } else { 2593 if (th->th_flags & TH_SYN) 2594 tlen++; 2595 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2596 (tcp_seq)0, TH_RST|TH_ACK); 2597 } 2598 return; 2599drop: 2600 m_freem(m); 2601 return; 2602} 2603 2604/* 2605 * Parse TCP options and place in tcpopt. 2606 */ 2607static void 2608tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 2609{ 2610 int opt, optlen; 2611 2612 to->to_flags = 0; 2613 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2614 opt = cp[0]; 2615 if (opt == TCPOPT_EOL) 2616 break; 2617 if (opt == TCPOPT_NOP) 2618 optlen = 1; 2619 else { 2620 if (cnt < 2) 2621 break; 2622 optlen = cp[1]; 2623 if (optlen < 2 || optlen > cnt) 2624 break; 2625 } 2626 switch (opt) { 2627 case TCPOPT_MAXSEG: 2628 if (optlen != TCPOLEN_MAXSEG) 2629 continue; 2630 if (!(flags & TO_SYN)) 2631 continue; 2632 to->to_flags |= TOF_MSS; 2633 bcopy((char *)cp + 2, 2634 (char *)&to->to_mss, sizeof(to->to_mss)); 2635 to->to_mss = ntohs(to->to_mss); 2636 break; 2637 case TCPOPT_WINDOW: 2638 if (optlen != TCPOLEN_WINDOW) 2639 continue; 2640 if (!(flags & TO_SYN)) 2641 continue; 2642 to->to_flags |= TOF_SCALE; 2643 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 2644 break; 2645 case TCPOPT_TIMESTAMP: 2646 if (optlen != TCPOLEN_TIMESTAMP) 2647 continue; 2648 to->to_flags |= TOF_TS; 2649 bcopy((char *)cp + 2, 2650 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2651 to->to_tsval = ntohl(to->to_tsval); 2652 bcopy((char *)cp + 6, 2653 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2654 to->to_tsecr = ntohl(to->to_tsecr); 2655 break; 2656#ifdef TCP_SIGNATURE 2657 /* 2658 * XXX In order to reply to a host which has set the 2659 * TCP_SIGNATURE option in its initial SYN, we have to 2660 * record the fact that the option was observed here 2661 * for the syncache code to perform the correct response. 2662 */ 2663 case TCPOPT_SIGNATURE: 2664 if (optlen != TCPOLEN_SIGNATURE) 2665 continue; 2666 to->to_flags |= TOF_SIGNATURE; 2667 to->to_signature = cp + 2; 2668 break; 2669#endif 2670 case TCPOPT_SACK_PERMITTED: 2671 if (optlen != TCPOLEN_SACK_PERMITTED) 2672 continue; 2673 if (!(flags & TO_SYN)) 2674 continue; 2675 if (!tcp_do_sack) 2676 continue; 2677 to->to_flags |= TOF_SACKPERM; 2678 break; 2679 case TCPOPT_SACK: 2680 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2681 continue; 2682 if (flags & TO_SYN) 2683 continue; 2684 to->to_flags |= TOF_SACK; 2685 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2686 to->to_sacks = cp + 2; 2687 tcpstat.tcps_sack_rcv_blocks++; 2688 break; 2689 default: 2690 continue; 2691 } 2692 } 2693} 2694 2695/* 2696 * Pull out of band byte out of a segment so 2697 * it doesn't appear in the user's data queue. 2698 * It is still reflected in the segment length for 2699 * sequencing purposes. 2700 */ 2701static void 2702tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 2703 int off) 2704{ 2705 int cnt = off + th->th_urp - 1; 2706 2707 while (cnt >= 0) { 2708 if (m->m_len > cnt) { 2709 char *cp = mtod(m, caddr_t) + cnt; 2710 struct tcpcb *tp = sototcpcb(so); 2711 2712 tp->t_iobc = *cp; 2713 tp->t_oobflags |= TCPOOB_HAVEDATA; 2714 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2715 m->m_len--; 2716 if (m->m_flags & M_PKTHDR) 2717 m->m_pkthdr.len--; 2718 return; 2719 } 2720 cnt -= m->m_len; 2721 m = m->m_next; 2722 if (m == NULL) 2723 break; 2724 } 2725 panic("tcp_pulloutofband"); 2726} 2727 2728/* 2729 * Collect new round-trip time estimate 2730 * and update averages and current timeout. 2731 */ 2732static void 2733tcp_xmit_timer(struct tcpcb *tp, int rtt) 2734{ 2735 int delta; 2736 2737 INP_LOCK_ASSERT(tp->t_inpcb); 2738 2739 tcpstat.tcps_rttupdated++; 2740 tp->t_rttupdated++; 2741 if (tp->t_srtt != 0) { 2742 /* 2743 * srtt is stored as fixed point with 5 bits after the 2744 * binary point (i.e., scaled by 8). The following magic 2745 * is equivalent to the smoothing algorithm in rfc793 with 2746 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2747 * point). Adjust rtt to origin 0. 2748 */ 2749 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2750 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2751 2752 if ((tp->t_srtt += delta) <= 0) 2753 tp->t_srtt = 1; 2754 2755 /* 2756 * We accumulate a smoothed rtt variance (actually, a 2757 * smoothed mean difference), then set the retransmit 2758 * timer to smoothed rtt + 4 times the smoothed variance. 