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 /* NB: tcp_timewait unlocks the INP and frees the mbuf. */ 755 if (tcp_timewait(inp, &to, th, m, tlen)) 756 goto findpcb; 757 INP_INFO_WUNLOCK(&tcbinfo); 758 return; 759 } 760 /* 761 * The TCPCB may no longer exist if the connection is winding 762 * down or it is in the CLOSED state. Either way we drop the 763 * segment and send an appropriate response. 764 */ 765 tp = intotcpcb(inp); 766 if (tp == NULL) { 767 rstreason = BANDLIM_RST_CLOSEDPORT; 768 goto dropwithreset; 769 } 770 if (tp->t_state == TCPS_CLOSED) 771 goto dropunlock; /* XXX: dropwithreset??? */ 772 773 #ifdef MAC 774 INP_LOCK_ASSERT(inp); 775 if (mac_check_inpcb_deliver(inp, m)) 776 goto dropunlock; 777 #endif 778 so = inp->inp_socket; 779 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 780 #ifdef TCPDEBUG 781 if (so->so_options & SO_DEBUG) { 782 ostate = tp->t_state; 783 if (isipv6) 784 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 785 else 786 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 787 tcp_savetcp = *th; 788 } 789 #endif 790 /* 791 * When the socket is accepting connections (the INPCB is in LISTEN 792 * state) we look into the SYN cache if this is a new connection 793 * attempt or the completion of a previous one. 794 */ 795 if (so->so_options & SO_ACCEPTCONN) { 796 struct in_conninfo inc; 797 798 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 799 "tp not listening", __func__)); 800 801 bzero(&inc, sizeof(inc)); 802 inc.inc_isipv6 = isipv6; 803 #ifdef INET6 804 if (isipv6) { 805 inc.inc6_faddr = ip6->ip6_src; 806 inc.inc6_laddr = ip6->ip6_dst; 807 } else 808 #endif 809 { 810 inc.inc_faddr = ip->ip_src; 811 inc.inc_laddr = ip->ip_dst; 812 } 813 inc.inc_fport = th->th_sport; 814 inc.inc_lport = th->th_dport; 815 816 /* 817 * If the state is LISTEN then ignore segment if it contains 818 * a RST. If the segment contains an ACK then it is bad and 819 * send a RST. If it does not contain a SYN then it is not 820 * interesting; drop it. 821 * 822 * If the state is SYN_RECEIVED (syncache) and seg contains 823 * an ACK, but not for our SYN/ACK, send a RST. If the seg 824 * contains a RST, check the sequence number to see if it 825 * is a valid reset segment. 826 */ 827 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 828 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 829 /* 830 * Parse the TCP options here because 831 * syncookies need access to the reflected 832 * timestamp. 833 */ 834 tcp_dooptions(&to, optp, optlen, 0); 835 /* 836 * NB: syncache_expand() doesn't unlock 837 * inp and tcpinfo locks. 838 */ 839 if (!syncache_expand(&inc, &to, th, &so, m)) { 840 /* 841 * No syncache entry or ACK was not 842 * for our SYN/ACK. Send a RST. 843 */ 844 rstreason = BANDLIM_RST_OPENPORT; 845 goto dropwithreset; 846 } 847 if (so == NULL) { 848 /* 849 * We completed the 3-way handshake 850 * but could not allocate a socket 851 * either due to memory shortage, 852 * listen queue length limits or 853 * global socket limits. 854 */ 855 rstreason = BANDLIM_UNLIMITED; 856 goto dropwithreset; 857 } 858 /* 859 * Socket is created in state SYN_RECEIVED. 860 * Continue processing segment. 861 */ 862 INP_UNLOCK(inp); /* listen socket */ 863 inp = sotoinpcb(so); 864 INP_LOCK(inp); /* new connection */ 865 tp = intotcpcb(inp); 866 /* 867 * Process the segment and the data it 868 * contains. tcp_do_segment() consumes 869 * the mbuf chain and unlocks the inpcb. 870 * XXX: The potential return value of 871 * TIME_WAIT nuked is supposed to be 872 * handled above. 873 */ 874 if (tcp_do_segment(m, th, so, tp, 875 drop_hdrlen, tlen)) 876 goto findpcb; /* TIME_WAIT nuked */ 877 return; 878 } 879 if (thflags & TH_RST) { 880 syncache_chkrst(&inc, th); 881 goto dropunlock; 882 } 883 if (thflags & TH_ACK) { 884 syncache_badack(&inc); 885 tcpstat.tcps_badsyn++; 886 rstreason = BANDLIM_RST_OPENPORT; 887 goto dropwithreset; 888 } 889 goto dropunlock; 890 } 891 892 /* 893 * Segment's flags are (SYN) or (SYN|FIN). 894 */ 895 #ifdef INET6 896 /* 897 * If deprecated address is forbidden, 898 * we do not accept SYN to deprecated interface 899 * address to prevent any new inbound connection from 900 * getting established. 901 * When we do not accept SYN, we send a TCP RST, 902 * with deprecated source address (instead of dropping 903 * it). We compromise it as it is much better for peer 904 * to send a RST, and RST will be the final packet 905 * for the exchange. 906 * 907 * If we do not forbid deprecated addresses, we accept 908 * the SYN packet. RFC2462 does not suggest dropping 909 * SYN in this case. 910 * If we decipher RFC2462 5.5.4, it says like this: 911 * 1. use of deprecated addr with existing 912 * communication is okay - "SHOULD continue to be 913 * used" 914 * 2. use of it with new communication: 915 * (2a) "SHOULD NOT be used if alternate address 916 * with sufficient scope is available" 917 * (2b) nothing mentioned otherwise. 918 * Here we fall into (2b) case as we have no choice in 919 * our source address selection - we must obey the peer. 920 * 921 * The wording in RFC2462 is confusing, and there are 922 * multiple description text for deprecated address 923 * handling - worse, they are not exactly the same. 924 * I believe 5.5.4 is the best one, so we follow 5.5.4. 925 */ 926 if (isipv6 && !ip6_use_deprecated) { 927 struct in6_ifaddr *ia6; 928 929 if ((ia6 = ip6_getdstifaddr(m)) && 930 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 931 rstreason = BANDLIM_RST_OPENPORT; 932 goto dropwithreset; 933 } 934 } 935 #endif 936 /* 937 * Basic sanity checks on incoming SYN requests: 938 * 939 * Don't bother responding if the destination was a 940 * broadcast according to RFC1122 4.2.3.10, p. 104. 941 * 942 * If it is from this socket, drop it, it must be forged. 943 * 944 * Note that it is quite possible to receive unicast 945 * link-layer packets with a broadcast IP address. Use 946 * in_broadcast() to find them. 947 */ 948 if (m->m_flags & (M_BCAST|M_MCAST)) 949 goto dropunlock; 950 if (isipv6) { 951 #ifdef INET6 952 if (th->th_dport == th->th_sport && 953 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) 954 goto dropunlock; 955 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 956 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 957 goto dropunlock; 958 #endif 959 } else { 960 if (th->th_dport == th->th_sport && 961 ip->ip_dst.s_addr == ip->ip_src.s_addr) 962 goto dropunlock; 963 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 964 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 965 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 966 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 967 goto dropunlock; 968 } 969 /* 970 * SYN appears to be valid. Create compressed TCP state 971 * for syncache. 972 */ 973 #ifdef TCPDEBUG 974 if (so->so_options & SO_DEBUG) 975 tcp_trace(TA_INPUT, ostate, tp, 976 (void *)tcp_saveipgen, &tcp_savetcp, 0); 977 #endif 978 tcp_dooptions(&to, optp, optlen, TO_SYN); 979 syncache_add(&inc, &to, th, inp, &so, m); 980 /* 981 * Entry added to syncache and mbuf consumed. 982 * Everything unlocked already by syncache_add(). 983 */ 984 return; 985 } 986 987 /* 988 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or 989 * later state. tcp_do_segment() always consumes the mbuf chain 990 * and unlocks the inpcb. 991 */ 992 if (tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen)) 993 goto findpcb; /* XXX: TIME_WAIT was nuked. */ 994 return; 995 996 dropwithreset: 997 INP_INFO_WLOCK_ASSERT(&tcbinfo); 998 tcp_dropwithreset(m, th, tp, tlen, rstreason); 999 m = NULL; /* mbuf chain got consumed. */ 1000 dropunlock: 1001 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1002 if (inp != NULL) 1003 INP_UNLOCK(inp); 1004 INP_INFO_WUNLOCK(&tcbinfo); 1005 drop: 1006 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 1007 if (m != NULL) 1008 m_freem(m); 1009 return; 1010 } 1011 1012 static int 1013 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1014 struct tcpcb *tp, int drop_hdrlen, int tlen) 1015 { 1016 int thflags, acked, ourfinisacked, needoutput = 0; 1017 int headlocked = 1; 1018 int rstreason, todrop, win; 1019 u_long tiwin; 1020 struct tcpopt to; 1021 1022 #ifdef TCPDEBUG 1023 /* 1024 * The size of tcp_saveipgen must be the size of the max ip header, 1025 * now IPv6. 