1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_ipfw.h" /* for ipfw_fwd */ 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #include "opt_ipsec.h" 37 #include "opt_mac.h" 38 #include "opt_tcpdebug.h" 39 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> 99 100 static const int tcprexmtthresh = 3; 101 102 struct tcpstat tcpstat; 103 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 104 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 105 106 static int tcp_log_in_vain = 0; 107 SYSCTL_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"); 109 110 static int blackhole = 0; 111 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 112 &blackhole, 0, "Do not send RST on segments to closed ports"); 113 114 int tcp_delack_enabled = 1; 115 SYSCTL_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"); 118 119 static int drop_synfin = 0; 120 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 121 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 122 123 static int tcp_do_rfc3042 = 1; 124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 125 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 126 127 static int tcp_do_rfc3390 = 1; 128 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 129 &tcp_do_rfc3390, 0, 130 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 131 132 static int tcp_insecure_rst = 0; 133 SYSCTL_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"); 136 137 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 138 "TCP Segment Reassembly Queue"); 139 140 static int tcp_reass_maxseg = 0; 141 SYSCTL_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"); 144 145 int tcp_reass_qsize = 0; 146 SYSCTL_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"); 149 150 static int tcp_reass_maxqlen = 48; 151 SYSCTL_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"); 154 155 static int tcp_reass_overflows = 0; 156 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 157 &tcp_reass_overflows, 0, 158 "Global number of TCP Segment Reassembly Queue Overflows"); 159 160 int tcp_do_autorcvbuf = 1; 161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 162 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing"); 163 164 int tcp_autorcvbuf_inc = 16*1024; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 166 &tcp_autorcvbuf_inc, 0, 167 "Incrementor step size of automatic receive buffer"); 168 169 int tcp_autorcvbuf_max = 256*1024; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 171 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer"); 172 173 struct inpcbhead tcb; 174 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 175 struct inpcbinfo tcbinfo; 176 177 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 178 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 179 struct socket *, struct tcpcb *, int, int); 180 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 181 struct tcpcb *, int, int); 182 static void tcp_pulloutofband(struct socket *, 183 struct tcphdr *, struct mbuf *, int); 184 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 185 struct mbuf *); 186 static void tcp_xmit_timer(struct tcpcb *, int); 187 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 188 static 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) \ 194 do { \ 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 */ 217 static void 218 tcp_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 } 224 225 uma_zone_t tcp_reass_zone; 226 void 227 tcp_reass_init(void) 228 { 229 230 tcp_reass_maxseg = nmbclusters / 16; 231 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 232 &tcp_reass_maxseg); 233 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 234 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 235 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 236 EVENTHANDLER_REGISTER(nmbclusters_change, 237 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY); 238 } 239 240 static int 241 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m) 242 { 243 struct tseg_qent *q; 244 struct tseg_qent *p = NULL; 245 struct tseg_qent *nq; 246 struct tseg_qent *te = NULL; 247 struct socket *so = tp->t_inpcb->inp_socket; 248 int flags; 249 250 INP_LOCK_ASSERT(tp->t_inpcb); 251 252 /* 253 * XXX: tcp_reass() is rather inefficient with its data structures 254 * and should be rewritten (see NetBSD for optimizations). While 255 * doing that it should move to its own file tcp_reass.c. 256 */ 257 258 /* 259 * Call with th==NULL after become established to 260 * force pre-ESTABLISHED data up to user socket. 261 */ 262 if (th == NULL) 263 goto present; 264 265 /* 266 * Limit the number of segments in the reassembly queue to prevent 267 * holding on to too many segments (and thus running out of mbufs). 268 * Make sure to let the missing segment through which caused this 269 * queue. Always keep one global queue entry spare to be able to 270 * process the missing segment. 271 */ 272 if (th->th_seq != tp->rcv_nxt && 273 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 274 tp->t_segqlen >= tcp_reass_maxqlen)) { 275 tcp_reass_overflows++; 276 tcpstat.tcps_rcvmemdrop++; 277 m_freem(m); 278 *tlenp = 0; 279 return (0); 280 } 281 282 /* 283 * Allocate a new queue entry. If we can't, or hit the zone limit 284 * just drop the pkt. 285 */ 286 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 287 if (te == NULL) { 288 tcpstat.tcps_rcvmemdrop++; 289 m_freem(m); 290 *tlenp = 0; 291 return (0); 292 } 293 tp->t_segqlen++; 294 tcp_reass_qsize++; 295 296 /* 297 * Find a segment which begins after this one does. 298 */ 299 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 300 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 301 break; 302 p = q; 303 } 304 305 /* 306 * If there is a preceding segment, it may provide some of 307 * our data already. If so, drop the data from the incoming 308 * segment. If it provides all of our data, drop us. 309 */ 310 if (p != NULL) { 311 int i; 312 /* conversion to int (in i) handles seq wraparound */ 313 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 314 if (i > 0) { 315 if (i >= *tlenp) { 316 tcpstat.tcps_rcvduppack++; 317 tcpstat.tcps_rcvdupbyte += *tlenp; 318 m_freem(m); 319 uma_zfree(tcp_reass_zone, te); 320 tp->t_segqlen--; 321 tcp_reass_qsize--; 322 /* 323 * Try to present any queued data 324 * at the left window edge to the user. 325 * This is needed after the 3-WHS 326 * completes. 327 */ 328 goto present; /* ??? */ 329 } 330 m_adj(m, i); 331 *tlenp -= i; 332 th->th_seq += i; 333 } 334 } 335 tcpstat.tcps_rcvoopack++; 336 tcpstat.tcps_rcvoobyte += *tlenp; 337 338 /* 339 * While we overlap succeeding segments trim them or, 340 * if they are completely covered, dequeue them. 341 */ 342 while (q) { 343 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 344 if (i <= 0) 345 break; 346 if (i < q->tqe_len) { 347 q->tqe_th->th_seq += i; 348 q->tqe_len -= i; 349 m_adj(q->tqe_m, i); 350 break; 351 } 352 353 nq = LIST_NEXT(q, tqe_q); 354 LIST_REMOVE(q, tqe_q); 355 m_freem(q->tqe_m); 356 uma_zfree(tcp_reass_zone, q); 357 tp->t_segqlen--; 358 tcp_reass_qsize--; 359 q = nq; 360 } 361 362 /* Insert the new segment queue entry into place. */ 363 te->tqe_m = m; 364 te->tqe_th = th; 365 te->tqe_len = *tlenp; 366 367 if (p == NULL) { 368 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 369 } else { 370 LIST_INSERT_AFTER(p, te, tqe_q); 371 } 372 373 present: 374 /* 375 * Present data to user, advancing rcv_nxt through 376 * completed sequence space. 377 */ 378 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 379 return (0); 380 q = LIST_FIRST(&tp->t_segq); 381 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 382 return (0); 383 SOCKBUF_LOCK(&so->so_rcv); 384 do { 385 tp->rcv_nxt += q->tqe_len; 386 flags = q->tqe_th->th_flags & TH_FIN; 387 nq = LIST_NEXT(q, tqe_q); 388 LIST_REMOVE(q, tqe_q); 389 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 390 m_freem(q->tqe_m); 391 else 392 sbappendstream_locked(&so->so_rcv, q->tqe_m); 393 uma_zfree(tcp_reass_zone, q); 394 tp->t_segqlen--; 395 tcp_reass_qsize--; 396 q = nq; 397 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 398 ND6_HINT(tp); 399 sorwakeup_locked(so); 400 return (flags); 401 } 402 403 /* 404 * TCP input routine, follows pages 65-76 of the 405 * protocol specification dated September, 1981 very closely. 406 */ 407 #ifdef INET6 408 int 409 tcp6_input(struct mbuf **mp, int *offp, int proto) 410 { 411 struct mbuf *m = *mp; 412 struct in6_ifaddr *ia6; 413 414 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 415 416 /* 417 * draft-itojun-ipv6-tcp-to-anycast 418 * better place to put this in? 419 */ 420 ia6 = ip6_getdstifaddr(m); 421 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 422 struct ip6_hdr *ip6; 423 424 ip6 = mtod(m, struct ip6_hdr *); 425 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 426 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 427 return IPPROTO_DONE; 428 } 429 430 tcp_input(m, *offp); 431 return IPPROTO_DONE; 432 } 433 #endif 434 435 void 436 tcp_input(struct mbuf *m, int off0) 437 { 438 struct tcphdr *th; 439 struct ip *ip = NULL; 440 struct ipovly *ipov; 441 struct inpcb *inp = NULL; 442 struct tcpcb *tp = NULL; 443 struct socket *so = NULL; 444 u_char *optp = NULL; 445 int optlen = 0; 446 int len, tlen, off; 447 int drop_hdrlen; 448 int thflags; 449 int rstreason = 0; /* For badport_bandlim accounting purposes */ 450 #ifdef IPFIREWALL_FORWARD 451 struct m_tag *fwd_tag; 452 #endif 453 #ifdef INET6 454 struct ip6_hdr *ip6 = NULL; 455 int isipv6; 456 char ip6buf[INET6_ADDRSTRLEN]; 457 #else 458 const int isipv6 = 0; 459 #endif 460 struct tcpopt to; /* options in this segment */ 461 462 #ifdef TCPDEBUG 463 /* 464 * The size of tcp_saveipgen must be the size of the max ip header, 465 * now IPv6. 466 */ 467 u_char tcp_saveipgen[IP6_HDR_LEN]; 468 struct tcphdr tcp_savetcp; 469 short ostate = 0; 470 #endif 471 472 #ifdef INET6 473 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 474 #endif 475 476 to.to_flags = 0; 477 tcpstat.tcps_rcvtotal++; 478 479 if (isipv6) { 480 #ifdef INET6 481 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 482 ip6 = mtod(m, struct ip6_hdr *); 483 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 484 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 485 tcpstat.tcps_rcvbadsum++; 486 goto drop; 487 } 488 th = (struct tcphdr *)((caddr_t)ip6 + off0); 489 490 /* 491 * Be proactive about unspecified IPv6 address in source. 492 * As we use all-zero to indicate unbounded/unconnected pcb, 493 * unspecified IPv6 address can be used to confuse us. 494 * 495 * Note that packets with unspecified IPv6 destination is 496 * already dropped in ip6_input. 497 */ 498 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 499 /* XXX stat */ 500 goto drop; 501 } 502 #else 503 th = NULL; /* XXX: avoid compiler warning */ 504 #endif 505 } else { 506 /* 507 * Get IP and TCP header together in first mbuf. 508 * Note: IP leaves IP header in first mbuf. 509 */ 510 if (off0 > sizeof (struct ip)) { 511 ip_stripoptions(m, (struct mbuf *)0); 512 off0 = sizeof(struct ip); 513 } 514 if (m->m_len < sizeof (struct tcpiphdr)) { 515 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 516 == NULL) { 517 tcpstat.tcps_rcvshort++; 518 return; 519 } 520 } 521 ip = mtod(m, struct ip *); 522 ipov = (struct ipovly *)ip; 523 th = (struct tcphdr *)((caddr_t)ip + off0); 524 tlen = ip->ip_len; 525 526 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 527 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 528 th->th_sum = m->m_pkthdr.csum_data; 529 else 530 th->th_sum = in_pseudo(ip->ip_src.s_addr, 531 ip->ip_dst.s_addr, 532 htonl(m->m_pkthdr.csum_data + 533 ip->ip_len + 534 IPPROTO_TCP)); 535 th->th_sum ^= 0xffff; 536 #ifdef TCPDEBUG 537 ipov->ih_len = (u_short)tlen; 538 ipov->ih_len = htons(ipov->ih_len); 539 #endif 540 } else { 541 /* 542 * Checksum extended TCP header and data. 543 */ 544 len = sizeof (struct ip) + tlen; 545 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 546 ipov->ih_len = (u_short)tlen; 547 ipov->ih_len = htons(ipov->ih_len); 548 th->th_sum = in_cksum(m, len); 549 } 550 if (th->th_sum) { 551 tcpstat.tcps_rcvbadsum++; 552 goto drop; 553 } 554 /* Re-initialization for later version check */ 555 ip->ip_v = IPVERSION; 556 } 557 558 /* 559 * Check that TCP offset makes sense, 560 * pull out TCP options and adjust length. XXX 561 */ 562 off = th->th_off << 2; 563 if (off < sizeof (struct tcphdr) || off > tlen) { 564 tcpstat.tcps_rcvbadoff++; 565 goto drop; 566 } 567 tlen -= off; /* tlen is used instead of ti->ti_len */ 568 if (off > sizeof (struct tcphdr)) { 569 if (isipv6) { 570 #ifdef INET6 571 IP6_EXTHDR_CHECK(m, off0, off, ); 572 ip6 = mtod(m, struct ip6_hdr *); 573 th = (struct tcphdr *)((caddr_t)ip6 + off0); 574 #endif 575 } else { 576 if (m->m_len < sizeof(struct ip) + off) { 577 if ((m = m_pullup(m, sizeof (struct ip) + off)) 578 == NULL) { 579 tcpstat.tcps_rcvshort++; 580 return; 581 } 582 ip = mtod(m, struct ip *); 583 ipov = (struct ipovly *)ip; 584 th = (struct tcphdr *)((caddr_t)ip + off0); 585 } 586 } 587 optlen = off - sizeof (struct tcphdr); 588 optp = (u_char *)(th + 1); 589 } 590 thflags = th->th_flags; 591 592 /* 593 * If the drop_synfin option is enabled, drop all packets with 594 * both the SYN and FIN bits set. This prevents e.g. nmap from 595 * identifying the TCP/IP stack. 