1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_ipfw.h" /* for ipfw_fwd */ 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #include "opt_ipsec.h" 37 #include "opt_mac.h" 38 #include "opt_tcpdebug.h" 39 40 #include <sys/param.h> 41 #include <sys/kernel.h> 42 #include <sys/malloc.h> 43 #include <sys/mbuf.h> 44 #include <sys/proc.h> /* for proc0 declaration */ 45 #include <sys/protosw.h> 46 #include <sys/signalvar.h> 47 #include <sys/socket.h> 48 #include <sys/socketvar.h> 49 #include <sys/sysctl.h> 50 #include <sys/syslog.h> 51 #include <sys/systm.h> 52 53 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 54 55 #include <vm/uma.h> 56 57 #include <net/if.h> 58 #include <net/route.h> 59 60 #include <netinet/in.h> 61 #include <netinet/in_pcb.h> 62 #include <netinet/in_systm.h> 63 #include <netinet/in_var.h> 64 #include <netinet/ip.h> 65 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 66 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 67 #include <netinet/ip_var.h> 68 #include <netinet/ip_options.h> 69 #include <netinet/ip6.h> 70 #include <netinet/icmp6.h> 71 #include <netinet6/in6_pcb.h> 72 #include <netinet6/ip6_var.h> 73 #include <netinet6/nd6.h> 74 #include <netinet/tcp.h> 75 #include <netinet/tcp_fsm.h> 76 #include <netinet/tcp_seq.h> 77 #include <netinet/tcp_timer.h> 78 #include <netinet/tcp_var.h> 79 #include <netinet6/tcp6_var.h> 80 #include <netinet/tcpip.h> 81 #ifdef TCPDEBUG 82 #include <netinet/tcp_debug.h> 83 #endif /* TCPDEBUG */ 84 85 #ifdef FAST_IPSEC 86 #include <netipsec/ipsec.h> 87 #include <netipsec/ipsec6.h> 88 #endif /*FAST_IPSEC*/ 89 90 #ifdef IPSEC 91 #include <netinet6/ipsec.h> 92 #include <netinet6/ipsec6.h> 93 #include <netkey/key.h> 94 #endif /*IPSEC*/ 95 96 #include <machine/in_cksum.h> 97 98 #include <security/mac/mac_framework.h> 99 100 static const int tcprexmtthresh = 3; 101 102 struct tcpstat tcpstat; 103 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 104 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 105 106 static int tcp_log_in_vain = 0; 107 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 108 &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports"); 109 110 static int blackhole = 0; 111 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 112 &blackhole, 0, "Do not send RST on segments to closed ports"); 113 114 int tcp_delack_enabled = 1; 115 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 116 &tcp_delack_enabled, 0, 117 "Delay ACK to try and piggyback it onto a data packet"); 118 119 static int drop_synfin = 0; 120 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 121 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 122 123 static int tcp_do_rfc3042 = 1; 124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 125 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 126 127 static int tcp_do_rfc3390 = 1; 128 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 129 &tcp_do_rfc3390, 0, 130 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 131 132 static int tcp_insecure_rst = 0; 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 134 &tcp_insecure_rst, 0, 135 "Follow the old (insecure) criteria for accepting RST packets"); 136 137 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 138 "TCP Segment Reassembly Queue"); 139 140 static int tcp_reass_maxseg = 0; 141 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 142 &tcp_reass_maxseg, 0, 143 "Global maximum number of TCP Segments in Reassembly Queue"); 144 145 int tcp_reass_qsize = 0; 146 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 147 &tcp_reass_qsize, 0, 148 "Global number of TCP Segments currently in Reassembly Queue"); 149 150 static int tcp_reass_maxqlen = 48; 151 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 152 &tcp_reass_maxqlen, 0, 153 "Maximum number of TCP Segments per individual Reassembly Queue"); 154 155 static int tcp_reass_overflows = 0; 156 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 157 &tcp_reass_overflows, 0, 158 "Global number of TCP Segment Reassembly Queue Overflows"); 159 160 int tcp_do_autorcvbuf = 1; 161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 162 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing"); 163 164 int tcp_autorcvbuf_inc = 16*1024; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 166 &tcp_autorcvbuf_inc, 0, 167 "Incrementor step size of automatic receive buffer"); 168 169 int tcp_autorcvbuf_max = 256*1024; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 171 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer"); 172 173 struct inpcbhead tcb; 174 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 175 struct inpcbinfo tcbinfo; 176 177 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 178 static int tcp_do_segment(struct mbuf *, struct tcphdr *, 179 struct socket *, struct tcpcb *, int, int); 180 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 181 struct tcpcb *, int, int); 182 static void tcp_pulloutofband(struct socket *, 183 struct tcphdr *, struct mbuf *, int); 184 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 185 struct mbuf *); 186 static void tcp_xmit_timer(struct tcpcb *, int); 187 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 188 static int tcp_timewait(struct inpcb *, struct tcpopt *, 189 struct tcphdr *, struct mbuf *, int); 190 191 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 192 #ifdef INET6 193 #define ND6_HINT(tp) \ 194 do { \ 195 if ((tp) && (tp)->t_inpcb && \ 196 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 197 nd6_nud_hint(NULL, NULL, 0); \ 198 } while (0) 199 #else 200 #define ND6_HINT(tp) 201 #endif 202 203 /* 204 * Indicate whether this ack should be delayed. We can delay the ack if 205 * - there is no delayed ack timer in progress and 206 * - our last ack wasn't a 0-sized window. We never want to delay 207 * the ack that opens up a 0-sized window and 208 * - delayed acks are enabled or 209 * - this is a half-synchronized T/TCP connection. 210 */ 211 #define DELAY_ACK(tp) \ 212 ((!tcp_timer_active(tp, TT_DELACK) && \ 213 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 214 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 215 216 /* Initialize TCP reassembly queue */ 217 static void 218 tcp_reass_zone_change(void *tag) 219 { 220 221 tcp_reass_maxseg = nmbclusters / 16; 222 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 223 } 224 225 uma_zone_t tcp_reass_zone; 226 void 227 tcp_reass_init() 228 { 229 tcp_reass_maxseg = nmbclusters / 16; 230 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 231 &tcp_reass_maxseg); 232 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 233 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 234 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 235 EVENTHANDLER_REGISTER(nmbclusters_change, 236 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY); 237 } 238 239 static int 240 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m) 241 { 242 struct tseg_qent *q; 243 struct tseg_qent *p = NULL; 244 struct tseg_qent *nq; 245 struct tseg_qent *te = NULL; 246 struct socket *so = tp->t_inpcb->inp_socket; 247 int flags; 248 249 INP_LOCK_ASSERT(tp->t_inpcb); 250 251 /* 252 * XXX: tcp_reass() is rather inefficient with its data structures 253 * and should be rewritten (see NetBSD for optimizations). While 254 * doing that it should move to its own file tcp_reass.c. 255 */ 256 257 /* 258 * Call with th==NULL after become established to 259 * force pre-ESTABLISHED data up to user socket. 260 */ 261 if (th == NULL) 262 goto present; 263 264 /* 265 * Limit the number of segments in the reassembly queue to prevent 266 * holding on to too many segments (and thus running out of mbufs). 267 * Make sure to let the missing segment through which caused this 268 * queue. Always keep one global queue entry spare to be able to 269 * process the missing segment. 270 */ 271 if (th->th_seq != tp->rcv_nxt && 272 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 273 tp->t_segqlen >= tcp_reass_maxqlen)) { 274 tcp_reass_overflows++; 275 tcpstat.tcps_rcvmemdrop++; 276 m_freem(m); 277 *tlenp = 0; 278 return (0); 279 } 280 281 /* 282 * Allocate a new queue entry. If we can't, or hit the zone limit 283 * just drop the pkt. 284 */ 285 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 286 if (te == NULL) { 287 tcpstat.tcps_rcvmemdrop++; 288 m_freem(m); 289 *tlenp = 0; 290 return (0); 291 } 292 tp->t_segqlen++; 293 tcp_reass_qsize++; 294 295 /* 296 * Find a segment which begins after this one does. 297 */ 298 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 299 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 300 break; 301 p = q; 302 } 303 304 /* 305 * If there is a preceding segment, it may provide some of 306 * our data already. If so, drop the data from the incoming 307 * segment. If it provides all of our data, drop us. 308 */ 309 if (p != NULL) { 310 int i; 311 /* conversion to int (in i) handles seq wraparound */ 312 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 313 if (i > 0) { 314 if (i >= *tlenp) { 315 tcpstat.tcps_rcvduppack++; 316 tcpstat.tcps_rcvdupbyte += *tlenp; 317 m_freem(m); 318 uma_zfree(tcp_reass_zone, te); 319 tp->t_segqlen--; 320 tcp_reass_qsize--; 321 /* 322 * Try to present any queued data 323 * at the left window edge to the user. 324 * This is needed after the 3-WHS 325 * completes. 326 */ 327 goto present; /* ??? */ 328 } 329 m_adj(m, i); 330 *tlenp -= i; 331 th->th_seq += i; 332 } 333 } 334 tcpstat.tcps_rcvoopack++; 335 tcpstat.tcps_rcvoobyte += *tlenp; 336 337 /* 338 * While we overlap succeeding segments trim them or, 339 * if they are completely covered, dequeue them. 340 */ 341 while (q) { 342 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 343 if (i <= 0) 344 break; 345 if (i < q->tqe_len) { 346 q->tqe_th->th_seq += i; 347 q->tqe_len -= i; 348 m_adj(q->tqe_m, i); 349 break; 350 } 351 352 nq = LIST_NEXT(q, tqe_q); 353 LIST_REMOVE(q, tqe_q); 354 m_freem(q->tqe_m); 355 uma_zfree(tcp_reass_zone, q); 356 tp->t_segqlen--; 357 tcp_reass_qsize--; 358 q = nq; 359 } 360 361 /* Insert the new segment queue entry into place. */ 362 te->tqe_m = m; 363 te->tqe_th = th; 364 te->tqe_len = *tlenp; 365 366 if (p == NULL) { 367 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 368 } else { 369 LIST_INSERT_AFTER(p, te, tqe_q); 370 } 371 372 present: 373 /* 374 * Present data to user, advancing rcv_nxt through 375 * completed sequence space. 376 */ 377 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 378 return (0); 379 q = LIST_FIRST(&tp->t_segq); 380 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 381 return (0); 382 SOCKBUF_LOCK(&so->so_rcv); 383 do { 384 tp->rcv_nxt += q->tqe_len; 385 flags = q->tqe_th->th_flags & TH_FIN; 386 nq = LIST_NEXT(q, tqe_q); 387 LIST_REMOVE(q, tqe_q); 388 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 389 m_freem(q->tqe_m); 390 else 391 sbappendstream_locked(&so->so_rcv, q->tqe_m); 392 uma_zfree(tcp_reass_zone, q); 393 tp->t_segqlen--; 394 tcp_reass_qsize--; 395 q = nq; 396 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 397 ND6_HINT(tp); 398 sorwakeup_locked(so); 399 return (flags); 400 } 401 402 /* 403 * TCP input routine, follows pages 65-76 of the 404 * protocol specification dated September, 1981 very closely. 405 */ 406 #ifdef INET6 407 int 408 tcp6_input(struct mbuf **mp, int *offp, int proto) 409 { 410 struct mbuf *m = *mp; 411 struct in6_ifaddr *ia6; 412 413 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 414 415 /* 416 * draft-itojun-ipv6-tcp-to-anycast 417 * better place to put this in? 418 */ 419 ia6 = ip6_getdstifaddr(m); 420 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 421 struct ip6_hdr *ip6; 422 423 ip6 = mtod(m, struct ip6_hdr *); 424 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 425 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 426 return IPPROTO_DONE; 427 } 428 429 tcp_input(m, *offp); 430 return IPPROTO_DONE; 431 } 432 #endif 433 434 void 435 tcp_input(struct mbuf *m, int off0) 436 { 437 struct tcphdr *th; 438 struct ip *ip = NULL; 439 struct ipovly *ipov; 440 struct inpcb *inp = NULL; 441 struct tcpcb *tp = NULL; 442 struct socket *so = NULL; 443 u_char *optp = NULL; 444 int optlen = 0; 445 int len, tlen, off; 446 int drop_hdrlen; 447 int thflags; 448 int rstreason = 0; /* For badport_bandlim accounting purposes */ 449 #ifdef IPFIREWALL_FORWARD 450 struct m_tag *fwd_tag; 451 #endif 452 #ifdef INET6 453 struct ip6_hdr *ip6 = NULL; 454 int isipv6; 455 char ip6buf[INET6_ADDRSTRLEN]; 456 #else 457 const int isipv6 = 0; 458 #endif 459 struct tcpopt to; /* options in this segment */ 460 461 #ifdef TCPDEBUG 462 /* 463 * The size of tcp_saveipgen must be the size of the max ip header, 464 * now IPv6. 