1026 */ 1027 u_char tcp_saveipgen[IP6_HDR_LEN]; 1028 struct tcphdr tcp_savetcp; 1029 short ostate = 0; 1030 #endif 1031 thflags = th->th_flags; 1032 1033 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1034 INP_LOCK_ASSERT(tp->t_inpcb); 1035 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", __func__)); 1036 1037 /* 1038 * Segment received on connection. 1039 * Reset idle time and keep-alive timer. 1040 */ 1041 tp->t_rcvtime = ticks; 1042 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1043 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1044 1045 /* 1046 * Unscale the window into a 32-bit value. 1047 * This value is bogus for the TCPS_SYN_SENT state 1048 * and is overwritten later. 1049 */ 1050 tiwin = th->th_win << tp->snd_scale; 1051 1052 /* 1053 * Parse options on any incoming segment. 1054 */ 1055 tcp_dooptions(&to, (u_char *)(th + 1), 1056 (th->th_off << 2) - sizeof(struct tcphdr), 1057 (thflags & TH_SYN) ? TO_SYN : 0); 1058 1059 /* 1060 * If echoed timestamp is later than the current time, 1061 * fall back to non RFC1323 RTT calculation. Normalize 1062 * timestamp if syncookies were used when this connection 1063 * was established. 1064 */ 1065 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1066 to.to_tsecr -= tp->ts_offset; 1067 if (TSTMP_GT(to.to_tsecr, ticks)) 1068 to.to_tsecr = 0; 1069 } 1070 1071 /* 1072 * Process options only when we get SYN/ACK back. The SYN case 1073 * for incoming connections is handled in tcp_syncache. 1074 * XXX this is traditional behavior, may need to be cleaned up. 1075 */ 1076 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1077 if ((to.to_flags & TOF_SCALE) && 1078 (tp->t_flags & TF_REQ_SCALE)) { 1079 tp->t_flags |= TF_RCVD_SCALE; 1080 tp->snd_scale = to.to_wscale; 1081 tp->snd_wnd = th->th_win << tp->snd_scale; 1082 tiwin = tp->snd_wnd; 1083 } 1084 if (to.to_flags & TOF_TS) { 1085 tp->t_flags |= TF_RCVD_TSTMP; 1086 tp->ts_recent = to.to_tsval; 1087 tp->ts_recent_age = ticks; 1088 } 1089 /* Initial send window, already scaled. */ 1090 tp->snd_wnd = th->th_win; 1091 if (to.to_flags & TOF_MSS) 1092 tcp_mss(tp, to.to_mss); 1093 if (tp->sack_enable) { 1094 if (!(to.to_flags & TOF_SACKPERM)) 1095 tp->sack_enable = 0; 1096 else 1097 tp->t_flags |= TF_SACK_PERMIT; 1098 } 1099 1100 } 1101 1102 /* 1103 * Header prediction: check for the two common cases 1104 * of a uni-directional data xfer. If the packet has 1105 * no control flags, is in-sequence, the window didn't 1106 * change and we're not retransmitting, it's a 1107 * candidate. If the length is zero and the ack moved 1108 * forward, we're the sender side of the xfer. Just 1109 * free the data acked & wake any higher level process 1110 * that was blocked waiting for space. If the length 1111 * is non-zero and the ack didn't move, we're the 1112 * receiver side. If we're getting packets in-order 1113 * (the reassembly queue is empty), add the data to 1114 * the socket buffer and note that we need a delayed ack. 1115 * Make sure that the hidden state-flags are also off. 1116 * Since we check for TCPS_ESTABLISHED above, it can only 1117 * be TH_NEEDSYN. 1118 */ 1119 if (tp->t_state == TCPS_ESTABLISHED && 1120 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1121 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1122 ((to.to_flags & TOF_TS) == 0 || 1123 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1124 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1125 tp->snd_nxt == tp->snd_max) { 1126 1127 /* 1128 * If last ACK falls within this segment's sequence numbers, 1129 * record the timestamp. 1130 * NOTE that the test is modified according to the latest 1131 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1132 */ 1133 if ((to.to_flags & TOF_TS) != 0 && 1134 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1135 tp->ts_recent_age = ticks; 1136 tp->ts_recent = to.to_tsval; 1137 } 1138 1139 if (tlen == 0) { 1140 if (SEQ_GT(th->th_ack, tp->snd_una) && 1141 SEQ_LEQ(th->th_ack, tp->snd_max) && 1142 tp->snd_cwnd >= tp->snd_wnd && 1143 ((!tcp_do_newreno && !tp->sack_enable && 1144 tp->t_dupacks < tcprexmtthresh) || 1145 ((tcp_do_newreno || tp->sack_enable) && 1146 !IN_FASTRECOVERY(tp) && 1147 (to.to_flags & TOF_SACK) == 0 && 1148 TAILQ_EMPTY(&tp->snd_holes)))) { 1149 KASSERT(headlocked, 1150 ("%s: headlocked", __func__)); 1151 INP_INFO_WUNLOCK(&tcbinfo); 1152 headlocked = 0; 1153 /* 1154 * this is a pure ack for outstanding data. 1155 */ 1156 ++tcpstat.tcps_predack; 1157 /* 1158 * "bad retransmit" recovery 1159 */ 1160 if (tp->t_rxtshift == 1 && 1161 ticks < tp->t_badrxtwin) { 1162 ++tcpstat.tcps_sndrexmitbad; 1163 tp->snd_cwnd = tp->snd_cwnd_prev; 1164 tp->snd_ssthresh = 1165 tp->snd_ssthresh_prev; 1166 tp->snd_recover = tp->snd_recover_prev; 1167 if (tp->t_flags & TF_WASFRECOVERY) 1168 ENTER_FASTRECOVERY(tp); 1169 tp->snd_nxt = tp->snd_max; 1170 tp->t_badrxtwin = 0; 1171 } 1172 1173 /* 1174 * Recalculate the transmit timer / rtt. 1175 * 1176 * Some boxes send broken timestamp replies 1177 * during the SYN+ACK phase, ignore 1178 * timestamps of 0 or we could calculate a 1179 * huge RTT and blow up the retransmit timer. 1180 */ 1181 if ((to.to_flags & TOF_TS) != 0 && 1182 to.to_tsecr) { 1183 if (!tp->t_rttlow || 1184 tp->t_rttlow > ticks - to.to_tsecr) 1185 tp->t_rttlow = ticks - to.to_tsecr; 1186 tcp_xmit_timer(tp, 1187 ticks - to.to_tsecr + 1); 1188 } else if (tp->t_rtttime && 1189 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1190 if (!tp->t_rttlow || 1191 tp->t_rttlow > ticks - tp->t_rtttime) 1192 tp->t_rttlow = ticks - tp->t_rtttime; 1193 tcp_xmit_timer(tp, 1194 ticks - tp->t_rtttime); 1195 } 1196 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1197 acked = th->th_ack - tp->snd_una; 1198 tcpstat.tcps_rcvackpack++; 1199 tcpstat.tcps_rcvackbyte += acked; 1200 sbdrop(&so->so_snd, acked); 1201 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1202 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1203 tp->snd_recover = th->th_ack - 1; 1204 tp->snd_una = th->th_ack; 1205 /* 1206 * pull snd_wl2 up to prevent seq wrap relative 1207 * to th_ack. 1208 */ 1209 tp->snd_wl2 = th->th_ack; 1210 tp->t_dupacks = 0; 1211 m_freem(m); 1212 ND6_HINT(tp); /* some progress has been done */ 1213 1214 /* 1215 * If all outstanding data are acked, stop 1216 * retransmit timer, otherwise restart timer 1217 * using current (possibly backed-off) value. 1218 * If process is waiting for space, 1219 * wakeup/selwakeup/signal. If data 1220 * are ready to send, let tcp_output 1221 * decide between more output or persist. 