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) 1982, 1986, 1988, 1990, 1993, 1994, 1995 36 * The Regents of the University of California. All rights reserved. 37 * 38 * Redistribution and use in source and binary forms, with or without 39 * modification, are permitted provided that the following conditions 40 * are met: 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. All advertising materials mentioning features or use of this software 47 * must display the following acknowledgement: 48 * This product includes software developed by the University of 49 * California, Berkeley and its contributors. 50 * 4. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 67 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $ 68 */ 69 70 #include "opt_inet.h" 71 #include "opt_inet6.h" 72 #include "opt_ipsec.h" 73 #include "opt_tcpdebug.h" 74 #include "opt_tcp_input.h" 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/kernel.h> 79 #include <sys/sysctl.h> 80 #include <sys/malloc.h> 81 #include <sys/mbuf.h> 82 #include <sys/proc.h> /* for proc0 declaration */ 83 #include <sys/protosw.h> 84 #include <sys/socket.h> 85 #include <sys/socketvar.h> 86 #include <sys/syslog.h> 87 #include <sys/in_cksum.h> 88 89 #include <sys/socketvar2.h> 90 91 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 92 #include <machine/stdarg.h> 93 94 #include <net/if.h> 95 #include <net/route.h> 96 97 #include <netinet/in.h> 98 #include <netinet/in_systm.h> 99 #include <netinet/ip.h> 100 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 101 #include <netinet/in_var.h> 102 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 103 #include <netinet/in_pcb.h> 104 #include <netinet/ip_var.h> 105 #include <netinet/ip6.h> 106 #include <netinet/icmp6.h> 107 #include <netinet6/nd6.h> 108 #include <netinet6/ip6_var.h> 109 #include <netinet6/in6_pcb.h> 110 #include <netinet/tcp.h> 111 #include <netinet/tcp_fsm.h> 112 #include <netinet/tcp_seq.h> 113 #include <netinet/tcp_timer.h> 114 #include <netinet/tcp_timer2.h> 115 #include <netinet/tcp_var.h> 116 #include <netinet6/tcp6_var.h> 117 #include <netinet/tcpip.h> 118 119 #ifdef TCPDEBUG 120 #include <netinet/tcp_debug.h> 121 122 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 123 struct tcphdr tcp_savetcp; 124 #endif 125 126 #ifdef FAST_IPSEC 127 #include <netproto/ipsec/ipsec.h> 128 #include <netproto/ipsec/ipsec6.h> 129 #endif 130 131 #ifdef IPSEC 132 #include <netinet6/ipsec.h> 133 #include <netinet6/ipsec6.h> 134 #include <netproto/key/key.h> 135 #endif 136 137 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 138 139 static int log_in_vain = 0; 140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 141 &log_in_vain, 0, "Log all incoming TCP connections"); 142 143 static int blackhole = 0; 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 145 &blackhole, 0, "Do not send RST when dropping refused connections"); 146 147 int tcp_delack_enabled = 1; 148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 149 &tcp_delack_enabled, 0, 150 "Delay ACK to try and piggyback it onto a data packet"); 151 152 #ifdef TCP_DROP_SYNFIN 153 static int drop_synfin = 0; 154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 155 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 156 #endif 157 158 static int tcp_do_limitedtransmit = 1; 159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW, 160 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)"); 161 162 static int tcp_do_early_retransmit = 1; 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW, 164 &tcp_do_early_retransmit, 0, "Early retransmit"); 165 166 int tcp_aggregate_acks = 1; 167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW, 168 &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack"); 169 170 static int tcp_do_eifel_detect = 1; 171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW, 172 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)"); 173 174 static int tcp_do_abc = 1; 175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW, 176 &tcp_do_abc, 0, 177 "TCP Appropriate Byte Counting (RFC 3465)"); 178 179 /* 180 * Define as tunable for easy testing with SACK on and off. 181 * Warning: do not change setting in the middle of an existing active TCP flow, 182 * else strange things might happen to that flow. 183 */ 184 int tcp_do_sack = 1; 185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 186 &tcp_do_sack, 0, "Enable SACK Algorithms"); 187 188 int tcp_do_smartsack = 1; 189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW, 190 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms"); 191 192 int tcp_do_rescuesack = 1; 193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack, CTLFLAG_RW, 194 &tcp_do_rescuesack, 0, "Rescue retransmission for SACK"); 195 196 int tcp_aggressive_rescuesack = 0; 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack_agg, CTLFLAG_RW, 198 &tcp_aggressive_rescuesack, 0, "Aggressive rescue retransmission for SACK"); 199 200 int tcp_do_rfc3517bis = 0; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3517bis, CTLFLAG_RW, 202 &tcp_do_rfc3517bis, 0, "Enable RFC3517 update"); 203 204 int tcp_rfc3517bis_rxt = 0; 205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3517bis_rxt, CTLFLAG_RW, 206 &tcp_rfc3517bis_rxt, 0, "Enable RFC3517 retransmit update"); 207 208 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 209 "TCP Segment Reassembly Queue"); 210 211 int tcp_reass_maxseg = 0; 212 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD, 213 &tcp_reass_maxseg, 0, 214 "Global maximum number of TCP Segments in Reassembly Queue"); 215 216 int tcp_reass_qsize = 0; 217 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 218 &tcp_reass_qsize, 0, 219 "Global number of TCP Segments currently in Reassembly Queue"); 220 221 static int tcp_reass_overflows = 0; 222 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 223 &tcp_reass_overflows, 0, 224 "Global number of TCP Segment Reassembly Queue Overflows"); 225 226 int tcp_do_autorcvbuf = 1; 227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 228 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing"); 229 230 int tcp_autorcvbuf_inc = 16*1024; 231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 232 &tcp_autorcvbuf_inc, 0, 233 "Incrementor step size of automatic receive buffer"); 234 235 int tcp_autorcvbuf_max = 2*1024*1024; 236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 237 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer"); 238 239 int tcp_sosend_agglim = 2; 240 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_agglim, CTLFLAG_RW, 241 &tcp_sosend_agglim, 0, "TCP sosend mbuf aggregation limit"); 242 243 int tcp_sosend_async = 1; 244 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_async, CTLFLAG_RW, 245 &tcp_sosend_async, 0, "TCP asynchronized pru_send"); 246 247 static int tcp_ignore_redun_dsack = 1; 248 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ignore_redun_dsack, CTLFLAG_RW, 249 &tcp_ignore_redun_dsack, 0, "Ignore redundant DSACK"); 250 251 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t, 252 tcp_seq); 253 static void tcp_pulloutofband(struct socket *, 254 struct tcphdr *, struct mbuf *, int); 255 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 256 struct mbuf *); 257 static void tcp_xmit_timer(struct tcpcb *, int, tcp_seq); 258 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int); 259 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *); 260 static boolean_t tcp_sack_limitedxmit(struct tcpcb *); 261 static int tcp_rmx_msl(const struct tcpcb *); 262 static void tcp_established(struct tcpcb *); 263 264 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 265 #ifdef INET6 266 #define ND6_HINT(tp) \ 267 do { \ 268 if ((tp) && (tp)->t_inpcb && \ 269 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \ 270 (tp)->t_inpcb->in6p_route.ro_rt) \ 271 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 272 } while (0) 273 #else 274 #define ND6_HINT(tp) 275 #endif 276 277 /* 278 * Indicate whether this ack should be delayed. We can delay the ack if 279 * - delayed acks are enabled and 280 * - there is no delayed ack timer in progress and 281 * - our last ack wasn't a 0-sized window. We never want to delay 282 * the ack that opens up a 0-sized window. 283 */ 284 #define DELAY_ACK(tp) \ 285 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \ 286 !(tp->t_flags & TF_RXWIN0SENT)) 287 288 #define acceptable_window_update(tp, th, tiwin) \ 289 (SEQ_LT(tp->snd_wl1, th->th_seq) || \ 290 (tp->snd_wl1 == th->th_seq && \ 291 (SEQ_LT(tp->snd_wl2, th->th_ack) || \ 292 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) 293 294 #define iceildiv(n, d) (((n)+(d)-1) / (d)) 295 #define need_early_retransmit(tp, ownd) \ 296 (tcp_do_early_retransmit && \ 297 (tcp_do_eifel_detect && (tp->t_flags & TF_RCVD_TSTMP)) && \ 298 ownd < ((tp->t_rxtthresh + 1) * tp->t_maxseg) && \ 299 tp->t_dupacks + 1 >= iceildiv(ownd, tp->t_maxseg) && \ 300 (!TCP_DO_SACK(tp) || ownd <= tp->t_maxseg || \ 301 tcp_sack_has_sacked(&tp->scb, ownd - tp->t_maxseg))) 302 303 static int 304 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m) 305 { 306 struct tseg_qent *q; 307 struct tseg_qent *p = NULL; 308 struct tseg_qent *te; 309 struct socket *so = tp->t_inpcb->inp_socket; 310 int flags; 311 312 /* 313 * Call with th == NULL after become established to 314 * force pre-ESTABLISHED data up to user socket. 315 */ 316 if (th == NULL) 317 goto present; 318 319 /* 320 * Limit the number of segments in the reassembly queue to prevent 321 * holding on to too many segments (and thus running out of mbufs). 322 * Make sure to let the missing segment through which caused this 323 * queue. Always keep one global queue entry spare to be able to 324 * process the missing segment. 325 */ 326 if (th->th_seq != tp->rcv_nxt && 327 tcp_reass_qsize + 1 >= tcp_reass_maxseg) { 328 tcp_reass_overflows++; 329 tcpstat.tcps_rcvmemdrop++; 330 m_freem(m); 331 /* no SACK block to report */ 332 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 333 return (0); 334 } 335 336 /* Allocate a new queue entry. */ 337 te = kmalloc(sizeof(struct tseg_qent), M_TSEGQ, M_INTWAIT | M_NULLOK); 338 if (te == NULL) { 339 tcpstat.tcps_rcvmemdrop++; 340 m_freem(m); 341 /* no SACK block to report */ 342 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 343 return (0); 344 } 345 atomic_add_int(&tcp_reass_qsize, 1); 346 347 /* 348 * Find a segment which begins after this one does. 349 */ 350 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 351 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 352 break; 353 p = q; 354 } 355 356 /* 357 * If there is a preceding segment, it may provide some of 358 * our data already. If so, drop the data from the incoming 359 * segment. If it provides all of our data, drop us. 360 */ 361 if (p != NULL) { 362 tcp_seq_diff_t i; 363 364 /* conversion to int (in i) handles seq wraparound */ 365 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 366 if (i > 0) { /* overlaps preceding segment */ 367 tp->sack_flags |= 368 (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG); 369 /* enclosing block starts w/ preceding segment */ 370 tp->encloseblk.rblk_start = p->tqe_th->th_seq; 371 if (i >= *tlenp) { 372 /* preceding encloses incoming segment */ 373 tp->encloseblk.rblk_end = TCP_SACK_BLKEND( 374 p->tqe_th->th_seq + p->tqe_len, 375 p->tqe_th->th_flags); 376 tcpstat.tcps_rcvduppack++; 377 tcpstat.tcps_rcvdupbyte += *tlenp; 378 m_freem(m); 379 kfree(te, M_TSEGQ); 380 atomic_add_int(&tcp_reass_qsize, -1); 381 /* 382 * Try to present any queued data 383 * at the left window edge to the user. 384 * This is needed after the 3-WHS 385 * completes. 386 */ 387 goto present; /* ??? */ 388 } 389 m_adj(m, i); 390 *tlenp -= i; 391 th->th_seq += i; 392 /* incoming segment end is enclosing block end */ 393 tp->encloseblk.rblk_end = TCP_SACK_BLKEND( 394 th->th_seq + *tlenp, th->th_flags); 395 /* trim end of reported D-SACK block */ 396 tp->reportblk.rblk_end = th->th_seq; 397 } 398 } 399 tcpstat.tcps_rcvoopack++; 400 tcpstat.tcps_rcvoobyte += *tlenp; 401 402 /* 403 * While we overlap succeeding segments trim them or, 404 * if they are completely covered, dequeue them. 405 */ 406 while (q) { 407 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 408 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len; 409 tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags); 410 struct tseg_qent *nq; 411 412 if (i <= 0) 413 break; 414 if (!(tp->sack_flags & TSACK_F_DUPSEG)) { 415 /* first time through */ 416 tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG); 417 tp->encloseblk = tp->reportblk; 418 /* report trailing duplicate D-SACK segment */ 419 tp->reportblk.rblk_start = q->tqe_th->th_seq; 420 } 421 if ((tp->sack_flags & TSACK_F_ENCLOSESEG) && 422 SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) { 423 /* extend enclosing block if one exists */ 424 tp->encloseblk.rblk_end = qend_sack; 425 } 426 if (i < q->tqe_len) { 427 q->tqe_th->th_seq += i; 428 q->tqe_len -= i; 429 m_adj(q->tqe_m, i); 430 break; 431 } 432 433 nq = LIST_NEXT(q, tqe_q); 434 LIST_REMOVE(q, tqe_q); 435 m_freem(q->tqe_m); 436 kfree(q, M_TSEGQ); 437 atomic_add_int(&tcp_reass_qsize, -1); 438 q = nq; 439 } 440 441 /* Insert the new segment queue entry into place. */ 442 te->tqe_m = m; 443 te->tqe_th = th; 444 te->tqe_len = *tlenp; 445 446 /* check if can coalesce with following segment */ 447 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) { 448 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len; 449 tcp_seq tend_sack = TCP_SACK_BLKEND(tend, te->tqe_th->th_flags); 450 451 te->tqe_len += q->tqe_len; 452 if (q->tqe_th->th_flags & TH_FIN) 453 te->tqe_th->th_flags |= TH_FIN; 454 m_cat(te->tqe_m, q->tqe_m); 455 tp->encloseblk.rblk_end = tend_sack; 456 /* 457 * When not reporting a duplicate segment, use 458 * the larger enclosing block as the SACK block. 