465 */ 466 u_char tcp_saveipgen[IP6_HDR_LEN]; 467 struct tcphdr tcp_savetcp; 468 short ostate = 0; 469 #endif 470 471 #ifdef INET6 472 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 473 #endif 474 475 to.to_flags = 0; 476 tcpstat.tcps_rcvtotal++; 477 478 if (isipv6) { 479 #ifdef INET6 480 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 481 ip6 = mtod(m, struct ip6_hdr *); 482 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 483 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 484 tcpstat.tcps_rcvbadsum++; 485 goto drop; 486 } 487 th = (struct tcphdr *)((caddr_t)ip6 + off0); 488 489 /* 490 * Be proactive about unspecified IPv6 address in source. 491 * As we use all-zero to indicate unbounded/unconnected pcb, 492 * unspecified IPv6 address can be used to confuse us. 493 * 494 * Note that packets with unspecified IPv6 destination is 495 * already dropped in ip6_input. 496 */ 497 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 498 /* XXX stat */ 499 goto drop; 500 } 501 #else 502 th = NULL; /* XXX: avoid compiler warning */ 503 #endif 504 } else { 505 /* 506 * Get IP and TCP header together in first mbuf. 507 * Note: IP leaves IP header in first mbuf. 508 */ 509 if (off0 > sizeof (struct ip)) { 510 ip_stripoptions(m, (struct mbuf *)0); 511 off0 = sizeof(struct ip); 512 } 513 if (m->m_len < sizeof (struct tcpiphdr)) { 514 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 515 == NULL) { 516 tcpstat.tcps_rcvshort++; 517 return; 518 } 519 } 520 ip = mtod(m, struct ip *); 521 ipov = (struct ipovly *)ip; 522 th = (struct tcphdr *)((caddr_t)ip + off0); 523 tlen = ip->ip_len; 524 525 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 526 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 527 th->th_sum = m->m_pkthdr.csum_data; 528 else 529 th->th_sum = in_pseudo(ip->ip_src.s_addr, 530 ip->ip_dst.s_addr, 531 htonl(m->m_pkthdr.csum_data + 532 ip->ip_len + 533 IPPROTO_TCP)); 534 th->th_sum ^= 0xffff; 535 #ifdef TCPDEBUG 536 ipov->ih_len = (u_short)tlen; 537 ipov->ih_len = htons(ipov->ih_len); 538 #endif 539 } else { 540 /* 541 * Checksum extended TCP header and data. 542 */ 543 len = sizeof (struct ip) + tlen; 544 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 545 ipov->ih_len = (u_short)tlen; 546 ipov->ih_len = htons(ipov->ih_len); 547 th->th_sum = in_cksum(m, len); 548 } 549 if (th->th_sum) { 550 tcpstat.tcps_rcvbadsum++; 551 goto drop; 552 } 553 /* Re-initialization for later version check */ 554 ip->ip_v = IPVERSION; 555 } 556 557 /* 558 * Check that TCP offset makes sense, 559 * pull out TCP options and adjust length. XXX 560 */ 561 off = th->th_off << 2; 562 if (off < sizeof (struct tcphdr) || off > tlen) { 563 tcpstat.tcps_rcvbadoff++; 564 goto drop; 565 } 566 tlen -= off; /* tlen is used instead of ti->ti_len */ 567 if (off > sizeof (struct tcphdr)) { 568 if (isipv6) { 569 #ifdef INET6 570 IP6_EXTHDR_CHECK(m, off0, off, ); 571 ip6 = mtod(m, struct ip6_hdr *); 572 th = (struct tcphdr *)((caddr_t)ip6 + off0); 573 #endif 574 } else { 575 if (m->m_len < sizeof(struct ip) + off) { 576 if ((m = m_pullup(m, sizeof (struct ip) + off)) 577 == NULL) { 578 tcpstat.tcps_rcvshort++; 579 return; 580 } 581 ip = mtod(m, struct ip *); 582 ipov = (struct ipovly *)ip; 583 th = (struct tcphdr *)((caddr_t)ip + off0); 584 } 585 } 586 optlen = off - sizeof (struct tcphdr); 587 optp = (u_char *)(th + 1); 588 } 589 thflags = th->th_flags; 590 591 /* 592 * If the drop_synfin option is enabled, drop all packets with 593 * both the SYN and FIN bits set. This prevents e.g. nmap from 594 * identifying the TCP/IP stack. 595 * 596 * This is a violation of the TCP specification. 597 */ 598 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 599 goto drop; 600 601 /* 602 * Convert TCP protocol specific fields to host format. 603 */ 604 th->th_seq = ntohl(th->th_seq); 605 th->th_ack = ntohl(th->th_ack); 606 th->th_win = ntohs(th->th_win); 607 th->th_urp = ntohs(th->th_urp); 608 609 /* 610 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 611 */ 612 drop_hdrlen = off0 + off; 613 614 /* 615 * Locate pcb for segment. 616 */ 617 INP_INFO_WLOCK(&tcbinfo); 618 findpcb: 619 INP_INFO_WLOCK_ASSERT(&tcbinfo); 620 #ifdef IPFIREWALL_FORWARD 621 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 622 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 623 624 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 625 struct sockaddr_in *next_hop; 626 627 next_hop = (struct sockaddr_in *)(fwd_tag+1); 628 /* 629 * Transparently forwarded. Pretend to be the destination. 630 * already got one like this? 631 */ 632 inp = in_pcblookup_hash(&tcbinfo, 633 ip->ip_src, th->th_sport, 634 ip->ip_dst, th->th_dport, 635 0, m->m_pkthdr.rcvif); 636 if (!inp) { 637 /* It's new. Try to find the ambushing socket. */ 638 inp = in_pcblookup_hash(&tcbinfo, 639 ip->ip_src, th->th_sport, 640 next_hop->sin_addr, 641 next_hop->sin_port ? 642 ntohs(next_hop->sin_port) : 643 th->th_dport, 644 INPLOOKUP_WILDCARD, 645 m->m_pkthdr.rcvif); 646 } 647 /* Remove the tag from the packet. We don't need it anymore. */ 648 m_tag_delete(m, fwd_tag); 649 } else 650 #endif /* IPFIREWALL_FORWARD */ 651 { 652 if (isipv6) { 653 #ifdef INET6 654 inp = in6_pcblookup_hash(&tcbinfo, 655 &ip6->ip6_src, th->th_sport, 656 &ip6->ip6_dst, th->th_dport, 657 INPLOOKUP_WILDCARD, 658 m->m_pkthdr.rcvif); 659 #endif 660 } else 661 inp = in_pcblookup_hash(&tcbinfo, 662 ip->ip_src, th->th_sport, 663 ip->ip_dst, th->th_dport, 664 INPLOOKUP_WILDCARD, 665 m->m_pkthdr.rcvif); 666 } 667 668 #if defined(IPSEC) || defined(FAST_IPSEC) 669 #ifdef INET6 670 if (isipv6 && inp != NULL && ipsec6_in_reject(m, inp)) { 671 #ifdef IPSEC 672 ipsec6stat.in_polvio++; 673 #endif 674 goto dropunlock; 675 } else 676 #endif /* INET6 */ 677 if (inp != NULL && ipsec4_in_reject(m, inp)) { 678 #ifdef IPSEC 679 ipsecstat.in_polvio++; 680 #endif 681 goto dropunlock; 682 } 683 #endif /*IPSEC || FAST_IPSEC*/ 684 685 /* 686 * If the INPCB does not exist then all data in the incoming 687 * segment is discarded and an appropriate RST is sent back. 688 */ 689 if (inp == NULL) { 690 /* 691 * Log communication attempts to ports that are not 692 * in use. 693 */ 694 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 695 tcp_log_in_vain == 2) { 696 #ifndef INET6 697 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 698 #else 699 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 700 if (isipv6) { 701 strcpy(dbuf, "["); 702 strcat(dbuf, 703 ip6_sprintf(ip6buf, &ip6->ip6_dst)); 704 strcat(dbuf, "]"); 705 strcpy(sbuf, "["); 706 strcat(sbuf, 707 ip6_sprintf(ip6buf, &ip6->ip6_src)); 708 strcat(sbuf, "]"); 709 } else 710 #endif /* INET6 */ 711 { 712 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 713 strcpy(sbuf, inet_ntoa(ip->ip_src)); 714 } 715 log(LOG_INFO, 716 "Connection attempt to TCP %s:%d " 717 "from %s:%d flags:0x%02x\n", 718 dbuf, ntohs(th->th_dport), sbuf, 719 ntohs(th->th_sport), thflags); 720 } 721 /* 722 * When blackholing do not respond with a RST but 723 * completely ignore the segment and drop it. 724 */ 725 if ((blackhole == 1 && (thflags & TH_SYN)) || 726 blackhole == 2) 727 goto dropunlock; 728 729 rstreason = BANDLIM_RST_CLOSEDPORT; 730 goto dropwithreset; 731 } 732 INP_LOCK(inp); 733 734 /* Check the minimum TTL for socket. */ 735 if (inp->inp_ip_minttl != 0) { 736 #ifdef INET6 737 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 738 goto dropunlock; 739 else 740 #endif 741 if (inp->inp_ip_minttl > ip->ip_ttl) 742 goto dropunlock; 743 } 744 745 /* 746 * A previous connection in TIMEWAIT state is supposed to catch 747 * stray or duplicate segments arriving late. If this segment 748 * was a legitimate new connection attempt the old INPCB gets 749 * removed and we can try again to find a listening socket. 750 */ 751 if (inp->inp_vflag & INP_TIMEWAIT) { 752 if (thflags & TH_SYN) 753 tcp_dooptions(&to, optp, optlen, TO_SYN); 754 if (tcp_timewait(inp, &to, th, m, tlen)) 755 goto findpcb; 756 /* tcp_timewait unlocks inp. */ 757 INP_INFO_WUNLOCK(&tcbinfo); 758 return; 759 } 760 /* 761 * The TCPCB may no longer exist if the connection is winding 762 * down or it is in the CLOSED state. Either way we drop the 763 * segment and send an appropriate response. 764 */ 765 tp = intotcpcb(inp); 766 if (tp == NULL) { 767 INP_UNLOCK(inp); 768 rstreason = BANDLIM_RST_CLOSEDPORT; 769 goto dropwithreset; 770 } 771 if (tp->t_state == TCPS_CLOSED) 772 goto dropunlock; /* XXX: dropwithreset??? */ 773 774 #ifdef MAC 775 INP_LOCK_ASSERT(inp); 776 if (mac_check_inpcb_deliver(inp, m)) 777 goto dropunlock; 778 #endif 779 so = inp->inp_socket; 780 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 781 #ifdef TCPDEBUG 782 if (so->so_options & SO_DEBUG) { 783 ostate = tp->t_state; 784 if (isipv6) 785 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 786 else 787 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 788 tcp_savetcp = *th; 789 } 790 #endif 791 /* 792 * When the socket is accepting connections (the INPCB is in LISTEN 793 * state) we look into the SYN cache if this is a new connection 794 * attempt or the completion of a previous one. 795 */ 796 if (so->so_options & SO_ACCEPTCONN) { 797 struct in_conninfo inc; 798 799 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 800 "tp not listening", __func__)); 801 802 bzero(&inc, sizeof(inc)); 803 inc.inc_isipv6 = isipv6; 804 #ifdef INET6 805 if (isipv6) { 806 inc.inc6_faddr = ip6->ip6_src; 807 inc.inc6_laddr = ip6->ip6_dst; 808 } else 809 #endif 810 { 811 inc.inc_faddr = ip->ip_src; 812 inc.inc_laddr = ip->ip_dst; 813 } 814 inc.inc_fport = th->th_sport; 815 inc.inc_lport = th->th_dport; 816 817 /* 818 * If the state is LISTEN then ignore segment if it contains 819 * a RST. If the segment contains an ACK then it is bad and 820 * send a RST. If it does not contain a SYN then it is not 821 * interesting; drop it. 822 * 823 * If the state is SYN_RECEIVED (syncache) and seg contains 824 * an ACK, but not for our SYN/ACK, send a RST. If the seg 825 * contains a RST, check the sequence number to see if it 826 * is a valid reset segment. 827 */ 828 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 829 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 830 /* 831 * Parse the TCP options here because 832 * syncookies need access to the reflected 833 * timestamp. 834 */ 835 tcp_dooptions(&to, optp, optlen, 0); 836 if (!syncache_expand(&inc, &to, th, &so, m)) { 837 /* 838 * No syncache entry or ACK was not 839 * for our SYN/ACK. Send a RST. 840 */ 841 rstreason = BANDLIM_RST_OPENPORT; 842 goto dropwithreset; 843 } 844 if (so == NULL) { 845 /* 846 * We completed the 3-way handshake 847 * but could not allocate a socket 848 * either due to memory shortage, 849 * listen queue length limits or 850 * global socket limits. 851 */ 852 rstreason = BANDLIM_UNLIMITED; 853 goto dropwithreset; 854 } 855 /* 856 * Socket is created in state SYN_RECEIVED. 857 * Continue processing segment. 858 */ 859 INP_UNLOCK(inp); /* listen socket */ 860 inp = sotoinpcb(so); 861 INP_LOCK(inp); /* new connection */ 862 tp = intotcpcb(inp); 863 /* 864 * Process the segment and the data it 865 * contains. tcp_do_segment() consumes 866 * the mbuf chain and unlocks the inpcb. 867 * XXX: The potential return value of 868 * TIME_WAIT nuked is supposed to be 869 * handled above. 870 */ 871 if (tcp_do_segment(m, th, so, tp, 872 drop_hdrlen, tlen)) 873 goto findpcb; /* TIME_WAIT nuked */ 874 return; 875 } 876 if (thflags & TH_RST) { 877 syncache_chkrst(&inc, th); 878 goto dropunlock; 879 } 880 if (thflags & TH_ACK) { 881 syncache_badack(&inc); 882 tcpstat.tcps_badsyn++; 883 rstreason = BANDLIM_RST_OPENPORT; 884 goto dropwithreset; 885 } 886 goto dropunlock; 887 } 888 889 /* 890 * Segment's flags are (SYN) or (SYN|FIN). 891 */ 892 #ifdef INET6 893 /* 894 * If deprecated address is forbidden, 895 * we do not accept SYN to deprecated interface 896 * address to prevent any new inbound connection from 897 * getting established. 898 * When we do not accept SYN, we send a TCP RST, 899 * with deprecated source address (instead of dropping 900 * it). We compromise it as it is much better for peer 901 * to send a RST, and RST will be the final packet 902 * for the exchange. 903 * 904 * If we do not forbid deprecated addresses, we accept 905 * the SYN packet. RFC2462 does not suggest dropping 906 * SYN in this case. 907 * If we decipher RFC2462 5.5.4, it says like this: 908 * 1. use of deprecated addr with existing 909 * communication is okay - "SHOULD continue to be 910 * used" 911 * 2. use of it with new communication: 912 * (2a) "SHOULD NOT be used if alternate address 913 * with sufficient scope is available" 914 * (2b) nothing mentioned otherwise. 915 * Here we fall into (2b) case as we have no choice in 916 * our source address selection - we must obey the peer. 917 * 918 * The wording in RFC2462 is confusing, and there are 919 * multiple description text for deprecated address 920 * handling - worse, they are not exactly the same. 921 * I believe 5.5.4 is the best one, so we follow 5.5.4. 