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. Neither the name of the University nor the names of its contributors 47 * may be used to endorse or promote products derived from this software 48 * without specific prior written permission. 49 * 50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 * 62 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 63 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $ 64 */ 65 66 #include "opt_inet.h" 67 #include "opt_inet6.h" 68 #include "opt_ipsec.h" 69 #include "opt_tcpdebug.h" 70 #include "opt_tcp_input.h" 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/kernel.h> 75 #include <sys/sysctl.h> 76 #include <sys/malloc.h> 77 #include <sys/mbuf.h> 78 #include <sys/proc.h> /* for proc0 declaration */ 79 #include <sys/protosw.h> 80 #include <sys/socket.h> 81 #include <sys/socketvar.h> 82 #include <sys/syslog.h> 83 #include <sys/in_cksum.h> 84 85 #include <sys/socketvar2.h> 86 87 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 88 #include <machine/stdarg.h> 89 90 #include <net/if.h> 91 #include <net/route.h> 92 93 #include <netinet/in.h> 94 #include <netinet/in_systm.h> 95 #include <netinet/ip.h> 96 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 97 #include <netinet/in_var.h> 98 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 99 #include <netinet/in_pcb.h> 100 #include <netinet/ip_var.h> 101 #include <netinet/ip6.h> 102 #include <netinet/icmp6.h> 103 #include <netinet6/nd6.h> 104 #include <netinet6/ip6_var.h> 105 #include <netinet6/in6_pcb.h> 106 #include <netinet/tcp.h> 107 #include <netinet/tcp_fsm.h> 108 #include <netinet/tcp_seq.h> 109 #include <netinet/tcp_timer.h> 110 #include <netinet/tcp_timer2.h> 111 #include <netinet/tcp_var.h> 112 #include <netinet6/tcp6_var.h> 113 #include <netinet/tcpip.h> 114 115 #ifdef TCPDEBUG 116 #include <netinet/tcp_debug.h> 117 118 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 119 struct tcphdr tcp_savetcp; 120 #endif 121 122 #ifdef FAST_IPSEC 123 #include <netproto/ipsec/ipsec.h> 124 #include <netproto/ipsec/ipsec6.h> 125 #endif 126 127 #ifdef IPSEC 128 #include <netinet6/ipsec.h> 129 #include <netinet6/ipsec6.h> 130 #include <netproto/key/key.h> 131 #endif 132 133 /* 134 * Limit burst of new packets during SACK based fast recovery 135 * or extended limited transmit. 136 */ 137 #define TCP_SACK_MAXBURST 4 138 139 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 140 141 static int log_in_vain = 0; 142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 143 &log_in_vain, 0, "Log all incoming TCP connections"); 144 145 static int blackhole = 0; 146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 147 &blackhole, 0, "Do not send RST when dropping refused connections"); 148 149 int tcp_delack_enabled = 1; 150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 151 &tcp_delack_enabled, 0, 152 "Delay ACK to try and piggyback it onto a data packet"); 153 154 #ifdef TCP_DROP_SYNFIN 155 static int drop_synfin = 0; 156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 157 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 158 #endif 159 160 static int tcp_do_limitedtransmit = 1; 161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW, 162 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)"); 163 164 static int tcp_do_early_retransmit = 1; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW, 166 &tcp_do_early_retransmit, 0, "Early retransmit"); 167 168 int tcp_aggregate_acks = 1; 169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW, 170 &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack"); 171 172 static int tcp_do_eifel_detect = 1; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW, 174 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)"); 175 176 static int tcp_do_abc = 1; 177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW, 178 &tcp_do_abc, 0, 179 "TCP Appropriate Byte Counting (RFC 3465)"); 180 181 /* 182 * The following value actually takes range [25ms, 250ms], 183 * given that most modern systems use 1ms ~ 10ms as the unit 184 * of timestamp option. 185 */ 186 static u_int tcp_paws_tolerance = 25; 187 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, paws_tolerance, CTLFLAG_RW, 188 &tcp_paws_tolerance, 0, "RFC1323 PAWS tolerance"); 189 190 /* 191 * Define as tunable for easy testing with SACK on and off. 192 * Warning: do not change setting in the middle of an existing active TCP flow, 193 * else strange things might happen to that flow. 194 */ 195 int tcp_do_sack = 1; 196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 197 &tcp_do_sack, 0, "Enable SACK Algorithms"); 198 199 int tcp_do_smartsack = 1; 200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW, 201 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms"); 202 203 int tcp_do_rescuesack = 1; 204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack, CTLFLAG_RW, 205 &tcp_do_rescuesack, 0, "Rescue retransmission for SACK"); 206 207 int tcp_aggressive_rescuesack = 0; 208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack_agg, CTLFLAG_RW, 209 &tcp_aggressive_rescuesack, 0, "Aggressive rescue retransmission for SACK"); 210 211 static int tcp_force_sackrxt = 1; 212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, force_sackrxt, CTLFLAG_RW, 213 &tcp_force_sackrxt, 0, "Allowed forced SACK retransmit burst"); 214 215 int tcp_do_rfc6675 = 1; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675, CTLFLAG_RW, 217 &tcp_do_rfc6675, 0, "Enable RFC6675"); 218 219 int tcp_rfc6675_rxt = 0; 220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_rxt, CTLFLAG_RW, 221 &tcp_rfc6675_rxt, 0, "Enable RFC6675 retransmit"); 222 223 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 224 "TCP Segment Reassembly Queue"); 225 226 int tcp_reass_maxseg = 0; 227 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD, 228 &tcp_reass_maxseg, 0, 229 "Global maximum number of TCP Segments in Reassembly Queue"); 230 231 int tcp_reass_qsize = 0; 232 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 233 &tcp_reass_qsize, 0, 234 "Global number of TCP Segments currently in Reassembly Queue"); 235 236 static int tcp_reass_overflows = 0; 237 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 238 &tcp_reass_overflows, 0, 239 "Global number of TCP Segment Reassembly Queue Overflows"); 240 241 int tcp_do_autorcvbuf = 1; 242 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 243 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing"); 244 245 int tcp_autorcvbuf_inc = 16*1024; 246 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 247 &tcp_autorcvbuf_inc, 0, 248 "Incrementor step size of automatic receive buffer"); 249 250 int tcp_autorcvbuf_max = 2*1024*1024; 251 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 252 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer"); 253 254 int tcp_sosend_agglim = 2; 255 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_agglim, CTLFLAG_RW, 256 &tcp_sosend_agglim, 0, "TCP sosend mbuf aggregation limit"); 257 258 int tcp_sosend_async = 1; 259 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_async, CTLFLAG_RW, 260 &tcp_sosend_async, 0, "TCP asynchronized pru_send"); 261 262 int tcp_sosend_jcluster = 1; 263 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_jcluster, CTLFLAG_RW, 264 &tcp_sosend_jcluster, 0, "TCP output uses jcluster"); 265 266 static int tcp_ignore_redun_dsack = 1; 267 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ignore_redun_dsack, CTLFLAG_RW, 268 &tcp_ignore_redun_dsack, 0, "Ignore redundant DSACK"); 269 270 static int tcp_reuseport_ext = 1; 271 SYSCTL_INT(_net_inet_tcp, OID_AUTO, reuseport_ext, CTLFLAG_RW, 272 &tcp_reuseport_ext, 0, "SO_REUSEPORT extension"); 273 274 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t, 275 tcp_seq); 276 static void tcp_pulloutofband(struct socket *, 277 struct tcphdr *, struct mbuf *, int); 278 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 279 struct mbuf *); 280 static void tcp_xmit_timer(struct tcpcb *, int, tcp_seq); 281 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int); 282 static void tcp_sack_rexmt(struct tcpcb *, boolean_t); 283 static boolean_t tcp_sack_limitedxmit(struct tcpcb *); 284 static int tcp_rmx_msl(const struct tcpcb *); 285 static void tcp_established(struct tcpcb *); 286 static boolean_t tcp_recv_dupack(struct tcpcb *, tcp_seq, u_int); 287 288 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 289 #ifdef INET6 290 #define ND6_HINT(tp) \ 291 do { \ 292 if ((tp) && (tp)->t_inpcb && \ 293 INP_ISIPV6((tp)->t_inpcb) && \ 294 (tp)->t_inpcb->in6p_route.ro_rt) \ 295 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 296 } while (0) 297 #else 298 #define ND6_HINT(tp) 299 #endif 300 301 /* 302 * Indicate whether this ack should be delayed. We can delay the ack if 303 * - delayed acks are enabled and 304 * - there is no delayed ack timer in progress and 305 * - our last ack wasn't a 0-sized window. We never want to delay 306 * the ack that opens up a 0-sized window. 