2759 * rttvar is stored as fixed point with 4 bits after the 2760 * binary point (scaled by 16). The following is 2761 * equivalent to rfc793 smoothing with an alpha of .75 2762 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2763 * rfc793's wired-in beta. 2764 */ 2765 if (delta < 0) 2766 delta = -delta; 2767 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2768 if ((tp->t_rttvar += delta) <= 0) 2769 tp->t_rttvar = 1; 2770 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2771 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2772 } else { 2773 /* 2774 * No rtt measurement yet - use the unsmoothed rtt. 2775 * Set the variance to half the rtt (so our first 2776 * retransmit happens at 3*rtt). 2777 */ 2778 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2779 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2780 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2781 } 2782 tp->t_rtttime = 0; 2783 tp->t_rxtshift = 0; 2784 2785 /* 2786 * the retransmit should happen at rtt + 4 * rttvar. 2787 * Because of the way we do the smoothing, srtt and rttvar 2788 * will each average +1/2 tick of bias. When we compute 2789 * the retransmit timer, we want 1/2 tick of rounding and 2790 * 1 extra tick because of +-1/2 tick uncertainty in the 2791 * firing of the timer. The bias will give us exactly the 2792 * 1.5 tick we need. But, because the bias is 2793 * statistical, we have to test that we don't drop below 2794 * the minimum feasible timer (which is 2 ticks). 2795 */ 2796 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2797 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2798 2799 /* 2800 * We received an ack for a packet that wasn't retransmitted; 2801 * it is probably safe to discard any error indications we've 2802 * received recently. This isn't quite right, but close enough 2803 * for now (a route might have failed after we sent a segment, 2804 * and the return path might not be symmetrical). 2805 */ 2806 tp->t_softerror = 0; 2807} 2808 2809/* 2810 * Determine a reasonable value for maxseg size. 2811 * If the route is known, check route for mtu. 2812 * If none, use an mss that can be handled on the outgoing 2813 * interface without forcing IP to fragment; if bigger than 2814 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2815 * to utilize large mbufs. If no route is found, route has no mtu, 2816 * or the destination isn't local, use a default, hopefully conservative 2817 * size (usually 512 or the default IP max size, but no more than the mtu 2818 * of the interface), as we can't discover anything about intervening 2819 * gateways or networks. We also initialize the congestion/slow start 2820 * window to be a single segment if the destination isn't local. 2821 * While looking at the routing entry, we also initialize other path-dependent 2822 * parameters from pre-set or cached values in the routing entry. 2823 * 2824 * Also take into account the space needed for options that we 2825 * send regularly. Make maxseg shorter by that amount to assure 2826 * that we can send maxseg amount of data even when the options 2827 * are present. Store the upper limit of the length of options plus 2828 * data in maxopd. 2829 * 2830 * 2831 * In case of T/TCP, we call this routine during implicit connection 2832 * setup as well (offer = -1), to initialize maxseg from the cached 2833 * MSS of our peer. 2834 * 2835 * NOTE that this routine is only called when we process an incoming 2836 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2837 */ 2838void 2839tcp_mss(struct tcpcb *tp, int offer) 2840{ 2841 int rtt, mss; 2842 u_long bufsize; 2843 u_long maxmtu; 2844 struct inpcb *inp = tp->t_inpcb; 2845 struct socket *so; 2846 struct hc_metrics_lite metrics; 2847 int origoffer = offer; 2848 int mtuflags = 0; 2849#ifdef INET6 2850 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2851 size_t min_protoh = isipv6 ? 