459 */ 460 if (!(tp->sack_flags & TSACK_F_DUPSEG)) 461 tp->reportblk.rblk_end = tend_sack; 462 LIST_REMOVE(q, tqe_q); 463 kfree(q, M_TSEGQ); 464 atomic_add_int(&tcp_reass_qsize, -1); 465 } 466 467 if (p == NULL) { 468 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 469 } else { 470 /* check if can coalesce with preceding segment */ 471 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) { 472 p->tqe_len += te->tqe_len; 473 m_cat(p->tqe_m, te->tqe_m); 474 tp->encloseblk.rblk_start = p->tqe_th->th_seq; 475 /* 476 * When not reporting a duplicate segment, use 477 * the larger enclosing block as the SACK block. 478 */ 479 if (!(tp->sack_flags & TSACK_F_DUPSEG)) 480 tp->reportblk.rblk_start = p->tqe_th->th_seq; 481 kfree(te, M_TSEGQ); 482 atomic_add_int(&tcp_reass_qsize, -1); 483 } else { 484 LIST_INSERT_AFTER(p, te, tqe_q); 485 } 486 } 487 488 present: 489 /* 490 * Present data to user, advancing rcv_nxt through 491 * completed sequence space. 492 */ 493 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 494 return (0); 495 q = LIST_FIRST(&tp->t_segq); 496 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt) 497 return (0); 498 tp->rcv_nxt += q->tqe_len; 499 if (!(tp->sack_flags & TSACK_F_DUPSEG)) { 500 /* no SACK block to report since ACK advanced */ 501 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 502 } 503 /* no enclosing block to report since ACK advanced */ 504 tp->sack_flags &= ~TSACK_F_ENCLOSESEG; 505 flags = q->tqe_th->th_flags & TH_FIN; 506 LIST_REMOVE(q, tqe_q); 507 KASSERT(LIST_EMPTY(&tp->t_segq) || 508 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt, 509 ("segment not coalesced")); 510 if (so->so_state & SS_CANTRCVMORE) { 511 m_freem(q->tqe_m); 512 } else { 513 lwkt_gettoken(&so->so_rcv.ssb_token); 514 ssb_appendstream(&so->so_rcv, q->tqe_m); 515 lwkt_reltoken(&so->so_rcv.ssb_token); 516 } 517 kfree(q, M_TSEGQ); 518 atomic_add_int(&tcp_reass_qsize, -1); 519 ND6_HINT(tp); 520 sorwakeup(so); 521 return (flags); 522 } 523 524 /* 525 * TCP input routine, follows pages 65-76 of the 526 * protocol specification dated September, 1981 very closely. 527 */ 528 #ifdef INET6 529 int 530 tcp6_input(struct mbuf **mp, int *offp, int proto) 531 { 532 struct mbuf *m = *mp; 533 struct in6_ifaddr *ia6; 534 535 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 536 537 /* 538 * draft-itojun-ipv6-tcp-to-anycast 539 * better place to put this in? 540 */ 541 ia6 = ip6_getdstifaddr(m); 542 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 543 struct ip6_hdr *ip6; 544 545 ip6 = mtod(m, struct ip6_hdr *); 546 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 547 offsetof(struct ip6_hdr, ip6_dst)); 548 return (IPPROTO_DONE); 549 } 550 551 tcp_input(mp, offp, proto); 552 return (IPPROTO_DONE); 553 } 554 #endif 555 556 int 557 tcp_input(struct mbuf **mp, int *offp, int proto) 558 { 559 int off0; 560 struct tcphdr *th; 561 struct ip *ip = NULL; 562 struct ipovly *ipov; 563 struct inpcb *inp = NULL; 564 u_char *optp = NULL; 565 int optlen = 0; 566 int tlen, off; 567 int len = 0; 568 int drop_hdrlen; 569 struct tcpcb *tp = NULL; 570 int thflags; 571 struct socket *so = NULL; 572 int todrop, acked; 573 boolean_t ourfinisacked, needoutput = FALSE; 574 u_long tiwin; 575 int recvwin; 576 struct tcpopt to; /* options in this segment */ 577 struct sockaddr_in *next_hop = NULL; 578 int rstreason; /* For badport_bandlim accounting purposes */ 579 int cpu; 580 struct ip6_hdr *ip6 = NULL; 581 struct mbuf *m; 582 #ifdef INET6 583 boolean_t isipv6; 584 #else 585 const boolean_t isipv6 = FALSE; 586 #endif 587 #ifdef TCPDEBUG 588 short ostate = 0; 589 #endif 590 591 off0 = *offp; 592 m = *mp; 593 *mp = NULL; 594 595 tcpstat.tcps_rcvtotal++; 596 597 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) { 598 struct m_tag *mtag; 599 600 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 601 KKASSERT(mtag != NULL); 602 next_hop = m_tag_data(mtag); 603 } 604 605 #ifdef INET6 606 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE; 607 #endif 608 609 if (isipv6) { 610 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 611 ip6 = mtod(m, struct ip6_hdr *); 612 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0; 613 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 614 tcpstat.tcps_rcvbadsum++; 615 goto drop; 616 } 617 th = (struct tcphdr *)((caddr_t)ip6 + off0); 618 619 /* 620 * Be proactive about unspecified IPv6 address in source. 621 * As we use all-zero to indicate unbounded/unconnected pcb, 622 * unspecified IPv6 address can be used to confuse us. 623 * 624 * Note that packets with unspecified IPv6 destination is 625 * already dropped in ip6_input. 626 */ 627 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 628 /* XXX stat */ 629 goto drop; 630 } 631 } else { 632 /* 633 * Get IP and TCP header together in first mbuf. 634 * Note: IP leaves IP header in first mbuf. 635 */ 636 if (off0 > sizeof(struct ip)) { 637 ip_stripoptions(m); 638 off0 = sizeof(struct ip); 639 } 640 /* already checked and pulled up in ip_demux() */ 641 KASSERT(m->m_len >= sizeof(struct tcpiphdr), 642 ("TCP header not in one mbuf: m->m_len %d", m->m_len)); 643 ip = mtod(m, struct ip *); 644 ipov = (struct ipovly *)ip; 645 th = (struct tcphdr *)((caddr_t)ip + off0); 646 tlen = ip->ip_len; 647 648 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 649 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 650 th->th_sum = m->m_pkthdr.csum_data; 651 else 652 th->th_sum = in_pseudo(ip->ip_src.s_addr, 653 ip->ip_dst.s_addr, 654 htonl(m->m_pkthdr.csum_data + 655 ip->ip_len + 656 IPPROTO_TCP)); 657 th->th_sum ^= 0xffff; 658 } else { 659 /* 660 * Checksum extended TCP header and data. 661 */ 662 len = sizeof(struct ip) + tlen; 663 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 664 ipov->ih_len = (u_short)tlen; 665 ipov->ih_len = htons(ipov->ih_len); 666 th->th_sum = in_cksum(m, len); 667 } 668 if (th->th_sum) { 669 tcpstat.tcps_rcvbadsum++; 670 goto drop; 671 } 672 #ifdef INET6 673 /* Re-initialization for later version check */ 674 ip->ip_v = IPVERSION; 675 #endif 676 } 677 678 /* 679 * Check that TCP offset makes sense, 680 * pull out TCP options and adjust length. XXX 681 */ 682 off = th->th_off << 2; 683 /* already checked and pulled up in ip_demux() */ 684 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen, 685 ("bad TCP data offset %d (tlen %d)", off, tlen)); 686 tlen -= off; /* tlen is used instead of ti->ti_len */ 687 if (off > sizeof(struct tcphdr)) { 688 if (isipv6) { 689 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 690 ip6 = mtod(m, struct ip6_hdr *); 691 th = (struct tcphdr *)((caddr_t)ip6 + off0); 692 } else { 693 /* already pulled up in ip_demux() */ 694 KASSERT(m->m_len >= sizeof(struct ip) + off, 695 ("TCP header and options not in one mbuf: " 696 "m_len %d, off %d", m->m_len, off)); 697 } 698 optlen = off - sizeof(struct tcphdr); 699 optp = (u_char *)(th + 1); 700 } 701 thflags = th->th_flags; 702 703 #ifdef TCP_DROP_SYNFIN 704 /* 705 * If the drop_synfin option is enabled, drop all packets with 706 * both the SYN and FIN bits set. This prevents e.g. nmap from 707 * identifying the TCP/IP stack. 708 * 709 * This is a violation of the TCP specification. 710 */ 711 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN)) 712 goto drop; 713 #endif 714 715 /* 716 * Convert TCP protocol specific fields to host format. 717 */ 718 th->th_seq = ntohl(th->th_seq); 719 th->th_ack = ntohl(th->th_ack); 720 th->th_win = ntohs(th->th_win); 721 th->th_urp = ntohs(th->th_urp); 722 723 /* 724 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 725 * until after ip6_savecontrol() is called and before other functions 726 * which don't want those proto headers. 727 * Because ip6_savecontrol() is going to parse the mbuf to 728 * search for data to be passed up to user-land, it wants mbuf 729 * parameters to be unchanged. 730 * XXX: the call of ip6_savecontrol() has been obsoleted based on 731 * latest version of the advanced API (20020110). 732 */ 733 drop_hdrlen = off0 + off; 734 735 /* 736 * Locate pcb for segment. 737 */ 738 findpcb: 739 /* IPFIREWALL_FORWARD section */ 740 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */ 741 /* 742 * Transparently forwarded. Pretend to be the destination. 743 * already got one like this? 744 */ 745 cpu = mycpu->gd_cpuid; 746 inp = in_pcblookup_hash(&tcbinfo[cpu], 747 ip->ip_src, th->th_sport, 748 ip->ip_dst, th->th_dport, 749 0, m->m_pkthdr.rcvif); 750 if (!inp) { 751 /* 752 * It's new. Try to find the ambushing socket. 753 */ 754 755 /* 756 * The rest of the ipfw code stores the port in 757 * host order. XXX 758 * (The IP address is still in network order.) 759 */ 760 in_port_t dport = next_hop->sin_port ? 761 htons(next_hop->sin_port) : 762 th->th_dport; 763 764 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport, 765 next_hop->sin_addr.s_addr, dport); 766 inp = in_pcblookup_hash(&tcbinfo[cpu], 767 ip->ip_src, th->th_sport, 768 next_hop->sin_addr, dport, 769 1, m->m_pkthdr.rcvif); 770 } 771 } else { 772 if (isipv6) { 773 inp = in6_pcblookup_hash(&tcbinfo[0], 774 &ip6->ip6_src, th->th_sport, 775 &ip6->ip6_dst, th->th_dport, 776 1, m->m_pkthdr.rcvif); 777 } else { 778 cpu = mycpu->gd_cpuid; 779 inp = in_pcblookup_hash(&tcbinfo[cpu], 780 ip->ip_src, th->th_sport, 781 ip->ip_dst, th->th_dport, 782 1, m->m_pkthdr.rcvif); 783 } 784 } 785 786 /* 787 * If the state is CLOSED (i.e., TCB does not exist) then 788 * all data in the incoming segment is discarded. 789 * If the TCB exists but is in CLOSED state, it is embryonic, 790 * but should either do a listen or a connect soon. 791 */ 792 if (inp == NULL) { 793 if (log_in_vain) { 794 #ifdef INET6 795 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 796 #else 797 char dbuf[sizeof "aaa.bbb.ccc.ddd"]; 798 char sbuf[sizeof "aaa.bbb.ccc.ddd"]; 799 #endif 800 if (isipv6) { 801 strcpy(dbuf, "["); 802 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 803 strcat(dbuf, "]"); 804 strcpy(sbuf, "["); 805 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 806 strcat(sbuf, "]"); 807 } else { 808 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 809 strcpy(sbuf, inet_ntoa(ip->ip_src)); 810 } 811 switch (log_in_vain) { 812 case 1: 813 if (!(thflags & TH_SYN)) 814 break; 815 case 2: 816 log(LOG_INFO, 817 "Connection attempt to TCP %s:%d " 818 "from %s:%d flags:0x%02x\n", 819 dbuf, ntohs(th->th_dport), sbuf, 820 ntohs(th->th_sport), thflags); 821 break; 822 default: 823 break; 824 } 825 } 826 if (blackhole) { 827 switch (blackhole) { 828 case 1: 829 if (thflags & TH_SYN) 830 goto drop; 831 break; 832 case 2: 833 goto drop; 834 default: 835 goto drop; 836 } 837 } 838 rstreason = BANDLIM_RST_CLOSEDPORT; 839 goto dropwithreset; 840 } 841 842 #ifdef IPSEC 843 if (isipv6) { 844 if (ipsec6_in_reject_so(m, inp->inp_socket)) { 845 ipsec6stat.in_polvio++; 846 goto drop; 847 } 848 } else { 849 if (ipsec4_in_reject_so(m, inp->inp_socket)) { 850 ipsecstat.in_polvio++; 851 goto drop; 852 } 853 } 854 #endif 855 #ifdef FAST_IPSEC 856 if (isipv6) { 857 if (ipsec6_in_reject(m, inp)) 858 goto drop; 859 } else { 860 if (ipsec4_in_reject(m, inp)) 861 goto drop; 862 } 863 #endif 864 /* Check the minimum TTL for socket. */ 865 #ifdef INET6 866 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl) 867 goto drop; 868 #endif 869 870 tp = intotcpcb(inp); 871 if (tp == NULL) { 872 rstreason = BANDLIM_RST_CLOSEDPORT; 873 goto dropwithreset; 874 } 875 if (tp->t_state <= TCPS_CLOSED) 876 goto drop; 877 878 so = inp->inp_socket; 879 880 #ifdef TCPDEBUG 881 if (so->so_options & SO_DEBUG) { 882 ostate = tp->t_state; 883 if (isipv6) 884 bcopy(ip6, tcp_saveipgen, sizeof(*ip6)); 885 else 886 bcopy(ip, tcp_saveipgen, sizeof(*ip)); 887 tcp_savetcp = *th; 888 } 889 #endif 890 891 bzero(&to, sizeof to); 892 893 if (so->so_options & SO_ACCEPTCONN) { 894 struct in_conninfo inc; 895 896 #ifdef INET6 897 inc.inc_isipv6 = (isipv6 == TRUE); 898 #endif 899 if (isipv6) { 900 inc.inc6_faddr = ip6->ip6_src; 901 inc.inc6_laddr = ip6->ip6_dst; 902 inc.inc6_route.ro_rt = NULL; /* XXX */ 903 } else { 904 inc.inc_faddr = ip->ip_src; 905 inc.inc_laddr = ip->ip_dst; 906 inc.inc_route.ro_rt = NULL; /* XXX */ 907 } 908 inc.inc_fport = th->th_sport; 909 inc.inc_lport = th->th_dport; 910 911 /* 912 * If the state is LISTEN then ignore segment if it contains 913 * a RST. If the segment contains an ACK then it is bad and 914 * send a RST. If it does not contain a SYN then it is not 915 * interesting; drop it. 916 * 917 * If the state is SYN_RECEIVED (syncache) and seg contains 918 * an ACK, but not for our SYN/ACK, send a RST. If the seg 919 * contains a RST, check the sequence number to see if it 920 * is a valid reset segment. 