307 */ 308 #define DELAY_ACK(tp) \ 309 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \ 310 !(tp->t_flags & TF_RXWIN0SENT)) 311 312 #define acceptable_window_update(tp, th, tiwin) \ 313 (SEQ_LT(tp->snd_wl1, th->th_seq) || \ 314 (tp->snd_wl1 == th->th_seq && \ 315 (SEQ_LT(tp->snd_wl2, th->th_ack) || \ 316 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) 317 318 #define iceildiv(n, d) (((n)+(d)-1) / (d)) 319 #define need_early_retransmit(tp, ownd) \ 320 (tcp_do_early_retransmit && \ 321 (tcp_do_eifel_detect && (tp->t_flags & TF_RCVD_TSTMP)) && \ 322 ownd < ((tp->t_rxtthresh + 1) * tp->t_maxseg) && \ 323 tp->t_dupacks + 1 >= iceildiv(ownd, tp->t_maxseg) && \ 324 (!TCP_DO_SACK(tp) || ownd <= tp->t_maxseg || \ 325 tcp_sack_has_sacked(&tp->scb, ownd - tp->t_maxseg))) 326 327 /* 328 * Returns TRUE, if this segment can be merged with the last 329 * pending segment in the reassemble queue and this segment 330 * does not overlap with the pending segment immediately 331 * preceeding the last pending segment. 332 */ 333 static __inline boolean_t 334 tcp_paws_canreasslast(const struct tcpcb *tp, const struct tcphdr *th, int tlen) 335 { 336 const struct tseg_qent *last, *prev; 337 338 last = TAILQ_LAST(&tp->t_segq, tsegqe_head); 339 if (last == NULL) 340 return FALSE; 341 342 /* This segment comes immediately after the last pending segment */ 343 if (last->tqe_th->th_seq + last->tqe_len == th->th_seq) { 344 if (last->tqe_th->th_flags & TH_FIN) { 345 /* No segments should follow segment w/ FIN */ 346 return FALSE; 347 } 348 return TRUE; 349 } 350 351 if (th->th_seq + tlen != last->tqe_th->th_seq) 352 return FALSE; 353 /* This segment comes immediately before the last pending segment */ 354 355 prev = TAILQ_PREV(last, tsegqe_head, tqe_q); 356 if (prev == NULL) { 357 /* 358 * No pending preceeding segment, we assume this segment 359 * could be reassembled. 360 */ 361 return TRUE; 362 } 363 364 /* This segment does not overlap with the preceeding segment */ 365 if (SEQ_GEQ(th->th_seq, prev->tqe_th->th_seq + prev->tqe_len)) 366 return TRUE; 367 368 return FALSE; 369 } 370 371 static __inline void 372 tcp_ncr_update_rxtthresh(struct tcpcb *tp) 373 { 374 int old_rxtthresh = tp->t_rxtthresh; 375 uint32_t ownd = tp->snd_max - tp->snd_una; 376 377 tp->t_rxtthresh = min(tcp_ncr_rxtthresh_max, 378 max(tcprexmtthresh, ((ownd / tp->t_maxseg) >> 1))); 379 if (tp->t_rxtthresh != old_rxtthresh) { 380 tcp_sack_update_lostseq(&tp->scb, tp->snd_una, 381 tp->t_maxseg, tp->t_rxtthresh); 382 } 383 } 384 385 static int 386 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m) 387 { 388 struct tseg_qent *q; 389 struct tseg_qent *p = NULL; 390 struct tseg_qent *te; 391 struct socket *so = tp->t_inpcb->inp_socket; 392 int flags; 393 394 /* 395 * Call with th == NULL after become established to 396 * force pre-ESTABLISHED data up to user socket. 397 */ 398 if (th == NULL) 399 goto present; 400 401 /* 402 * Limit the number of segments in the reassembly queue to prevent 403 * holding on to too many segments (and thus running out of mbufs). 404 * Make sure to let the missing segment through which caused this 405 * queue. Always keep one global queue entry spare to be able to 406 * process the missing segment. 407 */ 408 if (th->th_seq != tp->rcv_nxt && 409 tcp_reass_qsize + 1 >= tcp_reass_maxseg) { 410 tcp_reass_overflows++; 411 tcpstat.tcps_rcvmemdrop++; 412 m_freem(m); 413 /* no SACK block to report */ 414 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 415 return (0); 416 } 417 418 /* Allocate a new queue entry. */ 419 te = kmalloc(sizeof(struct tseg_qent), M_TSEGQ, M_INTWAIT | M_NULLOK); 420 if (te == NULL) { 421 tcpstat.tcps_rcvmemdrop++; 422 m_freem(m); 423 /* no SACK block to report */ 424 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 425 return (0); 426 } 427 atomic_add_int(&tcp_reass_qsize, 1); 428 429 if (th->th_flags & TH_FIN) 430 tp->t_flags |= TF_QUEDFIN; 431 432 /* 433 * Find a segment which begins after this one does. 434 */ 435 TAILQ_FOREACH(q, &tp->t_segq, tqe_q) { 436 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 437 break; 438 p = q; 439 } 440 441 /* 442 * If there is a preceding segment, it may provide some of 443 * our data already. If so, drop the data from the incoming 444 * segment. If it provides all of our data, drop us. 445 */ 446 if (p != NULL) { 447 tcp_seq_diff_t i; 448 449 /* conversion to int (in i) handles seq wraparound */ 450 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 451 if (i > 0) { /* overlaps preceding segment */ 452 tp->sack_flags |= 453 (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG); 454 /* enclosing block starts w/ preceding segment */ 455 tp->encloseblk.rblk_start = p->tqe_th->th_seq; 456 if (i >= *tlenp) { 457 if (th->th_flags & TH_FIN) 458 p->tqe_th->th_flags |= TH_FIN; 459 460 /* preceding encloses incoming segment */ 461 tp->encloseblk.rblk_end = TCP_SACK_BLKEND( 462 p->tqe_th->th_seq + p->tqe_len, 463 p->tqe_th->th_flags); 464 tcpstat.tcps_rcvduppack++; 465 tcpstat.tcps_rcvdupbyte += *tlenp; 466 m_freem(m); 467 kfree(te, M_TSEGQ); 468 atomic_add_int(&tcp_reass_qsize, -1); 469 /* 470 * Try to present any queued data 471 * at the left window edge to the user. 472 * This is needed after the 3-WHS 473 * completes. 474 */ 475 goto present; /* ??? */ 476 } 477 m_adj(m, i); 478 *tlenp -= i; 479 th->th_seq += i; 480 /* incoming segment end is enclosing block end */ 481 tp->encloseblk.rblk_end = TCP_SACK_BLKEND( 482 th->th_seq + *tlenp, th->th_flags); 483 /* trim end of reported D-SACK block */ 484 tp->reportblk.rblk_end = th->th_seq; 485 } 486 } 487 tcpstat.tcps_rcvoopack++; 488 tcpstat.tcps_rcvoobyte += *tlenp; 489 490 /* 491 * While we overlap succeeding segments trim them or, 492 * if they are completely covered, dequeue them. 493 */ 494 while (q) { 495 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 496 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len; 497 tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags); 498 struct tseg_qent *nq; 499 500 if (i <= 0) 501 break; 502 if (!(tp->sack_flags & TSACK_F_DUPSEG)) { 503 /* first time through */ 504 tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG); 505 tp->encloseblk = tp->reportblk; 506 /* report trailing duplicate D-SACK segment */ 507 tp->reportblk.rblk_start = q->tqe_th->th_seq; 508 } 509 if ((tp->sack_flags & TSACK_F_ENCLOSESEG) && 510 SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) { 511 /* extend enclosing block if one exists */ 512 tp->encloseblk.rblk_end = qend_sack; 513 } 514 if (i < q->tqe_len) { 515 q->tqe_th->th_seq += i; 516 q->tqe_len -= i; 517 m_adj(q->tqe_m, i); 518 break; 519 } 520 521 if (q->tqe_th->th_flags & TH_FIN) 522 th->th_flags |= TH_FIN; 523 524 nq = TAILQ_NEXT(q, tqe_q); 525 TAILQ_REMOVE(&tp->t_segq, q, tqe_q); 526 m_freem(q->tqe_m); 527 kfree(q, M_TSEGQ); 528 atomic_add_int(&tcp_reass_qsize, -1); 529 q = nq; 530 } 531 532 /* Insert the new segment queue entry into place. */ 533 te->tqe_m = m; 534 te->tqe_th = th; 535 te->tqe_len = *tlenp; 536 537 /* check if can coalesce with following segment */ 538 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) { 539 tcp_seq tend_sack; 540 541 te->tqe_len += q->tqe_len; 542 if (q->tqe_th->th_flags & TH_FIN) 543 te->tqe_th->th_flags |= TH_FIN; 544 tend_sack = TCP_SACK_BLKEND(te->tqe_th->th_seq + te->tqe_len, 545 te->tqe_th->th_flags); 546 547 m_cat(te->tqe_m, q->tqe_m); 548 tp->encloseblk.rblk_end = tend_sack; 549 /* 550 * When not reporting a duplicate segment, use 551 * the larger enclosing block as the SACK block. 