596 * 597 * This is a violation of the TCP specification. 598 */ 599 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 600 goto drop; 601 602 /* 603 * Convert TCP protocol specific fields to host format. 604 */ 605 th->th_seq = ntohl(th->th_seq); 606 th->th_ack = ntohl(th->th_ack); 607 th->th_win = ntohs(th->th_win); 608 th->th_urp = ntohs(th->th_urp); 609 610 /* 611 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 612 */ 613 drop_hdrlen = off0 + off; 614 615 /* 616 * Locate pcb for segment. 617 */ 618 INP_INFO_WLOCK(&tcbinfo); 619 findpcb: 620 INP_INFO_WLOCK_ASSERT(&tcbinfo); 621 #ifdef IPFIREWALL_FORWARD 622 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 623 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 624 625 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 626 struct sockaddr_in *next_hop; 627 628 next_hop = (struct sockaddr_in *)(fwd_tag+1); 629 /* 630 * Transparently forwarded. Pretend to be the destination. 631 * already got one like this? 632 */ 633 inp = in_pcblookup_hash(&tcbinfo, 634 ip->ip_src, th->th_sport, 635 ip->ip_dst, th->th_dport, 636 0, m->m_pkthdr.rcvif); 637 if (!inp) { 638 /* It's new. Try to find the ambushing socket. */ 639 inp = in_pcblookup_hash(&tcbinfo, 640 ip->ip_src, th->th_sport, 641 next_hop->sin_addr, 642 next_hop->sin_port ? 643 ntohs(next_hop->sin_port) : 644 th->th_dport, 645 INPLOOKUP_WILDCARD, 646 m->m_pkthdr.rcvif); 647 } 648 /* Remove the tag from the packet. We don't need it anymore. */ 649 m_tag_delete(m, fwd_tag); 650 } else 651 #endif /* IPFIREWALL_FORWARD */ 652 { 653 if (isipv6) { 654 #ifdef INET6 655 inp = in6_pcblookup_hash(&tcbinfo, 656 &ip6->ip6_src, th->th_sport, 657 &ip6->ip6_dst, th->th_dport, 658 INPLOOKUP_WILDCARD, 659 m->m_pkthdr.rcvif); 660 #endif 661 } else 662 inp = in_pcblookup_hash(&tcbinfo, 663 ip->ip_src, th->th_sport, 664 ip->ip_dst, th->th_dport, 665 INPLOOKUP_WILDCARD, 666 m->m_pkthdr.rcvif); 667 } 668 669 #if defined(IPSEC) || defined(FAST_IPSEC) 670 #ifdef INET6 671 if (isipv6 && inp != NULL && ipsec6_in_reject(m, inp)) { 672 #ifdef IPSEC 673 ipsec6stat.in_polvio++; 674 #endif 675 goto dropunlock; 676 } else 677 #endif /* INET6 */ 678 if (inp != NULL && ipsec4_in_reject(m, inp)) { 679 #ifdef IPSEC 680 ipsecstat.in_polvio++; 681 #endif 682 goto dropunlock; 683 } 684 #endif /*IPSEC || FAST_IPSEC*/ 685 686 /* 687 * If the INPCB does not exist then all data in the incoming 688 * segment is discarded and an appropriate RST is sent back. 689 */ 690 if (inp == NULL) { 691 /* 692 * Log communication attempts to ports that are not 693 * in use. 694 */ 695 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 696 tcp_log_in_vain == 2) { 697 #ifndef INET6 698 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 699 #else 700 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 701 if (isipv6) { 702 strcpy(dbuf, "["); 703 strcat(dbuf, 704 ip6_sprintf(ip6buf, &ip6->ip6_dst)); 705 strcat(dbuf, "]"); 706 strcpy(sbuf, "["); 707 strcat(sbuf, 708 ip6_sprintf(ip6buf, &ip6->ip6_src)); 709 strcat(sbuf, "]"); 710 } else 711 #endif /* INET6 */ 712 { 713 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 714 strcpy(sbuf, inet_ntoa(ip->ip_src)); 715 } 716 log(LOG_INFO, 717 "Connection attempt to TCP %s:%d " 718 "from %s:%d flags:0x%02x\n", 719 dbuf, ntohs(th->th_dport), sbuf, 720 ntohs(th->th_sport), thflags); 721 } 722 /* 723 * When blackholing do not respond with a RST but 724 * completely ignore the segment and drop it. 725 */ 726 if ((blackhole == 1 && (thflags & TH_SYN)) || 727 blackhole == 2) 728 goto dropunlock; 729 730 rstreason = BANDLIM_RST_CLOSEDPORT; 731 goto dropwithreset; 732 } 733 INP_LOCK(inp); 734 735 /* Check the minimum TTL for socket. */ 736 if (inp->inp_ip_minttl != 0) { 737 #ifdef INET6 738 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 739 goto dropunlock; 740 else 741 #endif 742 if (inp->inp_ip_minttl > ip->ip_ttl) 743 goto dropunlock; 744 } 745 746 /* 747 * A previous connection in TIMEWAIT state is supposed to catch 748 * stray or duplicate segments arriving late. If this segment 749 * was a legitimate new connection attempt the old INPCB gets 750 * removed and we can try again to find a listening socket. 751 */ 752 if (inp->inp_vflag & INP_TIMEWAIT) { 753 if (thflags & TH_SYN) 754 tcp_dooptions(&to, optp, optlen, TO_SYN); 755 /* NB: tcp_timewait unlocks the INP and frees the mbuf. */ 756 if (tcp_timewait(inp, &to, th, m, tlen)) 757 goto findpcb; 758 INP_INFO_WUNLOCK(&tcbinfo); 759 return; 760 } 761 /* 762 * The TCPCB may no longer exist if the connection is winding 763 * down or it is in the CLOSED state. Either way we drop the 764 * segment and send an appropriate response. 765 */ 766 tp = intotcpcb(inp); 767 if (tp == NULL) { 768 rstreason = BANDLIM_RST_CLOSEDPORT; 769 goto dropwithreset; 770 } 771 if (tp->t_state == TCPS_CLOSED) 772 goto dropunlock; /* XXX: dropwithreset??? */ 773 774 #ifdef MAC 775 INP_LOCK_ASSERT(inp); 776 if (mac_check_inpcb_deliver(inp, m)) 777 goto dropunlock; 778 #endif 779 so = inp->inp_socket; 780 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 781 #ifdef TCPDEBUG 782 if (so->so_options & SO_DEBUG) { 783 ostate = tp->t_state; 784 if (isipv6) { 785 #ifdef INET6 786 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 787 #endif 788 } else 789 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 790 tcp_savetcp = *th; 791 } 792 #endif 793 /* 794 * When the socket is accepting connections (the INPCB is in LISTEN 795 * state) we look into the SYN cache if this is a new connection 796 * attempt or the completion of a previous one. 797 */ 798 if (so->so_options & SO_ACCEPTCONN) { 799 struct in_conninfo inc; 800 801 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 802 "tp not listening", __func__)); 803 804 bzero(&inc, sizeof(inc)); 805 inc.inc_isipv6 = isipv6; 806 #ifdef INET6 807 if (isipv6) { 808 inc.inc6_faddr = ip6->ip6_src; 809 inc.inc6_laddr = ip6->ip6_dst; 810 } else 811 #endif 812 { 813 inc.inc_faddr = ip->ip_src; 814 inc.inc_laddr = ip->ip_dst; 815 } 816 inc.inc_fport = th->th_sport; 817 inc.inc_lport = th->th_dport; 818 819 /* 820 * If the state is LISTEN then ignore segment if it contains 821 * a RST. If the segment contains an ACK then it is bad and 822 * send a RST. If it does not contain a SYN then it is not 823 * interesting; drop it. 824 * 825 * If the state is SYN_RECEIVED (syncache) and seg contains 826 * an ACK, but not for our SYN/ACK, send a RST. If the seg 827 * contains a RST, check the sequence number to see if it 828 * is a valid reset segment. 829 */ 830 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 831 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 832 /* 833 * Parse the TCP options here because 834 * syncookies need access to the reflected 835 * timestamp. 836 */ 837 tcp_dooptions(&to, optp, optlen, 0); 838 /* 839 * NB: syncache_expand() doesn't unlock 840 * inp and tcpinfo locks. 841 */ 842 if (!syncache_expand(&inc, &to, th, &so, m)) { 843 /* 844 * No syncache entry or ACK was not 845 * for our SYN/ACK. Send a RST. 846 */ 847 rstreason = BANDLIM_RST_OPENPORT; 848 goto dropwithreset; 849 } 850 if (so == NULL) { 851 /* 852 * We completed the 3-way handshake 853 * but could not allocate a socket 854 * either due to memory shortage, 855 * listen queue length limits or 856 * global socket limits. 857 */ 858 rstreason = BANDLIM_UNLIMITED; 859 goto dropwithreset; 860 } 861 /* 862 * Socket is created in state SYN_RECEIVED. 863 * Continue processing segment. 864 */ 865 INP_UNLOCK(inp); /* listen socket */ 866 inp = sotoinpcb(so); 867 INP_LOCK(inp); /* new connection */ 868 tp = intotcpcb(inp); 869 /* 870 * Process the segment and the data it 871 * contains. tcp_do_segment() consumes 872 * the mbuf chain and unlocks the inpcb. 873 */ 874 tcp_do_segment(m, th, so, tp, drop_hdrlen, 875 tlen); 876 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 877 return; 878 } 879 if (thflags & TH_RST) { 880 syncache_chkrst(&inc, th); 881 goto dropunlock; 882 } 883 if (thflags & TH_ACK) { 884 syncache_badack(&inc); 885 tcpstat.tcps_badsyn++; 886 rstreason = BANDLIM_RST_OPENPORT; 887 goto dropwithreset; 888 } 889 goto dropunlock; 890 } 891 892 /* 893 * Segment's flags are (SYN) or (SYN|FIN). 894 */ 895 #ifdef INET6 896 /* 897 * If deprecated address is forbidden, 898 * we do not accept SYN to deprecated interface 899 * address to prevent any new inbound connection from 900 * getting established. 901 * When we do not accept SYN, we send a TCP RST, 902 * with deprecated source address (instead of dropping 903 * it). We compromise it as it is much better for peer 904 * to send a RST, and RST will be the final packet 905 * for the exchange. 906 * 907 * If we do not forbid deprecated addresses, we accept 908 * the SYN packet. RFC2462 does not suggest dropping 909 * SYN in this case. 910 * If we decipher RFC2462 5.5.4, it says like this: 911 * 1. use of deprecated addr with existing 912 * communication is okay - "SHOULD continue to be 913 * used" 914 * 2. use of it with new communication: 915 * (2a) "SHOULD NOT be used if alternate address 916 * with sufficient scope is available" 917 * (2b) nothing mentioned otherwise. 918 * Here we fall into (2b) case as we have no choice in 919 * our source address selection - we must obey the peer. 920 * 921 * The wording in RFC2462 is confusing, and there are 922 * multiple description text for deprecated address 923 * handling - worse, they are not exactly the same. 924 * I believe 5.5.4 is the best one, so we follow 5.5.4. 925 */ 926 if (isipv6 && !ip6_use_deprecated) { 927 struct in6_ifaddr *ia6; 928 929 if ((ia6 = ip6_getdstifaddr(m)) && 930 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 931 rstreason = BANDLIM_RST_OPENPORT; 932 goto dropwithreset; 933 } 934 } 935 #endif 936 /* 937 * Basic sanity checks on incoming SYN requests: 938 * 939 * Don't bother responding if the destination was a 940 * broadcast according to RFC1122 4.2.3.10, p. 104. 941 * 942 * If it is from this socket, drop it, it must be forged. 943 * 944 * Note that it is quite possible to receive unicast 945 * link-layer packets with a broadcast IP address. Use 946 * in_broadcast() to find them. 947 */ 948 if (m->m_flags & (M_BCAST|M_MCAST)) 949 goto dropunlock; 950 if (isipv6) { 951 #ifdef INET6 952 if (th->th_dport == th->th_sport && 953 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) 954 goto dropunlock; 955 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 956 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 957 goto dropunlock; 958 #endif 959 } else { 960 if (th->th_dport == th->th_sport && 961 ip->ip_dst.s_addr == ip->ip_src.s_addr) 962 goto dropunlock; 963 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 964 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 965 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 966 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 967 goto dropunlock; 968 } 969 /* 970 * SYN appears to be valid. Create compressed TCP state 971 * for syncache. 972 */ 973 #ifdef TCPDEBUG 974 if (so->so_options & SO_DEBUG) 975 tcp_trace(TA_INPUT, ostate, tp, 976 (void *)tcp_saveipgen, &tcp_savetcp, 0); 977 #endif 978 tcp_dooptions(&to, optp, optlen, TO_SYN); 979 syncache_add(&inc, &to, th, inp, &so, m); 980 /* 981 * Entry added to syncache and mbuf consumed. 