921 */ 922 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) { 923 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) { 924 if (!syncache_expand(&inc, th, &so, m)) { 925 /* 926 * No syncache entry, or ACK was not 927 * for our SYN/ACK. Send a RST. 928 */ 929 tcpstat.tcps_badsyn++; 930 rstreason = BANDLIM_RST_OPENPORT; 931 goto dropwithreset; 932 } 933 934 /* 935 * Could not complete 3-way handshake, 936 * connection is being closed down, and 937 * syncache will free mbuf. 938 */ 939 if (so == NULL) 940 return(IPPROTO_DONE); 941 942 /* 943 * We must be in the correct protocol thread 944 * for this connection. 945 */ 946 KKASSERT(so->so_port == &curthread->td_msgport); 947 948 /* 949 * Socket is created in state SYN_RECEIVED. 950 * Continue processing segment. 951 */ 952 inp = so->so_pcb; 953 tp = intotcpcb(inp); 954 /* 955 * This is what would have happened in 956 * tcp_output() when the SYN,ACK was sent. 957 */ 958 tp->snd_up = tp->snd_una; 959 tp->snd_max = tp->snd_nxt = tp->iss + 1; 960 tp->last_ack_sent = tp->rcv_nxt; 961 962 goto after_listen; 963 } 964 if (thflags & TH_RST) { 965 syncache_chkrst(&inc, th); 966 goto drop; 967 } 968 if (thflags & TH_ACK) { 969 syncache_badack(&inc); 970 tcpstat.tcps_badsyn++; 971 rstreason = BANDLIM_RST_OPENPORT; 972 goto dropwithreset; 973 } 974 goto drop; 975 } 976 977 /* 978 * Segment's flags are (SYN) or (SYN | FIN). 979 */ 980 #ifdef INET6 981 /* 982 * If deprecated address is forbidden, 983 * we do not accept SYN to deprecated interface 984 * address to prevent any new inbound connection from 985 * getting established. 986 * When we do not accept SYN, we send a TCP RST, 987 * with deprecated source address (instead of dropping 988 * it). We compromise it as it is much better for peer 989 * to send a RST, and RST will be the final packet 990 * for the exchange. 991 * 992 * If we do not forbid deprecated addresses, we accept 993 * the SYN packet. RFC2462 does not suggest dropping 994 * SYN in this case. 995 * If we decipher RFC2462 5.5.4, it says like this: 996 * 1. use of deprecated addr with existing 997 * communication is okay - "SHOULD continue to be 998 * used" 999 * 2. use of it with new communication: 1000 * (2a) "SHOULD NOT be used if alternate address 1001 * with sufficient scope is available" 1002 * (2b) nothing mentioned otherwise. 1003 * Here we fall into (2b) case as we have no choice in 1004 * our source address selection - we must obey the peer. 1005 * 1006 * The wording in RFC2462 is confusing, and there are 1007 * multiple description text for deprecated address 1008 * handling - worse, they are not exactly the same. 1009 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1010 */ 1011 if (isipv6 && !ip6_use_deprecated) { 1012 struct in6_ifaddr *ia6; 1013 1014 if ((ia6 = ip6_getdstifaddr(m)) && 1015 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1016 tp = NULL; 1017 rstreason = BANDLIM_RST_OPENPORT; 1018 goto dropwithreset; 1019 } 1020 } 1021 #endif 1022 /* 1023 * If it is from this socket, drop it, it must be forged. 1024 * Don't bother responding if the destination was a broadcast. 1025 */ 1026 if (th->th_dport == th->th_sport) { 1027 if (isipv6) { 1028 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 1029 &ip6->ip6_src)) 1030 goto drop; 1031 } else { 1032 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 1033 goto drop; 1034 } 1035 } 1036 /* 1037 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1038 * 1039 * Note that it is quite possible to receive unicast 1040 * link-layer packets with a broadcast IP address. Use 1041 * in_broadcast() to find them. 1042 */ 1043 if (m->m_flags & (M_BCAST | M_MCAST)) 1044 goto drop; 1045 if (isipv6) { 1046 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1047 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 1048 goto drop; 1049 } else { 1050 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1051 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1052 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1053 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1054 goto drop; 1055 } 1056 /* 1057 * SYN appears to be valid; create compressed TCP state 1058 * for syncache, or perform t/tcp connection. 1059 */ 1060 if (so->so_qlen <= so->so_qlimit) { 1061 tcp_dooptions(&to, optp, optlen, TRUE, th->th_ack); 1062 if (!syncache_add(&inc, &to, th, so, m)) 1063 goto drop; 1064 1065 /* 1066 * Entry added to syncache, mbuf used to 1067 * send SYN,ACK packet. 1068 */ 1069 return(IPPROTO_DONE); 1070 } 1071 goto drop; 1072 } 1073 1074 after_listen: 1075 /* 1076 * Should not happen - syncache should pick up these connections. 1077 * 1078 * Once we are past handling listen sockets we must be in the 1079 * correct protocol processing thread. 1080 */ 1081 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state")); 1082 KKASSERT(so->so_port == &curthread->td_msgport); 1083 1084 /* Unscale the window into a 32-bit value. */ 1085 if (!(thflags & TH_SYN)) 1086 tiwin = th->th_win << tp->snd_scale; 1087 else 1088 tiwin = th->th_win; 1089 1090 /* 1091 * This is the second part of the MSS DoS prevention code (after 1092 * minmss on the sending side) and it deals with too many too small 1093 * tcp packets in a too short timeframe (1 second). 1094 * 1095 * XXX Removed. This code was crap. It does not scale to network 1096 * speed, and default values break NFS. Gone. 1097 */ 1098 /* REMOVED */ 1099 1100 /* 1101 * Segment received on connection. 1102 * 1103 * Reset idle time and keep-alive timer. Don't waste time if less 1104 * then a second has elapsed. 1105 */ 1106 if ((int)(ticks - tp->t_rcvtime) > hz) 1107 tcp_timer_keep_activity(tp, thflags); 1108 1109 /* 1110 * Process options. 1111 * XXX this is tradtitional behavior, may need to be cleaned up. 1112 */ 1113 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0, th->th_ack); 1114 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1115 if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) { 1116 tp->t_flags |= TF_RCVD_SCALE; 1117 tp->snd_scale = to.to_requested_s_scale; 1118 } 1119 1120 /* 1121 * Initial send window; will be updated upon next ACK 1122 */ 1123 tp->snd_wnd = th->th_win; 1124 1125 if (to.to_flags & TOF_TS) { 1126 tp->t_flags |= TF_RCVD_TSTMP; 1127 tp->ts_recent = to.to_tsval; 1128 tp->ts_recent_age = ticks; 1129 } 1130 if (!(to.to_flags & TOF_MSS)) 1131 to.to_mss = 0; 1132 tcp_mss(tp, to.to_mss); 1133 /* 1134 * Only set the TF_SACK_PERMITTED per-connection flag 1135 * if we got a SACK_PERMITTED option from the other side 1136 * and the global tcp_do_sack variable is true. 1137 */ 1138 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED)) 1139 tp->t_flags |= TF_SACK_PERMITTED; 1140 } 1141 1142 /* 1143 * Header prediction: check for the two common cases 1144 * of a uni-directional data xfer. If the packet has 1145 * no control flags, is in-sequence, the window didn't 1146 * change and we're not retransmitting, it's a 1147 * candidate. If the length is zero and the ack moved 1148 * forward, we're the sender side of the xfer. Just 1149 * free the data acked & wake any higher level process 1150 * that was blocked waiting for space. If the length 1151 * is non-zero and the ack didn't move, we're the 1152 * receiver side. If we're getting packets in-order 1153 * (the reassembly queue is empty), add the data to 1154 * the socket buffer and note that we need a delayed ack. 1155 * Make sure that the hidden state-flags are also off. 1156 * Since we check for TCPS_ESTABLISHED above, it can only 1157 * be TH_NEEDSYN. 1158 */ 1159 if (tp->t_state == TCPS_ESTABLISHED && 1160 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1161 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) && 1162 (!(to.to_flags & TOF_TS) || 1163 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1164 th->th_seq == tp->rcv_nxt && 1165 tp->snd_nxt == tp->snd_max) { 1166 1167 /* 1168 * If last ACK falls within this segment's sequence numbers, 1169 * record the timestamp. 1170 * NOTE that the test is modified according to the latest 1171 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1172 */ 1173 if ((to.to_flags & TOF_TS) && 1174 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1175 tp->ts_recent_age = ticks; 1176 tp->ts_recent = to.to_tsval; 1177 } 1178 1179 if (tlen == 0) { 1180 if (SEQ_GT(th->th_ack, tp->snd_una) && 1181 SEQ_LEQ(th->th_ack, tp->snd_max) && 1182 tp->snd_cwnd >= tp->snd_wnd && 1183 !IN_FASTRECOVERY(tp)) { 1184 /* 1185 * This is a pure ack for outstanding data. 1186 */ 1187 ++tcpstat.tcps_predack; 1188 /* 1189 * "bad retransmit" recovery 1190 * 1191 * If Eifel detection applies, then 1192 * it is deterministic, so use it 1193 * unconditionally over the old heuristic. 1194 * Otherwise, fall back to the old heuristic. 1195 */ 1196 if (tcp_do_eifel_detect && 1197 (to.to_flags & TOF_TS) && to.to_tsecr && 1198 (tp->rxt_flags & TRXT_F_FIRSTACCACK)) { 1199 /* Eifel detection applicable. */ 1200 if (to.to_tsecr < tp->t_rexmtTS) { 1201 tcp_revert_congestion_state(tp); 1202 ++tcpstat.tcps_eifeldetected; 1203 if (tp->t_rxtshift != 1 || 1204 ticks >= tp->t_badrxtwin) 1205 ++tcpstat.tcps_rttcantdetect; 1206 } 1207 } else if (tp->t_rxtshift == 1 && 1208 ticks < tp->t_badrxtwin) { 1209 tcp_revert_congestion_state(tp); 1210 ++tcpstat.tcps_rttdetected; 1211 } 1212 tp->rxt_flags &= ~(TRXT_F_FIRSTACCACK | 1213 TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT); 1214 /* 1215 * Recalculate the retransmit timer / rtt. 1216 * 1217 * Some machines (certain windows boxes) 1218 * send broken timestamp replies during the 1219 * SYN+ACK phase, ignore timestamps of 0. 1220 */ 1221 if ((to.to_flags & TOF_TS) && to.to_tsecr) { 1222 tcp_xmit_timer(tp, 1223 ticks - to.to_tsecr + 1, 1224 th->th_ack); 1225 } else if (tp->t_rtttime && 1226 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1227 tcp_xmit_timer(tp, 1228 ticks - tp->t_rtttime, 1229 th->th_ack); 1230 } 1231 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1232 acked = th->th_ack - tp->snd_una; 1233 tcpstat.tcps_rcvackpack++; 1234 tcpstat.tcps_rcvackbyte += acked; 1235 sbdrop(&so->so_snd.sb, acked); 1236 tp->snd_recover = th->th_ack - 1; 1237 tp->snd_una = th->th_ack; 1238 tp->t_dupacks = 0; 1239 /* 1240 * Update window information. 1241 */ 1242 if (tiwin != tp->snd_wnd && 1243 acceptable_window_update(tp, th, tiwin)) { 1244 /* keep track of pure window updates */ 1245 if (tp->snd_wl2 == th->th_ack && 1246 tiwin > tp->snd_wnd) 1247 tcpstat.tcps_rcvwinupd++; 1248 tp->snd_wnd = tiwin; 1249 tp->snd_wl1 = th->th_seq; 1250 tp->snd_wl2 = th->th_ack; 1251 if (tp->snd_wnd > tp->max_sndwnd) 1252 tp->max_sndwnd = tp->snd_wnd; 1253 } 1254 m_freem(m); 1255 ND6_HINT(tp); /* some progress has been done */ 1256 /* 1257 * If all outstanding data are acked, stop 1258 * retransmit timer, otherwise restart timer 1259 * using current (possibly backed-off) value. 1260 * If process is waiting for space, 1261 * wakeup/selwakeup/signal. If data 1262 * are ready to send, let tcp_output 1263 * decide between more output or persist. 1264 */ 1265 if (tp->snd_una == tp->snd_max) { 1266 tcp_callout_stop(tp, tp->tt_rexmt); 1267 } else if (!tcp_callout_active(tp, 1268 tp->tt_persist)) { 1269 tcp_callout_reset(tp, tp->tt_rexmt, 1270 tp->t_rxtcur, tcp_timer_rexmt); 1271 } 1272 sowwakeup(so); 1273 if (so->so_snd.ssb_cc > 0) 1274 tcp_output(tp); 1275 return(IPPROTO_DONE); 1276 } 1277 } else if (tiwin == tp->snd_wnd && 1278 th->th_ack == tp->snd_una && 1279 LIST_EMPTY(&tp->t_segq) && 1280 tlen <= ssb_space(&so->so_rcv)) { 1281 u_long newsize = 0; /* automatic sockbuf scaling */ 1282 /* 1283 * This is a pure, in-sequence data packet 1284 * with nothing on the reassembly queue and 1285 * we have enough buffer space to take it. 1286 */ 1287 ++tcpstat.tcps_preddat; 1288 tp->rcv_nxt += tlen; 1289 tcpstat.tcps_rcvpack++; 1290 tcpstat.tcps_rcvbyte += tlen; 1291 ND6_HINT(tp); /* some progress has been done */ 1292 /* 1293 * Automatic sizing of receive socket buffer. Often the send 1294 * buffer size is not optimally adjusted to the actual network 1295 * conditions at hand (delay bandwidth product). Setting the 1296 * buffer size too small limits throughput on links with high 1297 * bandwidth and high delay (eg. trans-continental/oceanic links). 1298 * 1299 * On the receive side the socket buffer memory is only rarely 1300 * used to any significant extent. This allows us to be much 1301 * more aggressive in scaling the receive socket buffer. For 1302 * the case that the buffer space is actually used to a large 1303 * extent and we run out of kernel memory we can simply drop 1304 * the new segments; TCP on the sender will just retransmit it 1305 * later. Setting the buffer size too big may only consume too 1306 * much kernel memory if the application doesn't read() from 1307 * the socket or packet loss or reordering makes use of the 1308 * reassembly queue. 1309 * 1310 * The criteria to step up the receive buffer one notch are: 1311 * 1. the number of bytes received during the time it takes 1312 * one timestamp to be reflected back to us (the RTT); 1313 * 2. received bytes per RTT is within seven eighth of the 1314 * current socket buffer size; 1315 * 3. receive buffer size has not hit maximal automatic size; 1316 * 1317 * This algorithm does one step per RTT at most and only if 1318 * we receive a bulk stream w/o packet losses or reorderings. 