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 1.54 2005/03/09 06:57:29 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_do_rfc3390 = 1; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 184 &tcp_do_rfc3390, 0, 185 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 186 187 static int tcp_do_eifel_detect = 1; 188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW, 189 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)"); 190 191 /* 192 * Define as tunable for easy testing with SACK on and off. 193 * Warning: do not change setting in the middle of an existing active TCP flow, 194 * else strange things might happen to that flow. 195 */ 196 int tcp_do_sack = 1; 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 198 &tcp_do_sack, 0, "Enable SACK Algorithms"); 199 200 int tcp_do_smartsack = 1; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW, 202 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms"); 203 204 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 205 "TCP Segment Reassembly Queue"); 206 207 int tcp_reass_maxseg = 0; 208 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD, 209 &tcp_reass_maxseg, 0, 210 "Global maximum number of TCP Segments in Reassembly Queue"); 211 212 int tcp_reass_qsize = 0; 213 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 214 &tcp_reass_qsize, 0, 215 "Global number of TCP Segments currently in Reassembly Queue"); 216 217 static int tcp_reass_overflows = 0; 218 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 219 &tcp_reass_overflows, 0, 220 "Global number of TCP Segment Reassembly Queue Overflows"); 221 222 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t); 223 static void tcp_pulloutofband(struct socket *, 224 struct tcphdr *, struct mbuf *, int); 225 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 226 struct mbuf *); 227 static void tcp_xmit_timer(struct tcpcb *, int); 228 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int); 229 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *); 230 231 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 232 #ifdef INET6 233 #define ND6_HINT(tp) \ 234 do { \ 235 if ((tp) && (tp)->t_inpcb && \ 236 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \ 237 (tp)->t_inpcb->in6p_route.ro_rt) \ 238 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 239 } while (0) 240 #else 241 #define ND6_HINT(tp) 242 #endif 243 244 /* 245 * Indicate whether this ack should be delayed. We can delay the ack if 246 * - delayed acks are enabled and 247 * - there is no delayed ack timer in progress and 248 * - our last ack wasn't a 0-sized window. We never want to delay 249 * the ack that opens up a 0-sized window. 250 */ 251 #define DELAY_ACK(tp) \ 252 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 253 !(tp->t_flags & TF_RXWIN0SENT)) 254 255 #define acceptable_window_update(tp, th, tiwin) \ 256 (SEQ_LT(tp->snd_wl1, th->th_seq) || \ 257 (tp->snd_wl1 == th->th_seq && \ 258 (SEQ_LT(tp->snd_wl2, th->th_ack) || \ 259 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) 260 261 static int 262 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m) 263 { 264 struct tseg_qent *q; 265 struct tseg_qent *p = NULL; 266 struct tseg_qent *te; 267 struct socket *so = tp->t_inpcb->inp_socket; 268 int flags; 269 270 /* 271 * Call with th == NULL after become established to 272 * force pre-ESTABLISHED data up to user socket. 273 */ 274 if (th == NULL) 275 goto present; 276 277 /* 278 * Limit the number of segments in the reassembly queue to prevent 279 * holding on to too many segments (and thus running out of mbufs). 280 * Make sure to let the missing segment through which caused this 281 * queue. Always keep one global queue entry spare to be able to 282 * process the missing segment. 283 */ 284 if (th->th_seq != tp->rcv_nxt && 285 tcp_reass_qsize + 1 >= tcp_reass_maxseg) { 286 tcp_reass_overflows++; 287 tcpstat.tcps_rcvmemdrop++; 288 m_freem(m); 289 /* no SACK block to report */ 290 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 291 return (0); 292 } 293 294 /* Allocate a new queue entry. */ 295 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ, 296 M_INTWAIT | M_NULLOK); 297 if (te == NULL) { 298 tcpstat.tcps_rcvmemdrop++; 299 m_freem(m); 300 /* no SACK block to report */ 301 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 302 return (0); 303 } 304 tcp_reass_qsize++; 305 306 /* 307 * Find a segment which begins after this one does. 308 */ 309 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 310 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 311 break; 312 p = q; 313 } 314 315 /* 316 * If there is a preceding segment, it may provide some of 317 * our data already. If so, drop the data from the incoming 318 * segment. If it provides all of our data, drop us. 319 */ 320 if (p != NULL) { 321 tcp_seq_diff_t i; 322 323 /* conversion to int (in i) handles seq wraparound */ 324 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 325 if (i > 0) { /* overlaps preceding segment */ 326 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG); 327 /* enclosing block starts w/ preceding segment */ 328 tp->encloseblk.rblk_start = p->tqe_th->th_seq; 329 if (i >= *tlenp) { 330 /* preceding encloses incoming segment */ 331 tp->encloseblk.rblk_end = p->tqe_th->th_seq + 332 p->tqe_len; 333 tcpstat.tcps_rcvduppack++; 334 tcpstat.tcps_rcvdupbyte += *tlenp; 335 m_freem(m); 336 free(te, M_TSEGQ); 337 tcp_reass_qsize--; 338 /* 339 * Try to present any queued data 340 * at the left window edge to the user. 341 * This is needed after the 3-WHS 342 * completes. 343 */ 344 goto present; /* ??? */ 345 } 346 m_adj(m, i); 347 *tlenp -= i; 348 th->th_seq += i; 349 /* incoming segment end is enclosing block end */ 350 tp->encloseblk.rblk_end = th->th_seq + *tlenp + 351 ((th->th_flags & TH_FIN) != 0); 352 /* trim end of reported D-SACK block */ 353 tp->reportblk.rblk_end = th->th_seq; 354 } 355 } 356 tcpstat.tcps_rcvoopack++; 357 tcpstat.tcps_rcvoobyte += *tlenp; 358 359 /* 360 * While we overlap succeeding segments trim them or, 361 * if they are completely covered, dequeue them. 362 */ 363 while (q) { 364 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 365 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len; 366 struct tseg_qent *nq; 367 368 if (i <= 0) 369 break; 370 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */ 371 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG); 372 tp->encloseblk = tp->reportblk; 373 /* report trailing duplicate D-SACK segment */ 374 tp->reportblk.rblk_start = q->tqe_th->th_seq; 375 } 376 if ((tp->t_flags & TF_ENCLOSESEG) && 377 SEQ_GT(qend, tp->encloseblk.rblk_end)) { 378 /* extend enclosing block if one exists */ 379 tp->encloseblk.rblk_end = qend; 380 } 381 if (i < q->tqe_len) { 382 q->tqe_th->th_seq += i; 383 q->tqe_len -= i; 384 m_adj(q->tqe_m, i); 385 break; 386 } 387 388 nq = LIST_NEXT(q, tqe_q); 389 LIST_REMOVE(q, tqe_q); 390 m_freem(q->tqe_m); 391 free(q, M_TSEGQ); 392 tcp_reass_qsize--; 393 q = nq; 394 } 395 396 /* Insert the new segment queue entry into place. */ 397 te->tqe_m = m; 398 te->tqe_th = th; 399 te->tqe_len = *tlenp; 400 401 /* check if can coalesce with following segment */ 402 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) { 403 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len; 404 405 te->tqe_len += q->tqe_len; 406 if (q->tqe_th->th_flags & TH_FIN) 407 te->tqe_th->th_flags |= TH_FIN; 408 m_cat(te->tqe_m, q->tqe_m); 409 tp->encloseblk.rblk_end = tend; 410 /* 411 * When not reporting a duplicate segment, use 412 * the larger enclosing block as the SACK block. 413 */ 414 if (!(tp->t_flags & TF_DUPSEG)) 415 tp->reportblk.rblk_end = tend; 416 LIST_REMOVE(q, tqe_q); 417 free(q, M_TSEGQ); 418 tcp_reass_qsize--; 419 } 420 421 if (p == NULL) { 422 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 423 } else { 424 /* check if can coalesce with preceding segment */ 425 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) { 426 p->tqe_len += te->tqe_len; 427 m_cat(p->tqe_m, te->tqe_m); 428 tp->encloseblk.rblk_start = p->tqe_th->th_seq; 429 /* 430 * When not reporting a duplicate segment, use 431 * the larger enclosing block as the SACK block. 432 */ 433 if (!(tp->t_flags & TF_DUPSEG)) 434 tp->reportblk.rblk_start = p->tqe_th->th_seq; 435 free(te, M_TSEGQ); 436 tcp_reass_qsize--; 437 } else 438 LIST_INSERT_AFTER(p, te, tqe_q); 439 } 440 441 present: 442 /* 443 * Present data to user, advancing rcv_nxt through 444 * completed sequence space. 445 */ 446 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 447 return (0); 448 q = LIST_FIRST(&tp->t_segq); 449 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt) 450 return (0); 451 tp->rcv_nxt += q->tqe_len; 452 if (!(tp->t_flags & TF_DUPSEG)) { 453 /* no SACK block to report since ACK advanced */ 454 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 455 } 456 /* no enclosing block to report since ACK advanced */ 457 tp->t_flags &= ~TF_ENCLOSESEG; 458 flags = q->tqe_th->th_flags & TH_FIN; 459 LIST_REMOVE(q, tqe_q); 460 KASSERT(LIST_EMPTY(&tp->t_segq) || 461 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt, 462 ("segment not coalesced")); 463 if (so->so_state & SS_CANTRCVMORE) 464 m_freem(q->tqe_m); 465 else 466 sbappendstream(&so->so_rcv, q->tqe_m); 467 free(q, M_TSEGQ); 468 tcp_reass_qsize--; 469 ND6_HINT(tp); 470 sorwakeup(so); 471 return (flags); 472 } 473 474 /* 475 * TCP input routine, follows pages 65-76 of the 476 * protocol specification dated September, 1981 very closely. 477 */ 478 #ifdef INET6 479 int 480 tcp6_input(struct mbuf **mp, int *offp, int proto) 481 { 482 struct mbuf *m = *mp; 483 struct in6_ifaddr *ia6; 484 485 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 486 487 /* 488 * draft-itojun-ipv6-tcp-to-anycast 489 * better place to put this in? 490 */ 491 ia6 = ip6_getdstifaddr(m); 492 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 493 struct ip6_hdr *ip6; 494 495 ip6 = mtod(m, struct ip6_hdr *); 496 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 497 offsetof(struct ip6_hdr, ip6_dst)); 498 return (IPPROTO_DONE); 499 } 500 501 tcp_input(m, *offp, proto); 502 return (IPPROTO_DONE); 503 } 504 #endif 505 506 void 507 tcp_input(struct mbuf *m, ...) 508 { 509 __va_list ap; 510 int off0, proto; 511 struct tcphdr *th; 512 struct ip *ip = NULL; 513 struct ipovly *ipov; 514 struct inpcb *inp = NULL; 515 u_char *optp = NULL; 516 int optlen = 0; 517 int len, tlen, off; 518 int drop_hdrlen; 519 struct tcpcb *tp = NULL; 520 int thflags; 521 struct socket *so = 0; 522 int todrop, acked; 523 boolean_t ourfinisacked, needoutput = FALSE; 524 u_long tiwin; 525 int recvwin; 526 struct tcpopt to; /* options in this segment */ 527 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 528 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 529 struct sockaddr_in *next_hop = NULL; 530 int rstreason; /* For badport_bandlim accounting purposes */ 531 int cpu; 532 struct ip6_hdr *ip6 = NULL; 533 #ifdef INET6 534 boolean_t isipv6; 535 #else 536 const boolean_t isipv6 = FALSE; 537 #endif 538 #ifdef TCPDEBUG 539 short ostate = 0; 540 #endif 541 542 __va_start(ap, m); 543 off0 = __va_arg(ap, int); 544 proto = __va_arg(ap, int); 545 __va_end(ap); 546 547 tcpstat.tcps_rcvtotal++; 548 549 /* Grab info from and strip MT_TAG mbufs prepended to the chain. */ 550 while (m->m_type == MT_TAG) { 551 if (m->_m_tag_id == PACKET_TAG_IPFORWARD) 552 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data; 553 m = m->m_next; 554 } 555 556 #ifdef INET6 557 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE; 558 #endif 559 560 if (isipv6) { 561 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 562 ip6 = mtod(m, struct ip6_hdr *); 563 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0; 564 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 565 tcpstat.tcps_rcvbadsum++; 566 goto drop; 567 } 568 th = (struct tcphdr *)((caddr_t)ip6 + off0); 569 570 /* 571 * Be proactive about unspecified IPv6 address in source. 