1222 1223 #ifdef TCPDEBUG 1224 if (so->so_options & SO_DEBUG) 1225 tcp_trace(TA_INPUT, ostate, tp, 1226 (void *)tcp_saveipgen, 1227 &tcp_savetcp, 0); 1228 #endif 1229 */ 1230 if (tp->snd_una == tp->snd_max) 1231 tcp_timer_activate(tp, TT_REXMT, 0); 1232 else if (!tcp_timer_active(tp, TT_PERSIST)) 1233 tcp_timer_activate(tp, TT_REXMT, 1234 tp->t_rxtcur); 1235 1236 sowwakeup(so); 1237 if (so->so_snd.sb_cc) 1238 (void) tcp_output(tp); 1239 goto check_delack; 1240 } 1241 } else if (th->th_ack == tp->snd_una && 1242 LIST_EMPTY(&tp->t_segq) && 1243 tlen <= sbspace(&so->so_rcv)) { 1244 int newsize = 0; /* automatic sockbuf scaling */ 1245 1246 KASSERT(headlocked, ("%s: headlocked", __func__)); 1247 INP_INFO_WUNLOCK(&tcbinfo); 1248 headlocked = 0; 1249 /* 1250 * this is a pure, in-sequence data packet 1251 * with nothing on the reassembly queue and 1252 * we have enough buffer space to take it. 1253 */ 1254 /* Clean receiver SACK report if present */ 1255 if (tp->sack_enable && tp->rcv_numsacks) 1256 tcp_clean_sackreport(tp); 1257 ++tcpstat.tcps_preddat; 1258 tp->rcv_nxt += tlen; 1259 /* 1260 * Pull snd_wl1 up to prevent seq wrap relative to 1261 * th_seq. 1262 */ 1263 tp->snd_wl1 = th->th_seq; 1264 /* 1265 * Pull rcv_up up to prevent seq wrap relative to 1266 * rcv_nxt. 1267 */ 1268 tp->rcv_up = tp->rcv_nxt; 1269 tcpstat.tcps_rcvpack++; 1270 tcpstat.tcps_rcvbyte += tlen; 1271 ND6_HINT(tp); /* some progress has been done */ 1272 #ifdef TCPDEBUG 1273 if (so->so_options & SO_DEBUG) 1274 tcp_trace(TA_INPUT, ostate, tp, 1275 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1276 #endif 1277 /* 1278 * Automatic sizing of receive socket buffer. Often the send 1279 * buffer size is not optimally adjusted to the actual network 1280 * conditions at hand (delay bandwidth product). Setting the 1281 * buffer size too small limits throughput on links with high 1282 * bandwidth and high delay (eg. trans-continental/oceanic links). 1283 * 1284 * On the receive side the socket buffer memory is only rarely 1285 * used to any significant extent. This allows us to be much 1286 * more aggressive in scaling the receive socket buffer. For 1287 * the case that the buffer space is actually used to a large 1288 * extent and we run out of kernel memory we can simply drop 1289 * the new segments; TCP on the sender will just retransmit it 1290 * later. Setting the buffer size too big may only consume too 1291 * much kernel memory if the application doesn't read() from 1292 * the socket or packet loss or reordering makes use of the 1293 * reassembly queue. 1294 * 1295 * The criteria to step up the receive buffer one notch are: 1296 * 1. the number of bytes received during the time it takes 1297 * one timestamp to be reflected back to us (the RTT); 1298 * 2. received bytes per RTT is within seven eighth of the 1299 * current socket buffer size; 1300 * 3. receive buffer size has not hit maximal automatic size; 1301 * 1302 * This algorithm does one step per RTT at most and only if 1303 * we receive a bulk stream w/o packet losses or reorderings. 1304 * Shrinking the buffer during idle times is not necessary as 1305 * it doesn't consume any memory when idle. 1306 * 1307 * TODO: Only step up if the application is actually serving 1308 * the buffer to better manage the socket buffer resources. 1309 */ 1310 if (tcp_do_autorcvbuf && 1311 to.to_tsecr && 1312 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1313 if (to.to_tsecr > tp->rfbuf_ts && 1314 to.to_tsecr - tp->rfbuf_ts < hz) { 1315 if (tp->rfbuf_cnt > 1316 (so->so_rcv.sb_hiwat / 8 * 7) && 1317 so->so_rcv.sb_hiwat < 1318 tcp_autorcvbuf_max) { 1319 newsize = 1320 min(so->so_rcv.sb_hiwat + 1321 tcp_autorcvbuf_inc, 1322 tcp_autorcvbuf_max); 1323 } 1324 /* Start over with next RTT. */ 1325 tp->rfbuf_ts = 0; 1326 tp->rfbuf_cnt = 0; 1327 } else 1328 tp->rfbuf_cnt += tlen; /* add up */ 1329 } 1330 1331 /* Add data to socket buffer. */ 1332 SOCKBUF_LOCK(&so->so_rcv); 1333 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1334 m_freem(m); 1335 } else { 1336 /* 1337 * Set new socket buffer size. 1338 * Give up when limit is reached. 1339 */ 1340 if (newsize) 1341 if (!sbreserve_locked(&so->so_rcv, 1342 newsize, so, curthread)) 1343 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1344 m_adj(m, drop_hdrlen); /* delayed header drop */ 1345 sbappendstream_locked(&so->so_rcv, m); 1346 } 1347 sorwakeup_locked(so); 1348 if (DELAY_ACK(tp)) { 1349 tp->t_flags |= TF_DELACK; 1350 } else { 1351 tp->t_flags |= TF_ACKNOW; 1352 tcp_output(tp); 1353 } 1354 goto check_delack; 1355 } 1356 } 1357 1358 /* 1359 * Calculate amount of space in receive window, 1360 * and then do TCP input processing. 1361 * Receive window is amount of space in rcv queue, 1362 * but not less than advertised window. 1363 */ 1364 win = sbspace(&so->so_rcv); 1365 if (win < 0) 1366 win = 0; 1367 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1368 1369 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1370 tp->rfbuf_ts = 0; 1371 tp->rfbuf_cnt = 0; 1372 1373 switch (tp->t_state) { 1374 1375 /* 1376 * If the state is SYN_RECEIVED: 1377 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1378 */ 1379 case TCPS_SYN_RECEIVED: 1380 if ((thflags & TH_ACK) && 1381 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1382 SEQ_GT(th->th_ack, tp->snd_max))) { 1383 rstreason = BANDLIM_RST_OPENPORT; 1384 goto dropwithreset; 1385 } 1386 break; 1387 1388 /* 1389 * If the state is SYN_SENT: 1390 * if seg contains an ACK, but not for our SYN, drop the input. 1391 * if seg contains a RST, then drop the connection. 1392 * if seg does not contain SYN, then drop it. 1393 * Otherwise this is an acceptable SYN segment 1394 * initialize tp->rcv_nxt and tp->irs 1395 * if seg contains ack then advance tp->snd_una 1396 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1397 * arrange for segment to be acked (eventually) 1398 * continue processing rest of data/controls, beginning with URG 1399 */ 1400 case TCPS_SYN_SENT: 1401 if ((thflags & TH_ACK) && 1402 (SEQ_LEQ(th->th_ack, tp->iss) || 1403 SEQ_GT(th->th_ack, tp->snd_max))) { 1404 rstreason = BANDLIM_UNLIMITED; 1405 goto dropwithreset; 1406 } 1407 if (thflags & TH_RST) { 1408 if (thflags & TH_ACK) { 1409 KASSERT(headlocked, ("%s: after_listen: " 1410 "tcp_drop.2: head not locked", __func__)); 1411 tp = tcp_drop(tp, ECONNREFUSED); 1412 } 1413 goto drop; 1414 } 1415 if ((thflags & TH_SYN) == 0) 1416 goto drop; 1417 1418 tp->irs = th->th_seq; 1419 tcp_rcvseqinit(tp); 1420 if (thflags & TH_ACK) { 1421 tcpstat.tcps_connects++; 1422 soisconnected(so); 1423 #ifdef MAC 1424 SOCK_LOCK(so); 1425 mac_set_socket_peer_from_mbuf(m, so); 1426 SOCK_UNLOCK(so); 1427 #endif 1428 /* Do window scaling on this connection? */ 1429 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1430 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1431 tp->rcv_scale = tp->request_r_scale; 1432 } 1433 tp->rcv_adv += tp->rcv_wnd; 1434 tp->snd_una++; /* SYN is acked */ 1435 /* 1436 * If there's data, delay ACK; if there's also a FIN 1437 * ACKNOW will be turned on later. 1438 */ 1439 if (DELAY_ACK(tp) && tlen != 0) 1440 tcp_timer_activate(tp, TT_DELACK, 1441 tcp_delacktime); 1442 else 1443 tp->t_flags |= TF_ACKNOW; 1444 /* 1445 * Received <SYN,ACK> in SYN_SENT[*] state. 1446 * Transitions: 1447 * SYN_SENT --> ESTABLISHED 1448 * SYN_SENT* --> FIN_WAIT_1 1449 */ 1450 tp->t_starttime = ticks; 1451 if (tp->t_flags & TF_NEEDFIN) { 1452 tp->t_state = TCPS_FIN_WAIT_1; 1453 tp->t_flags &= ~TF_NEEDFIN; 1454 thflags &= ~TH_SYN; 1455 } else { 1456 tp->t_state = TCPS_ESTABLISHED; 1457 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1458 } 1459 } else { 1460 /* 1461 * Received initial SYN in SYN-SENT[*] state => 1462 * simultaneous open. If segment contains CC option 1463 * and there is a cached CC, apply TAO test. 1464 * If it succeeds, connection is * half-synchronized. 1465 * Otherwise, do 3-way handshake: 1466 * SYN-SENT -> SYN-RECEIVED 1467 * SYN-SENT* -> SYN-RECEIVED* 1468 * If there was no CC option, clear cached CC value. 1469 */ 1470 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1471 tcp_timer_activate(tp, TT_REXMT, 0); 1472 tp->t_state = TCPS_SYN_RECEIVED; 1473 } 1474 1475 KASSERT(headlocked, ("%s: trimthenstep6: head not locked", 1476 __func__)); 1477 INP_LOCK_ASSERT(tp->t_inpcb); 1478 1479 /* 1480 * Advance th->th_seq to correspond to first data byte. 1481 * If data, trim to stay within window, 1482 * dropping FIN if necessary. 