1319 * Shrinking the buffer during idle times is not necessary as 1320 * it doesn't consume any memory when idle. 1321 * 1322 * TODO: Only step up if the application is actually serving 1323 * the buffer to better manage the socket buffer resources. 1324 */ 1325 if (tcp_do_autorcvbuf && 1326 to.to_tsecr && 1327 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) { 1328 if (to.to_tsecr > tp->rfbuf_ts && 1329 to.to_tsecr - tp->rfbuf_ts < hz) { 1330 if (tp->rfbuf_cnt > 1331 (so->so_rcv.ssb_hiwat / 8 * 7) && 1332 so->so_rcv.ssb_hiwat < 1333 tcp_autorcvbuf_max) { 1334 newsize = 1335 ulmin(so->so_rcv.ssb_hiwat + 1336 tcp_autorcvbuf_inc, 1337 tcp_autorcvbuf_max); 1338 } 1339 /* Start over with next RTT. */ 1340 tp->rfbuf_ts = 0; 1341 tp->rfbuf_cnt = 0; 1342 } else 1343 tp->rfbuf_cnt += tlen; /* add up */ 1344 } 1345 /* 1346 * Add data to socket buffer. 1347 */ 1348 if (so->so_state & SS_CANTRCVMORE) { 1349 m_freem(m); 1350 } else { 1351 /* 1352 * Set new socket buffer size, give up when 1353 * limit is reached. 1354 * 1355 * Adjusting the size can mess up ACK 1356 * sequencing when pure window updates are 1357 * being avoided (which is the default), 1358 * so force an ack. 1359 */ 1360 lwkt_gettoken(&so->so_rcv.ssb_token); 1361 if (newsize) { 1362 tp->t_flags |= TF_RXRESIZED; 1363 if (!ssb_reserve(&so->so_rcv, newsize, 1364 so, NULL)) { 1365 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE); 1366 } 1367 if (newsize >= 1368 (TCP_MAXWIN << tp->rcv_scale)) { 1369 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE); 1370 } 1371 } 1372 m_adj(m, drop_hdrlen); /* delayed header drop */ 1373 ssb_appendstream(&so->so_rcv, m); 1374 lwkt_reltoken(&so->so_rcv.ssb_token); 1375 } 1376 sorwakeup(so); 1377 /* 1378 * This code is responsible for most of the ACKs 1379 * the TCP stack sends back after receiving a data 1380 * packet. Note that the DELAY_ACK check fails if 1381 * the delack timer is already running, which results 1382 * in an ack being sent every other packet (which is 1383 * what we want). 1384 * 1385 * We then further aggregate acks by not actually 1386 * sending one until the protocol thread has completed 1387 * processing the current backlog of packets. This 1388 * does not delay the ack any further, but allows us 1389 * to take advantage of the packet aggregation that 1390 * high speed NICs do (usually blocks of 8-10 packets) 1391 * to send a single ack rather then four or five acks, 1392 * greatly reducing the ack rate, the return channel 1393 * bandwidth, and the protocol overhead on both ends. 1394 * 1395 * Since this also has the effect of slowing down 1396 * the exponential slow-start ramp-up, systems with 1397 * very large bandwidth-delay products might want 1398 * to turn the feature off. 1399 */ 1400 if (DELAY_ACK(tp)) { 1401 tcp_callout_reset(tp, tp->tt_delack, 1402 tcp_delacktime, tcp_timer_delack); 1403 } else if (tcp_aggregate_acks) { 1404 tp->t_flags |= TF_ACKNOW; 1405 if (!(tp->t_flags & TF_ONOUTPUTQ)) { 1406 tp->t_flags |= TF_ONOUTPUTQ; 1407 tp->tt_cpu = mycpu->gd_cpuid; 1408 TAILQ_INSERT_TAIL( 1409 &tcpcbackq[tp->tt_cpu], 1410 tp, t_outputq); 1411 } 1412 } else { 1413 tp->t_flags |= TF_ACKNOW; 1414 tcp_output(tp); 1415 } 1416 return(IPPROTO_DONE); 1417 } 1418 } 1419 1420 /* 1421 * Calculate amount of space in receive window, 1422 * and then do TCP input processing. 1423 * Receive window is amount of space in rcv queue, 1424 * but not less than advertised window. 1425 */ 1426 recvwin = ssb_space(&so->so_rcv); 1427 if (recvwin < 0) 1428 recvwin = 0; 1429 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt)); 1430 1431 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1432 tp->rfbuf_ts = 0; 1433 tp->rfbuf_cnt = 0; 1434 1435 switch (tp->t_state) { 1436 /* 1437 * If the state is SYN_RECEIVED: 1438 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1439 */ 1440 case TCPS_SYN_RECEIVED: 1441 if ((thflags & TH_ACK) && 1442 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1443 SEQ_GT(th->th_ack, tp->snd_max))) { 1444 rstreason = BANDLIM_RST_OPENPORT; 1445 goto dropwithreset; 1446 } 1447 break; 1448 1449 /* 1450 * If the state is SYN_SENT: 1451 * if seg contains an ACK, but not for our SYN, drop the input. 1452 * if seg contains a RST, then drop the connection. 1453 * if seg does not contain SYN, then drop it. 1454 * Otherwise this is an acceptable SYN segment 1455 * initialize tp->rcv_nxt and tp->irs 1456 * if seg contains ack then advance tp->snd_una 1457 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1458 * arrange for segment to be acked (eventually) 1459 * continue processing rest of data/controls, beginning with URG 1460 */ 1461 case TCPS_SYN_SENT: 1462 if ((thflags & TH_ACK) && 1463 (SEQ_LEQ(th->th_ack, tp->iss) || 1464 SEQ_GT(th->th_ack, tp->snd_max))) { 1465 rstreason = BANDLIM_UNLIMITED; 1466 goto dropwithreset; 1467 } 1468 if (thflags & TH_RST) { 1469 if (thflags & TH_ACK) 1470 tp = tcp_drop(tp, ECONNREFUSED); 1471 goto drop; 1472 } 1473 if (!(thflags & TH_SYN)) 1474 goto drop; 1475 1476 tp->irs = th->th_seq; 1477 tcp_rcvseqinit(tp); 1478 if (thflags & TH_ACK) { 1479 /* Our SYN was acked. */ 1480 tcpstat.tcps_connects++; 1481 soisconnected(so); 1482 /* Do window scaling on this connection? */ 1483 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 1484 (TF_RCVD_SCALE | TF_REQ_SCALE)) 1485 tp->rcv_scale = tp->request_r_scale; 1486 tp->rcv_adv += tp->rcv_wnd; 1487 tp->snd_una++; /* SYN is acked */ 1488 tcp_callout_stop(tp, tp->tt_rexmt); 1489 /* 1490 * If there's data, delay ACK; if there's also a FIN 1491 * ACKNOW will be turned on later. 1492 */ 1493 if (DELAY_ACK(tp) && tlen != 0) { 1494 tcp_callout_reset(tp, tp->tt_delack, 1495 tcp_delacktime, tcp_timer_delack); 1496 } else { 1497 tp->t_flags |= TF_ACKNOW; 1498 } 1499 /* 1500 * Received <SYN,ACK> in SYN_SENT[*] state. 1501 * Transitions: 1502 * SYN_SENT --> ESTABLISHED 1503 * SYN_SENT* --> FIN_WAIT_1 1504 */ 1505 tp->t_starttime = ticks; 1506 if (tp->t_flags & TF_NEEDFIN) { 1507 tp->t_state = TCPS_FIN_WAIT_1; 1508 tp->t_flags &= ~TF_NEEDFIN; 1509 thflags &= ~TH_SYN; 1510 } else { 1511 tcp_established(tp); 1512 } 1513 } else { 1514 /* 1515 * Received initial SYN in SYN-SENT[*] state => 1516 * simultaneous open. 1517 * Do 3-way handshake: 1518 * SYN-SENT -> SYN-RECEIVED 1519 * SYN-SENT* -> SYN-RECEIVED* 1520 */ 1521 tp->t_flags |= TF_ACKNOW; 1522 tcp_callout_stop(tp, tp->tt_rexmt); 1523 tp->t_state = TCPS_SYN_RECEIVED; 1524 } 1525 1526 /* 1527 * Advance th->th_seq to correspond to first data byte. 1528 * If data, trim to stay within window, 1529 * dropping FIN if necessary. 1530 */ 1531 th->th_seq++; 1532 if (tlen > tp->rcv_wnd) { 1533 todrop = tlen - tp->rcv_wnd; 1534 m_adj(m, -todrop); 1535 tlen = tp->rcv_wnd; 1536 thflags &= ~TH_FIN; 1537 tcpstat.tcps_rcvpackafterwin++; 1538 tcpstat.tcps_rcvbyteafterwin += todrop; 1539 } 1540 tp->snd_wl1 = th->th_seq - 1; 1541 tp->rcv_up = th->th_seq; 1542 /* 1543 * Client side of transaction: already sent SYN and data. 1544 * If the remote host used T/TCP to validate the SYN, 1545 * our data will be ACK'd; if so, enter normal data segment 1546 * processing in the middle of step 5, ack processing. 1547 * Otherwise, goto step 6. 1548 */ 1549 if (thflags & TH_ACK) 1550 goto process_ACK; 1551 1552 goto step6; 1553 1554 /* 1555 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1556 * do normal processing (we no longer bother with T/TCP). 1557 */ 1558 case TCPS_LAST_ACK: 1559 case TCPS_CLOSING: 1560 case TCPS_TIME_WAIT: 1561 break; /* continue normal processing */ 1562 } 1563 1564 /* 1565 * States other than LISTEN or SYN_SENT. 1566 * First check the RST flag and sequence number since reset segments 1567 * are exempt from the timestamp and connection count tests. This 1568 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1569 * below which allowed reset segments in half the sequence space 1570 * to fall though and be processed (which gives forged reset 1571 * segments with a random sequence number a 50 percent chance of 1572 * killing a connection). 1573 * Then check timestamp, if present. 1574 * Then check the connection count, if present. 1575 * Then check that at least some bytes of segment are within 1576 * receive window. If segment begins before rcv_nxt, 1577 * drop leading data (and SYN); if nothing left, just ack. 1578 * 1579 * 1580 * If the RST bit is set, check the sequence number to see 1581 * if this is a valid reset segment. 1582 * RFC 793 page 37: 1583 * In all states except SYN-SENT, all reset (RST) segments 1584 * are validated by checking their SEQ-fields. A reset is 1585 * valid if its sequence number is in the window. 1586 * Note: this does not take into account delayed ACKs, so 1587 * we should test against last_ack_sent instead of rcv_nxt. 1588 * The sequence number in the reset segment is normally an 1589 * echo of our outgoing acknowledgement numbers, but some hosts 1590 * send a reset with the sequence number at the rightmost edge 1591 * of our receive window, and we have to handle this case. 1592 * If we have multiple segments in flight, the intial reset 1593 * segment sequence numbers will be to the left of last_ack_sent, 1594 * but they will eventually catch up. 1595 * In any case, it never made sense to trim reset segments to 1596 * fit the receive window since RFC 1122 says: 1597 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1598 * 1599 * A TCP SHOULD allow a received RST segment to include data. 1600 * 1601 * DISCUSSION 1602 * It has been suggested that a RST segment could contain 1603 * ASCII text that encoded and explained the cause of the 1604 * RST. No standard has yet been established for such 1605 * data. 1606 * 1607 * If the reset segment passes the sequence number test examine 1608 * the state: 1609 * SYN_RECEIVED STATE: 1610 * If passive open, return to LISTEN state. 1611 * If active open, inform user that connection was refused. 1612 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1613 * Inform user that connection was reset, and close tcb. 1614 * CLOSING, LAST_ACK STATES: 1615 * Close the tcb. 1616 * TIME_WAIT STATE: 1617 * Drop the segment - see Stevens, vol. 2, p. 964 and 1618 * RFC 1337. 1619 */ 1620 if (thflags & TH_RST) { 1621 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1622 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1623 switch (tp->t_state) { 1624 1625 case TCPS_SYN_RECEIVED: 1626 so->so_error = ECONNREFUSED; 1627 goto close; 1628 1629 case TCPS_ESTABLISHED: 1630 case TCPS_FIN_WAIT_1: 1631 case TCPS_FIN_WAIT_2: 1632 case TCPS_CLOSE_WAIT: 1633 so->so_error = ECONNRESET; 1634 close: 1635 tp->t_state = TCPS_CLOSED; 1636 tcpstat.tcps_drops++; 1637 tp = tcp_close(tp); 1638 break; 1639 1640 case TCPS_CLOSING: 1641 case TCPS_LAST_ACK: 1642 tp = tcp_close(tp); 1643 break; 1644 1645 case TCPS_TIME_WAIT: 1646 break; 1647 } 1648 } 1649 goto drop; 1650 } 1651 1652 /* 1653 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1654 * and it's less than ts_recent, drop it. 1655 */ 1656 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 && 1657 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1658 1659 /* Check to see if ts_recent is over 24 days old. */ 1660 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1661 /* 1662 * Invalidate ts_recent. If this segment updates 1663 * ts_recent, the age will be reset later and ts_recent 1664 * will get a valid value. If it does not, setting 1665 * ts_recent to zero will at least satisfy the 1666 * requirement that zero be placed in the timestamp 1667 * echo reply when ts_recent isn't valid. The 1668 * age isn't reset until we get a valid ts_recent 1669 * because we don't want out-of-order segments to be 1670 * dropped when ts_recent is old. 1671 */ 1672 tp->ts_recent = 0; 1673 } else { 1674 tcpstat.tcps_rcvduppack++; 1675 tcpstat.tcps_rcvdupbyte += tlen; 1676 tcpstat.tcps_pawsdrop++; 1677 if (tlen) 1678 goto dropafterack; 1679 goto drop; 1680 } 1681 } 1682 1683 /* 1684 * In the SYN-RECEIVED state, validate that the packet belongs to 1685 * this connection before trimming the data to fit the receive 1686 * window. Check the sequence number versus IRS since we know 1687 * the sequence numbers haven't wrapped. This is a partial fix 1688 * for the "LAND" DoS attack. 1689 */ 1690 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1691 rstreason = BANDLIM_RST_OPENPORT; 1692 goto dropwithreset; 1693 } 1694 1695 todrop = tp->rcv_nxt - th->th_seq; 1696 if (todrop > 0) { 1697 if (TCP_DO_SACK(tp)) { 1698 /* Report duplicate segment at head of packet. */ 1699 tp->reportblk.rblk_start = th->th_seq; 1700 tp->reportblk.rblk_end = TCP_SACK_BLKEND( 1701 th->th_seq + tlen, thflags); 1702 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt)) 1703 tp->reportblk.rblk_end = tp->rcv_nxt; 1704 tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_SACKLEFT); 1705 tp->t_flags |= TF_ACKNOW; 1706 } 1707 if (thflags & TH_SYN) { 1708 thflags &= ~TH_SYN; 1709 th->th_seq++; 1710 if (th->th_urp > 1) 1711 th->th_urp--; 1712 else 1713 thflags &= ~TH_URG; 1714 todrop--; 1715 } 1716 /* 1717 * Following if statement from Stevens, vol. 2, p. 960. 1718 */ 1719 if (todrop > tlen || 1720 (todrop == tlen && !(thflags & TH_FIN))) { 1721 /* 1722 * Any valid FIN must be to the left of the window. 