552 */ 553 if (!(tp->sack_flags & TSACK_F_DUPSEG)) 554 tp->reportblk.rblk_end = tend_sack; 555 TAILQ_REMOVE(&tp->t_segq, q, tqe_q); 556 kfree(q, M_TSEGQ); 557 atomic_add_int(&tcp_reass_qsize, -1); 558 } 559 560 if (p == NULL) { 561 TAILQ_INSERT_HEAD(&tp->t_segq, te, tqe_q); 562 } else { 563 /* check if can coalesce with preceding segment */ 564 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) { 565 if (te->tqe_th->th_flags & TH_FIN) 566 p->tqe_th->th_flags |= TH_FIN; 567 p->tqe_len += te->tqe_len; 568 m_cat(p->tqe_m, te->tqe_m); 569 tp->encloseblk.rblk_start = p->tqe_th->th_seq; 570 /* 571 * When not reporting a duplicate segment, use 572 * the larger enclosing block as the SACK block. 573 */ 574 if (!(tp->sack_flags & TSACK_F_DUPSEG)) 575 tp->reportblk.rblk_start = p->tqe_th->th_seq; 576 kfree(te, M_TSEGQ); 577 atomic_add_int(&tcp_reass_qsize, -1); 578 } else { 579 TAILQ_INSERT_AFTER(&tp->t_segq, p, te, tqe_q); 580 } 581 } 582 583 present: 584 /* 585 * Present data to user, advancing rcv_nxt through 586 * completed sequence space. 587 */ 588 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 589 return (0); 590 q = TAILQ_FIRST(&tp->t_segq); 591 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt) 592 return (0); 593 tp->rcv_nxt += q->tqe_len; 594 if (!(tp->sack_flags & TSACK_F_DUPSEG)) { 595 /* no SACK block to report since ACK advanced */ 596 tp->reportblk.rblk_start = tp->reportblk.rblk_end; 597 } 598 /* no enclosing block to report since ACK advanced */ 599 tp->sack_flags &= ~TSACK_F_ENCLOSESEG; 600 flags = q->tqe_th->th_flags & TH_FIN; 601 TAILQ_REMOVE(&tp->t_segq, q, tqe_q); 602 KASSERT(TAILQ_EMPTY(&tp->t_segq) || 603 TAILQ_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt, 604 ("segment not coalesced")); 605 if (so->so_state & SS_CANTRCVMORE) { 606 m_freem(q->tqe_m); 607 } else { 608 lwkt_gettoken(&so->so_rcv.ssb_token); 609 ssb_appendstream(&so->so_rcv, q->tqe_m); 610 lwkt_reltoken(&so->so_rcv.ssb_token); 611 } 612 kfree(q, M_TSEGQ); 613 atomic_add_int(&tcp_reass_qsize, -1); 614 ND6_HINT(tp); 615 sorwakeup(so); 616 return (flags); 617 } 618 619 /* 620 * TCP input routine, follows pages 65-76 of the 621 * protocol specification dated September, 1981 very closely. 622 */ 623 #ifdef INET6 624 int 625 tcp6_input(struct mbuf **mp, int *offp, int proto) 626 { 627 struct mbuf *m = *mp; 628 struct in6_ifaddr *ia6; 629 630 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 631 632 /* 633 * draft-itojun-ipv6-tcp-to-anycast 634 * better place to put this in? 635 */ 636 ia6 = ip6_getdstifaddr(m); 637 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 638 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 639 offsetof(struct ip6_hdr, ip6_dst)); 640 return (IPPROTO_DONE); 641 } 642 643 tcp_input(mp, offp, proto); 644 return (IPPROTO_DONE); 645 } 646 #endif 647 648 int 649 tcp_input(struct mbuf **mp, int *offp, int proto) 650 { 651 int off0; 652 struct tcphdr *th; 653 struct ip *ip = NULL; 654 struct ipovly *ipov; 655 struct inpcb *inp = NULL; 656 u_char *optp = NULL; 657 int optlen = 0; 658 int tlen, off; 659 int len = 0; 660 int drop_hdrlen; 661 struct tcpcb *tp = NULL; 662 int thflags; 663 struct socket *so = NULL; 664 int todrop, acked; 665 boolean_t ourfinisacked, needoutput = FALSE, delayed_dupack = FALSE; 666 tcp_seq th_dupack = 0; /* XXX gcc warning */ 667 u_int to_flags = 0; /* XXX gcc warning */ 668 u_long tiwin; 669 int recvwin; 670 struct tcpopt to; /* options in this segment */ 671 struct sockaddr_in *next_hop = NULL; 672 int rstreason; /* For badport_bandlim accounting purposes */ 673 int cpu; 674 struct ip6_hdr *ip6 = NULL; 675 struct mbuf *m; 676 #ifdef INET6 677 boolean_t isipv6; 678 #else 679 const boolean_t isipv6 = FALSE; 680 #endif 681 #ifdef TCPDEBUG 682 short ostate = 0; 683 #endif 684 685 off0 = *offp; 686 m = *mp; 687 *mp = NULL; 688 689 tcpstat.tcps_rcvtotal++; 690 691 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) { 692 struct m_tag *mtag; 693 694 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 695 KKASSERT(mtag != NULL); 696 next_hop = m_tag_data(mtag); 697 } 698 699 #ifdef INET6 700 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE; 701 #endif 702 703 if (isipv6) { 704 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 705 ip6 = mtod(m, struct ip6_hdr *); 706 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0; 707 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 708 tcpstat.tcps_rcvbadsum++; 709 goto drop; 710 } 711 th = (struct tcphdr *)((caddr_t)ip6 + off0); 712 713 /* 714 * Be proactive about unspecified IPv6 address in source. 715 * As we use all-zero to indicate unbounded/unconnected pcb, 716 * unspecified IPv6 address can be used to confuse us. 717 * 718 * Note that packets with unspecified IPv6 destination is 719 * already dropped in ip6_input. 720 */ 721 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 722 /* XXX stat */ 723 goto drop; 724 } 725 } else { 726 /* 727 * Get IP and TCP header together in first mbuf. 728 * Note: IP leaves IP header in first mbuf. 729 */ 730 if (off0 > sizeof(struct ip)) { 731 ip_stripoptions(m); 732 off0 = sizeof(struct ip); 733 } 734 /* already checked and pulled up in ip_demux() */ 735 KASSERT(m->m_len >= sizeof(struct tcpiphdr), 736 ("TCP header not in one mbuf: m->m_len %d", m->m_len)); 737 ip = mtod(m, struct ip *); 738 ipov = (struct ipovly *)ip; 739 th = (struct tcphdr *)((caddr_t)ip + off0); 740 tlen = ip->ip_len; 741 742 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 743 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 744 th->th_sum = m->m_pkthdr.csum_data; 745 else 746 th->th_sum = in_pseudo(ip->ip_src.s_addr, 747 ip->ip_dst.s_addr, 748 htonl(m->m_pkthdr.csum_data + 749 ip->ip_len + 750 IPPROTO_TCP)); 751 th->th_sum ^= 0xffff; 752 } else { 753 /* 754 * Checksum extended TCP header and data. 755 */ 756 len = sizeof(struct ip) + tlen; 757 bzero(ipov->ih_x1, sizeof ipov->ih_x1); 758 ipov->ih_len = (u_short)tlen; 759 ipov->ih_len = htons(ipov->ih_len); 760 th->th_sum = in_cksum(m, len); 761 } 762 if (th->th_sum) { 763 tcpstat.tcps_rcvbadsum++; 764 goto drop; 765 } 766 #ifdef INET6 767 /* Re-initialization for later version check */ 768 ip->ip_v = IPVERSION; 769 #endif 770 } 771 772 /* 773 * Check that TCP offset makes sense, 774 * pull out TCP options and adjust length. XXX 775 */ 776 off = th->th_off << 2; 777 /* already checked and pulled up in ip_demux() */ 778 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen, 779 ("bad TCP data offset %d (tlen %d)", off, tlen)); 780 tlen -= off; /* tlen is used instead of ti->ti_len */ 781 if (off > sizeof(struct tcphdr)) { 782 if (isipv6) { 783 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 784 ip6 = mtod(m, struct ip6_hdr *); 785 th = (struct tcphdr *)((caddr_t)ip6 + off0); 786 } else { 787 /* already pulled up in ip_demux() */ 788 KASSERT(m->m_len >= sizeof(struct ip) + off, 789 ("TCP header and options not in one mbuf: " 790 "m_len %d, off %d", m->m_len, off)); 791 } 792 optlen = off - sizeof(struct tcphdr); 793 optp = (u_char *)(th + 1); 794 } 795 thflags = th->th_flags; 796 797 #ifdef TCP_DROP_SYNFIN 798 /* 799 * If the drop_synfin option is enabled, drop all packets with 800 * both the SYN and FIN bits set. This prevents e.g. nmap from 801 * identifying the TCP/IP stack. 