922 */ 923 if (isipv6 && !ip6_use_deprecated) { 924 struct in6_ifaddr *ia6; 925 926 if ((ia6 = ip6_getdstifaddr(m)) && 927 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 928 INP_UNLOCK(inp); 929 tp = NULL; 930 rstreason = BANDLIM_RST_OPENPORT; 931 goto dropwithreset; 932 } 933 } 934 #endif 935 /* 936 * Basic sanity checks on incoming SYN requests: 937 * 938 * Don't bother responding if the destination was a 939 * broadcast according to RFC1122 4.2.3.10, p. 104. 940 * 941 * If it is from this socket, drop it, it must be forged. 942 * 943 * Note that it is quite possible to receive unicast 944 * link-layer packets with a broadcast IP address. Use 945 * in_broadcast() to find them. 946 */ 947 if (m->m_flags & (M_BCAST|M_MCAST)) 948 goto dropunlock; 949 if (isipv6) { 950 #ifdef INET6 951 if (th->th_dport == th->th_sport && 952 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) 953 goto dropunlock; 954 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 955 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 956 goto dropunlock; 957 #endif 958 } else { 959 if (th->th_dport == th->th_sport && 960 ip->ip_dst.s_addr == ip->ip_src.s_addr) 961 goto dropunlock; 962 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 963 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 964 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 965 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 966 goto dropunlock; 967 } 968 /* 969 * SYN appears to be valid. Create compressed TCP state 970 * for syncache. 971 */ 972 #ifdef TCPDEBUG 973 if (so->so_options & SO_DEBUG) 974 tcp_trace(TA_INPUT, ostate, tp, 975 (void *)tcp_saveipgen, &tcp_savetcp, 0); 976 #endif 977 tcp_dooptions(&to, optp, optlen, TO_SYN); 978 syncache_add(&inc, &to, th, inp, &so, m); 979 /* 980 * Entry added to syncache and mbuf consumed. 981 * Everything unlocked already by syncache_add(). 982 */ 983 return; 984 } 985 986 /* 987 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or 988 * later state. tcp_do_segment() always consumes the mbuf chain 989 * and unlocks the inpcb. 990 */ 991 if (tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen)) 992 goto findpcb; /* XXX: TIME_WAIT was nuked. */ 993 return; 994 995 dropwithreset: 996 INP_INFO_WLOCK_ASSERT(&tcbinfo); 997 tcp_dropwithreset(m, th, tp, tlen, rstreason); 998 m = NULL; /* mbuf chain got consumed. */ 999 dropunlock: 1000 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1001 if (tp != NULL) 1002 INP_UNLOCK(inp); 1003 INP_INFO_WUNLOCK(&tcbinfo); 1004 drop: 1005 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 1006 if (m != NULL) 1007 m_freem(m); 1008 return; 1009 } 1010 1011 static int 1012 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1013 struct tcpcb *tp, int drop_hdrlen, int tlen) 1014 { 1015 int thflags, acked, ourfinisacked, needoutput = 0; 1016 int headlocked = 1; 1017 int rstreason, todrop, win; 1018 u_long tiwin; 1019 struct tcpopt to; 1020 1021 #ifdef TCPDEBUG 1022 /* 1023 * The size of tcp_saveipgen must be the size of the max ip header, 1024 * now IPv6. 1025 */ 1026 u_char tcp_saveipgen[IP6_HDR_LEN]; 1027 struct tcphdr tcp_savetcp; 1028 short ostate = 0; 1029 #endif 1030 thflags = th->th_flags; 1031 1032 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1033 INP_LOCK_ASSERT(tp->t_inpcb); 1034 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", __func__)); 1035 1036 /* 1037 * Segment received on connection. 1038 * Reset idle time and keep-alive timer. 1039 */ 1040 tp->t_rcvtime = ticks; 1041 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1042 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1043 1044 /* 1045 * Unscale the window into a 32-bit value. 1046 * This value is bogus for the TCPS_SYN_SENT state 1047 * and is overwritten later. 1048 */ 1049 tiwin = th->th_win << tp->snd_scale; 1050 1051 /* 1052 * Parse options on any incoming segment. 1053 */ 1054 tcp_dooptions(&to, (u_char *)(th + 1), 1055 (th->th_off << 2) - sizeof(struct tcphdr), 1056 (thflags & TH_SYN) ? TO_SYN : 0); 1057 1058 /* 1059 * If echoed timestamp is later than the current time, 1060 * fall back to non RFC1323 RTT calculation. Normalize 1061 * timestamp if syncookies were used when this connection 1062 * was established. 1063 */ 1064 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1065 to.to_tsecr -= tp->ts_offset; 1066 if (TSTMP_GT(to.to_tsecr, ticks)) 1067 to.to_tsecr = 0; 1068 } 1069 1070 /* 1071 * Process options only when we get SYN/ACK back. The SYN case 1072 * for incoming connections is handled in tcp_syncache. 1073 * XXX this is traditional behavior, may need to be cleaned up. 1074 */ 1075 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1076 if ((to.to_flags & TOF_SCALE) && 1077 (tp->t_flags & TF_REQ_SCALE)) { 1078 tp->t_flags |= TF_RCVD_SCALE; 1079 tp->snd_scale = to.to_wscale; 1080 tp->snd_wnd = th->th_win << tp->snd_scale; 1081 tiwin = tp->snd_wnd; 1082 } 1083 if (to.to_flags & TOF_TS) { 1084 tp->t_flags |= TF_RCVD_TSTMP; 1085 tp->ts_recent = to.to_tsval; 1086 tp->ts_recent_age = ticks; 1087 } 1088 /* Initial send window, already scaled. */ 1089 tp->snd_wnd = th->th_win; 1090 if (to.to_flags & TOF_MSS) 1091 tcp_mss(tp, to.to_mss); 1092 if (tp->sack_enable) { 1093 if (!(to.to_flags & TOF_SACKPERM)) 1094 tp->sack_enable = 0; 1095 else 1096 tp->t_flags |= TF_SACK_PERMIT; 1097 } 1098 1099 } 1100 1101 /* 1102 * Header prediction: check for the two common cases 1103 * of a uni-directional data xfer. If the packet has 1104 * no control flags, is in-sequence, the window didn't 1105 * change and we're not retransmitting, it's a 1106 * candidate. If the length is zero and the ack moved 1107 * forward, we're the sender side of the xfer. Just 1108 * free the data acked & wake any higher level process 1109 * that was blocked waiting for space. If the length 1110 * is non-zero and the ack didn't move, we're the 1111 * receiver side. If we're getting packets in-order 1112 * (the reassembly queue is empty), add the data to 1113 * the socket buffer and note that we need a delayed ack. 1114 * Make sure that the hidden state-flags are also off. 1115 * Since we check for TCPS_ESTABLISHED above, it can only 1116 * be TH_NEEDSYN. 1117 */ 1118 if (tp->t_state == TCPS_ESTABLISHED && 1119 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1120 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1121 ((to.to_flags & TOF_TS) == 0 || 1122 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1123 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1124 tp->snd_nxt == tp->snd_max) { 1125 1126 /* 1127 * If last ACK falls within this segment's sequence numbers, 1128 * record the timestamp. 1129 * NOTE that the test is modified according to the latest 1130 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1131 */ 1132 if ((to.to_flags & TOF_TS) != 0 && 1133 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1134 tp->ts_recent_age = ticks; 1135 tp->ts_recent = to.to_tsval; 1136 } 1137 1138 if (tlen == 0) { 1139 if (SEQ_GT(th->th_ack, tp->snd_una) && 1140 SEQ_LEQ(th->th_ack, tp->snd_max) && 1141 tp->snd_cwnd >= tp->snd_wnd && 1142 ((!tcp_do_newreno && !tp->sack_enable && 1143 tp->t_dupacks < tcprexmtthresh) || 1144 ((tcp_do_newreno || tp->sack_enable) && 1145 !IN_FASTRECOVERY(tp) && 1146 (to.to_flags & TOF_SACK) == 0 && 1147 TAILQ_EMPTY(&tp->snd_holes)))) { 1148 KASSERT(headlocked, 1149 ("%s: headlocked", __func__)); 1150 INP_INFO_WUNLOCK(&tcbinfo); 1151 headlocked = 0; 1152 /* 1153 * this is a pure ack for outstanding data. 1154 */ 1155 ++tcpstat.tcps_predack; 1156 /* 1157 * "bad retransmit" recovery 1158 */ 1159 if (tp->t_rxtshift == 1 && 1160 ticks < tp->t_badrxtwin) { 1161 ++tcpstat.tcps_sndrexmitbad; 1162 tp->snd_cwnd = tp->snd_cwnd_prev; 1163 tp->snd_ssthresh = 1164 tp->snd_ssthresh_prev; 1165 tp->snd_recover = tp->snd_recover_prev; 1166 if (tp->t_flags & TF_WASFRECOVERY) 1167 ENTER_FASTRECOVERY(tp); 1168 tp->snd_nxt = tp->snd_max; 1169 tp->t_badrxtwin = 0; 1170 } 1171 1172 /* 1173 * Recalculate the transmit timer / rtt. 1174 * 1175 * Some boxes send broken timestamp replies 1176 * during the SYN+ACK phase, ignore 1177 * timestamps of 0 or we could calculate a 1178 * huge RTT and blow up the retransmit timer. 1179 */ 1180 if ((to.to_flags & TOF_TS) != 0 && 1181 to.to_tsecr) { 1182 if (!tp->t_rttlow || 1183 tp->t_rttlow > ticks - to.to_tsecr) 1184 tp->t_rttlow = ticks - to.to_tsecr; 1185 tcp_xmit_timer(tp, 1186 ticks - to.to_tsecr + 1); 1187 } else if (tp->t_rtttime && 1188 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1189 if (!tp->t_rttlow || 1190 tp->t_rttlow > ticks - tp->t_rtttime) 1191 tp->t_rttlow = ticks - tp->t_rtttime; 1192 tcp_xmit_timer(tp, 1193 ticks - tp->t_rtttime); 1194 } 1195 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1196 acked = th->th_ack - tp->snd_una; 1197 tcpstat.tcps_rcvackpack++; 1198 tcpstat.tcps_rcvackbyte += acked; 1199 sbdrop(&so->so_snd, acked); 1200 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1201 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1202 tp->snd_recover = th->th_ack - 1; 1203 tp->snd_una = th->th_ack; 1204 /* 1205 * pull snd_wl2 up to prevent seq wrap relative 1206 * to th_ack. 1207 */ 1208 tp->snd_wl2 = th->th_ack; 1209 tp->t_dupacks = 0; 1210 m_freem(m); 1211 ND6_HINT(tp); /* some progress has been done */ 1212 1213 /* 1214 * If all outstanding data are acked, stop 1215 * retransmit timer, otherwise restart timer 1216 * using current (possibly backed-off) value. 1217 * If process is waiting for space, 1218 * wakeup/selwakeup/signal. If data 1219 * are ready to send, let tcp_output 1220 * decide between more output or persist. 1221 1222 #ifdef TCPDEBUG 1223 if (so->so_options & SO_DEBUG) 1224 tcp_trace(TA_INPUT, ostate, tp, 1225 (void *)tcp_saveipgen, 1226 &tcp_savetcp, 0); 1227 #endif 1228 */ 1229 if (tp->snd_una == tp->snd_max) 1230 tcp_timer_activate(tp, TT_REXMT, 0); 1231 else if (!tcp_timer_active(tp, TT_PERSIST)) 1232 tcp_timer_activate(tp, TT_REXMT, 1233 tp->t_rxtcur); 1234 1235 sowwakeup(so); 1236 if (so->so_snd.sb_cc) 1237 (void) tcp_output(tp); 1238 goto check_delack; 1239 } 1240 } else if (th->th_ack == tp->snd_una && 1241 LIST_EMPTY(&tp->t_segq) && 1242 tlen <= sbspace(&so->so_rcv)) { 1243 int newsize = 0; /* automatic sockbuf scaling */ 1244 1245 KASSERT(headlocked, ("%s: headlocked", __func__)); 1246 INP_INFO_WUNLOCK(&tcbinfo); 1247 headlocked = 0; 1248 /* 1249 * this is a pure, in-sequence data packet 1250 * with nothing on the reassembly queue and 1251 * we have enough buffer space to take it. 1252 */ 1253 /* Clean receiver SACK report if present */ 1254 if (tp->sack_enable && tp->rcv_numsacks) 1255 tcp_clean_sackreport(tp); 1256 ++tcpstat.tcps_preddat; 1257 tp->rcv_nxt += tlen; 1258 /* 1259 * Pull snd_wl1 up to prevent seq wrap relative to 1260 * th_seq. 1261 */ 1262 tp->snd_wl1 = th->th_seq; 1263 /* 1264 * Pull rcv_up up to prevent seq wrap relative to 1265 * rcv_nxt. 1266 */ 1267 tp->rcv_up = tp->rcv_nxt; 1268 tcpstat.tcps_rcvpack++; 1269 tcpstat.tcps_rcvbyte += tlen; 1270 ND6_HINT(tp); /* some progress has been done */ 1271 #ifdef TCPDEBUG 1272 if (so->so_options & SO_DEBUG) 1273 tcp_trace(TA_INPUT, ostate, tp, 1274 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1275 #endif 1276 /* 1277 * Automatic sizing of receive socket buffer. Often the send 1278 * buffer size is not optimally adjusted to the actual network 1279 * conditions at hand (delay bandwidth product). Setting the 1280 * buffer size too small limits throughput on links with high 1281 * bandwidth and high delay (eg. trans-continental/oceanic links). 1282 * 1283 * On the receive side the socket buffer memory is only rarely 1284 * used to any significant extent. This allows us to be much 1285 * more aggressive in scaling the receive socket buffer. For 1286 * the case that the buffer space is actually used to a large 1287 * extent and we run out of kernel memory we can simply drop 1288 * the new segments; TCP on the sender will just retransmit it 1289 * later. Setting the buffer size too big may only consume too 1290 * much kernel memory if the application doesn't read() from 1291 * the socket or packet loss or reordering makes use of the 1292 * reassembly queue. 1293 * 1294 * The criteria to step up the receive buffer one notch are: 1295 * 1. the number of bytes received during the time it takes 1296 * one timestamp to be reflected back to us (the RTT); 1297 * 2. received bytes per RTT is within seven eighth of the 1298 * current socket buffer size; 1299 * 3. receive buffer size has not hit maximal automatic size; 1300 * 1301 * This algorithm does one step per RTT at most and only if 1302 * we receive a bulk stream w/o packet losses or reorderings. 