572 * As we use all-zero to indicate unbounded/unconnected pcb, 573 * unspecified IPv6 address can be used to confuse us. 574 * 575 * Note that packets with unspecified IPv6 destination is 576 * already dropped in ip6_input. 577 */ 578 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 579 /* XXX stat */ 580 goto drop; 581 } 582 } else { 583 /* 584 * Get IP and TCP header together in first mbuf. 585 * Note: IP leaves IP header in first mbuf. 586 */ 587 if (off0 > sizeof(struct ip)) { 588 ip_stripoptions(m); 589 off0 = sizeof(struct ip); 590 } 591 /* already checked and pulled up in ip_demux() */ 592 KASSERT(m->m_len >= sizeof(struct tcpiphdr), 593 ("TCP header not in one mbuf: m->m_len %d", m->m_len)); 594 ip = mtod(m, struct ip *); 595 ipov = (struct ipovly *)ip; 596 th = (struct tcphdr *)((caddr_t)ip + off0); 597 tlen = ip->ip_len; 598 599 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 600 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 601 th->th_sum = m->m_pkthdr.csum_data; 602 else 603 th->th_sum = in_pseudo(ip->ip_src.s_addr, 604 ip->ip_dst.s_addr, 605 htonl(m->m_pkthdr.csum_data + 606 ip->ip_len + 607 IPPROTO_TCP)); 608 th->th_sum ^= 0xffff; 609 } else { 610 /* 611 * Checksum extended TCP header and data. 612 */ 613 len = sizeof(struct ip) + tlen; 614 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 615 ipov->ih_len = (u_short)tlen; 616 ipov->ih_len = htons(ipov->ih_len); 617 th->th_sum = in_cksum(m, len); 618 } 619 if (th->th_sum) { 620 tcpstat.tcps_rcvbadsum++; 621 goto drop; 622 } 623 #ifdef INET6 624 /* Re-initialization for later version check */ 625 ip->ip_v = IPVERSION; 626 #endif 627 } 628 629 /* 630 * Check that TCP offset makes sense, 631 * pull out TCP options and adjust length. XXX 632 */ 633 off = th->th_off << 2; 634 /* already checked and pulled up in ip_demux() */ 635 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen, 636 ("bad TCP data offset %d (tlen %d)", off, tlen)); 637 tlen -= off; /* tlen is used instead of ti->ti_len */ 638 if (off > sizeof(struct tcphdr)) { 639 if (isipv6) { 640 IP6_EXTHDR_CHECK(m, off0, off, ); 641 ip6 = mtod(m, struct ip6_hdr *); 642 th = (struct tcphdr *)((caddr_t)ip6 + off0); 643 } else { 644 /* already pulled up in ip_demux() */ 645 KASSERT(m->m_len >= sizeof(struct ip) + off, 646 ("TCP header and options not in one mbuf: " 647 "m_len %d, off %d", m->m_len, off)); 648 } 649 optlen = off - sizeof(struct tcphdr); 650 optp = (u_char *)(th + 1); 651 } 652 thflags = th->th_flags; 653 654 #ifdef TCP_DROP_SYNFIN 655 /* 656 * If the drop_synfin option is enabled, drop all packets with 657 * both the SYN and FIN bits set. This prevents e.g. nmap from 658 * identifying the TCP/IP stack. 659 * 660 * This is a violation of the TCP specification. 661 */ 662 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN)) 663 goto drop; 664 #endif 665 666 /* 667 * Convert TCP protocol specific fields to host format. 668 */ 669 th->th_seq = ntohl(th->th_seq); 670 th->th_ack = ntohl(th->th_ack); 671 th->th_win = ntohs(th->th_win); 672 th->th_urp = ntohs(th->th_urp); 673 674 /* 675 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 676 * until after ip6_savecontrol() is called and before other functions 677 * which don't want those proto headers. 678 * Because ip6_savecontrol() is going to parse the mbuf to 679 * search for data to be passed up to user-land, it wants mbuf 680 * parameters to be unchanged. 681 * XXX: the call of ip6_savecontrol() has been obsoleted based on 682 * latest version of the advanced API (20020110). 683 */ 684 drop_hdrlen = off0 + off; 685 686 /* 687 * Locate pcb for segment. 688 */ 689 findpcb: 690 /* IPFIREWALL_FORWARD section */ 691 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */ 692 /* 693 * Transparently forwarded. Pretend to be the destination. 694 * already got one like this? 695 */ 696 cpu = mycpu->gd_cpuid; 697 inp = in_pcblookup_hash(&tcbinfo[cpu], 698 ip->ip_src, th->th_sport, 699 ip->ip_dst, th->th_dport, 700 0, m->m_pkthdr.rcvif); 701 if (!inp) { 702 /* 703 * It's new. Try to find the ambushing socket. 704 */ 705 706 /* 707 * The rest of the ipfw code stores the port in 708 * host order. XXX 709 * (The IP address is still in network order.) 710 */ 711 in_port_t dport = next_hop->sin_port ? 712 htons(next_hop->sin_port) : 713 th->th_dport; 714 715 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport, 716 next_hop->sin_addr.s_addr, dport); 717 inp = in_pcblookup_hash(&tcbinfo[cpu], 718 ip->ip_src, th->th_sport, 719 next_hop->sin_addr, dport, 720 1, m->m_pkthdr.rcvif); 721 } 722 } else { 723 if (isipv6) { 724 inp = in6_pcblookup_hash(&tcbinfo[0], 725 &ip6->ip6_src, th->th_sport, 726 &ip6->ip6_dst, th->th_dport, 727 1, m->m_pkthdr.rcvif); 728 } else { 729 cpu = mycpu->gd_cpuid; 730 inp = in_pcblookup_hash(&tcbinfo[cpu], 731 ip->ip_src, th->th_sport, 732 ip->ip_dst, th->th_dport, 733 1, m->m_pkthdr.rcvif); 734 } 735 } 736 737 /* 738 * If the state is CLOSED (i.e., TCB does not exist) then 739 * all data in the incoming segment is discarded. 740 * If the TCB exists but is in CLOSED state, it is embryonic, 741 * but should either do a listen or a connect soon. 742 */ 743 if (inp == NULL) { 744 if (log_in_vain) { 745 #ifdef INET6 746 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 747 #else 748 char dbuf[sizeof "aaa.bbb.ccc.ddd"]; 749 char sbuf[sizeof "aaa.bbb.ccc.ddd"]; 750 #endif 751 if (isipv6) { 752 strcpy(dbuf, "["); 753 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 754 strcat(dbuf, "]"); 755 strcpy(sbuf, "["); 756 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 757 strcat(sbuf, "]"); 758 } else { 759 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 760 strcpy(sbuf, inet_ntoa(ip->ip_src)); 761 } 762 switch (log_in_vain) { 763 case 1: 764 if (!(thflags & TH_SYN)) 765 break; 766 case 2: 767 log(LOG_INFO, 768 "Connection attempt to TCP %s:%d " 769 "from %s:%d flags:0x%02x\n", 770 dbuf, ntohs(th->th_dport), sbuf, 771 ntohs(th->th_sport), thflags); 772 break; 773 default: 774 break; 775 } 776 } 777 if (blackhole) { 778 switch (blackhole) { 779 case 1: 780 if (thflags & TH_SYN) 781 goto drop; 782 break; 783 case 2: 784 goto drop; 785 default: 786 goto drop; 787 } 788 } 789 rstreason = BANDLIM_RST_CLOSEDPORT; 790 goto dropwithreset; 791 } 792 793 #ifdef IPSEC 794 if (isipv6) { 795 if (ipsec6_in_reject_so(m, inp->inp_socket)) { 796 ipsec6stat.in_polvio++; 797 goto drop; 798 } 799 } else { 800 if (ipsec4_in_reject_so(m, inp->inp_socket)) { 801 ipsecstat.in_polvio++; 802 goto drop; 803 } 804 } 805 #endif 806 #ifdef FAST_IPSEC 807 if (isipv6) { 808 if (ipsec6_in_reject(m, inp)) 809 goto drop; 810 } else { 811 if (ipsec4_in_reject(m, inp)) 812 goto drop; 813 } 814 #endif 815 816 tp = intotcpcb(inp); 817 if (tp == NULL) { 818 rstreason = BANDLIM_RST_CLOSEDPORT; 819 goto dropwithreset; 820 } 821 if (tp->t_state <= TCPS_CLOSED) 822 goto drop; 823 824 /* Unscale the window into a 32-bit value. */ 825 if (!(thflags & TH_SYN)) 826 tiwin = th->th_win << tp->snd_scale; 827 else 828 tiwin = th->th_win; 829 830 so = inp->inp_socket; 831 832 #ifdef TCPDEBUG 833 if (so->so_options & SO_DEBUG) { 834 ostate = tp->t_state; 835 if (isipv6) 836 bcopy(ip6, tcp_saveipgen, sizeof(*ip6)); 837 else 838 bcopy(ip, tcp_saveipgen, sizeof(*ip)); 839 tcp_savetcp = *th; 840 } 841 #endif 842 843 bzero(&to, sizeof to); 844 845 if (so->so_options & SO_ACCEPTCONN) { 846 struct in_conninfo inc; 847 848 #ifdef INET6 849 inc.inc_isipv6 = (isipv6 == TRUE); 850 #endif 851 if (isipv6) { 852 inc.inc6_faddr = ip6->ip6_src; 853 inc.inc6_laddr = ip6->ip6_dst; 854 inc.inc6_route.ro_rt = NULL; /* XXX */ 855 } else { 856 inc.inc_faddr = ip->ip_src; 857 inc.inc_laddr = ip->ip_dst; 858 inc.inc_route.ro_rt = NULL; /* XXX */ 859 } 860 inc.inc_fport = th->th_sport; 861 inc.inc_lport = th->th_dport; 862 863 /* 864 * If the state is LISTEN then ignore segment if it contains 865 * a RST. If the segment contains an ACK then it is bad and 866 * send a RST. If it does not contain a SYN then it is not 867 * interesting; drop it. 868 * 869 * If the state is SYN_RECEIVED (syncache) and seg contains 870 * an ACK, but not for our SYN/ACK, send a RST. If the seg 871 * contains a RST, check the sequence number to see if it 872 * is a valid reset segment. 873 */ 874 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) { 875 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) { 876 if (!syncache_expand(&inc, th, &so, m)) { 877 /* 878 * No syncache entry, or ACK was not 879 * for our SYN/ACK. Send a RST. 880 */ 881 tcpstat.tcps_badsyn++; 882 rstreason = BANDLIM_RST_OPENPORT; 883 goto dropwithreset; 884 } 885 if (so == NULL) 886 /* 887 * Could not complete 3-way handshake, 888 * connection is being closed down, and 889 * syncache will free mbuf. 890 */ 891 return; 892 /* 893 * Socket is created in state SYN_RECEIVED. 894 * Continue processing segment. 895 */ 896 inp = so->so_pcb; 897 tp = intotcpcb(inp); 898 /* 899 * This is what would have happened in 900 * tcp_output() when the SYN,ACK was sent. 901 */ 902 tp->snd_up = tp->snd_una; 903 tp->snd_max = tp->snd_nxt = tp->iss + 1; 904 tp->last_ack_sent = tp->rcv_nxt; 905 /* 906 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 907 * until the _second_ ACK is received: 908 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 909 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 910 * move to ESTAB, set snd_wnd to tiwin. 911 */ 912 tp->snd_wnd = tiwin; /* unscaled */ 913 goto after_listen; 914 } 915 if (thflags & TH_RST) { 916 syncache_chkrst(&inc, th); 917 goto drop; 918 } 919 if (thflags & TH_ACK) { 920 syncache_badack(&inc); 921 tcpstat.tcps_badsyn++; 922 rstreason = BANDLIM_RST_OPENPORT; 923 goto dropwithreset; 924 } 925 goto drop; 926 } 927 928 /* 929 * Segment's flags are (SYN) or (SYN | FIN). 930 */ 931 #ifdef INET6 932 /* 933 * If deprecated address is forbidden, 934 * we do not accept SYN to deprecated interface 935 * address to prevent any new inbound connection from 936 * getting established. 937 * When we do not accept SYN, we send a TCP RST, 938 * with deprecated source address (instead of dropping 939 * it). We compromise it as it is much better for peer 940 * to send a RST, and RST will be the final packet 941 * for the exchange. 942 * 943 * If we do not forbid deprecated addresses, we accept 944 * the SYN packet. RFC2462 does not suggest dropping 945 * SYN in this case. 946 * If we decipher RFC2462 5.5.4, it says like this: 947 * 1. use of deprecated addr with existing 948 * communication is okay - "SHOULD continue to be 949 * used" 950 * 2. use of it with new communication: 951 * (2a) "SHOULD NOT be used if alternate address 952 * with sufficient scope is available" 953 * (2b) nothing mentioned otherwise. 