982 * Everything unlocked already by syncache_add(). 983 */ 984 return; 985 } 986 987 /* 988 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or late 989 * state. tcp_do_segment() always consumes the mbuf chain, unlocks the 990 * inpcb, and unlocks the pcbinfo. 991 */ 992 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen); 993 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 994 return; 995 996 dropwithreset: 997 INP_INFO_WLOCK_ASSERT(&tcbinfo); 998 tcp_dropwithreset(m, th, tp, tlen, rstreason); 999 m = NULL; /* mbuf chain got consumed. */ 1000 dropunlock: 1001 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1002 if (inp != NULL) 1003 INP_UNLOCK(inp); 1004 INP_INFO_WUNLOCK(&tcbinfo); 1005 drop: 1006 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 1007 if (m != NULL) 1008 m_freem(m); 1009 return; 1010 } 1011 1012 static void 1013 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1014 struct tcpcb *tp, int drop_hdrlen, int tlen) 1015 { 1016 int thflags, acked, ourfinisacked, needoutput = 0; 1017 int headlocked = 1; 1018 int rstreason, todrop, win; 1019 u_long tiwin; 1020 struct tcpopt to; 1021 1022 #ifdef TCPDEBUG 1023 /* 1024 * The size of tcp_saveipgen must be the size of the max ip header, 1025 * now IPv6. 1026 */ 1027 u_char tcp_saveipgen[IP6_HDR_LEN]; 1028 struct tcphdr tcp_savetcp; 1029 short ostate = 0; 1030 #endif 1031 thflags = th->th_flags; 1032 1033 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1034 INP_LOCK_ASSERT(tp->t_inpcb); 1035 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1036 __func__)); 1037 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1038 __func__)); 1039 1040 /* 1041 * Segment received on connection. 1042 * Reset idle time and keep-alive timer. 1043 */ 1044 tp->t_rcvtime = ticks; 1045 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1046 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1047 1048 /* 1049 * Unscale the window into a 32-bit value. 1050 * This value is bogus for the TCPS_SYN_SENT state 1051 * and is overwritten later. 1052 */ 1053 tiwin = th->th_win << tp->snd_scale; 1054 1055 /* 1056 * Parse options on any incoming segment. 1057 */ 1058 tcp_dooptions(&to, (u_char *)(th + 1), 1059 (th->th_off << 2) - sizeof(struct tcphdr), 1060 (thflags & TH_SYN) ? TO_SYN : 0); 1061 1062 /* 1063 * If echoed timestamp is later than the current time, 1064 * fall back to non RFC1323 RTT calculation. Normalize 1065 * timestamp if syncookies were used when this connection 1066 * was established. 1067 */ 1068 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1069 to.to_tsecr -= tp->ts_offset; 1070 if (TSTMP_GT(to.to_tsecr, ticks)) 1071 to.to_tsecr = 0; 1072 } 1073 1074 /* 1075 * Process options only when we get SYN/ACK back. The SYN case 1076 * for incoming connections is handled in tcp_syncache. 1077 * XXX this is traditional behavior, may need to be cleaned up. 1078 */ 1079 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1080 if ((to.to_flags & TOF_SCALE) && 1081 (tp->t_flags & TF_REQ_SCALE)) { 1082 tp->t_flags |= TF_RCVD_SCALE; 1083 tp->snd_scale = to.to_wscale; 1084 tp->snd_wnd = th->th_win << tp->snd_scale; 1085 tiwin = tp->snd_wnd; 1086 } 1087 if (to.to_flags & TOF_TS) { 1088 tp->t_flags |= TF_RCVD_TSTMP; 1089 tp->ts_recent = to.to_tsval; 1090 tp->ts_recent_age = ticks; 1091 } 1092 /* Initial send window, already scaled. */ 1093 tp->snd_wnd = th->th_win; 1094 if (to.to_flags & TOF_MSS) 1095 tcp_mss(tp, to.to_mss); 1096 if ((tp->t_flags & TF_SACK_PERMIT) && 1097 (to.to_flags & TOF_SACKPERM) == 0) 1098 tp->t_flags &= ~TF_SACK_PERMIT; 1099 } 1100 1101 /* 1102 * Header prediction: check for the two common cases 1103 * of a uni-directional data xfer. If the packet has 1104 * no control flags, is in-sequence, the window didn't 1105 * change and we're not retransmitting, it's a 1106 * candidate. If the length is zero and the ack moved 1107 * forward, we're the sender side of the xfer. Just 1108 * free the data acked & wake any higher level process 1109 * that was blocked waiting for space. If the length 1110 * is non-zero and the ack didn't move, we're the 1111 * receiver side. If we're getting packets in-order 1112 * (the reassembly queue is empty), add the data to 1113 * the socket buffer and note that we need a delayed ack. 1114 * Make sure that the hidden state-flags are also off. 1115 * Since we check for TCPS_ESTABLISHED first, it can only 1116 * be TH_NEEDSYN. 1117 */ 1118 if (tp->t_state == TCPS_ESTABLISHED && 1119 th->th_seq == tp->rcv_nxt && 1120 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1121 tp->snd_nxt == tp->snd_max && 1122 tiwin && tiwin == tp->snd_wnd && 1123 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1124 LIST_EMPTY(&tp->t_segq) && 1125 ((to.to_flags & TOF_TS) == 0 || 1126 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1127 1128 /* 1129 * If last ACK falls within this segment's sequence numbers, 1130 * record the timestamp. 1131 * NOTE that the test is modified according to the latest 1132 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1133 */ 1134 if ((to.to_flags & TOF_TS) != 0 && 1135 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1136 tp->ts_recent_age = ticks; 1137 tp->ts_recent = to.to_tsval; 1138 } 1139 1140 if (tlen == 0) { 1141 if (SEQ_GT(th->th_ack, tp->snd_una) && 1142 SEQ_LEQ(th->th_ack, tp->snd_max) && 1143 tp->snd_cwnd >= tp->snd_wnd && 1144 ((!tcp_do_newreno && 1145 !(tp->t_flags & TF_SACK_PERMIT) && 1146 tp->t_dupacks < tcprexmtthresh) || 1147 ((tcp_do_newreno || 1148 (tp->t_flags & TF_SACK_PERMIT)) && 1149 !IN_FASTRECOVERY(tp) && 1150 (to.to_flags & TOF_SACK) == 0 && 1151 TAILQ_EMPTY(&tp->snd_holes)))) { 1152 KASSERT(headlocked, 1153 ("%s: headlocked", __func__)); 1154 INP_INFO_WUNLOCK(&tcbinfo); 1155 headlocked = 0; 1156 /* 1157 * this is a pure ack for outstanding data. 1158 */ 1159 ++tcpstat.tcps_predack; 1160 /* 1161 * "bad retransmit" recovery 1162 */ 1163 if (tp->t_rxtshift == 1 && 1164 ticks < tp->t_badrxtwin) { 1165 ++tcpstat.tcps_sndrexmitbad; 1166 tp->snd_cwnd = tp->snd_cwnd_prev; 1167 tp->snd_ssthresh = 1168 tp->snd_ssthresh_prev; 1169 tp->snd_recover = tp->snd_recover_prev; 1170 if (tp->t_flags & TF_WASFRECOVERY) 1171 ENTER_FASTRECOVERY(tp); 1172 tp->snd_nxt = tp->snd_max; 1173 tp->t_badrxtwin = 0; 1174 } 1175 1176 /* 1177 * Recalculate the transmit timer / rtt. 1178 * 1179 * Some boxes send broken timestamp replies 1180 * during the SYN+ACK phase, ignore 1181 * timestamps of 0 or we could calculate a 1182 * huge RTT and blow up the retransmit timer. 1183 */ 1184 if ((to.to_flags & TOF_TS) != 0 && 1185 to.to_tsecr) { 1186 if (!tp->t_rttlow || 1187 tp->t_rttlow > ticks - to.to_tsecr) 1188 tp->t_rttlow = ticks - to.to_tsecr; 1189 tcp_xmit_timer(tp, 1190 ticks - to.to_tsecr + 1); 1191 } else if (tp->t_rtttime && 1192 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1193 if (!tp->t_rttlow || 1194 tp->t_rttlow > ticks - tp->t_rtttime) 1195 tp->t_rttlow = ticks - tp->t_rtttime; 1196 tcp_xmit_timer(tp, 1197 ticks - tp->t_rtttime); 1198 } 1199 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1200 acked = th->th_ack - tp->snd_una; 1201 tcpstat.tcps_rcvackpack++; 1202 tcpstat.tcps_rcvackbyte += acked; 1203 sbdrop(&so->so_snd, acked); 1204 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1205 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1206 tp->snd_recover = th->th_ack - 1; 1207 tp->snd_una = th->th_ack; 1208 /* 1209 * pull snd_wl2 up to prevent seq wrap relative 1210 * to th_ack. 1211 */ 1212 tp->snd_wl2 = th->th_ack; 1213 tp->t_dupacks = 0; 1214 m_freem(m); 1215 ND6_HINT(tp); /* some progress has been done */ 1216 1217 /* 1218 * If all outstanding data are acked, stop 1219 * retransmit timer, otherwise restart timer 1220 * using current (possibly backed-off) value. 1221 * If process is waiting for space, 1222 * wakeup/selwakeup/signal. If data 1223 * are ready to send, let tcp_output 1224 * decide between more output or persist. 1225 1226 #ifdef TCPDEBUG 1227 if (so->so_options & SO_DEBUG) 1228 tcp_trace(TA_INPUT, ostate, tp, 1229 (void *)tcp_saveipgen, 1230 &tcp_savetcp, 0); 1231 #endif 1232 */ 1233 if (tp->snd_una == tp->snd_max) 1234 tcp_timer_activate(tp, TT_REXMT, 0); 1235 else if (!tcp_timer_active(tp, TT_PERSIST)) 1236 tcp_timer_activate(tp, TT_REXMT, 1237 tp->t_rxtcur); 1238 1239 sowwakeup(so); 1240 if (so->so_snd.sb_cc) 1241 (void) tcp_output(tp); 1242 goto check_delack; 1243 } 1244 } else if (th->th_ack == tp->snd_una && 1245 tlen <= sbspace(&so->so_rcv)) { 1246 int newsize = 0; /* automatic sockbuf scaling */ 1247 1248 KASSERT(headlocked, ("%s: headlocked", __func__)); 1249 INP_INFO_WUNLOCK(&tcbinfo); 1250 headlocked = 0; 1251 /* 1252 * this is a pure, in-sequence data packet 1253 * with nothing on the reassembly queue and 1254 * we have enough buffer space to take it. 1255 */ 1256 /* Clean receiver SACK report if present */ 1257 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1258 tcp_clean_sackreport(tp); 1259 ++tcpstat.tcps_preddat; 1260 tp->rcv_nxt += tlen; 1261 /* 1262 * Pull snd_wl1 up to prevent seq wrap relative to 1263 * th_seq. 1264 */ 1265 tp->snd_wl1 = th->th_seq; 1266 /* 1267 * Pull rcv_up up to prevent seq wrap relative to 1268 * rcv_nxt. 1269 */ 1270 tp->rcv_up = tp->rcv_nxt; 1271 tcpstat.tcps_rcvpack++; 1272 tcpstat.tcps_rcvbyte += tlen; 1273 ND6_HINT(tp); /* some progress has been done */ 1274 #ifdef TCPDEBUG 1275 if (so->so_options & SO_DEBUG) 1276 tcp_trace(TA_INPUT, ostate, tp, 1277 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1278 #endif 1279 /* 1280 * Automatic sizing of receive socket buffer. Often the send 1281 * buffer size is not optimally adjusted to the actual network 1282 * conditions at hand (delay bandwidth product). Setting the 1283 * buffer size too small limits throughput on links with high 1284 * bandwidth and high delay (eg. trans-continental/oceanic links). 1285 * 1286 * On the receive side the socket buffer memory is only rarely 1287 * used to any significant extent. This allows us to be much 1288 * more aggressive in scaling the receive socket buffer. For 1289 * the case that the buffer space is actually used to a large 1290 * extent and we run out of kernel memory we can simply drop 1291 * the new segments; TCP on the sender will just retransmit it 1292 * later. Setting the buffer size too big may only consume too 1293 * much kernel memory if the application doesn't read() from 1294 * the socket or packet loss or reordering makes use of the 1295 * reassembly queue. 1296 * 1297 * The criteria to step up the receive buffer one notch are: 1298 * 1. the number of bytes received during the time it takes 1299 * one timestamp to be reflected back to us (the RTT); 1300 * 2. received bytes per RTT is within seven eighth of the 1301 * current socket buffer size; 1302 * 3. receive buffer size has not hit maximal automatic size; 1303 * 1304 * This algorithm does one step per RTT at most and only if 1305 * we receive a bulk stream w/o packet losses or reorderings. 1306 * Shrinking the buffer during idle times is not necessary as 1307 * it doesn't consume any memory when idle. 1308 * 1309 * TODO: Only step up if the application is actually serving 1310 * the buffer to better manage the socket buffer resources. 1311 */ 1312 if (tcp_do_autorcvbuf && 1313 to.to_tsecr && 1314 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1315 if (to.to_tsecr > tp->rfbuf_ts && 1316 to.to_tsecr - tp->rfbuf_ts < hz) { 1317 if (tp->rfbuf_cnt > 1318 (so->so_rcv.sb_hiwat / 8 * 7) && 1319 so->so_rcv.sb_hiwat < 1320 tcp_autorcvbuf_max) { 1321 newsize = 1322 min(so->so_rcv.sb_hiwat + 1323 tcp_autorcvbuf_inc, 1324 tcp_autorcvbuf_max); 1325 } 1326 /* Start over with next RTT. */ 1327 tp->rfbuf_ts = 0; 1328 tp->rfbuf_cnt = 0; 1329 } else 1330 tp->rfbuf_cnt += tlen; /* add up */ 1331 } 1332 1333 /* Add data to socket buffer. */ 1334 SOCKBUF_LOCK(&so->so_rcv); 1335 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1336 m_freem(m); 1337 } else { 1338 /* 1339 * Set new socket buffer size. 1340 * Give up when limit is reached. 