1303 * Shrinking the buffer during idle times is not necessary as 1304 * it doesn't consume any memory when idle. 1305 * 1306 * TODO: Only step up if the application is actually serving 1307 * the buffer to better manage the socket buffer resources. 1308 */ 1309 if (tcp_do_autorcvbuf && 1310 to.to_tsecr && 1311 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1312 if (to.to_tsecr > tp->rfbuf_ts && 1313 to.to_tsecr - tp->rfbuf_ts < hz) { 1314 if (tp->rfbuf_cnt > 1315 (so->so_rcv.sb_hiwat / 8 * 7) && 1316 so->so_rcv.sb_hiwat < 1317 tcp_autorcvbuf_max) { 1318 newsize = 1319 min(so->so_rcv.sb_hiwat + 1320 tcp_autorcvbuf_inc, 1321 tcp_autorcvbuf_max); 1322 } 1323 /* Start over with next RTT. */ 1324 tp->rfbuf_ts = 0; 1325 tp->rfbuf_cnt = 0; 1326 } else 1327 tp->rfbuf_cnt += tlen; /* add up */ 1328 } 1329 1330 /* Add data to socket buffer. */ 1331 SOCKBUF_LOCK(&so->so_rcv); 1332 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1333 m_freem(m); 1334 } else { 1335 /* 1336 * Set new socket buffer size. 1337 * Give up when limit is reached. 1338 */ 1339 if (newsize) 1340 if (!sbreserve_locked(&so->so_rcv, 1341 newsize, so, curthread)) 1342 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1343 m_adj(m, drop_hdrlen); /* delayed header drop */ 1344 sbappendstream_locked(&so->so_rcv, m); 1345 } 1346 sorwakeup_locked(so); 1347 if (DELAY_ACK(tp)) { 1348 tp->t_flags |= TF_DELACK; 1349 } else { 1350 tp->t_flags |= TF_ACKNOW; 1351 tcp_output(tp); 1352 } 1353 goto check_delack; 1354 } 1355 } 1356 1357 /* 1358 * Calculate amount of space in receive window, 1359 * and then do TCP input processing. 1360 * Receive window is amount of space in rcv queue, 1361 * but not less than advertised window. 1362 */ 1363 win = sbspace(&so->so_rcv); 1364 if (win < 0) 1365 win = 0; 1366 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1367 1368 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1369 tp->rfbuf_ts = 0; 1370 tp->rfbuf_cnt = 0; 1371 1372 switch (tp->t_state) { 1373 1374 /* 1375 * If the state is SYN_RECEIVED: 1376 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1377 */ 1378 case TCPS_SYN_RECEIVED: 1379 if ((thflags & TH_ACK) && 1380 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1381 SEQ_GT(th->th_ack, tp->snd_max))) { 1382 rstreason = BANDLIM_RST_OPENPORT; 1383 goto dropwithreset; 1384 } 1385 break; 1386 1387 /* 1388 * If the state is SYN_SENT: 1389 * if seg contains an ACK, but not for our SYN, drop the input. 1390 * if seg contains a RST, then drop the connection. 1391 * if seg does not contain SYN, then drop it. 1392 * Otherwise this is an acceptable SYN segment 1393 * initialize tp->rcv_nxt and tp->irs 1394 * if seg contains ack then advance tp->snd_una 1395 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1396 * arrange for segment to be acked (eventually) 1397 * continue processing rest of data/controls, beginning with URG 1398 */ 1399 case TCPS_SYN_SENT: 1400 if ((thflags & TH_ACK) && 1401 (SEQ_LEQ(th->th_ack, tp->iss) || 1402 SEQ_GT(th->th_ack, tp->snd_max))) { 1403 rstreason = BANDLIM_UNLIMITED; 1404 goto dropwithreset; 1405 } 1406 if (thflags & TH_RST) { 1407 if (thflags & TH_ACK) { 1408 KASSERT(headlocked, ("%s: after_listen: " 1409 "tcp_drop.2: head not locked", __func__)); 1410 tp = tcp_drop(tp, ECONNREFUSED); 1411 } 1412 goto drop; 1413 } 1414 if ((thflags & TH_SYN) == 0) 1415 goto drop; 1416 1417 tp->irs = th->th_seq; 1418 tcp_rcvseqinit(tp); 1419 if (thflags & TH_ACK) { 1420 tcpstat.tcps_connects++; 1421 soisconnected(so); 1422 #ifdef MAC 1423 SOCK_LOCK(so); 1424 mac_set_socket_peer_from_mbuf(m, so); 1425 SOCK_UNLOCK(so); 1426 #endif 1427 /* Do window scaling on this connection? */ 1428 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1429 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1430 tp->rcv_scale = tp->request_r_scale; 1431 } 1432 tp->rcv_adv += tp->rcv_wnd; 1433 tp->snd_una++; /* SYN is acked */ 1434 /* 1435 * If there's data, delay ACK; if there's also a FIN 1436 * ACKNOW will be turned on later. 1437 */ 1438 if (DELAY_ACK(tp) && tlen != 0) 1439 tcp_timer_activate(tp, TT_DELACK, 1440 tcp_delacktime); 1441 else 1442 tp->t_flags |= TF_ACKNOW; 1443 /* 1444 * Received <SYN,ACK> in SYN_SENT[*] state. 1445 * Transitions: 1446 * SYN_SENT --> ESTABLISHED 1447 * SYN_SENT* --> FIN_WAIT_1 1448 */ 1449 tp->t_starttime = ticks; 1450 if (tp->t_flags & TF_NEEDFIN) { 1451 tp->t_state = TCPS_FIN_WAIT_1; 1452 tp->t_flags &= ~TF_NEEDFIN; 1453 thflags &= ~TH_SYN; 1454 } else { 1455 tp->t_state = TCPS_ESTABLISHED; 1456 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1457 } 1458 } else { 1459 /* 1460 * Received initial SYN in SYN-SENT[*] state => 1461 * simultaneous open. If segment contains CC option 1462 * and there is a cached CC, apply TAO test. 1463 * If it succeeds, connection is * half-synchronized. 1464 * Otherwise, do 3-way handshake: 1465 * SYN-SENT -> SYN-RECEIVED 1466 * SYN-SENT* -> SYN-RECEIVED* 1467 * If there was no CC option, clear cached CC value. 1468 */ 1469 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1470 tcp_timer_activate(tp, TT_REXMT, 0); 1471 tp->t_state = TCPS_SYN_RECEIVED; 1472 } 1473 1474 KASSERT(headlocked, ("%s: trimthenstep6: head not locked", 1475 __func__)); 1476 INP_LOCK_ASSERT(tp->t_inpcb); 1477 1478 /* 1479 * Advance th->th_seq to correspond to first data byte. 1480 * If data, trim to stay within window, 1481 * dropping FIN if necessary. 1482 */ 1483 th->th_seq++; 1484 if (tlen > tp->rcv_wnd) { 1485 todrop = tlen - tp->rcv_wnd; 1486 m_adj(m, -todrop); 1487 tlen = tp->rcv_wnd; 1488 thflags &= ~TH_FIN; 1489 tcpstat.tcps_rcvpackafterwin++; 1490 tcpstat.tcps_rcvbyteafterwin += todrop; 1491 } 1492 tp->snd_wl1 = th->th_seq - 1; 1493 tp->rcv_up = th->th_seq; 1494 /* 1495 * Client side of transaction: already sent SYN and data. 1496 * If the remote host used T/TCP to validate the SYN, 1497 * our data will be ACK'd; if so, enter normal data segment 1498 * processing in the middle of step 5, ack processing. 1499 * Otherwise, goto step 6. 1500 */ 1501 if (thflags & TH_ACK) 1502 goto process_ACK; 1503 1504 goto step6; 1505 1506 /* 1507 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1508 * do normal processing. 1509 * 1510 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1511 */ 1512 case TCPS_LAST_ACK: 1513 case TCPS_CLOSING: 1514 case TCPS_TIME_WAIT: 1515 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: timewait", 1516 __func__)); 1517 break; /* continue normal processing */ 1518 } 1519 1520 /* 1521 * States other than LISTEN or SYN_SENT. 1522 * First check the RST flag and sequence number since reset segments 1523 * are exempt from the timestamp and connection count tests. This 1524 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1525 * below which allowed reset segments in half the sequence space 1526 * to fall though and be processed (which gives forged reset 1527 * segments with a random sequence number a 50 percent chance of 1528 * killing a connection). 1529 * Then check timestamp, if present. 1530 * Then check the connection count, if present. 1531 * Then check that at least some bytes of segment are within 1532 * receive window. If segment begins before rcv_nxt, 1533 * drop leading data (and SYN); if nothing left, just ack. 1534 * 1535 * 1536 * If the RST bit is set, check the sequence number to see 1537 * if this is a valid reset segment. 1538 * RFC 793 page 37: 1539 * In all states except SYN-SENT, all reset (RST) segments 1540 * are validated by checking their SEQ-fields. A reset is 1541 * valid if its sequence number is in the window. 1542 * Note: this does not take into account delayed ACKs, so 1543 * we should test against last_ack_sent instead of rcv_nxt. 1544 * The sequence number in the reset segment is normally an 1545 * echo of our outgoing acknowlegement numbers, but some hosts 1546 * send a reset with the sequence number at the rightmost edge 1547 * of our receive window, and we have to handle this case. 1548 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1549 * that brute force RST attacks are possible. To combat this, 1550 * we use a much stricter check while in the ESTABLISHED state, 1551 * only accepting RSTs where the sequence number is equal to 1552 * last_ack_sent. In all other states (the states in which a 1553 * RST is more likely), the more permissive check is used. 1554 * If we have multiple segments in flight, the intial reset 1555 * segment sequence numbers will be to the left of last_ack_sent, 1556 * but they will eventually catch up. 1557 * In any case, it never made sense to trim reset segments to 1558 * fit the receive window since RFC 1122 says: 1559 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1560 * 1561 * A TCP SHOULD allow a received RST segment to include data. 1562 * 1563 * DISCUSSION 1564 * It has been suggested that a RST segment could contain 1565 * ASCII text that encoded and explained the cause of the 1566 * RST. No standard has yet been established for such 1567 * data. 1568 * 1569 * If the reset segment passes the sequence number test examine 1570 * the state: 1571 * SYN_RECEIVED STATE: 1572 * If passive open, return to LISTEN state. 1573 * If active open, inform user that connection was refused. 1574 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1575 * Inform user that connection was reset, and close tcb. 1576 * CLOSING, LAST_ACK STATES: 1577 * Close the tcb. 1578 * TIME_WAIT STATE: 1579 * Drop the segment - see Stevens, vol. 2, p. 964 and 1580 * RFC 1337. 1581 */ 1582 if (thflags & TH_RST) { 1583 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1584 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1585 switch (tp->t_state) { 1586 1587 case TCPS_SYN_RECEIVED: 1588 so->so_error = ECONNREFUSED; 1589 goto close; 1590 1591 case TCPS_ESTABLISHED: 1592 if (tcp_insecure_rst == 0 && 1593 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 1594 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 1595 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1596 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 1597 tcpstat.tcps_badrst++; 1598 goto drop; 1599 } 1600 case TCPS_FIN_WAIT_1: 1601 case TCPS_FIN_WAIT_2: 1602 case TCPS_CLOSE_WAIT: 1603 so->so_error = ECONNRESET; 1604 close: 1605 tp->t_state = TCPS_CLOSED; 1606 tcpstat.tcps_drops++; 1607 KASSERT(headlocked, ("%s: trimthenstep6: " 1608 "tcp_close: head not locked", __func__)); 1609 tp = tcp_close(tp); 1610 break; 1611 1612 case TCPS_CLOSING: 1613 case TCPS_LAST_ACK: 1614 KASSERT(headlocked, ("%s: trimthenstep6: " 1615 "tcp_close.2: head not locked", __func__)); 1616 tp = tcp_close(tp); 1617 break; 1618 1619 case TCPS_TIME_WAIT: 1620 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1621 ("%s: timewait", __func__)); 1622 break; 1623 } 1624 } 1625 goto drop; 1626 } 1627 1628 /* 1629 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1630 * and it's less than ts_recent, drop it. 1631 */ 1632 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1633 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1634 1635 /* Check to see if ts_recent is over 24 days old. */ 1636 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1637 /* 1638 * Invalidate ts_recent. If this segment updates 1639 * ts_recent, the age will be reset later and ts_recent 1640 * will get a valid value. If it does not, setting 1641 * ts_recent to zero will at least satisfy the 1642 * requirement that zero be placed in the timestamp 1643 * echo reply when ts_recent isn't valid. The 1644 * age isn't reset until we get a valid ts_recent 1645 * because we don't want out-of-order segments to be 1646 * dropped when ts_recent is old. 1647 */ 1648 tp->ts_recent = 0; 1649 } else { 1650 tcpstat.tcps_rcvduppack++; 1651 tcpstat.tcps_rcvdupbyte += tlen; 1652 tcpstat.tcps_pawsdrop++; 1653 if (tlen) 1654 goto dropafterack; 1655 goto drop; 1656 } 1657 } 1658 1659 /* 1660 * In the SYN-RECEIVED state, validate that the packet belongs to 1661 * this connection before trimming the data to fit the receive 1662 * window. Check the sequence number versus IRS since we know 1663 * the sequence numbers haven't wrapped. This is a partial fix 1664 * for the "LAND" DoS attack. 1665 */ 1666 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1667 rstreason = BANDLIM_RST_OPENPORT; 1668 goto dropwithreset; 1669 } 1670 1671 todrop = tp->rcv_nxt - th->th_seq; 1672 if (todrop > 0) { 1673 if (thflags & TH_SYN) { 1674 thflags &= ~TH_SYN; 1675 th->th_seq++; 1676 if (th->th_urp > 1) 1677 th->th_urp--; 1678 else 1679 thflags &= ~TH_URG; 1680 todrop--; 1681 } 1682 /* 1683 * Following if statement from Stevens, vol. 2, p. 960. 1684 */ 1685 if (todrop > tlen 1686 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1687 /* 1688 * Any valid FIN must be to the left of the window. 