802 * 803 * This is a violation of the TCP specification. 804 */ 805 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN)) 806 goto drop; 807 #endif 808 809 /* 810 * Convert TCP protocol specific fields to host format. 811 */ 812 th->th_seq = ntohl(th->th_seq); 813 th->th_ack = ntohl(th->th_ack); 814 th->th_win = ntohs(th->th_win); 815 th->th_urp = ntohs(th->th_urp); 816 817 /* 818 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 819 * until after ip6_savecontrol() is called and before other functions 820 * which don't want those proto headers. 821 * Because ip6_savecontrol() is going to parse the mbuf to 822 * search for data to be passed up to user-land, it wants mbuf 823 * parameters to be unchanged. 824 * XXX: the call of ip6_savecontrol() has been obsoleted based on 825 * latest version of the advanced API (20020110). 826 */ 827 drop_hdrlen = off0 + off; 828 829 /* 830 * Locate pcb for segment. 831 */ 832 findpcb: 833 /* IPFIREWALL_FORWARD section */ 834 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */ 835 /* 836 * Transparently forwarded. Pretend to be the destination. 837 * already got one like this? 838 */ 839 cpu = mycpu->gd_cpuid; 840 inp = in_pcblookup_hash(&tcbinfo[cpu], 841 ip->ip_src, th->th_sport, 842 ip->ip_dst, th->th_dport, 843 0, m->m_pkthdr.rcvif); 844 if (!inp) { 845 /* 846 * It's new. Try to find the ambushing socket. 847 */ 848 849 /* 850 * The rest of the ipfw code stores the port in 851 * host order. XXX 852 * (The IP address is still in network order.) 853 */ 854 in_port_t dport = next_hop->sin_port ? 855 htons(next_hop->sin_port) : 856 th->th_dport; 857 858 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport, 859 next_hop->sin_addr.s_addr, dport); 860 inp = in_pcblookup_hash(&tcbinfo[cpu], 861 ip->ip_src, th->th_sport, 862 next_hop->sin_addr, dport, 863 1, m->m_pkthdr.rcvif); 864 } 865 } else { 866 if (isipv6) { 867 inp = in6_pcblookup_hash(&tcbinfo[0], 868 &ip6->ip6_src, th->th_sport, 869 &ip6->ip6_dst, th->th_dport, 870 1, m->m_pkthdr.rcvif); 871 } else { 872 cpu = mycpu->gd_cpuid; 873 inp = in_pcblookup_pkthash(&tcbinfo[cpu], 874 ip->ip_src, th->th_sport, 875 ip->ip_dst, th->th_dport, 876 1, m->m_pkthdr.rcvif, 877 tcp_reuseport_ext ? m : NULL); 878 } 879 } 880 881 /* 882 * If the state is CLOSED (i.e., TCB does not exist) then 883 * all data in the incoming segment is discarded. 884 * If the TCB exists but is in CLOSED state, it is embryonic, 885 * but should either do a listen or a connect soon. 886 */ 887 if (inp == NULL) { 888 if (log_in_vain) { 889 #ifdef INET6 890 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 891 #else 892 char dbuf[INET_ADDRSTRLEN], sbuf[INET_ADDRSTRLEN]; 893 #endif 894 if (isipv6) { 895 strcpy(dbuf, "["); 896 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 897 strcat(dbuf, "]"); 898 strcpy(sbuf, "["); 899 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 900 strcat(sbuf, "]"); 901 } else { 902 kinet_ntoa(ip->ip_dst, dbuf); 903 kinet_ntoa(ip->ip_src, sbuf); 904 } 905 switch (log_in_vain) { 906 case 1: 907 if (!(thflags & TH_SYN)) 908 break; 909 case 2: 910 log(LOG_INFO, 911 "Connection attempt to TCP %s:%d " 912 "from %s:%d flags:0x%02x\n", 913 dbuf, ntohs(th->th_dport), sbuf, 914 ntohs(th->th_sport), thflags); 915 break; 916 default: 917 break; 918 } 919 } 920 if (blackhole) { 921 switch (blackhole) { 922 case 1: 923 if (thflags & TH_SYN) 924 goto drop; 925 break; 926 case 2: 927 goto drop; 928 default: 929 goto drop; 930 } 931 } 932 rstreason = BANDLIM_RST_CLOSEDPORT; 933 goto dropwithreset; 934 } 935 936 #ifdef IPSEC 937 if (isipv6) { 938 if (ipsec6_in_reject_so(m, inp->inp_socket)) { 939 ipsec6stat.in_polvio++; 940 goto drop; 941 } 942 } else { 943 if (ipsec4_in_reject_so(m, inp->inp_socket)) { 944 ipsecstat.in_polvio++; 945 goto drop; 946 } 947 } 948 #endif 949 #ifdef FAST_IPSEC 950 if (isipv6) { 951 if (ipsec6_in_reject(m, inp)) 952 goto drop; 953 } else { 954 if (ipsec4_in_reject(m, inp)) 955 goto drop; 956 } 957 #endif 958 /* Check the minimum TTL for socket. */ 959 #ifdef INET6 960 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl) 961 goto drop; 962 #endif 963 964 tp = intotcpcb(inp); 965 KASSERT(tp != NULL, ("tcp_input: tp is NULL")); 966 if (tp->t_state <= TCPS_CLOSED) 967 goto drop; 968 969 so = inp->inp_socket; 970 971 #ifdef TCPDEBUG 972 if (so->so_options & SO_DEBUG) { 973 ostate = tp->t_state; 974 if (isipv6) 975 bcopy(ip6, tcp_saveipgen, sizeof(*ip6)); 976 else 977 bcopy(ip, tcp_saveipgen, sizeof(*ip)); 978 tcp_savetcp = *th; 979 } 980 #endif 981 982 bzero(&to, sizeof to); 983 984 if (so->so_options & SO_ACCEPTCONN) { 985 struct in_conninfo inc; 986 987 #ifdef INET6 988 inc.inc_isipv6 = (isipv6 == TRUE); 989 #endif 990 if (isipv6) { 991 inc.inc6_faddr = ip6->ip6_src; 992 inc.inc6_laddr = ip6->ip6_dst; 993 inc.inc6_route.ro_rt = NULL; /* XXX */ 994 } else { 995 inc.inc_faddr = ip->ip_src; 996 inc.inc_laddr = ip->ip_dst; 997 inc.inc_route.ro_rt = NULL; /* XXX */ 998 } 999 inc.inc_fport = th->th_sport; 1000 inc.inc_lport = th->th_dport; 1001 1002 /* 1003 * If the state is LISTEN then ignore segment if it contains 1004 * a RST. If the segment contains an ACK then it is bad and 1005 * send a RST. If it does not contain a SYN then it is not 1006 * interesting; drop it. 1007 * 1008 * If the state is SYN_RECEIVED (syncache) and seg contains 1009 * an ACK, but not for our SYN/ACK, send a RST. If the seg 1010 * contains a RST, check the sequence number to see if it 1011 * is a valid reset segment. 1012 */ 1013 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) { 1014 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) { 1015 if (!syncache_expand(&inc, th, &so, m)) { 1016 /* 1017 * No syncache entry, or ACK was not 1018 * for our SYN/ACK. Send a RST. 1019 */ 1020 tcpstat.tcps_badsyn++; 1021 rstreason = BANDLIM_RST_OPENPORT; 1022 goto dropwithreset; 1023 } 1024 1025 /* 1026 * Could not complete 3-way handshake, 1027 * connection is being closed down, and 1028 * syncache will free mbuf. 1029 */ 1030 if (so == NULL) 1031 return(IPPROTO_DONE); 1032 1033 /* 1034 * We must be in the correct protocol thread 1035 * for this connection. 1036 */ 1037 KKASSERT(so->so_port == &curthread->td_msgport); 1038 1039 /* 1040 * Socket is created in state SYN_RECEIVED. 1041 * Continue processing segment. 1042 */ 1043 inp = so->so_pcb; 1044 tp = intotcpcb(inp); 1045 /* 1046 * This is what would have happened in 1047 * tcp_output() when the SYN,ACK was sent. 1048 */ 1049 tp->snd_up = tp->snd_una; 1050 tp->snd_max = tp->snd_nxt = tp->iss + 1; 1051 tp->last_ack_sent = tp->rcv_nxt; 1052 1053 goto after_listen; 1054 } 1055 if (thflags & TH_RST) { 1056 syncache_chkrst(&inc, th); 1057 goto drop; 1058 } 1059 if (thflags & TH_ACK) { 1060 syncache_badack(&inc); 1061 tcpstat.tcps_badsyn++; 1062 rstreason = BANDLIM_RST_OPENPORT; 1063 goto dropwithreset; 1064 } 1065 goto drop; 1066 } 1067 1068 /* 1069 * Segment's flags are (SYN) or (SYN | FIN). 1070 */ 1071 #ifdef INET6 1072 /* 1073 * If deprecated address is forbidden, 1074 * we do not accept SYN to deprecated interface 1075 * address to prevent any new inbound connection from 1076 * getting established. 1077 * When we do not accept SYN, we send a TCP RST, 1078 * with deprecated source address (instead of dropping 1079 * it). We compromise it as it is much better for peer 1080 * to send a RST, and RST will be the final packet 1081 * for the exchange. 1082 * 1083 * If we do not forbid deprecated addresses, we accept 1084 * the SYN packet. RFC2462 does not suggest dropping 1085 * SYN in this case. 1086 * If we decipher RFC2462 5.5.4, it says like this: 1087 * 1. use of deprecated addr with existing 1088 * communication is okay - "SHOULD continue to be 1089 * used" 1090 * 2. use of it with new communication: 1091 * (2a) "SHOULD NOT be used if alternate address 1092 * with sufficient scope is available" 1093 * (2b) nothing mentioned otherwise. 