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 int 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() 228 { 229 tcp_reass_maxseg = nmbclusters / 16; 230 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 231 &tcp_reass_maxseg); 232 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 233 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 234 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 235 EVENTHANDLER_REGISTER(nmbclusters_change, 236 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY); 237 } 238 239 static int 240 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m) 241 { 242 struct tseg_qent *q; 243 struct tseg_qent *p = NULL; 244 struct tseg_qent *nq; 245 struct tseg_qent *te = NULL; 246 struct socket *so = tp->t_inpcb->inp_socket; 247 int flags; 248 249 INP_LOCK_ASSERT(tp->t_inpcb); 250 251 /* 252 * XXX: tcp_reass() is rather inefficient with its data structures 253 * and should be rewritten (see NetBSD for optimizations). While 254 * doing that it should move to its own file tcp_reass.c. 255 */ 256 257 /* 258 * Call with th==NULL after become established to 259 * force pre-ESTABLISHED data up to user socket. 260 */ 261 if (th == NULL) 262 goto present; 263 264 /* 265 * Limit the number of segments in the reassembly queue to prevent 266 * holding on to too many segments (and thus running out of mbufs). 267 * Make sure to let the missing segment through which caused this 268 * queue. Always keep one global queue entry spare to be able to 269 * process the missing segment. 270 */ 271 if (th->th_seq != tp->rcv_nxt && 272 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 273 tp->t_segqlen >= tcp_reass_maxqlen)) { 274 tcp_reass_overflows++; 275 tcpstat.tcps_rcvmemdrop++; 276 m_freem(m); 277 *tlenp = 0; 278 return (0); 279 } 280 281 /* 282 * Allocate a new queue entry. If we can't, or hit the zone limit 283 * just drop the pkt. 284 */ 285 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 286 if (te == NULL) { 287 tcpstat.tcps_rcvmemdrop++; 288 m_freem(m); 289 *tlenp = 0; 290 return (0); 291 } 292 tp->t_segqlen++; 293 tcp_reass_qsize++; 294 295 /* 296 * Find a segment which begins after this one does. 297 */ 298 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 299 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 300 break; 301 p = q; 302 } 303 304 /* 305 * If there is a preceding segment, it may provide some of 306 * our data already. If so, drop the data from the incoming 307 * segment. If it provides all of our data, drop us. 308 */ 309 if (p != NULL) { 310 int i; 311 /* conversion to int (in i) handles seq wraparound */ 312 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 313 if (i > 0) { 314 if (i >= *tlenp) { 315 tcpstat.tcps_rcvduppack++; 316 tcpstat.tcps_rcvdupbyte += *tlenp; 317 m_freem(m); 318 uma_zfree(tcp_reass_zone, te); 319 tp->t_segqlen--; 320 tcp_reass_qsize--; 321 /* 322 * Try to present any queued data 323 * at the left window edge to the user. 324 * This is needed after the 3-WHS 325 * completes. 326 */ 327 goto present; /* ??? */ 328 } 329 m_adj(m, i); 330 *tlenp -= i; 331 th->th_seq += i; 332 } 333 } 334 tcpstat.tcps_rcvoopack++; 335 tcpstat.tcps_rcvoobyte += *tlenp; 336 337 /* 338 * While we overlap succeeding segments trim them or, 339 * if they are completely covered, dequeue them. 340 */ 341 while (q) { 342 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 343 if (i <= 0) 344 break; 345 if (i < q->tqe_len) { 346 q->tqe_th->th_seq += i; 347 q->tqe_len -= i; 348 m_adj(q->tqe_m, i); 349 break; 350 } 351 352 nq = LIST_NEXT(q, tqe_q); 353 LIST_REMOVE(q, tqe_q); 354 m_freem(q->tqe_m); 355 uma_zfree(tcp_reass_zone, q); 356 tp->t_segqlen--; 357 tcp_reass_qsize--; 358 q = nq; 359 } 360 361 /* Insert the new segment queue entry into place. */ 362 te->tqe_m = m; 363 te->tqe_th = th; 364 te->tqe_len = *tlenp; 365 366 if (p == NULL) { 367 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 368 } else { 369 LIST_INSERT_AFTER(p, te, tqe_q); 370 } 371 372 present: 373 /* 374 * Present data to user, advancing rcv_nxt through 375 * completed sequence space. 376 */ 377 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 378 return (0); 379 q = LIST_FIRST(&tp->t_segq); 380 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 381 return (0); 382 SOCKBUF_LOCK(&so->so_rcv); 383 do { 384 tp->rcv_nxt += q->tqe_len; 385 flags = q->tqe_th->th_flags & TH_FIN; 386 nq = LIST_NEXT(q, tqe_q); 387 LIST_REMOVE(q, tqe_q); 388 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 389 m_freem(q->tqe_m); 390 else 391 sbappendstream_locked(&so->so_rcv, q->tqe_m); 392 uma_zfree(tcp_reass_zone, q); 393 tp->t_segqlen--; 394 tcp_reass_qsize--; 395 q = nq; 396 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 397 ND6_HINT(tp); 398 sorwakeup_locked(so); 399 return (flags); 400 } 401 402 /* 403 * TCP input routine, follows pages 65-76 of the 404 * protocol specification dated September, 1981 very closely. 405 */ 406 #ifdef INET6 407 int 408 tcp6_input(struct mbuf **mp, int *offp, int proto) 409 { 410 struct mbuf *m = *mp; 411 struct in6_ifaddr *ia6; 412 413 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 414 415 /* 416 * draft-itojun-ipv6-tcp-to-anycast 417 * better place to put this in? 418 */ 419 ia6 = ip6_getdstifaddr(m); 420 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 421 struct ip6_hdr *ip6; 422 423 ip6 = mtod(m, struct ip6_hdr *); 424 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 425 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 426 return IPPROTO_DONE; 427 } 428 429 tcp_input(m, *offp); 430 return IPPROTO_DONE; 431 } 432 #endif 433 434 void 435 tcp_input(struct mbuf *m, int off0) 436 { 437 struct tcphdr *th; 438 struct ip *ip = NULL; 439 struct ipovly *ipov; 440 struct inpcb *inp = NULL; 441 struct tcpcb *tp = NULL; 442 struct socket *so = NULL; 443 u_char *optp = NULL; 444 int optlen = 0; 445 int len, tlen, off; 446 int drop_hdrlen; 447 int thflags; 448 int rstreason = 0; /* For badport_bandlim accounting purposes */ 449 #ifdef IPFIREWALL_FORWARD 450 struct m_tag *fwd_tag; 451 #endif 452 #ifdef INET6 453 struct ip6_hdr *ip6 = NULL; 454 int isipv6; 455 char ip6buf[INET6_ADDRSTRLEN]; 456 #else 457 const int isipv6 = 0; 458 #endif 459 struct tcpopt to; /* options in this segment */ 460 461 #ifdef TCPDEBUG 462 /* 463 * The size of tcp_saveipgen must be the size of the max ip header, 464 * now IPv6. 465 */ 466 u_char tcp_saveipgen[IP6_HDR_LEN]; 467 struct tcphdr tcp_savetcp; 468 short ostate = 0; 469 #endif 470 471 #ifdef INET6 472 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 473 #endif 474 475 to.to_flags = 0; 476 tcpstat.tcps_rcvtotal++; 477 478 if (isipv6) { 479 #ifdef INET6 480 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 481 ip6 = mtod(m, struct ip6_hdr *); 482 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 483 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 484 tcpstat.tcps_rcvbadsum++; 485 goto drop; 486 } 487 th = (struct tcphdr *)((caddr_t)ip6 + off0); 488 489 /* 490 * Be proactive about unspecified IPv6 address in source. 491 * As we use all-zero to indicate unbounded/unconnected pcb, 492 * unspecified IPv6 address can be used to confuse us. 493 * 494 * Note that packets with unspecified IPv6 destination is 495 * already dropped in ip6_input. 496 */ 497 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 498 /* XXX stat */ 499 goto drop; 500 } 501 #else 502 th = NULL; /* XXX: avoid compiler warning */ 503 #endif 504 } else { 505 /* 506 * Get IP and TCP header together in first mbuf. 507 * Note: IP leaves IP header in first mbuf. 508 */ 509 if (off0 > sizeof (struct ip)) { 510 ip_stripoptions(m, (struct mbuf *)0); 511 off0 = sizeof(struct ip); 512 } 513 if (m->m_len < sizeof (struct tcpiphdr)) { 514 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 515 == NULL) { 516 tcpstat.tcps_rcvshort++; 517 return; 518 } 519 } 520 ip = mtod(m, struct ip *); 521 ipov = (struct ipovly *)ip; 522 th = (struct tcphdr *)((caddr_t)ip + off0); 523 tlen = ip->ip_len; 524 525 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 526 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 527 th->th_sum = m->m_pkthdr.csum_data; 528 else 529 th->th_sum = in_pseudo(ip->ip_src.s_addr, 530 ip->ip_dst.s_addr, 531 htonl(m->m_pkthdr.csum_data + 532 ip->ip_len + 533 IPPROTO_TCP)); 534 th->th_sum ^= 0xffff; 535 #ifdef TCPDEBUG 536 ipov->ih_len = (u_short)tlen; 537 ipov->ih_len = htons(ipov->ih_len); 538 #endif 539 } else { 540 /* 541 * Checksum extended TCP header and data. 542 */ 543 len = sizeof (struct ip) + tlen; 544 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 545 ipov->ih_len = (u_short)tlen; 546 ipov->ih_len = htons(ipov->ih_len); 547 th->th_sum = in_cksum(m, len); 548 } 549 if (th->th_sum) { 550 tcpstat.tcps_rcvbadsum++; 551 goto drop; 552 } 553 /* Re-initialization for later version check */ 554 ip->ip_v = IPVERSION; 555 } 556 557 /* 558 * Check that TCP offset makes sense, 559 * pull out TCP options and adjust length. XXX 560 */ 561 off = th->th_off << 2; 562 if (off < sizeof (struct tcphdr) || off > tlen) { 563 tcpstat.tcps_rcvbadoff++; 564 goto drop; 565 } 566 tlen -= off; /* tlen is used instead of ti->ti_len */ 567 if (off > sizeof (struct tcphdr)) { 568 if (isipv6) { 569 #ifdef INET6 570 IP6_EXTHDR_CHECK(m, off0, off, ); 571 ip6 = mtod(m, struct ip6_hdr *); 572 th = (struct tcphdr *)((caddr_t)ip6 + off0); 573 #endif 574 } else { 575 if (m->m_len < sizeof(struct ip) + off) { 576 if ((m = m_pullup(m, sizeof (struct ip) + off)) 577 == NULL) { 578 tcpstat.tcps_rcvshort++; 579 return; 580 } 581 ip = mtod(m, struct ip *); 582 ipov = (struct ipovly *)ip; 583 th = (struct tcphdr *)((caddr_t)ip + off0); 584 } 585 } 586 optlen = off - sizeof (struct tcphdr); 587 optp = (u_char *)(th + 1); 588 } 589 thflags = th->th_flags; 590 591 /* 592 * If the drop_synfin option is enabled, drop all packets with 593 * both the SYN and FIN bits set. This prevents e.g. nmap from 594 * identifying the TCP/IP stack. 595 * 596 * This is a violation of the TCP specification. 597 */ 598 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 599 goto drop; 600 601 /* 602 * Convert TCP protocol specific fields to host format. 603 */ 604 th->th_seq = ntohl(th->th_seq); 605 th->th_ack = ntohl(th->th_ack); 606 th->th_win = ntohs(th->th_win); 607 th->th_urp = ntohs(th->th_urp); 608 609 /* 610 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 611 */ 612 drop_hdrlen = off0 + off; 613 614 /* 615 * Locate pcb for segment. 616 */ 617 INP_INFO_WLOCK(&tcbinfo); 618 findpcb: 619 INP_INFO_WLOCK_ASSERT(&tcbinfo); 620 #ifdef IPFIREWALL_FORWARD 621 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 622 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 623 624 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 625 struct sockaddr_in *next_hop; 626 627 next_hop = (struct sockaddr_in *)(fwd_tag+1); 628 /* 629 * Transparently forwarded. Pretend to be the destination. 630 * already got one like this? 631 */ 632 inp = in_pcblookup_hash(&tcbinfo, 633 ip->ip_src, th->th_sport, 634 ip->ip_dst, th->th_dport, 635 0, m->m_pkthdr.rcvif); 636 if (!inp) { 637 /* It's new. Try to find the ambushing socket. */ 638 inp = in_pcblookup_hash(&tcbinfo, 639 ip->ip_src, th->th_sport, 640 next_hop->sin_addr, 641 next_hop->sin_port ? 