1341 */ 1342 if (newsize) 1343 if (!sbreserve_locked(&so->so_rcv, 1344 newsize, so, curthread)) 1345 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1346 m_adj(m, drop_hdrlen); /* delayed header drop */ 1347 sbappendstream_locked(&so->so_rcv, m); 1348 } 1349 sorwakeup_locked(so); 1350 if (DELAY_ACK(tp)) { 1351 tp->t_flags |= TF_DELACK; 1352 } else { 1353 tp->t_flags |= TF_ACKNOW; 1354 tcp_output(tp); 1355 } 1356 goto check_delack; 1357 } 1358 } 1359 1360 /* 1361 * Calculate amount of space in receive window, 1362 * and then do TCP input processing. 1363 * Receive window is amount of space in rcv queue, 1364 * but not less than advertised window. 1365 */ 1366 win = sbspace(&so->so_rcv); 1367 if (win < 0) 1368 win = 0; 1369 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1370 1371 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1372 tp->rfbuf_ts = 0; 1373 tp->rfbuf_cnt = 0; 1374 1375 switch (tp->t_state) { 1376 1377 /* 1378 * If the state is SYN_RECEIVED: 1379 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1380 */ 1381 case TCPS_SYN_RECEIVED: 1382 if ((thflags & TH_ACK) && 1383 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1384 SEQ_GT(th->th_ack, tp->snd_max))) { 1385 rstreason = BANDLIM_RST_OPENPORT; 1386 goto dropwithreset; 1387 } 1388 break; 1389 1390 /* 1391 * If the state is SYN_SENT: 1392 * if seg contains an ACK, but not for our SYN, drop the input. 1393 * if seg contains a RST, then drop the connection. 1394 * if seg does not contain SYN, then drop it. 1395 * Otherwise this is an acceptable SYN segment 1396 * initialize tp->rcv_nxt and tp->irs 1397 * if seg contains ack then advance tp->snd_una 1398 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1399 * arrange for segment to be acked (eventually) 1400 * continue processing rest of data/controls, beginning with URG 1401 */ 1402 case TCPS_SYN_SENT: 1403 if ((thflags & TH_ACK) && 1404 (SEQ_LEQ(th->th_ack, tp->iss) || 1405 SEQ_GT(th->th_ack, tp->snd_max))) { 1406 rstreason = BANDLIM_UNLIMITED; 1407 goto dropwithreset; 1408 } 1409 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1410 tp = tcp_drop(tp, ECONNREFUSED); 1411 if (thflags & TH_RST) 1412 goto drop; 1413 if (!(thflags & TH_SYN)) 1414 goto drop; 1415 1416 tp->irs = th->th_seq; 1417 tcp_rcvseqinit(tp); 1418 if (thflags & TH_ACK) { 1419 tcpstat.tcps_connects++; 1420 soisconnected(so); 1421 #ifdef MAC 1422 SOCK_LOCK(so); 1423 mac_set_socket_peer_from_mbuf(m, so); 1424 SOCK_UNLOCK(so); 1425 #endif 1426 /* Do window scaling on this connection? */ 1427 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1428 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1429 tp->rcv_scale = tp->request_r_scale; 1430 } 1431 tp->rcv_adv += tp->rcv_wnd; 1432 tp->snd_una++; /* SYN is acked */ 1433 /* 1434 * If there's data, delay ACK; if there's also a FIN 1435 * ACKNOW will be turned on later. 1436 */ 1437 if (DELAY_ACK(tp) && tlen != 0) 1438 tcp_timer_activate(tp, TT_DELACK, 1439 tcp_delacktime); 1440 else 1441 tp->t_flags |= TF_ACKNOW; 1442 /* 1443 * Received <SYN,ACK> in SYN_SENT[*] state. 1444 * Transitions: 1445 * SYN_SENT --> ESTABLISHED 1446 * SYN_SENT* --> FIN_WAIT_1 1447 */ 1448 tp->t_starttime = ticks; 1449 if (tp->t_flags & TF_NEEDFIN) { 1450 tp->t_state = TCPS_FIN_WAIT_1; 1451 tp->t_flags &= ~TF_NEEDFIN; 1452 thflags &= ~TH_SYN; 1453 } else { 1454 tp->t_state = TCPS_ESTABLISHED; 1455 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1456 } 1457 } else { 1458 /* 1459 * Received initial SYN in SYN-SENT[*] state => 1460 * simultaneous open. If segment contains CC option 1461 * and there is a cached CC, apply TAO test. 1462 * If it succeeds, connection is * half-synchronized. 1463 * Otherwise, do 3-way handshake: 1464 * SYN-SENT -> SYN-RECEIVED 1465 * SYN-SENT* -> SYN-RECEIVED* 1466 * If there was no CC option, clear cached CC value. 1467 */ 1468 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1469 tcp_timer_activate(tp, TT_REXMT, 0); 1470 tp->t_state = TCPS_SYN_RECEIVED; 1471 } 1472 1473 KASSERT(headlocked, ("%s: trimthenstep6: head not locked", 1474 __func__)); 1475 INP_LOCK_ASSERT(tp->t_inpcb); 1476 1477 /* 1478 * Advance th->th_seq to correspond to first data byte. 1479 * If data, trim to stay within window, 1480 * dropping FIN if necessary. 1481 */ 1482 th->th_seq++; 1483 if (tlen > tp->rcv_wnd) { 1484 todrop = tlen - tp->rcv_wnd; 1485 m_adj(m, -todrop); 1486 tlen = tp->rcv_wnd; 1487 thflags &= ~TH_FIN; 1488 tcpstat.tcps_rcvpackafterwin++; 1489 tcpstat.tcps_rcvbyteafterwin += todrop; 1490 } 1491 tp->snd_wl1 = th->th_seq - 1; 1492 tp->rcv_up = th->th_seq; 1493 /* 1494 * Client side of transaction: already sent SYN and data. 1495 * If the remote host used T/TCP to validate the SYN, 1496 * our data will be ACK'd; if so, enter normal data segment 1497 * processing in the middle of step 5, ack processing. 1498 * Otherwise, goto step 6. 1499 */ 1500 if (thflags & TH_ACK) 1501 goto process_ACK; 1502 1503 goto step6; 1504 1505 /* 1506 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1507 * do normal processing. 1508 * 1509 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1510 */ 1511 case TCPS_LAST_ACK: 1512 case TCPS_CLOSING: 1513 break; /* continue normal processing */ 1514 } 1515 1516 /* 1517 * States other than LISTEN or SYN_SENT. 1518 * First check the RST flag and sequence number since reset segments 1519 * are exempt from the timestamp and connection count tests. This 1520 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1521 * below which allowed reset segments in half the sequence space 1522 * to fall though and be processed (which gives forged reset 1523 * segments with a random sequence number a 50 percent chance of 1524 * killing a connection). 1525 * Then check timestamp, if present. 1526 * Then check the connection count, if present. 1527 * Then check that at least some bytes of segment are within 1528 * receive window. If segment begins before rcv_nxt, 1529 * drop leading data (and SYN); if nothing left, just ack. 1530 * 1531 * 1532 * If the RST bit is set, check the sequence number to see 1533 * if this is a valid reset segment. 1534 * RFC 793 page 37: 1535 * In all states except SYN-SENT, all reset (RST) segments 1536 * are validated by checking their SEQ-fields. A reset is 1537 * valid if its sequence number is in the window. 1538 * Note: this does not take into account delayed ACKs, so 1539 * we should test against last_ack_sent instead of rcv_nxt. 1540 * The sequence number in the reset segment is normally an 1541 * echo of our outgoing acknowlegement numbers, but some hosts 1542 * send a reset with the sequence number at the rightmost edge 1543 * of our receive window, and we have to handle this case. 1544 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1545 * that brute force RST attacks are possible. To combat this, 1546 * we use a much stricter check while in the ESTABLISHED state, 1547 * only accepting RSTs where the sequence number is equal to 1548 * last_ack_sent. In all other states (the states in which a 1549 * RST is more likely), the more permissive check is used. 1550 * If we have multiple segments in flight, the intial reset 1551 * segment sequence numbers will be to the left of last_ack_sent, 1552 * but they will eventually catch up. 1553 * In any case, it never made sense to trim reset segments to 1554 * fit the receive window since RFC 1122 says: 1555 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1556 * 1557 * A TCP SHOULD allow a received RST segment to include data. 1558 * 1559 * DISCUSSION 1560 * It has been suggested that a RST segment could contain 1561 * ASCII text that encoded and explained the cause of the 1562 * RST. No standard has yet been established for such 1563 * data. 1564 * 1565 * If the reset segment passes the sequence number test examine 1566 * the state: 1567 * SYN_RECEIVED STATE: 1568 * If passive open, return to LISTEN state. 1569 * If active open, inform user that connection was refused. 1570 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1571 * Inform user that connection was reset, and close tcb. 1572 * CLOSING, LAST_ACK STATES: 1573 * Close the tcb. 1574 * TIME_WAIT STATE: 1575 * Drop the segment - see Stevens, vol. 2, p. 964 and 1576 * RFC 1337. 1577 */ 1578 if (thflags & TH_RST) { 1579 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1580 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1581 switch (tp->t_state) { 1582 1583 case TCPS_SYN_RECEIVED: 1584 so->so_error = ECONNREFUSED; 1585 goto close; 1586 1587 case TCPS_ESTABLISHED: 1588 if (tcp_insecure_rst == 0 && 1589 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 1590 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 1591 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1592 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 1593 tcpstat.tcps_badrst++; 1594 goto drop; 1595 } 1596 case TCPS_FIN_WAIT_1: 1597 case TCPS_FIN_WAIT_2: 1598 case TCPS_CLOSE_WAIT: 1599 so->so_error = ECONNRESET; 1600 close: 1601 tp->t_state = TCPS_CLOSED; 1602 tcpstat.tcps_drops++; 1603 KASSERT(headlocked, ("%s: trimthenstep6: " 1604 "tcp_close: head not locked", __func__)); 1605 tp = tcp_close(tp); 1606 break; 1607 1608 case TCPS_CLOSING: 1609 case TCPS_LAST_ACK: 1610 KASSERT(headlocked, ("%s: trimthenstep6: " 1611 "tcp_close.2: head not locked", __func__)); 1612 tp = tcp_close(tp); 1613 break; 1614 } 1615 } 1616 goto drop; 1617 } 1618 1619 /* 1620 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1621 * and it's less than ts_recent, drop it. 1622 */ 1623 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1624 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1625 1626 /* Check to see if ts_recent is over 24 days old. */ 1627 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1628 /* 1629 * Invalidate ts_recent. If this segment updates 1630 * ts_recent, the age will be reset later and ts_recent 1631 * will get a valid value. If it does not, setting 1632 * ts_recent to zero will at least satisfy the 1633 * requirement that zero be placed in the timestamp 1634 * echo reply when ts_recent isn't valid. The 1635 * age isn't reset until we get a valid ts_recent 1636 * because we don't want out-of-order segments to be 1637 * dropped when ts_recent is old. 1638 */ 1639 tp->ts_recent = 0; 1640 } else { 1641 tcpstat.tcps_rcvduppack++; 1642 tcpstat.tcps_rcvdupbyte += tlen; 1643 tcpstat.tcps_pawsdrop++; 1644 if (tlen) 1645 goto dropafterack; 1646 goto drop; 1647 } 1648 } 1649 1650 /* 1651 * In the SYN-RECEIVED state, validate that the packet belongs to 1652 * this connection before trimming the data to fit the receive 1653 * window. Check the sequence number versus IRS since we know 1654 * the sequence numbers haven't wrapped. This is a partial fix 1655 * for the "LAND" DoS attack. 1656 */ 1657 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1658 rstreason = BANDLIM_RST_OPENPORT; 1659 goto dropwithreset; 1660 } 1661 1662 todrop = tp->rcv_nxt - th->th_seq; 1663 if (todrop > 0) { 1664 if (thflags & TH_SYN) { 1665 thflags &= ~TH_SYN; 1666 th->th_seq++; 1667 if (th->th_urp > 1) 1668 th->th_urp--; 1669 else 1670 thflags &= ~TH_URG; 1671 todrop--; 1672 } 1673 /* 1674 * Following if statement from Stevens, vol. 2, p. 960. 1675 */ 1676 if (todrop > tlen 1677 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1678 /* 1679 * Any valid FIN must be to the left of the window. 