1483 */ 1484 th->th_seq++; 1485 if (tlen > tp->rcv_wnd) { 1486 todrop = tlen - tp->rcv_wnd; 1487 m_adj(m, -todrop); 1488 tlen = tp->rcv_wnd; 1489 thflags &= ~TH_FIN; 1490 tcpstat.tcps_rcvpackafterwin++; 1491 tcpstat.tcps_rcvbyteafterwin += todrop; 1492 } 1493 tp->snd_wl1 = th->th_seq - 1; 1494 tp->rcv_up = th->th_seq; 1495 /* 1496 * Client side of transaction: already sent SYN and data. 1497 * If the remote host used T/TCP to validate the SYN, 1498 * our data will be ACK'd; if so, enter normal data segment 1499 * processing in the middle of step 5, ack processing. 1500 * Otherwise, goto step 6. 1501 */ 1502 if (thflags & TH_ACK) 1503 goto process_ACK; 1504 1505 goto step6; 1506 1507 /* 1508 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1509 * do normal processing. 1510 * 1511 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1512 */ 1513 case TCPS_LAST_ACK: 1514 case TCPS_CLOSING: 1515 case TCPS_TIME_WAIT: 1516 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: timewait", 1517 __func__)); 1518 break; /* continue normal processing */ 1519 } 1520 1521 /* 1522 * States other than LISTEN or SYN_SENT. 1523 * First check the RST flag and sequence number since reset segments 1524 * are exempt from the timestamp and connection count tests. This 1525 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1526 * below which allowed reset segments in half the sequence space 1527 * to fall though and be processed (which gives forged reset 1528 * segments with a random sequence number a 50 percent chance of 1529 * killing a connection). 1530 * Then check timestamp, if present. 1531 * Then check the connection count, if present. 1532 * Then check that at least some bytes of segment are within 1533 * receive window. If segment begins before rcv_nxt, 1534 * drop leading data (and SYN); if nothing left, just ack. 1535 * 1536 * 1537 * If the RST bit is set, check the sequence number to see 1538 * if this is a valid reset segment. 1539 * RFC 793 page 37: 1540 * In all states except SYN-SENT, all reset (RST) segments 1541 * are validated by checking their SEQ-fields. A reset is 1542 * valid if its sequence number is in the window. 1543 * Note: this does not take into account delayed ACKs, so 1544 * we should test against last_ack_sent instead of rcv_nxt. 1545 * The sequence number in the reset segment is normally an 1546 * echo of our outgoing acknowlegement numbers, but some hosts 1547 * send a reset with the sequence number at the rightmost edge 1548 * of our receive window, and we have to handle this case. 1549 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1550 * that brute force RST attacks are possible. To combat this, 1551 * we use a much stricter check while in the ESTABLISHED state, 1552 * only accepting RSTs where the sequence number is equal to 1553 * last_ack_sent. In all other states (the states in which a 1554 * RST is more likely), the more permissive check is used. 1555 * If we have multiple segments in flight, the intial reset 1556 * segment sequence numbers will be to the left of last_ack_sent, 1557 * but they will eventually catch up. 1558 * In any case, it never made sense to trim reset segments to 1559 * fit the receive window since RFC 1122 says: 1560 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1561 * 1562 * A TCP SHOULD allow a received RST segment to include data. 1563 * 1564 * DISCUSSION 1565 * It has been suggested that a RST segment could contain 1566 * ASCII text that encoded and explained the cause of the 1567 * RST. No standard has yet been established for such 1568 * data. 1569 * 1570 * If the reset segment passes the sequence number test examine 1571 * the state: 1572 * SYN_RECEIVED STATE: 1573 * If passive open, return to LISTEN state. 1574 * If active open, inform user that connection was refused. 1575 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1576 * Inform user that connection was reset, and close tcb. 1577 * CLOSING, LAST_ACK STATES: 1578 * Close the tcb. 1579 * TIME_WAIT STATE: 1580 * Drop the segment - see Stevens, vol. 2, p. 964 and 1581 * RFC 1337. 1582 */ 1583 if (thflags & TH_RST) { 1584 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1585 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1586 switch (tp->t_state) { 1587 1588 case TCPS_SYN_RECEIVED: 1589 so->so_error = ECONNREFUSED; 1590 goto close; 1591 1592 case TCPS_ESTABLISHED: 1593 if (tcp_insecure_rst == 0 && 1594 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 1595 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 1596 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1597 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 1598 tcpstat.tcps_badrst++; 1599 goto drop; 1600 } 1601 case TCPS_FIN_WAIT_1: 1602 case TCPS_FIN_WAIT_2: 1603 case TCPS_CLOSE_WAIT: 1604 so->so_error = ECONNRESET; 1605 close: 1606 tp->t_state = TCPS_CLOSED; 1607 tcpstat.tcps_drops++; 1608 KASSERT(headlocked, ("%s: trimthenstep6: " 1609 "tcp_close: head not locked", __func__)); 1610 tp = tcp_close(tp); 1611 break; 1612 1613 case TCPS_CLOSING: 1614 case TCPS_LAST_ACK: 1615 KASSERT(headlocked, ("%s: trimthenstep6: " 1616 "tcp_close.2: head not locked", __func__)); 1617 tp = tcp_close(tp); 1618 break; 1619 1620 case TCPS_TIME_WAIT: 1621 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1622 ("%s: timewait", __func__)); 1623 break; 1624 } 1625 } 1626 goto drop; 1627 } 1628 1629 /* 1630 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1631 * and it's less than ts_recent, drop it. 1632 */ 1633 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1634 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1635 1636 /* Check to see if ts_recent is over 24 days old. */ 1637 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1638 /* 1639 * Invalidate ts_recent. If this segment updates 1640 * ts_recent, the age will be reset later and ts_recent 1641 * will get a valid value. If it does not, setting 1642 * ts_recent to zero will at least satisfy the 1643 * requirement that zero be placed in the timestamp 1644 * echo reply when ts_recent isn't valid. The 1645 * age isn't reset until we get a valid ts_recent 1646 * because we don't want out-of-order segments to be 1647 * dropped when ts_recent is old. 1648 */ 1649 tp->ts_recent = 0; 1650 } else { 1651 tcpstat.tcps_rcvduppack++; 1652 tcpstat.tcps_rcvdupbyte += tlen; 1653 tcpstat.tcps_pawsdrop++; 1654 if (tlen) 1655 goto dropafterack; 1656 goto drop; 1657 } 1658 } 1659 1660 /* 1661 * In the SYN-RECEIVED state, validate that the packet belongs to 1662 * this connection before trimming the data to fit the receive 1663 * window. Check the sequence number versus IRS since we know 1664 * the sequence numbers haven't wrapped. This is a partial fix 1665 * for the "LAND" DoS attack. 1666 */ 1667 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1668 rstreason = BANDLIM_RST_OPENPORT; 1669 goto dropwithreset; 1670 } 1671 1672 todrop = tp->rcv_nxt - th->th_seq; 1673 if (todrop > 0) { 1674 if (thflags & TH_SYN) { 1675 thflags &= ~TH_SYN; 1676 th->th_seq++; 1677 if (th->th_urp > 1) 1678 th->th_urp--; 1679 else 1680 thflags &= ~TH_URG; 1681 todrop--; 1682 } 1683 /* 1684 * Following if statement from Stevens, vol. 2, p. 960. 1685 */ 1686 if (todrop > tlen 1687 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1688 /* 1689 * Any valid FIN must be to the left of the window. 