642 ntohs(next_hop->sin_port) : 643 th->th_dport, 644 INPLOOKUP_WILDCARD, 645 m->m_pkthdr.rcvif); 646 } 647 /* Remove the tag from the packet. We don't need it anymore. */ 648 m_tag_delete(m, fwd_tag); 649 } else 650 #endif /* IPFIREWALL_FORWARD */ 651 { 652 if (isipv6) { 653 #ifdef INET6 654 inp = in6_pcblookup_hash(&tcbinfo, 655 &ip6->ip6_src, th->th_sport, 656 &ip6->ip6_dst, th->th_dport, 657 INPLOOKUP_WILDCARD, 658 m->m_pkthdr.rcvif); 659 #endif 660 } else 661 inp = in_pcblookup_hash(&tcbinfo, 662 ip->ip_src, th->th_sport, 663 ip->ip_dst, th->th_dport, 664 INPLOOKUP_WILDCARD, 665 m->m_pkthdr.rcvif); 666 } 667 668 #if defined(IPSEC) || defined(FAST_IPSEC) 669 #ifdef INET6 670 if (isipv6 && inp != NULL && ipsec6_in_reject(m, inp)) { 671 #ifdef IPSEC 672 ipsec6stat.in_polvio++; 673 #endif 674 goto dropunlock; 675 } else 676 #endif /* INET6 */ 677 if (inp != NULL && ipsec4_in_reject(m, inp)) { 678 #ifdef IPSEC 679 ipsecstat.in_polvio++; 680 #endif 681 goto dropunlock; 682 } 683 #endif /*IPSEC || FAST_IPSEC*/ 684 685 /* 686 * If the INPCB does not exist then all data in the incoming 687 * segment is discarded and an appropriate RST is sent back. 688 */ 689 if (inp == NULL) { 690 /* 691 * Log communication attempts to ports that are not 692 * in use. 693 */ 694 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 695 tcp_log_in_vain == 2) { 696 #ifndef INET6 697 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 698 #else 699 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 700 if (isipv6) { 701 strcpy(dbuf, "["); 702 strcat(dbuf, 703 ip6_sprintf(ip6buf, &ip6->ip6_dst)); 704 strcat(dbuf, "]"); 705 strcpy(sbuf, "["); 706 strcat(sbuf, 707 ip6_sprintf(ip6buf, &ip6->ip6_src)); 708 strcat(sbuf, "]"); 709 } else 710 #endif /* INET6 */ 711 { 712 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 713 strcpy(sbuf, inet_ntoa(ip->ip_src)); 714 } 715 log(LOG_INFO, 716 "Connection attempt to TCP %s:%d " 717 "from %s:%d flags:0x%02x\n", 718 dbuf, ntohs(th->th_dport), sbuf, 719 ntohs(th->th_sport), thflags); 720 } 721 /* 722 * When blackholing do not respond with a RST but 723 * completely ignore the segment and drop it. 724 */ 725 if ((blackhole == 1 && (thflags & TH_SYN)) || 726 blackhole == 2) 727 goto dropunlock; 728 729 rstreason = BANDLIM_RST_CLOSEDPORT; 730 goto dropwithreset; 731 } 732 INP_LOCK(inp); 733 734 /* Check the minimum TTL for socket. */ 735 if (inp->inp_ip_minttl != 0) { 736 #ifdef INET6 737 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 738 goto dropunlock; 739 else 740 #endif 741 if (inp->inp_ip_minttl > ip->ip_ttl) 742 goto dropunlock; 743 } 744 745 /* 746 * A previous connection in TIMEWAIT state is supposed to catch 747 * stray or duplicate segments arriving late. If this segment 748 * was a legitimate new connection attempt the old INPCB gets 749 * removed and we can try again to find a listening socket. 750 */ 751 if (inp->inp_vflag & INP_TIMEWAIT) { 752 if (thflags & TH_SYN) 753 tcp_dooptions(&to, optp, optlen, TO_SYN); 754 if (tcp_timewait(inp, &to, th, m, tlen)) 755 goto findpcb; 756 /* tcp_timewait unlocks inp. */ 757 INP_INFO_WUNLOCK(&tcbinfo); 758 return; 759 } 760 /* 761 * The TCPCB may no longer exist if the connection is winding 762 * down or it is in the CLOSED state. Either way we drop the 763 * segment and send an appropriate response. 764 */ 765 tp = intotcpcb(inp); 766 if (tp == NULL) { 767 INP_UNLOCK(inp); 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 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 786 else 787 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 788 tcp_savetcp = *th; 789 } 790 #endif 791 /* 792 * When the socket is accepting connections (the INPCB is in LISTEN 793 * state) we look into the SYN cache if this is a new connection 794 * attempt or the completion of a previous one. 795 */ 796 if (so->so_options & SO_ACCEPTCONN) { 797 struct in_conninfo inc; 798 799 bzero(&inc, sizeof(inc)); 800 inc.inc_isipv6 = isipv6; 801 #ifdef INET6 802 if (isipv6) { 803 inc.inc6_faddr = ip6->ip6_src; 804 inc.inc6_laddr = ip6->ip6_dst; 805 } else 806 #endif 807 { 808 inc.inc_faddr = ip->ip_src; 809 inc.inc_laddr = ip->ip_dst; 810 } 811 inc.inc_fport = th->th_sport; 812 inc.inc_lport = th->th_dport; 813 814 /* 815 * If the state is LISTEN then ignore segment if it contains 816 * a RST. If the segment contains an ACK then it is bad and 817 * send a RST. If it does not contain a SYN then it is not 818 * interesting; drop it. 819 * 820 * If the state is SYN_RECEIVED (syncache) and seg contains 821 * an ACK, but not for our SYN/ACK, send a RST. If the seg 822 * contains a RST, check the sequence number to see if it 823 * is a valid reset segment. 824 */ 825 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 826 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 827 /* 828 * Parse the TCP options here because 829 * syncookies need access to the reflected 830 * timestamp. 831 */ 832 tcp_dooptions(&to, optp, optlen, 0); 833 if (!syncache_expand(&inc, &to, th, &so, m)) { 834 /* 835 * No syncache entry, or ACK was not 836 * for our SYN/ACK. Send a RST. 837 */ 838 tcpstat.tcps_badsyn++; 839 rstreason = BANDLIM_RST_OPENPORT; 840 goto dropwithreset; 841 } 842 if (so == NULL) { 843 /* 844 * Could not complete 3-way handshake, 845 * connection is being closed down, and 846 * syncache has free'd mbuf. 847 */ 848 INP_UNLOCK(inp); 849 INP_INFO_WUNLOCK(&tcbinfo); 850 return; 851 } 852 /* 853 * Socket is created in state SYN_RECEIVED. 854 * Continue processing segment. 855 */ 856 INP_UNLOCK(inp); /* listen socket */ 857 inp = sotoinpcb(so); 858 INP_LOCK(inp); /* new connection */ 859 tp = intotcpcb(inp); 860 /* 861 * Process the segment and the data it 862 * contains. tcp_do_segment() consumes 863 * the mbuf chain and unlocks the inpcb. 864 * XXX: The potential return value of 865 * TIME_WAIT nuked is supposed to be 866 * handled above. 867 */ 868 if (tcp_do_segment(m, th, so, tp, 869 drop_hdrlen, tlen)) 870 goto findpcb; /* TIME_WAIT nuked */ 871 return; 872 } 873 if (thflags & TH_RST) { 874 syncache_chkrst(&inc, th); 875 goto dropunlock; 876 } 877 if (thflags & TH_ACK) { 878 syncache_badack(&inc); 879 tcpstat.tcps_badsyn++; 880 rstreason = BANDLIM_RST_OPENPORT; 881 goto dropwithreset; 882 } 883 goto dropunlock; 884 } 885 886 /* 887 * Segment's flags are (SYN) or (SYN|FIN). 888 */ 889 #ifdef INET6 890 /* 891 * If deprecated address is forbidden, 892 * we do not accept SYN to deprecated interface 893 * address to prevent any new inbound connection from 894 * getting established. 895 * When we do not accept SYN, we send a TCP RST, 896 * with deprecated source address (instead of dropping 897 * it). We compromise it as it is much better for peer 898 * to send a RST, and RST will be the final packet 899 * for the exchange. 900 * 901 * If we do not forbid deprecated addresses, we accept 902 * the SYN packet. RFC2462 does not suggest dropping 903 * SYN in this case. 904 * If we decipher RFC2462 5.5.4, it says like this: 905 * 1. use of deprecated addr with existing 906 * communication is okay - "SHOULD continue to be 907 * used" 908 * 2. use of it with new communication: 909 * (2a) "SHOULD NOT be used if alternate address 910 * with sufficient scope is available" 911 * (2b) nothing mentioned otherwise. 912 * Here we fall into (2b) case as we have no choice in 913 * our source address selection - we must obey the peer. 914 * 915 * The wording in RFC2462 is confusing, and there are 916 * multiple description text for deprecated address 917 * handling - worse, they are not exactly the same. 918 * I believe 5.5.4 is the best one, so we follow 5.5.4. 919 */ 920 if (isipv6 && !ip6_use_deprecated) { 921 struct in6_ifaddr *ia6; 922 923 if ((ia6 = ip6_getdstifaddr(m)) && 924 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 925 INP_UNLOCK(inp); 926 tp = NULL; 927 rstreason = BANDLIM_RST_OPENPORT; 928 goto dropwithreset; 929 } 930 } 931 #endif 932 /* 933 * Basic sanity checks on incoming SYN requests: 934 * 935 * Don't bother responding if the destination was a 936 * broadcast according to RFC1122 4.2.3.10, p. 104. 937 * 938 * If it is from this socket, drop it, it must be forged. 939 * 940 * Note that it is quite possible to receive unicast 941 * link-layer packets with a broadcast IP address. Use 942 * in_broadcast() to find them. 943 */ 944 if (m->m_flags & (M_BCAST|M_MCAST)) 945 goto dropunlock; 946 if (isipv6) { 947 #ifdef INET6 948 if (th->th_dport == th->th_sport && 949 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) 950 goto dropunlock; 951 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 952 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 953 goto dropunlock; 954 #endif 955 } else { 956 if (th->th_dport == th->th_sport && 957 ip->ip_dst.s_addr == ip->ip_src.s_addr) 958 goto dropunlock; 959 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 960 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 961 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 962 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 963 goto dropunlock; 964 } 965 /* 966 * SYN appears to be valid. Create compressed TCP state 967 * for syncache. 968 */ 969 if (so->so_qlen <= so->so_qlimit) { 970 #ifdef TCPDEBUG 971 if (so->so_options & SO_DEBUG) 972 tcp_trace(TA_INPUT, ostate, tp, 973 (void *)tcp_saveipgen, &tcp_savetcp, 0); 974 #endif 975 tcp_dooptions(&to, optp, optlen, TO_SYN); 976 if (!syncache_add(&inc, &to, th, inp, &so, m)) 977 goto dropunlock; 978 /* 979 * Entry added to syncache, mbuf used to 980 * send SYN-ACK packet. Everything unlocked 981 * already. 982 */ 983 return; 984 } 985 /* Catch all. Everthing that makes it down here is junk. */ 986 goto dropunlock; 987 } 988 989 /* 990 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or 991 * later state. tcp_do_segment() always consumes the mbuf chain 992 * and unlocks the inpcb. 993 */ 994 if (tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen)) 995 goto findpcb; /* XXX: TIME_WAIT was nuked. */ 996 return; 997 998 dropwithreset: 999 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1000 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1001 m = NULL; /* mbuf chain got consumed. */ 1002 dropunlock: 1003 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1004 if (tp != NULL) 1005 INP_UNLOCK(inp); 1006 INP_INFO_WUNLOCK(&tcbinfo); 1007 drop: 1008 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 1009 if (m != NULL) 1010 m_freem(m); 1011 return; 1012 } 1013 1014 static int 1015 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1016 struct tcpcb *tp, int drop_hdrlen, int tlen) 1017 { 1018 int thflags, acked, ourfinisacked, needoutput = 0; 1019 int headlocked = 1; 1020 int rstreason, todrop, win; 1021 u_long tiwin; 1022 struct tcpopt to; 1023 1024 #ifdef TCPDEBUG 1025 /* 1026 * The size of tcp_saveipgen must be the size of the max ip header, 1027 * now IPv6. 1028 */ 1029 u_char tcp_saveipgen[IP6_HDR_LEN]; 1030 struct tcphdr tcp_savetcp; 1031 short ostate = 0; 1032 #endif 1033 thflags = th->th_flags; 1034 1035 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1036 INP_LOCK_ASSERT(tp->t_inpcb); 1037 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", __func__)); 1038 1039 /* 1040 * Segment received on connection. 1041 * Reset idle time and keep-alive timer. 