2852 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2853 sizeof (struct tcpiphdr); 2854#else 2855 const size_t min_protoh = sizeof(struct tcpiphdr); 2856#endif 2857 2858 /* initialize */ 2859#ifdef INET6 2860 if (isipv6) { 2861 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags); 2862 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2863 } else 2864#endif 2865 { 2866 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags); 2867 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2868 } 2869 so = inp->inp_socket; 2870 2871 /* 2872 * no route to sender, stay with default mss and return 2873 */ 2874 if (maxmtu == 0) 2875 return; 2876 2877 /* what have we got? */ 2878 switch (offer) { 2879 case 0: 2880 /* 2881 * Offer == 0 means that there was no MSS on the SYN 2882 * segment, in this case we use tcp_mssdflt. 2883 */ 2884 offer = 2885#ifdef INET6 2886 isipv6 ? tcp_v6mssdflt : 2887#endif 2888 tcp_mssdflt; 2889 break; 2890 2891 case -1: 2892 /* 2893 * Offer == -1 means that we didn't receive SYN yet. 2894 */ 2895 /* FALLTHROUGH */ 2896 2897 default: 2898 /* 2899 * Prevent DoS attack with too small MSS. Round up 2900 * to at least minmss. 2901 */ 2902 offer = max(offer, tcp_minmss); 2903 /* 2904 * Sanity check: make sure that maxopd will be large 2905 * enough to allow some data on segments even if the 2906 * all the option space is used (40bytes). Otherwise 2907 * funny things may happen in tcp_output. 2908 */ 2909 offer = max(offer, 64); 2910 } 2911 2912 /* 2913 * rmx information is now retrieved from tcp_hostcache 2914 */ 2915 tcp_hc_get(&inp->inp_inc, &metrics); 2916 2917 /* 2918 * if there's a discovered mtu int tcp hostcache, use it 2919 * else, use the link mtu. 2920 */ 2921 if (metrics.rmx_mtu) 2922 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2923 else { 2924#ifdef INET6 2925 if (isipv6) { 2926 mss = maxmtu - min_protoh; 2927 if (!path_mtu_discovery && 2928 !in6_localaddr(&inp->in6p_faddr)) 2929 mss = min(mss, tcp_v6mssdflt); 2930 } else 2931#endif 2932 { 2933 mss = maxmtu - min_protoh; 2934 if (!path_mtu_discovery && 2935 !in_localaddr(inp->inp_faddr)) 2936 mss = min(mss, tcp_mssdflt); 2937 } 2938 } 2939 mss = min(mss, offer); 2940 2941 /* 2942 * maxopd stores the maximum length of data AND options 2943 * in a segment; maxseg is the amount of data in a normal 2944 * segment. We need to store this value (maxopd) apart 2945 * from maxseg, because now every segment carries options 2946 * and thus we normally have somewhat less data in segments. 2947 */ 2948 tp->t_maxopd = mss; 2949 2950 /* 2951 * origoffer==-1 indicates, that no segments were received yet. 2952 * In this case we just guess. 2953 */ 2954 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2955 (origoffer == -1 || 2956 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2957 mss -= TCPOLEN_TSTAMP_APPA; 2958 tp->t_maxseg = mss; 2959 2960#if (MCLBYTES & (MCLBYTES - 1)) == 0 2961 if (mss > MCLBYTES) 2962 mss &= ~(MCLBYTES-1); 2963#else 2964 if (mss > MCLBYTES) 2965 mss = mss / MCLBYTES * MCLBYTES; 2966#endif 2967 tp->t_maxseg = mss; 2968 2969 /* 2970 * If there's a pipesize, change the socket buffer to that size, 2971 * don't change if sb_hiwat is different than default (then it 2972 * has been changed on purpose with setsockopt). 2973 * Make the socket buffers an integral number of mss units; 2974 * if the mss is larger than the socket buffer, decrease the mss. 2975 */ 2976 SOCKBUF_LOCK(&so->so_snd); 2977 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2978 bufsize = metrics.rmx_sendpipe; 2979 else 2980 bufsize = so->so_snd.sb_hiwat; 2981 if (bufsize < mss) 2982 mss = bufsize; 2983 else { 2984 bufsize = roundup(bufsize, mss); 2985 if (bufsize > sb_max) 2986 bufsize = sb_max; 2987 if (bufsize > so->so_snd.sb_hiwat) 2988 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2989 } 2990 SOCKBUF_UNLOCK(&so->so_snd); 2991 tp->t_maxseg = mss; 2992 2993 SOCKBUF_LOCK(&so->so_rcv); 2994 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2995 bufsize = metrics.rmx_recvpipe; 2996 else 2997 bufsize = so->so_rcv.sb_hiwat; 2998 if (bufsize > mss) { 2999 bufsize = roundup(bufsize, mss); 3000 if (bufsize > sb_max) 3001 bufsize = sb_max; 3002 if (bufsize > so->so_rcv.sb_hiwat) 3003 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3004 } 3005 SOCKBUF_UNLOCK(&so->so_rcv); 3006 /* 3007 * While we're here, check the others too 3008 */ 3009 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 3010 tp->t_srtt = rtt; 3011 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3012 tcpstat.tcps_usedrtt++; 3013 if (metrics.rmx_rttvar) { 3014 tp->t_rttvar = metrics.rmx_rttvar; 3015 tcpstat.tcps_usedrttvar++; 3016 } else { 3017 /* default variation is +- 1 rtt */ 3018 tp->t_rttvar = 3019 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3020 } 3021 TCPT_RANGESET(tp->t_rxtcur, 3022 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3023 tp->t_rttmin, TCPTV_REXMTMAX); 3024 } 3025 if (metrics.rmx_ssthresh) { 3026 /* 3027 * There's some sort of gateway or interface 3028 * buffer limit on the path. Use this to set 3029 * the slow start threshhold, but set the 3030 * threshold to no less than 2*mss. 3031 */ 3032 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3033 tcpstat.tcps_usedssthresh++; 3034 } 3035 if (metrics.rmx_bandwidth) 3036 tp->snd_bandwidth = metrics.rmx_bandwidth; 3037 3038 /* 3039 * Set the slow-start flight size depending on whether this 3040 * is a local network or not. 3041 * 3042 * Extend this so we cache the cwnd too and retrieve it here. 3043 * Make cwnd even bigger than RFC3390 suggests but only if we 3044 * have previous experience with the remote host. Be careful 3045 * not make cwnd bigger than remote receive window or our own 3046 * send socket buffer. Maybe put some additional upper bound 3047 * on the retrieved cwnd. Should do incremental updates to 3048 * hostcache when cwnd collapses so next connection doesn't 3049 * overloads the path again. 3050 * 3051 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3052 * We currently check only in syncache_socket for that. 3053 */ 3054#define TCP_METRICS_CWND 3055#ifdef TCP_METRICS_CWND 3056 if (metrics.rmx_cwnd) 3057 tp->snd_cwnd = max(mss, 3058 min(metrics.rmx_cwnd / 2, 3059 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3060 else 3061#endif 3062 if (tcp_do_rfc3390) 3063 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3064#ifdef INET6 3065 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3066 (!isipv6 && in_localaddr(inp->inp_faddr))) 3067#else 3068 else if (in_localaddr(inp->inp_faddr)) 3069#endif 3070 tp->snd_cwnd = mss * ss_fltsz_local; 3071 else 3072 tp->snd_cwnd = mss * ss_fltsz; 3073 3074 /* Check the interface for TSO capabilities. */ 3075 if (mtuflags & CSUM_TSO) 3076 tp->t_flags |= TF_TSO; 3077} 3078 3079/* 3080 * Determine the MSS option to send on an outgoing SYN. 3081 */ 3082int 3083tcp_mssopt(struct in_conninfo *inc) 3084{ 3085 int mss = 0; 3086 u_long maxmtu = 0; 3087 u_long thcmtu = 0; 3088 size_t min_protoh; 3089#ifdef INET6 3090 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3091#endif 3092 3093 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3094 3095#ifdef INET6 3096 if (isipv6) { 3097 mss = tcp_v6mssdflt; 3098 maxmtu = tcp_maxmtu6(inc, NULL); 3099 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3100 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3101 } else 3102#endif 3103 { 3104 mss = tcp_mssdflt; 3105 maxmtu = tcp_maxmtu(inc, NULL); 3106 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3107 min_protoh = sizeof(struct tcpiphdr); 3108 } 3109 if (maxmtu && thcmtu) 3110 mss = min(maxmtu, thcmtu) - min_protoh; 3111 else if (maxmtu || thcmtu) 3112 mss = max(maxmtu, thcmtu) - min_protoh; 3113 3114 return (mss); 3115} 3116 3117 3118/* 3119 * On a partial ack arrives, force the retransmission of the 3120 * next unacknowledged segment. Do not clear tp->t_dupacks. 3121 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3122 * be started again. 3123 */ 3124static void 3125tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3126{ 3127 tcp_seq onxt = tp->snd_nxt; 3128 u_long ocwnd = tp->snd_cwnd; 3129 3130 tcp_timer_activate(tp, TT_REXMT, 0); 3131 tp->t_rtttime = 0; 3132 tp->snd_nxt = th->th_ack; 3133 /* 3134 * Set snd_cwnd to one segment beyond acknowledged offset. 3135 * (tp->snd_una has not yet been updated when this function is called.) 