1 /* 2 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2002, 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved. 36 * 37 * License terms: all terms for the DragonFly license above plus the following: 38 * 39 * 4. All advertising materials mentioning features or use of this software 40 * must display the following acknowledgement: 41 * 42 * This product includes software developed by Jeffrey M. Hsu 43 * for the DragonFly Project. 44 * 45 * This requirement may be waived with permission from Jeffrey Hsu. 46 * This requirement will sunset and may be removed on July 8 2005, 47 * after which the standard DragonFly license (as shown above) will 48 * apply. 49 */ 50 51 /* 52 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 53 * The Regents of the University of California. All rights reserved. 54 * 55 * Redistribution and use in source and binary forms, with or without 56 * modification, are permitted provided that the following conditions 57 * are met: 58 * 1. Redistributions of source code must retain the above copyright 59 * notice, this list of conditions and the following disclaimer. 60 * 2. Redistributions in binary form must reproduce the above copyright 61 * notice, this list of conditions and the following disclaimer in the 62 * documentation and/or other materials provided with the distribution. 63 * 3. All advertising materials mentioning features or use of this software 64 * must display the following acknowledgement: 65 * This product includes software developed by the University of 66 * California, Berkeley and its contributors. 67 * 4. Neither the name of the University nor the names of its contributors 68 * may be used to endorse or promote products derived from this software 69 * without specific prior written permission. 70 * 71 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 72 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 73 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 74 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 75 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 76 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 77 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 78 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 79 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 80 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 81 * SUCH DAMAGE. 82 * 83 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 84 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $ 85 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.59 2005/04/18 22:41:23 hsu Exp $ 86 */ 87 88 #include "opt_ipfw.h" /* for ipfw_fwd */ 89 #include "opt_inet6.h" 90 #include "opt_ipsec.h" 91 #include "opt_tcpdebug.h" 92 #include "opt_tcp_input.h" 93 94 #include <sys/param.h> 95 #include <sys/systm.h> 96 #include <sys/kernel.h> 97 #include <sys/sysctl.h> 98 #include <sys/malloc.h> 99 #include <sys/mbuf.h> 100 #include <sys/proc.h> /* for proc0 declaration */ 101 #include <sys/protosw.h> 102 #include <sys/socket.h> 103 #include <sys/socketvar.h> 104 #include <sys/syslog.h> 105 #include <sys/in_cksum.h> 106 107 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 108 #include <machine/stdarg.h> 109 110 #include <net/if.h> 111 #include <net/route.h> 112 113 #include <netinet/in.h> 114 #include <netinet/in_systm.h> 115 #include <netinet/ip.h> 116 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 117 #include <netinet/in_var.h> 118 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 119 #include <netinet/in_pcb.h> 120 #include <netinet/ip_var.h> 121 #include <netinet/ip6.h> 122 #include <netinet/icmp6.h> 123 #include <netinet6/nd6.h> 124 #include <netinet6/ip6_var.h> 125 #include <netinet6/in6_pcb.h> 126 #include <netinet/tcp.h> 127 #include <netinet/tcp_fsm.h> 128 #include <netinet/tcp_seq.h> 129 #include <netinet/tcp_timer.h> 130 #include <netinet/tcp_var.h> 131 #include <netinet6/tcp6_var.h> 132 #include <netinet/tcpip.h> 133 134 #ifdef TCPDEBUG 135 #include <netinet/tcp_debug.h> 136 137 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 138 struct tcphdr tcp_savetcp; 139 #endif 140 141 #ifdef FAST_IPSEC 142 #include <netproto/ipsec/ipsec.h> 143 #include <netproto/ipsec/ipsec6.h> 144 #endif 145 146 #ifdef IPSEC 147 #include <netinet6/ipsec.h> 148 #include <netinet6/ipsec6.h> 149 #include <netproto/key/key.h> 150 #endif 151 152 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 153 154 tcp_cc tcp_ccgen; 155 static int log_in_vain = 0; 156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 157 &log_in_vain, 0, "Log all incoming TCP connections"); 158 159 static int blackhole = 0; 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 161 &blackhole, 0, "Do not send RST when dropping refused connections"); 162 163 int tcp_delack_enabled = 1; 164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 165 &tcp_delack_enabled, 0, 166 "Delay ACK to try and piggyback it onto a data packet"); 167 168 #ifdef TCP_DROP_SYNFIN 169 static int drop_synfin = 0; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 171 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 172 #endif 173 174 static int tcp_do_limitedtransmit = 1; 175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW, 176 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)"); 177 178 static int tcp_do_early_retransmit = 1; 179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW, 180 &tcp_do_early_retransmit, 0, "Early retransmit"); 181 182 static int tcp_aggregate_acks = 1; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW, 184 &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack"); 185 186 int tcp_do_rfc3390 = 1; 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 188 &tcp_do_rfc3390, 0, 189 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 190 191 static int tcp_do_eifel_detect = 1; 192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW, 193 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)"); 194 195 /* 196 * Define as tunable for easy testing with SACK on and off. 197 * Warning: do not change setting in the middle of an existing active TCP flow, 198 * else strange things might happen to that flow. 199 */ 200 int tcp_do_sack = 1; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 202 &tcp_do_sack, 0, "Enable SACK Algorithms"); 203 204 int tcp_do_smartsack = 1; 205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW, 206 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms"); 207 208 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 209 "TCP Segment Reassembly Queue"); 210 211 int tcp_reass_maxseg = 0; 212 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD, 213 &tcp_reass_maxseg, 0, 214 "Global maximum number of TCP Segments in Reassembly Queue"); 215 216 int tcp_reass_qsize = 0; 217 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 218 &tcp_reass_qsize, 0, 219 "Global number of TCP Segments currently in Reassembly Queue"); 220 221 static int tcp_reass_overflows = 0; 222 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 223 &tcp_reass_overflows, 0, 224 "Global number of TCP Segment Reassembly Queue Overflows"); 225 226 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t); 227 static void tcp_pulloutofband(struct socket *, 228 struct tcphdr *, struct mbuf *, int); 229 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 230 struct mbuf *); 231 static void tcp_xmit_timer(struct tcpcb *, int); 232 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int); 233 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *); 234 235 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 236 #ifdef INET6 237 #define ND6_HINT(tp) \ 238 do { \ 239 if ((tp) && (tp)->t_inpcb && \ 240 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \ 241 (tp)->t_inpcb->in6p_route.ro_rt) \ 242 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 243 } while (0) 244 #else 245 #define ND6_HINT(tp) 246 #endif 247 248 /* 249 * Indicate whether this ack should be delayed. We can delay the ack if 250 * - delayed acks are enabled and 251 * - there is no delayed ack timer in progress and 252 * - our last ack wasn't a 0-sized window. We never want to delay 253 * the ack that opens up a 0-sized window. 254 */ 255 #define DELAY_ACK(tp) \ 256 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 257 !(tp->t_flags & TF_RXWIN0SENT)) 258 259 #define acceptable_window_update(tp, th, tiwin) \ 260 (SEQ_LT(tp->snd_wl1, th->th_seq) || \ 261 (tp->snd_wl1 == th->th_seq && \ 262 (SEQ_LT(tp->snd_wl2, th->th_ack) || \ 263 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) 264 265 static int 266 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m) 267 { 268 struct tseg_qent *q; 269 struct tseg_qent *p = NULL; 270 struct tseg_qent *te; 271 struct socket *so = tp->t_inpcb->inp_socket; 272 int flags; 273 274 /* 275 * Call with th == NULL after become established to 276 * force pre-ESTABLISHED data up to user socket. 277 */ 278 if (th == NULL) 279 goto present; 280 281 /* 282 * Limit the number of segments in the reassembly queue to prevent 283 * holding on to too many segments (and thus running out of mbufs). 284 * Make sure to let the missing segment through which caused this 285 * queue. Always keep one global queue entry spare to be able to 286 * process the missing segment. 287 */ 288 if (th->th_seq != tp->rcv_nxt && 289 tcp_reass_qsize + 1 >= tcp_reass_maxseg) { 290 tcp_reass_overflows++; 291 tcpstat.tcps_rcvmemdrop++; 292 m_freem(m); 293 /* no SACK block to report */ 294 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 295 return (0); 296 } 297 298 /* Allocate a new queue entry. */ 299 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ, 300 M_INTWAIT | M_NULLOK); 301 if (te == NULL) { 302 tcpstat.tcps_rcvmemdrop++; 303 m_freem(m); 304 /* no SACK block to report */ 305 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 306 return (0); 307 } 308 tcp_reass_qsize++; 309 310 /* 311 * Find a segment which begins after this one does. 312 */ 313 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 314 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 315 break; 316 p = q; 317 } 318 319 /* 320 * If there is a preceding segment, it may provide some of 321 * our data already. If so, drop the data from the incoming 322 * segment. If it provides all of our data, drop us. 323 */ 324 if (p != NULL) { 325 tcp_seq_diff_t i; 326 327 /* conversion to int (in i) handles seq wraparound */ 328 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 329 if (i > 0) { /* overlaps preceding segment */ 330 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG); 331 /* enclosing block starts w/ preceding segment */ 332 tp->encloseblk.rblk_start = p->tqe_th->th_seq; 333 if (i >= *tlenp) { 334 /* preceding encloses incoming segment */ 335 tp->encloseblk.rblk_end = p->tqe_th->th_seq + 336 p->tqe_len; 337 tcpstat.tcps_rcvduppack++; 338 tcpstat.tcps_rcvdupbyte += *tlenp; 339 m_freem(m); 340 free(te, M_TSEGQ); 341 tcp_reass_qsize--; 342 /* 343 * Try to present any queued data 344 * at the left window edge to the user. 345 * This is needed after the 3-WHS 346 * completes. 347 */ 348 goto present; /* ??? */ 349 } 350 m_adj(m, i); 351 *tlenp -= i; 352 th->th_seq += i; 353 /* incoming segment end is enclosing block end */ 354 tp->encloseblk.rblk_end = th->th_seq + *tlenp + 355 ((th->th_flags & TH_FIN) != 0); 356 /* trim end of reported D-SACK block */ 357 tp->reportblk.rblk_end = th->th_seq; 358 } 359 } 360 tcpstat.tcps_rcvoopack++; 361 tcpstat.tcps_rcvoobyte += *tlenp; 362 363 /* 364 * While we overlap succeeding segments trim them or, 365 * if they are completely covered, dequeue them. 366 */ 367 while (q) { 368 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 369 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len; 370 struct tseg_qent *nq; 371 372 if (i <= 0) 373 break; 374 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */ 375 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG); 376 tp->encloseblk = tp->reportblk; 377 /* report trailing duplicate D-SACK segment */ 378 tp->reportblk.rblk_start = q->tqe_th->th_seq; 379 } 380 if ((tp->t_flags & TF_ENCLOSESEG) && 381 SEQ_GT(qend, tp->encloseblk.rblk_end)) { 382 /* extend enclosing block if one exists */ 383 tp->encloseblk.rblk_end = qend; 384 } 385 if (i < q->tqe_len) { 386 q->tqe_th->th_seq += i; 387 q->tqe_len -= i; 388 m_adj(q->tqe_m, i); 389 break; 390 } 391 392 nq = LIST_NEXT(q, tqe_q); 393 LIST_REMOVE(q, tqe_q); 394 m_freem(q->tqe_m); 395 free(q, M_TSEGQ); 396 tcp_reass_qsize--; 397 q = nq; 398 } 399 400 /* Insert the new segment queue entry into place. */ 401 te->tqe_m = m; 402 te->tqe_th = th; 403 te->tqe_len = *tlenp; 404 405 /* check if can coalesce with following segment */ 406 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) { 407 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len; 408 409 te->tqe_len += q->tqe_len; 410 if (q->tqe_th->th_flags & TH_FIN) 411 te->tqe_th->th_flags |= TH_FIN; 412 m_cat(te->tqe_m, q->tqe_m); 413 tp->encloseblk.rblk_end = tend; 414 /* 415 * When not reporting a duplicate segment, use 416 * the larger enclosing block as the SACK block. 