1690 * At this point the FIN must be a duplicate or out 1691 * of sequence; drop it. 1692 */ 1693 thflags &= ~TH_FIN; 1694 1695 /* 1696 * Send an ACK to resynchronize and drop any data. 1697 * But keep on processing for RST or ACK. 1698 */ 1699 tp->t_flags |= TF_ACKNOW; 1700 todrop = tlen; 1701 tcpstat.tcps_rcvduppack++; 1702 tcpstat.tcps_rcvdupbyte += todrop; 1703 } else { 1704 tcpstat.tcps_rcvpartduppack++; 1705 tcpstat.tcps_rcvpartdupbyte += todrop; 1706 } 1707 drop_hdrlen += todrop; /* drop from the top afterwards */ 1708 th->th_seq += todrop; 1709 tlen -= todrop; 1710 if (th->th_urp > todrop) 1711 th->th_urp -= todrop; 1712 else { 1713 thflags &= ~TH_URG; 1714 th->th_urp = 0; 1715 } 1716 } 1717 1718 /* 1719 * If new data are received on a connection after the 1720 * user processes are gone, then RST the other end. 1721 */ 1722 if ((so->so_state & SS_NOFDREF) && 1723 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1724 KASSERT(headlocked, ("%s: trimthenstep6: tcp_close.3: head " 1725 "not locked", __func__)); 1726 tp = tcp_close(tp); 1727 tcpstat.tcps_rcvafterclose++; 1728 rstreason = BANDLIM_UNLIMITED; 1729 goto dropwithreset; 1730 } 1731 1732 /* 1733 * If segment ends after window, drop trailing data 1734 * (and PUSH and FIN); if nothing left, just ACK. 1735 */ 1736 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1737 if (todrop > 0) { 1738 tcpstat.tcps_rcvpackafterwin++; 1739 if (todrop >= tlen) { 1740 tcpstat.tcps_rcvbyteafterwin += tlen; 1741 /* 1742 * If a new connection request is received 1743 * while in TIME_WAIT, drop the old connection 1744 * and start over if the sequence numbers 1745 * are above the previous ones. 1746 */ 1747 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: timewait", 1748 __func__)); 1749 if (thflags & TH_SYN && 1750 tp->t_state == TCPS_TIME_WAIT && 1751 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1752 KASSERT(headlocked, ("%s: trimthenstep6: " 1753 "tcp_close.4: head not locked", __func__)); 1754 tp = tcp_close(tp); 1755 /* XXX: Shouldn't be possible. */ 1756 return (1); 1757 } 1758 /* 1759 * If window is closed can only take segments at 1760 * window edge, and have to drop data and PUSH from 1761 * incoming segments. Continue processing, but 1762 * remember to ack. Otherwise, drop segment 1763 * and ack. 1764 */ 1765 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1766 tp->t_flags |= TF_ACKNOW; 1767 tcpstat.tcps_rcvwinprobe++; 1768 } else 1769 goto dropafterack; 1770 } else 1771 tcpstat.tcps_rcvbyteafterwin += todrop; 1772 m_adj(m, -todrop); 1773 tlen -= todrop; 1774 thflags &= ~(TH_PUSH|TH_FIN); 1775 } 1776 1777 /* 1778 * If last ACK falls within this segment's sequence numbers, 1779 * record its timestamp. 1780 * NOTE: 1781 * 1) That the test incorporates suggestions from the latest 1782 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1783 * 2) That updating only on newer timestamps interferes with 1784 * our earlier PAWS tests, so this check should be solely 1785 * predicated on the sequence space of this segment. 1786 * 3) That we modify the segment boundary check to be 1787 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1788 * instead of RFC1323's 1789 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1790 * This modified check allows us to overcome RFC1323's 1791 * limitations as described in Stevens TCP/IP Illustrated 1792 * Vol. 2 p.869. In such cases, we can still calculate the 1793 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1794 */ 1795 if ((to.to_flags & TOF_TS) != 0 && 1796 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1797 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1798 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1799 tp->ts_recent_age = ticks; 1800 tp->ts_recent = to.to_tsval; 1801 } 1802 1803 /* 1804 * If a SYN is in the window, then this is an 1805 * error and we send an RST and drop the connection. 1806 */ 1807 if (thflags & TH_SYN) { 1808 KASSERT(headlocked, ("%s: tcp_drop: trimthenstep6: " 1809 "head not locked", __func__)); 1810 tp = tcp_drop(tp, ECONNRESET); 1811 rstreason = BANDLIM_UNLIMITED; 1812 goto drop; 1813 } 1814 1815 /* 1816 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1817 * flag is on (half-synchronized state), then queue data for 1818 * later processing; else drop segment and return. 1819 */ 1820 if ((thflags & TH_ACK) == 0) { 1821 if (tp->t_state == TCPS_SYN_RECEIVED || 1822 (tp->t_flags & TF_NEEDSYN)) 1823 goto step6; 1824 else if (tp->t_flags & TF_ACKNOW) 1825 goto dropafterack; 1826 else 1827 goto drop; 1828 } 1829 1830 /* 1831 * Ack processing. 1832 */ 1833 switch (tp->t_state) { 1834 1835 /* 1836 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1837 * ESTABLISHED state and continue processing. 1838 * The ACK was checked above. 1839 */ 1840 case TCPS_SYN_RECEIVED: 1841 1842 tcpstat.tcps_connects++; 1843 soisconnected(so); 1844 /* Do window scaling? */ 1845 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1846 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1847 tp->rcv_scale = tp->request_r_scale; 1848 tp->snd_wnd = tiwin; 1849 } 1850 /* 1851 * Make transitions: 1852 * SYN-RECEIVED -> ESTABLISHED 1853 * SYN-RECEIVED* -> FIN-WAIT-1 1854 */ 1855 tp->t_starttime = ticks; 1856 if (tp->t_flags & TF_NEEDFIN) { 1857 tp->t_state = TCPS_FIN_WAIT_1; 1858 tp->t_flags &= ~TF_NEEDFIN; 1859 } else { 1860 tp->t_state = TCPS_ESTABLISHED; 1861 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1862 } 1863 /* 1864 * If segment contains data or ACK, will call tcp_reass() 1865 * later; if not, do so now to pass queued data to user. 1866 */ 1867 if (tlen == 0 && (thflags & TH_FIN) == 0) 1868 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1869 (struct mbuf *)0); 1870 tp->snd_wl1 = th->th_seq - 1; 1871 /* FALLTHROUGH */ 1872 1873 /* 1874 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1875 * ACKs. If the ack is in the range 1876 * tp->snd_una < th->th_ack <= tp->snd_max 1877 * then advance tp->snd_una to th->th_ack and drop 1878 * data from the retransmission queue. If this ACK reflects 1879 * more up to date window information we update our window information. 1880 */ 1881 case TCPS_ESTABLISHED: 1882 case TCPS_FIN_WAIT_1: 1883 case TCPS_FIN_WAIT_2: 1884 case TCPS_CLOSE_WAIT: 1885 case TCPS_CLOSING: 1886 case TCPS_LAST_ACK: 1887 case TCPS_TIME_WAIT: 1888 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: timewait", 1889 __func__)); 1890 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1891 tcpstat.tcps_rcvacktoomuch++; 1892 goto dropafterack; 1893 } 1894 if (tp->sack_enable && 1895 ((to.to_flags & TOF_SACK) || 1896 !TAILQ_EMPTY(&tp->snd_holes))) 1897 tcp_sack_doack(tp, &to, th->th_ack); 1898 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1899 if (tlen == 0 && tiwin == tp->snd_wnd) { 1900 tcpstat.tcps_rcvdupack++; 1901 /* 1902 * If we have outstanding data (other than 1903 * a window probe), this is a completely 1904 * duplicate ack (ie, window info didn't 1905 * change), the ack is the biggest we've 1906 * seen and we've seen exactly our rexmt 1907 * threshhold of them, assume a packet 1908 * has been dropped and retransmit it. 1909 * Kludge snd_nxt & the congestion 1910 * window so we send only this one 1911 * packet. 1912 * 1913 * We know we're losing at the current 1914 * window size so do congestion avoidance 1915 * (set ssthresh to half the current window 1916 * and pull our congestion window back to 1917 * the new ssthresh). 1918 * 1919 * Dup acks mean that packets have left the 1920 * network (they're now cached at the receiver) 1921 * so bump cwnd by the amount in the receiver 1922 * to keep a constant cwnd packets in the 1923 * network. 1924 */ 1925 if (!tcp_timer_active(tp, TT_REXMT) || 1926 th->th_ack != tp->snd_una) 1927 tp->t_dupacks = 0; 1928 else if (++tp->t_dupacks > tcprexmtthresh || 1929 ((tcp_do_newreno || tp->sack_enable) && 1930 IN_FASTRECOVERY(tp))) { 1931 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 1932 int awnd; 1933 1934 /* 1935 * Compute the amount of data in flight first. 1936 * We can inject new data into the pipe iff 1937 * we have less than 1/2 the original window's 1938 * worth of data in flight. 1939 */ 1940 awnd = (tp->snd_nxt - tp->snd_fack) + 1941 tp->sackhint.sack_bytes_rexmit; 1942 if (awnd < tp->snd_ssthresh) { 1943 tp->snd_cwnd += tp->t_maxseg; 1944 if (tp->snd_cwnd > tp->snd_ssthresh) 1945 tp->snd_cwnd = tp->snd_ssthresh; 1946 } 1947 } else 1948 tp->snd_cwnd += tp->t_maxseg; 1949 (void) tcp_output(tp); 1950 goto drop; 1951 } else if (tp->t_dupacks == tcprexmtthresh) { 1952 tcp_seq onxt = tp->snd_nxt; 1953 u_int win; 1954 1955 /* 1956 * If we're doing sack, check to 1957 * see if we're already in sack 1958 * recovery. If we're not doing sack, 1959 * check to see if we're in newreno 1960 * recovery. 