1680 * At this point the FIN must be a duplicate or out 1681 * of sequence; drop it. 1682 */ 1683 thflags &= ~TH_FIN; 1684 1685 /* 1686 * Send an ACK to resynchronize and drop any data. 1687 * But keep on processing for RST or ACK. 1688 */ 1689 tp->t_flags |= TF_ACKNOW; 1690 todrop = tlen; 1691 tcpstat.tcps_rcvduppack++; 1692 tcpstat.tcps_rcvdupbyte += todrop; 1693 } else { 1694 tcpstat.tcps_rcvpartduppack++; 1695 tcpstat.tcps_rcvpartdupbyte += todrop; 1696 } 1697 drop_hdrlen += todrop; /* drop from the top afterwards */ 1698 th->th_seq += todrop; 1699 tlen -= todrop; 1700 if (th->th_urp > todrop) 1701 th->th_urp -= todrop; 1702 else { 1703 thflags &= ~TH_URG; 1704 th->th_urp = 0; 1705 } 1706 } 1707 1708 /* 1709 * If new data are received on a connection after the 1710 * user processes are gone, then RST the other end. 1711 */ 1712 if ((so->so_state & SS_NOFDREF) && 1713 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1714 KASSERT(headlocked, ("%s: trimthenstep6: tcp_close.3: head " 1715 "not locked", __func__)); 1716 tp = tcp_close(tp); 1717 tcpstat.tcps_rcvafterclose++; 1718 rstreason = BANDLIM_UNLIMITED; 1719 goto dropwithreset; 1720 } 1721 1722 /* 1723 * If segment ends after window, drop trailing data 1724 * (and PUSH and FIN); if nothing left, just ACK. 1725 */ 1726 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 1727 if (todrop > 0) { 1728 tcpstat.tcps_rcvpackafterwin++; 1729 if (todrop >= tlen) { 1730 tcpstat.tcps_rcvbyteafterwin += tlen; 1731 /* 1732 * If window is closed can only take segments at 1733 * window edge, and have to drop data and PUSH from 1734 * incoming segments. Continue processing, but 1735 * remember to ack. Otherwise, drop segment 1736 * and ack. 1737 */ 1738 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1739 tp->t_flags |= TF_ACKNOW; 1740 tcpstat.tcps_rcvwinprobe++; 1741 } else 1742 goto dropafterack; 1743 } else 1744 tcpstat.tcps_rcvbyteafterwin += todrop; 1745 m_adj(m, -todrop); 1746 tlen -= todrop; 1747 thflags &= ~(TH_PUSH|TH_FIN); 1748 } 1749 1750 /* 1751 * If last ACK falls within this segment's sequence numbers, 1752 * record its timestamp. 1753 * NOTE: 1754 * 1) That the test incorporates suggestions from the latest 1755 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1756 * 2) That updating only on newer timestamps interferes with 1757 * our earlier PAWS tests, so this check should be solely 1758 * predicated on the sequence space of this segment. 1759 * 3) That we modify the segment boundary check to be 1760 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1761 * instead of RFC1323's 1762 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1763 * This modified check allows us to overcome RFC1323's 1764 * limitations as described in Stevens TCP/IP Illustrated 1765 * Vol. 2 p.869. In such cases, we can still calculate the 1766 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1767 */ 1768 if ((to.to_flags & TOF_TS) != 0 && 1769 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1770 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1771 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1772 tp->ts_recent_age = ticks; 1773 tp->ts_recent = to.to_tsval; 1774 } 1775 1776 /* 1777 * If a SYN is in the window, then this is an 1778 * error and we send an RST and drop the connection. 1779 */ 1780 if (thflags & TH_SYN) { 1781 KASSERT(headlocked, ("%s: tcp_drop: trimthenstep6: " 1782 "head not locked", __func__)); 1783 tp = tcp_drop(tp, ECONNRESET); 1784 rstreason = BANDLIM_UNLIMITED; 1785 goto drop; 1786 } 1787 1788 /* 1789 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1790 * flag is on (half-synchronized state), then queue data for 1791 * later processing; else drop segment and return. 1792 */ 1793 if ((thflags & TH_ACK) == 0) { 1794 if (tp->t_state == TCPS_SYN_RECEIVED || 1795 (tp->t_flags & TF_NEEDSYN)) 1796 goto step6; 1797 else if (tp->t_flags & TF_ACKNOW) 1798 goto dropafterack; 1799 else 1800 goto drop; 1801 } 1802 1803 /* 1804 * Ack processing. 1805 */ 1806 switch (tp->t_state) { 1807 1808 /* 1809 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1810 * ESTABLISHED state and continue processing. 1811 * The ACK was checked above. 1812 */ 1813 case TCPS_SYN_RECEIVED: 1814 1815 tcpstat.tcps_connects++; 1816 soisconnected(so); 1817 /* Do window scaling? */ 1818 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1819 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1820 tp->rcv_scale = tp->request_r_scale; 1821 tp->snd_wnd = tiwin; 1822 } 1823 /* 1824 * Make transitions: 1825 * SYN-RECEIVED -> ESTABLISHED 1826 * SYN-RECEIVED* -> FIN-WAIT-1 1827 */ 1828 tp->t_starttime = ticks; 1829 if (tp->t_flags & TF_NEEDFIN) { 1830 tp->t_state = TCPS_FIN_WAIT_1; 1831 tp->t_flags &= ~TF_NEEDFIN; 1832 } else { 1833 tp->t_state = TCPS_ESTABLISHED; 1834 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1835 } 1836 /* 1837 * If segment contains data or ACK, will call tcp_reass() 1838 * later; if not, do so now to pass queued data to user. 1839 */ 1840 if (tlen == 0 && (thflags & TH_FIN) == 0) 1841 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1842 (struct mbuf *)0); 1843 tp->snd_wl1 = th->th_seq - 1; 1844 /* FALLTHROUGH */ 1845 1846 /* 1847 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1848 * ACKs. If the ack is in the range 1849 * tp->snd_una < th->th_ack <= tp->snd_max 1850 * then advance tp->snd_una to th->th_ack and drop 1851 * data from the retransmission queue. If this ACK reflects 1852 * more up to date window information we update our window information. 1853 */ 1854 case TCPS_ESTABLISHED: 1855 case TCPS_FIN_WAIT_1: 1856 case TCPS_FIN_WAIT_2: 1857 case TCPS_CLOSE_WAIT: 1858 case TCPS_CLOSING: 1859 case TCPS_LAST_ACK: 1860 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1861 tcpstat.tcps_rcvacktoomuch++; 1862 goto dropafterack; 1863 } 1864 if ((tp->t_flags & TF_SACK_PERMIT) && 1865 ((to.to_flags & TOF_SACK) || 1866 !TAILQ_EMPTY(&tp->snd_holes))) 1867 tcp_sack_doack(tp, &to, th->th_ack); 1868 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1869 if (tlen == 0 && tiwin == tp->snd_wnd) { 1870 tcpstat.tcps_rcvdupack++; 1871 /* 1872 * If we have outstanding data (other than 1873 * a window probe), this is a completely 1874 * duplicate ack (ie, window info didn't 1875 * change), the ack is the biggest we've 1876 * seen and we've seen exactly our rexmt 1877 * threshhold of them, assume a packet 1878 * has been dropped and retransmit it. 1879 * Kludge snd_nxt & the congestion 1880 * window so we send only this one 1881 * packet. 1882 * 1883 * We know we're losing at the current 1884 * window size so do congestion avoidance 1885 * (set ssthresh to half the current window 1886 * and pull our congestion window back to 1887 * the new ssthresh). 1888 * 1889 * Dup acks mean that packets have left the 1890 * network (they're now cached at the receiver) 1891 * so bump cwnd by the amount in the receiver 1892 * to keep a constant cwnd packets in the 1893 * network. 1894 */ 1895 if (!tcp_timer_active(tp, TT_REXMT) || 1896 th->th_ack != tp->snd_una) 1897 tp->t_dupacks = 0; 1898 else if (++tp->t_dupacks > tcprexmtthresh || 1899 ((tcp_do_newreno || 1900 (tp->t_flags & TF_SACK_PERMIT)) && 1901 IN_FASTRECOVERY(tp))) { 1902 if ((tp->t_flags & TF_SACK_PERMIT) && 1903 IN_FASTRECOVERY(tp)) { 1904 int awnd; 1905 1906 /* 1907 * Compute the amount of data in flight first. 1908 * We can inject new data into the pipe iff 1909 * we have less than 1/2 the original window's 1910 * worth of data in flight. 1911 */ 1912 awnd = (tp->snd_nxt - tp->snd_fack) + 1913 tp->sackhint.sack_bytes_rexmit; 1914 if (awnd < tp->snd_ssthresh) { 1915 tp->snd_cwnd += tp->t_maxseg; 1916 if (tp->snd_cwnd > tp->snd_ssthresh) 1917 tp->snd_cwnd = tp->snd_ssthresh; 1918 } 1919 } else 1920 tp->snd_cwnd += tp->t_maxseg; 1921 (void) tcp_output(tp); 1922 goto drop; 1923 } else if (tp->t_dupacks == tcprexmtthresh) { 1924 tcp_seq onxt = tp->snd_nxt; 1925 u_int win; 1926 1927 /* 1928 * If we're doing sack, check to 1929 * see if we're already in sack 1930 * recovery. If we're not doing sack, 1931 * check to see if we're in newreno 1932 * recovery. 1933 */ 1934 if (tp->t_flags & TF_SACK_PERMIT) { 1935 if (IN_FASTRECOVERY(tp)) { 1936 tp->t_dupacks = 0; 1937 break; 1938 } 1939 } else if (tcp_do_newreno) { 1940 if (SEQ_LEQ(th->th_ack, 1941 tp->snd_recover)) { 1942 tp->t_dupacks = 0; 1943 break; 1944 } 1945 } 1946 win = min(tp->snd_wnd, tp->snd_cwnd) / 1947 2 / tp->t_maxseg; 1948 if (win < 2) 1949 win = 2; 1950 tp->snd_ssthresh = win * tp->t_maxseg; 1951 ENTER_FASTRECOVERY(tp); 1952 tp->snd_recover = tp->snd_max; 1953 tcp_timer_activate(tp, TT_REXMT, 0); 1954 tp->t_rtttime = 0; 1955 if (tp->t_flags & TF_SACK_PERMIT) { 1956 tcpstat.tcps_sack_recovery_episode++; 1957 tp->sack_newdata = tp->snd_nxt; 1958 tp->snd_cwnd = tp->t_maxseg; 1959 (void) tcp_output(tp); 1960 goto drop; 1961 } 1962 tp->snd_nxt = th->th_ack; 1963 tp->snd_cwnd = tp->t_maxseg; 1964 (void) tcp_output(tp); 1965 KASSERT(tp->snd_limited <= 2, 1966 ("%s: tp->snd_limited too big", 1967 __func__)); 1968 tp->snd_cwnd = tp->snd_ssthresh + 1969 tp->t_maxseg * 1970 (tp->t_dupacks - tp->snd_limited); 1971 if (SEQ_GT(onxt, tp->snd_nxt)) 1972 tp->snd_nxt = onxt; 1973 goto drop; 1974 } else if (tcp_do_rfc3042) { 1975 u_long oldcwnd = tp->snd_cwnd; 1976 tcp_seq oldsndmax = tp->snd_max; 1977 u_int sent; 1978 1979 KASSERT(tp->t_dupacks == 1 || 1980 tp->t_dupacks == 2, 1981 ("%s: dupacks not 1 or 2", 1982 __func__)); 1983 if (tp->t_dupacks == 1) 1984 tp->snd_limited = 0; 1985 tp->snd_cwnd = 1986 (tp->snd_nxt - tp->snd_una) + 1987 (tp->t_dupacks - tp->snd_limited) * 1988 tp->t_maxseg; 1989 (void) tcp_output(tp); 1990 sent = tp->snd_max - oldsndmax; 1991 if (sent > tp->t_maxseg) { 1992 KASSERT((tp->t_dupacks == 2 && 1993 tp->snd_limited == 0) || 1994 (sent == tp->t_maxseg + 1 && 1995 tp->t_flags & TF_SENTFIN), 1996 ("%s: sent too much", 1997 __func__)); 1998 tp->snd_limited = 2; 1999 } else if (sent > 0) 2000 ++tp->snd_limited; 2001 tp->snd_cwnd = oldcwnd; 2002 goto drop; 2003 } 2004 } else 2005 tp->t_dupacks = 0; 2006 break; 2007 } 2008 2009 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2010 ("%s: th_ack <= snd_una", __func__)); 2011 2012 /* 2013 * If the congestion window was inflated to account 2014 * for the other side's cached packets, retract it. 2015 */ 2016 if (tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) { 2017 if (IN_FASTRECOVERY(tp)) { 2018 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2019 if (tp->t_flags & TF_SACK_PERMIT) 2020 tcp_sack_partialack(tp, th); 2021 else 2022 tcp_newreno_partial_ack(tp, th); 2023 } else { 2024 /* 2025 * Out of fast recovery. 2026 * Window inflation should have left us 2027 * with approximately snd_ssthresh 2028 * outstanding data. 2029 * But in case we would be inclined to 2030 * send a burst, better to do it via 2031 * the slow start mechanism. 2032 */ 2033 if (SEQ_GT(th->th_ack + 2034 tp->snd_ssthresh, 2035 tp->snd_max)) 2036 tp->snd_cwnd = tp->snd_max - 2037 th->th_ack + 2038 tp->t_maxseg; 2039 else 2040 tp->snd_cwnd = tp->snd_ssthresh; 2041 } 2042 } 2043 } else { 2044 if (tp->t_dupacks >= tcprexmtthresh && 2045 tp->snd_cwnd > tp->snd_ssthresh) 2046 tp->snd_cwnd = tp->snd_ssthresh; 2047 } 2048 tp->t_dupacks = 0; 2049 /* 2050 * If we reach this point, ACK is not a duplicate, 2051 * i.e., it ACKs something we sent. 2052 */ 2053 if (tp->t_flags & TF_NEEDSYN) { 2054 /* 2055 * T/TCP: Connection was half-synchronized, and our 2056 * SYN has been ACK'd (so connection is now fully 2057 * synchronized). Go to non-starred state, 2058 * increment snd_una for ACK of SYN, and check if 2059 * we can do window scaling. 2060 */ 2061 tp->t_flags &= ~TF_NEEDSYN; 2062 tp->snd_una++; 2063 /* Do window scaling? */ 2064 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2065 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2066 tp->rcv_scale = tp->request_r_scale; 2067 /* Send window already scaled. */ 2068 } 2069 } 2070 2071 process_ACK: 2072 KASSERT(headlocked, ("%s: process_ACK: head not locked", 2073 __func__)); 2074 INP_LOCK_ASSERT(tp->t_inpcb); 2075 2076 acked = th->th_ack - tp->snd_una; 2077 tcpstat.tcps_rcvackpack++; 2078 tcpstat.tcps_rcvackbyte += acked; 2079 2080 /* 2081 * If we just performed our first retransmit, and the ACK 2082 * arrives within our recovery window, then it was a mistake 2083 * to do the retransmit in the first place. Recover our 2084 * original cwnd and ssthresh, and proceed to transmit where 2085 * we left off. 2086 */ 2087 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2088 ++tcpstat.tcps_sndrexmitbad; 2089 tp->snd_cwnd = tp->snd_cwnd_prev; 2090 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2091 tp->snd_recover = tp->snd_recover_prev; 2092 if (tp->t_flags & TF_WASFRECOVERY) 2093 ENTER_FASTRECOVERY(tp); 2094 tp->snd_nxt = tp->snd_max; 2095 tp->t_badrxtwin = 0; /* XXX probably not required */ 2096 } 2097 2098 /* 2099 * If we have a timestamp reply, update smoothed 2100 * round trip time. If no timestamp is present but 2101 * transmit timer is running and timed sequence 2102 * number was acked, update smoothed round trip time. 2103 * Since we now have an rtt measurement, cancel the 2104 * timer backoff (cf., Phil Karn's retransmit alg.). 2105 * Recompute the initial retransmit timer. 