417 */ 418 if (!(tp->t_flags & TF_DUPSEG)) 419 tp->reportblk.rblk_end = tend; 420 LIST_REMOVE(q, tqe_q); 421 free(q, M_TSEGQ); 422 tcp_reass_qsize--; 423 } 424 425 if (p == NULL) { 426 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 427 } else { 428 /* check if can coalesce with preceding segment */ 429 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) { 430 p->tqe_len += te->tqe_len; 431 m_cat(p->tqe_m, te->tqe_m); 432 tp->encloseblk.rblk_start = p->tqe_th->th_seq; 433 /* 434 * When not reporting a duplicate segment, use 435 * the larger enclosing block as the SACK block. 436 */ 437 if (!(tp->t_flags & TF_DUPSEG)) 438 tp->reportblk.rblk_start = p->tqe_th->th_seq; 439 free(te, M_TSEGQ); 440 tcp_reass_qsize--; 441 } else 442 LIST_INSERT_AFTER(p, te, tqe_q); 443 } 444 445 present: 446 /* 447 * Present data to user, advancing rcv_nxt through 448 * completed sequence space. 449 */ 450 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 451 return (0); 452 q = LIST_FIRST(&tp->t_segq); 453 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt) 454 return (0); 455 tp->rcv_nxt += q->tqe_len; 456 if (!(tp->t_flags & TF_DUPSEG)) { 457 /* no SACK block to report since ACK advanced */ 458 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 459 } 460 /* no enclosing block to report since ACK advanced */ 461 tp->t_flags &= ~TF_ENCLOSESEG; 462 flags = q->tqe_th->th_flags & TH_FIN; 463 LIST_REMOVE(q, tqe_q); 464 KASSERT(LIST_EMPTY(&tp->t_segq) || 465 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt, 466 ("segment not coalesced")); 467 if (so->so_state & SS_CANTRCVMORE) 468 m_freem(q->tqe_m); 469 else 470 sbappendstream(&so->so_rcv, q->tqe_m); 471 free(q, M_TSEGQ); 472 tcp_reass_qsize--; 473 ND6_HINT(tp); 474 sorwakeup(so); 475 return (flags); 476 } 477 478 /* 479 * TCP input routine, follows pages 65-76 of the 480 * protocol specification dated September, 1981 very closely. 481 */ 482 #ifdef INET6 483 int 484 tcp6_input(struct mbuf **mp, int *offp, int proto) 485 { 486 struct mbuf *m = *mp; 487 struct in6_ifaddr *ia6; 488 489 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 490 491 /* 492 * draft-itojun-ipv6-tcp-to-anycast 493 * better place to put this in? 494 */ 495 ia6 = ip6_getdstifaddr(m); 496 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 497 struct ip6_hdr *ip6; 498 499 ip6 = mtod(m, struct ip6_hdr *); 500 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 501 offsetof(struct ip6_hdr, ip6_dst)); 502 return (IPPROTO_DONE); 503 } 504 505 tcp_input(m, *offp, proto); 506 return (IPPROTO_DONE); 507 } 508 #endif 509 510 void 511 tcp_input(struct mbuf *m, ...) 512 { 513 __va_list ap; 514 int off0, proto; 515 struct tcphdr *th; 516 struct ip *ip = NULL; 517 struct ipovly *ipov; 518 struct inpcb *inp = NULL; 519 u_char *optp = NULL; 520 int optlen = 0; 521 int len, tlen, off; 522 int drop_hdrlen; 523 struct tcpcb *tp = NULL; 524 int thflags; 525 struct socket *so = 0; 526 int todrop, acked; 527 boolean_t ourfinisacked, needoutput = FALSE; 528 u_long tiwin; 529 int recvwin; 530 struct tcpopt to; /* options in this segment */ 531 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 532 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 533 struct sockaddr_in *next_hop = NULL; 534 int rstreason; /* For badport_bandlim accounting purposes */ 535 int cpu; 536 struct ip6_hdr *ip6 = NULL; 537 #ifdef INET6 538 boolean_t isipv6; 539 #else 540 const boolean_t isipv6 = FALSE; 541 #endif 542 #ifdef TCPDEBUG 543 short ostate = 0; 544 #endif 545 546 __va_start(ap, m); 547 off0 = __va_arg(ap, int); 548 proto = __va_arg(ap, int); 549 __va_end(ap); 550 551 tcpstat.tcps_rcvtotal++; 552 553 /* Grab info from and strip MT_TAG mbufs prepended to the chain. */ 554 while (m->m_type == MT_TAG) { 555 if (m->_m_tag_id == PACKET_TAG_IPFORWARD) 556 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data; 557 m = m->m_next; 558 } 559 560 #ifdef INET6 561 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE; 562 #endif 563 564 if (isipv6) { 565 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 566 ip6 = mtod(m, struct ip6_hdr *); 567 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0; 568 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 569 tcpstat.tcps_rcvbadsum++; 570 goto drop; 571 } 572 th = (struct tcphdr *)((caddr_t)ip6 + off0); 573 574 /* 575 * Be proactive about unspecified IPv6 address in source. 576 * As we use all-zero to indicate unbounded/unconnected pcb, 577 * unspecified IPv6 address can be used to confuse us. 578 * 579 * Note that packets with unspecified IPv6 destination is 580 * already dropped in ip6_input. 581 */ 582 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 583 /* XXX stat */ 584 goto drop; 585 } 586 } else { 587 /* 588 * Get IP and TCP header together in first mbuf. 589 * Note: IP leaves IP header in first mbuf. 590 */ 591 if (off0 > sizeof(struct ip)) { 592 ip_stripoptions(m); 593 off0 = sizeof(struct ip); 594 } 595 /* already checked and pulled up in ip_demux() */ 596 KASSERT(m->m_len >= sizeof(struct tcpiphdr), 597 ("TCP header not in one mbuf: m->m_len %d", m->m_len)); 598 ip = mtod(m, struct ip *); 599 ipov = (struct ipovly *)ip; 600 th = (struct tcphdr *)((caddr_t)ip + off0); 601 tlen = ip->ip_len; 602 603 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 604 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 605 th->th_sum = m->m_pkthdr.csum_data; 606 else 607 th->th_sum = in_pseudo(ip->ip_src.s_addr, 608 ip->ip_dst.s_addr, 609 htonl(m->m_pkthdr.csum_data + 610 ip->ip_len + 611 IPPROTO_TCP)); 612 th->th_sum ^= 0xffff; 613 } else { 614 /* 615 * Checksum extended TCP header and data. 616 */ 617 len = sizeof(struct ip) + tlen; 618 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 619 ipov->ih_len = (u_short)tlen; 620 ipov->ih_len = htons(ipov->ih_len); 621 th->th_sum = in_cksum(m, len); 622 } 623 if (th->th_sum) { 624 tcpstat.tcps_rcvbadsum++; 625 goto drop; 626 } 627 #ifdef INET6 628 /* Re-initialization for later version check */ 629 ip->ip_v = IPVERSION; 630 #endif 631 } 632 633 /* 634 * Check that TCP offset makes sense, 635 * pull out TCP options and adjust length. XXX 636 */ 637 off = th->th_off << 2; 638 /* already checked and pulled up in ip_demux() */ 639 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen, 640 ("bad TCP data offset %d (tlen %d)", off, tlen)); 641 tlen -= off; /* tlen is used instead of ti->ti_len */ 642 if (off > sizeof(struct tcphdr)) { 643 if (isipv6) { 644 IP6_EXTHDR_CHECK(m, off0, off, ); 645 ip6 = mtod(m, struct ip6_hdr *); 646 th = (struct tcphdr *)((caddr_t)ip6 + off0); 647 } else { 648 /* already pulled up in ip_demux() */ 649 KASSERT(m->m_len >= sizeof(struct ip) + off, 650 ("TCP header and options not in one mbuf: " 651 "m_len %d, off %d", m->m_len, off)); 652 } 653 optlen = off - sizeof(struct tcphdr); 654 optp = (u_char *)(th + 1); 655 } 656 thflags = th->th_flags; 657 658 #ifdef TCP_DROP_SYNFIN 659 /* 660 * If the drop_synfin option is enabled, drop all packets with 661 * both the SYN and FIN bits set. This prevents e.g. nmap from 662 * identifying the TCP/IP stack. 663 * 664 * This is a violation of the TCP specification. 665 */ 666 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN)) 667 goto drop; 668 #endif 669 670 /* 671 * Convert TCP protocol specific fields to host format. 672 */ 673 th->th_seq = ntohl(th->th_seq); 674 th->th_ack = ntohl(th->th_ack); 675 th->th_win = ntohs(th->th_win); 676 th->th_urp = ntohs(th->th_urp); 677 678 /* 679 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 680 * until after ip6_savecontrol() is called and before other functions 681 * which don't want those proto headers. 682 * Because ip6_savecontrol() is going to parse the mbuf to 683 * search for data to be passed up to user-land, it wants mbuf 684 * parameters to be unchanged. 685 * XXX: the call of ip6_savecontrol() has been obsoleted based on 686 * latest version of the advanced API (20020110). 687 */ 688 drop_hdrlen = off0 + off; 689 690 /* 691 * Locate pcb for segment. 692 */ 693 findpcb: 694 /* IPFIREWALL_FORWARD section */ 695 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */ 696 /* 697 * Transparently forwarded. Pretend to be the destination. 698 * already got one like this? 699 */ 700 cpu = mycpu->gd_cpuid; 701 inp = in_pcblookup_hash(&tcbinfo[cpu], 702 ip->ip_src, th->th_sport, 703 ip->ip_dst, th->th_dport, 704 0, m->m_pkthdr.rcvif); 705 if (!inp) { 706 /* 707 * It's new. Try to find the ambushing socket. 708 */ 709 710 /* 711 * The rest of the ipfw code stores the port in 712 * host order. XXX 713 * (The IP address is still in network order.) 714 */ 715 in_port_t dport = next_hop->sin_port ? 716 htons(next_hop->sin_port) : 717 th->th_dport; 718 719 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport, 720 next_hop->sin_addr.s_addr, dport); 721 inp = in_pcblookup_hash(&tcbinfo[cpu], 722 ip->ip_src, th->th_sport, 723 next_hop->sin_addr, dport, 724 1, m->m_pkthdr.rcvif); 725 } 726 } else { 727 if (isipv6) { 728 inp = in6_pcblookup_hash(&tcbinfo[0], 729 &ip6->ip6_src, th->th_sport, 730 &ip6->ip6_dst, th->th_dport, 731 1, m->m_pkthdr.rcvif); 732 } else { 733 cpu = mycpu->gd_cpuid; 734 inp = in_pcblookup_hash(&tcbinfo[cpu], 735 ip->ip_src, th->th_sport, 736 ip->ip_dst, th->th_dport, 737 1, m->m_pkthdr.rcvif); 738 } 739 } 740 741 /* 742 * If the state is CLOSED (i.e., TCB does not exist) then 743 * all data in the incoming segment is discarded. 744 * If the TCB exists but is in CLOSED state, it is embryonic, 745 * but should either do a listen or a connect soon. 746 */ 747 if (inp == NULL) { 748 if (log_in_vain) { 749 #ifdef INET6 750 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 751 #else 752 char dbuf[sizeof "aaa.bbb.ccc.ddd"]; 753 char sbuf[sizeof "aaa.bbb.ccc.ddd"]; 754 #endif 755 if (isipv6) { 756 strcpy(dbuf, "["); 757 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 758 strcat(dbuf, "]"); 759 strcpy(sbuf, "["); 760 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 761 strcat(sbuf, "]"); 762 } else { 763 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 764 strcpy(sbuf, inet_ntoa(ip->ip_src)); 765 } 766 switch (log_in_vain) { 767 case 1: 768 if (!(thflags & TH_SYN)) 769 break; 770 case 2: 771 log(LOG_INFO, 772 "Connection attempt to TCP %s:%d " 773 "from %s:%d flags:0x%02x\n", 774 dbuf, ntohs(th->th_dport), sbuf, 775 ntohs(th->th_sport), thflags); 776 break; 777 default: 778 break; 779 } 780 } 781 if (blackhole) { 782 switch (blackhole) { 783 case 1: 784 if (thflags & TH_SYN) 785 goto drop; 786 break; 787 case 2: 788 goto drop; 789 default: 790 goto drop; 791 } 792 } 793 rstreason = BANDLIM_RST_CLOSEDPORT; 794 goto dropwithreset; 795 } 796 797 #ifdef IPSEC 798 if (isipv6) { 799 if (ipsec6_in_reject_so(m, inp->inp_socket)) { 800 ipsec6stat.in_polvio++; 801 goto drop; 802 } 803 } else { 804 if (ipsec4_in_reject_so(m, inp->inp_socket)) { 805 ipsecstat.in_polvio++; 806 goto drop; 807 } 808 } 809 #endif 810 #ifdef FAST_IPSEC 811 if (isipv6) { 812 if (ipsec6_in_reject(m, inp)) 813 goto drop; 814 } else { 815 if (ipsec4_in_reject(m, inp)) 816 goto drop; 817 } 818 #endif 819 820 tp = intotcpcb(inp); 821 if (tp == NULL) { 822 rstreason = BANDLIM_RST_CLOSEDPORT; 823 goto dropwithreset; 824 } 825 if (tp->t_state <= TCPS_CLOSED) 826 goto drop; 827 828 /* Unscale the window into a 32-bit value. */ 829 if (!