1689 * At this point the FIN must be a duplicate or out 1690 * of sequence; drop it. 1691 */ 1692 thflags &= ~TH_FIN; 1693 1694 /* 1695 * Send an ACK to resynchronize and drop any data. 1696 * But keep on processing for RST or ACK. 1697 */ 1698 tp->t_flags |= TF_ACKNOW; 1699 todrop = tlen; 1700 tcpstat.tcps_rcvduppack++; 1701 tcpstat.tcps_rcvdupbyte += todrop; 1702 } else { 1703 tcpstat.tcps_rcvpartduppack++; 1704 tcpstat.tcps_rcvpartdupbyte += todrop; 1705 } 1706 drop_hdrlen += todrop; /* drop from the top afterwards */ 1707 th->th_seq += todrop; 1708 tlen -= todrop; 1709 if (th->th_urp > todrop) 1710 th->th_urp -= todrop; 1711 else { 1712 thflags &= ~TH_URG; 1713 th->th_urp = 0; 1714 } 1715 } 1716 1717 /* 1718 * If new data are received on a connection after the 1719 * user processes are gone, then RST the other end. 1720 */ 1721 if ((so->so_state & SS_NOFDREF) && 1722 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1723 KASSERT(headlocked, ("%s: trimthenstep6: tcp_close.3: head " 1724 "not locked", __func__)); 1725 tp = tcp_close(tp); 1726 tcpstat.tcps_rcvafterclose++; 1727 rstreason = BANDLIM_UNLIMITED; 1728 goto dropwithreset; 1729 } 1730 1731 /* 1732 * If segment ends after window, drop trailing data 1733 * (and PUSH and FIN); if nothing left, just ACK. 1734 */ 1735 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1736 if (todrop > 0) { 1737 tcpstat.tcps_rcvpackafterwin++; 1738 if (todrop >= tlen) { 1739 tcpstat.tcps_rcvbyteafterwin += tlen; 1740 /* 1741 * If a new connection request is received 1742 * while in TIME_WAIT, drop the old connection 1743 * and start over if the sequence numbers 1744 * are above the previous ones. 1745 */ 1746 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: timewait", 1747 __func__)); 1748 if (thflags & TH_SYN && 1749 tp->t_state == TCPS_TIME_WAIT && 1750 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1751 KASSERT(headlocked, ("%s: trimthenstep6: " 1752 "tcp_close.4: head not locked", __func__)); 1753 tp = tcp_close(tp); 1754 /* XXX: Shouldn't be possible. */ 1755 return (1); 1756 } 1757 /* 1758 * If window is closed can only take segments at 1759 * window edge, and have to drop data and PUSH from 1760 * incoming segments. Continue processing, but 1761 * remember to ack. Otherwise, drop segment 1762 * and ack. 1763 */ 1764 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1765 tp->t_flags |= TF_ACKNOW; 1766 tcpstat.tcps_rcvwinprobe++; 1767 } else 1768 goto dropafterack; 1769 } else 1770 tcpstat.tcps_rcvbyteafterwin += todrop; 1771 m_adj(m, -todrop); 1772 tlen -= todrop; 1773 thflags &= ~(TH_PUSH|TH_FIN); 1774 } 1775 1776 /* 1777 * If last ACK falls within this segment's sequence numbers, 1778 * record its timestamp. 1779 * NOTE: 1780 * 1) That the test incorporates suggestions from the latest 1781 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1782 * 2) That updating only on newer timestamps interferes with 1783 * our earlier PAWS tests, so this check should be solely 1784 * predicated on the sequence space of this segment. 1785 * 3) That we modify the segment boundary check to be 1786 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1787 * instead of RFC1323's 1788 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1789 * This modified check allows us to overcome RFC1323's 1790 * limitations as described in Stevens TCP/IP Illustrated 1791 * Vol. 2 p.869. In such cases, we can still calculate the 1792 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1793 */ 1794 if ((to.to_flags & TOF_TS) != 0 && 1795 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1796 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1797 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1798 tp->ts_recent_age = ticks; 1799 tp->ts_recent = to.to_tsval; 1800 } 1801 1802 /* 1803 * If a SYN is in the window, then this is an 1804 * error and we send an RST and drop the connection. 1805 */ 1806 if (thflags & TH_SYN) { 1807 KASSERT(headlocked, ("%s: tcp_drop: trimthenstep6: " 1808 "head not locked", __func__)); 1809 tp = tcp_drop(tp, ECONNRESET); 1810 rstreason = BANDLIM_UNLIMITED; 1811 goto drop; 1812 } 1813 1814 /* 1815 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1816 * flag is on (half-synchronized state), then queue data for 1817 * later processing; else drop segment and return. 1818 */ 1819 if ((thflags & TH_ACK) == 0) { 1820 if (tp->t_state == TCPS_SYN_RECEIVED || 1821 (tp->t_flags & TF_NEEDSYN)) 1822 goto step6; 1823 else if (tp->t_flags & TF_ACKNOW) 1824 goto dropafterack; 1825 else 1826 goto drop; 1827 } 1828 1829 /* 1830 * Ack processing. 1831 */ 1832 switch (tp->t_state) { 1833 1834 /* 1835 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1836 * ESTABLISHED state and continue processing. 1837 * The ACK was checked above. 1838 */ 1839 case TCPS_SYN_RECEIVED: 1840 1841 tcpstat.tcps_connects++; 1842 soisconnected(so); 1843 /* Do window scaling? */ 1844 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1845 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1846 tp->rcv_scale = tp->request_r_scale; 1847 tp->snd_wnd = tiwin; 1848 } 1849 /* 1850 * Make transitions: 1851 * SYN-RECEIVED -> ESTABLISHED 1852 * SYN-RECEIVED* -> FIN-WAIT-1 1853 */ 1854 tp->t_starttime = ticks; 1855 if (tp->t_flags & TF_NEEDFIN) { 1856 tp->t_state = TCPS_FIN_WAIT_1; 1857 tp->t_flags &= ~TF_NEEDFIN; 1858 } else { 1859 tp->t_state = TCPS_ESTABLISHED; 1860 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1861 } 1862 /* 1863 * If segment contains data or ACK, will call tcp_reass() 1864 * later; if not, do so now to pass queued data to user. 1865 */ 1866 if (tlen == 0 && (thflags & TH_FIN) == 0) 1867 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1868 (struct mbuf *)0); 1869 tp->snd_wl1 = th->th_seq - 1; 1870 /* FALLTHROUGH */ 1871 1872 /* 1873 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1874 * ACKs. If the ack is in the range 1875 * tp->snd_una < th->th_ack <= tp->snd_max 1876 * then advance tp->snd_una to th->th_ack and drop 1877 * data from the retransmission queue. If this ACK reflects 1878 * more up to date window information we update our window information. 1879 */ 1880 case TCPS_ESTABLISHED: 1881 case TCPS_FIN_WAIT_1: 1882 case TCPS_FIN_WAIT_2: 1883 case TCPS_CLOSE_WAIT: 1884 case TCPS_CLOSING: 1885 case TCPS_LAST_ACK: 1886 case TCPS_TIME_WAIT: 1887 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: timewait", 1888 __func__)); 1889 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1890 tcpstat.tcps_rcvacktoomuch++; 1891 goto dropafterack; 1892 } 1893 if (tp->sack_enable && 1894 ((to.to_flags & TOF_SACK) || 1895 !TAILQ_EMPTY(&tp->snd_holes))) 1896 tcp_sack_doack(tp, &to, th->th_ack); 1897 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1898 if (tlen == 0 && tiwin == tp->snd_wnd) { 1899 tcpstat.tcps_rcvdupack++; 1900 /* 1901 * If we have outstanding data (other than 1902 * a window probe), this is a completely 1903 * duplicate ack (ie, window info didn't 1904 * change), the ack is the biggest we've 1905 * seen and we've seen exactly our rexmt 1906 * threshhold of them, assume a packet 1907 * has been dropped and retransmit it. 1908 * Kludge snd_nxt & the congestion 1909 * window so we send only this one 1910 * packet. 1911 * 1912 * We know we're losing at the current 1913 * window size so do congestion avoidance 1914 * (set ssthresh to half the current window 1915 * and pull our congestion window back to 1916 * the new ssthresh). 1917 * 1918 * Dup acks mean that packets have left the 1919 * network (they're now cached at the receiver) 1920 * so bump cwnd by the amount in the receiver 1921 * to keep a constant cwnd packets in the 1922 * network. 1923 */ 1924 if (!tcp_timer_active(tp, TT_REXMT) || 1925 th->th_ack != tp->snd_una) 1926 tp->t_dupacks = 0; 1927 else if (++tp->t_dupacks > tcprexmtthresh || 1928 ((tcp_do_newreno || tp->sack_enable) && 1929 IN_FASTRECOVERY(tp))) { 1930 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 1931 int awnd; 1932 1933 /* 1934 * Compute the amount of data in flight first. 1935 * We can inject new data into the pipe iff 1936 * we have less than 1/2 the original window's 1937 * worth of data in flight. 1938 */ 1939 awnd = (tp->snd_nxt - tp->snd_fack) + 1940 tp->sackhint.sack_bytes_rexmit; 1941 if (awnd < tp->snd_ssthresh) { 1942 tp->snd_cwnd += tp->t_maxseg; 1943 if (tp->snd_cwnd > tp->snd_ssthresh) 1944 tp->snd_cwnd = tp->snd_ssthresh; 1945 } 1946 } else 1947 tp->snd_cwnd += tp->t_maxseg; 1948 (void) tcp_output(tp); 1949 goto drop; 1950 } else if (tp->t_dupacks == tcprexmtthresh) { 1951 tcp_seq onxt = tp->snd_nxt; 1952 u_int win; 1953 1954 /* 1955 * If we're doing sack, check to 1956 * see if we're already in sack 1957 * recovery. If we're not doing sack, 1958 * check to see if we're in newreno 1959 * recovery. 1960 */ 1961 if (tp->sack_enable) { 1962 if (IN_FASTRECOVERY(tp)) { 1963 tp->t_dupacks = 0; 1964 break; 1965 } 1966 } else if (tcp_do_newreno) { 1967 if (SEQ_LEQ(th->th_ack, 1968 tp->snd_recover)) { 1969 tp->t_dupacks = 0; 1970 break; 1971 } 1972 } 1973 win = min(tp->snd_wnd, tp->snd_cwnd) / 1974 2 / tp->t_maxseg; 1975 if (win < 2) 1976 win = 2; 1977 tp->snd_ssthresh = win * tp->t_maxseg; 1978 ENTER_FASTRECOVERY(tp); 1979 tp->snd_recover = tp->snd_max; 1980 tcp_timer_activate(tp, TT_REXMT, 0); 1981 tp->t_rtttime = 0; 1982 if (tp->sack_enable) { 1983 tcpstat.tcps_sack_recovery_episode++; 1984 tp->sack_newdata = tp->snd_nxt; 1985 tp->snd_cwnd = tp->t_maxseg; 1986 (void) tcp_output(tp); 1987 goto drop; 1988 } 1989 tp->snd_nxt = th->th_ack; 1990 tp->snd_cwnd = tp->t_maxseg; 1991 (void) tcp_output(tp); 1992 KASSERT(tp->snd_limited <= 2, 1993 ("%s: tp->snd_limited too big", 1994 __func__)); 1995 tp->snd_cwnd = tp->snd_ssthresh + 1996 tp->t_maxseg * 1997 (tp->t_dupacks - tp->snd_limited); 1998 if (SEQ_GT(onxt, tp->snd_nxt)) 1999 tp->snd_nxt = onxt; 2000 goto drop; 2001 } else if (tcp_do_rfc3042) { 2002 u_long oldcwnd = tp->snd_cwnd; 2003 tcp_seq oldsndmax = tp->snd_max; 2004 u_int sent; 2005 2006 KASSERT(tp->t_dupacks == 1 || 2007 tp->t_dupacks == 2, 2008 ("%s: dupacks not 1 or 2", 2009 __func__)); 2010 if (tp->t_dupacks == 1) 2011 tp->snd_limited = 0; 2012 tp->snd_cwnd = 2013 (tp->snd_nxt - tp->snd_una) + 2014 (tp->t_dupacks - tp->snd_limited) * 2015 tp->t_maxseg; 2016 (void) tcp_output(tp); 2017 sent = tp->snd_max - oldsndmax; 2018 if (sent > tp->t_maxseg) { 2019 KASSERT((tp->t_dupacks == 2 && 2020 tp->snd_limited == 0) || 2021 (sent == tp->t_maxseg + 1 && 2022 tp->t_flags & TF_SENTFIN), 2023 ("%s: sent too much", 2024 __func__)); 2025 tp->snd_limited = 2; 2026 } else if (sent > 0) 2027 ++tp->snd_limited; 2028 tp->snd_cwnd = oldcwnd; 2029 goto drop; 2030 } 2031 } else 2032 tp->t_dupacks = 0; 2033 break; 2034 } 2035 2036 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2037 ("%s: th_ack <= snd_una", __func__)); 2038 2039 /* 2040 * If the congestion window was inflated to account 2041 * for the other side's cached packets, retract it. 2042 */ 2043 if (tcp_do_newreno || tp->sack_enable) { 2044 if (IN_FASTRECOVERY(tp)) { 2045 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2046 if (tp->sack_enable) 2047 tcp_sack_partialack(tp, th); 2048 else 2049 tcp_newreno_partial_ack(tp, th); 2050 } else { 2051 /* 2052 * Out of fast recovery. 2053 * Window inflation should have left us 2054 * with approximately snd_ssthresh 2055 * outstanding data. 2056 * But in case we would be inclined to 2057 * send a burst, better to do it via 2058 * the slow start mechanism. 2059 */ 2060 if (SEQ_GT(th->th_ack + 2061 tp->snd_ssthresh, 2062 tp->snd_max)) 2063 tp->snd_cwnd = tp->snd_max - 2064 th->th_ack + 2065 tp->t_maxseg; 2066 else 2067 tp->snd_cwnd = tp->snd_ssthresh; 2068 } 2069 } 2070 } else { 2071 if (tp->t_dupacks >= tcprexmtthresh && 2072 tp->snd_cwnd > tp->snd_ssthresh) 2073 tp->snd_cwnd = tp->snd_ssthresh; 2074 } 2075 tp->t_dupacks = 0; 2076 /* 2077 * If we reach this point, ACK is not a duplicate, 2078 * i.e., it ACKs something we sent. 2079 */ 2080 if (tp->t_flags & TF_NEEDSYN) { 2081 /* 2082 * T/TCP: Connection was half-synchronized, and our 2083 * SYN has been ACK'd (so connection is now fully 2084 * synchronized). Go to non-starred state, 2085 * increment snd_una for ACK of SYN, and check if 2086 * we can do window scaling. 