1094 * Here we fall into (2b) case as we have no choice in 1095 * our source address selection - we must obey the peer. 1096 * 1097 * The wording in RFC2462 is confusing, and there are 1098 * multiple description text for deprecated address 1099 * handling - worse, they are not exactly the same. 1100 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1101 */ 1102 if (isipv6 && !ip6_use_deprecated) { 1103 struct in6_ifaddr *ia6; 1104 1105 if ((ia6 = ip6_getdstifaddr(m)) && 1106 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1107 tp = NULL; 1108 rstreason = BANDLIM_RST_OPENPORT; 1109 goto dropwithreset; 1110 } 1111 } 1112 #endif 1113 /* 1114 * If it is from this socket, drop it, it must be forged. 1115 * Don't bother responding if the destination was a broadcast. 1116 */ 1117 if (th->th_dport == th->th_sport) { 1118 if (isipv6) { 1119 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 1120 &ip6->ip6_src)) 1121 goto drop; 1122 } else { 1123 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 1124 goto drop; 1125 } 1126 } 1127 /* 1128 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1129 * 1130 * Note that it is quite possible to receive unicast 1131 * link-layer packets with a broadcast IP address. Use 1132 * in_broadcast() to find them. 1133 */ 1134 if (m->m_flags & (M_BCAST | M_MCAST)) 1135 goto drop; 1136 if (isipv6) { 1137 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1138 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 1139 goto drop; 1140 } else { 1141 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1142 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1143 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1144 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 1145 goto drop; 1146 } 1147 /* 1148 * SYN appears to be valid; create compressed TCP state 1149 * for syncache. 1150 */ 1151 if (so->so_qlen <= so->so_qlimit) { 1152 tcp_dooptions(&to, optp, optlen, TRUE, th->th_ack); 1153 if (!syncache_add(&inc, &to, th, so, m)) 1154 goto drop; 1155 1156 /* 1157 * Entry added to syncache, mbuf used to 1158 * send SYN,ACK packet. 1159 */ 1160 return(IPPROTO_DONE); 1161 } 1162 goto drop; 1163 } 1164 1165 after_listen: 1166 /* 1167 * Should not happen - syncache should pick up these connections. 1168 * 1169 * Once we are past handling listen sockets we must be in the 1170 * correct protocol processing thread. 1171 */ 1172 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state")); 1173 KKASSERT(so->so_port == &curthread->td_msgport); 1174 1175 /* Unscale the window into a 32-bit value. */ 1176 if (!(thflags & TH_SYN)) 1177 tiwin = th->th_win << tp->snd_scale; 1178 else 1179 tiwin = th->th_win; 1180 1181 /* 1182 * This is the second part of the MSS DoS prevention code (after 1183 * minmss on the sending side) and it deals with too many too small 1184 * tcp packets in a too short timeframe (1 second). 1185 * 1186 * XXX Removed. This code was crap. It does not scale to network 1187 * speed, and default values break NFS. Gone. 1188 */ 1189 /* REMOVED */ 1190 1191 /* 1192 * Segment received on connection. 1193 * 1194 * Reset idle time and keep-alive timer. Don't waste time if less 1195 * then a second has elapsed. 1196 */ 1197 if ((int)(ticks - tp->t_rcvtime) > hz) 1198 tcp_timer_keep_activity(tp, thflags); 1199 1200 /* 1201 * Process options. 1202 * XXX this is tradtitional behavior, may need to be cleaned up. 1203 */ 1204 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0, th->th_ack); 1205 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1206 if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) { 1207 tp->t_flags |= TF_RCVD_SCALE; 1208 tp->snd_scale = to.to_requested_s_scale; 1209 } 1210 1211 /* 1212 * Initial send window; will be updated upon next ACK 1213 */ 1214 tp->snd_wnd = th->th_win; 1215 1216 if (to.to_flags & TOF_TS) { 1217 tp->t_flags |= TF_RCVD_TSTMP; 1218 tp->ts_recent = to.to_tsval; 1219 tp->ts_recent_age = ticks; 1220 } 1221 if (!(to.to_flags & TOF_MSS)) 1222 to.to_mss = 0; 1223 tcp_rmx_init(tp, to.to_mss); 1224 /* 1225 * Only set the TF_SACK_PERMITTED per-connection flag 1226 * if we got a SACK_PERMITTED option from the other side 1227 * and the global tcp_do_sack variable is true. 1228 */ 1229 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED)) 1230 tp->t_flags |= TF_SACK_PERMITTED; 1231 } 1232 1233 /* 1234 * Header prediction: check for the two common cases 1235 * of a uni-directional data xfer. If the packet has 1236 * no control flags, is in-sequence, the window didn't 1237 * change and we're not retransmitting, it's a 1238 * candidate. If the length is zero and the ack moved 1239 * forward, we're the sender side of the xfer. Just 1240 * free the data acked & wake any higher level process 1241 * that was blocked waiting for space. If the length 1242 * is non-zero and the ack didn't move, we're the 1243 * receiver side. If we're getting packets in-order 1244 * (the reassembly queue is empty), add the data to 1245 * the socket buffer and note that we need a delayed ack. 1246 * Make sure that the hidden state-flags are also off. 1247 * Since we check for TCPS_ESTABLISHED above, it can only 1248 * be TH_NEEDSYN. 1249 */ 1250 if (tp->t_state == TCPS_ESTABLISHED && 1251 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1252 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) && 1253 (!(to.to_flags & TOF_TS) || 1254 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1255 th->th_seq == tp->rcv_nxt && 1256 tp->snd_nxt == tp->snd_max) { 1257 1258 /* 1259 * If last ACK falls within this segment's sequence numbers, 1260 * record the timestamp. 1261 * NOTE that the test is modified according to the latest 1262 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1263 */ 1264 if ((to.to_flags & TOF_TS) && 1265 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1266 tp->ts_recent_age = ticks; 1267 tp->ts_recent = to.to_tsval; 1268 } 1269 1270 if (tlen == 0) { 1271 if (SEQ_GT(th->th_ack, tp->snd_una) && 1272 SEQ_LEQ(th->th_ack, tp->snd_max) && 1273 tp->snd_cwnd >= tp->snd_wnd && 1274 !IN_FASTRECOVERY(tp)) { 1275 /* 1276 * This is a pure ack for outstanding data. 1277 */ 1278 ++tcpstat.tcps_predack; 1279 /* 1280 * "bad retransmit" recovery 1281 * 1282 * If Eifel detection applies, then 1283 * it is deterministic, so use it 1284 * unconditionally over the old heuristic. 1285 * Otherwise, fall back to the old heuristic. 1286 */ 1287 if (tcp_do_eifel_detect && 1288 (to.to_flags & TOF_TS) && to.to_tsecr && 1289 (tp->rxt_flags & TRXT_F_FIRSTACCACK)) { 1290 /* Eifel detection applicable. */ 1291 if (to.to_tsecr < tp->t_rexmtTS) { 1292 tcp_revert_congestion_state(tp); 1293 ++tcpstat.tcps_eifeldetected; 1294 if (tp->t_rxtshift != 1 || 1295 ticks >= tp->t_badrxtwin) 1296 ++tcpstat.tcps_rttcantdetect; 1297 } 1298 } else if (tp->t_rxtshift == 1 && 1299 ticks < tp->t_badrxtwin) { 1300 tcp_revert_congestion_state(tp); 1301 ++tcpstat.tcps_rttdetected; 1302 } 1303 tp->rxt_flags &= ~(TRXT_F_FIRSTACCACK | 1304 TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT); 1305 /* 1306 * Recalculate the retransmit timer / rtt. 1307 * 1308 * Some machines (certain windows boxes) 1309 * send broken timestamp replies during the 1310 * SYN+ACK phase, ignore timestamps of 0. 1311 */ 1312 if ((to.to_flags & TOF_TS) && to.to_tsecr) { 1313 tcp_xmit_timer(tp, 1314 ticks - to.to_tsecr + 1, 1315 th->th_ack); 1316 } else if (tp->t_rtttime && 1317 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1318 tcp_xmit_timer(tp, 1319 ticks - tp->t_rtttime + 1, 1320 th->th_ack); 1321 } 1322 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1323 acked = th->th_ack - tp->snd_una; 1324 tcpstat.tcps_rcvackpack++; 1325 tcpstat.tcps_rcvackbyte += acked; 1326 sbdrop(&so->so_snd.sb, acked); 1327 tp->snd_recover = th->th_ack - 1; 1328 tp->snd_una = th->th_ack; 1329 tp->t_dupacks = 0; 1330 /* 1331 * Update window information. 