(thflags & TH_SYN)) 830 tiwin = th->th_win << tp->snd_scale; 831 else 832 tiwin = th->th_win; 833 834 so = inp->inp_socket; 835 836 #ifdef TCPDEBUG 837 if (so->so_options & SO_DEBUG) { 838 ostate = tp->t_state; 839 if (isipv6) 840 bcopy(ip6, tcp_saveipgen, sizeof(*ip6)); 841 else 842 bcopy(ip, tcp_saveipgen, sizeof(*ip)); 843 tcp_savetcp = *th; 844 } 845 #endif 846 847 bzero(&to, sizeof to); 848 849 if (so->so_options & SO_ACCEPTCONN) { 850 struct in_conninfo inc; 851 852 #ifdef INET6 853 inc.inc_isipv6 = (isipv6 == TRUE); 854 #endif 855 if (isipv6) { 856 inc.inc6_faddr = ip6->ip6_src; 857 inc.inc6_laddr = ip6->ip6_dst; 858 inc.inc6_route.ro_rt = NULL; /* XXX */ 859 } else { 860 inc.inc_faddr = ip->ip_src; 861 inc.inc_laddr = ip->ip_dst; 862 inc.inc_route.ro_rt = NULL; /* XXX */ 863 } 864 inc.inc_fport = th->th_sport; 865 inc.inc_lport = th->th_dport; 866 867 /* 868 * If the state is LISTEN then ignore segment if it contains 869 * a RST. If the segment contains an ACK then it is bad and 870 * send a RST. If it does not contain a SYN then it is not 871 * interesting; drop it. 872 * 873 * If the state is SYN_RECEIVED (syncache) and seg contains 874 * an ACK, but not for our SYN/ACK, send a RST. If the seg 875 * contains a RST, check the sequence number to see if it 876 * is a valid reset segment. 877 */ 878 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) { 879 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) { 880 if (!syncache_expand(&inc, th, &so, m)) { 881 /* 882 * No syncache entry, or ACK was not 883 * for our SYN/ACK. Send a RST. 884 */ 885 tcpstat.tcps_badsyn++; 886 rstreason = BANDLIM_RST_OPENPORT; 887 goto dropwithreset; 888 } 889 if (so == NULL) 890 /* 891 * Could not complete 3-way handshake, 892 * connection is being closed down, and 893 * syncache will free mbuf. 894 */ 895 return; 896 /* 897 * Socket is created in state SYN_RECEIVED. 898 * Continue processing segment. 899 */ 900 inp = so->so_pcb; 901 tp = intotcpcb(inp); 902 /* 903 * This is what would have happened in 904 * tcp_output() when the SYN,ACK was sent. 905 */ 906 tp->snd_up = tp->snd_una; 907 tp->snd_max = tp->snd_nxt = tp->iss + 1; 908 tp->last_ack_sent = tp->rcv_nxt; 909 /* 910 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 911 * until the _second_ ACK is received: 912 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 913 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 914 * move to ESTAB, set snd_wnd to tiwin. 915 */ 916 tp->snd_wnd = tiwin; /* unscaled */ 917 goto after_listen; 918 } 919 if (thflags & TH_RST) { 920 syncache_chkrst(&inc, th); 921 goto drop; 922 } 923 if (thflags & TH_ACK) { 924 syncache_badack(&inc); 925 tcpstat.tcps_badsyn++; 926 rstreason = BANDLIM_RST_OPENPORT; 927 goto dropwithreset; 928 } 929 goto drop; 930 } 931 932 /* 933 * Segment's flags are (SYN) or (SYN | FIN). 934 */ 935 #ifdef INET6 936 /* 937 * If deprecated address is forbidden, 938 * we do not accept SYN to deprecated interface 939 * address to prevent any new inbound connection from 940 * getting established. 941 * When we do not accept SYN, we send a TCP RST, 942 * with deprecated source address (instead of dropping 943 * it). We compromise it as it is much better for peer 944 * to send a RST, and RST will be the final packet 945 * for the exchange. 946 * 947 * If we do not forbid deprecated addresses, we accept 948 * the SYN packet. RFC2462 does not suggest dropping 949 * SYN in this case. 950 * If we decipher RFC2462 5.5.4, it says like this: 951 * 1. use of deprecated addr with existing 952 * communication is okay - "SHOULD continue to be 953 * used" 954 * 2. use of it with new communication: 955 * (2a) "SHOULD NOT be used if alternate address 956 * with sufficient scope is available" 957 * (2b) nothing mentioned otherwise. 958 * Here we fall into (2b) case as we have no choice in 959 * our source address selection - we must obey the peer. 960 * 961 * The wording in RFC2462 is confusing, and there are 962 * multiple description text for deprecated address 963 * handling - worse, they are not exactly the same. 964 * I believe 5.5.4 is the best one, so we follow 5.5.4. 965 */ 966 if (isipv6 && !ip6_use_deprecated) { 967 struct in6_ifaddr *ia6; 968 969 if ((ia6 = ip6_getdstifaddr(m)) && 970 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 971 tp = NULL; 972 rstreason = BANDLIM_RST_OPENPORT; 973 goto dropwithreset; 974 } 975 } 976 #endif 977 /* 978 * If it is from this socket, drop it, it must be forged. 979 * Don't bother responding if the destination was a broadcast. 980 */ 981 if (th->th_dport == th->th_sport) { 982 if (isipv6) { 983 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 984 &ip6->ip6_src)) 985 goto drop; 986 } else { 987 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 988 goto drop; 989 } 990 } 991 /* 992 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 993 * 994 * Note that it is quite possible to receive unicast 995 * link-layer packets with a broadcast IP address. Use 996 * in_broadcast() to find them. 997 */ 998 if (m->m_flags & (M_BCAST | M_MCAST)) 999 goto drop; 1000 if (isipv6) { 1001 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1002 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 1003 goto drop; 1004 } else { 1005 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1006 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1007 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1008 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1009 goto drop; 1010 } 1011 /* 1012 * SYN appears to be valid; create compressed TCP state 1013 * for syncache, or perform t/tcp connection. 1014 */ 1015 if (so->so_qlen <= so->so_qlimit) { 1016 tcp_dooptions(&to, optp, optlen, TRUE); 1017 if (!syncache_add(&inc, &to, th, &so, m)) 1018 goto drop; 1019 if (so == NULL) 1020 /* 1021 * Entry added to syncache, mbuf used to 1022 * send SYN,ACK packet. 1023 */ 1024 return; 1025 /* 1026 * Segment passed TAO tests. 1027 */ 1028 inp = so->so_pcb; 1029 tp = intotcpcb(inp); 1030 tp->snd_wnd = tiwin; 1031 tp->t_starttime = ticks; 1032 tp->t_state = TCPS_ESTABLISHED; 1033 1034 /* 1035 * If there is a FIN, or if there is data and the 1036 * connection is local, then delay SYN,ACK(SYN) in 1037 * the hope of piggy-backing it on a response 1038 * segment. Otherwise must send ACK now in case 1039 * the other side is slow starting. 1040 */ 1041 if (DELAY_ACK(tp) && 1042 ((thflags & TH_FIN) || 1043 (tlen != 0 && 1044 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 1045 (!isipv6 && in_localaddr(inp->inp_faddr)))))) { 1046 callout_reset(tp->tt_delack, tcp_delacktime, 1047 tcp_timer_delack, tp); 1048 tp->t_flags |= TF_NEEDSYN; 1049 } else 1050 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1051 1052 tcpstat.tcps_connects++; 1053 soisconnected(so); 1054 goto trimthenstep6; 1055 } 1056 goto drop; 1057 } 1058 after_listen: 1059 1060 /* should not happen - syncache should pick up these connections */ 1061 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state")); 1062 1063 /* 1064 * Segment received on connection. 1065 * Reset idle time and keep-alive timer. 1066 */ 1067 tp->t_rcvtime = ticks; 1068 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1069 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1070 1071 /* 1072 * Process options. 1073 * XXX this is tradtitional behavior, may need to be cleaned up. 1074 */ 1075 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0); 1076 if (thflags & TH_SYN) { 1077 if (to.to_flags & TOF_SCALE) { 1078 tp->t_flags |= TF_RCVD_SCALE; 1079 tp->requested_s_scale = to.to_requested_s_scale; 1080 } 1081 if (to.to_flags & TOF_TS) { 1082 tp->t_flags |= TF_RCVD_TSTMP; 1083 tp->ts_recent = to.to_tsval; 1084 tp->ts_recent_age = ticks; 1085 } 1086 if (to.to_flags & (TOF_CC | TOF_CCNEW)) 1087 tp->t_flags |= TF_RCVD_CC; 1088 if (to.to_flags & TOF_MSS) 1089 tcp_mss(tp, to.to_mss); 1090 /* 1091 * Only set the TF_SACK_PERMITTED per-connection flag 1092 * if we got a SACK_PERMITTED option from the other side 1093 * and the global tcp_do_sack variable is true. 1094 */ 1095 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED)) 1096 tp->t_flags |= TF_SACK_PERMITTED; 1097 } 1098 1099 /* 1100 * Header prediction: check for the two common cases 1101 * of a uni-directional data xfer. If the packet has 1102 * no control flags, is in-sequence, the window didn't 1103 * change and we're not retransmitting, it's a 1104 * candidate. If the length is zero and the ack moved 1105 * forward, we're the sender side of the xfer. Just 1106 * free the data acked & wake any higher level process 1107 * that was blocked waiting for space. If the length 1108 * is non-zero and the ack didn't move, we're the 1109 * receiver side. If we're getting packets in-order 1110 * (the reassembly queue is empty), add the data to 1111 * the socket buffer and note that we need a delayed ack. 1112 * Make sure that the hidden state-flags are also off. 1113 * Since we check for TCPS_ESTABLISHED above, it can only 1114 * be TH_NEEDSYN. 1115 */ 1116 if (tp->t_state == TCPS_ESTABLISHED && 1117 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1118 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) && 1119 (!(to.to_flags & TOF_TS) || 1120 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1121 /* 1122 * Using the CC option is compulsory if once started: 1123 * the segment is OK if no T/TCP was negotiated or 1124 * if the segment has a CC option equal to CCrecv 1125 */ 1126 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 1127 ((to.to_flags & TOF_CC) && to.to_cc == tp->cc_recv)) && 1128 th->th_seq == tp->rcv_nxt && 1129 tp->snd_nxt == tp->snd_max) { 1130 1131 /* 1132 * If last ACK falls within this segment's sequence numbers, 1133 * record the timestamp. 1134 * NOTE that the test is modified according to the latest 1135 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1136 */ 1137 if ((to.to_flags & TOF_TS) && 1138 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1139 tp->ts_recent_age = ticks; 1140 tp->ts_recent = to.to_tsval; 1141 } 1142 1143 if (tlen == 0) { 1144 if (SEQ_GT(th->th_ack, tp->snd_una) && 1145 SEQ_LEQ(th->th_ack, tp->snd_max) && 1146 tp->snd_cwnd >= tp->snd_wnd && 1147 !IN_FASTRECOVERY(tp)) { 1148 /* 1149 * This is a pure ack for outstanding data. 1150 */ 1151 ++tcpstat.tcps_predack; 1152 /* 1153 * "bad retransmit" recovery 1154 * 1155 * If Eifel detection applies, then 1156 * it is deterministic, so use it 1157 * unconditionally over the old heuristic. 1158 * Otherwise, fall back to the old heuristic. 1159 */ 1160 if (tcp_do_eifel_detect && 1161 (to.to_flags & TOF_TS) && to.to_tsecr && 1162 (tp->t_flags & TF_FIRSTACCACK)) { 1163 /* Eifel detection applicable. */ 1164 if (to.to_tsecr < tp->t_rexmtTS) { 1165 tcp_revert_congestion_state(tp); 1166 ++tcpstat.tcps_eifeldetected; 1167 } 1168 } else if (tp->t_rxtshift == 1 && 1169 ticks < tp->t_badrxtwin) { 1170 tcp_revert_congestion_state(tp); 1171 ++tcpstat.tcps_rttdetected; 1172 } 1173 tp->t_flags &= ~(TF_FIRSTACCACK | 1174 TF_FASTREXMT | TF_EARLYREXMT); 1175 /* 1176 * Recalculate the retransmit timer / rtt. 1177 * 1178 * Some machines (certain windows boxes) 1179 * send broken timestamp replies during the 1180 * SYN+ACK phase, ignore timestamps of 0. 1181 */ 1182 if ((to.to_flags & TOF_TS) && to.to_tsecr) { 1183 tcp_xmit_timer(tp, 1184 ticks - to.to_tsecr + 1); 1185 } else if (tp->t_rtttime && 1186 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1187 tcp_xmit_timer(tp, 1188 ticks - tp->t_rtttime); 1189 } 1190 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1191 acked = th->th_ack - tp->snd_una; 1192 tcpstat.tcps_rcvackpack++; 1193 tcpstat.tcps_rcvackbyte += acked; 1194 sbdrop(&so->so_snd, acked); 1195 tp->snd_recover = th->th_ack - 1; 1196 tp->snd_una = th->th_ack; 1197 tp->t_dupacks = 0; 1198 /* 1199 * Update window information. 1200 */ 1201 if (tiwin != tp->snd_wnd && 1202 acceptable_window_update(tp, th, tiwin)) { 1203 /* keep track of pure window updates */ 1204 if (tp->snd_wl2 == th->th_ack && 1205 tiwin > tp->snd_wnd) 1206 tcpstat.tcps_rcvwinupd++; 1207 tp->snd_wnd = tiwin; 1208 tp->snd_wl1 = th->th_seq; 1209 tp->snd_wl2 = th->th_ack; 1210 if (tp->snd_wnd > tp->max_sndwnd) 1211 tp->max_sndwnd = tp->snd_wnd; 1212 } 1213 m_freem(m); 1214 ND6_HINT(tp); /* some progress has been done */ 1215 /* 1216 * If all outstanding data are acked, stop 1217 * retransmit timer, otherwise restart timer 1218 * using current (possibly backed-off) value. 