1042 */ 1043 tp->t_rcvtime = ticks; 1044 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1045 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1046 1047 /* 1048 * Unscale the window into a 32-bit value. 1049 * This value is bogus for the TCPS_SYN_SENT state 1050 * and is overwritten later. 1051 */ 1052 tiwin = th->th_win << tp->snd_scale; 1053 1054 /* 1055 * Parse options on any incoming segment. 1056 */ 1057 tcp_dooptions(&to, (u_char *)(th + 1), 1058 (th->th_off << 2) - sizeof(struct tcphdr), 1059 (thflags & TH_SYN) ? TO_SYN : 0); 1060 1061 /* 1062 * If echoed timestamp is later than the current time, 1063 * fall back to non RFC1323 RTT calculation. Normalize 1064 * timestamp if syncookies were used when this connection 1065 * was established. 1066 */ 1067 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1068 to.to_tsecr -= tp->ts_offset; 1069 if (TSTMP_GT(to.to_tsecr, ticks)) 1070 to.to_tsecr = 0; 1071 } 1072 1073 /* 1074 * Process options only when we get SYN/ACK back. The SYN case 1075 * for incoming connections is handled in tcp_syncache. 1076 * XXX this is traditional behavior, may need to be cleaned up. 1077 */ 1078 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1079 if ((to.to_flags & TOF_SCALE) && 1080 (tp->t_flags & TF_REQ_SCALE)) { 1081 tp->t_flags |= TF_RCVD_SCALE; 1082 tp->snd_scale = to.to_wscale; 1083 tp->snd_wnd = th->th_win << tp->snd_scale; 1084 tiwin = tp->snd_wnd; 1085 } 1086 if (to.to_flags & TOF_TS) { 1087 tp->t_flags |= TF_RCVD_TSTMP; 1088 tp->ts_recent = to.to_tsval; 1089 tp->ts_recent_age = ticks; 1090 } 1091 /* Initial send window, already scaled. */ 1092 tp->snd_wnd = th->th_win; 1093 if (to.to_flags & TOF_MSS) 1094 tcp_mss(tp, to.to_mss); 1095 if (tp->sack_enable) { 1096 if (!(to.to_flags & TOF_SACKPERM)) 1097 tp->sack_enable = 0; 1098 else 1099 tp->t_flags |= TF_SACK_PERMIT; 1100 } 1101 1102 } 1103 1104 /* 1105 * Header prediction: check for the two common cases 1106 * of a uni-directional data xfer. If the packet has 1107 * no control flags, is in-sequence, the window didn't 1108 * change and we're not retransmitting, it's a 1109 * candidate. If the length is zero and the ack moved 1110 * forward, we're the sender side of the xfer. Just 1111 * free the data acked & wake any higher level process 1112 * that was blocked waiting for space. If the length 1113 * is non-zero and the ack didn't move, we're the 1114 * receiver side. If we're getting packets in-order 1115 * (the reassembly queue is empty), add the data to 1116 * the socket buffer and note that we need a delayed ack. 1117 * Make sure that the hidden state-flags are also off. 1118 * Since we check for TCPS_ESTABLISHED above, it can only 1119 * be TH_NEEDSYN. 1120 */ 1121 if (tp->t_state == TCPS_ESTABLISHED && 1122 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1123 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1124 ((to.to_flags & TOF_TS) == 0 || 1125 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1126 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1127 tp->snd_nxt == tp->snd_max) { 1128 1129 /* 1130 * If last ACK falls within this segment's sequence numbers, 1131 * record the timestamp. 1132 * NOTE that the test is modified according to the latest 1133 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1134 */ 1135 if ((to.to_flags & TOF_TS) != 0 && 1136 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1137 tp->ts_recent_age = ticks; 1138 tp->ts_recent = to.to_tsval; 1139 } 1140 1141 if (tlen == 0) { 1142 if (SEQ_GT(th->th_ack, tp->snd_una) && 1143 SEQ_LEQ(th->th_ack, tp->snd_max) && 1144 tp->snd_cwnd >= tp->snd_wnd && 1145 ((!tcp_do_newreno && !tp->sack_enable && 1146 tp->t_dupacks < tcprexmtthresh) || 1147 ((tcp_do_newreno || tp->sack_enable) && 1148 !IN_FASTRECOVERY(tp) && 1149 (to.to_flags & TOF_SACK) == 0 && 1150 TAILQ_EMPTY(&tp->snd_holes)))) { 1151 KASSERT(headlocked, 1152 ("%s: headlocked", __func__)); 1153 INP_INFO_WUNLOCK(&tcbinfo); 1154 headlocked = 0; 1155 /* 1156 * this is a pure ack for outstanding data. 1157 */ 1158 ++tcpstat.tcps_predack; 1159 /* 1160 * "bad retransmit" recovery 1161 */ 1162 if (tp->t_rxtshift == 1 && 1163 ticks < tp->t_badrxtwin) { 1164 ++tcpstat.tcps_sndrexmitbad; 1165 tp->snd_cwnd = tp->snd_cwnd_prev; 1166 tp->snd_ssthresh = 1167 tp->snd_ssthresh_prev; 1168 tp->snd_recover = tp->snd_recover_prev; 1169 if (tp->t_flags & TF_WASFRECOVERY) 1170 ENTER_FASTRECOVERY(tp); 1171 tp->snd_nxt = tp->snd_max; 1172 tp->t_badrxtwin = 0; 1173 } 1174 1175 /* 1176 * Recalculate the transmit timer / rtt. 1177 * 1178 * Some boxes send broken timestamp replies 1179 * during the SYN+ACK phase, ignore 1180 * timestamps of 0 or we could calculate a 1181 * huge RTT and blow up the retransmit timer. 1182 */ 1183 if ((to.to_flags & TOF_TS) != 0 && 1184 to.to_tsecr) { 1185 if (!tp->t_rttlow || 1186 tp->t_rttlow > ticks - to.to_tsecr) 1187 tp->t_rttlow = ticks - to.to_tsecr; 1188 tcp_xmit_timer(tp, 1189 ticks - to.to_tsecr + 1); 1190 } else if (tp->t_rtttime && 1191 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1192 if (!tp->t_rttlow || 1193 tp->t_rttlow > ticks - tp->t_rtttime) 1194 tp->t_rttlow = ticks - tp->t_rtttime; 1195 tcp_xmit_timer(tp, 1196 ticks - tp->t_rtttime); 1197 } 1198 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1199 acked = th->th_ack - tp->snd_una; 1200 tcpstat.tcps_rcvackpack++; 1201 tcpstat.tcps_rcvackbyte += acked; 1202 sbdrop(&so->so_snd, acked); 1203 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1204 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1205 tp->snd_recover = th->th_ack - 1; 1206 tp->snd_una = th->th_ack; 1207 /* 1208 * pull snd_wl2 up to prevent seq wrap relative 1209 * to th_ack. 1210 */ 1211 tp->snd_wl2 = th->th_ack; 1212 tp->t_dupacks = 0; 1213 m_freem(m); 1214 ND6_HINT(tp); /* some progress has been done */ 1215 1216 /* 1217 * If all outstanding data are acked, stop 1218 * retransmit timer, otherwise restart timer 1219 * using current (possibly backed-off) value. 1220 * If process is waiting for space, 1221 * wakeup/selwakeup/signal. If data 1222 * are ready to send, let tcp_output 1223 * decide between more output or persist. 1224 1225 #ifdef TCPDEBUG 1226 if (so->so_options & SO_DEBUG) 1227 tcp_trace(TA_INPUT, ostate, tp, 1228 (void *)tcp_saveipgen, 1229 &tcp_savetcp, 0); 1230 #endif 1231 */ 1232 if (tp->snd_una == tp->snd_max) 1233 tcp_timer_activate(tp, TT_REXMT, 0); 1234 else if (!tcp_timer_active(tp, TT_PERSIST)) 1235 tcp_timer_activate(tp, TT_REXMT, 1236 tp->t_rxtcur); 1237 1238 sowwakeup(so); 1239 if (so->so_snd.sb_cc) 1240 (void) tcp_output(tp); 1241 goto check_delack; 1242 } 1243 } else if (th->th_ack == tp->snd_una && 1244 LIST_EMPTY(&tp->t_segq) && 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->sack_enable && 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_RST) { 1410 if (thflags & TH_ACK) { 1411 KASSERT(headlocked, ("%s: after_listen: " 1412 "tcp_drop.2: head not locked", __func__)); 1413 tp = tcp_drop(tp, ECONNREFUSED); 1414 } 1415 goto drop; 1416 } 1417 if ((thflags & TH_SYN) == 0) 1418 goto drop; 1419 1420 tp->irs = th->th_seq; 1421 tcp_rcvseqinit(tp); 1422 if (thflags & TH_ACK) { 1423 tcpstat.tcps_connects++; 1424 soisconnected(so); 1425 #ifdef MAC 1426 SOCK_LOCK(so); 1427 mac_set_socket_peer_from_mbuf(m, so); 1428 SOCK_UNLOCK(so); 1429 #endif 1430 /* Do window scaling on this connection? */ 1431 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1432 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1433 tp->rcv_scale = tp->request_r_scale; 1434 } 1435 tp->rcv_adv += tp->rcv_wnd; 1436 tp->snd_una++; /* SYN is acked */ 1437 /* 1438 * If there's data, delay ACK; if there's also a FIN 1439 * ACKNOW will be turned on later. 1440 */ 1441 if (DELAY_ACK(tp) && tlen != 0) 1442 tcp_timer_activate(tp, TT_DELACK, 1443 tcp_delacktime); 1444 else 1445 tp->t_flags |= TF_ACKNOW; 1446 /* 1447 * Received <SYN,ACK> in SYN_SENT[*] state. 1448 * Transitions: 1449 * SYN_SENT --> ESTABLISHED 1450 * SYN_SENT* --> FIN_WAIT_1 1451 */ 1452 tp->t_starttime = ticks; 1453 if (tp->t_flags & TF_NEEDFIN) { 1454 tp->t_state = TCPS_FIN_WAIT_1; 1455 tp->t_flags &= ~TF_NEEDFIN; 1456 thflags &= ~TH_SYN; 1457 } else { 1458 tp->t_state = TCPS_ESTABLISHED; 1459 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1460 } 1461 } else { 1462 /* 1463 * Received initial SYN in SYN-SENT[*] state => 1464 * simultaneous open. If segment contains CC option 1465 * and there is a cached CC, apply TAO test. 1466 * If it succeeds, connection is * half-synchronized. 1467 * Otherwise, do 3-way handshake: 1468 * SYN-SENT -> SYN-RECEIVED 1469 * SYN-SENT* -> SYN-RECEIVED* 1470 * If there was no CC option, clear cached CC value. 1471 */ 1472 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1473 tcp_timer_activate(tp, TT_REXMT, 0); 1474 tp->t_state = TCPS_SYN_RECEIVED; 1475 } 1476 1477 KASSERT(headlocked, ("%s: trimthenstep6: head not locked", 1478 __func__)); 1479 INP_LOCK_ASSERT(tp->t_inpcb); 1480 1481 /* 1482 * Advance th->th_seq to correspond to first data byte. 1483 * If data, trim to stay within window, 1484 * dropping FIN if necessary. 1485 */ 1486 th->th_seq++; 1487 if (tlen > tp->rcv_wnd) { 1488 todrop = tlen - tp->rcv_wnd; 1489 m_adj(m, -todrop); 1490 tlen = tp->rcv_wnd; 1491 thflags &= ~TH_FIN; 1492 tcpstat.tcps_rcvpackafterwin++; 1493 tcpstat.tcps_rcvbyteafterwin += todrop; 1494 } 1495 tp->snd_wl1 = th->th_seq - 1; 1496 tp->rcv_up = th->th_seq; 1497 /* 1498 * Client side of transaction: already sent SYN and data. 1499 * If the remote host used T/TCP to validate the SYN, 1500 * our data will be ACK'd; if so, enter normal data segment 1501 * processing in the middle of step 5, ack processing. 1502 * Otherwise, goto step 6. 1503 */ 1504 if (thflags & TH_ACK) 1505 goto process_ACK; 1506 1507 goto step6; 1508 1509 /* 1510 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1511 * do normal processing. 1512 * 1513 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1514 */ 1515 case TCPS_LAST_ACK: 1516 case TCPS_CLOSING: 1517 case TCPS_TIME_WAIT: 1518 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: timewait", 1519 __func__)); 1520 break; /* continue normal processing */ 1521 } 1522 1523 /* 1524 * States other than LISTEN or SYN_SENT. 1525 * First check the RST flag and sequence number since reset segments 1526 * are exempt from the timestamp and connection count tests. This 1527 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1528 * below which allowed reset segments in half the sequence space 1529 * to fall though and be processed (which gives forged reset 1530 * segments with a random sequence number a 50 percent chance of 1531 * killing a connection). 1532 * Then check timestamp, if present. 1533 * Then check the connection count, if present. 1534 * Then check that at least some bytes of segment are within 1535 * receive window. If segment begins before rcv_nxt, 1536 * drop leading data (and SYN); if nothing left, just ack. 1537 * 1538 * 1539 * If the RST bit is set, check the sequence number to see 1540 * if this is a valid reset segment. 1541 * RFC 793 page 37: 1542 * In all states except SYN-SENT, all reset (RST) segments 1543 * are validated by checking their SEQ-fields. A reset is 1544 * valid if its sequence number is in the window. 