2106 * 2107 * Some boxes send broken timestamp replies 2108 * during the SYN+ACK phase, ignore 2109 * timestamps of 0 or we could calculate a 2110 * huge RTT and blow up the retransmit timer. 2111 */ 2112 if ((to.to_flags & TOF_TS) != 0 && 2113 to.to_tsecr) { 2114 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2115 tp->t_rttlow = ticks - to.to_tsecr; 2116 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2117 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2118 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2119 tp->t_rttlow = ticks - tp->t_rtttime; 2120 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2121 } 2122 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2123 2124 /* 2125 * If all outstanding data is acked, stop retransmit 2126 * timer and remember to restart (more output or persist). 2127 * If there is more data to be acked, restart retransmit 2128 * timer, using current (possibly backed-off) value. 2129 */ 2130 if (th->th_ack == tp->snd_max) { 2131 tcp_timer_activate(tp, TT_REXMT, 0); 2132 needoutput = 1; 2133 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2134 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2135 2136 /* 2137 * If no data (only SYN) was ACK'd, 2138 * skip rest of ACK processing. 2139 */ 2140 if (acked == 0) 2141 goto step6; 2142 2143 /* 2144 * When new data is acked, open the congestion window. 2145 * If the window gives us less than ssthresh packets 2146 * in flight, open exponentially (maxseg per packet). 2147 * Otherwise open linearly: maxseg per window 2148 * (maxseg^2 / cwnd per packet). 2149 */ 2150 if ((!tcp_do_newreno && !(tp->t_flags & TF_SACK_PERMIT)) || 2151 !IN_FASTRECOVERY(tp)) { 2152 u_int cw = tp->snd_cwnd; 2153 u_int incr = tp->t_maxseg; 2154 if (cw > tp->snd_ssthresh) 2155 incr = incr * incr / cw; 2156 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2157 } 2158 SOCKBUF_LOCK(&so->so_snd); 2159 if (acked > so->so_snd.sb_cc) { 2160 tp->snd_wnd -= so->so_snd.sb_cc; 2161 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2162 ourfinisacked = 1; 2163 } else { 2164 sbdrop_locked(&so->so_snd, acked); 2165 tp->snd_wnd -= acked; 2166 ourfinisacked = 0; 2167 } 2168 sowwakeup_locked(so); 2169 /* detect una wraparound */ 2170 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2171 !IN_FASTRECOVERY(tp) && 2172 SEQ_GT(tp->snd_una, tp->snd_recover) && 2173 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2174 tp->snd_recover = th->th_ack - 1; 2175 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2176 IN_FASTRECOVERY(tp) && 2177 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2178 EXIT_FASTRECOVERY(tp); 2179 tp->snd_una = th->th_ack; 2180 if (tp->t_flags & TF_SACK_PERMIT) { 2181 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2182 tp->snd_recover = tp->snd_una; 2183 } 2184 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2185 tp->snd_nxt = tp->snd_una; 2186 2187 switch (tp->t_state) { 2188 2189 /* 2190 * In FIN_WAIT_1 STATE in addition to the processing 2191 * for the ESTABLISHED state if our FIN is now acknowledged 2192 * then enter FIN_WAIT_2. 2193 */ 2194 case TCPS_FIN_WAIT_1: 2195 if (ourfinisacked) { 2196 /* 2197 * If we can't receive any more 2198 * data, then closing user can proceed. 2199 * Starting the timer is contrary to the 2200 * specification, but if we don't get a FIN 2201 * we'll hang forever. 2202 */ 2203 /* XXXjl 2204 * we should release the tp also, and use a 2205 * compressed state. 2206 */ 2207 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2208 int timeout; 2209 2210 soisdisconnected(so); 2211 timeout = (tcp_fast_finwait2_recycle) ? 2212 tcp_finwait2_timeout : tcp_maxidle; 2213 tcp_timer_activate(tp, TT_2MSL, timeout); 2214 } 2215 tp->t_state = TCPS_FIN_WAIT_2; 2216 } 2217 break; 2218 2219 /* 2220 * In CLOSING STATE in addition to the processing for 2221 * the ESTABLISHED state if the ACK acknowledges our FIN 2222 * then enter the TIME-WAIT state, otherwise ignore 2223 * the segment. 2224 */ 2225 case TCPS_CLOSING: 2226 if (ourfinisacked) { 2227 KASSERT(headlocked, ("%s: process_ACK: " 2228 "head not locked", __func__)); 2229 tcp_twstart(tp); 2230 INP_INFO_WUNLOCK(&tcbinfo); 2231 headlocked = 0; 2232 m_freem(m); 2233 return; 2234 } 2235 break; 2236 2237 /* 2238 * In LAST_ACK, we may still be waiting for data to drain 2239 * and/or to be acked, as well as for the ack of our FIN. 2240 * If our FIN is now acknowledged, delete the TCB, 2241 * enter the closed state and return. 2242 */ 2243 case TCPS_LAST_ACK: 2244 if (ourfinisacked) { 2245 KASSERT(headlocked, ("%s: process_ACK: " 2246 "tcp_close: head not locked", __func__)); 2247 tp = tcp_close(tp); 2248 goto drop; 2249 } 2250 break; 2251 } 2252 } 2253 2254 step6: 2255 KASSERT(headlocked, ("%s: step6: head not locked", __func__)); 2256 INP_LOCK_ASSERT(tp->t_inpcb); 2257 2258 /* 2259 * Update window information. 2260 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2261 */ 2262 if ((thflags & TH_ACK) && 2263 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2264 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2265 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2266 /* keep track of pure window updates */ 2267 if (tlen == 0 && 2268 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2269 tcpstat.tcps_rcvwinupd++; 2270 tp->snd_wnd = tiwin; 2271 tp->snd_wl1 = th->th_seq; 2272 tp->snd_wl2 = th->th_ack; 2273 if (tp->snd_wnd > tp->max_sndwnd) 2274 tp->max_sndwnd = tp->snd_wnd; 2275 needoutput = 1; 2276 } 2277 2278 /* 2279 * Process segments with URG. 2280 */ 2281 if ((thflags & TH_URG) && th->th_urp && 2282 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2283 /* 2284 * This is a kludge, but if we receive and accept 2285 * random urgent pointers, we'll crash in 2286 * soreceive. It's hard to imagine someone 2287 * actually wanting to send this much urgent data. 2288 */ 2289 SOCKBUF_LOCK(&so->so_rcv); 2290 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2291 th->th_urp = 0; /* XXX */ 2292 thflags &= ~TH_URG; /* XXX */ 2293 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2294 goto dodata; /* XXX */ 2295 } 2296 /* 2297 * If this segment advances the known urgent pointer, 2298 * then mark the data stream. This should not happen 2299 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2300 * a FIN has been received from the remote side. 2301 * In these states we ignore the URG. 2302 * 2303 * According to RFC961 (Assigned Protocols), 2304 * the urgent pointer points to the last octet 2305 * of urgent data. We continue, however, 2306 * to consider it to indicate the first octet 2307 * of data past the urgent section as the original 2308 * spec states (in one of two places). 2309 */ 2310 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2311 tp->rcv_up = th->th_seq + th->th_urp; 2312 so->so_oobmark = so->so_rcv.sb_cc + 2313 (tp->rcv_up - tp->rcv_nxt) - 1; 2314 if (so->so_oobmark == 0) 2315 so->so_rcv.sb_state |= SBS_RCVATMARK; 2316 sohasoutofband(so); 2317 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2318 } 2319 SOCKBUF_UNLOCK(&so->so_rcv); 2320 /* 2321 * Remove out of band data so doesn't get presented to user. 2322 * This can happen independent of advancing the URG pointer, 2323 * but if two URG's are pending at once, some out-of-band 2324 * data may creep in... ick. 2325 */ 2326 if (th->th_urp <= (u_long)tlen && 2327 !(so->so_options & SO_OOBINLINE)) { 2328 /* hdr drop is delayed */ 2329 tcp_pulloutofband(so, th, m, drop_hdrlen); 2330 } 2331 } else { 2332 /* 2333 * If no out of band data is expected, 2334 * pull receive urgent pointer along 2335 * with the receive window. 2336 */ 2337 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2338 tp->rcv_up = tp->rcv_nxt; 2339 } 2340 dodata: /* XXX */ 2341 KASSERT(headlocked, ("%s: dodata: head not locked", __func__)); 2342 INP_LOCK_ASSERT(tp->t_inpcb); 2343 2344 /* 2345 * Process the segment text, merging it into the TCP sequencing queue, 2346 * and arranging for acknowledgment of receipt if necessary. 2347 * This process logically involves adjusting tp->rcv_wnd as data 2348 * is presented to the user (this happens in tcp_usrreq.c, 2349 * case PRU_RCVD). If a FIN has already been received on this 2350 * connection then we just ignore the text. 2351 */ 2352 if ((tlen || (thflags & TH_FIN)) && 2353 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2354 tcp_seq save_start = th->th_seq; 2355 tcp_seq save_end = th->th_seq + tlen; 2356 m_adj(m, drop_hdrlen); /* delayed header drop */ 2357 /* 2358 * Insert segment which includes th into TCP reassembly queue 2359 * with control block tp. Set thflags to whether reassembly now 2360 * includes a segment with FIN. This handles the common case 2361 * inline (segment is the next to be received on an established 2362 * connection, and the queue is empty), avoiding linkage into 2363 * and removal from the queue and repetition of various 2364 * conversions. 2365 * Set DELACK for segments received in order, but ack 2366 * immediately when segments are out of order (so 2367 * fast retransmit can work). 2368 */ 2369 if (th->th_seq == tp->rcv_nxt && 2370 LIST_EMPTY(&tp->t_segq) && 2371 TCPS_HAVEESTABLISHED(tp->t_state)) { 2372 if (DELAY_ACK(tp)) 2373 tp->t_flags |= TF_DELACK; 2374 else 2375 tp->t_flags |= TF_ACKNOW; 2376 tp->rcv_nxt += tlen; 2377 thflags = th->th_flags & TH_FIN; 2378 tcpstat.tcps_rcvpack++; 2379 tcpstat.tcps_rcvbyte += tlen; 2380 ND6_HINT(tp); 2381 SOCKBUF_LOCK(&so->so_rcv); 2382 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2383 m_freem(m); 2384 else 2385 sbappendstream_locked(&so->so_rcv, m); 2386 sorwakeup_locked(so); 2387 } else { 2388 thflags = tcp_reass(tp, th, &tlen, m); 2389 tp->t_flags |= TF_ACKNOW; 2390 } 2391 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2392 tcp_update_sack_list(tp, save_start, save_end); 2393 #if 0 2394 /* 2395 * Note the amount of data that peer has sent into 2396 * our window, in order to estimate the sender's 2397 * buffer size. 2398 * XXX: Unused. 2399 */ 2400 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2401 #endif 2402 } else { 2403 m_freem(m); 2404 thflags &= ~TH_FIN; 2405 } 2406 2407 /* 2408 * If FIN is received ACK the FIN and let the user know 2409 * that the connection is closing. 2410 */ 2411 if (thflags & TH_FIN) { 2412 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2413 socantrcvmore(so); 2414 /* 2415 * If connection is half-synchronized 2416 * (ie NEEDSYN flag on) then delay ACK, 2417 * so it may be piggybacked when SYN is sent. 2418 * Otherwise, since we received a FIN then no 2419 * more input can be expected, send ACK now. 2420 */ 2421 if (tp->t_flags & TF_NEEDSYN) 2422 tp->t_flags |= TF_DELACK; 2423 else 2424 tp->t_flags |= TF_ACKNOW; 2425 tp->rcv_nxt++; 2426 } 2427 switch (tp->t_state) { 2428 2429 /* 2430 * In SYN_RECEIVED and ESTABLISHED STATES 2431 * enter the CLOSE_WAIT state. 2432 */ 2433 case TCPS_SYN_RECEIVED: 2434 tp->t_starttime = ticks; 2435 /*FALLTHROUGH*/ 2436 case TCPS_ESTABLISHED: 2437 tp->t_state = TCPS_CLOSE_WAIT; 2438 break; 2439 2440 /* 2441 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2442 * enter the CLOSING state. 2443 */ 2444 case TCPS_FIN_WAIT_1: 2445 tp->t_state = TCPS_CLOSING; 2446 break; 2447 2448 /* 2449 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2450 * starting the time-wait timer, turning off the other 2451 * standard timers. 2452 */ 2453 case TCPS_FIN_WAIT_2: 2454 KASSERT(headlocked == 1, ("%s: dodata: " 2455 "TCP_FIN_WAIT_2: head not locked", __func__)); 2456 tcp_twstart(tp); 2457 INP_INFO_WUNLOCK(&tcbinfo); 2458 return; 2459 } 2460 } 2461 INP_INFO_WUNLOCK(&tcbinfo); 2462 headlocked = 0; 2463 #ifdef TCPDEBUG 2464 if (so->so_options & SO_DEBUG) 2465 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2466 &tcp_savetcp, 0); 2467 #endif 2468 2469 /* 2470 * Return any desired output. 2471 */ 2472 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2473 (void) tcp_output(tp); 2474 2475 check_delack: 2476 KASSERT(headlocked == 0, ("%s: check_delack: head locked", 2477 __func__)); 2478 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 2479 INP_LOCK_ASSERT(tp->t_inpcb); 2480 if (tp->t_flags & TF_DELACK) { 2481 tp->t_flags &= ~TF_DELACK; 2482 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2483 } 2484 INP_UNLOCK(tp->t_inpcb); 2485 return; 2486 2487 dropafterack: 2488 KASSERT(headlocked, ("%s: dropafterack: head not locked", __func__)); 2489 /* 2490 * Generate an ACK dropping incoming segment if it occupies 2491 * sequence space, where the ACK reflects our state. 