1961 */ 1962 if (tp->sack_enable) { 1963 if (IN_FASTRECOVERY(tp)) { 1964 tp->t_dupacks = 0; 1965 break; 1966 } 1967 } else if (tcp_do_newreno) { 1968 if (SEQ_LEQ(th->th_ack, 1969 tp->snd_recover)) { 1970 tp->t_dupacks = 0; 1971 break; 1972 } 1973 } 1974 win = min(tp->snd_wnd, tp->snd_cwnd) / 1975 2 / tp->t_maxseg; 1976 if (win < 2) 1977 win = 2; 1978 tp->snd_ssthresh = win * tp->t_maxseg; 1979 ENTER_FASTRECOVERY(tp); 1980 tp->snd_recover = tp->snd_max; 1981 tcp_timer_activate(tp, TT_REXMT, 0); 1982 tp->t_rtttime = 0; 1983 if (tp->sack_enable) { 1984 tcpstat.tcps_sack_recovery_episode++; 1985 tp->sack_newdata = tp->snd_nxt; 1986 tp->snd_cwnd = tp->t_maxseg; 1987 (void) tcp_output(tp); 1988 goto drop; 1989 } 1990 tp->snd_nxt = th->th_ack; 1991 tp->snd_cwnd = tp->t_maxseg; 1992 (void) tcp_output(tp); 1993 KASSERT(tp->snd_limited <= 2, 1994 ("%s: tp->snd_limited too big", 1995 __func__)); 1996 tp->snd_cwnd = tp->snd_ssthresh + 1997 tp->t_maxseg * 1998 (tp->t_dupacks - tp->snd_limited); 1999 if (SEQ_GT(onxt, tp->snd_nxt)) 2000 tp->snd_nxt = onxt; 2001 goto drop; 2002 } else if (tcp_do_rfc3042) { 2003 u_long oldcwnd = tp->snd_cwnd; 2004 tcp_seq oldsndmax = tp->snd_max; 2005 u_int sent; 2006 2007 KASSERT(tp->t_dupacks == 1 || 2008 tp->t_dupacks == 2, 2009 ("%s: dupacks not 1 or 2", 2010 __func__)); 2011 if (tp->t_dupacks == 1) 2012 tp->snd_limited = 0; 2013 tp->snd_cwnd = 2014 (tp->snd_nxt - tp->snd_una) + 2015 (tp->t_dupacks - tp->snd_limited) * 2016 tp->t_maxseg; 2017 (void) tcp_output(tp); 2018 sent = tp->snd_max - oldsndmax; 2019 if (sent > tp->t_maxseg) { 2020 KASSERT((tp->t_dupacks == 2 && 2021 tp->snd_limited == 0) || 2022 (sent == tp->t_maxseg + 1 && 2023 tp->t_flags & TF_SENTFIN), 2024 ("%s: sent too much", 2025 __func__)); 2026 tp->snd_limited = 2; 2027 } else if (sent > 0) 2028 ++tp->snd_limited; 2029 tp->snd_cwnd = oldcwnd; 2030 goto drop; 2031 } 2032 } else 2033 tp->t_dupacks = 0; 2034 break; 2035 } 2036 2037 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2038 ("%s: th_ack <= snd_una", __func__)); 2039 2040 /* 2041 * If the congestion window was inflated to account 2042 * for the other side's cached packets, retract it. 2043 */ 2044 if (tcp_do_newreno || tp->sack_enable) { 2045 if (IN_FASTRECOVERY(tp)) { 2046 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2047 if (tp->sack_enable) 2048 tcp_sack_partialack(tp, th); 2049 else 2050 tcp_newreno_partial_ack(tp, th); 2051 } else { 2052 /* 2053 * Out of fast recovery. 2054 * Window inflation should have left us 2055 * with approximately snd_ssthresh 2056 * outstanding data. 2057 * But in case we would be inclined to 2058 * send a burst, better to do it via 2059 * the slow start mechanism. 2060 */ 2061 if (SEQ_GT(th->th_ack + 2062 tp->snd_ssthresh, 2063 tp->snd_max)) 2064 tp->snd_cwnd = tp->snd_max - 2065 th->th_ack + 2066 tp->t_maxseg; 2067 else 2068 tp->snd_cwnd = tp->snd_ssthresh; 2069 } 2070 } 2071 } else { 2072 if (tp->t_dupacks >= tcprexmtthresh && 2073 tp->snd_cwnd > tp->snd_ssthresh) 2074 tp->snd_cwnd = tp->snd_ssthresh; 2075 } 2076 tp->t_dupacks = 0; 2077 /* 2078 * If we reach this point, ACK is not a duplicate, 2079 * i.e., it ACKs something we sent. 2080 */ 2081 if (tp->t_flags & TF_NEEDSYN) { 2082 /* 2083 * T/TCP: Connection was half-synchronized, and our 2084 * SYN has been ACK'd (so connection is now fully 2085 * synchronized). Go to non-starred state, 2086 * increment snd_una for ACK of SYN, and check if 2087 * we can do window scaling. 2088 */ 2089 tp->t_flags &= ~TF_NEEDSYN; 2090 tp->snd_una++; 2091 /* Do window scaling? */ 2092 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2093 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2094 tp->rcv_scale = tp->request_r_scale; 2095 /* Send window already scaled. */ 2096 } 2097 } 2098 2099 process_ACK: 2100 KASSERT(headlocked, ("%s: process_ACK: head not locked", 2101 __func__)); 2102 INP_LOCK_ASSERT(tp->t_inpcb); 2103 2104 acked = th->th_ack - tp->snd_una; 2105 tcpstat.tcps_rcvackpack++; 2106 tcpstat.tcps_rcvackbyte += acked; 2107 2108 /* 2109 * If we just performed our first retransmit, and the ACK 2110 * arrives within our recovery window, then it was a mistake 2111 * to do the retransmit in the first place. Recover our 2112 * original cwnd and ssthresh, and proceed to transmit where 2113 * we left off. 2114 */ 2115 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2116 ++tcpstat.tcps_sndrexmitbad; 2117 tp->snd_cwnd = tp->snd_cwnd_prev; 2118 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2119 tp->snd_recover = tp->snd_recover_prev; 2120 if (tp->t_flags & TF_WASFRECOVERY) 2121 ENTER_FASTRECOVERY(tp); 2122 tp->snd_nxt = tp->snd_max; 2123 tp->t_badrxtwin = 0; /* XXX probably not required */ 2124 } 2125 2126 /* 2127 * If we have a timestamp reply, update smoothed 2128 * round trip time. If no timestamp is present but 2129 * transmit timer is running and timed sequence 2130 * number was acked, update smoothed round trip time. 2131 * Since we now have an rtt measurement, cancel the 2132 * timer backoff (cf., Phil Karn's retransmit alg.). 2133 * Recompute the initial retransmit timer. 2134 * 2135 * Some boxes send broken timestamp replies 2136 * during the SYN+ACK phase, ignore 2137 * timestamps of 0 or we could calculate a 2138 * huge RTT and blow up the retransmit timer. 2139 */ 2140 if ((to.to_flags & TOF_TS) != 0 && 2141 to.to_tsecr) { 2142 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2143 tp->t_rttlow = ticks - to.to_tsecr; 2144 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2145 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2146 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2147 tp->t_rttlow = ticks - tp->t_rtttime; 2148 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2149 } 2150 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2151 2152 /* 2153 * If all outstanding data is acked, stop retransmit 2154 * timer and remember to restart (more output or persist). 2155 * If there is more data to be acked, restart retransmit 2156 * timer, using current (possibly backed-off) value. 2157 */ 2158 if (th->th_ack == tp->snd_max) { 2159 tcp_timer_activate(tp, TT_REXMT, 0); 2160 needoutput = 1; 2161 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2162 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2163 2164 /* 2165 * If no data (only SYN) was ACK'd, 2166 * skip rest of ACK processing. 2167 */ 2168 if (acked == 0) 2169 goto step6; 2170 2171 /* 2172 * When new data is acked, open the congestion window. 2173 * If the window gives us less than ssthresh packets 2174 * in flight, open exponentially (maxseg per packet). 2175 * Otherwise open linearly: maxseg per window 2176 * (maxseg^2 / cwnd per packet). 2177 */ 2178 if ((!tcp_do_newreno && !tp->sack_enable) || 2179 !IN_FASTRECOVERY(tp)) { 2180 u_int cw = tp->snd_cwnd; 2181 u_int incr = tp->t_maxseg; 2182 if (cw > tp->snd_ssthresh) 2183 incr = incr * incr / cw; 2184 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2185 } 2186 SOCKBUF_LOCK(&so->so_snd); 2187 if (acked > so->so_snd.sb_cc) { 2188 tp->snd_wnd -= so->so_snd.sb_cc; 2189 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2190 ourfinisacked = 1; 2191 } else { 2192 sbdrop_locked(&so->so_snd, acked); 2193 tp->snd_wnd -= acked; 2194 ourfinisacked = 0; 2195 } 2196 sowwakeup_locked(so); 2197 /* detect una wraparound */ 2198 if ((tcp_do_newreno || tp->sack_enable) && 2199 !IN_FASTRECOVERY(tp) && 2200 SEQ_GT(tp->snd_una, tp->snd_recover) && 2201 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2202 tp->snd_recover = th->th_ack - 1; 2203 if ((tcp_do_newreno || tp->sack_enable) && 2204 IN_FASTRECOVERY(tp) && 2205 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2206 EXIT_FASTRECOVERY(tp); 2207 tp->snd_una = th->th_ack; 2208 if (tp->sack_enable) { 2209 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2210 tp->snd_recover = tp->snd_una; 2211 } 2212 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2213 tp->snd_nxt = tp->snd_una; 2214 2215 switch (tp->t_state) { 2216 2217 /* 2218 * In FIN_WAIT_1 STATE in addition to the processing 2219 * for the ESTABLISHED state if our FIN is now acknowledged 2220 * then enter FIN_WAIT_2. 