1219 * If process is waiting for space, 1220 * wakeup/selwakeup/signal. If data 1221 * are ready to send, let tcp_output 1222 * decide between more output or persist. 1223 */ 1224 if (tp->snd_una == tp->snd_max) 1225 callout_stop(tp->tt_rexmt); 1226 else if (!callout_active(tp->tt_persist)) 1227 callout_reset(tp->tt_rexmt, 1228 tp->t_rxtcur, 1229 tcp_timer_rexmt, tp); 1230 sowwakeup(so); 1231 if (so->so_snd.sb_cc > 0) 1232 tcp_output(tp); 1233 return; 1234 } 1235 } else if (tiwin == tp->snd_wnd && 1236 th->th_ack == tp->snd_una && 1237 LIST_EMPTY(&tp->t_segq) && 1238 tlen <= sbspace(&so->so_rcv)) { 1239 /* 1240 * This is a pure, in-sequence data packet 1241 * with nothing on the reassembly queue and 1242 * we have enough buffer space to take it. 1243 */ 1244 ++tcpstat.tcps_preddat; 1245 tp->rcv_nxt += tlen; 1246 tcpstat.tcps_rcvpack++; 1247 tcpstat.tcps_rcvbyte += tlen; 1248 ND6_HINT(tp); /* some progress has been done */ 1249 /* 1250 * Add data to socket buffer. 1251 */ 1252 if (so->so_state & SS_CANTRCVMORE) { 1253 m_freem(m); 1254 } else { 1255 m_adj(m, drop_hdrlen); /* delayed header drop */ 1256 sbappendstream(&so->so_rcv, m); 1257 } 1258 sorwakeup(so); 1259 /* 1260 * This code is responsible for most of the ACKs 1261 * the TCP stack sends back after receiving a data 1262 * packet. Note that the DELAY_ACK check fails if 1263 * the delack timer is already running, which results 1264 * in an ack being sent every other packet (which is 1265 * what we want). 1266 * 1267 * We then further aggregate acks by not actually 1268 * sending one until the protocol thread has completed 1269 * processing the current backlog of packets. This 1270 * does not delay the ack any further, but allows us 1271 * to take advantage of the packet aggregation that 1272 * high speed NICs do (usually blocks of 8-10 packets) 1273 * to send a single ack rather then four or five acks, 1274 * greatly reducing the ack rate, the return channel 1275 * bandwidth, and the protocol overhead on both ends. 1276 * 1277 * Since this also has the effect of slowing down 1278 * the exponential slow-start ramp-up, systems with 1279 * very large bandwidth-delay products might want 1280 * to turn the feature off. 1281 */ 1282 if (DELAY_ACK(tp)) { 1283 callout_reset(tp->tt_delack, tcp_delacktime, 1284 tcp_timer_delack, tp); 1285 } else if (tcp_aggregate_acks) { 1286 tp->t_flags |= TF_ACKNOW; 1287 if (!(tp->t_flags & TF_ONOUTPUTQ)) { 1288 tp->t_flags |= TF_ONOUTPUTQ; 1289 tp->tt_cpu = mycpu->gd_cpuid; 1290 TAILQ_INSERT_TAIL( 1291 &tcpcbackq[tp->tt_cpu], 1292 tp, t_outputq); 1293 } 1294 } else { 1295 tp->t_flags |= TF_ACKNOW; 1296 tcp_output(tp); 1297 } 1298 return; 1299 } 1300 } 1301 1302 /* 1303 * Calculate amount of space in receive window, 1304 * and then do TCP input processing. 1305 * Receive window is amount of space in rcv queue, 1306 * but not less than advertised window. 1307 */ 1308 recvwin = sbspace(&so->so_rcv); 1309 if (recvwin < 0) 1310 recvwin = 0; 1311 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt)); 1312 1313 switch (tp->t_state) { 1314 /* 1315 * If the state is SYN_RECEIVED: 1316 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1317 */ 1318 case TCPS_SYN_RECEIVED: 1319 if ((thflags & TH_ACK) && 1320 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1321 SEQ_GT(th->th_ack, tp->snd_max))) { 1322 rstreason = BANDLIM_RST_OPENPORT; 1323 goto dropwithreset; 1324 } 1325 break; 1326 1327 /* 1328 * If the state is SYN_SENT: 1329 * if seg contains an ACK, but not for our SYN, drop the input. 1330 * if seg contains a RST, then drop the connection. 1331 * if seg does not contain SYN, then drop it. 1332 * Otherwise this is an acceptable SYN segment 1333 * initialize tp->rcv_nxt and tp->irs 1334 * if seg contains ack then advance tp->snd_una 1335 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1336 * arrange for segment to be acked (eventually) 1337 * continue processing rest of data/controls, beginning with URG 1338 */ 1339 case TCPS_SYN_SENT: 1340 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1341 taop = &tao_noncached; 1342 bzero(taop, sizeof *taop); 1343 } 1344 1345 if ((thflags & TH_ACK) && 1346 (SEQ_LEQ(th->th_ack, tp->iss) || 1347 SEQ_GT(th->th_ack, tp->snd_max))) { 1348 /* 1349 * If we have a cached CCsent for the remote host, 1350 * hence we haven't just crashed and restarted, 1351 * do not send a RST. This may be a retransmission 1352 * from the other side after our earlier ACK was lost. 1353 * Our new SYN, when it arrives, will serve as the 1354 * needed ACK. 1355 */ 1356 if (taop->tao_ccsent != 0) 1357 goto drop; 1358 else { 1359 rstreason = BANDLIM_UNLIMITED; 1360 goto dropwithreset; 1361 } 1362 } 1363 if (thflags & TH_RST) { 1364 if (thflags & TH_ACK) 1365 tp = tcp_drop(tp, ECONNREFUSED); 1366 goto drop; 1367 } 1368 if (!(thflags & TH_SYN)) 1369 goto drop; 1370 tp->snd_wnd = th->th_win; /* initial send window */ 1371 tp->cc_recv = to.to_cc; /* foreign CC */ 1372 1373 tp->irs = th->th_seq; 1374 tcp_rcvseqinit(tp); 1375 if (thflags & TH_ACK) { 1376 /* 1377 * Our SYN was acked. If segment contains CC.ECHO 1378 * option, check it to make sure this segment really 1379 * matches our SYN. If not, just drop it as old 1380 * duplicate, but send an RST if we're still playing 1381 * by the old rules. If no CC.ECHO option, make sure 1382 * we don't get fooled into using T/TCP. 1383 */ 1384 if (to.to_flags & TOF_CCECHO) { 1385 if (tp->cc_send != to.to_ccecho) { 1386 if (taop->tao_ccsent != 0) 1387 goto drop; 1388 else { 1389 rstreason = BANDLIM_UNLIMITED; 1390 goto dropwithreset; 1391 } 1392 } 1393 } else 1394 tp->t_flags &= ~TF_RCVD_CC; 1395 tcpstat.tcps_connects++; 1396 soisconnected(so); 1397 /* Do window scaling on this connection? */ 1398 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 1399 (TF_RCVD_SCALE | TF_REQ_SCALE)) { 1400 tp->snd_scale = tp->requested_s_scale; 1401 tp->rcv_scale = tp->request_r_scale; 1402 } 1403 /* Segment is acceptable, update cache if undefined. */ 1404 if (taop->tao_ccsent == 0) 1405 taop->tao_ccsent = to.to_ccecho; 1406 1407 tp->rcv_adv += tp->rcv_wnd; 1408 tp->snd_una++; /* SYN is acked */ 1409 callout_stop(tp->tt_rexmt); 1410 /* 1411 * If there's data, delay ACK; if there's also a FIN 1412 * ACKNOW will be turned on later. 1413 */ 1414 if (DELAY_ACK(tp) && tlen != 0) 1415 callout_reset(tp->tt_delack, tcp_delacktime, 1416 tcp_timer_delack, tp); 1417 else 1418 tp->t_flags |= TF_ACKNOW; 1419 /* 1420 * Received <SYN,ACK> in SYN_SENT[*] state. 1421 * Transitions: 1422 * SYN_SENT --> ESTABLISHED 1423 * SYN_SENT* --> FIN_WAIT_1 1424 */ 1425 tp->t_starttime = ticks; 1426 if (tp->t_flags & TF_NEEDFIN) { 1427 tp->t_state = TCPS_FIN_WAIT_1; 1428 tp->t_flags &= ~TF_NEEDFIN; 1429 thflags &= ~TH_SYN; 1430 } else { 1431 tp->t_state = TCPS_ESTABLISHED; 1432 callout_reset(tp->tt_keep, tcp_keepidle, 1433 tcp_timer_keep, tp); 1434 } 1435 } else { 1436 /* 1437 * Received initial SYN in SYN-SENT[*] state => 1438 * simultaneous open. If segment contains CC option 1439 * and there is a cached CC, apply TAO test. 1440 * If it succeeds, connection is * half-synchronized. 1441 * Otherwise, do 3-way handshake: 1442 * SYN-SENT -> SYN-RECEIVED 1443 * SYN-SENT* -> SYN-RECEIVED* 1444 * If there was no CC option, clear cached CC value. 1445 */ 1446 tp->t_flags |= TF_ACKNOW; 1447 callout_stop(tp->tt_rexmt); 1448 if (to.to_flags & TOF_CC) { 1449 if (taop->tao_cc != 0 && 1450 CC_GT(to.to_cc, taop->tao_cc)) { 1451 /* 1452 * update cache and make transition: 1453 * SYN-SENT -> ESTABLISHED* 1454 * SYN-SENT* -> FIN-WAIT-1* 1455 */ 1456 taop->tao_cc = to.to_cc; 1457 tp->t_starttime = ticks; 1458 if (tp->t_flags & TF_NEEDFIN) { 1459 tp->t_state = TCPS_FIN_WAIT_1; 1460 tp->t_flags &= ~TF_NEEDFIN; 1461 } else { 1462 tp->t_state = TCPS_ESTABLISHED; 1463 callout_reset(tp->tt_keep, 1464 tcp_keepidle, 1465 tcp_timer_keep, 1466 tp); 1467 } 1468 tp->t_flags |= TF_NEEDSYN; 1469 } else 1470 tp->t_state = TCPS_SYN_RECEIVED; 1471 } else { 1472 /* CC.NEW or no option => invalidate cache */ 1473 taop->tao_cc = 0; 1474 tp->t_state = TCPS_SYN_RECEIVED; 1475 } 1476 } 1477 1478 trimthenstep6: 1479 /* 1480 * Advance th->th_seq to correspond to first data byte. 1481 * If data, trim to stay within window, 1482 * dropping FIN if necessary. 1483 */ 1484 th->th_seq++; 1485 if (tlen > tp->rcv_wnd) { 1486 todrop = tlen - tp->rcv_wnd; 1487 m_adj(m, -todrop); 1488 tlen = tp->rcv_wnd; 1489 thflags &= ~TH_FIN; 1490 tcpstat.tcps_rcvpackafterwin++; 1491 tcpstat.tcps_rcvbyteafterwin += todrop; 1492 } 1493 tp->snd_wl1 = th->th_seq - 1; 1494 tp->rcv_up = th->th_seq; 1495 /* 1496 * Client side of transaction: already sent SYN and data. 1497 * If the remote host used T/TCP to validate the SYN, 1498 * our data will be ACK'd; if so, enter normal data segment 1499 * processing in the middle of step 5, ack processing. 1500 * Otherwise, goto step 6. 1501 */ 1502 if (thflags & TH_ACK) 1503 goto process_ACK; 1504 1505 goto step6; 1506 1507 /* 1508 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1509 * if segment contains a SYN and CC [not CC.NEW] option: 1510 * if state == TIME_WAIT and connection duration > MSL, 1511 * drop packet and send RST; 1512 * 1513 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1514 * ack the FIN (and data) in retransmission queue. 1515 * Complete close and delete TCPCB. Then reprocess 1516 * segment, hoping to find new TCPCB in LISTEN state; 1517 * 1518 * else must be old SYN; drop it. 1519 * else do normal processing. 1520 */ 1521 case TCPS_LAST_ACK: 1522 case TCPS_CLOSING: 1523 case TCPS_TIME_WAIT: 1524 if ((thflags & TH_SYN) && 1525 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1526 if (tp->t_state == TCPS_TIME_WAIT && 1527 (ticks - tp->t_starttime) > tcp_msl) { 1528 rstreason = BANDLIM_UNLIMITED; 1529 goto dropwithreset; 1530 } 1531 if (CC_GT(to.to_cc, tp->cc_recv)) { 1532 tp = tcp_close(tp); 1533 goto findpcb; 1534 } 1535 else 1536 goto drop; 1537 } 1538 break; /* continue normal processing */ 1539 } 1540 1541 /* 1542 * States other than LISTEN or SYN_SENT. 1543 * First check the RST flag and sequence number since reset segments 1544 * are exempt from the timestamp and connection count tests. This 1545 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1546 * below which allowed reset segments in half the sequence space 1547 * to fall though and be processed (which gives forged reset 1548 * segments with a random sequence number a 50 percent chance of 1549 * killing a connection). 1550 * Then check timestamp, if present. 1551 * Then check the connection count, if present. 1552 * Then check that at least some bytes of segment are within 1553 * receive window. If segment begins before rcv_nxt, 1554 * drop leading data (and SYN); if nothing left, just ack. 1555 * 1556 * 1557 * If the RST bit is set, check the sequence number to see 1558 * if this is a valid reset segment. 1559 * RFC 793 page 37: 1560 * In all states except SYN-SENT, all reset (RST) segments 1561 * are validated by checking their SEQ-fields. A reset is 1562 * valid if its sequence number is in the window. 1563 * Note: this does not take into account delayed ACKs, so 1564 * we should test against last_ack_sent instead of rcv_nxt. 1565 * The sequence number in the reset segment is normally an 1566 * echo of our outgoing acknowledgement numbers, but some hosts 1567 * send a reset with the sequence number at the rightmost edge 1568 * of our receive window, and we have to handle this case. 1569 * If we have multiple segments in flight, the intial reset 1570 * segment sequence numbers will be to the left of last_ack_sent, 1571 * but they will eventually catch up. 1572 * In any case, it never made sense to trim reset segments to 1573 * fit the receive window since RFC 1122 says: 1574 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1575 * 1576 * A TCP SHOULD allow a received RST segment to include data. 1577 * 1578 * DISCUSSION 1579 * It has been suggested that a RST segment could contain 1580 * ASCII text that encoded and explained the cause of the 1581 * RST. No standard has yet been established for such 1582 * data. 1583 * 1584 * If the reset segment passes the sequence number test examine 1585 * the state: 1586 * SYN_RECEIVED STATE: 1587 * If passive open, return to LISTEN state. 