1545 * Note: this does not take into account delayed ACKs, so 1546 * we should test against last_ack_sent instead of rcv_nxt. 1547 * The sequence number in the reset segment is normally an 1548 * echo of our outgoing acknowlegement numbers, but some hosts 1549 * send a reset with the sequence number at the rightmost edge 1550 * of our receive window, and we have to handle this case. 1551 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1552 * that brute force RST attacks are possible. To combat this, 1553 * we use a much stricter check while in the ESTABLISHED state, 1554 * only accepting RSTs where the sequence number is equal to 1555 * last_ack_sent. In all other states (the states in which a 1556 * RST is more likely), the more permissive check is used. 1557 * If we have multiple segments in flight, the intial reset 1558 * segment sequence numbers will be to the left of last_ack_sent, 1559 * but they will eventually catch up. 1560 * In any case, it never made sense to trim reset segments to 1561 * fit the receive window since RFC 1122 says: 1562 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1563 * 1564 * A TCP SHOULD allow a received RST segment to include data. 1565 * 1566 * DISCUSSION 1567 * It has been suggested that a RST segment could contain 1568 * ASCII text that encoded and explained the cause of the 1569 * RST. No standard has yet been established for such 1570 * data. 1571 * 1572 * If the reset segment passes the sequence number test examine 1573 * the state: 1574 * SYN_RECEIVED STATE: 1575 * If passive open, return to LISTEN state. 1576 * If active open, inform user that connection was refused. 1577 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1578 * Inform user that connection was reset, and close tcb. 1579 * CLOSING, LAST_ACK STATES: 1580 * Close the tcb. 1581 * TIME_WAIT STATE: 1582 * Drop the segment - see Stevens, vol. 2, p. 964 and 1583 * RFC 1337. 1584 */ 1585 if (thflags & TH_RST) { 1586 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1587 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1588 switch (tp->t_state) { 1589 1590 case TCPS_SYN_RECEIVED: 1591 so->so_error = ECONNREFUSED; 1592 goto close; 1593 1594 case TCPS_ESTABLISHED: 1595 if (tcp_insecure_rst == 0 && 1596 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 1597 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 1598 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1599 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 1600 tcpstat.tcps_badrst++; 1601 goto drop; 1602 } 1603 case TCPS_FIN_WAIT_1: 1604 case TCPS_FIN_WAIT_2: 1605 case TCPS_CLOSE_WAIT: 1606 so->so_error = ECONNRESET; 1607 close: 1608 tp->t_state = TCPS_CLOSED; 1609 tcpstat.tcps_drops++; 1610 KASSERT(headlocked, ("%s: trimthenstep6: " 1611 "tcp_close: head not locked", __func__)); 1612 tp = tcp_close(tp); 1613 break; 1614 1615 case TCPS_CLOSING: 1616 case TCPS_LAST_ACK: 1617 KASSERT(headlocked, ("%s: trimthenstep6: " 1618 "tcp_close.2: head not locked", __func__)); 1619 tp = tcp_close(tp); 1620 break; 1621 1622 case TCPS_TIME_WAIT: 1623 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1624 ("%s: timewait", __func__)); 1625 break; 1626 } 1627 } 1628 goto drop; 1629 } 1630 1631 /* 1632 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1633 * and it's less than ts_recent, drop it. 1634 */ 1635 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1636 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1637 1638 /* Check to see if ts_recent is over 24 days old. */ 1639 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1640 /* 1641 * Invalidate ts_recent. If this segment updates 1642 * ts_recent, the age will be reset later and ts_recent 1643 * will get a valid value. If it does not, setting 1644 * ts_recent to zero will at least satisfy the 1645 * requirement that zero be placed in the timestamp 1646 * echo reply when ts_recent isn't valid. The 1647 * age isn't reset until we get a valid ts_recent 1648 * because we don't want out-of-order segments to be 1649 * dropped when ts_recent is old. 1650 */ 1651 tp->ts_recent = 0; 1652 } else { 1653 tcpstat.tcps_rcvduppack++; 1654 tcpstat.tcps_rcvdupbyte += tlen; 1655 tcpstat.tcps_pawsdrop++; 1656 if (tlen) 1657 goto dropafterack; 1658 goto drop; 1659 } 1660 } 1661 1662 /* 1663 * In the SYN-RECEIVED state, validate that the packet belongs to 1664 * this connection before trimming the data to fit the receive 1665 * window. Check the sequence number versus IRS since we know 1666 * the sequence numbers haven't wrapped. This is a partial fix 1667 * for the "LAND" DoS attack. 1668 */ 1669 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1670 rstreason = BANDLIM_RST_OPENPORT; 1671 goto dropwithreset; 1672 } 1673 1674 todrop = tp->rcv_nxt - th->th_seq; 1675 if (todrop > 0) { 1676 if (thflags & TH_SYN) { 1677 thflags &= ~TH_SYN; 1678 th->th_seq++; 1679 if (th->th_urp > 1) 1680 th->th_urp--; 1681 else 1682 thflags &= ~TH_URG; 1683 todrop--; 1684 } 1685 /* 1686 * Following if statement from Stevens, vol. 2, p. 960. 1687 */ 1688 if (todrop > tlen 1689 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1690 /* 1691 * Any valid FIN must be to the left of the window. 1692 * At this point the FIN must be a duplicate or out 1693 * of sequence; drop it. 1694 */ 1695 thflags &= ~TH_FIN; 1696 1697 /* 1698 * Send an ACK to resynchronize and drop any data. 1699 * But keep on processing for RST or ACK. 1700 */ 1701 tp->t_flags |= TF_ACKNOW; 1702 todrop = tlen; 1703 tcpstat.tcps_rcvduppack++; 1704 tcpstat.tcps_rcvdupbyte += todrop; 1705 } else { 1706 tcpstat.tcps_rcvpartduppack++; 1707 tcpstat.tcps_rcvpartdupbyte += todrop; 1708 } 1709 drop_hdrlen += todrop; /* drop from the top afterwards */ 1710 th->th_seq += todrop; 1711 tlen -= todrop; 1712 if (th->th_urp > todrop) 1713 th->th_urp -= todrop; 1714 else { 1715 thflags &= ~TH_URG; 1716 th->th_urp = 0; 1717 } 1718 } 1719 1720 /* 1721 * If new data are received on a connection after the 1722 * user processes are gone, then RST the other end. 1723 */ 1724 if ((so->so_state & SS_NOFDREF) && 1725 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1726 KASSERT(headlocked, ("%s: trimthenstep6: tcp_close.3: head " 1727 "not locked", __func__)); 1728 tp = tcp_close(tp); 1729 tcpstat.tcps_rcvafterclose++; 1730 rstreason = BANDLIM_UNLIMITED; 1731 goto dropwithreset; 1732 } 1733 1734 /* 1735 * If segment ends after window, drop trailing data 1736 * (and PUSH and FIN); if nothing left, just ACK. 1737 */ 1738 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1739 if (todrop > 0) { 1740 tcpstat.tcps_rcvpackafterwin++; 1741 if (todrop >= tlen) { 1742 tcpstat.tcps_rcvbyteafterwin += tlen; 1743 /* 1744 * If a new connection request is received 1745 * while in TIME_WAIT, drop the old connection 1746 * and start over if the sequence numbers 1747 * are above the previous ones. 1748 */ 1749 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: timewait", 1750 __func__)); 1751 if (thflags & TH_SYN && 1752 tp->t_state == TCPS_TIME_WAIT && 1753 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1754 KASSERT(headlocked, ("%s: trimthenstep6: " 1755 "tcp_close.4: head not locked", __func__)); 1756 tp = tcp_close(tp); 1757 /* XXX: Shouldn't be possible. */ 1758 return (1); 1759 } 1760 /* 1761 * If window is closed can only take segments at 1762 * window edge, and have to drop data and PUSH from 1763 * incoming segments. Continue processing, but 1764 * remember to ack. Otherwise, drop segment 1765 * and ack. 1766 */ 1767 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1768 tp->t_flags |= TF_ACKNOW; 1769 tcpstat.tcps_rcvwinprobe++; 1770 } else 1771 goto dropafterack; 1772 } else 1773 tcpstat.tcps_rcvbyteafterwin += todrop; 1774 m_adj(m, -todrop); 1775 tlen -= todrop; 1776 thflags &= ~(TH_PUSH|TH_FIN); 1777 } 1778 1779 /* 1780 * If last ACK falls within this segment's sequence numbers, 1781 * record its timestamp. 1782 * NOTE: 1783 * 1) That the test incorporates suggestions from the latest 1784 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1785 * 2) That updating only on newer timestamps interferes with 1786 * our earlier PAWS tests, so this check should be solely 1787 * predicated on the sequence space of this segment. 1788 * 3) That we modify the segment boundary check to be 1789 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1790 * instead of RFC1323's 1791 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1792 * This modified check allows us to overcome RFC1323's 1793 * limitations as described in Stevens TCP/IP Illustrated 1794 * Vol. 2 p.869. In such cases, we can still calculate the 1795 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1796 */ 1797 if ((to.to_flags & TOF_TS) != 0 && 1798 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1799 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1800 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1801 tp->ts_recent_age = ticks; 1802 tp->ts_recent = to.to_tsval; 1803 } 1804 1805 /* 1806 * If a SYN is in the window, then this is an 1807 * error and we send an RST and drop the connection. 1808 */ 1809 if (thflags & TH_SYN) { 1810 KASSERT(headlocked, ("%s: tcp_drop: trimthenstep6: " 1811 "head not locked", __func__)); 1812 tp = tcp_drop(tp, ECONNRESET); 1813 rstreason = BANDLIM_UNLIMITED; 1814 goto drop; 1815 } 1816 1817 /* 1818 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1819 * flag is on (half-synchronized state), then queue data for 1820 * later processing; else drop segment and return. 1821 */ 1822 if ((thflags & TH_ACK) == 0) { 1823 if (tp->t_state == TCPS_SYN_RECEIVED || 1824 (tp->t_flags & TF_NEEDSYN)) 1825 goto step6; 1826 else if (tp->t_flags & TF_ACKNOW) 1827 goto dropafterack; 1828 else 1829 goto drop; 1830 } 1831 1832 /* 1833 * Ack processing. 1834 */ 1835 switch (tp->t_state) { 1836 1837 /* 1838 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1839 * ESTABLISHED state and continue processing. 1840 * The ACK was checked above. 1841 */ 1842 case TCPS_SYN_RECEIVED: 1843 1844 tcpstat.tcps_connects++; 1845 soisconnected(so); 1846 /* Do window scaling? */ 1847 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1848 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1849 tp->rcv_scale = tp->request_r_scale; 1850 tp->snd_wnd = tiwin; 1851 } 1852 /* 1853 * Make transitions: 1854 * SYN-RECEIVED -> ESTABLISHED 1855 * SYN-RECEIVED* -> FIN-WAIT-1 1856 */ 1857 tp->t_starttime = ticks; 1858 if (tp->t_flags & TF_NEEDFIN) { 1859 tp->t_state = TCPS_FIN_WAIT_1; 1860 tp->t_flags &= ~TF_NEEDFIN; 1861 } else { 1862 tp->t_state = TCPS_ESTABLISHED; 1863 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1864 } 1865 /* 1866 * If segment contains data or ACK, will call tcp_reass() 1867 * later; if not, do so now to pass queued data to user. 1868 */ 1869 if (tlen == 0 && (thflags & TH_FIN) == 0) 1870 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1871 (struct mbuf *)0); 1872 tp->snd_wl1 = th->th_seq - 1; 1873 /* FALLTHROUGH */ 1874 1875 /* 1876 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1877 * ACKs. If the ack is in the range 1878 * tp->snd_una < th->th_ack <= tp->snd_max 1879 * then advance tp->snd_una to th->th_ack and drop 1880 * data from the retransmission queue. If this ACK reflects 1881 * more up to date window information we update our window information. 