1332 */ 1333 if (tiwin != tp->snd_wnd && 1334 acceptable_window_update(tp, th, tiwin)) { 1335 /* keep track of pure window updates */ 1336 if (tp->snd_wl2 == th->th_ack && 1337 tiwin > tp->snd_wnd) 1338 tcpstat.tcps_rcvwinupd++; 1339 tp->snd_wnd = tiwin; 1340 tp->snd_wl1 = th->th_seq; 1341 tp->snd_wl2 = th->th_ack; 1342 if (tp->snd_wnd > tp->max_sndwnd) 1343 tp->max_sndwnd = tp->snd_wnd; 1344 } 1345 m_freem(m); 1346 ND6_HINT(tp); /* some progress has been done */ 1347 /* 1348 * If all outstanding data are acked, stop 1349 * retransmit timer, otherwise restart timer 1350 * using current (possibly backed-off) value. 1351 * If process is waiting for space, 1352 * wakeup/selwakeup/signal. If data 1353 * are ready to send, let tcp_output 1354 * decide between more output or persist. 1355 */ 1356 if (tp->snd_una == tp->snd_max) { 1357 tcp_callout_stop(tp, tp->tt_rexmt); 1358 } else if (!tcp_callout_active(tp, 1359 tp->tt_persist)) { 1360 tcp_callout_reset(tp, tp->tt_rexmt, 1361 tp->t_rxtcur, tcp_timer_rexmt); 1362 } 1363 sowwakeup(so); 1364 if (so->so_snd.ssb_cc > 0 && 1365 !tcp_output_pending(tp)) 1366 tcp_output_fair(tp); 1367 return(IPPROTO_DONE); 1368 } 1369 } else if (tiwin == tp->snd_wnd && 1370 th->th_ack == tp->snd_una && 1371 TAILQ_EMPTY(&tp->t_segq) && 1372 tlen <= ssb_space(&so->so_rcv)) { 1373 u_long newsize = 0; /* automatic sockbuf scaling */ 1374 /* 1375 * This is a pure, in-sequence data packet 1376 * with nothing on the reassembly queue and 1377 * we have enough buffer space to take it. 1378 */ 1379 ++tcpstat.tcps_preddat; 1380 tp->rcv_nxt += tlen; 1381 tcpstat.tcps_rcvpack++; 1382 tcpstat.tcps_rcvbyte += tlen; 1383 ND6_HINT(tp); /* some progress has been done */ 1384 /* 1385 * Automatic sizing of receive socket buffer. Often the send 1386 * buffer size is not optimally adjusted to the actual network 1387 * conditions at hand (delay bandwidth product). Setting the 1388 * buffer size too small limits throughput on links with high 1389 * bandwidth and high delay (eg. trans-continental/oceanic links). 1390 * 1391 * On the receive side the socket buffer memory is only rarely 1392 * used to any significant extent. This allows us to be much 1393 * more aggressive in scaling the receive socket buffer. For 1394 * the case that the buffer space is actually used to a large 1395 * extent and we run out of kernel memory we can simply drop 1396 * the new segments; TCP on the sender will just retransmit it 1397 * later. Setting the buffer size too big may only consume too 1398 * much kernel memory if the application doesn't read() from 1399 * the socket or packet loss or reordering makes use of the 1400 * reassembly queue. 1401 * 1402 * The criteria to step up the receive buffer one notch are: 1403 * 1. the number of bytes received during the time it takes 1404 * one timestamp to be reflected back to us (the RTT); 1405 * 2. received bytes per RTT is within seven eighth of the 1406 * current socket buffer size; 1407 * 3. receive buffer size has not hit maximal automatic size; 1408 * 1409 * This algorithm does one step per RTT at most and only if 1410 * we receive a bulk stream w/o packet losses or reorderings. 1411 * Shrinking the buffer during idle times is not necessary as 1412 * it doesn't consume any memory when idle. 1413 * 1414 * TODO: Only step up if the application is actually serving 1415 * the buffer to better manage the socket buffer resources. 1416 */ 1417 if (tcp_do_autorcvbuf && 1418 to.to_tsecr && 1419 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) { 1420 if (to.to_tsecr > tp->rfbuf_ts && 1421 to.to_tsecr - tp->rfbuf_ts < hz) { 1422 if (tp->rfbuf_cnt > 1423 (so->so_rcv.ssb_hiwat / 8 * 7) && 1424 so->so_rcv.ssb_hiwat < 1425 tcp_autorcvbuf_max) { 1426 newsize = 1427 ulmin(so->so_rcv.ssb_hiwat + 1428 tcp_autorcvbuf_inc, 1429 tcp_autorcvbuf_max); 1430 } 1431 /* Start over with next RTT. */ 1432 tp->rfbuf_ts = 0; 1433 tp->rfbuf_cnt = 0; 1434 } else 1435 tp->rfbuf_cnt += tlen; /* add up */ 1436 } 1437 /* 1438 * Add data to socket buffer. 1439 */ 1440 if (so->so_state & SS_CANTRCVMORE) { 1441 m_freem(m); 1442 } else { 1443 /* 1444 * Set new socket buffer size, give up when 1445 * limit is reached. 1446 * 1447 * Adjusting the size can mess up ACK 1448 * sequencing when pure window updates are 1449 * being avoided (which is the default), 1450 * so force an ack. 1451 */ 1452 lwkt_gettoken(&so->so_rcv.ssb_token); 1453 if (newsize) { 1454 tp->t_flags |= TF_RXRESIZED; 1455 if (!ssb_reserve(&so->so_rcv, newsize, 1456 so, NULL)) { 1457 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE); 1458 } 1459 if (newsize >= 1460 (TCP_MAXWIN << tp->rcv_scale)) { 1461 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE); 1462 } 1463 } 1464 m_adj(m, drop_hdrlen); /* delayed header drop */ 1465 ssb_appendstream(&so->so_rcv, m); 1466 lwkt_reltoken(&so->so_rcv.ssb_token); 1467 } 1468 sorwakeup(so); 1469 /* 1470 * This code is responsible for most of the ACKs 1471 * the TCP stack sends back after receiving a data 1472 * packet. Note that the DELAY_ACK check fails if 1473 * the delack timer is already running, which results 1474 * in an ack being sent every other packet (which is 1475 * what we want). 1476 * 1477 * We then further aggregate acks by not actually 1478 * sending one until the protocol thread has completed 1479 * processing the current backlog of packets. This 1480 * does not delay the ack any further, but allows us 1481 * to take advantage of the packet aggregation that 1482 * high speed NICs do (usually blocks of 8-10 packets) 1483 * to send a single ack rather then four or five acks, 1484 * greatly reducing the ack rate, the return channel 1485 * bandwidth, and the protocol overhead on both ends. 1486 * 1487 * Since this also has the effect of slowing down 1488 * the exponential slow-start ramp-up, systems with 1489 * very large bandwidth-delay products might want 1490 * to turn the feature off. 1491 */ 1492 if (DELAY_ACK(tp)) { 1493 tcp_callout_reset(tp, tp->tt_delack, 1494 tcp_delacktime, tcp_timer_delack); 1495 } else if (tcp_aggregate_acks) { 1496 tp->t_flags |= TF_ACKNOW; 1497 if (!(tp->t_flags & TF_ONOUTPUTQ)) { 1498 tp->t_flags |= TF_ONOUTPUTQ; 1499 tp->tt_cpu = mycpu->gd_cpuid; 1500 TAILQ_INSERT_TAIL( 1501 &tcpcbackq[tp->tt_cpu].head, 1502 tp, t_outputq); 1503 } 1504 } else { 1505 tp->t_flags |= TF_ACKNOW; 1506 tcp_output(tp); 1507 } 1508 return(IPPROTO_DONE); 1509 } 1510 } 1511 1512 /* 1513 * Calculate amount of space in receive window, 1514 * and then do TCP input processing. 1515 * Receive window is amount of space in rcv queue, 1516 * but not less than advertised window. 1517 */ 1518 recvwin = ssb_space(&so->so_rcv); 1519 if (recvwin < 0) 1520 recvwin = 0; 1521 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt)); 1522 1523 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1524 tp->rfbuf_ts = 0; 1525 tp->rfbuf_cnt = 0; 1526 1527 switch (tp->t_state) { 1528 /* 1529 * If the state is SYN_RECEIVED: 1530 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1531 */ 1532 case TCPS_SYN_RECEIVED: 1533 if ((thflags & TH_ACK) && 1534 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1535 SEQ_GT(th->th_ack, tp->snd_max))) { 1536 rstreason = BANDLIM_RST_OPENPORT; 1537 goto dropwithreset; 1538 } 1539 break; 1540 1541 /* 1542 * If the state is SYN_SENT: 1543 * if seg contains an ACK, but not for our SYN, drop the input. 1544 * if seg contains a RST, then drop the connection. 1545 * if seg does not contain SYN, then drop it. 1546 * Otherwise this is an acceptable SYN segment 1547 * initialize tp->rcv_nxt and tp->irs 1548 * if seg contains ack then advance tp->snd_una 1549 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1550 * arrange for segment to be acked (eventually) 1551 * continue processing rest of data/controls, beginning with URG 1552 */ 1553 case TCPS_SYN_SENT: 1554 if ((thflags & TH_ACK) && 1555 (SEQ_LEQ(th->th_ack, tp->iss) || 1556 SEQ_GT(th->th_ack, tp->snd_max))) { 1557 rstreason = BANDLIM_UNLIMITED; 1558 goto dropwithreset; 1559 } 1560 if (thflags & TH_RST) { 1561 if (thflags & TH_ACK) 1562 tp = tcp_drop(tp, ECONNREFUSED); 1563 goto drop; 1564 } 1565 if (!