1588 * If active open, inform user that connection was refused. 1589 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1590 * Inform user that connection was reset, and close tcb. 1591 * CLOSING, LAST_ACK STATES: 1592 * Close the tcb. 1593 * TIME_WAIT STATE: 1594 * Drop the segment - see Stevens, vol. 2, p. 964 and 1595 * RFC 1337. 1596 */ 1597 if (thflags & TH_RST) { 1598 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1599 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1600 switch (tp->t_state) { 1601 1602 case TCPS_SYN_RECEIVED: 1603 so->so_error = ECONNREFUSED; 1604 goto close; 1605 1606 case TCPS_ESTABLISHED: 1607 case TCPS_FIN_WAIT_1: 1608 case TCPS_FIN_WAIT_2: 1609 case TCPS_CLOSE_WAIT: 1610 so->so_error = ECONNRESET; 1611 close: 1612 tp->t_state = TCPS_CLOSED; 1613 tcpstat.tcps_drops++; 1614 tp = tcp_close(tp); 1615 break; 1616 1617 case TCPS_CLOSING: 1618 case TCPS_LAST_ACK: 1619 tp = tcp_close(tp); 1620 break; 1621 1622 case TCPS_TIME_WAIT: 1623 break; 1624 } 1625 } 1626 goto drop; 1627 } 1628 1629 /* 1630 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1631 * and it's less than ts_recent, drop it. 1632 */ 1633 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 && 1634 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1635 1636 /* Check to see if ts_recent is over 24 days old. */ 1637 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1638 /* 1639 * Invalidate ts_recent. If this segment updates 1640 * ts_recent, the age will be reset later and ts_recent 1641 * will get a valid value. If it does not, setting 1642 * ts_recent to zero will at least satisfy the 1643 * requirement that zero be placed in the timestamp 1644 * echo reply when ts_recent isn't valid. The 1645 * age isn't reset until we get a valid ts_recent 1646 * because we don't want out-of-order segments to be 1647 * dropped when ts_recent is old. 1648 */ 1649 tp->ts_recent = 0; 1650 } else { 1651 tcpstat.tcps_rcvduppack++; 1652 tcpstat.tcps_rcvdupbyte += tlen; 1653 tcpstat.tcps_pawsdrop++; 1654 if (tlen) 1655 goto dropafterack; 1656 goto drop; 1657 } 1658 } 1659 1660 /* 1661 * T/TCP mechanism 1662 * If T/TCP was negotiated and the segment doesn't have CC, 1663 * or if its CC is wrong then drop the segment. 1664 * RST segments do not have to comply with this. 1665 */ 1666 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1667 (!(to.to_flags & TOF_CC) || tp->cc_recv != to.to_cc)) 1668 goto dropafterack; 1669 1670 /* 1671 * In the SYN-RECEIVED state, validate that the packet belongs to 1672 * this connection before trimming the data to fit the receive 1673 * window. Check the sequence number versus IRS since we know 1674 * the sequence numbers haven't wrapped. This is a partial fix 1675 * for the "LAND" DoS attack. 1676 */ 1677 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1678 rstreason = BANDLIM_RST_OPENPORT; 1679 goto dropwithreset; 1680 } 1681 1682 todrop = tp->rcv_nxt - th->th_seq; 1683 if (todrop > 0) { 1684 if (TCP_DO_SACK(tp)) { 1685 /* Report duplicate segment at head of packet. */ 1686 tp->reportblk.rblk_start = th->th_seq; 1687 tp->reportblk.rblk_end = th->th_seq + tlen; 1688 if (thflags & TH_FIN) 1689 ++tp->reportblk.rblk_end; 1690 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt)) 1691 tp->reportblk.rblk_end = tp->rcv_nxt; 1692 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW); 1693 } 1694 if (thflags & TH_SYN) { 1695 thflags &= ~TH_SYN; 1696 th->th_seq++; 1697 if (th->th_urp > 1) 1698 th->th_urp--; 1699 else 1700 thflags &= ~TH_URG; 1701 todrop--; 1702 } 1703 /* 1704 * Following if statement from Stevens, vol. 2, p. 960. 1705 */ 1706 if (todrop > tlen || 1707 (todrop == tlen && !(thflags & TH_FIN))) { 1708 /* 1709 * Any valid FIN must be to the left of the window. 1710 * At this point the FIN must be a duplicate or out 1711 * of sequence; drop it. 1712 */ 1713 thflags &= ~TH_FIN; 1714 1715 /* 1716 * Send an ACK to resynchronize and drop any data. 1717 * But keep on processing for RST or ACK. 1718 */ 1719 tp->t_flags |= TF_ACKNOW; 1720 todrop = tlen; 1721 tcpstat.tcps_rcvduppack++; 1722 tcpstat.tcps_rcvdupbyte += todrop; 1723 } else { 1724 tcpstat.tcps_rcvpartduppack++; 1725 tcpstat.tcps_rcvpartdupbyte += todrop; 1726 } 1727 drop_hdrlen += todrop; /* drop from the top afterwards */ 1728 th->th_seq += todrop; 1729 tlen -= todrop; 1730 if (th->th_urp > todrop) 1731 th->th_urp -= todrop; 1732 else { 1733 thflags &= ~TH_URG; 1734 th->th_urp = 0; 1735 } 1736 } 1737 1738 /* 1739 * If new data are received on a connection after the 1740 * user processes are gone, then RST the other end. 1741 */ 1742 if ((so->so_state & SS_NOFDREF) && 1743 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1744 tp = tcp_close(tp); 1745 tcpstat.tcps_rcvafterclose++; 1746 rstreason = BANDLIM_UNLIMITED; 1747 goto dropwithreset; 1748 } 1749 1750 /* 1751 * If segment ends after window, drop trailing data 1752 * (and PUSH and FIN); if nothing left, just ACK. 1753 */ 1754 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 1755 if (todrop > 0) { 1756 tcpstat.tcps_rcvpackafterwin++; 1757 if (todrop >= tlen) { 1758 tcpstat.tcps_rcvbyteafterwin += tlen; 1759 /* 1760 * If a new connection request is received 1761 * while in TIME_WAIT, drop the old connection 1762 * and start over if the sequence numbers 1763 * are above the previous ones. 1764 */ 1765 if (thflags & TH_SYN && 1766 tp->t_state == TCPS_TIME_WAIT && 1767 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1768 tp = tcp_close(tp); 1769 goto findpcb; 1770 } 1771 /* 1772 * If window is closed can only take segments at 1773 * window edge, and have to drop data and PUSH from 1774 * incoming segments. Continue processing, but 1775 * remember to ack. Otherwise, drop segment 1776 * and ack. 1777 */ 1778 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1779 tp->t_flags |= TF_ACKNOW; 1780 tcpstat.tcps_rcvwinprobe++; 1781 } else 1782 goto dropafterack; 1783 } else 1784 tcpstat.tcps_rcvbyteafterwin += todrop; 1785 m_adj(m, -todrop); 1786 tlen -= todrop; 1787 thflags &= ~(TH_PUSH | TH_FIN); 1788 } 1789 1790 /* 1791 * If last ACK falls within this segment's sequence numbers, 1792 * record its timestamp. 1793 * NOTE that the test is modified according to the latest 1794 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1795 */ 1796 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1797 tp->ts_recent_age = ticks; 1798 tp->ts_recent = to.to_tsval; 1799 } 1800 1801 /* 1802 * If a SYN is in the window, then this is an 1803 * error and we send an RST and drop the connection. 1804 */ 1805 if (thflags & TH_SYN) { 1806 tp = tcp_drop(tp, ECONNRESET); 1807 rstreason = BANDLIM_UNLIMITED; 1808 goto dropwithreset; 1809 } 1810 1811 /* 1812 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1813 * flag is on (half-synchronized state), then queue data for 1814 * later processing; else drop segment and return. 1815 */ 1816 if (!(thflags & TH_ACK)) { 1817 if (tp->t_state == TCPS_SYN_RECEIVED || 1818 (tp->t_flags & TF_NEEDSYN)) 1819 goto step6; 1820 else 1821 goto drop; 1822 } 1823 1824 /* 1825 * Ack processing. 1826 */ 1827 switch (tp->t_state) { 1828 /* 1829 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter 1830 * ESTABLISHED state and continue processing. 1831 * The ACK was checked above. 1832 */ 1833 case TCPS_SYN_RECEIVED: 1834 1835 tcpstat.tcps_connects++; 1836 soisconnected(so); 1837 /* Do window scaling? */ 1838 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 1839 (TF_RCVD_SCALE | TF_REQ_SCALE)) { 1840 tp->snd_scale = tp->requested_s_scale; 1841 tp->rcv_scale = tp->request_r_scale; 1842 } 1843 /* 1844 * Upon successful completion of 3-way handshake, 1845 * update cache.CC if it was undefined, pass any queued 1846 * data to the user, and advance state appropriately. 1847 */ 1848 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1849 taop->tao_cc == 0) 1850 taop->tao_cc = tp->cc_recv; 1851 1852 /* 1853 * Make transitions: 1854 * SYN-RECEIVED -> ESTABLISHED 1855 * SYN-RECEIVED* -> FIN-WAIT-1 1856 */ 1857 tp->t_starttime = ticks; 1858 if (tp->t_flags & TF_NEEDFIN) { 1859 tp->t_state = TCPS_FIN_WAIT_1; 1860 tp->t_flags &= ~TF_NEEDFIN; 1861 } else { 1862 tp->t_state = TCPS_ESTABLISHED; 1863 callout_reset(tp->tt_keep, tcp_keepidle, 1864 tcp_timer_keep, tp); 1865 } 1866 /* 1867 * If segment contains data or ACK, will call tcp_reass() 1868 * later; if not, do so now to pass queued data to user. 1869 */ 1870 if (tlen == 0 && !(thflags & TH_FIN)) 1871 tcp_reass(tp, NULL, NULL, NULL); 1872 /* fall into ... */ 1873 1874 /* 1875 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1876 * ACKs. If the ack is in the range 1877 * tp->snd_una < th->th_ack <= tp->snd_max 1878 * then advance tp->snd_una to th->th_ack and drop 1879 * data from the retransmission queue. If this ACK reflects 1880 * more up to date window information we update our window information. 1881 */ 1882 case TCPS_ESTABLISHED: 1883 case TCPS_FIN_WAIT_1: 1884 case TCPS_FIN_WAIT_2: 1885 case TCPS_CLOSE_WAIT: 1886 case TCPS_CLOSING: 1887 case TCPS_LAST_ACK: 1888 case TCPS_TIME_WAIT: 1889 1890 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1891 if (TCP_DO_SACK(tp)) 1892 tcp_sack_update_scoreboard(tp, &to); 1893 if (tlen != 0 || tiwin != tp->snd_wnd) { 1894 tp->t_dupacks = 0; 1895 break; 1896 } 1897 tcpstat.tcps_rcvdupack++; 1898 if (!callout_active(tp->tt_rexmt) || 1899 th->th_ack != tp->snd_una) { 1900 tp->t_dupacks = 0; 1901 break; 1902 } 1903 /* 1904 * We have outstanding data (other than 1905 * a window probe), this is a completely 1906 * duplicate ack (ie, window info didn't 1907 * change), the ack is the biggest we've 1908 * seen and we've seen exactly our rexmt 1909 * threshhold of them, so assume a packet 1910 * has been dropped and retransmit it. 1911 * Kludge snd_nxt & the congestion 1912 * window so we send only this one 1913 * packet. 1914 */ 1915 if (IN_FASTRECOVERY(tp)) { 1916 if (TCP_DO_SACK(tp)) { 1917 /* No artifical cwnd inflation. */ 1918 tcp_sack_rexmt(tp, th); 1919 } else { 1920 /* 1921 * Dup acks mean that packets 1922 * have left the network 1923 * (they're now cached at the 1924 * receiver) so bump cwnd by 1925 * the amount in the receiver 1926 * to keep a constant cwnd 1927 * packets in the network. 1928 */ 1929 tp->snd_cwnd += tp->t_maxseg; 1930 tcp_output(tp); 1931 } 1932 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1933 tp->t_dupacks = 0; 1934 break; 1935 } else if (++tp->t_dupacks == tcprexmtthresh) { 1936 tcp_seq old_snd_nxt; 1937 u_int win; 1938 1939 fastretransmit: 1940 if (tcp_do_eifel_detect && 1941 (tp->t_flags & TF_RCVD_TSTMP)) { 1942 tcp_save_congestion_state(tp); 1943 tp->t_flags |= TF_FASTREXMT; 1944 } 1945 /* 1946 * We know we're losing at the current 1947 * window size, so do congestion avoidance: 1948 * set ssthresh to half the current window 1949 * and pull our congestion window back to the 1950 * new ssthresh. 