1882 */ 1883 case TCPS_ESTABLISHED: 1884 case TCPS_FIN_WAIT_1: 1885 case TCPS_FIN_WAIT_2: 1886 case TCPS_CLOSE_WAIT: 1887 case TCPS_CLOSING: 1888 case TCPS_LAST_ACK: 1889 case TCPS_TIME_WAIT: 1890 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: timewait", 1891 __func__)); 1892 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1893 tcpstat.tcps_rcvacktoomuch++; 1894 goto dropafterack; 1895 } 1896 if (tp->sack_enable && 1897 ((to.to_flags & TOF_SACK) || 1898 !TAILQ_EMPTY(&tp->snd_holes))) 1899 tcp_sack_doack(tp, &to, th->th_ack); 1900 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1901 if (tlen == 0 && tiwin == tp->snd_wnd) { 1902 tcpstat.tcps_rcvdupack++; 1903 /* 1904 * If we have outstanding data (other than 1905 * a window probe), this is a completely 1906 * duplicate ack (ie, window info didn't 1907 * change), the ack is the biggest we've 1908 * seen and we've seen exactly our rexmt 1909 * threshhold of them, assume a packet 1910 * has been dropped and retransmit it. 1911 * Kludge snd_nxt & the congestion 1912 * window so we send only this one 1913 * packet. 1914 * 1915 * We know we're losing at the current 1916 * window size so do congestion avoidance 1917 * (set ssthresh to half the current window 1918 * and pull our congestion window back to 1919 * the new ssthresh). 1920 * 1921 * Dup acks mean that packets have left the 1922 * network (they're now cached at the receiver) 1923 * so bump cwnd by the amount in the receiver 1924 * to keep a constant cwnd packets in the 1925 * network. 1926 */ 1927 if (!tcp_timer_active(tp, TT_REXMT) || 1928 th->th_ack != tp->snd_una) 1929 tp->t_dupacks = 0; 1930 else if (++tp->t_dupacks > tcprexmtthresh || 1931 ((tcp_do_newreno || tp->sack_enable) && 1932 IN_FASTRECOVERY(tp))) { 1933 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 1934 int awnd; 1935 1936 /* 1937 * Compute the amount of data in flight first. 1938 * We can inject new data into the pipe iff 1939 * we have less than 1/2 the original window's 1940 * worth of data in flight. 1941 */ 1942 awnd = (tp->snd_nxt - tp->snd_fack) + 1943 tp->sackhint.sack_bytes_rexmit; 1944 if (awnd < tp->snd_ssthresh) { 1945 tp->snd_cwnd += tp->t_maxseg; 1946 if (tp->snd_cwnd > tp->snd_ssthresh) 1947 tp->snd_cwnd = tp->snd_ssthresh; 1948 } 1949 } else 1950 tp->snd_cwnd += tp->t_maxseg; 1951 (void) tcp_output(tp); 1952 goto drop; 1953 } else if (tp->t_dupacks == tcprexmtthresh) { 1954 tcp_seq onxt = tp->snd_nxt; 1955 u_int win; 1956 1957 /* 1958 * If we're doing sack, check to 1959 * see if we're already in sack 1960 * recovery. If we're not doing sack, 1961 * check to see if we're in newreno 1962 * recovery. 1963 */ 1964 if (tp->sack_enable) { 1965 if (IN_FASTRECOVERY(tp)) { 1966 tp->t_dupacks = 0; 1967 break; 1968 } 1969 } else if (tcp_do_newreno) { 1970 if (SEQ_LEQ(th->th_ack, 1971 tp->snd_recover)) { 1972 tp->t_dupacks = 0; 1973 break; 1974 } 1975 } 1976 win = min(tp->snd_wnd, tp->snd_cwnd) / 1977 2 / tp->t_maxseg; 1978 if (win < 2) 1979 win = 2; 1980 tp->snd_ssthresh = win * tp->t_maxseg; 1981 ENTER_FASTRECOVERY(tp); 1982 tp->snd_recover = tp->snd_max; 1983 tcp_timer_activate(tp, TT_REXMT, 0); 1984 tp->t_rtttime = 0; 1985 if (tp->sack_enable) { 1986 tcpstat.tcps_sack_recovery_episode++; 1987 tp->sack_newdata = tp->snd_nxt; 1988 tp->snd_cwnd = tp->t_maxseg; 1989 (void) tcp_output(tp); 1990 goto drop; 1991 } 1992 tp->snd_nxt = th->th_ack; 1993 tp->snd_cwnd = tp->t_maxseg; 1994 (void) tcp_output(tp); 1995 KASSERT(tp->snd_limited <= 2, 1996 ("%s: tp->snd_limited too big", 1997 __func__)); 1998 tp->snd_cwnd = tp->snd_ssthresh + 1999 tp->t_maxseg * 2000 (tp->t_dupacks - tp->snd_limited); 2001 if (SEQ_GT(onxt, tp->snd_nxt)) 2002 tp->snd_nxt = onxt; 2003 goto drop; 2004 } else if (tcp_do_rfc3042) { 2005 u_long oldcwnd = tp->snd_cwnd; 2006 tcp_seq oldsndmax = tp->snd_max; 2007 u_int sent; 2008 2009 KASSERT(tp->t_dupacks == 1 || 2010 tp->t_dupacks == 2, 2011 ("%s: dupacks not 1 or 2", 2012 __func__)); 2013 if (tp->t_dupacks == 1) 2014 tp->snd_limited = 0; 2015 tp->snd_cwnd = 2016 (tp->snd_nxt - tp->snd_una) + 2017 (tp->t_dupacks - tp->snd_limited) * 2018 tp->t_maxseg; 2019 (void) tcp_output(tp); 2020 sent = tp->snd_max - oldsndmax; 2021 if (sent > tp->t_maxseg) { 2022 KASSERT((tp->t_dupacks == 2 && 2023 tp->snd_limited == 0) || 2024 (sent == tp->t_maxseg + 1 && 2025 tp->t_flags & TF_SENTFIN), 2026 ("%s: sent too much", 2027 __func__)); 2028 tp->snd_limited = 2; 2029 } else if (sent > 0) 2030 ++tp->snd_limited; 2031 tp->snd_cwnd = oldcwnd; 2032 goto drop; 2033 } 2034 } else 2035 tp->t_dupacks = 0; 2036 break; 2037 } 2038 2039 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2040 ("%s: th_ack <= snd_una", __func__)); 2041 2042 /* 2043 * If the congestion window was inflated to account 2044 * for the other side's cached packets, retract it. 2045 */ 2046 if (tcp_do_newreno || tp->sack_enable) { 2047 if (IN_FASTRECOVERY(tp)) { 2048 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2049 if (tp->sack_enable) 2050 tcp_sack_partialack(tp, th); 2051 else 2052 tcp_newreno_partial_ack(tp, th); 2053 } else { 2054 /* 2055 * Out of fast recovery. 2056 * Window inflation should have left us 2057 * with approximately snd_ssthresh 2058 * outstanding data. 2059 * But in case we would be inclined to 2060 * send a burst, better to do it via 2061 * the slow start mechanism. 2062 */ 2063 if (SEQ_GT(th->th_ack + 2064 tp->snd_ssthresh, 2065 tp->snd_max)) 2066 tp->snd_cwnd = tp->snd_max - 2067 th->th_ack + 2068 tp->t_maxseg; 2069 else 2070 tp->snd_cwnd = tp->snd_ssthresh; 2071 } 2072 } 2073 } else { 2074 if (tp->t_dupacks >= tcprexmtthresh && 2075 tp->snd_cwnd > tp->snd_ssthresh) 2076 tp->snd_cwnd = tp->snd_ssthresh; 2077 } 2078 tp->t_dupacks = 0; 2079 /* 2080 * If we reach this point, ACK is not a duplicate, 2081 * i.e., it ACKs something we sent. 2082 */ 2083 if (tp->t_flags & TF_NEEDSYN) { 2084 /* 2085 * T/TCP: Connection was half-synchronized, and our 2086 * SYN has been ACK'd (so connection is now fully 2087 * synchronized). Go to non-starred state, 2088 * increment snd_una for ACK of SYN, and check if 2089 * we can do window scaling. 2090 */ 2091 tp->t_flags &= ~TF_NEEDSYN; 2092 tp->snd_una++; 2093 /* Do window scaling? */ 2094 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2095 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2096 tp->rcv_scale = tp->request_r_scale; 2097 /* Send window already scaled. */ 2098 } 2099 } 2100 2101 process_ACK: 2102 KASSERT(headlocked, ("%s: process_ACK: head not locked", 2103 __func__)); 2104 INP_LOCK_ASSERT(tp->t_inpcb); 2105 2106 acked = th->th_ack - tp->snd_una; 2107 tcpstat.tcps_rcvackpack++; 2108 tcpstat.tcps_rcvackbyte += acked; 2109 2110 /* 2111 * If we just performed our first retransmit, and the ACK 2112 * arrives within our recovery window, then it was a mistake 2113 * to do the retransmit in the first place. Recover our 2114 * original cwnd and ssthresh, and proceed to transmit where 2115 * we left off. 2116 */ 2117 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2118 ++tcpstat.tcps_sndrexmitbad; 2119 tp->snd_cwnd = tp->snd_cwnd_prev; 2120 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2121 tp->snd_recover = tp->snd_recover_prev; 2122 if (tp->t_flags & TF_WASFRECOVERY) 2123 ENTER_FASTRECOVERY(tp); 2124 tp->snd_nxt = tp->snd_max; 2125 tp->t_badrxtwin = 0; /* XXX probably not required */ 2126 } 2127 2128 /* 2129 * If we have a timestamp reply, update smoothed 2130 * round trip time. If no timestamp is present but 2131 * transmit timer is running and timed sequence 2132 * number was acked, update smoothed round trip time. 2133 * Since we now have an rtt measurement, cancel the 2134 * timer backoff (cf., Phil Karn's retransmit alg.). 2135 * Recompute the initial retransmit timer. 2136 * 2137 * Some boxes send broken timestamp replies 2138 * during the SYN+ACK phase, ignore 2139 * timestamps of 0 or we could calculate a 2140 * huge RTT and blow up the retransmit timer. 2141 */ 2142 if ((to.to_flags & TOF_TS) != 0 && 2143 to.to_tsecr) { 2144 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2145 tp->t_rttlow = ticks - to.to_tsecr; 2146 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2147 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2148 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2149 tp->t_rttlow = ticks - tp->t_rtttime; 2150 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2151 } 2152 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2153 2154 /* 2155 * If all outstanding data is acked, stop retransmit 2156 * timer and remember to restart (more output or persist). 2157 * If there is more data to be acked, restart retransmit 2158 * timer, using current (possibly backed-off) value. 2159 */ 2160 if (th->th_ack == tp->snd_max) { 2161 tcp_timer_activate(tp, TT_REXMT, 0); 2162 needoutput = 1; 2163 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2164 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2165 2166 /* 2167 * If no data (only SYN) was ACK'd, 2168 * skip rest of ACK processing. 2169 */ 2170 if (acked == 0) 2171 goto step6; 2172 2173 /* 2174 * When new data is acked, open the congestion window. 2175 * If the window gives us less than ssthresh packets 2176 * in flight, open exponentially (maxseg per packet). 2177 * Otherwise open linearly: maxseg per window 2178 * (maxseg^2 / cwnd per packet). 2179 */ 2180 if ((!tcp_do_newreno && !tp->sack_enable) || 2181 !IN_FASTRECOVERY(tp)) { 2182 u_int cw = tp->snd_cwnd; 2183 u_int incr = tp->t_maxseg; 2184 if (cw > tp->snd_ssthresh) 2185 incr = incr * incr / cw; 2186 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2187 } 2188 SOCKBUF_LOCK(&so->so_snd); 2189 if (acked > so->so_snd.sb_cc) { 2190 tp->snd_wnd -= so->so_snd.sb_cc; 2191 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2192 ourfinisacked = 1; 2193 } else { 2194 sbdrop_locked(&so->so_snd, acked); 2195 tp->snd_wnd -= acked; 2196 ourfinisacked = 0; 2197 } 2198 sowwakeup_locked(so); 2199 /* detect una wraparound */ 2200 if ((tcp_do_newreno || tp->sack_enable) && 2201 !IN_FASTRECOVERY(tp) && 2202 SEQ_GT(tp->snd_una, tp->snd_recover) && 2203 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2204 tp->snd_recover = th->th_ack - 1; 2205 if ((tcp_do_newreno || tp->sack_enable) && 2206 IN_FASTRECOVERY(tp) && 2207 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2208 EXIT_FASTRECOVERY(tp); 2209 tp->snd_una = th->th_ack; 2210 if (tp->sack_enable) { 2211 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2212 tp->snd_recover = tp->snd_una; 2213 } 2214 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2215 tp->snd_nxt = tp->snd_una; 2216 2217 switch (tp->t_state) { 2218 2219 /* 2220 * In FIN_WAIT_1 STATE in addition to the processing 2221 * for the ESTABLISHED state if our FIN is now acknowledged 2222 * then enter FIN_WAIT_2. 2223 */ 2224 case TCPS_FIN_WAIT_1: 2225 if (ourfinisacked) { 2226 /* 2227 * If we can't receive any more 2228 * data, then closing user can proceed. 2229 * Starting the timer is contrary to the 2230 * specification, but if we don't get a FIN 2231 * we'll hang forever. 2232 */ 2233 /* XXXjl 2234 * we should release the tp also, and use a 2235 * compressed state. 2236 */ 2237 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2238 int timeout; 2239 2240 soisdisconnected(so); 2241 timeout = (tcp_fast_finwait2_recycle) ? 