3136 */ 3137 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3138 tp->t_flags |= TF_ACKNOW; 3139 (void) tcp_output(tp); 3140 tp->snd_cwnd = ocwnd; 3141 if (SEQ_GT(onxt, tp->snd_nxt)) 3142 tp->snd_nxt = onxt; 3143 /* 3144 * Partial window deflation. Relies on fact that tp->snd_una 3145 * not updated yet. 3146 */ 3147 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3148 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3149 else 3150 tp->snd_cwnd = 0; 3151 tp->snd_cwnd += tp->t_maxseg; 3152} 3153 3154/* 3155 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3156 * looking for a pcb in the listen state. Returns 0 otherwise. 3157 */ 3158static int 3159tcp_timewait(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th, 3160 struct mbuf *m, int tlen) 3161{ 3162 struct tcptw *tw; 3163 int thflags; 3164 tcp_seq seq; 3165#ifdef INET6 3166 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3167#else 3168 const int isipv6 = 0; 3169#endif 3170 3171 /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */ 3172 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3173 INP_LOCK_ASSERT(inp); 3174 3175 /* 3176 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 3177 * still present. This is undesirable, but temporarily necessary 3178 * until we work out how to handle inpcb's who's timewait state has 3179 * been removed. 3180 */ 3181 tw = intotw(inp); 3182 if (tw == NULL) 3183 goto drop; 3184 3185 thflags = th->th_flags; 3186 3187 /* 3188 * NOTE: for FIN_WAIT_2 (to be added later), 3189 * must validate sequence number before accepting RST 3190 */ 3191 3192 /* 3193 * If the segment contains RST: 3194 * Drop the segment - see Stevens, vol. 2, p. 964 and 3195 * RFC 1337. 3196 */ 3197 if (thflags & TH_RST) 3198 goto drop; 3199 3200#if 0 3201/* PAWS not needed at the moment */ 3202 /* 3203 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3204 * and it's less than ts_recent, drop it. 3205 */ 3206 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3207 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3208 if ((thflags & TH_ACK) == 0) 3209 goto drop; 3210 goto ack; 3211 } 3212 /* 3213 * ts_recent is never updated because we never accept new segments. 3214 */ 3215#endif 3216 3217 /* 3218 * If a new connection request is received 3219 * while in TIME_WAIT, drop the old connection 3220 * and start over if the sequence numbers 3221 * are above the previous ones. 3222 */ 3223 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3224 tcp_twclose(tw, 0); 3225 return (1); 3226 } 3227 3228 /* 3229 * Drop the the segment if it does not contain an ACK. 3230 */ 3231 if ((thflags & TH_ACK) == 0) 3232 goto drop; 3233 3234 /* 3235 * Reset the 2MSL timer if this is a duplicate FIN. 3236 */ 3237 if (thflags & TH_FIN) { 3238 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3239 if (seq + 1 == tw->rcv_nxt) 3240 tcp_timer_2msl_reset(tw, 1); 3241 } 3242 3243 /* 3244 * Acknowledge the segment if it has data or is not a duplicate ACK. 3245 */ 3246 if (thflags != TH_ACK || tlen != 0 || 3247 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3248 tcp_twrespond(tw, TH_ACK); 3249 goto drop; 3250 3251 /* 3252 * Generate a RST, dropping incoming segment. 3253 * Make ACK acceptable to originator of segment. 3254 * Don't bother to respond if destination was broadcast/multicast. 3255 */ 3256 if (m->m_flags & (M_BCAST|M_MCAST)) 3257 goto drop; 3258 if (isipv6) { 3259 struct ip6_hdr *ip6; 3260 3261 /* IPv6 anycast check is done at tcp6_input() */ 3262 ip6 = mtod(m, struct ip6_hdr *); 3263 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3264 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3265 goto drop; 3266 } else { 3267 struct ip *ip; 3268 3269 ip = mtod(m, struct ip *); 3270 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3271 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3272 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3273 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3274 goto drop; 3275 } 3276 if (thflags & TH_ACK) { 3277 tcp_respond(NULL, 3278 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3279 } else { 3280 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3281 tcp_respond(NULL, 3282 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3283 } 3284 INP_UNLOCK(inp); 3285 return (0); 3286 3287drop: 3288 INP_UNLOCK(inp); 3289 m_freem(m); 3290 return (0); 3291} 3292