1951 */ 1952 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1953 tp->t_maxseg; 1954 if (win < 2) 1955 win = 2; 1956 tp->snd_ssthresh = win * tp->t_maxseg; 1957 ENTER_FASTRECOVERY(tp); 1958 tp->snd_recover = tp->snd_max; 1959 callout_stop(tp->tt_rexmt); 1960 tp->t_rtttime = 0; 1961 old_snd_nxt = tp->snd_nxt; 1962 tp->snd_nxt = th->th_ack; 1963 tp->snd_cwnd = tp->t_maxseg; 1964 tcp_output(tp); 1965 ++tcpstat.tcps_sndfastrexmit; 1966 tp->snd_cwnd = tp->snd_ssthresh; 1967 tp->rexmt_high = tp->snd_nxt; 1968 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 1969 tp->snd_nxt = old_snd_nxt; 1970 KASSERT(tp->snd_limited <= 2, 1971 ("tp->snd_limited too big")); 1972 if (TCP_DO_SACK(tp)) 1973 tcp_sack_rexmt(tp, th); 1974 else 1975 tp->snd_cwnd += tp->t_maxseg * 1976 (tp->t_dupacks - tp->snd_limited); 1977 } else if (tcp_do_limitedtransmit) { 1978 u_long oldcwnd = tp->snd_cwnd; 1979 tcp_seq oldsndmax = tp->snd_max; 1980 tcp_seq oldsndnxt = tp->snd_nxt; 1981 /* outstanding data */ 1982 uint32_t ownd = tp->snd_max - tp->snd_una; 1983 u_int sent; 1984 1985 #define iceildiv(n, d) (((n)+(d)-1) / (d)) 1986 1987 KASSERT(tp->t_dupacks == 1 || 1988 tp->t_dupacks == 2, 1989 ("dupacks not 1 or 2")); 1990 if (tp->t_dupacks == 1) 1991 tp->snd_limited = 0; 1992 tp->snd_nxt = tp->snd_max; 1993 tp->snd_cwnd = ownd + 1994 (tp->t_dupacks - tp->snd_limited) * 1995 tp->t_maxseg; 1996 tcp_output(tp); 1997 if (SEQ_LT(oldsndnxt, oldsndmax)) 1998 tp->snd_nxt = oldsndnxt; 1999 tp->snd_cwnd = oldcwnd; 2000 sent = tp->snd_max - oldsndmax; 2001 if (sent > tp->t_maxseg) { 2002 KASSERT((tp->t_dupacks == 2 && 2003 tp->snd_limited == 0) || 2004 (sent == tp->t_maxseg + 1 && 2005 tp->t_flags & TF_SENTFIN), 2006 ("sent too much")); 2007 KASSERT(sent <= tp->t_maxseg * 2, 2008 ("sent too many segments")); 2009 tp->snd_limited = 2; 2010 tcpstat.tcps_sndlimited += 2; 2011 } else if (sent > 0) { 2012 ++tp->snd_limited; 2013 ++tcpstat.tcps_sndlimited; 2014 } else if (tcp_do_early_retransmit && 2015 (tcp_do_eifel_detect && 2016 (tp->t_flags & TF_RCVD_TSTMP)) && 2017 ownd < 4 * tp->t_maxseg && 2018 tp->t_dupacks + 1 >= 2019 iceildiv(ownd, tp->t_maxseg) && 2020 (!TCP_DO_SACK(tp) || 2021 ownd <= tp->t_maxseg || 2022 tcp_sack_has_sacked(&tp->scb, 2023 ownd - tp->t_maxseg))) { 2024 ++tcpstat.tcps_sndearlyrexmit; 2025 tp->t_flags |= TF_EARLYREXMT; 2026 goto fastretransmit; 2027 } 2028 } 2029 goto drop; 2030 } 2031 2032 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 2033 tp->t_dupacks = 0; 2034 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2035 /* 2036 * Detected optimistic ACK attack. 2037 * Force slow-start to de-synchronize attack. 2038 */ 2039 tp->snd_cwnd = tp->t_maxseg; 2040 2041 tcpstat.tcps_rcvacktoomuch++; 2042 goto dropafterack; 2043 } 2044 /* 2045 * If we reach this point, ACK is not a duplicate, 2046 * i.e., it ACKs something we sent. 2047 */ 2048 if (tp->t_flags & TF_NEEDSYN) { 2049 /* 2050 * T/TCP: Connection was half-synchronized, and our 2051 * SYN has been ACK'd (so connection is now fully 2052 * synchronized). Go to non-starred state, 2053 * increment snd_una for ACK of SYN, and check if 2054 * we can do window scaling. 2055 */ 2056 tp->t_flags &= ~TF_NEEDSYN; 2057 tp->snd_una++; 2058 /* Do window scaling? */ 2059 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 2060 (TF_RCVD_SCALE | TF_REQ_SCALE)) { 2061 tp->snd_scale = tp->requested_s_scale; 2062 tp->rcv_scale = tp->request_r_scale; 2063 } 2064 } 2065 2066 process_ACK: 2067 acked = th->th_ack - tp->snd_una; 2068 tcpstat.tcps_rcvackpack++; 2069 tcpstat.tcps_rcvackbyte += acked; 2070 2071 if (tcp_do_eifel_detect && acked > 0 && 2072 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) && 2073 (tp->t_flags & TF_FIRSTACCACK)) { 2074 /* Eifel detection applicable. */ 2075 if (to.to_tsecr < tp->t_rexmtTS) { 2076 ++tcpstat.tcps_eifeldetected; 2077 tcp_revert_congestion_state(tp); 2078 if (tp->t_rxtshift == 1 && 2079 ticks >= tp->t_badrxtwin) 2080 ++tcpstat.tcps_rttcantdetect; 2081 } 2082 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2083 /* 2084 * If we just performed our first retransmit, 2085 * and the ACK arrives within our recovery window, 2086 * then it was a mistake to do the retransmit 2087 * in the first place. Recover our original cwnd 2088 * and ssthresh, and proceed to transmit where we 2089 * left off. 2090 */ 2091 tcp_revert_congestion_state(tp); 2092 ++tcpstat.tcps_rttdetected; 2093 } 2094 2095 /* 2096 * If we have a timestamp reply, update smoothed 2097 * round trip time. If no timestamp is present but 2098 * transmit timer is running and timed sequence 2099 * number was acked, update smoothed round trip time. 2100 * Since we now have an rtt measurement, cancel the 2101 * timer backoff (cf., Phil Karn's retransmit alg.). 2102 * Recompute the initial retransmit timer. 2103 * 2104 * Some machines (certain windows boxes) send broken 2105 * timestamp replies during the SYN+ACK phase, ignore 2106 * timestamps of 0. 2107 */ 2108 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) 2109 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2110 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2111 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2112 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2113 2114 /* 2115 * If no data (only SYN) was ACK'd, 2116 * skip rest of ACK processing. 2117 */ 2118 if (acked == 0) 2119 goto step6; 2120 2121 /* Stop looking for an acceptable ACK since one was received. */ 2122 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT); 2123 2124 if (acked > so->so_snd.sb_cc) { 2125 tp->snd_wnd -= so->so_snd.sb_cc; 2126 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2127 ourfinisacked = TRUE; 2128 } else { 2129 sbdrop(&so->so_snd, acked); 2130 tp->snd_wnd -= acked; 2131 ourfinisacked = FALSE; 2132 } 2133 sowwakeup(so); 2134 2135 /* 2136 * Update window information. 2137 * Don't look at window if no ACK: 2138 * TAC's send garbage on first SYN. 2139 */ 2140 if (SEQ_LT(tp->snd_wl1, th->th_seq) || 2141 (tp->snd_wl1 == th->th_seq && 2142 (SEQ_LT(tp->snd_wl2, th->th_ack) || 2143 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) { 2144 /* keep track of pure window updates */ 2145 if (tlen == 0 && tp->snd_wl2 == th->th_ack && 2146 tiwin > tp->snd_wnd) 2147 tcpstat.tcps_rcvwinupd++; 2148 tp->snd_wnd = tiwin; 2149 tp->snd_wl1 = th->th_seq; 2150 tp->snd_wl2 = th->th_ack; 2151 if (tp->snd_wnd > tp->max_sndwnd) 2152 tp->max_sndwnd = tp->snd_wnd; 2153 needoutput = TRUE; 2154 } 2155 2156 tp->snd_una = th->th_ack; 2157 if (TCP_DO_SACK(tp)) 2158 tcp_sack_update_scoreboard(tp, &to); 2159 if (IN_FASTRECOVERY(tp)) { 2160 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2161 EXIT_FASTRECOVERY(tp); 2162 needoutput = TRUE; 2163 /* 2164 * If the congestion window was inflated 2165 * to account for the other side's 2166 * cached packets, retract it. 2167 * 2168 * Window inflation should have left us 2169 * with approximately snd_ssthresh outstanding 2170 * data. But, in case we would be inclined 2171 * to send a burst, better do it using 2172 * slow start. 2173 */ 2174 if (!TCP_DO_SACK(tp)) 2175 tp->snd_cwnd = tp->snd_ssthresh; 2176 2177 if (SEQ_GT(th->th_ack + tp->snd_cwnd, 2178 tp->snd_max + 2 * tp->t_maxseg)) 2179 tp->snd_cwnd = 2180 (tp->snd_max - tp->snd_una) + 2181 2 * tp->t_maxseg; 2182 } else { 2183 if (TCP_DO_SACK(tp)) { 2184 tp->snd_max_rexmt = tp->snd_max; 2185 tcp_sack_rexmt(tp, th); 2186 } else { 2187 tcp_newreno_partial_ack(tp, th, acked); 2188 } 2189 needoutput = FALSE; 2190 } 2191 } else { 2192 /* 2193 * When new data is acked, open the congestion window. 2194 * If the window gives us less than ssthresh packets 2195 * in flight, open exponentially (maxseg per packet). 2196 * Otherwise open linearly: maxseg per window 2197 * (maxseg^2 / cwnd per packet). 2198 */ 2199 u_int cw = tp->snd_cwnd; 2200 u_int incr; 2201 2202 if (cw > tp->snd_ssthresh) 2203 incr = tp->t_maxseg * tp->t_maxseg / cw; 2204 else 2205 incr = tp->t_maxseg; 2206 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2207 tp->snd_recover = th->th_ack - 1; 2208 } 2209 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2210 tp->snd_nxt = tp->snd_una; 2211 2212 /* 2213 * If all outstanding data is acked, stop retransmit 2214 * timer and remember to restart (more output or persist). 2215 * If there is more data to be acked, restart retransmit 2216 * timer, using current (possibly backed-off) value. 2217 */ 2218 if (th->th_ack == tp->snd_max) { 2219 callout_stop(tp->tt_rexmt); 2220 needoutput = TRUE; 2221 } else if (!callout_active(tp->tt_persist)) 2222 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2223 tcp_timer_rexmt, tp); 2224 2225 switch (tp->t_state) { 2226 /* 2227 * In FIN_WAIT_1 STATE in addition to the processing 2228 * for the ESTABLISHED state if our FIN is now acknowledged 2229 * then enter FIN_WAIT_2. 2230 */ 2231 case TCPS_FIN_WAIT_1: 2232 if (ourfinisacked) { 2233 /* 2234 * If we can't receive any more 2235 * data, then closing user can proceed. 2236 * Starting the timer is contrary to the 2237 * specification, but if we don't get a FIN 2238 * we'll hang forever. 2239 */ 2240 if (so->so_state & SS_CANTRCVMORE) { 2241 soisdisconnected(so); 2242 callout_reset(tp->tt_2msl, tcp_maxidle, 2243 tcp_timer_2msl, tp); 2244 } 2245 tp->t_state = TCPS_FIN_WAIT_2; 2246 } 2247 break; 2248 2249 /* 2250 * In CLOSING STATE in addition to the processing for 2251 * the ESTABLISHED state if the ACK acknowledges our FIN 2252 * then enter the TIME-WAIT state, otherwise ignore 2253 * the segment. 2254 */ 2255 case TCPS_CLOSING: 2256 if (ourfinisacked) { 2257 tp->t_state = TCPS_TIME_WAIT; 2258 tcp_canceltimers(tp); 2259 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2260 if (tp->cc_recv != 0 && 2261 (ticks - tp->t_starttime) < tcp_msl) 2262 callout_reset(tp->tt_2msl, 2263 tp->t_rxtcur * TCPTV_TWTRUNC, 2264 tcp_timer_2msl, tp); 2265 else 2266 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2267 tcp_timer_2msl, tp); 2268 soisdisconnected(so); 2269 } 2270 break; 2271 2272 /* 2273 * In LAST_ACK, we may still be waiting for data to drain 2274 * and/or to be acked, as well as for the ack of our FIN. 2275 * If our FIN is now acknowledged, delete the TCB, 2276 * enter the closed state and return. 2277 */ 2278 case TCPS_LAST_ACK: 2279 if (ourfinisacked) { 2280 tp = tcp_close(tp); 2281 goto drop; 2282 } 2283 break; 2284 2285 /* 2286 * In TIME_WAIT state the only thing that should arrive 2287 * is a retransmission of the remote FIN. Acknowledge 2288 * it and restart the finack timer. 2289 */ 2290 case TCPS_TIME_WAIT: 2291 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2292 tcp_timer_2msl, tp); 2293 goto dropafterack; 2294 } 2295 } 2296 2297 step6: 2298 /* 2299 * Update window information. 2300 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2301 */ 2302 if ((thflags & TH_ACK) && 2303 acceptable_window_update(tp, th, tiwin)) { 2304 /* keep track of pure window updates */ 2305 if (tlen == 0 && tp->snd_wl2 == th->th_ack && 2306 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 = TRUE; 2314 } 2315 2316 /* 2317 * Process segments with URG. 2318 */ 2319 if ((thflags & TH_URG) && th->th_urp && 2320 !TCPS_HAVERCVDFIN(tp->t_state)) { 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 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2328 th->th_urp = 0; /* XXX */ 2329 thflags &= ~TH_URG; /* XXX */ 2330 goto dodata; /* XXX */ 2331 } 2332 /* 2333 * If this segment advances the known urgent pointer, 2334 * then mark the data stream. This should not happen 2335 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2336 * a FIN has been received from the remote side. 2337 * In these states we ignore the URG. 2338 * 2339 * According to RFC961 (Assigned Protocols), 2340 * the urgent pointer points to the last octet 2341 * of urgent data. We continue, however, 2342 * to consider it to indicate the first octet 2343 * of data past the urgent section as the original 2344 * spec states (in one of two places). 2345 */ 2346 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) { 2347 tp->rcv_up = th->th_seq + th->th_urp; 2348 so->so_oobmark = so->so_rcv.sb_cc + 2349 (tp->rcv_up - tp->rcv_nxt) - 1; 2350 if (so->so_oobmark == 0) 2351 so->so_state |= SS_RCVATMARK; 2352 sohasoutofband(so); 2353 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2354 } 2355 /* 2356 * Remove out of band data so doesn't get presented to user. 2357 * This can happen independent of advancing the URG pointer, 2358 * but if two URG's are pending at once, some out-of-band 2359 * data may creep in... ick. 2360 */ 2361 if (th->th_urp <= (u_long)tlen && 2362 !(so->so_options & SO_OOBINLINE)) { 2363 /* hdr drop is delayed */ 2364 tcp_pulloutofband(so, th, m, drop_hdrlen); 2365 } 2366 } else { 2367 /* 2368 * If no out of band data is expected, 2369 * pull receive urgent pointer along 2370 * with the receive window. 2371 */ 2372 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2373 tp->rcv_up = tp->rcv_nxt; 2374 } 2375 2376 dodata: /* XXX */ 2377 /* 2378 * Process the segment text, merging it into the TCP sequencing queue, 2379 * and arranging for acknowledgment of receipt if necessary. 