2087 */ 2088 tp->t_flags &= ~TF_NEEDSYN; 2089 tp->snd_una++; 2090 /* Do window scaling? */ 2091 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2092 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2093 tp->rcv_scale = tp->request_r_scale; 2094 /* Send window already scaled. */ 2095 } 2096 } 2097 2098 process_ACK: 2099 KASSERT(headlocked, ("%s: process_ACK: head not locked", 2100 __func__)); 2101 INP_LOCK_ASSERT(tp->t_inpcb); 2102 2103 acked = th->th_ack - tp->snd_una; 2104 tcpstat.tcps_rcvackpack++; 2105 tcpstat.tcps_rcvackbyte += acked; 2106 2107 /* 2108 * If we just performed our first retransmit, and the ACK 2109 * arrives within our recovery window, then it was a mistake 2110 * to do the retransmit in the first place. Recover our 2111 * original cwnd and ssthresh, and proceed to transmit where 2112 * we left off. 2113 */ 2114 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2115 ++tcpstat.tcps_sndrexmitbad; 2116 tp->snd_cwnd = tp->snd_cwnd_prev; 2117 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2118 tp->snd_recover = tp->snd_recover_prev; 2119 if (tp->t_flags & TF_WASFRECOVERY) 2120 ENTER_FASTRECOVERY(tp); 2121 tp->snd_nxt = tp->snd_max; 2122 tp->t_badrxtwin = 0; /* XXX probably not required */ 2123 } 2124 2125 /* 2126 * If we have a timestamp reply, update smoothed 2127 * round trip time. If no timestamp is present but 2128 * transmit timer is running and timed sequence 2129 * number was acked, update smoothed round trip time. 2130 * Since we now have an rtt measurement, cancel the 2131 * timer backoff (cf., Phil Karn's retransmit alg.). 2132 * Recompute the initial retransmit timer. 2133 * 2134 * Some boxes send broken timestamp replies 2135 * during the SYN+ACK phase, ignore 2136 * timestamps of 0 or we could calculate a 2137 * huge RTT and blow up the retransmit timer. 2138 */ 2139 if ((to.to_flags & TOF_TS) != 0 && 2140 to.to_tsecr) { 2141 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2142 tp->t_rttlow = ticks - to.to_tsecr; 2143 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2144 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2145 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2146 tp->t_rttlow = ticks - tp->t_rtttime; 2147 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2148 } 2149 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2150 2151 /* 2152 * If all outstanding data is acked, stop retransmit 2153 * timer and remember to restart (more output or persist). 2154 * If there is more data to be acked, restart retransmit 2155 * timer, using current (possibly backed-off) value. 2156 */ 2157 if (th->th_ack == tp->snd_max) { 2158 tcp_timer_activate(tp, TT_REXMT, 0); 2159 needoutput = 1; 2160 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2161 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2162 2163 /* 2164 * If no data (only SYN) was ACK'd, 2165 * skip rest of ACK processing. 2166 */ 2167 if (acked == 0) 2168 goto step6; 2169 2170 /* 2171 * When new data is acked, open the congestion window. 2172 * If the window gives us less than ssthresh packets 2173 * in flight, open exponentially (maxseg per packet). 2174 * Otherwise open linearly: maxseg per window 2175 * (maxseg^2 / cwnd per packet). 2176 */ 2177 if ((!tcp_do_newreno && !tp->sack_enable) || 2178 !IN_FASTRECOVERY(tp)) { 2179 u_int cw = tp->snd_cwnd; 2180 u_int incr = tp->t_maxseg; 2181 if (cw > tp->snd_ssthresh) 2182 incr = incr * incr / cw; 2183 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2184 } 2185 SOCKBUF_LOCK(&so->so_snd); 2186 if (acked > so->so_snd.sb_cc) { 2187 tp->snd_wnd -= so->so_snd.sb_cc; 2188 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2189 ourfinisacked = 1; 2190 } else { 2191 sbdrop_locked(&so->so_snd, acked); 2192 tp->snd_wnd -= acked; 2193 ourfinisacked = 0; 2194 } 2195 sowwakeup_locked(so); 2196 /* detect una wraparound */ 2197 if ((tcp_do_newreno || tp->sack_enable) && 2198 !IN_FASTRECOVERY(tp) && 2199 SEQ_GT(tp->snd_una, tp->snd_recover) && 2200 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2201 tp->snd_recover = th->th_ack - 1; 2202 if ((tcp_do_newreno || tp->sack_enable) && 2203 IN_FASTRECOVERY(tp) && 2204 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2205 EXIT_FASTRECOVERY(tp); 2206 tp->snd_una = th->th_ack; 2207 if (tp->sack_enable) { 2208 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2209 tp->snd_recover = tp->snd_una; 2210 } 2211 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2212 tp->snd_nxt = tp->snd_una; 2213 2214 switch (tp->t_state) { 2215 2216 /* 2217 * In FIN_WAIT_1 STATE in addition to the processing 2218 * for the ESTABLISHED state if our FIN is now acknowledged 2219 * then enter FIN_WAIT_2. 2220 */ 2221 case TCPS_FIN_WAIT_1: 2222 if (ourfinisacked) { 2223 /* 2224 * If we can't receive any more 2225 * data, then closing user can proceed. 2226 * Starting the timer is contrary to the 2227 * specification, but if we don't get a FIN 2228 * we'll hang forever. 2229 */ 2230 /* XXXjl 2231 * we should release the tp also, and use a 2232 * compressed state. 2233 */ 2234 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2235 int timeout; 2236 2237 soisdisconnected(so); 2238 timeout = (tcp_fast_finwait2_recycle) ? 2239 tcp_finwait2_timeout : tcp_maxidle; 2240 tcp_timer_activate(tp, TT_2MSL, timeout); 2241 } 2242 tp->t_state = TCPS_FIN_WAIT_2; 2243 } 2244 break; 2245 2246 /* 2247 * In CLOSING STATE in addition to the processing for 2248 * the ESTABLISHED state if the ACK acknowledges our FIN 2249 * then enter the TIME-WAIT state, otherwise ignore 2250 * the segment. 2251 */ 2252 case TCPS_CLOSING: 2253 if (ourfinisacked) { 2254 KASSERT(headlocked, ("%s: process_ACK: " 2255 "head not locked", __func__)); 2256 tcp_twstart(tp); 2257 INP_INFO_WUNLOCK(&tcbinfo); 2258 headlocked = 0; 2259 m_freem(m); 2260 return (0); 2261 } 2262 break; 2263 2264 /* 2265 * In LAST_ACK, we may still be waiting for data to drain 2266 * and/or to be acked, as well as for the ack of our FIN. 2267 * If our FIN is now acknowledged, delete the TCB, 2268 * enter the closed state and return. 2269 */ 2270 case TCPS_LAST_ACK: 2271 if (ourfinisacked) { 2272 KASSERT(headlocked, ("%s: process_ACK: " 2273 "tcp_close: head not locked", __func__)); 2274 tp = tcp_close(tp); 2275 goto drop; 2276 } 2277 break; 2278 2279 /* 2280 * In TIME_WAIT state the only thing that should arrive 2281 * is a retransmission of the remote FIN. Acknowledge 2282 * it and restart the finack timer. 2283 */ 2284 case TCPS_TIME_WAIT: 2285 KASSERT(tp->t_state != TCPS_TIME_WAIT, 2286 ("%s: timewait", __func__)); 2287 tcp_timer_activate(tp, TT_2MSL, 2 * tcp_msl); 2288 goto dropafterack; 2289 } 2290 } 2291 2292 step6: 2293 KASSERT(headlocked, ("%s: step6: head not locked", __func__)); 2294 INP_LOCK_ASSERT(tp->t_inpcb); 2295 2296 /* 2297 * Update window information. 2298 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2299 */ 2300 if ((thflags & TH_ACK) && 2301 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2302 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2303 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2304 /* keep track of pure window updates */ 2305 if (tlen == 0 && 2306 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2307 tcpstat.tcps_rcvwinupd++; 2308 tp->snd_wnd = tiwin; 2309 tp->snd_wl1 = th->th_seq; 2310 tp->snd_wl2 = th->th_ack; 2311 if (tp->snd_wnd > tp->max_sndwnd) 2312 tp->max_sndwnd = tp->snd_wnd; 2313 needoutput = 1; 2314 } 2315 2316 /* 2317 * Process segments with URG. 2318 */ 2319 if ((thflags & TH_URG) && th->th_urp && 2320 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2321 /* 2322 * This is a kludge, but if we receive and accept 2323 * random urgent pointers, we'll crash in 2324 * soreceive. It's hard to imagine someone 2325 * actually wanting to send this much urgent data. 2326 */ 2327 SOCKBUF_LOCK(&so->so_rcv); 2328 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2329 th->th_urp = 0; /* XXX */ 2330 thflags &= ~TH_URG; /* XXX */ 2331 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2332 goto dodata; /* XXX */ 2333 } 2334 /* 2335 * If this segment advances the known urgent pointer, 2336 * then mark the data stream. This should not happen 2337 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2338 * a FIN has been received from the remote side. 2339 * In these states we ignore the URG. 2340 * 2341 * According to RFC961 (Assigned Protocols), 2342 * the urgent pointer points to the last octet 2343 * of urgent data. We continue, however, 2344 * to consider it to indicate the first octet 2345 * of data past the urgent section as the original 2346 * spec states (in one of two places). 2347 */ 2348 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2349 tp->rcv_up = th->th_seq + th->th_urp; 2350 so->so_oobmark = so->so_rcv.sb_cc + 2351 (tp->rcv_up - tp->rcv_nxt) - 1; 2352 if (so->so_oobmark == 0) 2353 so->so_rcv.sb_state |= SBS_RCVATMARK; 2354 sohasoutofband(so); 2355 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2356 } 2357 SOCKBUF_UNLOCK(&so->so_rcv); 2358 /* 2359 * Remove out of band data so doesn't get presented to user. 2360 * This can happen independent of advancing the URG pointer, 2361 * but if two URG's are pending at once, some out-of-band 2362 * data may creep in... ick. 2363 */ 2364 if (th->th_urp <= (u_long)tlen && 2365 !(so->so_options & SO_OOBINLINE)) { 2366 /* hdr drop is delayed */ 2367 tcp_pulloutofband(so, th, m, drop_hdrlen); 2368 } 2369 } else { 2370 /* 2371 * If no out of band data is expected, 2372 * pull receive urgent pointer along 2373 * with the receive window. 2374 */ 2375 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2376 tp->rcv_up = tp->rcv_nxt; 2377 } 2378 dodata: /* XXX */ 2379 KASSERT(headlocked, ("%s: dodata: head not locked", __func__)); 2380 INP_LOCK_ASSERT(tp->t_inpcb); 2381 2382 /* 2383 * Process the segment text, merging it into the TCP sequencing queue, 2384 * and arranging for acknowledgment of receipt if necessary. 2385 * This process logically involves adjusting tp->rcv_wnd as data 2386 * is presented to the user (this happens in tcp_usrreq.c, 2387 * case PRU_RCVD). If a FIN has already been received on this 2388 * connection then we just ignore the text. 2389 */ 2390 if ((tlen || (thflags & TH_FIN)) && 2391 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2392 tcp_seq save_start = th->th_seq; 2393 tcp_seq save_end = th->th_seq + tlen; 2394 m_adj(m, drop_hdrlen); /* delayed header drop */ 2395 /* 2396 * Insert segment which includes th into TCP reassembly queue 2397 * with control block tp. Set thflags to whether reassembly now 2398 * includes a segment with FIN. This handles the common case 2399 * inline (segment is the next to be received on an established 2400 * connection, and the queue is empty), avoiding linkage into 2401 * and removal from the queue and repetition of various 2402 * conversions. 2403 * Set DELACK for segments received in order, but ack 2404 * immediately when segments are out of order (so 2405 * fast retransmit can work). 2406 */ 2407 if (th->th_seq == tp->rcv_nxt && 2408 LIST_EMPTY(&tp->t_segq) && 2409 TCPS_HAVEESTABLISHED(tp->t_state)) { 2410 if (DELAY_ACK(tp)) 2411 tp->t_flags |= TF_DELACK; 2412 else 2413 tp->t_flags |= TF_ACKNOW; 2414 tp->rcv_nxt += tlen; 2415 thflags = th->th_flags & TH_FIN; 2416 tcpstat.tcps_rcvpack++; 2417 tcpstat.tcps_rcvbyte += tlen; 2418 ND6_HINT(tp); 2419 SOCKBUF_LOCK(&so->so_rcv); 2420 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2421 m_freem(m); 2422 else 2423 sbappendstream_locked(&so->so_rcv, m); 2424 sorwakeup_locked(so); 2425 } else { 2426 thflags = tcp_reass(tp, th, &tlen, m); 2427 tp->t_flags |= TF_ACKNOW; 2428 } 2429 if (tlen > 0 && tp->sack_enable) 2430 tcp_update_sack_list(tp, save_start, save_end); 2431 #if 0 2432 /* 2433 * Note the amount of data that peer has sent into 2434 * our window, in order to estimate the sender's 2435 * buffer size. 2436 * XXX: Unused. 2437 */ 2438 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2439 #endif 2440 } else { 2441 m_freem(m); 2442 thflags &= ~TH_FIN; 2443 } 2444 2445 /* 2446 * If FIN is received ACK the FIN and let the user know 2447 * that the connection is closing. 2448 */ 2449 if (thflags & TH_FIN) { 2450 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2451 socantrcvmore(so); 2452 /* 2453 * If connection is half-synchronized 2454 * (ie NEEDSYN flag on) then delay ACK, 2455 * so it may be piggybacked when SYN is sent. 2456 * Otherwise, since we received a FIN then no 2457 * more input can be expected, send ACK now. 2458 */ 2459 if (tp->t_flags & TF_NEEDSYN) 2460 tp->t_flags |= TF_DELACK; 2461 else 2462 tp->t_flags |= TF_ACKNOW; 2463 tp->rcv_nxt++; 2464 } 2465 switch (tp->t_state) { 2466 2467 /* 2468 * In SYN_RECEIVED and ESTABLISHED STATES 2469 * enter the CLOSE_WAIT state. 2470 */ 2471 case TCPS_SYN_RECEIVED: 2472 tp->t_starttime = ticks; 2473 /*FALLTHROUGH*/ 2474 case TCPS_ESTABLISHED: 2475 tp->t_state = TCPS_CLOSE_WAIT; 2476 break; 2477 2478 /* 2479 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2480 * enter the CLOSING state. 2481 */ 2482 case TCPS_FIN_WAIT_1: 2483 tp->t_state = TCPS_CLOSING; 2484 break; 2485 2486 /* 2487 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2488 * starting the time-wait timer, turning off the other 2489 * standard timers. 