1723 * At this point the FIN must be a duplicate or out 1724 * of sequence; drop it. 1725 */ 1726 thflags &= ~TH_FIN; 1727 1728 /* 1729 * Send an ACK to resynchronize and drop any data. 1730 * But keep on processing for RST or ACK. 1731 */ 1732 tp->t_flags |= TF_ACKNOW; 1733 todrop = tlen; 1734 tcpstat.tcps_rcvduppack++; 1735 tcpstat.tcps_rcvdupbyte += todrop; 1736 } else { 1737 tcpstat.tcps_rcvpartduppack++; 1738 tcpstat.tcps_rcvpartdupbyte += todrop; 1739 } 1740 drop_hdrlen += todrop; /* drop from the top afterwards */ 1741 th->th_seq += todrop; 1742 tlen -= todrop; 1743 if (th->th_urp > todrop) 1744 th->th_urp -= todrop; 1745 else { 1746 thflags &= ~TH_URG; 1747 th->th_urp = 0; 1748 } 1749 } 1750 1751 /* 1752 * If new data are received on a connection after the 1753 * user processes are gone, then RST the other end. 1754 */ 1755 if ((so->so_state & SS_NOFDREF) && 1756 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1757 tp = tcp_close(tp); 1758 tcpstat.tcps_rcvafterclose++; 1759 rstreason = BANDLIM_UNLIMITED; 1760 goto dropwithreset; 1761 } 1762 1763 /* 1764 * If segment ends after window, drop trailing data 1765 * (and PUSH and FIN); if nothing left, just ACK. 1766 */ 1767 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 1768 if (todrop > 0) { 1769 tcpstat.tcps_rcvpackafterwin++; 1770 if (todrop >= tlen) { 1771 tcpstat.tcps_rcvbyteafterwin += tlen; 1772 /* 1773 * If a new connection request is received 1774 * while in TIME_WAIT, drop the old connection 1775 * and start over if the sequence numbers 1776 * are above the previous ones. 1777 */ 1778 if (thflags & TH_SYN && 1779 tp->t_state == TCPS_TIME_WAIT && 1780 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1781 tp = tcp_close(tp); 1782 goto findpcb; 1783 } 1784 /* 1785 * If window is closed can only take segments at 1786 * window edge, and have to drop data and PUSH from 1787 * incoming segments. Continue processing, but 1788 * remember to ack. Otherwise, drop segment 1789 * and ack. 1790 */ 1791 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1792 tp->t_flags |= TF_ACKNOW; 1793 tcpstat.tcps_rcvwinprobe++; 1794 } else 1795 goto dropafterack; 1796 } else 1797 tcpstat.tcps_rcvbyteafterwin += todrop; 1798 m_adj(m, -todrop); 1799 tlen -= todrop; 1800 thflags &= ~(TH_PUSH | TH_FIN); 1801 } 1802 1803 /* 1804 * If last ACK falls within this segment's sequence numbers, 1805 * record its timestamp. 1806 * NOTE: 1807 * 1) That the test incorporates suggestions from the latest 1808 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1809 * 2) That updating only on newer timestamps interferes with 1810 * our earlier PAWS tests, so this check should be solely 1811 * predicated on the sequence space of this segment. 1812 * 3) That we modify the segment boundary check to be 1813 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN 1814 * instead of RFC1323's 1815 * Last.ACK.Sent < SEG.SEQ + SEG.LEN, 1816 * This modified check allows us to overcome RFC1323's 1817 * limitations as described in Stevens TCP/IP Illustrated 1818 * Vol. 2 p.869. In such cases, we can still calculate the 1819 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1820 */ 1821 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1822 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen 1823 + ((thflags & TH_SYN) != 0) 1824 + ((thflags & TH_FIN) != 0)))) { 1825 tp->ts_recent_age = ticks; 1826 tp->ts_recent = to.to_tsval; 1827 } 1828 1829 /* 1830 * If a SYN is in the window, then this is an 1831 * error and we send an RST and drop the connection. 1832 */ 1833 if (thflags & TH_SYN) { 1834 tp = tcp_drop(tp, ECONNRESET); 1835 rstreason = BANDLIM_UNLIMITED; 1836 goto dropwithreset; 1837 } 1838 1839 /* 1840 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1841 * flag is on (half-synchronized state), then queue data for 1842 * later processing; else drop segment and return. 1843 */ 1844 if (!(thflags & TH_ACK)) { 1845 if (tp->t_state == TCPS_SYN_RECEIVED || 1846 (tp->t_flags & TF_NEEDSYN)) 1847 goto step6; 1848 else 1849 goto drop; 1850 } 1851 1852 /* 1853 * Ack processing. 1854 */ 1855 switch (tp->t_state) { 1856 /* 1857 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter 1858 * ESTABLISHED state and continue processing. 1859 * The ACK was checked above. 1860 */ 1861 case TCPS_SYN_RECEIVED: 1862 1863 tcpstat.tcps_connects++; 1864 soisconnected(so); 1865 /* Do window scaling? */ 1866 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 1867 (TF_RCVD_SCALE | TF_REQ_SCALE)) 1868 tp->rcv_scale = tp->request_r_scale; 1869 /* 1870 * Make transitions: 1871 * SYN-RECEIVED -> ESTABLISHED 1872 * SYN-RECEIVED* -> FIN-WAIT-1 1873 */ 1874 tp->t_starttime = ticks; 1875 if (tp->t_flags & TF_NEEDFIN) { 1876 tp->t_state = TCPS_FIN_WAIT_1; 1877 tp->t_flags &= ~TF_NEEDFIN; 1878 } else { 1879 tcp_established(tp); 1880 } 1881 /* 1882 * If segment contains data or ACK, will call tcp_reass() 1883 * later; if not, do so now to pass queued data to user. 1884 */ 1885 if (tlen == 0 && !(thflags & TH_FIN)) 1886 tcp_reass(tp, NULL, NULL, NULL); 1887 /* fall into ... */ 1888 1889 /* 1890 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1891 * ACKs. If the ack is in the range 1892 * tp->snd_una < th->th_ack <= tp->snd_max 1893 * then advance tp->snd_una to th->th_ack and drop 1894 * data from the retransmission queue. If this ACK reflects 1895 * more up to date window information we update our window information. 1896 */ 1897 case TCPS_ESTABLISHED: 1898 case TCPS_FIN_WAIT_1: 1899 case TCPS_FIN_WAIT_2: 1900 case TCPS_CLOSE_WAIT: 1901 case TCPS_CLOSING: 1902 case TCPS_LAST_ACK: 1903 case TCPS_TIME_WAIT: 1904 1905 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1906 if (TCP_DO_SACK(tp)) 1907 tcp_sack_update_scoreboard(tp, &to); 1908 if (!tcp_callout_active(tp, tp->tt_rexmt) || 1909 th->th_ack != tp->snd_una) { 1910 if (tlen == 0 && tiwin == tp->snd_wnd) 1911 tcpstat.tcps_rcvdupack++; 1912 tp->t_dupacks = 0; 1913 break; 1914 } 1915 if (tlen != 0 || tiwin != tp->snd_wnd) { 1916 if (!tcp_do_rfc3517bis || 1917 !TCP_DO_SACK(tp) || 1918 (to.to_flags & 1919 (TOF_SACK | TOF_SACK_REDUNDANT)) 1920 != TOF_SACK) { 1921 tp->t_dupacks = 0; 1922 break; 1923 } 1924 /* 1925 * Update window information. 1926 */ 1927 if (tiwin != tp->snd_wnd && 1928 acceptable_window_update(tp, th, tiwin)) { 1929 /* keep track of pure window updates */ 1930 if (tlen == 0 && 1931 tp->snd_wl2 == th->th_ack && 1932 tiwin > tp->snd_wnd) 1933 tcpstat.tcps_rcvwinupd++; 1934 tp->snd_wnd = tiwin; 1935 tp->snd_wl1 = th->th_seq; 1936 tp->snd_wl2 = th->th_ack; 1937 if (tp->snd_wnd > tp->max_sndwnd) 1938 tp->max_sndwnd = tp->snd_wnd; 1939 } 1940 } 1941 tcpstat.tcps_rcvdupack++; 1942 1943 /* 1944 * We have outstanding data (other than 1945 * a window probe), this is a completely 1946 * duplicate ack (ie, window info didn't 1947 * change), the ack is the biggest we've 1948 * seen and we've seen exactly our rexmt 1949 * threshhold of them, so assume a packet 1950 * has been dropped and retransmit it. 1951 * Kludge snd_nxt & the congestion 1952 * window so we send only this one 1953 * packet. 1954 */ 1955 if (IN_FASTRECOVERY(tp)) { 1956 if (TCP_DO_SACK(tp)) { 1957 /* No artifical cwnd inflation. */ 1958 tcp_sack_rexmt(tp, th); 1959 } else { 1960 /* 1961 * Dup acks mean that packets 1962 * have left the network 1963 * (they're now cached at the 1964 * receiver) so bump cwnd by 1965 * the amount in the receiver 1966 * to keep a constant cwnd 1967 * packets in the network. 1968 */ 1969 tp->snd_cwnd += tp->t_maxseg; 1970 tcp_output(tp); 1971 } 1972 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1973 tp->t_dupacks = 0; 1974 break; 1975 } else if (tcp_ignore_redun_dsack && TCP_DO_SACK(tp) && 1976 (to.to_flags & (TOF_DSACK | TOF_SACK_REDUNDANT)) == 1977 (TOF_DSACK | TOF_SACK_REDUNDANT)) { 1978 /* 1979 * If the ACK carries DSACK and other 1980 * SACK blocks carry information that 1981 * we have already known, don't count 1982 * this ACK as duplicate ACK. This 1983 * prevents spurious early retransmit 1984 * and fast retransmit. This also 1985 * meets the requirement of RFC3042 1986 * that new segments should not be sent 1987 * if the SACK blocks do not contain 1988 * new information (XXX we actually 1989 * loosen the requirment that only DSACK 1990 * is checked here). 1991 * 1992 * This kind of ACKs are usually sent 1993 * after spurious retransmit. 1994 */ 1995 /* Do nothing; don't change t_dupacks */ 1996 } else if (++tp->t_dupacks == tp->t_rxtthresh) { 1997 tcp_seq old_snd_nxt; 1998 u_int win; 1999 2000 fastretransmit: 2001 if (tcp_do_eifel_detect && 2002 (tp->t_flags & TF_RCVD_TSTMP)) { 2003 tcp_save_congestion_state(tp); 2004 tp->rxt_flags |= TRXT_F_FASTREXMT; 2005 } 2006 /* 2007 * We know we're losing at the current 2008 * window size, so do congestion avoidance: 2009 * set ssthresh to half the current window 2010 * and pull our congestion window back to the 2011 * new ssthresh. 2012 */ 2013 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / 2014 tp->t_maxseg; 2015 if (win < 2) 2016 win = 2; 2017 tp->snd_ssthresh = win * tp->t_maxseg; 2018 ENTER_FASTRECOVERY(tp); 2019 tp->snd_recover = tp->snd_max; 2020 tcp_callout_stop(tp, tp->tt_rexmt); 2021 tp->t_rtttime = 0; 2022 old_snd_nxt = tp->snd_nxt; 2023 tp->snd_nxt = th->th_ack; 2024 tp->snd_cwnd = tp->t_maxseg; 2025 tcp_output(tp); 2026 ++tcpstat.tcps_sndfastrexmit; 2027 tp->snd_cwnd = tp->snd_ssthresh; 2028 tp->rexmt_high = tp->snd_nxt; 2029 tp->sack_flags &= ~TSACK_F_SACKRESCUED; 2030 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 2031 tp->snd_nxt = old_snd_nxt; 2032 KASSERT(tp->snd_limited <= 2, 2033 ("tp->snd_limited too big")); 2034 if (TCP_DO_SACK(tp)) 2035 tcp_sack_rexmt(tp, th); 2036 else 2037 tp->snd_cwnd += tp->t_maxseg * 2038 (tp->t_dupacks - tp->snd_limited); 2039 } else if (tcp_do_rfc3517bis && TCP_DO_SACK(tp)) { 2040 if (tcp_rfc3517bis_rxt && 2041 tcp_sack_islost(&tp->scb, tp->snd_una)) 2042 goto fastretransmit; 2043 if (tcp_do_limitedtransmit) { 2044 /* outstanding data */ 2045 uint32_t ownd = 2046 tp->snd_max - tp->snd_una; 2047 2048 if (!tcp_sack_limitedxmit(tp) && 2049 need_early_retransmit(tp, ownd)) { 2050 ++tcpstat.tcps_sndearlyrexmit; 2051 tp->rxt_flags |= 2052 TRXT_F_EARLYREXMT; 2053 goto fastretransmit; 2054 } 2055 } 2056 } else if (tcp_do_limitedtransmit) { 2057 u_long oldcwnd = tp->snd_cwnd; 2058 tcp_seq oldsndmax = tp->snd_max; 2059 tcp_seq oldsndnxt = tp->snd_nxt; 2060 /* outstanding data */ 2061 uint32_t ownd = tp->snd_max - tp->snd_una; 2062 u_int sent; 2063 2064 KASSERT(tp->t_dupacks == 1 || 2065 tp->t_dupacks == 2, 2066 ("dupacks not 1 or 2")); 2067 if (tp->t_dupacks == 1) 2068 tp->snd_limited = 0; 2069 tp->snd_nxt = tp->snd_max; 2070 tp->snd_cwnd = ownd + 2071 (tp->t_dupacks - tp->snd_limited) * 2072 tp->t_maxseg; 2073 tcp_output(tp); 2074 2075 if (SEQ_LT(oldsndnxt, oldsndmax)) { 2076 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una), 2077 ("snd_una moved in other threads")); 2078 tp->snd_nxt = oldsndnxt; 2079 } 2080 tp->snd_cwnd = oldcwnd; 2081 sent = tp->snd_max - oldsndmax; 2082 if (sent > tp->t_maxseg) { 2083 KASSERT((tp->t_dupacks == 2 && 2084 tp->snd_limited == 0) || 2085 (sent == tp->t_maxseg + 1 && 2086 tp->t_flags & TF_SENTFIN), 2087 ("sent too much")); 2088 KASSERT(sent <= tp->t_maxseg * 2, 2089 ("sent too many segments")); 2090 tp->snd_limited = 2; 2091 tcpstat.tcps_sndlimited += 2; 2092 } else if (sent > 0) { 2093 ++tp->snd_limited; 2094 ++tcpstat.tcps_sndlimited; 2095 } else if (need_early_retransmit(tp, ownd)) { 2096 ++tcpstat.tcps_sndearlyrexmit; 2097 tp->rxt_flags |= TRXT_F_EARLYREXMT; 2098 goto fastretransmit; 2099 } 2100 } 2101 if (tlen != 0) 2102 break; 2103 else 2104 goto drop; 2105 } 2106 2107 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 2108 tp->t_dupacks = 0; 2109 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2110 /* 2111 * Detected optimistic ACK attack. 2112 * Force slow-start to de-synchronize attack. 2113 */ 2114 tp->snd_cwnd = tp->t_maxseg; 2115 tp->snd_wacked = 0; 2116 2117 tcpstat.tcps_rcvacktoomuch++; 2118 goto dropafterack; 2119 } 2120 /* 2121 * If we reach this point, ACK is not a duplicate, 2122 * i.e., it ACKs something we sent. 2123 */ 2124 if (tp->t_flags & TF_NEEDSYN) { 2125 /* 2126 * T/TCP: Connection was half-synchronized, and our 2127 * SYN has been ACK'd (so connection is now fully 2128 * synchronized). Go to non-starred state, 2129 * increment snd_una for ACK of SYN, and check if 2130 * we can do window scaling. 2131 */ 2132 tp->t_flags &= ~TF_NEEDSYN; 2133 tp->snd_una++; 2134 /* Do window scaling? */ 2135 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 2136 (TF_RCVD_SCALE | TF_REQ_SCALE)) 2137 tp->rcv_scale = tp->request_r_scale; 2138 } 2139 2140 process_ACK: 2141 acked = th->th_ack - tp->snd_una; 2142 tcpstat.tcps_rcvackpack++; 2143 tcpstat.tcps_rcvackbyte += acked; 2144 2145 if (tcp_do_eifel_detect && acked > 0 && 2146 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) && 2147 (tp->rxt_flags & TRXT_F_FIRSTACCACK)) { 2148 /* Eifel detection applicable. */ 2149 if (to.to_tsecr < tp->t_rexmtTS) { 2150 ++tcpstat.tcps_eifeldetected; 2151 tcp_revert_congestion_state(tp); 2152 if (tp->t_rxtshift != 1 || 2153 ticks >= tp->t_badrxtwin) 2154 ++tcpstat.tcps_rttcantdetect; 2155 } 2156 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2157 /* 2158 * If we just performed our first retransmit, 2159 * and the ACK arrives within our recovery window, 2160 * then it was a mistake to do the retransmit 2161 * in the first place. Recover our original cwnd 2162 * and ssthresh, and proceed to transmit where we 2163 * left off. 2164 */ 2165 tcp_revert_congestion_state(tp); 2166 ++tcpstat.tcps_rttdetected; 2167 } 2168 2169 /* 2170 * If we have a timestamp reply, update smoothed 2171 * round trip time. If no timestamp is present but 2172 * transmit timer is running and timed sequence 2173 * number was acked, update smoothed round trip time. 2174 * Since we now have an rtt measurement, cancel the 2175 * timer backoff (cf., Phil Karn's retransmit alg.). 