2380 * This process logically involves adjusting tp->rcv_wnd as data 2381 * is presented to the user (this happens in tcp_usrreq.c, 2382 * case PRU_RCVD). If a FIN has already been received on this 2383 * connection then we just ignore the text. 2384 */ 2385 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) { 2386 m_adj(m, drop_hdrlen); /* delayed header drop */ 2387 /* 2388 * Insert segment which includes th into TCP reassembly queue 2389 * with control block tp. Set thflags to whether reassembly now 2390 * includes a segment with FIN. This handles the common case 2391 * inline (segment is the next to be received on an established 2392 * connection, and the queue is empty), avoiding linkage into 2393 * and removal from the queue and repetition of various 2394 * conversions. 2395 * Set DELACK for segments received in order, but ack 2396 * immediately when segments are out of order (so 2397 * fast retransmit can work). 2398 */ 2399 if (th->th_seq == tp->rcv_nxt && 2400 LIST_EMPTY(&tp->t_segq) && 2401 TCPS_HAVEESTABLISHED(tp->t_state)) { 2402 if (DELAY_ACK(tp)) 2403 callout_reset(tp->tt_delack, tcp_delacktime, 2404 tcp_timer_delack, tp); 2405 else 2406 tp->t_flags |= TF_ACKNOW; 2407 tp->rcv_nxt += tlen; 2408 thflags = th->th_flags & TH_FIN; 2409 tcpstat.tcps_rcvpack++; 2410 tcpstat.tcps_rcvbyte += tlen; 2411 ND6_HINT(tp); 2412 if (so->so_state & SS_CANTRCVMORE) 2413 m_freem(m); 2414 else 2415 sbappendstream(&so->so_rcv, m); 2416 sorwakeup(so); 2417 } else { 2418 if (!(tp->t_flags & TF_DUPSEG)) { 2419 /* Initialize SACK report block. */ 2420 tp->reportblk.rblk_start = th->th_seq; 2421 tp->reportblk.rblk_end = th->th_seq + tlen + 2422 ((thflags & TH_FIN) != 0); 2423 } 2424 thflags = tcp_reass(tp, th, &tlen, m); 2425 tp->t_flags |= TF_ACKNOW; 2426 } 2427 2428 /* 2429 * Note the amount of data that peer has sent into 2430 * our window, in order to estimate the sender's 2431 * buffer size. 2432 */ 2433 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2434 } else { 2435 m_freem(m); 2436 thflags &= ~TH_FIN; 2437 } 2438 2439 /* 2440 * If FIN is received ACK the FIN and let the user know 2441 * that the connection is closing. 2442 */ 2443 if (thflags & TH_FIN) { 2444 if (!TCPS_HAVERCVDFIN(tp->t_state)) { 2445 socantrcvmore(so); 2446 /* 2447 * If connection is half-synchronized 2448 * (ie NEEDSYN flag on) then delay ACK, 2449 * so it may be piggybacked when SYN is sent. 2450 * Otherwise, since we received a FIN then no 2451 * more input can be expected, send ACK now. 2452 */ 2453 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2454 callout_reset(tp->tt_delack, tcp_delacktime, 2455 tcp_timer_delack, tp); 2456 else 2457 tp->t_flags |= TF_ACKNOW; 2458 tp->rcv_nxt++; 2459 } 2460 2461 switch (tp->t_state) { 2462 /* 2463 * In SYN_RECEIVED and ESTABLISHED STATES 2464 * enter the CLOSE_WAIT state. 2465 */ 2466 case TCPS_SYN_RECEIVED: 2467 tp->t_starttime = ticks; 2468 /*FALLTHROUGH*/ 2469 case TCPS_ESTABLISHED: 2470 tp->t_state = TCPS_CLOSE_WAIT; 2471 break; 2472 2473 /* 2474 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2475 * enter the CLOSING state. 2476 */ 2477 case TCPS_FIN_WAIT_1: 2478 tp->t_state = TCPS_CLOSING; 2479 break; 2480 2481 /* 2482 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2483 * starting the time-wait timer, turning off the other 2484 * standard timers. 2485 */ 2486 case TCPS_FIN_WAIT_2: 2487 tp->t_state = TCPS_TIME_WAIT; 2488 tcp_canceltimers(tp); 2489 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2490 if (tp->cc_recv != 0 && 2491 (ticks - tp->t_starttime) < tcp_msl) { 2492 callout_reset(tp->tt_2msl, 2493 tp->t_rxtcur * TCPTV_TWTRUNC, 2494 tcp_timer_2msl, tp); 2495 /* For transaction client, force ACK now. */ 2496 tp->t_flags |= TF_ACKNOW; 2497 } 2498 else 2499 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2500 tcp_timer_2msl, tp); 2501 soisdisconnected(so); 2502 break; 2503 2504 /* 2505 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2506 */ 2507 case TCPS_TIME_WAIT: 2508 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2509 tcp_timer_2msl, tp); 2510 break; 2511 } 2512 } 2513 2514 #ifdef TCPDEBUG 2515 if (so->so_options & SO_DEBUG) 2516 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2517 #endif 2518 2519 /* 2520 * Return any desired output. 2521 */ 2522 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2523 tcp_output(tp); 2524 return; 2525 2526 dropafterack: 2527 /* 2528 * Generate an ACK dropping incoming segment if it occupies 2529 * sequence space, where the ACK reflects our state. 2530 * 2531 * We can now skip the test for the RST flag since all 2532 * paths to this code happen after packets containing 2533 * RST have been dropped. 2534 * 2535 * In the SYN-RECEIVED state, don't send an ACK unless the 2536 * segment we received passes the SYN-RECEIVED ACK test. 2537 * If it fails send a RST. This breaks the loop in the 2538 * "LAND" DoS attack, and also prevents an ACK storm 2539 * between two listening ports that have been sent forged 2540 * SYN segments, each with the source address of the other. 2541 */ 2542 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2543 (SEQ_GT(tp->snd_una, th->th_ack) || 2544 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2545 rstreason = BANDLIM_RST_OPENPORT; 2546 goto dropwithreset; 2547 } 2548 #ifdef TCPDEBUG 2549 if (so->so_options & SO_DEBUG) 2550 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2551 #endif 2552 m_freem(m); 2553 tp->t_flags |= TF_ACKNOW; 2554 tcp_output(tp); 2555 return; 2556 2557 dropwithreset: 2558 /* 2559 * Generate a RST, dropping incoming segment. 2560 * Make ACK acceptable to originator of segment. 2561 * Don't bother to respond if destination was broadcast/multicast. 2562 */ 2563 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST)) 2564 goto drop; 2565 if (isipv6) { 2566 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2567 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2568 goto drop; 2569 } else { 2570 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2571 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2572 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2573 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2574 goto drop; 2575 } 2576 /* IPv6 anycast check is done at tcp6_input() */ 2577 2578 /* 2579 * Perform bandwidth limiting. 2580 */ 2581 #ifdef ICMP_BANDLIM 2582 if (badport_bandlim(rstreason) < 0) 2583 goto drop; 2584 #endif 2585 2586 #ifdef TCPDEBUG 2587 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2588 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2589 #endif 2590 if (thflags & TH_ACK) 2591 /* mtod() below is safe as long as hdr dropping is delayed */ 2592 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2593 TH_RST); 2594 else { 2595 if (thflags & TH_SYN) 2596 tlen++; 2597 /* mtod() below is safe as long as hdr dropping is delayed */ 2598 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen, 2599 (tcp_seq)0, TH_RST | TH_ACK); 2600 } 2601 return; 2602 2603 drop: 2604 /* 2605 * Drop space held by incoming segment and return. 2606 */ 2607 #ifdef TCPDEBUG 2608 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2609 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2610 #endif 2611 m_freem(m); 2612 return; 2613 } 2614 2615 /* 2616 * Parse TCP options and place in tcpopt. 2617 */ 2618 static void 2619 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn) 2620 { 2621 int opt, optlen, i; 2622 2623 to->to_flags = 0; 2624 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2625 opt = cp[0]; 2626 if (opt == TCPOPT_EOL) 2627 break; 2628 if (opt == TCPOPT_NOP) 2629 optlen = 1; 2630 else { 2631 if (cnt < 2) 2632 break; 2633 optlen = cp[1]; 2634 if (optlen < 2 || optlen > cnt) 2635 break; 2636 } 2637 switch (opt) { 2638 case TCPOPT_MAXSEG: 2639 if (optlen != TCPOLEN_MAXSEG) 2640 continue; 2641 if (!is_syn) 2642 continue; 2643 to->to_flags |= TOF_MSS; 2644 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss); 2645 to->to_mss = ntohs(to->to_mss); 2646 break; 2647 case TCPOPT_WINDOW: 2648 if (optlen != TCPOLEN_WINDOW) 2649 continue; 2650 if (!is_syn) 2651 continue; 2652 to->to_flags |= TOF_SCALE; 2653 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2654 break; 2655 case TCPOPT_TIMESTAMP: 2656 if (optlen != TCPOLEN_TIMESTAMP) 2657 continue; 2658 to->to_flags |= TOF_TS; 2659 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval); 2660 to->to_tsval = ntohl(to->to_tsval); 2661 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr); 2662 to->to_tsecr = ntohl(to->to_tsecr); 2663 break; 2664 case TCPOPT_CC: 2665 if (optlen != TCPOLEN_CC) 2666 continue; 2667 to->to_flags |= TOF_CC; 2668 bcopy(cp + 2, &to->to_cc, sizeof to->to_cc); 2669 to->to_cc = ntohl(to->to_cc); 2670 break; 2671 case TCPOPT_CCNEW: 2672 if (optlen != TCPOLEN_CC) 2673 continue; 2674 if (!is_syn) 2675 continue; 2676 to->to_flags |= TOF_CCNEW; 2677 bcopy(cp + 2, &to->to_cc, sizeof to->to_cc); 2678 to->to_cc = ntohl(to->to_cc); 2679 break; 2680 case TCPOPT_CCECHO: 2681 if (optlen != TCPOLEN_CC) 2682 continue; 2683 if (!is_syn) 2684 continue; 2685 to->to_flags |= TOF_CCECHO; 2686 bcopy(cp + 2, &to->to_ccecho, sizeof to->to_ccecho); 2687 to->to_ccecho = ntohl(to->to_ccecho); 2688 break; 2689 case TCPOPT_SACK_PERMITTED: 2690 if (optlen != TCPOLEN_SACK_PERMITTED) 2691 continue; 2692 if (!is_syn) 2693 continue; 2694 to->to_flags |= TOF_SACK_PERMITTED; 2695 break; 2696 case TCPOPT_SACK: 2697 if ((optlen - 2) & 0x07) /* not multiple of 8 */ 2698 continue; 2699 to->to_nsackblocks = (optlen - 2) / 8; 2700 to->to_sackblocks = (struct raw_sackblock *) (cp + 2); 2701 to->to_flags |= TOF_SACK; 2702 for (i = 0; i < to->to_nsackblocks; i++) { 2703 struct raw_sackblock *r = &to->to_sackblocks[i]; 2704 2705 r->rblk_start = ntohl(r->rblk_start); 2706 r->rblk_end = ntohl(r->rblk_end); 2707 } 2708 break; 2709 default: 2710 continue; 2711 } 2712 } 2713 } 2714 2715 /* 2716 * Pull out of band byte out of a segment so 2717 * it doesn't appear in the user's data queue. 2718 * It is still reflected in the segment length for 2719 * sequencing purposes. 2720 * "off" is the delayed to be dropped hdrlen. 2721 */ 2722 static void 2723 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off) 2724 { 2725 int cnt = off + th->th_urp - 1; 2726 2727 while (cnt >= 0) { 2728 if (m->m_len > cnt) { 2729 char *cp = mtod(m, caddr_t) + cnt; 2730 struct tcpcb *tp = sototcpcb(so); 2731 2732 tp->t_iobc = *cp; 2733 tp->t_oobflags |= TCPOOB_HAVEDATA; 2734 bcopy(cp + 1, cp, m->m_len - cnt - 1); 2735 m->m_len--; 2736 if (m->m_flags & M_PKTHDR) 2737 m->m_pkthdr.len--; 2738 return; 2739 } 2740 cnt -= m->m_len; 2741 m = m->m_next; 2742 if (m == 0) 2743 break; 2744 } 2745 panic("tcp_pulloutofband"); 2746 } 2747 2748 /* 2749 * Collect new round-trip time estimate 2750 * and update averages and current timeout. 2751 */ 2752 static void 2753 tcp_xmit_timer(struct tcpcb *tp, int rtt) 2754 { 2755 int delta; 2756 2757 tcpstat.tcps_rttupdated++; 2758 tp->t_rttupdated++; 2759 if (tp->t_srtt != 0) { 2760 /* 2761 * srtt is stored as fixed point with 5 bits after the 2762 * binary point (i.e., scaled by 8). The following magic 2763 * is equivalent to the smoothing algorithm in rfc793 with 2764 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2765 * point). Adjust rtt to origin 0. 2766 */ 2767 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2768 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2769 2770 if ((tp->t_srtt += delta) <= 0) 2771 tp->t_srtt = 1; 2772 2773 /* 2774 * We accumulate a smoothed rtt variance (actually, a 2775 * smoothed mean difference), then set the retransmit 2776 * timer to smoothed rtt + 4 times the smoothed variance. 2777 * rttvar is stored as fixed point with 4 bits after the 2778 * binary point (scaled by 16). The following is 2779 * equivalent to rfc793 smoothing with an alpha of .75 2780 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2781 * rfc793's wired-in beta. 2782 */ 2783 if (delta < 0) 2784 delta = -delta; 2785 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2786 if ((tp->t_rttvar += delta) <= 0) 2787 tp->t_rttvar = 1; 2788 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2789 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2790 } else { 2791 /* 2792 * No rtt measurement yet - use the unsmoothed rtt. 2793 * Set the variance to half the rtt (so our first 2794 * retransmit happens at 3*rtt). 