2242 tcp_finwait2_timeout : tcp_maxidle; 2243 tcp_timer_activate(tp, TT_2MSL, timeout); 2244 } 2245 tp->t_state = TCPS_FIN_WAIT_2; 2246 } 2247 break; 2248 2249 /* 2250 * In CLOSING STATE in addition to the processing for 2251 * the ESTABLISHED state if the ACK acknowledges our FIN 2252 * then enter the TIME-WAIT state, otherwise ignore 2253 * the segment. 2254 */ 2255 case TCPS_CLOSING: 2256 if (ourfinisacked) { 2257 KASSERT(headlocked, ("%s: process_ACK: " 2258 "head not locked", __func__)); 2259 tcp_twstart(tp); 2260 INP_INFO_WUNLOCK(&tcbinfo); 2261 headlocked = 0; 2262 m_freem(m); 2263 return (0); 2264 } 2265 break; 2266 2267 /* 2268 * In LAST_ACK, we may still be waiting for data to drain 2269 * and/or to be acked, as well as for the ack of our FIN. 2270 * If our FIN is now acknowledged, delete the TCB, 2271 * enter the closed state and return. 2272 */ 2273 case TCPS_LAST_ACK: 2274 if (ourfinisacked) { 2275 KASSERT(headlocked, ("%s: process_ACK: " 2276 "tcp_close: head not locked", __func__)); 2277 tp = tcp_close(tp); 2278 goto drop; 2279 } 2280 break; 2281 2282 /* 2283 * In TIME_WAIT state the only thing that should arrive 2284 * is a retransmission of the remote FIN. Acknowledge 2285 * it and restart the finack timer. 2286 */ 2287 case TCPS_TIME_WAIT: 2288 KASSERT(tp->t_state != TCPS_TIME_WAIT, 2289 ("%s: timewait", __func__)); 2290 tcp_timer_activate(tp, TT_2MSL, 2 * tcp_msl); 2291 goto dropafterack; 2292 } 2293 } 2294 2295 step6: 2296 KASSERT(headlocked, ("%s: step6: head not locked", __func__)); 2297 INP_LOCK_ASSERT(tp->t_inpcb); 2298 2299 /* 2300 * Update window information. 2301 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2302 */ 2303 if ((thflags & TH_ACK) && 2304 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2305 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2306 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2307 /* keep track of pure window updates */ 2308 if (tlen == 0 && 2309 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2310 tcpstat.tcps_rcvwinupd++; 2311 tp->snd_wnd = tiwin; 2312 tp->snd_wl1 = th->th_seq; 2313 tp->snd_wl2 = th->th_ack; 2314 if (tp->snd_wnd > tp->max_sndwnd) 2315 tp->max_sndwnd = tp->snd_wnd; 2316 needoutput = 1; 2317 } 2318 2319 /* 2320 * Process segments with URG. 2321 */ 2322 if ((thflags & TH_URG) && th->th_urp && 2323 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2324 /* 2325 * This is a kludge, but if we receive and accept 2326 * random urgent pointers, we'll crash in 2327 * soreceive. It's hard to imagine someone 2328 * actually wanting to send this much urgent data. 2329 */ 2330 SOCKBUF_LOCK(&so->so_rcv); 2331 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2332 th->th_urp = 0; /* XXX */ 2333 thflags &= ~TH_URG; /* XXX */ 2334 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2335 goto dodata; /* XXX */ 2336 } 2337 /* 2338 * If this segment advances the known urgent pointer, 2339 * then mark the data stream. This should not happen 2340 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2341 * a FIN has been received from the remote side. 2342 * In these states we ignore the URG. 2343 * 2344 * According to RFC961 (Assigned Protocols), 2345 * the urgent pointer points to the last octet 2346 * of urgent data. We continue, however, 2347 * to consider it to indicate the first octet 2348 * of data past the urgent section as the original 2349 * spec states (in one of two places). 2350 */ 2351 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2352 tp->rcv_up = th->th_seq + th->th_urp; 2353 so->so_oobmark = so->so_rcv.sb_cc + 2354 (tp->rcv_up - tp->rcv_nxt) - 1; 2355 if (so->so_oobmark == 0) 2356 so->so_rcv.sb_state |= SBS_RCVATMARK; 2357 sohasoutofband(so); 2358 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2359 } 2360 SOCKBUF_UNLOCK(&so->so_rcv); 2361 /* 2362 * Remove out of band data so doesn't get presented to user. 2363 * This can happen independent of advancing the URG pointer, 2364 * but if two URG's are pending at once, some out-of-band 2365 * data may creep in... ick. 2366 */ 2367 if (th->th_urp <= (u_long)tlen && 2368 !(so->so_options & SO_OOBINLINE)) { 2369 /* hdr drop is delayed */ 2370 tcp_pulloutofband(so, th, m, drop_hdrlen); 2371 } 2372 } else { 2373 /* 2374 * If no out of band data is expected, 2375 * pull receive urgent pointer along 2376 * with the receive window. 2377 */ 2378 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2379 tp->rcv_up = tp->rcv_nxt; 2380 } 2381 dodata: /* XXX */ 2382 KASSERT(headlocked, ("%s: dodata: head not locked", __func__)); 2383 INP_LOCK_ASSERT(tp->t_inpcb); 2384 2385 /* 2386 * Process the segment text, merging it into the TCP sequencing queue, 2387 * and arranging for acknowledgment of receipt if necessary. 2388 * This process logically involves adjusting tp->rcv_wnd as data 2389 * is presented to the user (this happens in tcp_usrreq.c, 2390 * case PRU_RCVD). If a FIN has already been received on this 2391 * connection then we just ignore the text. 2392 */ 2393 if ((tlen || (thflags & TH_FIN)) && 2394 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2395 tcp_seq save_start = th->th_seq; 2396 tcp_seq save_end = th->th_seq + tlen; 2397 m_adj(m, drop_hdrlen); /* delayed header drop */ 2398 /* 2399 * Insert segment which includes th into TCP reassembly queue 2400 * with control block tp. Set thflags to whether reassembly now 2401 * includes a segment with FIN. This handles the common case 2402 * inline (segment is the next to be received on an established 2403 * connection, and the queue is empty), avoiding linkage into 2404 * and removal from the queue and repetition of various 2405 * conversions. 2406 * Set DELACK for segments received in order, but ack 2407 * immediately when segments are out of order (so 2408 * fast retransmit can work). 2409 */ 2410 if (th->th_seq == tp->rcv_nxt && 2411 LIST_EMPTY(&tp->t_segq) && 2412 TCPS_HAVEESTABLISHED(tp->t_state)) { 2413 if (DELAY_ACK(tp)) 2414 tp->t_flags |= TF_DELACK; 2415 else 2416 tp->t_flags |= TF_ACKNOW; 2417 tp->rcv_nxt += tlen; 2418 thflags = th->th_flags & TH_FIN; 2419 tcpstat.tcps_rcvpack++; 2420 tcpstat.tcps_rcvbyte += tlen; 2421 ND6_HINT(tp); 2422 SOCKBUF_LOCK(&so->so_rcv); 2423 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2424 m_freem(m); 2425 else 2426 sbappendstream_locked(&so->so_rcv, m); 2427 sorwakeup_locked(so); 2428 } else { 2429 thflags = tcp_reass(tp, th, &tlen, m); 2430 tp->t_flags |= TF_ACKNOW; 2431 } 2432 if (tlen > 0 && tp->sack_enable) 2433 tcp_update_sack_list(tp, save_start, save_end); 2434 #if 0 2435 /* 2436 * Note the amount of data that peer has sent into 2437 * our window, in order to estimate the sender's 2438 * buffer size. 2439 * XXX: Unused. 2440 */ 2441 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2442 #endif 2443 } else { 2444 m_freem(m); 2445 thflags &= ~TH_FIN; 2446 } 2447 2448 /* 2449 * If FIN is received ACK the FIN and let the user know 2450 * that the connection is closing. 2451 */ 2452 if (thflags & TH_FIN) { 2453 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2454 socantrcvmore(so); 2455 /* 2456 * If connection is half-synchronized 2457 * (ie NEEDSYN flag on) then delay ACK, 2458 * so it may be piggybacked when SYN is sent. 2459 * Otherwise, since we received a FIN then no 2460 * more input can be expected, send ACK now. 2461 */ 2462 if (tp->t_flags & TF_NEEDSYN) 2463 tp->t_flags |= TF_DELACK; 2464 else 2465 tp->t_flags |= TF_ACKNOW; 2466 tp->rcv_nxt++; 2467 } 2468 switch (tp->t_state) { 2469 2470 /* 2471 * In SYN_RECEIVED and ESTABLISHED STATES 2472 * enter the CLOSE_WAIT state. 2473 */ 2474 case TCPS_SYN_RECEIVED: 2475 tp->t_starttime = ticks; 2476 /*FALLTHROUGH*/ 2477 case TCPS_ESTABLISHED: 2478 tp->t_state = TCPS_CLOSE_WAIT; 2479 break; 2480 2481 /* 2482 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2483 * enter the CLOSING state. 2484 */ 2485 case TCPS_FIN_WAIT_1: 2486 tp->t_state = TCPS_CLOSING; 2487 break; 2488 2489 /* 2490 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2491 * starting the time-wait timer, turning off the other 2492 * standard timers. 2493 */ 2494 case TCPS_FIN_WAIT_2: 2495 KASSERT(headlocked == 1, ("%s: dodata: " 2496 "TCP_FIN_WAIT_2: head not locked", __func__)); 2497 tcp_twstart(tp); 2498 INP_INFO_WUNLOCK(&tcbinfo); 2499 return (0); 2500 2501 /* 2502 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2503 */ 2504 case TCPS_TIME_WAIT: 2505 KASSERT(tp->t_state != TCPS_TIME_WAIT, 2506 ("%s: timewait", __func__)); 2507 tcp_timer_activate(tp, TT_2MSL, 2 * tcp_msl); 2508 break; 2509 } 2510 } 2511 INP_INFO_WUNLOCK(&tcbinfo); 2512 headlocked = 0; 2513 #ifdef TCPDEBUG 2514 if (so->so_options & SO_DEBUG) 2515 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2516 &tcp_savetcp, 0); 2517 #endif 2518 2519 /* 2520 * Return any desired output. 2521 */ 2522 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2523 (void) tcp_output(tp); 2524 2525 check_delack: 2526 KASSERT(headlocked == 0, ("%s: check_delack: head locked", 2527 __func__)); 2528 INP_LOCK_ASSERT(tp->t_inpcb); 2529 if (tp->t_flags & TF_DELACK) { 2530 tp->t_flags &= ~TF_DELACK; 2531 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2532 } 2533 INP_UNLOCK(tp->t_inpcb); 2534 return (0); 2535 2536 dropafterack: 2537 KASSERT(headlocked, ("%s: dropafterack: head not locked", __func__)); 2538 /* 2539 * Generate an ACK dropping incoming segment if it occupies 2540 * sequence space, where the ACK reflects our state. 2541 * 2542 * We can now skip the test for the RST flag since all 2543 * paths to this code happen after packets containing 2544 * RST have been dropped. 2545 * 2546 * In the SYN-RECEIVED state, don't send an ACK unless the 2547 * segment we received passes the SYN-RECEIVED ACK test. 2548 * If it fails send a RST. This breaks the loop in the 2549 * "LAND" DoS attack, and also prevents an ACK storm 2550 * between two listening ports that have been sent forged 2551 * SYN segments, each with the source address of the other. 2552 */ 2553 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2554 (SEQ_GT(tp->snd_una, th->th_ack) || 2555 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2556 rstreason = BANDLIM_RST_OPENPORT; 2557 goto dropwithreset; 2558 } 2559 #ifdef TCPDEBUG 2560 if (so->so_options & SO_DEBUG) 2561 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2562 &tcp_savetcp, 0); 2563 #endif 2564 KASSERT(headlocked, ("%s: headlocked should be 1", __func__)); 2565 INP_INFO_WUNLOCK(&tcbinfo); 2566 tp->t_flags |= TF_ACKNOW; 2567 (void) tcp_output(tp); 2568 INP_UNLOCK(tp->t_inpcb); 2569 m_freem(m); 2570 return (0); 2571 2572 dropwithreset: 2573 KASSERT(headlocked, ("%s: dropwithreset: head not locked", __func__)); 2574 2575 tcp_dropwithreset(m, th, tp, tlen, rstreason); 2576 2577 if (tp != NULL) 2578 INP_UNLOCK(tp->t_inpcb); 2579 if (headlocked) 2580 INP_INFO_WUNLOCK(&tcbinfo); 2581 return (0); 2582 2583 drop: 2584 /* 2585 * Drop space held by incoming segment and return. 2586 */ 2587 #ifdef TCPDEBUG 2588 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2589 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2590 &tcp_savetcp, 0); 2591 #endif 2592 if (tp != NULL) 2593 INP_UNLOCK(tp->t_inpcb); 2594 if (headlocked) 2595 INP_INFO_WUNLOCK(&tcbinfo); 2596 m_freem(m); 2597 return (0); 2598 } 2599 2600 2601 /* 2602 * Issue RST on TCP segment. The mbuf must still include the original 2603 * packet header. 