2176 * Recompute the initial retransmit timer. 2177 * 2178 * Some machines (certain windows boxes) send broken 2179 * timestamp replies during the SYN+ACK phase, ignore 2180 * timestamps of 0. 2181 */ 2182 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) 2183 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1, th->th_ack); 2184 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2185 tcp_xmit_timer(tp, ticks - tp->t_rtttime, th->th_ack); 2186 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2187 2188 /* 2189 * If no data (only SYN) was ACK'd, 2190 * skip rest of ACK processing. 2191 */ 2192 if (acked == 0) 2193 goto step6; 2194 2195 /* Stop looking for an acceptable ACK since one was received. */ 2196 tp->rxt_flags &= ~(TRXT_F_FIRSTACCACK | 2197 TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT); 2198 2199 if (acked > so->so_snd.ssb_cc) { 2200 tp->snd_wnd -= so->so_snd.ssb_cc; 2201 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc); 2202 ourfinisacked = TRUE; 2203 } else { 2204 sbdrop(&so->so_snd.sb, acked); 2205 tp->snd_wnd -= acked; 2206 ourfinisacked = FALSE; 2207 } 2208 sowwakeup(so); 2209 2210 /* 2211 * Update window information. 2212 */ 2213 if (acceptable_window_update(tp, th, tiwin)) { 2214 /* keep track of pure window updates */ 2215 if (tlen == 0 && tp->snd_wl2 == th->th_ack && 2216 tiwin > tp->snd_wnd) 2217 tcpstat.tcps_rcvwinupd++; 2218 tp->snd_wnd = tiwin; 2219 tp->snd_wl1 = th->th_seq; 2220 tp->snd_wl2 = th->th_ack; 2221 if (tp->snd_wnd > tp->max_sndwnd) 2222 tp->max_sndwnd = tp->snd_wnd; 2223 needoutput = TRUE; 2224 } 2225 2226 tp->snd_una = th->th_ack; 2227 if (TCP_DO_SACK(tp)) 2228 tcp_sack_update_scoreboard(tp, &to); 2229 if (IN_FASTRECOVERY(tp)) { 2230 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2231 EXIT_FASTRECOVERY(tp); 2232 needoutput = TRUE; 2233 /* 2234 * If the congestion window was inflated 2235 * to account for the other side's 2236 * cached packets, retract it. 2237 */ 2238 if (!TCP_DO_SACK(tp)) 2239 tp->snd_cwnd = tp->snd_ssthresh; 2240 2241 /* 2242 * Window inflation should have left us 2243 * with approximately snd_ssthresh outstanding 2244 * data. But, in case we would be inclined 2245 * to send a burst, better do it using 2246 * slow start. 2247 */ 2248 if (SEQ_GT(th->th_ack + tp->snd_cwnd, 2249 tp->snd_max + 2 * tp->t_maxseg)) 2250 tp->snd_cwnd = 2251 (tp->snd_max - tp->snd_una) + 2252 2 * tp->t_maxseg; 2253 2254 tp->snd_wacked = 0; 2255 } else { 2256 if (TCP_DO_SACK(tp)) { 2257 tp->snd_max_rexmt = tp->snd_max; 2258 tcp_sack_rexmt(tp, th); 2259 } else { 2260 tcp_newreno_partial_ack(tp, th, acked); 2261 } 2262 needoutput = FALSE; 2263 } 2264 } else { 2265 /* 2266 * Open the congestion window. When in slow-start, 2267 * open exponentially: maxseg per packet. Otherwise, 2268 * open linearly: maxseg per window. 2269 */ 2270 if (tp->snd_cwnd <= tp->snd_ssthresh) { 2271 u_int abc_sslimit = 2272 (SEQ_LT(tp->snd_nxt, tp->snd_max) ? 2273 tp->t_maxseg : 2 * tp->t_maxseg); 2274 2275 /* slow-start */ 2276 tp->snd_cwnd += tcp_do_abc ? 2277 min(acked, abc_sslimit) : tp->t_maxseg; 2278 } else { 2279 /* linear increase */ 2280 tp->snd_wacked += tcp_do_abc ? acked : 2281 tp->t_maxseg; 2282 if (tp->snd_wacked >= tp->snd_cwnd) { 2283 tp->snd_wacked -= tp->snd_cwnd; 2284 tp->snd_cwnd += tp->t_maxseg; 2285 } 2286 } 2287 tp->snd_cwnd = min(tp->snd_cwnd, 2288 TCP_MAXWIN << tp->snd_scale); 2289 tp->snd_recover = th->th_ack - 1; 2290 } 2291 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2292 tp->snd_nxt = tp->snd_una; 2293 2294 /* 2295 * If all outstanding data is acked, stop retransmit 2296 * timer and remember to restart (more output or persist). 2297 * If there is more data to be acked, restart retransmit 2298 * timer, using current (possibly backed-off) value. 2299 */ 2300 if (th->th_ack == tp->snd_max) { 2301 tcp_callout_stop(tp, tp->tt_rexmt); 2302 needoutput = TRUE; 2303 } else if (!tcp_callout_active(tp, tp->tt_persist)) { 2304 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur, 2305 tcp_timer_rexmt); 2306 } 2307 2308 switch (tp->t_state) { 2309 /* 2310 * In FIN_WAIT_1 STATE in addition to the processing 2311 * for the ESTABLISHED state if our FIN is now acknowledged 2312 * then enter FIN_WAIT_2. 2313 */ 2314 case TCPS_FIN_WAIT_1: 2315 if (ourfinisacked) { 2316 /* 2317 * If we can't receive any more 2318 * data, then closing user can proceed. 2319 * Starting the timer is contrary to the 2320 * specification, but if we don't get a FIN 2321 * we'll hang forever. 2322 */ 2323 if (so->so_state & SS_CANTRCVMORE) { 2324 soisdisconnected(so); 2325 tcp_callout_reset(tp, tp->tt_2msl, 2326 tp->t_maxidle, tcp_timer_2msl); 2327 } 2328 tp->t_state = TCPS_FIN_WAIT_2; 2329 } 2330 break; 2331 2332 /* 2333 * In CLOSING STATE in addition to the processing for 2334 * the ESTABLISHED state if the ACK acknowledges our FIN 2335 * then enter the TIME-WAIT state, otherwise ignore 2336 * the segment. 2337 */ 2338 case TCPS_CLOSING: 2339 if (ourfinisacked) { 2340 tp->t_state = TCPS_TIME_WAIT; 2341 tcp_canceltimers(tp); 2342 tcp_callout_reset(tp, tp->tt_2msl, 2343 2 * tcp_rmx_msl(tp), 2344 tcp_timer_2msl); 2345 soisdisconnected(so); 2346 } 2347 break; 2348 2349 /* 2350 * In LAST_ACK, we may still be waiting for data to drain 2351 * and/or to be acked, as well as for the ack of our FIN. 2352 * If our FIN is now acknowledged, delete the TCB, 2353 * enter the closed state and return. 2354 */ 2355 case TCPS_LAST_ACK: 2356 if (ourfinisacked) { 2357 tp = tcp_close(tp); 2358 goto drop; 2359 } 2360 break; 2361 2362 /* 2363 * In TIME_WAIT state the only thing that should arrive 2364 * is a retransmission of the remote FIN. Acknowledge 2365 * it and restart the finack timer. 2366 */ 2367 case TCPS_TIME_WAIT: 2368 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp), 2369 tcp_timer_2msl); 2370 goto dropafterack; 2371 } 2372 } 2373 2374 step6: 2375 /* 2376 * Update window information. 2377 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2378 */ 2379 if ((thflags & TH_ACK) && 2380 acceptable_window_update(tp, th, tiwin)) { 2381 /* keep track of pure window updates */ 2382 if (tlen == 0 && tp->snd_wl2 == th->th_ack && 2383 tiwin > tp->snd_wnd) 2384 tcpstat.tcps_rcvwinupd++; 2385 tp->snd_wnd = tiwin; 2386 tp->snd_wl1 = th->th_seq; 2387 tp->snd_wl2 = th->th_ack; 2388 if (tp->snd_wnd > tp->max_sndwnd) 2389 tp->max_sndwnd = tp->snd_wnd; 2390 needoutput = TRUE; 2391 } 2392 2393 /* 2394 * Process segments with URG. 2395 */ 2396 if ((thflags & TH_URG) && th->th_urp && 2397 !TCPS_HAVERCVDFIN(tp->t_state)) { 2398 /* 2399 * This is a kludge, but if we receive and accept 2400 * random urgent pointers, we'll crash in 2401 * soreceive. It's hard to imagine someone 2402 * actually wanting to send this much urgent data. 2403 */ 2404 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) { 2405 th->th_urp = 0; /* XXX */ 2406 thflags &= ~TH_URG; /* XXX */ 2407 goto dodata; /* XXX */ 2408 } 2409 /* 2410 * If this segment advances the known urgent pointer, 2411 * then mark the data stream. This should not happen 2412 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2413 * a FIN has been received from the remote side. 2414 * In these states we ignore the URG. 2415 * 2416 * According to RFC961 (Assigned Protocols), 2417 * the urgent pointer points to the last octet 2418 * of urgent data. We continue, however, 2419 * to consider it to indicate the first octet 2420 * of data past the urgent section as the original 2421 * spec states (in one of two places). 2422 */ 2423 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) { 2424 tp->rcv_up = th->th_seq + th->th_urp; 2425 so->so_oobmark = so->so_rcv.ssb_cc + 2426 (tp->rcv_up - tp->rcv_nxt) - 1; 2427 if (so->so_oobmark == 0) 2428 sosetstate(so, SS_RCVATMARK); 2429 sohasoutofband(so); 2430 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2431 } 2432 /* 2433 * Remove out of band data so doesn't get presented to user. 2434 * This can happen independent of advancing the URG pointer, 2435 * but if two URG's are pending at once, some out-of-band 2436 * data may creep in... ick. 2437 */ 2438 if (th->th_urp <= (u_long)tlen && 2439 !(so->so_options & SO_OOBINLINE)) { 2440 /* hdr drop is delayed */ 2441 tcp_pulloutofband(so, th, m, drop_hdrlen); 2442 } 2443 } else { 2444 /* 2445 * If no out of band data is expected, 2446 * pull receive urgent pointer along 2447 * with the receive window. 2448 */ 2449 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2450 tp->rcv_up = tp->rcv_nxt; 2451 } 2452 2453 dodata: /* XXX */ 2454 /* 2455 * Process the segment text, merging it into the TCP sequencing queue, 2456 * and arranging for acknowledgment of receipt if necessary. 2457 * This process logically involves adjusting tp->rcv_wnd as data 2458 * is presented to the user (this happens in tcp_usrreq.c, 2459 * case PRU_RCVD). If a FIN has already been received on this 2460 * connection then we just ignore the text. 2461 */ 2462 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) { 2463 m_adj(m, drop_hdrlen); /* delayed header drop */ 2464 /* 2465 * Insert segment which includes th into TCP reassembly queue 2466 * with control block tp. Set thflags to whether reassembly now 2467 * includes a segment with FIN. This handles the common case 2468 * inline (segment is the next to be received on an established 2469 * connection, and the queue is empty), avoiding linkage into 2470 * and removal from the queue and repetition of various 2471 * conversions. 2472 * Set DELACK for segments received in order, but ack 2473 * immediately when segments are out of order (so 2474 * fast retransmit can work). 2475 */ 2476 if (th->th_seq == tp->rcv_nxt && 2477 LIST_EMPTY(&tp->t_segq) && 2478 TCPS_HAVEESTABLISHED(tp->t_state)) { 2479 if (DELAY_ACK(tp)) { 2480 tcp_callout_reset(tp, tp->tt_delack, 2481 tcp_delacktime, tcp_timer_delack); 2482 } else { 2483 tp->t_flags |= TF_ACKNOW; 2484 } 2485 tp->rcv_nxt += tlen; 2486 thflags = th->th_flags & TH_FIN; 2487 tcpstat.tcps_rcvpack++; 2488 tcpstat.tcps_rcvbyte += tlen; 2489 ND6_HINT(tp); 2490 if (so->so_state & SS_CANTRCVMORE) { 2491 m_freem(m); 2492 } else { 2493 lwkt_gettoken(&so->so_rcv.ssb_token); 2494 ssb_appendstream(&so->so_rcv, m); 2495 lwkt_reltoken(&so->so_rcv.ssb_token); 2496 } 2497 sorwakeup(so); 2498 } else { 2499 if (!(tp->sack_flags & TSACK_F_DUPSEG)) { 2500 /* Initialize SACK report block. */ 2501 tp->reportblk.rblk_start = th->th_seq; 2502 tp->reportblk.rblk_end = TCP_SACK_BLKEND( 2503 th->th_seq + tlen, thflags); 2504 } 2505 thflags = tcp_reass(tp, th, &tlen, m); 2506 tp->t_flags |= TF_ACKNOW; 2507 } 2508 2509 /* 2510 * Note the amount of data that peer has sent into 2511 * our window, in order to estimate the sender's 2512 * buffer size. 2513 */ 2514 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2515 } else { 2516 m_freem(m); 2517 thflags &= ~TH_FIN; 2518 } 2519 2520 /* 2521 * If FIN is received ACK the FIN and let the user know 2522 * that the connection is closing. 2523 */ 2524 if (thflags & TH_FIN) { 2525 if (!TCPS_HAVERCVDFIN(tp->t_state)) { 2526 socantrcvmore(so); 2527 /* 2528 * If connection is half-synchronized 2529 * (ie NEEDSYN flag on) then delay ACK, 2530 * so it may be piggybacked when SYN is sent. 2531 * Otherwise, since we received a FIN then no 2532 * more input can be expected, send ACK now. 2533 */ 2534 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) { 2535 tcp_callout_reset(tp, tp->tt_delack, 2536 tcp_delacktime, tcp_timer_delack); 2537 } else { 2538 tp->t_flags |= TF_ACKNOW; 2539 } 2540 tp->rcv_nxt++; 2541 } 2542 2543 switch (tp->t_state) { 2544 /* 2545 * In SYN_RECEIVED and ESTABLISHED STATES 2546 * enter the CLOSE_WAIT state. 2547 */ 2548 case TCPS_SYN_RECEIVED: 2549 tp->t_starttime = ticks; 2550 /*FALLTHROUGH*/ 2551 case TCPS_ESTABLISHED: 2552 tp->t_state = TCPS_CLOSE_WAIT; 2553 break; 2554 2555 /* 2556 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2557 * enter the CLOSING state. 2558 */ 2559 case TCPS_FIN_WAIT_1: 2560 tp->t_state = TCPS_CLOSING; 2561 break; 2562 2563 /* 2564 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2565 * starting the time-wait timer, turning off the other 2566 * standard timers. 2567 */ 2568 case TCPS_FIN_WAIT_2: 2569 tp->t_state = TCPS_TIME_WAIT; 2570 tcp_canceltimers(tp); 2571 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp), 2572 tcp_timer_2msl); 2573 soisdisconnected(so); 2574 break; 2575 2576 /* 2577 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2578 */ 2579 case TCPS_TIME_WAIT: 2580 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp), 2581 tcp_timer_2msl); 2582 break; 2583 } 2584 } 2585 2586 #ifdef TCPDEBUG 2587 if (so->so_options & SO_DEBUG) 2588 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2589 #endif 2590 2591 /* 2592 * Return any desired output. 