2795 */ 2796 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2797 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2798 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2799 } 2800 tp->t_rtttime = 0; 2801 tp->t_rxtshift = 0; 2802 2803 /* 2804 * the retransmit should happen at rtt + 4 * rttvar. 2805 * Because of the way we do the smoothing, srtt and rttvar 2806 * will each average +1/2 tick of bias. When we compute 2807 * the retransmit timer, we want 1/2 tick of rounding and 2808 * 1 extra tick because of +-1/2 tick uncertainty in the 2809 * firing of the timer. The bias will give us exactly the 2810 * 1.5 tick we need. But, because the bias is 2811 * statistical, we have to test that we don't drop below 2812 * the minimum feasible timer (which is 2 ticks). 2813 */ 2814 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2815 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2816 2817 /* 2818 * We received an ack for a packet that wasn't retransmitted; 2819 * it is probably safe to discard any error indications we've 2820 * received recently. This isn't quite right, but close enough 2821 * for now (a route might have failed after we sent a segment, 2822 * and the return path might not be symmetrical). 2823 */ 2824 tp->t_softerror = 0; 2825 } 2826 2827 /* 2828 * Determine a reasonable value for maxseg size. 2829 * If the route is known, check route for mtu. 2830 * If none, use an mss that can be handled on the outgoing 2831 * interface without forcing IP to fragment; if bigger than 2832 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2833 * to utilize large mbufs. If no route is found, route has no mtu, 2834 * or the destination isn't local, use a default, hopefully conservative 2835 * size (usually 512 or the default IP max size, but no more than the mtu 2836 * of the interface), as we can't discover anything about intervening 2837 * gateways or networks. We also initialize the congestion/slow start 2838 * window to be a single segment if the destination isn't local. 2839 * While looking at the routing entry, we also initialize other path-dependent 2840 * parameters from pre-set or cached values in the routing entry. 2841 * 2842 * Also take into account the space needed for options that we 2843 * send regularly. Make maxseg shorter by that amount to assure 2844 * that we can send maxseg amount of data even when the options 2845 * are present. Store the upper limit of the length of options plus 2846 * data in maxopd. 2847 * 2848 * NOTE that this routine is only called when we process an incoming 2849 * segment, for outgoing segments only tcp_mssopt is called. 2850 * 2851 * In case of T/TCP, we call this routine during implicit connection 2852 * setup as well (offer = -1), to initialize maxseg from the cached 2853 * MSS of our peer. 2854 */ 2855 void 2856 tcp_mss(struct tcpcb *tp, int offer) 2857 { 2858 struct rtentry *rt; 2859 struct ifnet *ifp; 2860 int rtt, mss; 2861 u_long bufsize; 2862 struct inpcb *inp = tp->t_inpcb; 2863 struct socket *so; 2864 struct rmxp_tao *taop; 2865 int origoffer = offer; 2866 #ifdef INET6 2867 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE); 2868 size_t min_protoh = isipv6 ? 2869 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 2870 sizeof(struct tcpiphdr); 2871 #else 2872 const boolean_t isipv6 = FALSE; 2873 const size_t min_protoh = sizeof(struct tcpiphdr); 2874 #endif 2875 2876 if (isipv6) 2877 rt = tcp_rtlookup6(&inp->inp_inc); 2878 else 2879 rt = tcp_rtlookup(&inp->inp_inc); 2880 if (rt == NULL) { 2881 tp->t_maxopd = tp->t_maxseg = 2882 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2883 return; 2884 } 2885 ifp = rt->rt_ifp; 2886 so = inp->inp_socket; 2887 2888 taop = rmx_taop(rt->rt_rmx); 2889 /* 2890 * Offer == -1 means that we didn't receive SYN yet, 2891 * use cached value in that case; 2892 */ 2893 if (offer == -1) 2894 offer = taop->tao_mssopt; 2895 /* 2896 * Offer == 0 means that there was no MSS on the SYN segment, 2897 * in this case we use tcp_mssdflt. 2898 */ 2899 if (offer == 0) 2900 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2901 else 2902 /* 2903 * Sanity check: make sure that maxopd will be large 2904 * enough to allow some data on segments even is the 2905 * all the option space is used (40bytes). Otherwise 2906 * funny things may happen in tcp_output. 2907 */ 2908 offer = max(offer, 64); 2909 taop->tao_mssopt = offer; 2910 2911 /* 2912 * While we're here, check if there's an initial rtt 2913 * or rttvar. Convert from the route-table units 2914 * to scaled multiples of the slow timeout timer. 2915 */ 2916 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2917 /* 2918 * XXX the lock bit for RTT indicates that the value 2919 * is also a minimum value; this is subject to time. 2920 */ 2921 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2922 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2923 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2924 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2925 tcpstat.tcps_usedrtt++; 2926 if (rt->rt_rmx.rmx_rttvar) { 2927 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2928 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2929 tcpstat.tcps_usedrttvar++; 2930 } else { 2931 /* default variation is +- 1 rtt */ 2932 tp->t_rttvar = 2933 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2934 } 2935 TCPT_RANGESET(tp->t_rxtcur, 2936 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2937 tp->t_rttmin, TCPTV_REXMTMAX); 2938 } 2939 /* 2940 * if there's an mtu associated with the route, use it 2941 * else, use the link mtu. 2942 */ 2943 if (rt->rt_rmx.rmx_mtu) 2944 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2945 else { 2946 if (isipv6) { 2947 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh; 2948 if (!in6_localaddr(&inp->in6p_faddr)) 2949 mss = min(mss, tcp_v6mssdflt); 2950 } else { 2951 mss = ifp->if_mtu - min_protoh; 2952 if (!in_localaddr(inp->inp_faddr)) 2953 mss = min(mss, tcp_mssdflt); 2954 } 2955 } 2956 mss = min(mss, offer); 2957 /* 2958 * maxopd stores the maximum length of data AND options 2959 * in a segment; maxseg is the amount of data in a normal 2960 * segment. We need to store this value (maxopd) apart 2961 * from maxseg, because now every segment carries options 2962 * and thus we normally have somewhat less data in segments. 2963 */ 2964 tp->t_maxopd = mss; 2965 2966 /* 2967 * In case of T/TCP, origoffer==-1 indicates, that no segments 2968 * were received yet. In this case we just guess, otherwise 2969 * we do the same as before T/TCP. 2970 */ 2971 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP && 2972 (origoffer == -1 || 2973 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2974 mss -= TCPOLEN_TSTAMP_APPA; 2975 if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC && 2976 (origoffer == -1 || 2977 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2978 mss -= TCPOLEN_CC_APPA; 2979 2980 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2981 if (mss > MCLBYTES) 2982 mss &= ~(MCLBYTES-1); 2983 #else 2984 if (mss > MCLBYTES) 2985 mss = mss / MCLBYTES * MCLBYTES; 2986 #endif 2987 /* 2988 * If there's a pipesize, change the socket buffer 2989 * to that size. Make the socket buffers an integral 2990 * number of mss units; if the mss is larger than 2991 * the socket buffer, decrease the mss. 2992 */ 2993 #ifdef RTV_SPIPE 2994 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2995 #endif 2996 bufsize = so->so_snd.sb_hiwat; 2997 if (bufsize < mss) 2998 mss = bufsize; 2999 else { 3000 bufsize = roundup(bufsize, mss); 3001 if (bufsize > sb_max) 3002 bufsize = sb_max; 3003 if (bufsize > so->so_snd.sb_hiwat) 3004 sbreserve(&so->so_snd, bufsize, so, NULL); 3005 } 3006 tp->t_maxseg = mss; 3007 3008 #ifdef RTV_RPIPE 3009 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 3010 #endif 3011 bufsize = so->so_rcv.sb_hiwat; 3012 if (bufsize > mss) { 3013 bufsize = roundup(bufsize, mss); 3014 if (bufsize > sb_max) 3015 bufsize = sb_max; 3016 if (bufsize > so->so_rcv.sb_hiwat) 3017 sbreserve(&so->so_rcv, bufsize, so, NULL); 3018 } 3019 3020 /* 3021 * Set the slow-start flight size depending on whether this 3022 * is a local network or not. 3023 */ 3024 if (tcp_do_rfc3390) 3025 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3026 else 3027 tp->snd_cwnd = mss; 3028 3029 if (rt->rt_rmx.rmx_ssthresh) { 3030 /* 3031 * There's some sort of gateway or interface 3032 * buffer limit on the path. Use this to set 3033 * the slow start threshhold, but set the 3034 * threshold to no less than 2*mss. 3035 */ 3036 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 3037 tcpstat.tcps_usedssthresh++; 3038 } 3039 } 3040 3041 /* 3042 * Determine the MSS option to send on an outgoing SYN. 3043 */ 3044 int 3045 tcp_mssopt(struct tcpcb *tp) 3046 { 3047 struct rtentry *rt; 3048 #ifdef INET6 3049 boolean_t isipv6 = 3050 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE); 3051 int min_protoh = isipv6 ? 3052 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 3053 sizeof(struct tcpiphdr); 3054 #else 3055 const boolean_t isipv6 = FALSE; 3056 const size_t min_protoh = sizeof(struct tcpiphdr); 3057 #endif 3058 3059 if (isipv6) 3060 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 3061 else 3062 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 3063 if (rt == NULL) 3064 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 3065 3066 return (rt->rt_ifp->if_mtu - min_protoh); 3067 } 3068 3069 /* 3070 * When a partial ack arrives, force the retransmission of the 3071 * next unacknowledged segment. Do not exit Fast Recovery. 3072 * 3073 * Implement the Slow-but-Steady variant of NewReno by restarting the 3074 * the retransmission timer. Turn it off here so it can be restarted 3075 * later in tcp_output(). 3076 */ 3077 static void 3078 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked) 3079 { 3080 tcp_seq old_snd_nxt = tp->snd_nxt; 3081 u_long ocwnd = tp->snd_cwnd; 3082 3083 callout_stop(tp->tt_rexmt); 3084 tp->t_rtttime = 0; 3085 tp->snd_nxt = th->th_ack; 3086 /* Set snd_cwnd to one segment beyond acknowledged offset. */ 3087 tp->snd_cwnd = tp->t_maxseg; 3088 tp->t_flags |= TF_ACKNOW; 3089 tcp_output(tp); 3090 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 3091 tp->snd_nxt = old_snd_nxt; 3092 /* partial window deflation */ 3093 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg; 3094 } 3095 3096 /* 3097 * In contrast to the Slow-but-Steady NewReno variant, 3098 * we do not reset the retransmission timer for SACK retransmissions, 3099 * except when retransmitting snd_una. 3100 */ 3101 static void 3102 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th) 3103 { 3104 uint32_t pipe, seglen; 3105 tcp_seq nextrexmt; 3106 boolean_t lostdup; 3107 tcp_seq old_snd_nxt = tp->snd_nxt; 3108 u_long ocwnd = tp->snd_cwnd; 3109 int nseg = 0; /* consecutive new segments */ 3110 #define MAXBURST 4 /* limit burst of new packets on partial ack */ 3111 3112 tp->t_rtttime = 0; 3113 pipe = tcp_sack_compute_pipe(tp); 3114 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg && 3115 (!tcp_do_smartsack || nseg < MAXBURST) && 3116 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) { 3117 uint32_t sent; 3118 tcp_seq old_snd_max; 3119 int error; 3120 3121 if (nextrexmt == tp->snd_max) ++nseg; 3122 tp->snd_nxt = nextrexmt; 3123 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen; 3124 old_snd_max = tp->snd_max; 3125 if (nextrexmt == tp->snd_una) 3126 callout_stop(tp->tt_rexmt); 3127 error = tcp_output(tp); 3128 if (error != 0) 3129 break; 3130 sent = tp->snd_nxt - nextrexmt; 3131 if (sent <= 0) 3132 break; 3133 if (!lostdup) 3134 pipe += sent; 3135 tcpstat.tcps_sndsackpack++; 3136 tcpstat.tcps_sndsackbyte += sent; 3137 if (SEQ_LT(nextrexmt, old_snd_max) && 3138 SEQ_LT(tp->rexmt_high, tp->snd_nxt)) 3139 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max); 3140 } 3141 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 3142 tp->snd_nxt = old_snd_nxt; 3143 tp->snd_cwnd = ocwnd; 3144 } 3145