954 * Here we fall into (2b) case as we have no choice in 955 * our source address selection - we must obey the peer. 956 * 957 * The wording in RFC2462 is confusing, and there are 958 * multiple description text for deprecated address 959 * handling - worse, they are not exactly the same. 960 * I believe 5.5.4 is the best one, so we follow 5.5.4. 961 */ 962 if (isipv6 && !ip6_use_deprecated) { 963 struct in6_ifaddr *ia6; 964 965 if ((ia6 = ip6_getdstifaddr(m)) && 966 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 967 tp = NULL; 968 rstreason = BANDLIM_RST_OPENPORT; 969 goto dropwithreset; 970 } 971 } 972 #endif 973 /* 974 * If it is from this socket, drop it, it must be forged. 975 * Don't bother responding if the destination was a broadcast. 976 */ 977 if (th->th_dport == th->th_sport) { 978 if (isipv6) { 979 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 980 &ip6->ip6_src)) 981 goto drop; 982 } else { 983 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 984 goto drop; 985 } 986 } 987 /* 988 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 989 * 990 * Note that it is quite possible to receive unicast 991 * link-layer packets with a broadcast IP address. Use 992 * in_broadcast() to find them. 993 */ 994 if (m->m_flags & (M_BCAST | M_MCAST)) 995 goto drop; 996 if (isipv6) { 997 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 998 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 999 goto drop; 1000 } else { 1001 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1002 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1003 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1004 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1005 goto drop; 1006 } 1007 /* 1008 * SYN appears to be valid; create compressed TCP state 1009 * for syncache, or perform t/tcp connection. 1010 */ 1011 if (so->so_qlen <= so->so_qlimit) { 1012 tcp_dooptions(&to, optp, optlen, TRUE); 1013 if (!syncache_add(&inc, &to, th, &so, m)) 1014 goto drop; 1015 if (so == NULL) 1016 /* 1017 * Entry added to syncache, mbuf used to 1018 * send SYN,ACK packet. 1019 */ 1020 return; 1021 /* 1022 * Segment passed TAO tests. 1023 */ 1024 inp = so->so_pcb; 1025 tp = intotcpcb(inp); 1026 tp->snd_wnd = tiwin; 1027 tp->t_starttime = ticks; 1028 tp->t_state = TCPS_ESTABLISHED; 1029 1030 /* 1031 * If there is a FIN, or if there is data and the 1032 * connection is local, then delay SYN,ACK(SYN) in 1033 * the hope of piggy-backing it on a response 1034 * segment. Otherwise must send ACK now in case 1035 * the other side is slow starting. 1036 */ 1037 if (DELAY_ACK(tp) && 1038 ((thflags & TH_FIN) || 1039 (tlen != 0 && 1040 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 1041 (!isipv6 && in_localaddr(inp->inp_faddr)))))) { 1042 callout_reset(tp->tt_delack, tcp_delacktime, 1043 tcp_timer_delack, tp); 1044 tp->t_flags |= TF_NEEDSYN; 1045 } else 1046 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1047 1048 tcpstat.tcps_connects++; 1049 soisconnected(so); 1050 goto trimthenstep6; 1051 } 1052 goto drop; 1053 } 1054 after_listen: 1055 1056 /* should not happen - syncache should pick up these connections */ 1057 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state")); 1058 1059 /* 1060 * Segment received on connection. 1061 * Reset idle time and keep-alive timer. 1062 */ 1063 tp->t_rcvtime = ticks; 1064 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1065 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1066 1067 /* 1068 * Process options. 1069 * XXX this is tradtitional behavior, may need to be cleaned up. 1070 */ 1071 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0); 1072 if (thflags & TH_SYN) { 1073 if (to.to_flags & TOF_SCALE) { 1074 tp->t_flags |= TF_RCVD_SCALE; 1075 tp->requested_s_scale = to.to_requested_s_scale; 1076 } 1077 if (to.to_flags & TOF_TS) { 1078 tp->t_flags |= TF_RCVD_TSTMP; 1079 tp->ts_recent = to.to_tsval; 1080 tp->ts_recent_age = ticks; 1081 } 1082 if (to.to_flags & (TOF_CC | TOF_CCNEW)) 1083 tp->t_flags |= TF_RCVD_CC; 1084 if (to.to_flags & TOF_MSS) 1085 tcp_mss(tp, to.to_mss); 1086 /* 1087 * Only set the TF_SACK_PERMITTED per-connection flag 1088 * if we got a SACK_PERMITTED option from the other side 1089 * and the global tcp_do_sack variable is true. 1090 */ 1091 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED)) 1092 tp->t_flags |= TF_SACK_PERMITTED; 1093 } 1094 1095 /* 1096 * Header prediction: check for the two common cases 1097 * of a uni-directional data xfer. If the packet has 1098 * no control flags, is in-sequence, the window didn't 1099 * change and we're not retransmitting, it's a 1100 * candidate. If the length is zero and the ack moved 1101 * forward, we're the sender side of the xfer. Just 1102 * free the data acked & wake any higher level process 1103 * that was blocked waiting for space. If the length 1104 * is non-zero and the ack didn't move, we're the 1105 * receiver side. If we're getting packets in-order 1106 * (the reassembly queue is empty), add the data to 1107 * the socket buffer and note that we need a delayed ack. 1108 * Make sure that the hidden state-flags are also off. 1109 * Since we check for TCPS_ESTABLISHED above, it can only 1110 * be TH_NEEDSYN. 1111 */ 1112 if (tp->t_state == TCPS_ESTABLISHED && 1113 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1114 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) && 1115 (!(to.to_flags & TOF_TS) || 1116 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1117 /* 1118 * Using the CC option is compulsory if once started: 1119 * the segment is OK if no T/TCP was negotiated or 1120 * if the segment has a CC option equal to CCrecv 1121 */ 1122 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 1123 ((to.to_flags & TOF_CC) && to.to_cc == tp->cc_recv)) && 1124 th->th_seq == tp->rcv_nxt && 1125 tp->snd_nxt == tp->snd_max) { 1126 1127 /* 1128 * If last ACK falls within this segment's sequence numbers, 1129 * record the timestamp. 1130 * NOTE that the test is modified according to the latest 1131 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1132 */ 1133 if ((to.to_flags & TOF_TS) && 1134 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1135 tp->ts_recent_age = ticks; 1136 tp->ts_recent = to.to_tsval; 1137 } 1138 1139 if (tlen == 0) { 1140 if (SEQ_GT(th->th_ack, tp->snd_una) && 1141 SEQ_LEQ(th->th_ack, tp->snd_max) && 1142 tp->snd_cwnd >= tp->snd_wnd && 1143 !IN_FASTRECOVERY(tp)) { 1144 /* 1145 * This is a pure ack for outstanding data. 1146 */ 1147 ++tcpstat.tcps_predack; 1148 /* 1149 * "bad retransmit" recovery 1150 * 1151 * If Eifel detection applies, then 1152 * it is deterministic, so use it 1153 * unconditionally over the old heuristic. 1154 * Otherwise, fall back to the old heuristic. 1155 */ 1156 if (tcp_do_eifel_detect && 1157 (to.to_flags & TOF_TS) && to.to_tsecr && 1158 (tp->t_flags & TF_FIRSTACCACK)) { 1159 /* Eifel detection applicable. */ 1160 if (to.to_tsecr < tp->t_rexmtTS) { 1161 tcp_revert_congestion_state(tp); 1162 ++tcpstat.tcps_eifeldetected; 1163 } 1164 } else if (tp->t_rxtshift == 1 && 1165 ticks < tp->t_badrxtwin) { 1166 tcp_revert_congestion_state(tp); 1167 ++tcpstat.tcps_rttdetected; 1168 } 1169 tp->t_flags &= ~(TF_FIRSTACCACK | 1170 TF_FASTREXMT | TF_EARLYREXMT); 1171 /* 1172 * Recalculate the retransmit timer / rtt. 1173 * 1174 * Some machines (certain windows boxes) 1175 * send broken timestamp replies during the 1176 * SYN+ACK phase, ignore timestamps of 0. 1177 */ 1178 if ((to.to_flags & TOF_TS) && to.to_tsecr) { 1179 tcp_xmit_timer(tp, 1180 ticks - to.to_tsecr + 1); 1181 } else if (tp->t_rtttime && 1182 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1183 tcp_xmit_timer(tp, 1184 ticks - tp->t_rtttime); 1185 } 1186 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1187 acked = th->th_ack - tp->snd_una; 1188 tcpstat.tcps_rcvackpack++; 1189 tcpstat.tcps_rcvackbyte += acked; 1190 sbdrop(&so->so_snd, acked); 1191 tp->snd_recover = th->th_ack - 1; 1192 tp->snd_una = th->th_ack; 1193 tp->t_dupacks = 0; 1194 /* 1195 * Update window information. 1196 */ 1197 if (tiwin != tp->snd_wnd && 1198 acceptable_window_update(tp, th, tiwin)) { 1199 /* keep track of pure window updates */ 1200 if (tp->snd_wl2 == th->th_ack && 1201 tiwin > tp->snd_wnd) 1202 tcpstat.tcps_rcvwinupd++; 1203 tp->snd_wnd = tiwin; 1204 tp->snd_wl1 = th->th_seq; 1205 tp->snd_wl2 = th->th_ack; 1206 if (tp->snd_wnd > tp->max_sndwnd) 1207 tp->max_sndwnd = tp->snd_wnd; 1208 } 1209 m_freem(m); 1210 ND6_HINT(tp); /* some progress has been done */ 1211 /* 1212 * If all outstanding data are acked, stop 1213 * retransmit timer, otherwise restart timer 1214 * using current (possibly backed-off) value. 1215 * If process is waiting for space, 1216 * wakeup/selwakeup/signal. If data 1217 * are ready to send, let tcp_output 1218 * decide between more output or persist. 1219 */ 1220 if (tp->snd_una == tp->snd_max) 1221 callout_stop(tp->tt_rexmt); 1222 else if (!callout_active(tp->tt_persist)) 1223 callout_reset(tp->tt_rexmt, 1224 tp->t_rxtcur, 1225 tcp_timer_rexmt, tp); 1226 sowwakeup(so); 1227 if (so->so_snd.sb_cc > 0) 1228 tcp_output(tp); 1229 return; 1230 } 1231 } else if (tiwin == tp->snd_wnd && 1232 th->th_ack == tp->snd_una && 1233 LIST_EMPTY(&tp->t_segq) && 1234 tlen <= sbspace(&so->so_rcv)) { 1235 /* 1236 * This is a pure, in-sequence data packet 1237 * with nothing on the reassembly queue and 1238 * we have enough buffer space to take it. 1239 */ 1240 ++tcpstat.tcps_preddat; 1241 tp->rcv_nxt += tlen; 1242 tcpstat.tcps_rcvpack++; 1243 tcpstat.tcps_rcvbyte += tlen; 1244 ND6_HINT(tp); /* some progress has been done */ 1245 /* 1246 * Add data to socket buffer. 1247 */ 1248 if (so->so_state & SS_CANTRCVMORE) { 1249 m_freem(m); 1250 } else { 1251 m_adj(m, drop_hdrlen); /* delayed header drop */ 1252 sbappendstream(&so->so_rcv, m); 1253 } 1254 sorwakeup(so); 1255 /* 1256 * This code is responsible for most of the ACKs 1257 * the TCP stack sends back after receiving a data 1258 * packet. Note that the DELAY_ACK check fails if 1259 * the delack timer is already running, which results 1260 * in an ack being sent every other packet (which is 1261 * what we want). 1262 */ 1263 if (DELAY_ACK(tp)) { 1264 callout_reset(tp->tt_delack, tcp_delacktime, 1265 tcp_timer_delack, tp); 1266 } else { 1267 tp->t_flags |= TF_ACKNOW; 1268 if (!(tp->t_flags & TF_ONOUTPUTQ)) { 1269 tp->t_flags |= TF_ONOUTPUTQ; 1270 tp->tt_cpu = mycpu->gd_cpuid; 1271 TAILQ_INSERT_TAIL( 1272 &tcpcbackq[tp->tt_cpu], 1273 tp, t_outputq); 1274 } 1275 } 1276 return; 1277 } 1278 } 1279 1280 /* 1281 * Calculate amount of space in receive window, 1282 * and then do TCP input processing. 1283 * Receive window is amount of space in rcv queue, 1284 * but not less than advertised window. 1285 */ 1286 recvwin = sbspace(&so->so_rcv); 1287 if (recvwin < 0) 1288 recvwin = 0; 1289 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt)); 1290 1291 switch (tp->t_state) { 1292 /* 1293 * If the state is SYN_RECEIVED: 1294 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1295 */ 1296 case TCPS_SYN_RECEIVED: 1297 if ((thflags & TH_ACK) && 1298 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1299 SEQ_GT(th->th_ack, tp->snd_max))) { 1300 rstreason = BANDLIM_RST_OPENPORT; 1301 goto dropwithreset; 1302 } 1303 break; 1304 1305 /* 1306 * If the state is SYN_SENT: 1307 * if seg contains an ACK, but not for our SYN, drop the input. 1308 * if seg contains a RST, then drop the connection. 