2593 */ 2594 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2595 tcp_output(tp); 2596 tcp_sack_report_cleanup(tp); 2597 return(IPPROTO_DONE); 2598 2599 dropafterack: 2600 /* 2601 * Generate an ACK dropping incoming segment if it occupies 2602 * sequence space, where the ACK reflects our state. 2603 * 2604 * We can now skip the test for the RST flag since all 2605 * paths to this code happen after packets containing 2606 * RST have been dropped. 2607 * 2608 * In the SYN-RECEIVED state, don't send an ACK unless the 2609 * segment we received passes the SYN-RECEIVED ACK test. 2610 * If it fails send a RST. This breaks the loop in the 2611 * "LAND" DoS attack, and also prevents an ACK storm 2612 * between two listening ports that have been sent forged 2613 * SYN segments, each with the source address of the other. 2614 */ 2615 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2616 (SEQ_GT(tp->snd_una, th->th_ack) || 2617 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2618 rstreason = BANDLIM_RST_OPENPORT; 2619 goto dropwithreset; 2620 } 2621 #ifdef TCPDEBUG 2622 if (so->so_options & SO_DEBUG) 2623 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2624 #endif 2625 m_freem(m); 2626 tp->t_flags |= TF_ACKNOW; 2627 tcp_output(tp); 2628 tcp_sack_report_cleanup(tp); 2629 return(IPPROTO_DONE); 2630 2631 dropwithreset: 2632 /* 2633 * Generate a RST, dropping incoming segment. 2634 * Make ACK acceptable to originator of segment. 2635 * Don't bother to respond if destination was broadcast/multicast. 2636 */ 2637 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST)) 2638 goto drop; 2639 if (isipv6) { 2640 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2641 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2642 goto drop; 2643 } else { 2644 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2645 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2646 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2647 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2648 goto drop; 2649 } 2650 /* IPv6 anycast check is done at tcp6_input() */ 2651 2652 /* 2653 * Perform bandwidth limiting. 2654 */ 2655 #ifdef ICMP_BANDLIM 2656 if (badport_bandlim(rstreason) < 0) 2657 goto drop; 2658 #endif 2659 2660 #ifdef TCPDEBUG 2661 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2662 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2663 #endif 2664 if (thflags & TH_ACK) 2665 /* mtod() below is safe as long as hdr dropping is delayed */ 2666 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2667 TH_RST); 2668 else { 2669 if (thflags & TH_SYN) 2670 tlen++; 2671 /* mtod() below is safe as long as hdr dropping is delayed */ 2672 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen, 2673 (tcp_seq)0, TH_RST | TH_ACK); 2674 } 2675 if (tp != NULL) 2676 tcp_sack_report_cleanup(tp); 2677 return(IPPROTO_DONE); 2678 2679 drop: 2680 /* 2681 * Drop space held by incoming segment and return. 2682 */ 2683 #ifdef TCPDEBUG 2684 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2685 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2686 #endif 2687 m_freem(m); 2688 if (tp != NULL) 2689 tcp_sack_report_cleanup(tp); 2690 return(IPPROTO_DONE); 2691 } 2692 2693 /* 2694 * Parse TCP options and place in tcpopt. 2695 */ 2696 static void 2697 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn, 2698 tcp_seq ack) 2699 { 2700 int opt, optlen, i; 2701 2702 to->to_flags = 0; 2703 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2704 opt = cp[0]; 2705 if (opt == TCPOPT_EOL) 2706 break; 2707 if (opt == TCPOPT_NOP) 2708 optlen = 1; 2709 else { 2710 if (cnt < 2) 2711 break; 2712 optlen = cp[1]; 2713 if (optlen < 2 || optlen > cnt) 2714 break; 2715 } 2716 switch (opt) { 2717 case TCPOPT_MAXSEG: 2718 if (optlen != TCPOLEN_MAXSEG) 2719 continue; 2720 if (!is_syn) 2721 continue; 2722 to->to_flags |= TOF_MSS; 2723 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss); 2724 to->to_mss = ntohs(to->to_mss); 2725 break; 2726 case TCPOPT_WINDOW: 2727 if (optlen != TCPOLEN_WINDOW) 2728 continue; 2729 if (!is_syn) 2730 continue; 2731 to->to_flags |= TOF_SCALE; 2732 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2733 break; 2734 case TCPOPT_TIMESTAMP: 2735 if (optlen != TCPOLEN_TIMESTAMP) 2736 continue; 2737 to->to_flags |= TOF_TS; 2738 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval); 2739 to->to_tsval = ntohl(to->to_tsval); 2740 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr); 2741 to->to_tsecr = ntohl(to->to_tsecr); 2742 /* 2743 * If echoed timestamp is later than the current time, 2744 * fall back to non RFC1323 RTT calculation. 2745 */ 2746 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks)) 2747 to->to_tsecr = 0; 2748 break; 2749 case TCPOPT_SACK_PERMITTED: 2750 if (optlen != TCPOLEN_SACK_PERMITTED) 2751 continue; 2752 if (!is_syn) 2753 continue; 2754 to->to_flags |= TOF_SACK_PERMITTED; 2755 break; 2756 case TCPOPT_SACK: 2757 if ((optlen - 2) & 0x07) /* not multiple of 8 */ 2758 continue; 2759 to->to_nsackblocks = (optlen - 2) / 8; 2760 to->to_sackblocks = (struct raw_sackblock *) (cp + 2); 2761 to->to_flags |= TOF_SACK; 2762 for (i = 0; i < to->to_nsackblocks; i++) { 2763 struct raw_sackblock *r = &to->to_sackblocks[i]; 2764 2765 r->rblk_start = ntohl(r->rblk_start); 2766 r->rblk_end = ntohl(r->rblk_end); 2767 2768 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) { 2769 /* 2770 * Invalid SACK block; discard all 2771 * SACK blocks 2772 */ 2773 tcpstat.tcps_rcvbadsackopt++; 2774 to->to_nsackblocks = 0; 2775 to->to_sackblocks = NULL; 2776 to->to_flags &= ~TOF_SACK; 2777 break; 2778 } 2779 } 2780 if ((to->to_flags & TOF_SACK) && 2781 tcp_sack_ndsack_blocks(to->to_sackblocks, 2782 to->to_nsackblocks, ack)) 2783 to->to_flags |= TOF_DSACK; 2784 break; 2785 #ifdef TCP_SIGNATURE 2786 /* 2787 * XXX In order to reply to a host which has set the 2788 * TCP_SIGNATURE option in its initial SYN, we have to 2789 * record the fact that the option was observed here 2790 * for the syncache code to perform the correct response. 2791 */ 2792 case TCPOPT_SIGNATURE: 2793 if (optlen != TCPOLEN_SIGNATURE) 2794 continue; 2795 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2796 break; 2797 #endif /* TCP_SIGNATURE */ 2798 default: 2799 continue; 2800 } 2801 } 2802 } 2803 2804 /* 2805 * Pull out of band byte out of a segment so 2806 * it doesn't appear in the user's data queue. 2807 * It is still reflected in the segment length for 2808 * sequencing purposes. 2809 * "off" is the delayed to be dropped hdrlen. 2810 */ 2811 static void 2812 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off) 2813 { 2814 int cnt = off + th->th_urp - 1; 2815 2816 while (cnt >= 0) { 2817 if (m->m_len > cnt) { 2818 char *cp = mtod(m, caddr_t) + cnt; 2819 struct tcpcb *tp = sototcpcb(so); 2820 2821 tp->t_iobc = *cp; 2822 tp->t_oobflags |= TCPOOB_HAVEDATA; 2823 bcopy(cp + 1, cp, m->m_len - cnt - 1); 2824 m->m_len--; 2825 if (m->m_flags & M_PKTHDR) 2826 m->m_pkthdr.len--; 2827 return; 2828 } 2829 cnt -= m->m_len; 2830 m = m->m_next; 2831 if (m == NULL) 2832 break; 2833 } 2834 panic("tcp_pulloutofband"); 2835 } 2836 2837 /* 2838 * Collect new round-trip time estimate 2839 * and update averages and current timeout. 2840 */ 2841 static void 2842 tcp_xmit_timer(struct tcpcb *tp, int rtt, tcp_seq ack) 2843 { 2844 int rebaserto = 0; 2845 2846 tcpstat.tcps_rttupdated++; 2847 tp->t_rttupdated++; 2848 if ((tp->rxt_flags & TRXT_F_REBASERTO) && 2849 SEQ_GT(ack, tp->snd_max_prev)) { 2850 #ifdef DEBUG_EIFEL_RESPONSE 2851 kprintf("srtt/rttvar, prev %d/%d, cur %d/%d, ", 2852 tp->t_srtt_prev, tp->t_rttvar_prev, 2853 tp->t_srtt, tp->t_rttvar); 2854 #endif 2855 2856 tcpstat.tcps_eifelresponse++; 2857 rebaserto = 1; 2858 tp->rxt_flags &= ~TRXT_F_REBASERTO; 2859 tp->t_srtt = max(tp->t_srtt_prev, (rtt << TCP_RTT_SHIFT)); 2860 tp->t_rttvar = max(tp->t_rttvar_prev, 2861 (rtt << (TCP_RTTVAR_SHIFT - 1))); 2862 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2863 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2864 2865 #ifdef DEBUG_EIFEL_RESPONSE 2866 kprintf("new %d/%d ", tp->t_srtt, tp->t_rttvar); 2867 #endif 2868 } else if (tp->t_srtt != 0) { 2869 int delta; 2870 2871 /* 2872 * srtt is stored as fixed point with 5 bits after the 2873 * binary point (i.e., scaled by 8). The following magic 2874 * is equivalent to the smoothing algorithm in rfc793 with 2875 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2876 * point). Adjust rtt to origin 0. 2877 */ 2878 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2879 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2880 2881 if ((tp->t_srtt += delta) <= 0) 2882 tp->t_srtt = 1; 2883 2884 /* 2885 * We accumulate a smoothed rtt variance (actually, a 2886 * smoothed mean difference), then set the retransmit 2887 * timer to smoothed rtt + 4 times the smoothed variance. 2888 * rttvar is stored as fixed point with 4 bits after the 2889 * binary point (scaled by 16). The following is 2890 * equivalent to rfc793 smoothing with an alpha of .75 2891 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2892 * rfc793's wired-in beta. 2893 */ 2894 if (delta < 0) 2895 delta = -delta; 2896 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2897 if ((tp->t_rttvar += delta) <= 0) 2898 tp->t_rttvar = 1; 2899 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2900 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2901 } else { 2902 /* 2903 * No rtt measurement yet - use the unsmoothed rtt. 2904 * Set the variance to half the rtt (so our first 2905 * retransmit happens at 3*rtt). 2906 */ 2907 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2908 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2909 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2910 } 2911 tp->t_rtttime = 0; 2912 tp->t_rxtshift = 0; 2913 2914 #ifdef DEBUG_EIFEL_RESPONSE 2915 if (rebaserto) { 2916 kprintf("| rxtcur prev %d, old %d, ", 2917 tp->t_rxtcur_prev, tp->t_rxtcur); 2918 } 2919 #endif 2920 2921 /* 2922 * the retransmit should happen at rtt + 4 * rttvar. 2923 * Because of the way we do the smoothing, srtt and rttvar 2924 * will each average +1/2 tick of bias. When we compute 2925 * the retransmit timer, we want 1/2 tick of rounding and 2926 * 1 extra tick because of +-1/2 tick uncertainty in the 2927 * firing of the timer. The bias will give us exactly the 2928 * 1.5 tick we need. But, because the bias is 2929 * statistical, we have to test that we don't drop below 2930 * the minimum feasible timer (which is 2 ticks). 2931 */ 2932 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2933 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2934 2935 if (rebaserto) { 2936 if (tp->t_rxtcur < tp->t_rxtcur_prev + tcp_eifel_rtoinc) { 2937 /* 2938 * RFC4015 requires that the new RTO is at least 2939 * 2*G (tcp_eifel_rtoinc) greater then the RTO 2940 * (t_rxtcur_prev) when the spurious retransmit 2941 * timeout happens. 2942 * 2943 * The above condition could be true, if the SRTT 2944 * and RTTVAR used to calculate t_rxtcur_prev 2945 * resulted in a value less than t_rttmin. So 2946 * simply increasing SRTT by tcp_eifel_rtoinc when 2947 * preparing for the Eifel response in 2948 * tcp_save_congestion_state() could not ensure 2949 * that the new RTO will be tcp_eifel_rtoinc greater 2950 * t_rxtcur_prev. 2951 */ 2952 tp->t_rxtcur = tp->t_rxtcur_prev + tcp_eifel_rtoinc; 2953 } 2954 #ifdef DEBUG_EIFEL_RESPONSE 2955 kprintf("new %d\n", tp->t_rxtcur); 2956 #endif 2957 } 2958 2959 /* 2960 * We received an ack for a packet that wasn't retransmitted; 2961 * it is probably safe to discard any error indications we've 2962 * received recently. This isn't quite right, but close enough 2963 * for now (a route might have failed after we sent a segment, 2964 * and the return path might not be symmetrical). 2965 */ 2966 tp->t_softerror = 0; 2967 } 2968 2969 /* 2970 * Determine a reasonable value for maxseg size. 2971 * If the route is known, check route for mtu. 2972 * If none, use an mss that can be handled on the outgoing 2973 * interface without forcing IP to fragment; if bigger than 2974 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2975 * to utilize large mbufs. If no route is found, route has no mtu, 2976 * or the destination isn't local, use a default, hopefully conservative 2977 * size (usually 512 or the default IP max size, but no more than the mtu 2978 * of the interface), as we can't discover anything about intervening 2979 * gateways or networks. We also initialize the congestion/slow start 2980 * window to be a single segment if the destination isn't local. 2981 * While looking at the routing entry, we also initialize other path-dependent 2982 * parameters from pre-set or cached values in the routing entry. 2983 * 2984 * Also take into account the space needed for options that we 2985 * send regularly. Make maxseg shorter by that amount to assure 2986 * that we can send maxseg amount of data even when the options 2987 * are present. Store the upper limit of the length of options plus 2988 * data in maxopd. 2989 * 2990 * NOTE that this routine is only called when we process an incoming 2991 * segment, for outgoing segments only tcp_mssopt is called. 