2221 */ 2222 case TCPS_FIN_WAIT_1: 2223 if (ourfinisacked) { 2224 /* 2225 * If we can't receive any more 2226 * data, then closing user can proceed. 2227 * Starting the timer is contrary to the 2228 * specification, but if we don't get a FIN 2229 * we'll hang forever. 2230 */ 2231 /* XXXjl 2232 * we should release the tp also, and use a 2233 * compressed state. 2234 */ 2235 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2236 int timeout; 2237 2238 soisdisconnected(so); 2239 timeout = (tcp_fast_finwait2_recycle) ? 2240 tcp_finwait2_timeout : tcp_maxidle; 2241 tcp_timer_activate(tp, TT_2MSL, timeout); 2242 } 2243 tp->t_state = TCPS_FIN_WAIT_2; 2244 } 2245 break; 2246 2247 /* 2248 * In CLOSING STATE in addition to the processing for 2249 * the ESTABLISHED state if the ACK acknowledges our FIN 2250 * then enter the TIME-WAIT state, otherwise ignore 2251 * the segment. 2252 */ 2253 case TCPS_CLOSING: 2254 if (ourfinisacked) { 2255 KASSERT(headlocked, ("%s: process_ACK: " 2256 "head not locked", __func__)); 2257 tcp_twstart(tp); 2258 INP_INFO_WUNLOCK(&tcbinfo); 2259 headlocked = 0; 2260 m_freem(m); 2261 return (0); 2262 } 2263 break; 2264 2265 /* 2266 * In LAST_ACK, we may still be waiting for data to drain 2267 * and/or to be acked, as well as for the ack of our FIN. 2268 * If our FIN is now acknowledged, delete the TCB, 2269 * enter the closed state and return. 2270 */ 2271 case TCPS_LAST_ACK: 2272 if (ourfinisacked) { 2273 KASSERT(headlocked, ("%s: process_ACK: " 2274 "tcp_close: head not locked", __func__)); 2275 tp = tcp_close(tp); 2276 goto drop; 2277 } 2278 break; 2279 2280 /* 2281 * In TIME_WAIT state the only thing that should arrive 2282 * is a retransmission of the remote FIN. Acknowledge 2283 * it and restart the finack timer. 2284 */ 2285 case TCPS_TIME_WAIT: 2286 KASSERT(tp->t_state != TCPS_TIME_WAIT, 2287 ("%s: timewait", __func__)); 2288 tcp_timer_activate(tp, TT_2MSL, 2 * tcp_msl); 2289 goto dropafterack; 2290 } 2291 } 2292 2293 step6: 2294 KASSERT(headlocked, ("%s: step6: head not locked", __func__)); 2295 INP_LOCK_ASSERT(tp->t_inpcb); 2296 2297 /* 2298 * Update window information. 2299 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2300 */ 2301 if ((thflags & TH_ACK) && 2302 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2303 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2304 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2305 /* keep track of pure window updates */ 2306 if (tlen == 0 && 2307 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2308 tcpstat.tcps_rcvwinupd++; 2309 tp->snd_wnd = tiwin; 2310 tp->snd_wl1 = th->th_seq; 2311 tp->snd_wl2 = th->th_ack; 2312 if (tp->snd_wnd > tp->max_sndwnd) 2313 tp->max_sndwnd = tp->snd_wnd; 2314 needoutput = 1; 2315 } 2316 2317 /* 2318 * Process segments with URG. 2319 */ 2320 if ((thflags & TH_URG) && th->th_urp && 2321 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2322 /* 2323 * This is a kludge, but if we receive and accept 2324 * random urgent pointers, we'll crash in 2325 * soreceive. It's hard to imagine someone 2326 * actually wanting to send this much urgent data. 2327 */ 2328 SOCKBUF_LOCK(&so->so_rcv); 2329 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2330 th->th_urp = 0; /* XXX */ 2331 thflags &= ~TH_URG; /* XXX */ 2332 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2333 goto dodata; /* XXX */ 2334 } 2335 /* 2336 * If this segment advances the known urgent pointer, 2337 * then mark the data stream. This should not happen 2338 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2339 * a FIN has been received from the remote side. 2340 * In these states we ignore the URG. 2341 * 2342 * According to RFC961 (Assigned Protocols), 2343 * the urgent pointer points to the last octet 2344 * of urgent data. We continue, however, 2345 * to consider it to indicate the first octet 2346 * of data past the urgent section as the original 2347 * spec states (in one of two places). 2348 */ 2349 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2350 tp->rcv_up = th->th_seq + th->th_urp; 2351 so->so_oobmark = so->so_rcv.sb_cc + 2352 (tp->rcv_up - tp->rcv_nxt) - 1; 2353 if (so->so_oobmark == 0) 2354 so->so_rcv.sb_state |= SBS_RCVATMARK; 2355 sohasoutofband(so); 2356 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2357 } 2358 SOCKBUF_UNLOCK(&so->so_rcv); 2359 /* 2360 * Remove out of band data so doesn't get presented to user. 2361 * This can happen independent of advancing the URG pointer, 2362 * but if two URG's are pending at once, some out-of-band 2363 * data may creep in... ick. 2364 */ 2365 if (th->th_urp <= (u_long)tlen && 2366 !(so->so_options & SO_OOBINLINE)) { 2367 /* hdr drop is delayed */ 2368 tcp_pulloutofband(so, th, m, drop_hdrlen); 2369 } 2370 } else { 2371 /* 2372 * If no out of band data is expected, 2373 * pull receive urgent pointer along 2374 * with the receive window. 2375 */ 2376 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2377 tp->rcv_up = tp->rcv_nxt; 2378 } 2379 dodata: /* XXX */ 2380 KASSERT(headlocked, ("%s: dodata: head not locked", __func__)); 2381 INP_LOCK_ASSERT(tp->t_inpcb); 2382 2383 /* 2384 * Process the segment text, merging it into the TCP sequencing queue, 2385 * and arranging for acknowledgment of receipt if necessary. 2386 * This process logically involves adjusting tp->rcv_wnd as data 2387 * is presented to the user (this happens in tcp_usrreq.c, 2388 * case PRU_RCVD). If a FIN has already been received on this 2389 * connection then we just ignore the text. 2390 */ 2391 if ((tlen || (thflags & TH_FIN)) && 2392 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2393 tcp_seq save_start = th->th_seq; 2394 tcp_seq save_end = th->th_seq + tlen; 2395 m_adj(m, drop_hdrlen); /* delayed header drop */ 2396 /* 2397 * Insert segment which includes th into TCP reassembly queue 2398 * with control block tp. Set thflags to whether reassembly now 2399 * includes a segment with FIN. This handles the common case 2400 * inline (segment is the next to be received on an established 2401 * connection, and the queue is empty), avoiding linkage into 2402 * and removal from the queue and repetition of various 2403 * conversions. 2404 * Set DELACK for segments received in order, but ack 2405 * immediately when segments are out of order (so 2406 * fast retransmit can work). 2407 */ 2408 if (th->th_seq == tp->rcv_nxt && 2409 LIST_EMPTY(&tp->t_segq) && 2410 TCPS_HAVEESTABLISHED(tp->t_state)) { 2411 if (DELAY_ACK(tp)) 2412 tp->t_flags |= TF_DELACK; 2413 else 2414 tp->t_flags |= TF_ACKNOW; 2415 tp->rcv_nxt += tlen; 2416 thflags = th->th_flags & TH_FIN; 2417 tcpstat.tcps_rcvpack++; 2418 tcpstat.tcps_rcvbyte += tlen; 2419 ND6_HINT(tp); 2420 SOCKBUF_LOCK(&so->so_rcv); 2421 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2422 m_freem(m); 2423 else 2424 sbappendstream_locked(&so->so_rcv, m); 2425 sorwakeup_locked(so); 2426 } else { 2427 thflags = tcp_reass(tp, th, &tlen, m); 2428 tp->t_flags |= TF_ACKNOW; 2429 } 2430 if (tlen > 0 && tp->sack_enable) 2431 tcp_update_sack_list(tp, save_start, save_end); 2432 #if 0 2433 /* 2434 * Note the amount of data that peer has sent into 2435 * our window, in order to estimate the sender's 2436 * buffer size. 2437 * XXX: Unused. 2438 */ 2439 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2440 #endif 2441 } else { 2442 m_freem(m); 2443 thflags &= ~TH_FIN; 2444 } 2445 2446 /* 2447 * If FIN is received ACK the FIN and let the user know 2448 * that the connection is closing. 