(thflags & TH_SYN)) 1566 goto drop; 1567 1568 tp->irs = th->th_seq; 1569 tcp_rcvseqinit(tp); 1570 if (thflags & TH_ACK) { 1571 /* Our SYN was acked. */ 1572 tcpstat.tcps_connects++; 1573 soisconnected(so); 1574 /* Do window scaling on this connection? */ 1575 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 1576 (TF_RCVD_SCALE | TF_REQ_SCALE)) 1577 tp->rcv_scale = tp->request_r_scale; 1578 tp->rcv_adv += tp->rcv_wnd; 1579 tp->snd_una++; /* SYN is acked */ 1580 tcp_callout_stop(tp, tp->tt_rexmt); 1581 /* 1582 * If there's data, delay ACK; if there's also a FIN 1583 * ACKNOW will be turned on later. 1584 */ 1585 if (DELAY_ACK(tp) && tlen != 0) { 1586 tcp_callout_reset(tp, tp->tt_delack, 1587 tcp_delacktime, tcp_timer_delack); 1588 } else { 1589 tp->t_flags |= TF_ACKNOW; 1590 } 1591 /* 1592 * Received <SYN,ACK> in SYN_SENT[*] state. 1593 * Transitions: 1594 * SYN_SENT --> ESTABLISHED 1595 * SYN_SENT* --> FIN_WAIT_1 1596 */ 1597 tp->t_starttime = ticks; 1598 if (tp->t_flags & TF_NEEDFIN) { 1599 TCP_STATE_CHANGE(tp, TCPS_FIN_WAIT_1); 1600 tp->t_flags &= ~TF_NEEDFIN; 1601 thflags &= ~TH_SYN; 1602 } else { 1603 tcp_established(tp); 1604 } 1605 } else { 1606 /* 1607 * Received initial SYN in SYN-SENT[*] state => 1608 * simultaneous open. 1609 * Do 3-way handshake: 1610 * SYN-SENT -> SYN-RECEIVED 1611 * SYN-SENT* -> SYN-RECEIVED* 1612 */ 1613 tp->t_flags |= TF_ACKNOW; 1614 tcp_callout_stop(tp, tp->tt_rexmt); 1615 TCP_STATE_CHANGE(tp, TCPS_SYN_RECEIVED); 1616 } 1617 1618 /* 1619 * Advance th->th_seq to correspond to first data byte. 1620 * If data, trim to stay within window, 1621 * dropping FIN if necessary. 1622 */ 1623 th->th_seq++; 1624 if (tlen > tp->rcv_wnd) { 1625 todrop = tlen - tp->rcv_wnd; 1626 m_adj(m, -todrop); 1627 tlen = tp->rcv_wnd; 1628 thflags &= ~TH_FIN; 1629 tcpstat.tcps_rcvpackafterwin++; 1630 tcpstat.tcps_rcvbyteafterwin += todrop; 1631 } 1632 tp->snd_wl1 = th->th_seq - 1; 1633 tp->rcv_up = th->th_seq; 1634 /* 1635 * Client side of transaction: already sent SYN and data. 1636 * If the remote host used T/TCP to validate the SYN, 1637 * our data will be ACK'd; if so, enter normal data segment 1638 * processing in the middle of step 5, ack processing. 1639 * Otherwise, goto step 6. 1640 */ 1641 if (thflags & TH_ACK) 1642 goto process_ACK; 1643 1644 goto step6; 1645 1646 /* 1647 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1648 * do normal processing (we no longer bother with T/TCP). 1649 */ 1650 case TCPS_LAST_ACK: 1651 case TCPS_CLOSING: 1652 case TCPS_TIME_WAIT: 1653 break; /* continue normal processing */ 1654 } 1655 1656 /* 1657 * States other than LISTEN or SYN_SENT. 1658 * First check the RST flag and sequence number since reset segments 1659 * are exempt from the timestamp and connection count tests. This 1660 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1661 * below which allowed reset segments in half the sequence space 1662 * to fall though and be processed (which gives forged reset 1663 * segments with a random sequence number a 50 percent chance of 1664 * killing a connection). 1665 * Then check timestamp, if present. 1666 * Then check the connection count, if present. 1667 * Then check that at least some bytes of segment are within 1668 * receive window. If segment begins before rcv_nxt, 1669 * drop leading data (and SYN); if nothing left, just ack. 1670 * 1671 * 1672 * If the RST bit is set, check the sequence number to see 1673 * if this is a valid reset segment. 1674 * RFC 793 page 37: 1675 * In all states except SYN-SENT, all reset (RST) segments 1676 * are validated by checking their SEQ-fields. A reset is 1677 * valid if its sequence number is in the window. 1678 * Note: this does not take into account delayed ACKs, so 1679 * we should test against last_ack_sent instead of rcv_nxt. 1680 * The sequence number in the reset segment is normally an 1681 * echo of our outgoing acknowledgement numbers, but some hosts 1682 * send a reset with the sequence number at the rightmost edge 1683 * of our receive window, and we have to handle this case. 1684 * If we have multiple segments in flight, the intial reset 1685 * segment sequence numbers will be to the left of last_ack_sent, 1686 * but they will eventually catch up. 1687 * In any case, it never made sense to trim reset segments to 1688 * fit the receive window since RFC 1122 says: 1689 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1690 * 1691 * A TCP SHOULD allow a received RST segment to include data. 1692 * 1693 * DISCUSSION 1694 * It has been suggested that a RST segment could contain 1695 * ASCII text that encoded and explained the cause of the 1696 * RST. No standard has yet been established for such 1697 * data. 1698 * 1699 * If the reset segment passes the sequence number test examine 1700 * the state: 1701 * SYN_RECEIVED STATE: 1702 * If passive open, return to LISTEN state. 1703 * If active open, inform user that connection was refused. 1704 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1705 * Inform user that connection was reset, and close tcb. 1706 * CLOSING, LAST_ACK STATES: 1707 * Close the tcb. 1708 * TIME_WAIT STATE: 1709 * Drop the segment - see Stevens, vol. 2, p. 964 and 1710 * RFC 1337. 1711 */ 1712 if (thflags & TH_RST) { 1713 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1714 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1715 switch (tp->t_state) { 1716 1717 case TCPS_SYN_RECEIVED: 1718 so->so_error = ECONNREFUSED; 1719 goto close; 1720 1721 case TCPS_ESTABLISHED: 1722 case TCPS_FIN_WAIT_1: 1723 case TCPS_FIN_WAIT_2: 1724 case TCPS_CLOSE_WAIT: 1725 so->so_error = ECONNRESET; 1726 close: 1727 TCP_STATE_CHANGE(tp, TCPS_CLOSED); 1728 tcpstat.tcps_drops++; 1729 tp = tcp_close(tp); 1730 break; 1731 1732 case TCPS_CLOSING: 1733 case TCPS_LAST_ACK: 1734 tp = tcp_close(tp); 1735 break; 1736 1737 case TCPS_TIME_WAIT: 1738 break; 1739 } 1740 } 1741 goto drop; 1742 } 1743 1744 /* 1745 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1746 * and it's less than ts_recent, drop it. 1747 */ 1748 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 && 1749 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1750 /* Check to see if ts_recent is over 24 days old. */ 1751 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1752 /* 1753 * Invalidate ts_recent. If this segment updates 1754 * ts_recent, the age will be reset later and ts_recent 1755 * will get a valid value. If it does not, setting 1756 * ts_recent to zero will at least satisfy the 1757 * requirement that zero be placed in the timestamp 1758 * echo reply when ts_recent isn't valid. The 1759 * age isn't reset until we get a valid ts_recent 1760 * because we don't want out-of-order segments to be 1761 * dropped when ts_recent is old. 1762 */ 1763 tp->ts_recent = 0; 1764 } else if (tcp_paws_tolerance && tlen != 0 && 1765 tp->t_state == TCPS_ESTABLISHED && 1766 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK&& 1767 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) && 1768 th->th_ack == tp->snd_una && 1769 tiwin == tp->snd_wnd && 1770 TSTMP_GEQ(to.to_tsval + tcp_paws_tolerance, tp->ts_recent)&& 1771 (th->th_seq == tp->rcv_nxt || 1772 (SEQ_GT(th->th_seq, tp->rcv_nxt) && 1773 tcp_paws_canreasslast(tp, th, tlen)))) { 1774 /* 1775 * This tends to prevent valid new segments from being 1776 * dropped by the reordered segments sent by the fast 1777 * retransmission algorithm on the sending side, i.e. 1778 * the fast retransmitted segment w/ larger timestamp 1779 * arrives earlier than the previously sent new segments 1780 * w/ smaller timestamp. 