2490 */ 2491 case TCPS_FIN_WAIT_2: 2492 KASSERT(headlocked == 1, ("%s: dodata: " 2493 "TCP_FIN_WAIT_2: head not locked", __func__)); 2494 tcp_twstart(tp); 2495 INP_INFO_WUNLOCK(&tcbinfo); 2496 return (0); 2497 2498 /* 2499 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2500 */ 2501 case TCPS_TIME_WAIT: 2502 KASSERT(tp->t_state != TCPS_TIME_WAIT, 2503 ("%s: timewait", __func__)); 2504 tcp_timer_activate(tp, TT_2MSL, 2 * tcp_msl); 2505 break; 2506 } 2507 } 2508 INP_INFO_WUNLOCK(&tcbinfo); 2509 headlocked = 0; 2510 #ifdef TCPDEBUG 2511 if (so->so_options & SO_DEBUG) 2512 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2513 &tcp_savetcp, 0); 2514 #endif 2515 2516 /* 2517 * Return any desired output. 2518 */ 2519 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2520 (void) tcp_output(tp); 2521 2522 check_delack: 2523 KASSERT(headlocked == 0, ("%s: check_delack: head locked", 2524 __func__)); 2525 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 2526 INP_LOCK_ASSERT(tp->t_inpcb); 2527 if (tp->t_flags & TF_DELACK) { 2528 tp->t_flags &= ~TF_DELACK; 2529 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2530 } 2531 INP_UNLOCK(tp->t_inpcb); 2532 return (0); 2533 2534 dropafterack: 2535 KASSERT(headlocked, ("%s: dropafterack: head not locked", __func__)); 2536 /* 2537 * Generate an ACK dropping incoming segment if it occupies 2538 * sequence space, where the ACK reflects our state. 2539 * 2540 * We can now skip the test for the RST flag since all 2541 * paths to this code happen after packets containing 2542 * RST have been dropped. 2543 * 2544 * In the SYN-RECEIVED state, don't send an ACK unless the 2545 * segment we received passes the SYN-RECEIVED ACK test. 2546 * If it fails send a RST. This breaks the loop in the 2547 * "LAND" DoS attack, and also prevents an ACK storm 2548 * between two listening ports that have been sent forged 2549 * SYN segments, each with the source address of the other. 2550 */ 2551 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2552 (SEQ_GT(tp->snd_una, th->th_ack) || 2553 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2554 rstreason = BANDLIM_RST_OPENPORT; 2555 goto dropwithreset; 2556 } 2557 #ifdef TCPDEBUG 2558 if (so->so_options & SO_DEBUG) 2559 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2560 &tcp_savetcp, 0); 2561 #endif 2562 KASSERT(headlocked, ("%s: headlocked should be 1", __func__)); 2563 INP_INFO_WUNLOCK(&tcbinfo); 2564 tp->t_flags |= TF_ACKNOW; 2565 (void) tcp_output(tp); 2566 INP_UNLOCK(tp->t_inpcb); 2567 m_freem(m); 2568 return (0); 2569 2570 dropwithreset: 2571 KASSERT(headlocked, ("%s: dropwithreset: head not locked", __func__)); 2572 2573 tcp_dropwithreset(m, th, tp, tlen, rstreason); 2574 2575 if (tp != NULL) 2576 INP_UNLOCK(tp->t_inpcb); 2577 if (headlocked) 2578 INP_INFO_WUNLOCK(&tcbinfo); 2579 return (0); 2580 2581 drop: 2582 /* 2583 * Drop space held by incoming segment and return. 2584 */ 2585 #ifdef TCPDEBUG 2586 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2587 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2588 &tcp_savetcp, 0); 2589 #endif 2590 if (tp != NULL) 2591 INP_UNLOCK(tp->t_inpcb); 2592 if (headlocked) 2593 INP_INFO_WUNLOCK(&tcbinfo); 2594 m_freem(m); 2595 return (0); 2596 } 2597 2598 2599 /* 2600 * Issue RST on TCP segment. The mbuf must still include the original 2601 * packet header. 2602 */ 2603 static void 2604 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 2605 int tlen, int rstreason) 2606 { 2607 struct ip *ip; 2608 #ifdef INET6 2609 struct ip6_hdr *ip6; 2610 #endif 2611 /* 2612 * Generate a RST, dropping incoming segment. 2613 * Make ACK acceptable to originator of segment. 2614 * Don't bother to respond if destination was broadcast/multicast. 2615 * tp may be NULL. 2616 */ 2617 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2618 goto drop; 2619 #ifdef INET6 2620 if (mtod(m, struct ip *)->ip_v == 6) { 2621 ip6 = mtod(m, struct ip6_hdr *); 2622 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2623 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2624 goto drop; 2625 /* IPv6 anycast check is done at tcp6_input() */ 2626 } else 2627 #endif 2628 { 2629 ip = mtod(m, struct ip *); 2630 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2631 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2632 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2633 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2634 goto drop; 2635 } 2636 2637 /* Perform bandwidth limiting. */ 2638 if (badport_bandlim(rstreason) < 0) 2639 goto drop; 2640 2641 /* tcp_respond consumes the mbuf chain. */ 2642 if (th->th_flags & TH_ACK) { 2643 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 2644 th->th_ack, TH_RST); 2645 } else { 2646 if (th->th_flags & TH_SYN) 2647 tlen++; 2648 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2649 (tcp_seq)0, TH_RST|TH_ACK); 2650 } 2651 return; 2652 drop: 2653 m_freem(m); 2654 return; 2655 } 2656 2657 /* 2658 * Parse TCP options and place in tcpopt. 2659 */ 2660 static void 2661 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 2662 { 2663 int opt, optlen; 2664 2665 to->to_flags = 0; 2666 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2667 opt = cp[0]; 2668 if (opt == TCPOPT_EOL) 2669 break; 2670 if (opt == TCPOPT_NOP) 2671 optlen = 1; 2672 else { 2673 if (cnt < 2) 2674 break; 2675 optlen = cp[1]; 2676 if (optlen < 2 || optlen > cnt) 2677 break; 2678 } 2679 switch (opt) { 2680 case TCPOPT_MAXSEG: 2681 if (optlen != TCPOLEN_MAXSEG) 2682 continue; 2683 if (!(flags & TO_SYN)) 2684 continue; 2685 to->to_flags |= TOF_MSS; 2686 bcopy((char *)cp + 2, 2687 (char *)&to->to_mss, sizeof(to->to_mss)); 2688 to->to_mss = ntohs(to->to_mss); 2689 break; 2690 case TCPOPT_WINDOW: 2691 if (optlen != TCPOLEN_WINDOW) 2692 continue; 2693 if (!(flags & TO_SYN)) 2694 continue; 2695 to->to_flags |= TOF_SCALE; 2696 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 2697 break; 2698 case TCPOPT_TIMESTAMP: 2699 if (optlen != TCPOLEN_TIMESTAMP) 2700 continue; 2701 to->to_flags |= TOF_TS; 2702 bcopy((char *)cp + 2, 2703 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2704 to->to_tsval = ntohl(to->to_tsval); 2705 bcopy((char *)cp + 6, 2706 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2707 to->to_tsecr = ntohl(to->to_tsecr); 2708 break; 2709 #ifdef TCP_SIGNATURE 2710 /* 2711 * XXX In order to reply to a host which has set the 2712 * TCP_SIGNATURE option in its initial SYN, we have to 2713 * record the fact that the option was observed here 2714 * for the syncache code to perform the correct response. 2715 */ 2716 case TCPOPT_SIGNATURE: 2717 if (optlen != TCPOLEN_SIGNATURE) 2718 continue; 2719 to->to_flags |= TOF_SIGNATURE; 2720 to->to_signature = cp + 2; 2721 break; 2722 #endif 2723 case TCPOPT_SACK_PERMITTED: 2724 if (optlen != TCPOLEN_SACK_PERMITTED) 2725 continue; 2726 if (!(flags & TO_SYN)) 2727 continue; 2728 if (!tcp_do_sack) 2729 continue; 2730 to->to_flags |= TOF_SACKPERM; 2731 break; 2732 case TCPOPT_SACK: 2733 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2734 continue; 2735 if (flags & TO_SYN) 2736 continue; 2737 to->to_flags |= TOF_SACK; 2738 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2739 to->to_sacks = cp + 2; 2740 tcpstat.tcps_sack_rcv_blocks++; 2741 break; 2742 default: 2743 continue; 2744 } 2745 } 2746 } 2747 2748 /* 2749 * Pull out of band byte out of a segment so 2750 * it doesn't appear in the user's data queue. 2751 * It is still reflected in the segment length for 2752 * sequencing purposes. 2753 */ 2754 static void 2755 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 2756 int off) 2757 { 2758 int cnt = off + th->th_urp - 1; 2759 2760 while (cnt >= 0) { 2761 if (m->m_len > cnt) { 2762 char *cp = mtod(m, caddr_t) + cnt; 2763 struct tcpcb *tp = sototcpcb(so); 2764 2765 tp->t_iobc = *cp; 2766 tp->t_oobflags |= TCPOOB_HAVEDATA; 2767 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2768 m->m_len--; 2769 if (m->m_flags & M_PKTHDR) 2770 m->m_pkthdr.len--; 2771 return; 2772 } 2773 cnt -= m->m_len; 2774 m = m->m_next; 2775 if (m == NULL) 2776 break; 2777 } 2778 panic("tcp_pulloutofband"); 2779 } 2780 2781 /* 2782 * Collect new round-trip time estimate 2783 * and update averages and current timeout. 2784 */ 2785 static void 2786 tcp_xmit_timer(struct tcpcb *tp, int rtt) 2787 { 2788 int delta; 2789 2790 INP_LOCK_ASSERT(tp->t_inpcb); 2791 2792 tcpstat.tcps_rttupdated++; 2793 tp->t_rttupdated++; 2794 if (tp->t_srtt != 0) { 2795 /* 2796 * srtt is stored as fixed point with 5 bits after the 2797 * binary point (i.e., scaled by 8). The following magic 2798 * is equivalent to the smoothing algorithm in rfc793 with 2799 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2800 * point). Adjust rtt to origin 0. 2801 */ 2802 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2803 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2804 2805 if ((tp->t_srtt += delta) <= 0) 2806 tp->t_srtt = 1; 2807 2808 /* 2809 * We accumulate a smoothed rtt variance (actually, a 2810 * smoothed mean difference), then set the retransmit 2811 * timer to smoothed rtt + 4 times the smoothed variance. 2812 * rttvar is stored as fixed point with 4 bits after the 2813 * binary point (scaled by 16). The following is 2814 * equivalent to rfc793 smoothing with an alpha of .75 2815 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2816 * rfc793's wired-in beta. 2817 */ 2818 if (delta < 0) 2819 delta = -delta; 2820 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2821 if ((tp->t_rttvar += delta) <= 0) 2822 tp->t_rttvar = 1; 2823 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2824 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2825 } else { 2826 /* 2827 * No rtt measurement yet - use the unsmoothed rtt. 2828 * Set the variance to half the rtt (so our first 2829 * retransmit happens at 3*rtt). 2830 */ 2831 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2832 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2833 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2834 } 2835 tp->t_rtttime = 0; 2836 tp->t_rxtshift = 0; 2837 2838 /* 2839 * the retransmit should happen at rtt + 4 * rttvar. 2840 * Because of the way we do the smoothing, srtt and rttvar 2841 * will each average +1/2 tick of bias. When we compute 2842 * the retransmit timer, we want 1/2 tick of rounding and 2843 * 1 extra tick because of +-1/2 tick uncertainty in the 2844 * firing of the timer. The bias will give us exactly the 2845 * 1.5 tick we need. But, because the bias is 2846 * statistical, we have to test that we don't drop below 2847 * the minimum feasible timer (which is 2 ticks). 2848 */ 2849 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2850 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2851 2852 /* 2853 * We received an ack for a packet that wasn't retransmitted; 2854 * it is probably safe to discard any error indications we've 2855 * received recently. This isn't quite right, but close enough 2856 * for now (a route might have failed after we sent a segment, 2857 * and the return path might not be symmetrical). 2858 */ 2859 tp->t_softerror = 0; 2860 } 2861 2862 /* 2863 * Determine a reasonable value for maxseg size. 2864 * If the route is known, check route for mtu. 2865 * If none, use an mss that can be handled on the outgoing 2866 * interface without forcing IP to fragment; if bigger than 2867 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2868 * to utilize large mbufs. If no route is found, route has no mtu, 2869 * or the destination isn't local, use a default, hopefully conservative 2870 * size (usually 512 or the default IP max size, but no more than the mtu 2871 * of the interface), as we can't discover anything about intervening 2872 * gateways or networks. We also initialize the congestion/slow start 2873 * window to be a single segment if the destination isn't local. 2874 * While looking at the routing entry, we also initialize other path-dependent 2875 * parameters from pre-set or cached values in the routing entry. 2876 * 2877 * Also take into account the space needed for options that we 2878 * send regularly. Make maxseg shorter by that amount to assure 2879 * that we can send maxseg amount of data even when the options 2880 * are present. Store the upper limit of the length of options plus 2881 * data in maxopd. 2882 * 2883 * 2884 * In case of T/TCP, we call this routine during implicit connection 2885 * setup as well (offer = -1), to initialize maxseg from the cached 2886 * MSS of our peer. 2887 * 2888 * NOTE that this routine is only called when we process an incoming 2889 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2890 */ 2891 void 2892 tcp_mss(struct tcpcb *tp, int offer) 2893 { 2894 int rtt, mss; 2895 u_long bufsize; 2896 u_long maxmtu; 2897 struct inpcb *inp = tp->t_inpcb; 2898 struct socket *so; 2899 struct hc_metrics_lite metrics; 2900 int origoffer = offer; 2901 int mtuflags = 0; 2902 #ifdef INET6 2903 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2904 size_t min_protoh = isipv6 ? 