2492 * 2493 * We can now skip the test for the RST flag since all 2494 * paths to this code happen after packets containing 2495 * RST have been dropped. 2496 * 2497 * In the SYN-RECEIVED state, don't send an ACK unless the 2498 * segment we received passes the SYN-RECEIVED ACK test. 2499 * If it fails send a RST. This breaks the loop in the 2500 * "LAND" DoS attack, and also prevents an ACK storm 2501 * between two listening ports that have been sent forged 2502 * SYN segments, each with the source address of the other. 2503 */ 2504 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2505 (SEQ_GT(tp->snd_una, th->th_ack) || 2506 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2507 rstreason = BANDLIM_RST_OPENPORT; 2508 goto dropwithreset; 2509 } 2510 #ifdef TCPDEBUG 2511 if (so->so_options & SO_DEBUG) 2512 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2513 &tcp_savetcp, 0); 2514 #endif 2515 KASSERT(headlocked, ("%s: headlocked should be 1", __func__)); 2516 INP_INFO_WUNLOCK(&tcbinfo); 2517 tp->t_flags |= TF_ACKNOW; 2518 (void) tcp_output(tp); 2519 INP_UNLOCK(tp->t_inpcb); 2520 m_freem(m); 2521 return; 2522 2523 dropwithreset: 2524 KASSERT(headlocked, ("%s: dropwithreset: head not locked", __func__)); 2525 2526 tcp_dropwithreset(m, th, tp, tlen, rstreason); 2527 2528 if (tp != NULL) 2529 INP_UNLOCK(tp->t_inpcb); 2530 if (headlocked) 2531 INP_INFO_WUNLOCK(&tcbinfo); 2532 return; 2533 2534 drop: 2535 /* 2536 * Drop space held by incoming segment and return. 2537 */ 2538 #ifdef TCPDEBUG 2539 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2540 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2541 &tcp_savetcp, 0); 2542 #endif 2543 if (tp != NULL) 2544 INP_UNLOCK(tp->t_inpcb); 2545 if (headlocked) 2546 INP_INFO_WUNLOCK(&tcbinfo); 2547 m_freem(m); 2548 return; 2549 } 2550 2551 /* 2552 * Issue RST and make ACK acceptable to originator of segment. 2553 * The mbuf must still include the original packet header. 2554 * tp may be NULL. 2555 */ 2556 static void 2557 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 2558 int tlen, int rstreason) 2559 { 2560 struct ip *ip; 2561 #ifdef INET6 2562 struct ip6_hdr *ip6; 2563 #endif 2564 /* Don't bother if destination was broadcast/multicast. */ 2565 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2566 goto drop; 2567 #ifdef INET6 2568 if (mtod(m, struct ip *)->ip_v == 6) { 2569 ip6 = mtod(m, struct ip6_hdr *); 2570 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2571 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2572 goto drop; 2573 /* IPv6 anycast check is done at tcp6_input() */ 2574 } else 2575 #endif 2576 { 2577 ip = mtod(m, struct ip *); 2578 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2579 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2580 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2581 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2582 goto drop; 2583 } 2584 2585 /* Perform bandwidth limiting. */ 2586 if (badport_bandlim(rstreason) < 0) 2587 goto drop; 2588 2589 /* tcp_respond consumes the mbuf chain. */ 2590 if (th->th_flags & TH_ACK) { 2591 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 2592 th->th_ack, TH_RST); 2593 } else { 2594 if (th->th_flags & TH_SYN) 2595 tlen++; 2596 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2597 (tcp_seq)0, TH_RST|TH_ACK); 2598 } 2599 return; 2600 drop: 2601 m_freem(m); 2602 return; 2603 } 2604 2605 /* 2606 * Parse TCP options and place in tcpopt. 2607 */ 2608 static void 2609 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 2610 { 2611 int opt, optlen; 2612 2613 to->to_flags = 0; 2614 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2615 opt = cp[0]; 2616 if (opt == TCPOPT_EOL) 2617 break; 2618 if (opt == TCPOPT_NOP) 2619 optlen = 1; 2620 else { 2621 if (cnt < 2) 2622 break; 2623 optlen = cp[1]; 2624 if (optlen < 2 || optlen > cnt) 2625 break; 2626 } 2627 switch (opt) { 2628 case TCPOPT_MAXSEG: 2629 if (optlen != TCPOLEN_MAXSEG) 2630 continue; 2631 if (!(flags & TO_SYN)) 2632 continue; 2633 to->to_flags |= TOF_MSS; 2634 bcopy((char *)cp + 2, 2635 (char *)&to->to_mss, sizeof(to->to_mss)); 2636 to->to_mss = ntohs(to->to_mss); 2637 break; 2638 case TCPOPT_WINDOW: 2639 if (optlen != TCPOLEN_WINDOW) 2640 continue; 2641 if (!(flags & TO_SYN)) 2642 continue; 2643 to->to_flags |= TOF_SCALE; 2644 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 2645 break; 2646 case TCPOPT_TIMESTAMP: 2647 if (optlen != TCPOLEN_TIMESTAMP) 2648 continue; 2649 to->to_flags |= TOF_TS; 2650 bcopy((char *)cp + 2, 2651 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2652 to->to_tsval = ntohl(to->to_tsval); 2653 bcopy((char *)cp + 6, 2654 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2655 to->to_tsecr = ntohl(to->to_tsecr); 2656 break; 2657 #ifdef TCP_SIGNATURE 2658 /* 2659 * XXX In order to reply to a host which has set the 2660 * TCP_SIGNATURE option in its initial SYN, we have to 2661 * record the fact that the option was observed here 2662 * for the syncache code to perform the correct response. 2663 */ 2664 case TCPOPT_SIGNATURE: 2665 if (optlen != TCPOLEN_SIGNATURE) 2666 continue; 2667 to->to_flags |= TOF_SIGNATURE; 2668 to->to_signature = cp + 2; 2669 break; 2670 #endif 2671 case TCPOPT_SACK_PERMITTED: 2672 if (optlen != TCPOLEN_SACK_PERMITTED) 2673 continue; 2674 if (!(flags & TO_SYN)) 2675 continue; 2676 if (!tcp_do_sack) 2677 continue; 2678 to->to_flags |= TOF_SACKPERM; 2679 break; 2680 case TCPOPT_SACK: 2681 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2682 continue; 2683 if (flags & TO_SYN) 2684 continue; 2685 to->to_flags |= TOF_SACK; 2686 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2687 to->to_sacks = cp + 2; 2688 tcpstat.tcps_sack_rcv_blocks++; 2689 break; 2690 default: 2691 continue; 2692 } 2693 } 2694 } 2695 2696 /* 2697 * Pull out of band byte out of a segment so 2698 * it doesn't appear in the user's data queue. 2699 * It is still reflected in the segment length for 2700 * sequencing purposes. 2701 */ 2702 static void 2703 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 2704 int off) 2705 { 2706 int cnt = off + th->th_urp - 1; 2707 2708 while (cnt >= 0) { 2709 if (m->m_len > cnt) { 2710 char *cp = mtod(m, caddr_t) + cnt; 2711 struct tcpcb *tp = sototcpcb(so); 2712 2713 tp->t_iobc = *cp; 2714 tp->t_oobflags |= TCPOOB_HAVEDATA; 2715 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2716 m->m_len--; 2717 if (m->m_flags & M_PKTHDR) 2718 m->m_pkthdr.len--; 2719 return; 2720 } 2721 cnt -= m->m_len; 2722 m = m->m_next; 2723 if (m == NULL) 2724 break; 2725 } 2726 panic("tcp_pulloutofband"); 2727 } 2728 2729 /* 2730 * Collect new round-trip time estimate 2731 * and update averages and current timeout. 2732 */ 2733 static void 2734 tcp_xmit_timer(struct tcpcb *tp, int rtt) 2735 { 2736 int delta; 2737 2738 INP_LOCK_ASSERT(tp->t_inpcb); 2739 2740 tcpstat.tcps_rttupdated++; 2741 tp->t_rttupdated++; 2742 if (tp->t_srtt != 0) { 2743 /* 2744 * srtt is stored as fixed point with 5 bits after the 2745 * binary point (i.e., scaled by 8). The following magic 2746 * is equivalent to the smoothing algorithm in rfc793 with 2747 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2748 * point). Adjust rtt to origin 0. 2749 */ 2750 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2751 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2752 2753 if ((tp->t_srtt += delta) <= 0) 2754 tp->t_srtt = 1; 2755 2756 /* 2757 * We accumulate a smoothed rtt variance (actually, a 2758 * smoothed mean difference), then set the retransmit 2759 * timer to smoothed rtt + 4 times the smoothed variance. 2760 * rttvar is stored as fixed point with 4 bits after the 2761 * binary point (scaled by 16). The following is 2762 * equivalent to rfc793 smoothing with an alpha of .75 2763 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2764 * rfc793's wired-in beta. 2765 */ 2766 if (delta < 0) 2767 delta = -delta; 2768 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2769 if ((tp->t_rttvar += delta) <= 0) 2770 tp->t_rttvar = 1; 2771 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2772 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2773 } else { 2774 /* 2775 * No rtt measurement yet - use the unsmoothed rtt. 2776 * Set the variance to half the rtt (so our first 2777 * retransmit happens at 3*rtt). 2778 */ 2779 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2780 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2781 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2782 } 2783 tp->t_rtttime = 0; 2784 tp->t_rxtshift = 0; 2785 2786 /* 2787 * the retransmit should happen at rtt + 4 * rttvar. 2788 * Because of the way we do the smoothing, srtt and rttvar 2789 * will each average +1/2 tick of bias. When we compute 2790 * the retransmit timer, we want 1/2 tick of rounding and 2791 * 1 extra tick because of +-1/2 tick uncertainty in the 2792 * firing of the timer. The bias will give us exactly the 2793 * 1.5 tick we need. But, because the bias is 2794 * statistical, we have to test that we don't drop below 2795 * the minimum feasible timer (which is 2 ticks). 2796 */ 2797 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2798 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2799 2800 /* 2801 * We received an ack for a packet that wasn't retransmitted; 2802 * it is probably safe to discard any error indications we've 2803 * received recently. This isn't quite right, but close enough 2804 * for now (a route might have failed after we sent a segment, 2805 * and the return path might not be symmetrical). 2806 */ 2807 tp->t_softerror = 0; 2808 } 2809 2810 /* 2811 * Determine a reasonable value for maxseg size. 2812 * If the route is known, check route for mtu. 2813 * If none, use an mss that can be handled on the outgoing 2814 * interface without forcing IP to fragment; if bigger than 2815 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2816 * to utilize large mbufs. If no route is found, route has no mtu, 2817 * or the destination isn't local, use a default, hopefully conservative 2818 * size (usually 512 or the default IP max size, but no more than the mtu 2819 * of the interface), as we can't discover anything about intervening 2820 * gateways or networks. We also initialize the congestion/slow start 2821 * window to be a single segment if the destination isn't local. 2822 * While looking at the routing entry, we also initialize other path-dependent 2823 * parameters from pre-set or cached values in the routing entry. 2824 * 2825 * Also take into account the space needed for options that we 2826 * send regularly. Make maxseg shorter by that amount to assure 2827 * that we can send maxseg amount of data even when the options 2828 * are present. Store the upper limit of the length of options plus 2829 * data in maxopd. 2830 * 2831 * 2832 * In case of T/TCP, we call this routine during implicit connection 2833 * setup as well (offer = -1), to initialize maxseg from the cached 2834 * MSS of our peer. 2835 * 2836 * NOTE that this routine is only called when we process an incoming 2837 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2838 */ 2839 void 2840 tcp_mss(struct tcpcb *tp, int offer) 2841 { 2842 int rtt, mss; 2843 u_long bufsize; 2844 u_long maxmtu; 2845 struct inpcb *inp = tp->t_inpcb; 2846 struct socket *so; 2847 struct hc_metrics_lite metrics; 2848 int origoffer = offer; 2849 int mtuflags = 0; 2850 #ifdef INET6 2851 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2852 size_t min_protoh = isipv6 ? 2853 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2854 sizeof (struct tcpiphdr); 2855 #else 2856 const size_t min_protoh = sizeof(struct tcpiphdr); 2857 #endif 2858 2859 /* initialize */ 2860 #ifdef INET6 2861 if (isipv6) { 2862 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags); 2863 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2864 } else 2865 #endif 2866 { 2867 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags); 2868 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2869 } 2870 so = inp->inp_socket; 2871 2872 /* 2873 * no route to sender, stay with default mss and return 2874 */ 2875 if (maxmtu == 0) 2876 return; 2877 2878 /* what have we got? */ 2879 switch (offer) { 2880 case 0: 2881 /* 2882 * Offer == 0 means that there was no MSS on the SYN 2883 * segment, in this case we use tcp_mssdflt. 