2992 */ 2993 void 2994 tcp_mss(struct tcpcb *tp, int offer) 2995 { 2996 struct rtentry *rt; 2997 struct ifnet *ifp; 2998 int rtt, mss; 2999 u_long bufsize; 3000 struct inpcb *inp = tp->t_inpcb; 3001 struct socket *so; 3002 #ifdef INET6 3003 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE); 3004 size_t min_protoh = isipv6 ? 3005 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 3006 sizeof(struct tcpiphdr); 3007 #else 3008 const boolean_t isipv6 = FALSE; 3009 const size_t min_protoh = sizeof(struct tcpiphdr); 3010 #endif 3011 3012 if (isipv6) 3013 rt = tcp_rtlookup6(&inp->inp_inc); 3014 else 3015 rt = tcp_rtlookup(&inp->inp_inc); 3016 if (rt == NULL) { 3017 tp->t_maxopd = tp->t_maxseg = 3018 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 3019 return; 3020 } 3021 ifp = rt->rt_ifp; 3022 so = inp->inp_socket; 3023 3024 /* 3025 * Offer == 0 means that there was no MSS on the SYN segment, 3026 * in this case we use either the interface mtu or tcp_mssdflt. 3027 * 3028 * An offer which is too large will be cut down later. 3029 */ 3030 if (offer == 0) { 3031 if (isipv6) { 3032 if (in6_localaddr(&inp->in6p_faddr)) { 3033 offer = ND_IFINFO(rt->rt_ifp)->linkmtu - 3034 min_protoh; 3035 } else { 3036 offer = tcp_v6mssdflt; 3037 } 3038 } else { 3039 if (in_localaddr(inp->inp_faddr)) 3040 offer = ifp->if_mtu - min_protoh; 3041 else 3042 offer = tcp_mssdflt; 3043 } 3044 } 3045 3046 /* 3047 * Prevent DoS attack with too small MSS. Round up 3048 * to at least minmss. 3049 * 3050 * Sanity check: make sure that maxopd will be large 3051 * enough to allow some data on segments even is the 3052 * all the option space is used (40bytes). Otherwise 3053 * funny things may happen in tcp_output. 3054 */ 3055 offer = max(offer, tcp_minmss); 3056 offer = max(offer, 64); 3057 3058 rt->rt_rmx.rmx_mssopt = offer; 3059 3060 /* 3061 * While we're here, check if there's an initial rtt 3062 * or rttvar. Convert from the route-table units 3063 * to scaled multiples of the slow timeout timer. 3064 */ 3065 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 3066 /* 3067 * XXX the lock bit for RTT indicates that the value 3068 * is also a minimum value; this is subject to time. 3069 */ 3070 if (rt->rt_rmx.rmx_locks & RTV_RTT) 3071 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 3072 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 3073 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3074 tcpstat.tcps_usedrtt++; 3075 if (rt->rt_rmx.rmx_rttvar) { 3076 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 3077 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 3078 tcpstat.tcps_usedrttvar++; 3079 } else { 3080 /* default variation is +- 1 rtt */ 3081 tp->t_rttvar = 3082 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3083 } 3084 TCPT_RANGESET(tp->t_rxtcur, 3085 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3086 tp->t_rttmin, TCPTV_REXMTMAX); 3087 } 3088 3089 /* 3090 * if there's an mtu associated with the route, use it 3091 * else, use the link mtu. Take the smaller of mss or offer 3092 * as our final mss. 3093 */ 3094 if (rt->rt_rmx.rmx_mtu) { 3095 mss = rt->rt_rmx.rmx_mtu - min_protoh; 3096 } else { 3097 if (isipv6) 3098 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh; 3099 else 3100 mss = ifp->if_mtu - min_protoh; 3101 } 3102 mss = min(mss, offer); 3103 3104 /* 3105 * maxopd stores the maximum length of data AND options 3106 * in a segment; maxseg is the amount of data in a normal 3107 * segment. We need to store this value (maxopd) apart 3108 * from maxseg, because now every segment carries options 3109 * and thus we normally have somewhat less data in segments. 3110 */ 3111 tp->t_maxopd = mss; 3112 3113 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP && 3114 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3115 mss -= TCPOLEN_TSTAMP_APPA; 3116 3117 #if (MCLBYTES & (MCLBYTES - 1)) == 0 3118 if (mss > MCLBYTES) 3119 mss &= ~(MCLBYTES-1); 3120 #else 3121 if (mss > MCLBYTES) 3122 mss = mss / MCLBYTES * MCLBYTES; 3123 #endif 3124 /* 3125 * If there's a pipesize, change the socket buffer 3126 * to that size. Make the socket buffers an integral 3127 * number of mss units; if the mss is larger than 3128 * the socket buffer, decrease the mss. 3129 */ 3130 #ifdef RTV_SPIPE 3131 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 3132 #endif 3133 bufsize = so->so_snd.ssb_hiwat; 3134 if (bufsize < mss) 3135 mss = bufsize; 3136 else { 3137 bufsize = roundup(bufsize, mss); 3138 if (bufsize > sb_max) 3139 bufsize = sb_max; 3140 if (bufsize > so->so_snd.ssb_hiwat) 3141 ssb_reserve(&so->so_snd, bufsize, so, NULL); 3142 } 3143 tp->t_maxseg = mss; 3144 3145 #ifdef RTV_RPIPE 3146 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 3147 #endif 3148 bufsize = so->so_rcv.ssb_hiwat; 3149 if (bufsize > mss) { 3150 bufsize = roundup(bufsize, mss); 3151 if (bufsize > sb_max) 3152 bufsize = sb_max; 3153 if (bufsize > so->so_rcv.ssb_hiwat) { 3154 lwkt_gettoken(&so->so_rcv.ssb_token); 3155 ssb_reserve(&so->so_rcv, bufsize, so, NULL); 3156 lwkt_reltoken(&so->so_rcv.ssb_token); 3157 } 3158 } 3159 3160 /* 3161 * Set the slow-start flight size 3162 * 3163 * NOTE: t_maxseg must have been configured! 3164 */ 3165 tp->snd_cwnd = tcp_initial_window(tp); 3166 3167 if (rt->rt_rmx.rmx_ssthresh) { 3168 /* 3169 * There's some sort of gateway or interface 3170 * buffer limit on the path. Use this to set 3171 * the slow start threshhold, but set the 3172 * threshold to no less than 2*mss. 3173 */ 3174 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 3175 tcpstat.tcps_usedssthresh++; 3176 } 3177 } 3178 3179 /* 3180 * Determine the MSS option to send on an outgoing SYN. 3181 */ 3182 int 3183 tcp_mssopt(struct tcpcb *tp) 3184 { 3185 struct rtentry *rt; 3186 #ifdef INET6 3187 boolean_t isipv6 = 3188 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE); 3189 int min_protoh = isipv6 ? 3190 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 3191 sizeof(struct tcpiphdr); 3192 #else 3193 const boolean_t isipv6 = FALSE; 3194 const size_t min_protoh = sizeof(struct tcpiphdr); 3195 #endif 3196 3197 if (isipv6) 3198 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 3199 else 3200 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 3201 if (rt == NULL) 3202 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 3203 3204 return (rt->rt_ifp->if_mtu - min_protoh); 3205 } 3206 3207 /* 3208 * When a partial ack arrives, force the retransmission of the 3209 * next unacknowledged segment. Do not exit Fast Recovery. 3210 * 3211 * Implement the Slow-but-Steady variant of NewReno by restarting the 3212 * the retransmission timer. Turn it off here so it can be restarted 3213 * later in tcp_output(). 3214 */ 3215 static void 3216 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked) 3217 { 3218 tcp_seq old_snd_nxt = tp->snd_nxt; 3219 u_long ocwnd = tp->snd_cwnd; 3220 3221 tcp_callout_stop(tp, tp->tt_rexmt); 3222 tp->t_rtttime = 0; 3223 tp->snd_nxt = th->th_ack; 3224 /* Set snd_cwnd to one segment beyond acknowledged offset. */ 3225 tp->snd_cwnd = tp->t_maxseg; 3226 tp->t_flags |= TF_ACKNOW; 3227 tcp_output(tp); 3228 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 3229 tp->snd_nxt = old_snd_nxt; 3230 /* partial window deflation */ 3231 if (ocwnd > acked) 3232 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg; 3233 else 3234 tp->snd_cwnd = tp->t_maxseg; 3235 } 3236 3237 /* 3238 * In contrast to the Slow-but-Steady NewReno variant, 3239 * we do not reset the retransmission timer for SACK retransmissions, 3240 * except when retransmitting snd_una. 3241 */ 3242 static void 3243 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th) 3244 { 3245 tcp_seq old_snd_nxt = tp->snd_nxt; 3246 u_long ocwnd = tp->snd_cwnd; 3247 uint32_t pipe; 3248 int nseg = 0; /* consecutive new segments */ 3249 int nseg_rexmt = 0; /* retransmitted segments */ 3250 #define MAXBURST 4 /* limit burst of new packets on partial ack */ 3251 3252 tp->t_rtttime = 0; 3253 pipe = tcp_sack_compute_pipe(tp); 3254 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg && 3255 (!tcp_do_smartsack || nseg < MAXBURST)) { 3256 tcp_seq old_snd_max, old_rexmt_high, nextrexmt; 3257 uint32_t sent, seglen; 3258 boolean_t rescue; 3259 int error; 3260 3261 old_rexmt_high = tp->rexmt_high; 3262 if (!tcp_sack_nextseg(tp, &nextrexmt, &seglen, &rescue)) { 3263 tp->rexmt_high = old_rexmt_high; 3264 break; 3265 } 3266 3267 /* 3268 * If the next tranmission is a rescue retranmission, 3269 * we check whether we have already sent some data 3270 * (either new segments or retransmitted segments) 3271 * into the the network or not. Since the idea of rescue 3272 * retransmission is to sustain ACK clock, as long as 3273 * some segments are in the network, ACK clock will be 3274 * kept ticking. 3275 */ 3276 if (rescue && (nseg_rexmt > 0 || nseg > 0)) { 3277 tp->rexmt_high = old_rexmt_high; 3278 break; 3279 } 3280 3281 if (nextrexmt == tp->snd_max) 3282 ++nseg; 3283 else 3284 ++nseg_rexmt; 3285 tp->snd_nxt = nextrexmt; 3286 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen; 3287 old_snd_max = tp->snd_max; 3288 if (nextrexmt == tp->snd_una) 3289 tcp_callout_stop(tp, tp->tt_rexmt); 3290 error = tcp_output(tp); 3291 if (error != 0) { 3292 tp->rexmt_high = old_rexmt_high; 3293 break; 3294 } 3295 sent = tp->snd_nxt - nextrexmt; 3296 if (sent <= 0) { 3297 tp->rexmt_high = old_rexmt_high; 3298 break; 3299 } 3300 pipe += sent; 3301 tcpstat.tcps_sndsackpack++; 3302 tcpstat.tcps_sndsackbyte += sent; 3303 3304 if (rescue) { 3305 tcpstat.tcps_sackrescue++; 3306 tp->rexmt_rescue = tp->snd_nxt; 3307 tp->sack_flags |= TSACK_F_SACKRESCUED; 3308 break; 3309 } 3310 if (SEQ_LT(nextrexmt, old_snd_max) && 3311 SEQ_LT(tp->rexmt_high, tp->snd_nxt)) { 3312 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max); 3313 if (tcp_aggressive_rescuesack && 3314 (tp->sack_flags & TSACK_F_SACKRESCUED) && 3315 SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) { 3316 /* Drag RescueRxt along with HighRxt */ 3317 tp->rexmt_rescue = tp->rexmt_high; 3318 } 3319 } 3320 } 3321 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 3322 tp->snd_nxt = old_snd_nxt; 3323 tp->snd_cwnd = ocwnd; 3324 } 3325 3326 static boolean_t 3327 tcp_sack_limitedxmit(struct tcpcb *tp) 3328 { 3329 tcp_seq oldsndnxt = tp->snd_nxt; 3330 tcp_seq oldsndmax = tp->snd_max; 3331 u_long ocwnd = tp->snd_cwnd; 3332 uint32_t pipe; 3333 boolean_t ret = FALSE; 3334 3335 tp->rexmt_high = tp->snd_una - 1; 3336 pipe = tcp_sack_compute_pipe(tp); 3337 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg) { 3338 uint32_t sent; 3339 tcp_seq next; 3340 int error; 3341 3342 next = tp->snd_nxt = tp->snd_max; 3343 tp->snd_cwnd = tp->snd_nxt - tp->snd_una + tp->t_maxseg; 3344 3345 error = tcp_output(tp); 3346 if (error) 3347 break; 3348 3349 sent = tp->snd_nxt - next; 3350 if (sent <= 0) 3351 break; 3352 pipe += sent; 3353 ++tcpstat.tcps_sndlimited; 3354 ret = TRUE; 3355 } 3356 3357 if (SEQ_LT(oldsndnxt, oldsndmax)) { 3358 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una), 3359 ("snd_una moved in other threads")); 3360 tp->snd_nxt = oldsndnxt; 3361 } 3362 tp->snd_cwnd = ocwnd; 3363 3364 return ret; 3365 } 3366 3367 /* 3368 * Reset idle time and keep-alive timer, typically called when a valid 3369 * tcp packet is received but may also be called when FASTKEEP is set 3370 * to prevent the previous long-timeout from calculating to a drop. 3371 * 3372 * Only update t_rcvtime for non-SYN packets. 3373 * 3374 * Handle the case where one side thinks the connection is established 3375 * but the other side has, say, rebooted without cleaning out the 3376 * connection. The SYNs could be construed as an attack and wind 3377 * up ignored, but in case it isn't an attack we can validate the 3378 * connection by forcing a keepalive. 3379 */ 3380 void 3381 tcp_timer_keep_activity(struct tcpcb *tp, int thflags) 3382 { 3383 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3384 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) { 3385 tp->t_flags |= TF_KEEPALIVE; 3386 tcp_callout_reset(tp, tp->tt_keep, hz / 2, 3387 tcp_timer_keep); 3388 } else { 3389 tp->t_rcvtime = ticks; 3390 tp->t_flags &= ~TF_KEEPALIVE; 3391 tcp_callout_reset(tp, tp->tt_keep, 3392 tp->t_keepidle, 3393 tcp_timer_keep); 3394 } 3395 } 3396 } 3397 3398 static int 3399 tcp_rmx_msl(const struct tcpcb *tp) 3400 { 3401 struct rtentry *rt; 3402 struct inpcb *inp = tp->t_inpcb; 3403 int msl; 3404 #ifdef INET6 3405 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE); 3406 #else 3407 const boolean_t isipv6 = FALSE; 3408 #endif 3409 3410 if (isipv6) 3411 rt = tcp_rtlookup6(&inp->inp_inc); 3412 else 3413 rt = tcp_rtlookup(&inp->inp_inc); 3414 if (rt == NULL || rt->rt_rmx.rmx_msl == 0) 3415 return tcp_msl; 3416 3417 msl = (rt->rt_rmx.rmx_msl * hz) / 1000; 3418 if (msl == 0) 3419 msl = 1; 3420 3421 return msl; 3422 } 3423 3424 static void 3425 tcp_established(struct tcpcb *tp) 3426 { 3427 tp->t_state = TCPS_ESTABLISHED; 3428 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle, tcp_timer_keep); 3429 3430 if (tp->t_rxtsyn > 0) { 3431 /* 3432 * RFC6298: 3433 * "If the timer expires awaiting the ACK of a SYN segment 3434 * and the TCP implementation is using an RTO less than 3 3435 * seconds, the RTO MUST be re-initialized to 3 seconds 3436 * when data transmission begins" 3437 */ 3438 if (tp->t_rxtcur < TCPTV_RTOBASE3) 3439 tp->t_rxtcur = TCPTV_RTOBASE3; 3440 } 3441 } 3442