1309 * if seg does not contain SYN, then drop it. 1310 * Otherwise this is an acceptable SYN segment 1311 * initialize tp->rcv_nxt and tp->irs 1312 * if seg contains ack then advance tp->snd_una 1313 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1314 * arrange for segment to be acked (eventually) 1315 * continue processing rest of data/controls, beginning with URG 1316 */ 1317 case TCPS_SYN_SENT: 1318 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1319 taop = &tao_noncached; 1320 bzero(taop, sizeof *taop); 1321 } 1322 1323 if ((thflags & TH_ACK) && 1324 (SEQ_LEQ(th->th_ack, tp->iss) || 1325 SEQ_GT(th->th_ack, tp->snd_max))) { 1326 /* 1327 * If we have a cached CCsent for the remote host, 1328 * hence we haven't just crashed and restarted, 1329 * do not send a RST. This may be a retransmission 1330 * from the other side after our earlier ACK was lost. 1331 * Our new SYN, when it arrives, will serve as the 1332 * needed ACK. 1333 */ 1334 if (taop->tao_ccsent != 0) 1335 goto drop; 1336 else { 1337 rstreason = BANDLIM_UNLIMITED; 1338 goto dropwithreset; 1339 } 1340 } 1341 if (thflags & TH_RST) { 1342 if (thflags & TH_ACK) 1343 tp = tcp_drop(tp, ECONNREFUSED); 1344 goto drop; 1345 } 1346 if (!(thflags & TH_SYN)) 1347 goto drop; 1348 tp->snd_wnd = th->th_win; /* initial send window */ 1349 tp->cc_recv = to.to_cc; /* foreign CC */ 1350 1351 tp->irs = th->th_seq; 1352 tcp_rcvseqinit(tp); 1353 if (thflags & TH_ACK) { 1354 /* 1355 * Our SYN was acked. If segment contains CC.ECHO 1356 * option, check it to make sure this segment really 1357 * matches our SYN. If not, just drop it as old 1358 * duplicate, but send an RST if we're still playing 1359 * by the old rules. If no CC.ECHO option, make sure 1360 * we don't get fooled into using T/TCP. 1361 */ 1362 if (to.to_flags & TOF_CCECHO) { 1363 if (tp->cc_send != to.to_ccecho) { 1364 if (taop->tao_ccsent != 0) 1365 goto drop; 1366 else { 1367 rstreason = BANDLIM_UNLIMITED; 1368 goto dropwithreset; 1369 } 1370 } 1371 } else 1372 tp->t_flags &= ~TF_RCVD_CC; 1373 tcpstat.tcps_connects++; 1374 soisconnected(so); 1375 /* Do window scaling on this connection? */ 1376 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 1377 (TF_RCVD_SCALE | TF_REQ_SCALE)) { 1378 tp->snd_scale = tp->requested_s_scale; 1379 tp->rcv_scale = tp->request_r_scale; 1380 } 1381 /* Segment is acceptable, update cache if undefined. */ 1382 if (taop->tao_ccsent == 0) 1383 taop->tao_ccsent = to.to_ccecho; 1384 1385 tp->rcv_adv += tp->rcv_wnd; 1386 tp->snd_una++; /* SYN is acked */ 1387 callout_stop(tp->tt_rexmt); 1388 /* 1389 * If there's data, delay ACK; if there's also a FIN 1390 * ACKNOW will be turned on later. 1391 */ 1392 if (DELAY_ACK(tp) && tlen != 0) 1393 callout_reset(tp->tt_delack, tcp_delacktime, 1394 tcp_timer_delack, tp); 1395 else 1396 tp->t_flags |= TF_ACKNOW; 1397 /* 1398 * Received <SYN,ACK> in SYN_SENT[*] state. 1399 * Transitions: 1400 * SYN_SENT --> ESTABLISHED 1401 * SYN_SENT* --> FIN_WAIT_1 1402 */ 1403 tp->t_starttime = ticks; 1404 if (tp->t_flags & TF_NEEDFIN) { 1405 tp->t_state = TCPS_FIN_WAIT_1; 1406 tp->t_flags &= ~TF_NEEDFIN; 1407 thflags &= ~TH_SYN; 1408 } else { 1409 tp->t_state = TCPS_ESTABLISHED; 1410 callout_reset(tp->tt_keep, tcp_keepidle, 1411 tcp_timer_keep, tp); 1412 } 1413 } else { 1414 /* 1415 * Received initial SYN in SYN-SENT[*] state => 1416 * simultaneous open. If segment contains CC option 1417 * and there is a cached CC, apply TAO test. 1418 * If it succeeds, connection is * half-synchronized. 1419 * Otherwise, do 3-way handshake: 1420 * SYN-SENT -> SYN-RECEIVED 1421 * SYN-SENT* -> SYN-RECEIVED* 1422 * If there was no CC option, clear cached CC value. 1423 */ 1424 tp->t_flags |= TF_ACKNOW; 1425 callout_stop(tp->tt_rexmt); 1426 if (to.to_flags & TOF_CC) { 1427 if (taop->tao_cc != 0 && 1428 CC_GT(to.to_cc, taop->tao_cc)) { 1429 /* 1430 * update cache and make transition: 1431 * SYN-SENT -> ESTABLISHED* 1432 * SYN-SENT* -> FIN-WAIT-1* 1433 */ 1434 taop->tao_cc = to.to_cc; 1435 tp->t_starttime = ticks; 1436 if (tp->t_flags & TF_NEEDFIN) { 1437 tp->t_state = TCPS_FIN_WAIT_1; 1438 tp->t_flags &= ~TF_NEEDFIN; 1439 } else { 1440 tp->t_state = TCPS_ESTABLISHED; 1441 callout_reset(tp->tt_keep, 1442 tcp_keepidle, 1443 tcp_timer_keep, 1444 tp); 1445 } 1446 tp->t_flags |= TF_NEEDSYN; 1447 } else 1448 tp->t_state = TCPS_SYN_RECEIVED; 1449 } else { 1450 /* CC.NEW or no option => invalidate cache */ 1451 taop->tao_cc = 0; 1452 tp->t_state = TCPS_SYN_RECEIVED; 1453 } 1454 } 1455 1456 trimthenstep6: 1457 /* 1458 * Advance th->th_seq to correspond to first data byte. 1459 * If data, trim to stay within window, 1460 * dropping FIN if necessary. 1461 */ 1462 th->th_seq++; 1463 if (tlen > tp->rcv_wnd) { 1464 todrop = tlen - tp->rcv_wnd; 1465 m_adj(m, -todrop); 1466 tlen = tp->rcv_wnd; 1467 thflags &= ~TH_FIN; 1468 tcpstat.tcps_rcvpackafterwin++; 1469 tcpstat.tcps_rcvbyteafterwin += todrop; 1470 } 1471 tp->snd_wl1 = th->th_seq - 1; 1472 tp->rcv_up = th->th_seq; 1473 /* 1474 * Client side of transaction: already sent SYN and data. 1475 * If the remote host used T/TCP to validate the SYN, 1476 * our data will be ACK'd; if so, enter normal data segment 1477 * processing in the middle of step 5, ack processing. 1478 * Otherwise, goto step 6. 1479 */ 1480 if (thflags & TH_ACK) 1481 goto process_ACK; 1482 1483 goto step6; 1484 1485 /* 1486 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1487 * if segment contains a SYN and CC [not CC.NEW] option: 1488 * if state == TIME_WAIT and connection duration > MSL, 1489 * drop packet and send RST; 1490 * 1491 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1492 * ack the FIN (and data) in retransmission queue. 1493 * Complete close and delete TCPCB. Then reprocess 1494 * segment, hoping to find new TCPCB in LISTEN state; 1495 * 1496 * else must be old SYN; drop it. 1497 * else do normal processing. 1498 */ 1499 case TCPS_LAST_ACK: 1500 case TCPS_CLOSING: 1501 case TCPS_TIME_WAIT: 1502 if ((thflags & TH_SYN) && 1503 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1504 if (tp->t_state == TCPS_TIME_WAIT && 1505 (ticks - tp->t_starttime) > tcp_msl) { 1506 rstreason = BANDLIM_UNLIMITED; 1507 goto dropwithreset; 1508 } 1509 if (CC_GT(to.to_cc, tp->cc_recv)) { 1510 tp = tcp_close(tp); 1511 goto findpcb; 1512 } 1513 else 1514 goto drop; 1515 } 1516 break; /* continue normal processing */ 1517 } 1518 1519 /* 1520 * States other than LISTEN or SYN_SENT. 1521 * First check the RST flag and sequence number since reset segments 1522 * are exempt from the timestamp and connection count tests. This 1523 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1524 * below which allowed reset segments in half the sequence space 1525 * to fall though and be processed (which gives forged reset 1526 * segments with a random sequence number a 50 percent chance of 1527 * killing a connection). 1528 * Then check timestamp, if present. 1529 * Then check the connection count, if present. 1530 * Then check that at least some bytes of segment are within 1531 * receive window. If segment begins before rcv_nxt, 1532 * drop leading data (and SYN); if nothing left, just ack. 1533 * 1534 * 1535 * If the RST bit is set, check the sequence number to see 1536 * if this is a valid reset segment. 1537 * RFC 793 page 37: 1538 * In all states except SYN-SENT, all reset (RST) segments 1539 * are validated by checking their SEQ-fields. A reset is 1540 * valid if its sequence number is in the window. 1541 * Note: this does not take into account delayed ACKs, so 1542 * we should test against last_ack_sent instead of rcv_nxt. 1543 * The sequence number in the reset segment is normally an 1544 * echo of our outgoing acknowledgement numbers, but some hosts 1545 * send a reset with the sequence number at the rightmost edge 1546 * of our receive window, and we have to handle this case. 1547 * If we have multiple segments in flight, the intial reset 1548 * segment sequence numbers will be to the left of last_ack_sent, 1549 * but they will eventually catch up. 1550 * In any case, it never made sense to trim reset segments to 1551 * fit the receive window since RFC 1122 says: 1552 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1553 * 1554 * A TCP SHOULD allow a received RST segment to include data. 1555 * 1556 * DISCUSSION 1557 * It has been suggested that a RST segment could contain 1558 * ASCII text that encoded and explained the cause of the 1559 * RST. No standard has yet been established for such 1560 * data. 1561 * 1562 * If the reset segment passes the sequence number test examine 1563 * the state: 1564 * SYN_RECEIVED STATE: 1565 * If passive open, return to LISTEN state. 1566 * If active open, inform user that connection was refused. 1567 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1568 * Inform user that connection was reset, and close tcb. 1569 * CLOSING, LAST_ACK STATES: 1570 * Close the tcb. 1571 * TIME_WAIT STATE: 1572 * Drop the segment - see Stevens, vol. 2, p. 964 and 1573 * RFC 1337. 1574 */ 1575 if (thflags & TH_RST) { 1576 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1577 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1578 switch (tp->t_state) { 1579 1580 case TCPS_SYN_RECEIVED: 1581 so->so_error = ECONNREFUSED; 1582 goto close; 1583 1584 case TCPS_ESTABLISHED: 1585 case TCPS_FIN_WAIT_1: 1586 case TCPS_FIN_WAIT_2: 1587 case TCPS_CLOSE_WAIT: 1588 so->so_error = ECONNRESET; 1589 close: 1590 tp->t_state = TCPS_CLOSED; 1591 tcpstat.tcps_drops++; 1592 tp = tcp_close(tp); 1593 break; 1594 1595 case TCPS_CLOSING: 1596 case TCPS_LAST_ACK: 1597 tp = tcp_close(tp); 1598 break; 1599 1600 case TCPS_TIME_WAIT: 1601 break; 1602 } 1603 } 1604 goto drop; 1605 } 1606 1607 /* 1608 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1609 * and it's less than ts_recent, drop it. 1610 */ 1611 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 && 1612 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1613 1614 /* Check to see if ts_recent is over 24 days old. */ 1615 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1616 /* 1617 * Invalidate ts_recent. If this segment updates 1618 * ts_recent, the age will be reset later and ts_recent 1619 * will get a valid value. If it does not, setting 1620 * ts_recent to zero will at least satisfy the 1621 * requirement that zero be placed in the timestamp 1622 * echo reply when ts_recent isn't valid. The 1623 * age isn't reset until we get a valid ts_recent 1624 * because we don't want out-of-order segments to be 1625 * dropped when ts_recent is old. 1626 */ 1627 tp->ts_recent = 0; 1628 } else { 1629 tcpstat.tcps_rcvduppack++; 1630 tcpstat.tcps_rcvdupbyte += tlen; 1631 tcpstat.tcps_pawsdrop++; 1632 if (tlen) 1633 goto dropafterack; 1634 goto drop; 1635 } 1636 } 1637 1638 /* 1639 * T/TCP mechanism 1640 * If T/TCP was negotiated and the segment doesn't have CC, 1641 * or if its CC is wrong then drop the segment. 1642 * RST segments do not have to comply with this. 1643 */ 1644 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1645 (!(to.to_flags & TOF_CC) || tp->cc_recv != to.to_cc)) 1646 goto dropafterack; 1647 1648 /* 1649 * In the SYN-RECEIVED state, validate that the packet belongs to 1650 * this connection before trimming the data to fit the receive 1651 * window. Check the sequence number versus IRS since we know 1652 * the sequence numbers haven't wrapped. This is a partial fix 1653 * for the "LAND" DoS attack. 