2884 */ 2885 offer = 2886 #ifdef INET6 2887 isipv6 ? tcp_v6mssdflt : 2888 #endif 2889 tcp_mssdflt; 2890 break; 2891 2892 case -1: 2893 /* 2894 * Offer == -1 means that we didn't receive SYN yet. 2895 */ 2896 /* FALLTHROUGH */ 2897 2898 default: 2899 /* 2900 * Prevent DoS attack with too small MSS. Round up 2901 * to at least minmss. 2902 */ 2903 offer = max(offer, tcp_minmss); 2904 /* 2905 * Sanity check: make sure that maxopd will be large 2906 * enough to allow some data on segments even if the 2907 * all the option space is used (40bytes). Otherwise 2908 * funny things may happen in tcp_output. 2909 */ 2910 offer = max(offer, 64); 2911 } 2912 2913 /* 2914 * rmx information is now retrieved from tcp_hostcache 2915 */ 2916 tcp_hc_get(&inp->inp_inc, &metrics); 2917 2918 /* 2919 * if there's a discovered mtu int tcp hostcache, use it 2920 * else, use the link mtu. 2921 */ 2922 if (metrics.rmx_mtu) 2923 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2924 else { 2925 #ifdef INET6 2926 if (isipv6) { 2927 mss = maxmtu - min_protoh; 2928 if (!path_mtu_discovery && 2929 !in6_localaddr(&inp->in6p_faddr)) 2930 mss = min(mss, tcp_v6mssdflt); 2931 } else 2932 #endif 2933 { 2934 mss = maxmtu - min_protoh; 2935 if (!path_mtu_discovery && 2936 !in_localaddr(inp->inp_faddr)) 2937 mss = min(mss, tcp_mssdflt); 2938 } 2939 } 2940 mss = min(mss, offer); 2941 2942 /* 2943 * maxopd stores the maximum length of data AND options 2944 * in a segment; maxseg is the amount of data in a normal 2945 * segment. We need to store this value (maxopd) apart 2946 * from maxseg, because now every segment carries options 2947 * and thus we normally have somewhat less data in segments. 2948 */ 2949 tp->t_maxopd = mss; 2950 2951 /* 2952 * origoffer==-1 indicates, that no segments were received yet. 2953 * In this case we just guess. 2954 */ 2955 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2956 (origoffer == -1 || 2957 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2958 mss -= TCPOLEN_TSTAMP_APPA; 2959 tp->t_maxseg = mss; 2960 2961 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2962 if (mss > MCLBYTES) 2963 mss &= ~(MCLBYTES-1); 2964 #else 2965 if (mss > MCLBYTES) 2966 mss = mss / MCLBYTES * MCLBYTES; 2967 #endif 2968 tp->t_maxseg = mss; 2969 2970 /* 2971 * If there's a pipesize, change the socket buffer to that size, 2972 * don't change if sb_hiwat is different than default (then it 2973 * has been changed on purpose with setsockopt). 2974 * Make the socket buffers an integral number of mss units; 2975 * if the mss is larger than the socket buffer, decrease the mss. 2976 */ 2977 SOCKBUF_LOCK(&so->so_snd); 2978 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2979 bufsize = metrics.rmx_sendpipe; 2980 else 2981 bufsize = so->so_snd.sb_hiwat; 2982 if (bufsize < mss) 2983 mss = bufsize; 2984 else { 2985 bufsize = roundup(bufsize, mss); 2986 if (bufsize > sb_max) 2987 bufsize = sb_max; 2988 if (bufsize > so->so_snd.sb_hiwat) 2989 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2990 } 2991 SOCKBUF_UNLOCK(&so->so_snd); 2992 tp->t_maxseg = mss; 2993 2994 SOCKBUF_LOCK(&so->so_rcv); 2995 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2996 bufsize = metrics.rmx_recvpipe; 2997 else 2998 bufsize = so->so_rcv.sb_hiwat; 2999 if (bufsize > mss) { 3000 bufsize = roundup(bufsize, mss); 3001 if (bufsize > sb_max) 3002 bufsize = sb_max; 3003 if (bufsize > so->so_rcv.sb_hiwat) 3004 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3005 } 3006 SOCKBUF_UNLOCK(&so->so_rcv); 3007 /* 3008 * While we're here, check the others too 3009 */ 3010 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 3011 tp->t_srtt = rtt; 3012 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3013 tcpstat.tcps_usedrtt++; 3014 if (metrics.rmx_rttvar) { 3015 tp->t_rttvar = metrics.rmx_rttvar; 3016 tcpstat.tcps_usedrttvar++; 3017 } else { 3018 /* default variation is +- 1 rtt */ 3019 tp->t_rttvar = 3020 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3021 } 3022 TCPT_RANGESET(tp->t_rxtcur, 3023 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3024 tp->t_rttmin, TCPTV_REXMTMAX); 3025 } 3026 if (metrics.rmx_ssthresh) { 3027 /* 3028 * There's some sort of gateway or interface 3029 * buffer limit on the path. Use this to set 3030 * the slow start threshhold, but set the 3031 * threshold to no less than 2*mss. 3032 */ 3033 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3034 tcpstat.tcps_usedssthresh++; 3035 } 3036 if (metrics.rmx_bandwidth) 3037 tp->snd_bandwidth = metrics.rmx_bandwidth; 3038 3039 /* 3040 * Set the slow-start flight size depending on whether this 3041 * is a local network or not. 3042 * 3043 * Extend this so we cache the cwnd too and retrieve it here. 3044 * Make cwnd even bigger than RFC3390 suggests but only if we 3045 * have previous experience with the remote host. Be careful 3046 * not make cwnd bigger than remote receive window or our own 3047 * send socket buffer. Maybe put some additional upper bound 3048 * on the retrieved cwnd. Should do incremental updates to 3049 * hostcache when cwnd collapses so next connection doesn't 3050 * overloads the path again. 3051 * 3052 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3053 * We currently check only in syncache_socket for that. 3054 */ 3055 #define TCP_METRICS_CWND 3056 #ifdef TCP_METRICS_CWND 3057 if (metrics.rmx_cwnd) 3058 tp->snd_cwnd = max(mss, 3059 min(metrics.rmx_cwnd / 2, 3060 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3061 else 3062 #endif 3063 if (tcp_do_rfc3390) 3064 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3065 #ifdef INET6 3066 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3067 (!isipv6 && in_localaddr(inp->inp_faddr))) 3068 #else 3069 else if (in_localaddr(inp->inp_faddr)) 3070 #endif 3071 tp->snd_cwnd = mss * ss_fltsz_local; 3072 else 3073 tp->snd_cwnd = mss * ss_fltsz; 3074 3075 /* Check the interface for TSO capabilities. */ 3076 if (mtuflags & CSUM_TSO) 3077 tp->t_flags |= TF_TSO; 3078 } 3079 3080 /* 3081 * Determine the MSS option to send on an outgoing SYN. 3082 */ 3083 int 3084 tcp_mssopt(struct in_conninfo *inc) 3085 { 3086 int mss = 0; 3087 u_long maxmtu = 0; 3088 u_long thcmtu = 0; 3089 size_t min_protoh; 3090 #ifdef INET6 3091 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3092 #endif 3093 3094 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3095 3096 #ifdef INET6 3097 if (isipv6) { 3098 mss = tcp_v6mssdflt; 3099 maxmtu = tcp_maxmtu6(inc, NULL); 3100 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3101 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3102 } else 3103 #endif 3104 { 3105 mss = tcp_mssdflt; 3106 maxmtu = tcp_maxmtu(inc, NULL); 3107 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3108 min_protoh = sizeof(struct tcpiphdr); 3109 } 3110 if (maxmtu && thcmtu) 3111 mss = min(maxmtu, thcmtu) - min_protoh; 3112 else if (maxmtu || thcmtu) 3113 mss = max(maxmtu, thcmtu) - min_protoh; 3114 3115 return (mss); 3116 } 3117 3118 3119 /* 3120 * On a partial ack arrives, force the retransmission of the 3121 * next unacknowledged segment. Do not clear tp->t_dupacks. 3122 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3123 * be started again. 3124 */ 3125 static void 3126 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3127 { 3128 tcp_seq onxt = tp->snd_nxt; 3129 u_long ocwnd = tp->snd_cwnd; 3130 3131 tcp_timer_activate(tp, TT_REXMT, 0); 3132 tp->t_rtttime = 0; 3133 tp->snd_nxt = th->th_ack; 3134 /* 3135 * Set snd_cwnd to one segment beyond acknowledged offset. 3136 * (tp->snd_una has not yet been updated when this function is called.) 3137 */ 3138 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3139 tp->t_flags |= TF_ACKNOW; 3140 (void) tcp_output(tp); 3141 tp->snd_cwnd = ocwnd; 3142 if (SEQ_GT(onxt, tp->snd_nxt)) 3143 tp->snd_nxt = onxt; 3144 /* 3145 * Partial window deflation. Relies on fact that tp->snd_una 3146 * not updated yet. 3147 */ 3148 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3149 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3150 else 3151 tp->snd_cwnd = 0; 3152 tp->snd_cwnd += tp->t_maxseg; 3153 } 3154 3155 /* 3156 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3157 * looking for a pcb in the listen state. Returns 0 otherwise. 3158 */ 3159 static int 3160 tcp_timewait(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th, 3161 struct mbuf *m, int tlen) 3162 { 3163 struct tcptw *tw; 3164 int thflags; 3165 tcp_seq seq; 3166 #ifdef INET6 3167 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3168 #else 3169 const int isipv6 = 0; 3170 #endif 3171 3172 /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */ 3173 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3174 INP_LOCK_ASSERT(inp); 3175 3176 /* 3177 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 3178 * still present. This is undesirable, but temporarily necessary 3179 * until we work out how to handle inpcb's who's timewait state has 3180 * been removed. 3181 */ 3182 tw = intotw(inp); 3183 if (tw == NULL) 3184 goto drop; 3185 3186 thflags = th->th_flags; 3187 3188 /* 3189 * NOTE: for FIN_WAIT_2 (to be added later), 3190 * must validate sequence number before accepting RST 3191 */ 3192 3193 /* 3194 * If the segment contains RST: 3195 * Drop the segment - see Stevens, vol. 2, p. 964 and 3196 * RFC 1337. 3197 */ 3198 if (thflags & TH_RST) 3199 goto drop; 3200 3201 #if 0 3202 /* PAWS not needed at the moment */ 3203 /* 3204 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3205 * and it's less than ts_recent, drop it. 3206 */ 3207 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3208 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3209 if ((thflags & TH_ACK) == 0) 3210 goto drop; 3211 goto ack; 3212 } 3213 /* 3214 * ts_recent is never updated because we never accept new segments. 3215 */ 3216 #endif 3217 3218 /* 3219 * If a new connection request is received 3220 * while in TIME_WAIT, drop the old connection 3221 * and start over if the sequence numbers 3222 * are above the previous ones. 3223 */ 3224 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3225 tcp_twclose(tw, 0); 3226 return (1); 3227 } 3228 3229 /* 3230 * Drop the the segment if it does not contain an ACK. 3231 */ 3232 if ((thflags & TH_ACK) == 0) 3233 goto drop; 3234 3235 /* 3236 * Reset the 2MSL timer if this is a duplicate FIN. 3237 */ 3238 if (thflags & TH_FIN) { 3239 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3240 if (seq + 1 == tw->rcv_nxt) 3241 tcp_timer_2msl_reset(tw, 1); 3242 } 3243 3244 /* 3245 * Acknowledge the segment if it has data or is not a duplicate ACK. 3246 */ 3247 if (thflags != TH_ACK || tlen != 0 || 3248 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3249 tcp_twrespond(tw, TH_ACK); 3250 goto drop; 3251 3252 /* 3253 * Generate a RST, dropping incoming segment. 3254 * Make ACK acceptable to originator of segment. 3255 * Don't bother to respond if destination was broadcast/multicast. 3256 */ 3257 if (m->m_flags & (M_BCAST|M_MCAST)) 3258 goto drop; 3259 if (isipv6) { 3260 struct ip6_hdr *ip6; 3261 3262 /* IPv6 anycast check is done at tcp6_input() */ 3263 ip6 = mtod(m, struct ip6_hdr *); 3264 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3265 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3266 goto drop; 3267 } else { 3268 struct ip *ip; 3269 3270 ip = mtod(m, struct ip *); 3271 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3272 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3273 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3274 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3275 goto drop; 3276 } 3277 if (thflags & TH_ACK) { 3278 tcp_respond(NULL, 3279 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3280 } else { 3281 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3282 tcp_respond(NULL, 3283 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3284 } 3285 INP_UNLOCK(inp); 3286 return (0); 3287 3288 drop: 3289 INP_UNLOCK(inp); 3290 m_freem(m); 3291 return (0); 3292 } 3293