2604 */ 2605 static void 2606 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 2607 int tlen, int rstreason) 2608 { 2609 struct ip *ip; 2610 #ifdef INET6 2611 struct ip6_hdr *ip6; 2612 #endif 2613 /* 2614 * Generate a RST, dropping incoming segment. 2615 * Make ACK acceptable to originator of segment. 2616 * Don't bother to respond if destination was broadcast/multicast. 2617 * tp may be NULL. 2618 */ 2619 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2620 goto drop; 2621 #ifdef INET6 2622 if (mtod(m, struct ip *)->ip_v == 6) { 2623 ip6 = mtod(m, struct ip6_hdr *); 2624 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2625 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2626 goto drop; 2627 /* IPv6 anycast check is done at tcp6_input() */ 2628 } else 2629 #endif 2630 { 2631 ip = mtod(m, struct ip *); 2632 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2633 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2634 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2635 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2636 goto drop; 2637 } 2638 2639 /* Perform bandwidth limiting. */ 2640 if (badport_bandlim(rstreason) < 0) 2641 goto drop; 2642 2643 /* tcp_respond consumes the mbuf chain. */ 2644 if (th->th_flags & TH_ACK) { 2645 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 2646 th->th_ack, TH_RST); 2647 } else { 2648 if (th->th_flags & TH_SYN) 2649 tlen++; 2650 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2651 (tcp_seq)0, TH_RST|TH_ACK); 2652 } 2653 return; 2654 drop: 2655 m_freem(m); 2656 return; 2657 } 2658 2659 /* 2660 * Parse TCP options and place in tcpopt. 2661 */ 2662 static void 2663 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 2664 { 2665 int opt, optlen; 2666 2667 to->to_flags = 0; 2668 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2669 opt = cp[0]; 2670 if (opt == TCPOPT_EOL) 2671 break; 2672 if (opt == TCPOPT_NOP) 2673 optlen = 1; 2674 else { 2675 if (cnt < 2) 2676 break; 2677 optlen = cp[1]; 2678 if (optlen < 2 || optlen > cnt) 2679 break; 2680 } 2681 switch (opt) { 2682 case TCPOPT_MAXSEG: 2683 if (optlen != TCPOLEN_MAXSEG) 2684 continue; 2685 if (!(flags & TO_SYN)) 2686 continue; 2687 to->to_flags |= TOF_MSS; 2688 bcopy((char *)cp + 2, 2689 (char *)&to->to_mss, sizeof(to->to_mss)); 2690 to->to_mss = ntohs(to->to_mss); 2691 break; 2692 case TCPOPT_WINDOW: 2693 if (optlen != TCPOLEN_WINDOW) 2694 continue; 2695 if (!(flags & TO_SYN)) 2696 continue; 2697 to->to_flags |= TOF_SCALE; 2698 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 2699 break; 2700 case TCPOPT_TIMESTAMP: 2701 if (optlen != TCPOLEN_TIMESTAMP) 2702 continue; 2703 to->to_flags |= TOF_TS; 2704 bcopy((char *)cp + 2, 2705 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2706 to->to_tsval = ntohl(to->to_tsval); 2707 bcopy((char *)cp + 6, 2708 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2709 to->to_tsecr = ntohl(to->to_tsecr); 2710 break; 2711 #ifdef TCP_SIGNATURE 2712 /* 2713 * XXX In order to reply to a host which has set the 2714 * TCP_SIGNATURE option in its initial SYN, we have to 2715 * record the fact that the option was observed here 2716 * for the syncache code to perform the correct response. 2717 */ 2718 case TCPOPT_SIGNATURE: 2719 if (optlen != TCPOLEN_SIGNATURE) 2720 continue; 2721 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2722 break; 2723 #endif 2724 case TCPOPT_SACK_PERMITTED: 2725 if (optlen != TCPOLEN_SACK_PERMITTED) 2726 continue; 2727 if (!(flags & TO_SYN)) 2728 continue; 2729 if (!tcp_do_sack) 2730 continue; 2731 to->to_flags |= TOF_SACKPERM; 2732 break; 2733 case TCPOPT_SACK: 2734 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2735 continue; 2736 if (flags & TO_SYN) 2737 continue; 2738 to->to_flags |= TOF_SACK; 2739 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2740 to->to_sacks = cp + 2; 2741 tcpstat.tcps_sack_rcv_blocks++; 2742 break; 2743 default: 2744 continue; 2745 } 2746 } 2747 } 2748 2749 /* 2750 * Pull out of band byte out of a segment so 2751 * it doesn't appear in the user's data queue. 2752 * It is still reflected in the segment length for 2753 * sequencing purposes. 2754 */ 2755 static void 2756 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 2757 int off) 2758 { 2759 int cnt = off + th->th_urp - 1; 2760 2761 while (cnt >= 0) { 2762 if (m->m_len > cnt) { 2763 char *cp = mtod(m, caddr_t) + cnt; 2764 struct tcpcb *tp = sototcpcb(so); 2765 2766 tp->t_iobc = *cp; 2767 tp->t_oobflags |= TCPOOB_HAVEDATA; 2768 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2769 m->m_len--; 2770 if (m->m_flags & M_PKTHDR) 2771 m->m_pkthdr.len--; 2772 return; 2773 } 2774 cnt -= m->m_len; 2775 m = m->m_next; 2776 if (m == NULL) 2777 break; 2778 } 2779 panic("tcp_pulloutofband"); 2780 } 2781 2782 /* 2783 * Collect new round-trip time estimate 2784 * and update averages and current timeout. 2785 */ 2786 static void 2787 tcp_xmit_timer(struct tcpcb *tp, int rtt) 2788 { 2789 int delta; 2790 2791 INP_LOCK_ASSERT(tp->t_inpcb); 2792 2793 tcpstat.tcps_rttupdated++; 2794 tp->t_rttupdated++; 2795 if (tp->t_srtt != 0) { 2796 /* 2797 * srtt is stored as fixed point with 5 bits after the 2798 * binary point (i.e., scaled by 8). The following magic 2799 * is equivalent to the smoothing algorithm in rfc793 with 2800 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2801 * point). Adjust rtt to origin 0. 2802 */ 2803 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2804 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2805 2806 if ((tp->t_srtt += delta) <= 0) 2807 tp->t_srtt = 1; 2808 2809 /* 2810 * We accumulate a smoothed rtt variance (actually, a 2811 * smoothed mean difference), then set the retransmit 2812 * timer to smoothed rtt + 4 times the smoothed variance. 2813 * rttvar is stored as fixed point with 4 bits after the 2814 * binary point (scaled by 16). The following is 2815 * equivalent to rfc793 smoothing with an alpha of .75 2816 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2817 * rfc793's wired-in beta. 2818 */ 2819 if (delta < 0) 2820 delta = -delta; 2821 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2822 if ((tp->t_rttvar += delta) <= 0) 2823 tp->t_rttvar = 1; 2824 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2825 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2826 } else { 2827 /* 2828 * No rtt measurement yet - use the unsmoothed rtt. 2829 * Set the variance to half the rtt (so our first 2830 * retransmit happens at 3*rtt). 2831 */ 2832 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2833 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2834 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2835 } 2836 tp->t_rtttime = 0; 2837 tp->t_rxtshift = 0; 2838 2839 /* 2840 * the retransmit should happen at rtt + 4 * rttvar. 2841 * Because of the way we do the smoothing, srtt and rttvar 2842 * will each average +1/2 tick of bias. When we compute 2843 * the retransmit timer, we want 1/2 tick of rounding and 2844 * 1 extra tick because of +-1/2 tick uncertainty in the 2845 * firing of the timer. The bias will give us exactly the 2846 * 1.5 tick we need. But, because the bias is 2847 * statistical, we have to test that we don't drop below 2848 * the minimum feasible timer (which is 2 ticks). 2849 */ 2850 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2851 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2852 2853 /* 2854 * We received an ack for a packet that wasn't retransmitted; 2855 * it is probably safe to discard any error indications we've 2856 * received recently. This isn't quite right, but close enough 2857 * for now (a route might have failed after we sent a segment, 2858 * and the return path might not be symmetrical). 2859 */ 2860 tp->t_softerror = 0; 2861 } 2862 2863 /* 2864 * Determine a reasonable value for maxseg size. 2865 * If the route is known, check route for mtu. 2866 * If none, use an mss that can be handled on the outgoing 2867 * interface without forcing IP to fragment; if bigger than 2868 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2869 * to utilize large mbufs. If no route is found, route has no mtu, 2870 * or the destination isn't local, use a default, hopefully conservative 2871 * size (usually 512 or the default IP max size, but no more than the mtu 2872 * of the interface), as we can't discover anything about intervening 2873 * gateways or networks. We also initialize the congestion/slow start 2874 * window to be a single segment if the destination isn't local. 2875 * While looking at the routing entry, we also initialize other path-dependent 2876 * parameters from pre-set or cached values in the routing entry. 2877 * 2878 * Also take into account the space needed for options that we 2879 * send regularly. Make maxseg shorter by that amount to assure 2880 * that we can send maxseg amount of data even when the options 2881 * are present. Store the upper limit of the length of options plus 2882 * data in maxopd. 2883 * 2884 * 2885 * In case of T/TCP, we call this routine during implicit connection 2886 * setup as well (offer = -1), to initialize maxseg from the cached 2887 * MSS of our peer. 2888 * 2889 * NOTE that this routine is only called when we process an incoming 2890 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2891 */ 2892 void 2893 tcp_mss(struct tcpcb *tp, int offer) 2894 { 2895 int rtt, mss; 2896 u_long bufsize; 2897 u_long maxmtu; 2898 struct inpcb *inp = tp->t_inpcb; 2899 struct socket *so; 2900 struct hc_metrics_lite metrics; 2901 int origoffer = offer; 2902 int mtuflags = 0; 2903 #ifdef INET6 2904 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2905 size_t min_protoh = isipv6 ? 2906 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2907 sizeof (struct tcpiphdr); 2908 #else 2909 const size_t min_protoh = sizeof(struct tcpiphdr); 2910 #endif 2911 2912 /* initialize */ 2913 #ifdef INET6 2914 if (isipv6) { 2915 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags); 2916 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2917 } else 2918 #endif 2919 { 2920 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags); 2921 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2922 } 2923 so = inp->inp_socket; 2924 2925 /* 2926 * no route to sender, stay with default mss and return 2927 */ 2928 if (maxmtu == 0) 2929 return; 2930 2931 /* what have we got? */ 2932 switch (offer) { 2933 case 0: 2934 /* 2935 * Offer == 0 means that there was no MSS on the SYN 2936 * segment, in this case we use tcp_mssdflt. 2937 */ 2938 offer = 2939 #ifdef INET6 2940 isipv6 ? tcp_v6mssdflt : 2941 #endif 2942 tcp_mssdflt; 2943 break; 2944 2945 case -1: 2946 /* 2947 * Offer == -1 means that we didn't receive SYN yet. 2948 */ 2949 /* FALLTHROUGH */ 2950 2951 default: 2952 /* 2953 * Prevent DoS attack with too small MSS. Round up 2954 * to at least minmss. 2955 */ 2956 offer = max(offer, tcp_minmss); 2957 /* 2958 * Sanity check: make sure that maxopd will be large 2959 * enough to allow some data on segments even if the 2960 * all the option space is used (40bytes). Otherwise 2961 * funny things may happen in tcp_output. 