1654 */ 1655 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1656 rstreason = BANDLIM_RST_OPENPORT; 1657 goto dropwithreset; 1658 } 1659 1660 todrop = tp->rcv_nxt - th->th_seq; 1661 if (todrop > 0) { 1662 if (TCP_DO_SACK(tp)) { 1663 /* Report duplicate segment at head of packet. */ 1664 tp->reportblk.rblk_start = th->th_seq; 1665 tp->reportblk.rblk_end = th->th_seq + tlen; 1666 if (thflags & TH_FIN) 1667 ++tp->reportblk.rblk_end; 1668 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt)) 1669 tp->reportblk.rblk_end = tp->rcv_nxt; 1670 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW); 1671 } 1672 if (thflags & TH_SYN) { 1673 thflags &= ~TH_SYN; 1674 th->th_seq++; 1675 if (th->th_urp > 1) 1676 th->th_urp--; 1677 else 1678 thflags &= ~TH_URG; 1679 todrop--; 1680 } 1681 /* 1682 * Following if statement from Stevens, vol. 2, p. 960. 1683 */ 1684 if (todrop > tlen || 1685 (todrop == tlen && !(thflags & TH_FIN))) { 1686 /* 1687 * Any valid FIN must be to the left of the window. 1688 * At this point the FIN must be a duplicate or out 1689 * of sequence; drop it. 1690 */ 1691 thflags &= ~TH_FIN; 1692 1693 /* 1694 * Send an ACK to resynchronize and drop any data. 1695 * But keep on processing for RST or ACK. 1696 */ 1697 tp->t_flags |= TF_ACKNOW; 1698 todrop = tlen; 1699 tcpstat.tcps_rcvduppack++; 1700 tcpstat.tcps_rcvdupbyte += todrop; 1701 } else { 1702 tcpstat.tcps_rcvpartduppack++; 1703 tcpstat.tcps_rcvpartdupbyte += todrop; 1704 } 1705 drop_hdrlen += todrop; /* drop from the top afterwards */ 1706 th->th_seq += todrop; 1707 tlen -= todrop; 1708 if (th->th_urp > todrop) 1709 th->th_urp -= todrop; 1710 else { 1711 thflags &= ~TH_URG; 1712 th->th_urp = 0; 1713 } 1714 } 1715 1716 /* 1717 * If new data are received on a connection after the 1718 * user processes are gone, then RST the other end. 1719 */ 1720 if ((so->so_state & SS_NOFDREF) && 1721 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1722 tp = tcp_close(tp); 1723 tcpstat.tcps_rcvafterclose++; 1724 rstreason = BANDLIM_UNLIMITED; 1725 goto dropwithreset; 1726 } 1727 1728 /* 1729 * If segment ends after window, drop trailing data 1730 * (and PUSH and FIN); if nothing left, just ACK. 1731 */ 1732 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 1733 if (todrop > 0) { 1734 tcpstat.tcps_rcvpackafterwin++; 1735 if (todrop >= tlen) { 1736 tcpstat.tcps_rcvbyteafterwin += tlen; 1737 /* 1738 * If a new connection request is received 1739 * while in TIME_WAIT, drop the old connection 1740 * and start over if the sequence numbers 1741 * are above the previous ones. 1742 */ 1743 if (thflags & TH_SYN && 1744 tp->t_state == TCPS_TIME_WAIT && 1745 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1746 tp = tcp_close(tp); 1747 goto findpcb; 1748 } 1749 /* 1750 * If window is closed can only take segments at 1751 * window edge, and have to drop data and PUSH from 1752 * incoming segments. Continue processing, but 1753 * remember to ack. Otherwise, drop segment 1754 * and ack. 1755 */ 1756 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1757 tp->t_flags |= TF_ACKNOW; 1758 tcpstat.tcps_rcvwinprobe++; 1759 } else 1760 goto dropafterack; 1761 } else 1762 tcpstat.tcps_rcvbyteafterwin += todrop; 1763 m_adj(m, -todrop); 1764 tlen -= todrop; 1765 thflags &= ~(TH_PUSH | TH_FIN); 1766 } 1767 1768 /* 1769 * If last ACK falls within this segment's sequence numbers, 1770 * record its timestamp. 1771 * NOTE that the test is modified according to the latest 1772 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1773 */ 1774 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1775 tp->ts_recent_age = ticks; 1776 tp->ts_recent = to.to_tsval; 1777 } 1778 1779 /* 1780 * If a SYN is in the window, then this is an 1781 * error and we send an RST and drop the connection. 1782 */ 1783 if (thflags & TH_SYN) { 1784 tp = tcp_drop(tp, ECONNRESET); 1785 rstreason = BANDLIM_UNLIMITED; 1786 goto dropwithreset; 1787 } 1788 1789 /* 1790 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1791 * flag is on (half-synchronized state), then queue data for 1792 * later processing; else drop segment and return. 1793 */ 1794 if (!(thflags & TH_ACK)) { 1795 if (tp->t_state == TCPS_SYN_RECEIVED || 1796 (tp->t_flags & TF_NEEDSYN)) 1797 goto step6; 1798 else 1799 goto drop; 1800 } 1801 1802 /* 1803 * Ack processing. 1804 */ 1805 switch (tp->t_state) { 1806 /* 1807 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter 1808 * ESTABLISHED state and continue processing. 1809 * The ACK was checked above. 1810 */ 1811 case TCPS_SYN_RECEIVED: 1812 1813 tcpstat.tcps_connects++; 1814 soisconnected(so); 1815 /* Do window scaling? */ 1816 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 1817 (TF_RCVD_SCALE | TF_REQ_SCALE)) { 1818 tp->snd_scale = tp->requested_s_scale; 1819 tp->rcv_scale = tp->request_r_scale; 1820 } 1821 /* 1822 * Upon successful completion of 3-way handshake, 1823 * update cache.CC if it was undefined, pass any queued 1824 * data to the user, and advance state appropriately. 1825 */ 1826 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1827 taop->tao_cc == 0) 1828 taop->tao_cc = tp->cc_recv; 1829 1830 /* 1831 * Make transitions: 1832 * SYN-RECEIVED -> ESTABLISHED 1833 * SYN-RECEIVED* -> FIN-WAIT-1 1834 */ 1835 tp->t_starttime = ticks; 1836 if (tp->t_flags & TF_NEEDFIN) { 1837 tp->t_state = TCPS_FIN_WAIT_1; 1838 tp->t_flags &= ~TF_NEEDFIN; 1839 } else { 1840 tp->t_state = TCPS_ESTABLISHED; 1841 callout_reset(tp->tt_keep, tcp_keepidle, 1842 tcp_timer_keep, tp); 1843 } 1844 /* 1845 * If segment contains data or ACK, will call tcp_reass() 1846 * later; if not, do so now to pass queued data to user. 1847 */ 1848 if (tlen == 0 && !(thflags & TH_FIN)) 1849 tcp_reass(tp, NULL, NULL, NULL); 1850 /* fall into ... */ 1851 1852 /* 1853 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1854 * ACKs. If the ack is in the range 1855 * tp->snd_una < th->th_ack <= tp->snd_max 1856 * then advance tp->snd_una to th->th_ack and drop 1857 * data from the retransmission queue. If this ACK reflects 1858 * more up to date window information we update our window information. 1859 */ 1860 case TCPS_ESTABLISHED: 1861 case TCPS_FIN_WAIT_1: 1862 case TCPS_FIN_WAIT_2: 1863 case TCPS_CLOSE_WAIT: 1864 case TCPS_CLOSING: 1865 case TCPS_LAST_ACK: 1866 case TCPS_TIME_WAIT: 1867 1868 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1869 if (TCP_DO_SACK(tp)) 1870 tcp_sack_update_scoreboard(tp, &to); 1871 if (tlen != 0 || tiwin != tp->snd_wnd) { 1872 tp->t_dupacks = 0; 1873 break; 1874 } 1875 tcpstat.tcps_rcvdupack++; 1876 if (!callout_active(tp->tt_rexmt) || 1877 th->th_ack != tp->snd_una) { 1878 tp->t_dupacks = 0; 1879 break; 1880 } 1881 /* 1882 * We have outstanding data (other than 1883 * a window probe), this is a completely 1884 * duplicate ack (ie, window info didn't 1885 * change), the ack is the biggest we've 1886 * seen and we've seen exactly our rexmt 1887 * threshhold of them, so assume a packet 1888 * has been dropped and retransmit it. 1889 * Kludge snd_nxt & the congestion 1890 * window so we send only this one 1891 * packet. 1892 */ 1893 if (IN_FASTRECOVERY(tp)) { 1894 if (TCP_DO_SACK(tp)) { 1895 /* No artifical cwnd inflation. */ 1896 tcp_sack_rexmt(tp, th); 1897 } else { 1898 /* 1899 * Dup acks mean that packets 1900 * have left the network 1901 * (they're now cached at the 1902 * receiver) so bump cwnd by 1903 * the amount in the receiver 1904 * to keep a constant cwnd 1905 * packets in the network. 1906 */ 1907 tp->snd_cwnd += tp->t_maxseg; 1908 tcp_output(tp); 1909 } 1910 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1911 tp->t_dupacks = 0; 1912 break; 1913 } else if (++tp->t_dupacks == tcprexmtthresh) { 1914 tcp_seq old_snd_nxt; 1915 u_int win; 1916 1917 fastretransmit: 1918 if (tcp_do_eifel_detect && 1919 (tp->t_flags & TF_RCVD_TSTMP)) { 1920 tcp_save_congestion_state(tp); 1921 tp->t_flags |= TF_FASTREXMT; 1922 } 1923 /* 1924 * We know we're losing at the current 1925 * window size, so do congestion avoidance: 1926 * set ssthresh to half the current window 1927 * and pull our congestion window back to the 1928 * new ssthresh. 1929 */ 1930 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1931 tp->t_maxseg; 1932 if (win < 2) 1933 win = 2; 1934 tp->snd_ssthresh = win * tp->t_maxseg; 1935 ENTER_FASTRECOVERY(tp); 1936 tp->snd_recover = tp->snd_max; 1937 callout_stop(tp->tt_rexmt); 1938 tp->t_rtttime = 0; 1939 old_snd_nxt = tp->snd_nxt; 1940 tp->snd_nxt = th->th_ack; 1941 tp->snd_cwnd = tp->t_maxseg; 1942 tcp_output(tp); 1943 ++tcpstat.tcps_sndfastrexmit; 1944 tp->snd_cwnd = tp->snd_ssthresh; 1945 tp->rexmt_high = tp->snd_nxt; 1946 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 1947 tp->snd_nxt = old_snd_nxt; 1948 KASSERT(tp->snd_limited <= 2, 1949 ("tp->snd_limited too big")); 1950 if (TCP_DO_SACK(tp)) 1951 tcp_sack_rexmt(tp, th); 1952 else 1953 tp->snd_cwnd += tp->t_maxseg * 1954 (tp->t_dupacks - tp->snd_limited); 1955 } else if (tcp_do_limitedtransmit) { 1956 u_long oldcwnd = tp->snd_cwnd; 1957 tcp_seq oldsndmax = tp->snd_max; 1958 /* outstanding data */ 1959 uint32_t ownd = tp->snd_max - tp->snd_una; 1960 u_int sent; 1961 1962 #define iceildiv(n, d) (((n)+(d)-1) / (d)) 1963 1964 KASSERT(tp->t_dupacks == 1 || 1965 tp->t_dupacks == 2, 1966 ("dupacks not 1 or 2")); 1967 if (tp->t_dupacks == 1) 1968 tp->snd_limited = 0; 1969 tp->snd_cwnd = ownd + 1970 (tp->t_dupacks - tp->snd_limited) * 1971 tp->t_maxseg; 1972 tcp_output(tp); 1973 tp->snd_cwnd = oldcwnd; 1974 sent = tp->snd_max - oldsndmax; 1975 if (sent > tp->t_maxseg) { 1976 KASSERT((tp->t_dupacks == 2 && 1977 tp->snd_limited == 0) || 1978 (sent == tp->t_maxseg + 1 && 1979 tp->t_flags & TF_SENTFIN), 1980 ("sent too much")); 1981 KASSERT(sent <= tp->t_maxseg * 2, 1982 ("sent too many segments")); 1983 tp->snd_limited = 2; 1984 tcpstat.tcps_sndlimited += 2; 1985 } else if (sent > 0) { 1986 ++tp->snd_limited; 1987 ++tcpstat.tcps_sndlimited; 1988 } else if (tcp_do_early_retransmit && 1989 (tcp_do_eifel_detect && 1990 (tp->t_flags & TF_RCVD_TSTMP)) && 1991 ownd < 4 * tp->t_maxseg && 1992 tp->t_dupacks + 1 >= 1993 iceildiv(ownd, tp->t_maxseg) && 1994 (!TCP_DO_SACK(tp) || 1995 ownd <= tp->t_maxseg || 1996 tcp_sack_has_sacked(&tp->scb, 1997 ownd - tp->t_maxseg))) { 1998 ++tcpstat.tcps_sndearlyrexmit; 1999 tp->t_flags |= TF_EARLYREXMT; 2000 goto fastretransmit; 2001 } 2002 } 2003 goto drop; 2004 } 2005 2006 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 2007 tp->t_dupacks = 0; 2008 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2009 /* 2010 * Detected optimistic ACK attack. 2011 * Force slow-start to de-synchronize attack. 2012 */ 2013 tp->snd_cwnd = tp->t_maxseg; 2014 2015 tcpstat.tcps_rcvacktoomuch++; 2016 goto dropafterack; 2017 } 2018 /* 2019 * If we reach this point, ACK is not a duplicate, 2020 * i.e., it ACKs something we sent. 2021 */ 2022 if (tp->t_flags & TF_NEEDSYN) { 2023 /* 2024 * T/TCP: Connection was half-synchronized, and our 2025 * SYN has been ACK'd (so connection is now fully 2026 * synchronized). Go to non-starred state, 2027 * increment snd_una for ACK of SYN, and check if 2028 * we can do window scaling. 