2449 */ 2450 if (thflags & TH_FIN) { 2451 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2452 socantrcvmore(so); 2453 /* 2454 * If connection is half-synchronized 2455 * (ie NEEDSYN flag on) then delay ACK, 2456 * so it may be piggybacked when SYN is sent. 2457 * Otherwise, since we received a FIN then no 2458 * more input can be expected, send ACK now. 2459 */ 2460 if (tp->t_flags & TF_NEEDSYN) 2461 tp->t_flags |= TF_DELACK; 2462 else 2463 tp->t_flags |= TF_ACKNOW; 2464 tp->rcv_nxt++; 2465 } 2466 switch (tp->t_state) { 2467 2468 /* 2469 * In SYN_RECEIVED and ESTABLISHED STATES 2470 * enter the CLOSE_WAIT state. 2471 */ 2472 case TCPS_SYN_RECEIVED: 2473 tp->t_starttime = ticks; 2474 /*FALLTHROUGH*/ 2475 case TCPS_ESTABLISHED: 2476 tp->t_state = TCPS_CLOSE_WAIT; 2477 break; 2478 2479 /* 2480 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2481 * enter the CLOSING state. 2482 */ 2483 case TCPS_FIN_WAIT_1: 2484 tp->t_state = TCPS_CLOSING; 2485 break; 2486 2487 /* 2488 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2489 * starting the time-wait timer, turning off the other 2490 * standard timers. 2491 */ 2492 case TCPS_FIN_WAIT_2: 2493 KASSERT(headlocked == 1, ("%s: dodata: " 2494 "TCP_FIN_WAIT_2: head not locked", __func__)); 2495 tcp_twstart(tp); 2496 INP_INFO_WUNLOCK(&tcbinfo); 2497 return (0); 2498 2499 /* 2500 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2501 */ 2502 case TCPS_TIME_WAIT: 2503 KASSERT(tp->t_state != TCPS_TIME_WAIT, 2504 ("%s: timewait", __func__)); 2505 tcp_timer_activate(tp, TT_2MSL, 2 * tcp_msl); 2506 break; 2507 } 2508 } 2509 INP_INFO_WUNLOCK(&tcbinfo); 2510 headlocked = 0; 2511 #ifdef TCPDEBUG 2512 if (so->so_options & SO_DEBUG) 2513 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2514 &tcp_savetcp, 0); 2515 #endif 2516 2517 /* 2518 * Return any desired output. 2519 */ 2520 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2521 (void) tcp_output(tp); 2522 2523 check_delack: 2524 KASSERT(headlocked == 0, ("%s: check_delack: head locked", 2525 __func__)); 2526 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 2527 INP_LOCK_ASSERT(tp->t_inpcb); 2528 if (tp->t_flags & TF_DELACK) { 2529 tp->t_flags &= ~TF_DELACK; 2530 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2531 } 2532 INP_UNLOCK(tp->t_inpcb); 2533 return (0); 2534 2535 dropafterack: 2536 KASSERT(headlocked, ("%s: dropafterack: head not locked", __func__)); 2537 /* 2538 * Generate an ACK dropping incoming segment if it occupies 2539 * sequence space, where the ACK reflects our state. 2540 * 2541 * We can now skip the test for the RST flag since all 2542 * paths to this code happen after packets containing 2543 * RST have been dropped. 2544 * 2545 * In the SYN-RECEIVED state, don't send an ACK unless the 2546 * segment we received passes the SYN-RECEIVED ACK test. 2547 * If it fails send a RST. This breaks the loop in the 2548 * "LAND" DoS attack, and also prevents an ACK storm 2549 * between two listening ports that have been sent forged 2550 * SYN segments, each with the source address of the other. 2551 */ 2552 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2553 (SEQ_GT(tp->snd_una, th->th_ack) || 2554 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2555 rstreason = BANDLIM_RST_OPENPORT; 2556 goto dropwithreset; 2557 } 2558 #ifdef TCPDEBUG 2559 if (so->so_options & SO_DEBUG) 2560 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2561 &tcp_savetcp, 0); 2562 #endif 2563 KASSERT(headlocked, ("%s: headlocked should be 1", __func__)); 2564 INP_INFO_WUNLOCK(&tcbinfo); 2565 tp->t_flags |= TF_ACKNOW; 2566 (void) tcp_output(tp); 2567 INP_UNLOCK(tp->t_inpcb); 2568 m_freem(m); 2569 return (0); 2570 2571 dropwithreset: 2572 KASSERT(headlocked, ("%s: dropwithreset: head not locked", __func__)); 2573 2574 tcp_dropwithreset(m, th, tp, tlen, rstreason); 2575 2576 if (tp != NULL) 2577 INP_UNLOCK(tp->t_inpcb); 2578 if (headlocked) 2579 INP_INFO_WUNLOCK(&tcbinfo); 2580 return (0); 2581 2582 drop: 2583 /* 2584 * Drop space held by incoming segment and return. 2585 */ 2586 #ifdef TCPDEBUG 2587 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2588 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2589 &tcp_savetcp, 0); 2590 #endif 2591 if (tp != NULL) 2592 INP_UNLOCK(tp->t_inpcb); 2593 if (headlocked) 2594 INP_INFO_WUNLOCK(&tcbinfo); 2595 m_freem(m); 2596 return (0); 2597 } 2598 2599 2600 /* 2601 * Issue RST on TCP segment. The mbuf must still include the original 2602 * packet header. 2603 */ 2604 static void 2605 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 2606 int tlen, int rstreason) 2607 { 2608 struct ip *ip; 2609 #ifdef INET6 2610 struct ip6_hdr *ip6; 2611 #endif 2612 /* 2613 * Generate a RST, dropping incoming segment. 2614 * Make ACK acceptable to originator of segment. 2615 * Don't bother to respond if destination was broadcast/multicast. 2616 * tp may be NULL. 2617 */ 2618 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2619 goto drop; 2620 #ifdef INET6 2621 if (mtod(m, struct ip *)->ip_v == 6) { 2622 ip6 = mtod(m, struct ip6_hdr *); 2623 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2624 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2625 goto drop; 2626 /* IPv6 anycast check is done at tcp6_input() */ 2627 } else 2628 #endif 2629 { 2630 ip = mtod(m, struct ip *); 2631 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2632 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2633 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2634 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2635 goto drop; 2636 } 2637 2638 /* Perform bandwidth limiting. */ 2639 if (badport_bandlim(rstreason) < 0) 2640 goto drop; 2641 2642 /* tcp_respond consumes the mbuf chain. */ 2643 if (th->th_flags & TH_ACK) { 2644 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 2645 th->th_ack, TH_RST); 2646 } else { 2647 if (th->th_flags & TH_SYN) 2648 tlen++; 2649 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2650 (tcp_seq)0, TH_RST|TH_ACK); 2651 } 2652 return; 2653 drop: 2654 m_freem(m); 2655 return; 2656 } 2657 2658 /* 2659 * Parse TCP options and place in tcpopt. 2660 */ 2661 static void 2662 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 2663 { 2664 int opt, optlen; 2665 2666 to->to_flags = 0; 2667 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2668 opt = cp[0]; 2669 if (opt == TCPOPT_EOL) 2670 break; 2671 if (opt == TCPOPT_NOP) 2672 optlen = 1; 2673 else { 2674 if (cnt < 2) 2675 break; 2676 optlen = cp[1]; 2677 if (optlen < 2 || optlen > cnt) 2678 break; 2679 } 2680 switch (opt) { 2681 case TCPOPT_MAXSEG: 2682 if (optlen != TCPOLEN_MAXSEG) 2683 continue; 2684 if (!(flags & TO_SYN)) 2685 continue; 2686 to->to_flags |= TOF_MSS; 2687 bcopy((char *)cp + 2, 2688 (char *)&to->to_mss, sizeof(to->to_mss)); 2689 to->to_mss = ntohs(to->to_mss); 2690 break; 2691 case TCPOPT_WINDOW: 2692 if (optlen != TCPOLEN_WINDOW) 2693 continue; 2694 if (!(flags & TO_SYN)) 2695 continue; 2696 to->to_flags |= TOF_SCALE; 2697 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 2698 break; 2699 case TCPOPT_TIMESTAMP: 2700 if (optlen != TCPOLEN_TIMESTAMP) 2701 continue; 2702 to->to_flags |= TOF_TS; 2703 bcopy((char *)cp + 2, 2704 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2705 to->to_tsval = ntohl(to->to_tsval); 2706 bcopy((char *)cp + 6, 2707 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2708 to->to_tsecr = ntohl(to->to_tsecr); 2709 break; 2710 #ifdef TCP_SIGNATURE 2711 /* 2712 * XXX In order to reply to a host which has set the 2713 * TCP_SIGNATURE option in its initial SYN, we have to 2714 * record the fact that the option was observed here 2715 * for the syncache code to perform the correct response. 2716 */ 2717 case TCPOPT_SIGNATURE: 2718 if (optlen != TCPOLEN_SIGNATURE) 2719 continue; 2720 to->to_flags |= TOF_SIGNATURE; 2721 to->to_signature = cp + 2; 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