2962 */ 2963 offer = max(offer, 64); 2964 } 2965 2966 /* 2967 * rmx information is now retrieved from tcp_hostcache 2968 */ 2969 tcp_hc_get(&inp->inp_inc, &metrics); 2970 2971 /* 2972 * if there's a discovered mtu int tcp hostcache, use it 2973 * else, use the link mtu. 2974 */ 2975 if (metrics.rmx_mtu) 2976 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2977 else { 2978 #ifdef INET6 2979 if (isipv6) { 2980 mss = maxmtu - min_protoh; 2981 if (!path_mtu_discovery && 2982 !in6_localaddr(&inp->in6p_faddr)) 2983 mss = min(mss, tcp_v6mssdflt); 2984 } else 2985 #endif 2986 { 2987 mss = maxmtu - min_protoh; 2988 if (!path_mtu_discovery && 2989 !in_localaddr(inp->inp_faddr)) 2990 mss = min(mss, tcp_mssdflt); 2991 } 2992 } 2993 mss = min(mss, offer); 2994 2995 /* 2996 * maxopd stores the maximum length of data AND options 2997 * in a segment; maxseg is the amount of data in a normal 2998 * segment. We need to store this value (maxopd) apart 2999 * from maxseg, because now every segment carries options 3000 * and thus we normally have somewhat less data in segments. 3001 */ 3002 tp->t_maxopd = mss; 3003 3004 /* 3005 * origoffer==-1 indicates, that no segments were received yet. 3006 * In this case we just guess. 3007 */ 3008 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3009 (origoffer == -1 || 3010 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3011 mss -= TCPOLEN_TSTAMP_APPA; 3012 tp->t_maxseg = mss; 3013 3014 #if (MCLBYTES & (MCLBYTES - 1)) == 0 3015 if (mss > MCLBYTES) 3016 mss &= ~(MCLBYTES-1); 3017 #else 3018 if (mss > MCLBYTES) 3019 mss = mss / MCLBYTES * MCLBYTES; 3020 #endif 3021 tp->t_maxseg = mss; 3022 3023 /* 3024 * If there's a pipesize, change the socket buffer to that size, 3025 * don't change if sb_hiwat is different than default (then it 3026 * has been changed on purpose with setsockopt). 3027 * Make the socket buffers an integral number of mss units; 3028 * if the mss is larger than the socket buffer, decrease the mss. 3029 */ 3030 SOCKBUF_LOCK(&so->so_snd); 3031 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 3032 bufsize = metrics.rmx_sendpipe; 3033 else 3034 bufsize = so->so_snd.sb_hiwat; 3035 if (bufsize < mss) 3036 mss = bufsize; 3037 else { 3038 bufsize = roundup(bufsize, mss); 3039 if (bufsize > sb_max) 3040 bufsize = sb_max; 3041 if (bufsize > so->so_snd.sb_hiwat) 3042 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3043 } 3044 SOCKBUF_UNLOCK(&so->so_snd); 3045 tp->t_maxseg = mss; 3046 3047 SOCKBUF_LOCK(&so->so_rcv); 3048 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3049 bufsize = metrics.rmx_recvpipe; 3050 else 3051 bufsize = so->so_rcv.sb_hiwat; 3052 if (bufsize > mss) { 3053 bufsize = roundup(bufsize, mss); 3054 if (bufsize > sb_max) 3055 bufsize = sb_max; 3056 if (bufsize > so->so_rcv.sb_hiwat) 3057 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3058 } 3059 SOCKBUF_UNLOCK(&so->so_rcv); 3060 /* 3061 * While we're here, check the others too 3062 */ 3063 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 3064 tp->t_srtt = rtt; 3065 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3066 tcpstat.tcps_usedrtt++; 3067 if (metrics.rmx_rttvar) { 3068 tp->t_rttvar = metrics.rmx_rttvar; 3069 tcpstat.tcps_usedrttvar++; 3070 } else { 3071 /* default variation is +- 1 rtt */ 3072 tp->t_rttvar = 3073 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3074 } 3075 TCPT_RANGESET(tp->t_rxtcur, 3076 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3077 tp->t_rttmin, TCPTV_REXMTMAX); 3078 } 3079 if (metrics.rmx_ssthresh) { 3080 /* 3081 * There's some sort of gateway or interface 3082 * buffer limit on the path. Use this to set 3083 * the slow start threshhold, but set the 3084 * threshold to no less than 2*mss. 3085 */ 3086 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3087 tcpstat.tcps_usedssthresh++; 3088 } 3089 if (metrics.rmx_bandwidth) 3090 tp->snd_bandwidth = metrics.rmx_bandwidth; 3091 3092 /* 3093 * Set the slow-start flight size depending on whether this 3094 * is a local network or not. 3095 * 3096 * Extend this so we cache the cwnd too and retrieve it here. 3097 * Make cwnd even bigger than RFC3390 suggests but only if we 3098 * have previous experience with the remote host. Be careful 3099 * not make cwnd bigger than remote receive window or our own 3100 * send socket buffer. Maybe put some additional upper bound 3101 * on the retrieved cwnd. Should do incremental updates to 3102 * hostcache when cwnd collapses so next connection doesn't 3103 * overloads the path again. 3104 * 3105 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3106 * We currently check only in syncache_socket for that. 3107 */ 3108 #define TCP_METRICS_CWND 3109 #ifdef TCP_METRICS_CWND 3110 if (metrics.rmx_cwnd) 3111 tp->snd_cwnd = max(mss, 3112 min(metrics.rmx_cwnd / 2, 3113 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3114 else 3115 #endif 3116 if (tcp_do_rfc3390) 3117 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3118 #ifdef INET6 3119 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3120 (!isipv6 && in_localaddr(inp->inp_faddr))) 3121 #else 3122 else if (in_localaddr(inp->inp_faddr)) 3123 #endif 3124 tp->snd_cwnd = mss * ss_fltsz_local; 3125 else 3126 tp->snd_cwnd = mss * ss_fltsz; 3127 3128 /* Check the interface for TSO capabilities. */ 3129 if (mtuflags & CSUM_TSO) 3130 tp->t_flags |= TF_TSO; 3131 } 3132 3133 /* 3134 * Determine the MSS option to send on an outgoing SYN. 3135 */ 3136 int 3137 tcp_mssopt(struct in_conninfo *inc) 3138 { 3139 int mss = 0; 3140 u_long maxmtu = 0; 3141 u_long thcmtu = 0; 3142 size_t min_protoh; 3143 #ifdef INET6 3144 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3145 #endif 3146 3147 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3148 3149 #ifdef INET6 3150 if (isipv6) { 3151 mss = tcp_v6mssdflt; 3152 maxmtu = tcp_maxmtu6(inc, NULL); 3153 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3154 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3155 } else 3156 #endif 3157 { 3158 mss = tcp_mssdflt; 3159 maxmtu = tcp_maxmtu(inc, NULL); 3160 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3161 min_protoh = sizeof(struct tcpiphdr); 3162 } 3163 if (maxmtu && thcmtu) 3164 mss = min(maxmtu, thcmtu) - min_protoh; 3165 else if (maxmtu || thcmtu) 3166 mss = max(maxmtu, thcmtu) - min_protoh; 3167 3168 return (mss); 3169 } 3170 3171 3172 /* 3173 * On a partial ack arrives, force the retransmission of the 3174 * next unacknowledged segment. Do not clear tp->t_dupacks. 3175 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3176 * be started again. 3177 */ 3178 static void 3179 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3180 { 3181 tcp_seq onxt = tp->snd_nxt; 3182 u_long ocwnd = tp->snd_cwnd; 3183 3184 tcp_timer_activate(tp, TT_REXMT, 0); 3185 tp->t_rtttime = 0; 3186 tp->snd_nxt = th->th_ack; 3187 /* 3188 * Set snd_cwnd to one segment beyond acknowledged offset. 3189 * (tp->snd_una has not yet been updated when this function is called.) 3190 */ 3191 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3192 tp->t_flags |= TF_ACKNOW; 3193 (void) tcp_output(tp); 3194 tp->snd_cwnd = ocwnd; 3195 if (SEQ_GT(onxt, tp->snd_nxt)) 3196 tp->snd_nxt = onxt; 3197 /* 3198 * Partial window deflation. Relies on fact that tp->snd_una 3199 * not updated yet. 3200 */ 3201 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3202 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3203 else 3204 tp->snd_cwnd = 0; 3205 tp->snd_cwnd += tp->t_maxseg; 3206 } 3207 3208 /* 3209 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3210 * looking for a pcb in the listen state. Returns 0 otherwise. 3211 */ 3212 static int 3213 tcp_timewait(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th, 3214 struct mbuf *m, int tlen) 3215 { 3216 struct tcptw *tw; 3217 int thflags; 3218 tcp_seq seq; 3219 #ifdef INET6 3220 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3221 #else 3222 const int isipv6 = 0; 3223 #endif 3224 3225 /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */ 3226 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3227 INP_LOCK_ASSERT(inp); 3228 3229 /* 3230 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 3231 * still present. This is undesirable, but temporarily necessary 3232 * until we work out how to handle inpcb's who's timewait state has 3233 * been removed. 3234 */ 3235 tw = intotw(inp); 3236 if (tw == NULL) 3237 goto drop; 3238 3239 thflags = th->th_flags; 3240 3241 /* 3242 * NOTE: for FIN_WAIT_2 (to be added later), 3243 * must validate sequence number before accepting RST 3244 */ 3245 3246 /* 3247 * If the segment contains RST: 3248 * Drop the segment - see Stevens, vol. 2, p. 964 and 3249 * RFC 1337. 3250 */ 3251 if (thflags & TH_RST) 3252 goto drop; 3253 3254 #if 0 3255 /* PAWS not needed at the moment */ 3256 /* 3257 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3258 * and it's less than ts_recent, drop it. 3259 */ 3260 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3261 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3262 if ((thflags & TH_ACK) == 0) 3263 goto drop; 3264 goto ack; 3265 } 3266 /* 3267 * ts_recent is never updated because we never accept new segments. 3268 */ 3269 #endif 3270 3271 /* 3272 * If a new connection request is received 3273 * while in TIME_WAIT, drop the old connection 3274 * and start over if the sequence numbers 3275 * are above the previous ones. 3276 */ 3277 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3278 tcp_twclose(tw, 0); 3279 return (1); 3280 } 3281 3282 /* 3283 * Drop the the segment if it does not contain an ACK. 3284 */ 3285 if ((thflags & TH_ACK) == 0) 3286 goto drop; 3287 3288 /* 3289 * Reset the 2MSL timer if this is a duplicate FIN. 3290 */ 3291 if (thflags & TH_FIN) { 3292 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3293 if (seq + 1 == tw->rcv_nxt) 3294 tcp_timer_2msl_reset(tw, 1); 3295 } 3296 3297 /* 3298 * Acknowledge the segment if it has data or is not a duplicate ACK. 3299 */ 3300 if (thflags != TH_ACK || tlen != 0 || 3301 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3302 tcp_twrespond(tw, TH_ACK); 3303 goto drop; 3304 3305 /* 3306 * Generate a RST, dropping incoming segment. 3307 * Make ACK acceptable to originator of segment. 3308 * Don't bother to respond if destination was broadcast/multicast. 3309 */ 3310 if (m->m_flags & (M_BCAST|M_MCAST)) 3311 goto drop; 3312 if (isipv6) { 3313 struct ip6_hdr *ip6; 3314 3315 /* IPv6 anycast check is done at tcp6_input() */ 3316 ip6 = mtod(m, struct ip6_hdr *); 3317 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3318 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3319 goto drop; 3320 } else { 3321 struct ip *ip; 3322 3323 ip = mtod(m, struct ip *); 3324 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3325 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3326 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3327 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3328 goto drop; 3329 } 3330 if (thflags & TH_ACK) { 3331 tcp_respond(NULL, 3332 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3333 } else { 3334 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3335 tcp_respond(NULL, 3336 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3337 } 3338 INP_UNLOCK(inp); 3339 return (0); 3340 3341 drop: 3342 INP_UNLOCK(inp); 3343 m_freem(m); 3344 return (0); 3345 } 3346