2029 */ 2030 tp->t_flags &= ~TF_NEEDSYN; 2031 tp->snd_una++; 2032 /* Do window scaling? */ 2033 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 2034 (TF_RCVD_SCALE | TF_REQ_SCALE)) { 2035 tp->snd_scale = tp->requested_s_scale; 2036 tp->rcv_scale = tp->request_r_scale; 2037 } 2038 } 2039 2040 process_ACK: 2041 acked = th->th_ack - tp->snd_una; 2042 tcpstat.tcps_rcvackpack++; 2043 tcpstat.tcps_rcvackbyte += acked; 2044 2045 if (tcp_do_eifel_detect && acked > 0 && 2046 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) && 2047 (tp->t_flags & TF_FIRSTACCACK)) { 2048 /* Eifel detection applicable. */ 2049 if (to.to_tsecr < tp->t_rexmtTS) { 2050 ++tcpstat.tcps_eifeldetected; 2051 tcp_revert_congestion_state(tp); 2052 if (tp->t_rxtshift == 1 && 2053 ticks >= tp->t_badrxtwin) 2054 ++tcpstat.tcps_rttcantdetect; 2055 } 2056 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2057 /* 2058 * If we just performed our first retransmit, 2059 * and the ACK arrives within our recovery window, 2060 * then it was a mistake to do the retransmit 2061 * in the first place. Recover our original cwnd 2062 * and ssthresh, and proceed to transmit where we 2063 * left off. 2064 */ 2065 tcp_revert_congestion_state(tp); 2066 ++tcpstat.tcps_rttdetected; 2067 } 2068 2069 /* 2070 * If we have a timestamp reply, update smoothed 2071 * round trip time. If no timestamp is present but 2072 * transmit timer is running and timed sequence 2073 * number was acked, update smoothed round trip time. 2074 * Since we now have an rtt measurement, cancel the 2075 * timer backoff (cf., Phil Karn's retransmit alg.). 2076 * Recompute the initial retransmit timer. 2077 * 2078 * Some machines (certain windows boxes) send broken 2079 * timestamp replies during the SYN+ACK phase, ignore 2080 * timestamps of 0. 2081 */ 2082 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) 2083 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2084 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2085 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2086 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2087 2088 /* 2089 * If no data (only SYN) was ACK'd, 2090 * skip rest of ACK processing. 2091 */ 2092 if (acked == 0) 2093 goto step6; 2094 2095 /* Stop looking for an acceptable ACK since one was received. */ 2096 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT); 2097 2098 if (acked > so->so_snd.sb_cc) { 2099 tp->snd_wnd -= so->so_snd.sb_cc; 2100 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2101 ourfinisacked = TRUE; 2102 } else { 2103 sbdrop(&so->so_snd, acked); 2104 tp->snd_wnd -= acked; 2105 ourfinisacked = FALSE; 2106 } 2107 sowwakeup(so); 2108 2109 /* 2110 * Update window information. 2111 * Don't look at window if no ACK: 2112 * TAC's send garbage on first SYN. 2113 */ 2114 if (SEQ_LT(tp->snd_wl1, th->th_seq) || 2115 (tp->snd_wl1 == th->th_seq && 2116 (SEQ_LT(tp->snd_wl2, th->th_ack) || 2117 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) { 2118 /* keep track of pure window updates */ 2119 if (tlen == 0 && tp->snd_wl2 == th->th_ack && 2120 tiwin > tp->snd_wnd) 2121 tcpstat.tcps_rcvwinupd++; 2122 tp->snd_wnd = tiwin; 2123 tp->snd_wl1 = th->th_seq; 2124 tp->snd_wl2 = th->th_ack; 2125 if (tp->snd_wnd > tp->max_sndwnd) 2126 tp->max_sndwnd = tp->snd_wnd; 2127 needoutput = TRUE; 2128 } 2129 2130 tp->snd_una = th->th_ack; 2131 if (TCP_DO_SACK(tp)) 2132 tcp_sack_update_scoreboard(tp, &to); 2133 if (IN_FASTRECOVERY(tp)) { 2134 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2135 EXIT_FASTRECOVERY(tp); 2136 needoutput = TRUE; 2137 /* 2138 * If the congestion window was inflated 2139 * to account for the other side's 2140 * cached packets, retract it. 2141 * 2142 * Window inflation should have left us 2143 * with approximately snd_ssthresh outstanding 2144 * data. But, in case we would be inclined 2145 * to send a burst, better do it using 2146 * slow start. 2147 */ 2148 if (!TCP_DO_SACK(tp)) 2149 tp->snd_cwnd = tp->snd_ssthresh; 2150 2151 if (SEQ_GT(th->th_ack + tp->snd_cwnd, 2152 tp->snd_max + 2 * tp->t_maxseg)) 2153 tp->snd_cwnd = 2154 (tp->snd_max - tp->snd_una) + 2155 2 * tp->t_maxseg; 2156 } else { 2157 if (TCP_DO_SACK(tp)) { 2158 tp->snd_max_rexmt = tp->snd_max; 2159 tcp_sack_rexmt(tp, th); 2160 } else { 2161 tcp_newreno_partial_ack(tp, th, acked); 2162 } 2163 needoutput = FALSE; 2164 } 2165 } else { 2166 /* 2167 * When new data is acked, open the congestion window. 2168 * If the window gives us less than ssthresh packets 2169 * in flight, open exponentially (maxseg per packet). 2170 * Otherwise open linearly: maxseg per window 2171 * (maxseg^2 / cwnd per packet). 2172 */ 2173 u_int cw = tp->snd_cwnd; 2174 u_int incr; 2175 2176 if (cw > tp->snd_ssthresh) 2177 incr = tp->t_maxseg * tp->t_maxseg / cw; 2178 else 2179 incr = tp->t_maxseg; 2180 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2181 tp->snd_recover = th->th_ack - 1; 2182 } 2183 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2184 tp->snd_nxt = tp->snd_una; 2185 2186 /* 2187 * If all outstanding data is acked, stop retransmit 2188 * timer and remember to restart (more output or persist). 2189 * If there is more data to be acked, restart retransmit 2190 * timer, using current (possibly backed-off) value. 2191 */ 2192 if (th->th_ack == tp->snd_max) { 2193 callout_stop(tp->tt_rexmt); 2194 needoutput = TRUE; 2195 } else if (!callout_active(tp->tt_persist)) 2196 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2197 tcp_timer_rexmt, tp); 2198 2199 switch (tp->t_state) { 2200 /* 2201 * In FIN_WAIT_1 STATE in addition to the processing 2202 * for the ESTABLISHED state if our FIN is now acknowledged 2203 * then enter FIN_WAIT_2. 2204 */ 2205 case TCPS_FIN_WAIT_1: 2206 if (ourfinisacked) { 2207 /* 2208 * If we can't receive any more 2209 * data, then closing user can proceed. 2210 * Starting the timer is contrary to the 2211 * specification, but if we don't get a FIN 2212 * we'll hang forever. 2213 */ 2214 if (so->so_state & SS_CANTRCVMORE) { 2215 soisdisconnected(so); 2216 callout_reset(tp->tt_2msl, tcp_maxidle, 2217 tcp_timer_2msl, tp); 2218 } 2219 tp->t_state = TCPS_FIN_WAIT_2; 2220 } 2221 break; 2222 2223 /* 2224 * In CLOSING STATE in addition to the processing for 2225 * the ESTABLISHED state if the ACK acknowledges our FIN 2226 * then enter the TIME-WAIT state, otherwise ignore 2227 * the segment. 2228 */ 2229 case TCPS_CLOSING: 2230 if (ourfinisacked) { 2231 tp->t_state = TCPS_TIME_WAIT; 2232 tcp_canceltimers(tp); 2233 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2234 if (tp->cc_recv != 0 && 2235 (ticks - tp->t_starttime) < tcp_msl) 2236 callout_reset(tp->tt_2msl, 2237 tp->t_rxtcur * TCPTV_TWTRUNC, 2238 tcp_timer_2msl, tp); 2239 else 2240 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2241 tcp_timer_2msl, tp); 2242 soisdisconnected(so); 2243 } 2244 break; 2245 2246 /* 2247 * In LAST_ACK, we may still be waiting for data to drain 2248 * and/or to be acked, as well as for the ack of our FIN. 2249 * If our FIN is now acknowledged, delete the TCB, 2250 * enter the closed state and return. 2251 */ 2252 case TCPS_LAST_ACK: 2253 if (ourfinisacked) { 2254 tp = tcp_close(tp); 2255 goto drop; 2256 } 2257 break; 2258 2259 /* 2260 * In TIME_WAIT state the only thing that should arrive 2261 * is a retransmission of the remote FIN. Acknowledge 2262 * it and restart the finack timer. 2263 */ 2264 case TCPS_TIME_WAIT: 2265 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2266 tcp_timer_2msl, tp); 2267 goto dropafterack; 2268 } 2269 } 2270 2271 step6: 2272 /* 2273 * Update window information. 2274 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2275 */ 2276 if ((thflags & TH_ACK) && 2277 acceptable_window_update(tp, th, tiwin)) { 2278 /* keep track of pure window updates */ 2279 if (tlen == 0 && tp->snd_wl2 == th->th_ack && 2280 tiwin > tp->snd_wnd) 2281 tcpstat.tcps_rcvwinupd++; 2282 tp->snd_wnd = tiwin; 2283 tp->snd_wl1 = th->th_seq; 2284 tp->snd_wl2 = th->th_ack; 2285 if (tp->snd_wnd > tp->max_sndwnd) 2286 tp->max_sndwnd = tp->snd_wnd; 2287 needoutput = TRUE; 2288 } 2289 2290 /* 2291 * Process segments with URG. 2292 */ 2293 if ((thflags & TH_URG) && th->th_urp && 2294 !TCPS_HAVERCVDFIN(tp->t_state)) { 2295 /* 2296 * This is a kludge, but if we receive and accept 2297 * random urgent pointers, we'll crash in 2298 * soreceive. It's hard to imagine someone 2299 * actually wanting to send this much urgent data. 2300 */ 2301 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2302 th->th_urp = 0; /* XXX */ 2303 thflags &= ~TH_URG; /* XXX */ 2304 goto dodata; /* XXX */ 2305 } 2306 /* 2307 * If this segment advances the known urgent pointer, 2308 * then mark the data stream. This should not happen 2309 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2310 * a FIN has been received from the remote side. 2311 * In these states we ignore the URG. 2312 * 2313 * According to RFC961 (Assigned Protocols), 2314 * the urgent pointer points to the last octet 2315 * of urgent data. We continue, however, 2316 * to consider it to indicate the first octet 2317 * of data past the urgent section as the original 2318 * spec states (in one of two places). 2319 */ 2320 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) { 2321 tp->rcv_up = th->th_seq + th->th_urp; 2322 so->so_oobmark = so->so_rcv.sb_cc + 2323 (tp->rcv_up - tp->rcv_nxt) - 1; 2324 if (so->so_oobmark == 0) 2325 so->so_state |= SS_RCVATMARK; 2326 sohasoutofband(so); 2327 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2328 } 2329 /* 2330 * Remove out of band data so doesn't get presented to user. 2331 * This can happen independent of advancing the URG pointer, 2332 * but if two URG's are pending at once, some out-of-band 2333 * data may creep in... ick. 2334 */ 2335 if (th->th_urp <= (u_long)tlen && 2336 !(so->so_options & SO_OOBINLINE)) { 2337 /* hdr drop is delayed */ 2338 tcp_pulloutofband(so, th, m, drop_hdrlen); 2339 } 2340 } else { 2341 /* 2342 * If no out of band data is expected, 2343 * pull receive urgent pointer along 2344 * with the receive window. 2345 */ 2346 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2347 tp->rcv_up = tp->rcv_nxt; 2348 } 2349 2350 dodata: /* XXX */ 2351 /* 2352 * Process the segment text, merging it into the TCP sequencing queue, 2353 * and arranging for acknowledgment of receipt if necessary. 2354 * This process logically involves adjusting tp->rcv_wnd as data 2355 * is presented to the user (this happens in tcp_usrreq.c, 2356 * case PRU_RCVD). If a FIN has already been received on this 2357 * connection then we just ignore the text. 2358 */ 2359 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) { 2360 m_adj(m, drop_hdrlen); /* delayed header drop */ 2361 /* 2362 * Insert segment which includes th into TCP reassembly queue 2363 * with control block tp. Set thflags to whether reassembly now 2364 * includes a segment with FIN. This handles the common case 2365 * inline (segment is the next to be received on an established 2366 * connection, and the queue is empty), avoiding linkage into 2367 * and removal from the queue and repetition of various 2368 * conversions. 2369 * Set DELACK for segments received in order, but ack 2370 * immediately when segments are out of order (so 2371 * fast retransmit can work). 2372 */ 2373 if (th->th_seq == tp->rcv_nxt && 2374 LIST_EMPTY(&tp->t_segq) && 2375 TCPS_HAVEESTABLISHED(tp->t_state)) { 2376 if (DELAY_ACK(tp)) 2377 callout_reset(tp->tt_delack, tcp_delacktime, 2378 tcp_timer_delack, tp); 2379 else 2380 tp->t_flags |= TF_ACKNOW; 2381 tp->rcv_nxt += tlen; 2382 thflags = th->th_flags & TH_FIN; 2383 tcpstat.tcps_rcvpack++; 2384 tcpstat.tcps_rcvbyte += tlen; 2385 ND6_HINT(tp); 2386 if (so->so_state & SS_CANTRCVMORE) 2387 m_freem(m); 2388 else 2389 sbappendstream(&so->so_rcv, m); 2390 sorwakeup(so); 2391 } else { 2392 if (!