2905 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2906 sizeof (struct tcpiphdr); 2907 #else 2908 const size_t min_protoh = sizeof(struct tcpiphdr); 2909 #endif 2910 2911 /* initialize */ 2912 #ifdef INET6 2913 if (isipv6) { 2914 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags); 2915 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2916 } else 2917 #endif 2918 { 2919 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags); 2920 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2921 } 2922 so = inp->inp_socket; 2923 2924 /* 2925 * no route to sender, stay with default mss and return 2926 */ 2927 if (maxmtu == 0) 2928 return; 2929 2930 /* what have we got? */ 2931 switch (offer) { 2932 case 0: 2933 /* 2934 * Offer == 0 means that there was no MSS on the SYN 2935 * segment, in this case we use tcp_mssdflt. 2936 */ 2937 offer = 2938 #ifdef INET6 2939 isipv6 ? tcp_v6mssdflt : 2940 #endif 2941 tcp_mssdflt; 2942 break; 2943 2944 case -1: 2945 /* 2946 * Offer == -1 means that we didn't receive SYN yet. 2947 */ 2948 /* FALLTHROUGH */ 2949 2950 default: 2951 /* 2952 * Prevent DoS attack with too small MSS. Round up 2953 * to at least minmss. 2954 */ 2955 offer = max(offer, tcp_minmss); 2956 /* 2957 * Sanity check: make sure that maxopd will be large 2958 * enough to allow some data on segments even if the 2959 * all the option space is used (40bytes). Otherwise 2960 * funny things may happen in tcp_output. 2961 */ 2962 offer = max(offer, 64); 2963 } 2964 2965 /* 2966 * rmx information is now retrieved from tcp_hostcache 2967 */ 2968 tcp_hc_get(&inp->inp_inc, &metrics); 2969 2970 /* 2971 * if there's a discovered mtu int tcp hostcache, use it 2972 * else, use the link mtu. 2973 */ 2974 if (metrics.rmx_mtu) 2975 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2976 else { 2977 #ifdef INET6 2978 if (isipv6) { 2979 mss = maxmtu - min_protoh; 2980 if (!path_mtu_discovery && 2981 !in6_localaddr(&inp->in6p_faddr)) 2982 mss = min(mss, tcp_v6mssdflt); 2983 } else 2984 #endif 2985 { 2986 mss = maxmtu - min_protoh; 2987 if (!path_mtu_discovery && 2988 !in_localaddr(inp->inp_faddr)) 2989 mss = min(mss, tcp_mssdflt); 2990 } 2991 } 2992 mss = min(mss, offer); 2993 2994 /* 2995 * maxopd stores the maximum length of data AND options 2996 * in a segment; maxseg is the amount of data in a normal 2997 * segment. We need to store this value (maxopd) apart 2998 * from maxseg, because now every segment carries options 2999 * and thus we normally have somewhat less data in segments. 3000 */ 3001 tp->t_maxopd = mss; 3002 3003 /* 3004 * origoffer==-1 indicates, that no segments were received yet. 3005 * In this case we just guess. 3006 */ 3007 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3008 (origoffer == -1 || 3009 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3010 mss -= TCPOLEN_TSTAMP_APPA; 3011 tp->t_maxseg = mss; 3012 3013 #if (MCLBYTES & (MCLBYTES - 1)) == 0 3014 if (mss > MCLBYTES) 3015 mss &= ~(MCLBYTES-1); 3016 #else 3017 if (mss > MCLBYTES) 3018 mss = mss / MCLBYTES * MCLBYTES; 3019 #endif 3020 tp->t_maxseg = mss; 3021 3022 /* 3023 * If there's a pipesize, change the socket buffer to that size, 3024 * don't change if sb_hiwat is different than default (then it 3025 * has been changed on purpose with setsockopt). 3026 * Make the socket buffers an integral number of mss units; 3027 * if the mss is larger than the socket buffer, decrease the mss. 3028 */ 3029 SOCKBUF_LOCK(&so->so_snd); 3030 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 3031 bufsize = metrics.rmx_sendpipe; 3032 else 3033 bufsize = so->so_snd.sb_hiwat; 3034 if (bufsize < mss) 3035 mss = bufsize; 3036 else { 3037 bufsize = roundup(bufsize, mss); 3038 if (bufsize > sb_max) 3039 bufsize = sb_max; 3040 if (bufsize > so->so_snd.sb_hiwat) 3041 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3042 } 3043 SOCKBUF_UNLOCK(&so->so_snd); 3044 tp->t_maxseg = mss; 3045 3046 SOCKBUF_LOCK(&so->so_rcv); 3047 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3048 bufsize = metrics.rmx_recvpipe; 3049 else 3050 bufsize = so->so_rcv.sb_hiwat; 3051 if (bufsize > mss) { 3052 bufsize = roundup(bufsize, mss); 3053 if (bufsize > sb_max) 3054 bufsize = sb_max; 3055 if (bufsize > so->so_rcv.sb_hiwat) 3056 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3057 } 3058 SOCKBUF_UNLOCK(&so->so_rcv); 3059 /* 3060 * While we're here, check the others too 3061 */ 3062 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 3063 tp->t_srtt = rtt; 3064 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3065 tcpstat.tcps_usedrtt++; 3066 if (metrics.rmx_rttvar) { 3067 tp->t_rttvar = metrics.rmx_rttvar; 3068 tcpstat.tcps_usedrttvar++; 3069 } else { 3070 /* default variation is +- 1 rtt */ 3071 tp->t_rttvar = 3072 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3073 } 3074 TCPT_RANGESET(tp->t_rxtcur, 3075 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3076 tp->t_rttmin, TCPTV_REXMTMAX); 3077 } 3078 if (metrics.rmx_ssthresh) { 3079 /* 3080 * There's some sort of gateway or interface 3081 * buffer limit on the path. Use this to set 3082 * the slow start threshhold, but set the 3083 * threshold to no less than 2*mss. 3084 */ 3085 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3086 tcpstat.tcps_usedssthresh++; 3087 } 3088 if (metrics.rmx_bandwidth) 3089 tp->snd_bandwidth = metrics.rmx_bandwidth; 3090 3091 /* 3092 * Set the slow-start flight size depending on whether this 3093 * is a local network or not. 3094 * 3095 * Extend this so we cache the cwnd too and retrieve it here. 3096 * Make cwnd even bigger than RFC3390 suggests but only if we 3097 * have previous experience with the remote host. Be careful 3098 * not make cwnd bigger than remote receive window or our own 3099 * send socket buffer. Maybe put some additional upper bound 3100 * on the retrieved cwnd. Should do incremental updates to 3101 * hostcache when cwnd collapses so next connection doesn't 3102 * overloads the path again. 3103 * 3104 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3105 * We currently check only in syncache_socket for that. 3106 */ 3107 #define TCP_METRICS_CWND 3108 #ifdef TCP_METRICS_CWND 3109 if (metrics.rmx_cwnd) 3110 tp->snd_cwnd = max(mss, 3111 min(metrics.rmx_cwnd / 2, 3112 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3113 else 3114 #endif 3115 if (tcp_do_rfc3390) 3116 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3117 #ifdef INET6 3118 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3119 (!isipv6 && in_localaddr(inp->inp_faddr))) 3120 #else 3121 else if (in_localaddr(inp->inp_faddr)) 3122 #endif 3123 tp->snd_cwnd = mss * ss_fltsz_local; 3124 else 3125 tp->snd_cwnd = mss * ss_fltsz; 3126 3127 /* Check the interface for TSO capabilities. */ 3128 if (mtuflags & CSUM_TSO) 3129 tp->t_flags |= TF_TSO; 3130 } 3131 3132 /* 3133 * Determine the MSS option to send on an outgoing SYN. 3134 */ 3135 int 3136 tcp_mssopt(struct in_conninfo *inc) 3137 { 3138 int mss = 0; 3139 u_long maxmtu = 0; 3140 u_long thcmtu = 0; 3141 size_t min_protoh; 3142 #ifdef INET6 3143 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3144 #endif 3145 3146 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3147 3148 #ifdef INET6 3149 if (isipv6) { 3150 mss = tcp_v6mssdflt; 3151 maxmtu = tcp_maxmtu6(inc, NULL); 3152 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3153 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3154 } else 3155 #endif 3156 { 3157 mss = tcp_mssdflt; 3158 maxmtu = tcp_maxmtu(inc, NULL); 3159 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3160 min_protoh = sizeof(struct tcpiphdr); 3161 } 3162 if (maxmtu && thcmtu) 3163 mss = min(maxmtu, thcmtu) - min_protoh; 3164 else if (maxmtu || thcmtu) 3165 mss = max(maxmtu, thcmtu) - min_protoh; 3166 3167 return (mss); 3168 } 3169 3170 3171 /* 3172 * On a partial ack arrives, force the retransmission of the 3173 * next unacknowledged segment. Do not clear tp->t_dupacks. 3174 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3175 * be started again. 3176 */ 3177 static void 3178 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3179 { 3180 tcp_seq onxt = tp->snd_nxt; 3181 u_long ocwnd = tp->snd_cwnd; 3182 3183 tcp_timer_activate(tp, TT_REXMT, 0); 3184 tp->t_rtttime = 0; 3185 tp->snd_nxt = th->th_ack; 3186 /* 3187 * Set snd_cwnd to one segment beyond acknowledged offset. 3188 * (tp->snd_una has not yet been updated when this function is called.) 3189 */ 3190 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3191 tp->t_flags |= TF_ACKNOW; 3192 (void) tcp_output(tp); 3193 tp->snd_cwnd = ocwnd; 3194 if (SEQ_GT(onxt, tp->snd_nxt)) 3195 tp->snd_nxt = onxt; 3196 /* 3197 * Partial window deflation. Relies on fact that tp->snd_una 3198 * not updated yet. 3199 */ 3200 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3201 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3202 else 3203 tp->snd_cwnd = 0; 3204 tp->snd_cwnd += tp->t_maxseg; 3205 } 3206 3207 /* 3208 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3209 * looking for a pcb in the listen state. Returns 0 otherwise. 3210 */ 3211 static int 3212 tcp_timewait(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th, 3213 struct mbuf *m, int tlen) 3214 { 3215 struct tcptw *tw; 3216 int thflags; 3217 tcp_seq seq; 3218 #ifdef INET6 3219 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3220 #else 3221 const int isipv6 = 0; 3222 #endif 3223 3224 /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */ 3225 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3226 INP_LOCK_ASSERT(inp); 3227 3228 /* 3229 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 3230 * still present. This is undesirable, but temporarily necessary 3231 * until we work out how to handle inpcb's who's timewait state has 3232 * been removed. 3233 */ 3234 tw = intotw(inp); 3235 if (tw == NULL) 3236 goto drop; 3237 3238 thflags = th->th_flags; 3239 3240 /* 3241 * NOTE: for FIN_WAIT_2 (to be added later), 3242 * must validate sequence number before accepting RST 3243 */ 3244 3245 /* 3246 * If the segment contains RST: 3247 * Drop the segment - see Stevens, vol. 2, p. 964 and 3248 * RFC 1337. 3249 */ 3250 if (thflags & TH_RST) 3251 goto drop; 3252 3253 #if 0 3254 /* PAWS not needed at the moment */ 3255 /* 3256 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3257 * and it's less than ts_recent, drop it. 3258 */ 3259 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3260 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3261 if ((thflags & TH_ACK) == 0) 3262 goto drop; 3263 goto ack; 3264 } 3265 /* 3266 * ts_recent is never updated because we never accept new segments. 3267 */ 3268 #endif 3269 3270 /* 3271 * If a new connection request is received 3272 * while in TIME_WAIT, drop the old connection 3273 * and start over if the sequence numbers 3274 * are above the previous ones. 3275 */ 3276 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3277 tcp_twclose(tw, 0); 3278 return (1); 3279 } 3280 3281 /* 3282 * Drop the the segment if it does not contain an ACK. 3283 */ 3284 if ((thflags & TH_ACK) == 0) 3285 goto drop; 3286 3287 /* 3288 * Reset the 2MSL timer if this is a duplicate FIN. 3289 */ 3290 if (thflags & TH_FIN) { 3291 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3292 if (seq + 1 == tw->rcv_nxt) 3293 tcp_timer_2msl_reset(tw, 1); 3294 } 3295 3296 /* 3297 * Acknowledge the segment if it has data or is not a duplicate ACK. 3298 */ 3299 if (thflags != TH_ACK || tlen != 0 || 3300 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3301 tcp_twrespond(tw, TH_ACK); 3302 goto drop; 3303 3304 /* 3305 * Generate a RST, dropping incoming segment. 3306 * Make ACK acceptable to originator of segment. 3307 * Don't bother to respond if destination was broadcast/multicast. 3308 */ 3309 if (m->m_flags & (M_BCAST|M_MCAST)) 3310 goto drop; 3311 if (isipv6) { 3312 struct ip6_hdr *ip6; 3313 3314 /* IPv6 anycast check is done at tcp6_input() */ 3315 ip6 = mtod(m, struct ip6_hdr *); 3316 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3317 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3318 goto drop; 3319 } else { 3320 struct ip *ip; 3321 3322 ip = mtod(m, struct ip *); 3323 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3324 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3325 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3326 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3327 goto drop; 3328 } 3329 if (thflags & TH_ACK) { 3330 tcp_respond(NULL, 3331 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3332 } else { 3333 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3334 tcp_respond(NULL, 3335 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3336 } 3337 INP_UNLOCK(inp); 3338 return (0); 3339 3340 drop: 3341 INP_UNLOCK(inp); 3342 m_freem(m); 3343 return (0); 3344 } 3345