(tp->t_flags & TF_DUPSEG)) { 2393 /* Initialize SACK report block. */ 2394 tp->reportblk.rblk_start = th->th_seq; 2395 tp->reportblk.rblk_end = th->th_seq + tlen + 2396 ((thflags & TH_FIN) != 0); 2397 } 2398 thflags = tcp_reass(tp, th, &tlen, m); 2399 tp->t_flags |= TF_ACKNOW; 2400 } 2401 2402 /* 2403 * Note the amount of data that peer has sent into 2404 * our window, in order to estimate the sender's 2405 * buffer size. 2406 */ 2407 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2408 } else { 2409 m_freem(m); 2410 thflags &= ~TH_FIN; 2411 } 2412 2413 /* 2414 * If FIN is received ACK the FIN and let the user know 2415 * that the connection is closing. 2416 */ 2417 if (thflags & TH_FIN) { 2418 if (!TCPS_HAVERCVDFIN(tp->t_state)) { 2419 socantrcvmore(so); 2420 /* 2421 * If connection is half-synchronized 2422 * (ie NEEDSYN flag on) then delay ACK, 2423 * so it may be piggybacked when SYN is sent. 2424 * Otherwise, since we received a FIN then no 2425 * more input can be expected, send ACK now. 2426 */ 2427 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2428 callout_reset(tp->tt_delack, tcp_delacktime, 2429 tcp_timer_delack, tp); 2430 else 2431 tp->t_flags |= TF_ACKNOW; 2432 tp->rcv_nxt++; 2433 } 2434 2435 switch (tp->t_state) { 2436 /* 2437 * In SYN_RECEIVED and ESTABLISHED STATES 2438 * enter the CLOSE_WAIT state. 2439 */ 2440 case TCPS_SYN_RECEIVED: 2441 tp->t_starttime = ticks; 2442 /*FALLTHROUGH*/ 2443 case TCPS_ESTABLISHED: 2444 tp->t_state = TCPS_CLOSE_WAIT; 2445 break; 2446 2447 /* 2448 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2449 * enter the CLOSING state. 2450 */ 2451 case TCPS_FIN_WAIT_1: 2452 tp->t_state = TCPS_CLOSING; 2453 break; 2454 2455 /* 2456 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2457 * starting the time-wait timer, turning off the other 2458 * standard timers. 2459 */ 2460 case TCPS_FIN_WAIT_2: 2461 tp->t_state = TCPS_TIME_WAIT; 2462 tcp_canceltimers(tp); 2463 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2464 if (tp->cc_recv != 0 && 2465 (ticks - tp->t_starttime) < tcp_msl) { 2466 callout_reset(tp->tt_2msl, 2467 tp->t_rxtcur * TCPTV_TWTRUNC, 2468 tcp_timer_2msl, tp); 2469 /* For transaction client, force ACK now. */ 2470 tp->t_flags |= TF_ACKNOW; 2471 } 2472 else 2473 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2474 tcp_timer_2msl, tp); 2475 soisdisconnected(so); 2476 break; 2477 2478 /* 2479 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2480 */ 2481 case TCPS_TIME_WAIT: 2482 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2483 tcp_timer_2msl, tp); 2484 break; 2485 } 2486 } 2487 2488 #ifdef TCPDEBUG 2489 if (so->so_options & SO_DEBUG) 2490 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2491 #endif 2492 2493 /* 2494 * Return any desired output. 2495 */ 2496 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2497 tcp_output(tp); 2498 return; 2499 2500 dropafterack: 2501 /* 2502 * Generate an ACK dropping incoming segment if it occupies 2503 * sequence space, where the ACK reflects our state. 2504 * 2505 * We can now skip the test for the RST flag since all 2506 * paths to this code happen after packets containing 2507 * RST have been dropped. 2508 * 2509 * In the SYN-RECEIVED state, don't send an ACK unless the 2510 * segment we received passes the SYN-RECEIVED ACK test. 2511 * If it fails send a RST. This breaks the loop in the 2512 * "LAND" DoS attack, and also prevents an ACK storm 2513 * between two listening ports that have been sent forged 2514 * SYN segments, each with the source address of the other. 2515 */ 2516 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2517 (SEQ_GT(tp->snd_una, th->th_ack) || 2518 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2519 rstreason = BANDLIM_RST_OPENPORT; 2520 goto dropwithreset; 2521 } 2522 #ifdef TCPDEBUG 2523 if (so->so_options & SO_DEBUG) 2524 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2525 #endif 2526 m_freem(m); 2527 tp->t_flags |= TF_ACKNOW; 2528 tcp_output(tp); 2529 return; 2530 2531 dropwithreset: 2532 /* 2533 * Generate a RST, dropping incoming segment. 2534 * Make ACK acceptable to originator of segment. 2535 * Don't bother to respond if destination was broadcast/multicast. 2536 */ 2537 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST)) 2538 goto drop; 2539 if (isipv6) { 2540 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2541 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2542 goto drop; 2543 } else { 2544 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2545 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2546 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2547 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2548 goto drop; 2549 } 2550 /* IPv6 anycast check is done at tcp6_input() */ 2551 2552 /* 2553 * Perform bandwidth limiting. 2554 */ 2555 #ifdef ICMP_BANDLIM 2556 if (badport_bandlim(rstreason) < 0) 2557 goto drop; 2558 #endif 2559 2560 #ifdef TCPDEBUG 2561 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2562 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2563 #endif 2564 if (thflags & TH_ACK) 2565 /* mtod() below is safe as long as hdr dropping is delayed */ 2566 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2567 TH_RST); 2568 else { 2569 if (thflags & TH_SYN) 2570 tlen++; 2571 /* mtod() below is safe as long as hdr dropping is delayed */ 2572 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen, 2573 (tcp_seq)0, TH_RST | TH_ACK); 2574 } 2575 return; 2576 2577 drop: 2578 /* 2579 * Drop space held by incoming segment and return. 2580 */ 2581 #ifdef TCPDEBUG 2582 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2583 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2584 #endif 2585 m_freem(m); 2586 return; 2587 } 2588 2589 /* 2590 * Parse TCP options and place in tcpopt. 2591 */ 2592 static void 2593 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn) 2594 { 2595 int opt, optlen, i; 2596 2597 to->to_flags = 0; 2598 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2599 opt = cp[0]; 2600 if (opt == TCPOPT_EOL) 2601 break; 2602 if (opt == TCPOPT_NOP) 2603 optlen = 1; 2604 else { 2605 if (cnt < 2) 2606 break; 2607 optlen = cp[1]; 2608 if (optlen < 2 || optlen > cnt) 2609 break; 2610 } 2611 switch (opt) { 2612 case TCPOPT_MAXSEG: 2613 if (optlen != TCPOLEN_MAXSEG) 2614 continue; 2615 if (!is_syn) 2616 continue; 2617 to->to_flags |= TOF_MSS; 2618 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss); 2619 to->to_mss = ntohs(to->to_mss); 2620 break; 2621 case TCPOPT_WINDOW: 2622 if (optlen != TCPOLEN_WINDOW) 2623 continue; 2624 if (!is_syn) 2625 continue; 2626 to->to_flags |= TOF_SCALE; 2627 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2628 break; 2629 case TCPOPT_TIMESTAMP: 2630 if (optlen != TCPOLEN_TIMESTAMP) 2631 continue; 2632 to->to_flags |= TOF_TS; 2633 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval); 2634 to->to_tsval = ntohl(to->to_tsval); 2635 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr); 2636 to->to_tsecr = ntohl(to->to_tsecr); 2637 break; 2638 case TCPOPT_CC: 2639 if (optlen != TCPOLEN_CC) 2640 continue; 2641 to->to_flags |= TOF_CC; 2642 bcopy(cp + 2, &to->to_cc, sizeof to->to_cc); 2643 to->to_cc = ntohl(to->to_cc); 2644 break; 2645 case TCPOPT_CCNEW: 2646 if (optlen != TCPOLEN_CC) 2647 continue; 2648 if (!is_syn) 2649 continue; 2650 to->to_flags |= TOF_CCNEW; 2651 bcopy(cp + 2, &to->to_cc, sizeof to->to_cc); 2652 to->to_cc = ntohl(to->to_cc); 2653 break; 2654 case TCPOPT_CCECHO: 2655 if (optlen != TCPOLEN_CC) 2656 continue; 2657 if (!is_syn) 2658 continue; 2659 to->to_flags |= TOF_CCECHO; 2660 bcopy(cp + 2, &to->to_ccecho, sizeof to->to_ccecho); 2661 to->to_ccecho = ntohl(to->to_ccecho); 2662 break; 2663 case TCPOPT_SACK_PERMITTED: 2664 if (optlen != TCPOLEN_SACK_PERMITTED) 2665 continue; 2666 if (!is_syn) 2667 continue; 2668 to->to_flags |= TOF_SACK_PERMITTED; 2669 break; 2670 case TCPOPT_SACK: 2671 if ((optlen - 2) & 0x07) /* not multiple of 8 */ 2672 continue; 2673 to->to_nsackblocks = (optlen - 2) / 8; 2674 to->to_sackblocks = (struct raw_sackblock *) (cp + 2); 2675 to->to_flags |= TOF_SACK; 2676 for (i = 0; i < to->to_nsackblocks; i++) { 2677 struct raw_sackblock *r = &to->to_sackblocks[i]; 2678 2679 r->rblk_start = ntohl(r->rblk_start); 2680 r->rblk_end = ntohl(r->rblk_end); 2681 } 2682 break; 2683 default: 2684 continue; 2685 } 2686 } 2687 } 2688 2689 /* 2690 * Pull out of band byte out of a segment so 2691 * it doesn't appear in the user's data queue. 2692 * It is still reflected in the segment length for 2693 * sequencing purposes. 2694 * "off" is the delayed to be dropped hdrlen. 2695 */ 2696 static void 2697 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off) 2698 { 2699 int cnt = off + th->th_urp - 1; 2700 2701 while (cnt >= 0) { 2702 if (m->m_len > cnt) { 2703 char *cp = mtod(m, caddr_t) + cnt; 2704 struct tcpcb *tp = sototcpcb(so); 2705 2706 tp->t_iobc = *cp; 2707 tp->t_oobflags |= TCPOOB_HAVEDATA; 2708 bcopy(cp + 1, cp, m->m_len - cnt - 1); 2709 m->m_len--; 2710 if (m->m_flags & M_PKTHDR) 2711 m->m_pkthdr.len--; 2712 return; 2713 } 2714 cnt -= m->m_len; 2715 m = m->m_next; 2716 if (m == 0) 2717 break; 2718 } 2719 panic("tcp_pulloutofband"); 2720 } 2721 2722 /* 2723 * Collect new round-trip time estimate 2724 * and update averages and current timeout. 2725 */ 2726 static void 2727 tcp_xmit_timer(struct tcpcb *tp, int rtt) 2728 { 2729 int delta; 2730 2731 tcpstat.tcps_rttupdated++; 2732 tp->t_rttupdated++; 2733 if (tp->t_srtt != 0) { 2734 /* 2735 * srtt is stored as fixed point with 5 bits after the 2736 * binary point (i.e., scaled by 8). The following magic 2737 * is equivalent to the smoothing algorithm in rfc793 with 2738 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2739 * point). Adjust rtt to origin 0. 2740 */ 2741 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2742 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2743 2744 if ((tp->t_srtt += delta) <= 0) 2745 tp->t_srtt = 1; 2746 2747 /* 2748 * We accumulate a smoothed rtt variance (actually, a 2749 * smoothed mean difference), then set the retransmit 2750 * timer to smoothed rtt + 4 times the smoothed variance. 2751 * rttvar is stored as fixed point with 4 bits after the 2752 * binary point (scaled by 16). The following is 2753 * equivalent to rfc793 smoothing with an alpha of .75 2754 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2755 * rfc793's wired-in beta. 2756 */ 2757 if (delta < 0) 2758 delta = -delta; 2759 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2760 if ((tp->t_rttvar += delta) <= 0) 2761 tp->t_rttvar = 1; 2762 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2763 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2764 } else { 2765 /* 2766 * No rtt measurement yet - use the unsmoothed rtt. 2767 * Set the variance to half the rtt (so our first 2768 * retransmit happens at 3*rtt). 2769 */ 2770 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2771 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2772 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2773 } 2774 tp->t_rtttime = 0; 2775 tp->t_rxtshift = 0; 2776 2777 /* 2778 * the retransmit should happen at rtt + 4 * rttvar. 2779 * Because of the way we do the smoothing, srtt and rttvar 2780 * will each average +1/2 tick of bias. When we compute 2781 * the retransmit timer, we want 1/2 tick of rounding and 2782 * 1 extra tick because of +-1/2 tick uncertainty in the 2783 * firing of the timer. The bias will give us exactly the 2784 * 1.5 tick we need. But, because the bias is 2785 * statistical, we have to test that we don't drop below 2786 * the minimum feasible timer (which is 2 ticks). 2787 */ 2788 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2789 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2790 2791 /* 2792 * We received an ack for a packet that wasn't retransmitted; 2793 * it is probably safe to discard any error indications we've 2794 * received recently. This isn't quite right, but close enough 2795 * for now (a route might have failed after we sent a segment, 2796 * and the return path might not be symmetrical). 2797 */ 2798 tp->t_softerror = 0; 2799 } 2800 2801 /* 2802 * Determine a reasonable value for maxseg size. 2803 * If the route is known, check route for mtu. 2804 * If none, use an mss that can be handled on the outgoing 2805 * interface without forcing IP to fragment; if bigger than 2806 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2807 * to utilize large mbufs. If no route is found, route has no mtu, 2808 * or the destination isn't local, use a default, hopefully conservative 2809 * size (usually 512 or the default IP max size, but no more than the mtu 2810 * of the interface), as we can't discover anything about intervening 2811 * gateways or networks. We also initialize the congestion/slow start 2812 * window to be a single segment if the destination isn't local. 2813 * While looking at the routing entry, we also initialize other path-dependent 2814 * parameters from pre-set or cached values in the routing entry. 2815 * 2816 * Also take into account the space needed for options that we 2817 * send regularly. Make maxseg shorter by that amount to assure 2818 * that we can send maxseg amount of data even when the options 2819 * are present. Store the upper limit of the length of options plus 2820 * data in maxopd. 2821 * 2822 * NOTE that this routine is only called when we process an incoming 2823 * segment, for outgoing segments only tcp_mssopt is called. 2824 * 2825 * In case of T/TCP, we call this routine during implicit connection 2826 * setup as well (offer = -1), to initialize maxseg from the cached 2827 * MSS of our peer. 2828 */ 2829 void 2830 tcp_mss(struct tcpcb *tp, int offer) 2831 { 2832 struct rtentry *rt; 2833 struct ifnet *ifp; 2834 int rtt, mss; 2835 u_long bufsize; 2836 struct inpcb *inp = tp->t_inpcb; 2837 struct socket *so; 2838 struct rmxp_tao *taop; 2839 int origoffer = offer; 2840 #ifdef INET6 2841 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE); 2842 size_t min_protoh = isipv6 ? 2843 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 2844 sizeof(struct tcpiphdr); 2845 #else 2846 const boolean_t isipv6 = FALSE; 2847 const size_t min_protoh = sizeof(struct tcpiphdr); 2848 #endif 2849 2850 if (isipv6) 2851 rt = tcp_rtlookup6(&inp->inp_inc); 2852 else 2853 rt = tcp_rtlookup(&inp->inp_inc); 2854 if (rt == NULL) { 2855 tp->t_maxopd = tp->t_maxseg = 2856 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2857 return; 2858 } 2859 ifp = rt->rt_ifp; 2860 so = inp->inp_socket; 2861 2862 taop = rmx_taop(rt->rt_rmx); 2863 /* 2864 * Offer == -1 means that we didn't receive SYN yet, 2865 * use cached value in that case; 2866 */ 2867 if (offer == -1) 2868 offer = taop->tao_mssopt; 2869 /* 2870 * Offer == 0 means that there was no MSS on the SYN segment, 2871 * in this case we use tcp_mssdflt. 2872 */ 2873 if (offer == 0) 2874 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2875 else 2876 /* 2877 * Sanity check: make sure that maxopd will be large 2878 * enough to allow some data on segments even is the 2879 * all the option space is used (40bytes). Otherwise 2880 * funny things may happen in tcp_output. 2881 */ 2882 offer = max(offer, 64); 2883 taop->tao_mssopt = offer; 2884 2885 /* 2886 * While we're here, check if there's an initial rtt 2887 * or rttvar. Convert from the route-table units 2888 * to scaled multiples of the slow timeout timer. 2889 */ 2890 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2891 /* 2892 * XXX the lock bit for RTT indicates that the value 2893 * is also a minimum value; this is subject to time. 2894 */ 2895 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2896 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2897 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2898 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2899 tcpstat.tcps_usedrtt++; 2900 if (rt->rt_rmx.rmx_rttvar) { 2901 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2902 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2903 tcpstat.tcps_usedrttvar++; 2904 } else { 2905 /* default variation is +- 1 rtt */ 2906 tp->t_rttvar = 2907 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2908 } 2909 TCPT_RANGESET(tp->t_rxtcur, 2910 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2911 tp->t_rttmin, TCPTV_REXMTMAX); 2912 } 2913 /* 2914 * if there's an mtu associated with the route, use it 2915 * else, use the link mtu. 2916 */ 2917 if (rt->rt_rmx.rmx_mtu) 2918 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2919 else { 2920 if (isipv6) { 2921 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh; 2922 if (!in6_localaddr(&inp->in6p_faddr)) 2923 mss = min(mss, tcp_v6mssdflt); 2924 } else { 2925 mss = ifp->if_mtu - min_protoh; 2926 if (!in_localaddr(inp->inp_faddr)) 2927 mss = min(mss, tcp_mssdflt); 2928 } 2929 } 2930 mss = min(mss, offer); 2931 /* 2932 * maxopd stores the maximum length of data AND options 2933 * in a segment; maxseg is the amount of data in a normal 2934 * segment. We need to store this value (maxopd) apart 2935 * from maxseg, because now every segment carries options 2936 * and thus we normally have somewhat less data in segments. 2937 */ 2938 tp->t_maxopd = mss; 2939 2940 /* 2941 * In case of T/TCP, origoffer==-1 indicates, that no segments 2942 * were received yet. In this case we just guess, otherwise 2943 * we do the same as before T/TCP. 2944 */ 2945 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP && 2946 (origoffer == -1 || 2947 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2948 mss -= TCPOLEN_TSTAMP_APPA; 2949 if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC && 2950 (origoffer == -1 || 2951 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2952 mss -= TCPOLEN_CC_APPA; 2953 2954 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2955 if (mss > MCLBYTES) 2956 mss &= ~(MCLBYTES-1); 2957 #else 2958 if (mss > MCLBYTES) 2959 mss = mss / MCLBYTES * MCLBYTES; 2960 #endif 2961 /* 2962 * If there's a pipesize, change the socket buffer 2963 * to that size. Make the socket buffers an integral 2964 * number of mss units; if the mss is larger than 2965 * the socket buffer, decrease the mss. 2966 */ 2967 #ifdef RTV_SPIPE 2968 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2969 #endif 2970 bufsize = so->so_snd.sb_hiwat; 2971 if (bufsize < mss) 2972 mss = bufsize; 2973 else { 2974 bufsize = roundup(bufsize, mss); 2975 if (bufsize > sb_max) 2976 bufsize = sb_max; 2977 if (bufsize > so->so_snd.sb_hiwat) 2978 sbreserve(&so->so_snd, bufsize, so, NULL); 2979 } 2980 tp->t_maxseg = mss; 2981 2982 #ifdef RTV_RPIPE 2983 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2984 #endif 2985 bufsize = so->so_rcv.sb_hiwat; 2986 if (bufsize > mss) { 2987 bufsize = roundup(bufsize, mss); 2988 if (bufsize > sb_max) 2989 bufsize = sb_max; 2990 if (bufsize > so->so_rcv.sb_hiwat) 2991 sbreserve(&so->so_rcv, bufsize, so, NULL); 2992 } 2993 2994 /* 2995 * Set the slow-start flight size depending on whether this 2996 * is a local network or not. 2997 */ 2998 if (tcp_do_rfc3390) 2999 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3000 else 3001 tp->snd_cwnd = mss; 3002 3003 if (rt->rt_rmx.rmx_ssthresh) { 3004 /* 3005 * There's some sort of gateway or interface 3006 * buffer limit on the path. Use this to set 3007 * the slow start threshhold, but set the 3008 * threshold to no less than 2*mss. 3009 */ 3010 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 3011 tcpstat.tcps_usedssthresh++; 3012 } 3013 } 3014 3015 /* 3016 * Determine the MSS option to send on an outgoing SYN. 3017 */ 3018 int 3019 tcp_mssopt(struct tcpcb *tp) 3020 { 3021 struct rtentry *rt; 3022 #ifdef INET6 3023 boolean_t isipv6 = 3024 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE); 3025 int min_protoh = isipv6 ? 3026 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 3027 sizeof(struct tcpiphdr); 3028 #else 3029 const boolean_t isipv6 = FALSE; 3030 const size_t min_protoh = sizeof(struct tcpiphdr); 3031 #endif 3032 3033 if (isipv6) 3034 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 3035 else 3036 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 3037 if (rt == NULL) 3038 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 3039 3040 return (rt->rt_ifp->if_mtu - min_protoh); 3041 } 3042 3043 /* 3044 * When a partial ack arrives, force the retransmission of the 3045 * next unacknowledged segment. Do not exit Fast Recovery. 3046 * 3047 * Implement the Slow-but-Steady variant of NewReno by restarting the 3048 * the retransmission timer. Turn it off here so it can be restarted 3049 * later in tcp_output(). 3050 */ 3051 static void 3052 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked) 3053 { 3054 tcp_seq old_snd_nxt = tp->snd_nxt; 3055 u_long ocwnd = tp->snd_cwnd; 3056 3057 callout_stop(tp->tt_rexmt); 3058 tp->t_rtttime = 0; 3059 tp->snd_nxt = th->th_ack; 3060 /* Set snd_cwnd to one segment beyond acknowledged offset. */ 3061 tp->snd_cwnd = tp->t_maxseg; 3062 tp->t_flags |= TF_ACKNOW; 3063 tcp_output(tp); 3064 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 3065 tp->snd_nxt = old_snd_nxt; 3066 /* partial window deflation */ 3067 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg; 3068 } 3069 3070 /* 3071 * In contrast to the Slow-but-Steady NewReno variant, 3072 * we do not reset the retransmission timer for SACK retransmissions, 3073 * except when retransmitting snd_una. 3074 */ 3075 static void 3076 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th) 3077 { 3078 uint32_t pipe, seglen; 3079 tcp_seq nextrexmt; 3080 boolean_t lostdup; 3081 tcp_seq old_snd_nxt = tp->snd_nxt; 3082 u_long ocwnd = tp->snd_cwnd; 3083 int nseg = 0; /* consecutive new segments */ 3084 #define MAXBURST 4 /* limit burst of new packets on partial ack */ 3085 3086 tp->t_rtttime = 0; 3087 pipe = tcp_sack_compute_pipe(tp); 3088 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg && 3089 (!tcp_do_smartsack || nseg < MAXBURST) && 3090 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) { 3091 uint32_t sent; 3092 tcp_seq old_snd_max; 3093 int error; 3094 3095 if (nextrexmt == tp->snd_max) ++nseg; 3096 tp->snd_nxt = nextrexmt; 3097 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen; 3098 old_snd_max = tp->snd_max; 3099 if (nextrexmt == tp->snd_una) 3100 callout_stop(tp->tt_rexmt); 3101 error = tcp_output(tp); 3102 if (error != 0) 3103 break; 3104 sent = tp->snd_nxt - nextrexmt; 3105 if (sent <= 0) 3106 break; 3107 if (!lostdup) 3108 pipe += sent; 3109 tcpstat.tcps_sndsackpack++; 3110 tcpstat.tcps_sndsackbyte += sent; 3111 if (SEQ_LT(nextrexmt, old_snd_max) && 3112 SEQ_LT(tp->rexmt_high, tp->snd_nxt)) 3113 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max); 3114 } 3115 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 3116 tp->snd_nxt = old_snd_nxt; 3117 tp->snd_cwnd = ocwnd; 3118 } 3119