1781 * 1782 * If following conditions are met, the segment is 1783 * accepted: 1784 * - The segment contains data 1785 * - The connection is established 1786 * - The header does not contain important flags 1787 * - SYN or FIN is not needed 1788 * - It does not acknowledge new data 1789 * - Receive window is not changed 1790 * - The timestamp is within "acceptable" range 1791 * - The new segment is what we are expecting or 1792 * the new segment could be merged w/ the last 1793 * pending segment on the reassemble queue 1794 */ 1795 tcpstat.tcps_pawsaccept++; 1796 tcpstat.tcps_pawsdrop++; 1797 } else { 1798 tcpstat.tcps_rcvduppack++; 1799 tcpstat.tcps_rcvdupbyte += tlen; 1800 tcpstat.tcps_pawsdrop++; 1801 if (tlen) 1802 goto dropafterack; 1803 goto drop; 1804 } 1805 } 1806 1807 /* 1808 * In the SYN-RECEIVED state, validate that the packet belongs to 1809 * this connection before trimming the data to fit the receive 1810 * window. Check the sequence number versus IRS since we know 1811 * the sequence numbers haven't wrapped. This is a partial fix 1812 * for the "LAND" DoS attack. 1813 */ 1814 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1815 rstreason = BANDLIM_RST_OPENPORT; 1816 goto dropwithreset; 1817 } 1818 1819 todrop = tp->rcv_nxt - th->th_seq; 1820 if (todrop > 0) { 1821 if (TCP_DO_SACK(tp)) { 1822 /* Report duplicate segment at head of packet. */ 1823 tp->reportblk.rblk_start = th->th_seq; 1824 tp->reportblk.rblk_end = TCP_SACK_BLKEND( 1825 th->th_seq + tlen, thflags); 1826 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt)) 1827 tp->reportblk.rblk_end = tp->rcv_nxt; 1828 tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_SACKLEFT); 1829 tp->t_flags |= TF_ACKNOW; 1830 } 1831 if (thflags & TH_SYN) { 1832 thflags &= ~TH_SYN; 1833 th->th_seq++; 1834 if (th->th_urp > 1) 1835 th->th_urp--; 1836 else 1837 thflags &= ~TH_URG; 1838 todrop--; 1839 } 1840 /* 1841 * Following if statement from Stevens, vol. 2, p. 960. 1842 */ 1843 if (todrop > tlen || 1844 (todrop == tlen && !(thflags & TH_FIN))) { 1845 /* 1846 * Any valid FIN must be to the left of the window. 1847 * At this point the FIN must be a duplicate or out 1848 * of sequence; drop it. 1849 */ 1850 thflags &= ~TH_FIN; 1851 1852 /* 1853 * Send an ACK to resynchronize and drop any data. 1854 * But keep on processing for RST or ACK. 1855 */ 1856 tp->t_flags |= TF_ACKNOW; 1857 todrop = tlen; 1858 tcpstat.tcps_rcvduppack++; 1859 tcpstat.tcps_rcvdupbyte += todrop; 1860 } else { 1861 tcpstat.tcps_rcvpartduppack++; 1862 tcpstat.tcps_rcvpartdupbyte += todrop; 1863 } 1864 drop_hdrlen += todrop; /* drop from the top afterwards */ 1865 th->th_seq += todrop; 1866 tlen -= todrop; 1867 if (th->th_urp > todrop) 1868 th->th_urp -= todrop; 1869 else { 1870 thflags &= ~TH_URG; 1871 th->th_urp = 0; 1872 } 1873 } 1874 1875 /* 1876 * If new data are received on a connection after the 1877 * user processes are gone, then RST the other end. 1878 */ 1879 if ((so->so_state & SS_NOFDREF) && 1880 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1881 tp = tcp_close(tp); 1882 tcpstat.tcps_rcvafterclose++; 1883 rstreason = BANDLIM_UNLIMITED; 1884 goto dropwithreset; 1885 } 1886 1887 /* 1888 * If segment ends after window, drop trailing data 1889 * (and PUSH and FIN); if nothing left, just ACK. 1890 */ 1891 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 1892 if (todrop > 0) { 1893 tcpstat.tcps_rcvpackafterwin++; 1894 if (todrop >= tlen) { 1895 tcpstat.tcps_rcvbyteafterwin += tlen; 1896 /* 1897 * If a new connection request is received 1898 * while in TIME_WAIT, drop the old connection 1899 * and start over if the sequence numbers 1900 * are above the previous ones. 1901 */ 1902 if (thflags & TH_SYN && 1903 tp->t_state == TCPS_TIME_WAIT && 1904 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1905 tp = tcp_close(tp); 1906 goto findpcb; 1907 } 1908 /* 1909 * If window is closed can only take segments at 1910 * window edge, and have to drop data and PUSH from 1911 * incoming segments. Continue processing, but 1912 * remember to ack. Otherwise, drop segment 1913 * and ack. 1914 */ 1915 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1916 tp->t_flags |= TF_ACKNOW; 1917 tcpstat.tcps_rcvwinprobe++; 1918 } else 1919 goto dropafterack; 1920 } else 1921 tcpstat.tcps_rcvbyteafterwin += todrop; 1922 m_adj(m, -todrop); 1923 tlen -= todrop; 1924 thflags &= ~(TH_PUSH | TH_FIN); 1925 } 1926 1927 /* 1928 * If last ACK falls within this segment's sequence numbers, 1929 * record its timestamp. 1930 * NOTE: 1931 * 1) That the test incorporates suggestions from the latest 1932 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1933 * 2) That updating only on newer timestamps interferes with 1934 * our earlier PAWS tests, so this check should be solely 1935 * predicated on the sequence space of this segment. 1936 * 3) That we modify the segment boundary check to be 1937 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN 1938 * instead of RFC1323's 1939 * Last.ACK.Sent < SEG.SEQ + SEG.LEN, 1940 * This modified check allows us to overcome RFC1323's 1941 * limitations as described in Stevens TCP/IP Illustrated 1942 * Vol. 2 p.869. In such cases, we can still calculate the 1943 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1944 */ 1945 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1946 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen 1947 + ((thflags & TH_SYN) != 0) 1948 + ((thflags & TH_FIN) != 0)))) { 1949 tp->ts_recent_age = ticks; 1950 tp->ts_recent = to.to_tsval; 1951 } 1952 1953 /* 1954 * If a SYN is in the window, then this is an 1955 * error and we send an RST and drop the connection. 1956 */ 1957 if (thflags & TH_SYN) { 1958 tp = tcp_drop(tp, ECONNRESET); 1959 rstreason = BANDLIM_UNLIMITED; 1960 goto dropwithreset; 1961 } 1962 1963 /* 1964 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1965 * flag is on (half-synchronized state), then queue data for 1966 * later processing; else drop segment and return. 1967 */ 1968 if (!(thflags & TH_ACK)) { 1969 if (tp->t_state == TCPS_SYN_RECEIVED || 1970 (tp->t_flags & TF_NEEDSYN)) 1971 goto step6; 1972 else 1973 goto drop; 1974 } 1975 1976 /* 1977 * Ack processing. 1978 */ 1979 switch (tp->t_state) { 1980 /* 1981 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter 1982 * ESTABLISHED state and continue processing. 1983 * The ACK was checked above. 1984 */ 1985 case TCPS_SYN_RECEIVED: 1986 1987 tcpstat.tcps_connects++; 1988 soisconnected(so); 1989 /* Do window scaling? */ 1990 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 1991 (TF_RCVD_SCALE | TF_REQ_SCALE)) 1992 tp->rcv_scale = tp->request_r_scale; 1993 /* 1994 * Make transitions: 1995 * SYN-RECEIVED -> ESTABLISHED 1996 * SYN-RECEIVED* -> FIN-WAIT-1 1997 */ 1998 tp->t_starttime = ticks; 1999 if (tp->t_flags & TF_NEEDFIN) { 2000 TCP_STATE_CHANGE(tp, TCPS_FIN_WAIT_1); 2001 tp->t_flags &= ~TF_NEEDFIN; 2002 } else { 2003 tcp_established(tp); 2004 } 2005 /* 2006 * If segment contains data or ACK, will call tcp_reass() 2007 * later; if not, do so now to pass queued data to user. 2008 */ 2009 if (tlen == 0 && !(thflags & TH_FIN)) 2010 tcp_reass(tp, NULL, NULL, NULL); 2011 /* fall into ... */ 2012 2013 /* 2014 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2015 * ACKs. If the ack is in the range 2016 * tp->snd_una < th->th_ack <= tp->snd_max 2017 * then advance tp->snd_una to th->th_ack and drop 2018 * data from the retransmission queue. If this ACK reflects 2019 * more up to date window information we update our window information. 2020 */ 2021 case TCPS_ESTABLISHED: 2022 case TCPS_FIN_WAIT_1: 2023 case TCPS_FIN_WAIT_2: 2024 case TCPS_CLOSE_WAIT: 2025 case TCPS_CLOSING: 2026 case TCPS_LAST_ACK: 2027 case TCPS_TIME_WAIT: 2028 2029 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2030 boolean_t maynotdup = FALSE; 2031 2032 if (TCP_DO_SACK(tp)) 2033 tcp_sack_update_scoreboard(tp, &to); 2034 2035 if (tlen != 0 || tiwin != tp->snd_wnd || 2036 ((thflags & TH_FIN) && !(tp->t_flags & TF_SAWFIN))) 2037 maynotdup = TRUE; 2038 2039 if (!tcp_callout_active(tp, tp->tt_rexmt) || 2040 th->th_ack != tp->snd_una) { 2041 if (!maynotdup) 2042 tcpstat.tcps_rcvdupack++; 2043 tp->t_dupacks = 0; 2044 break; 2045 } 2046 2047 #define DELAY_DUPACK \ 2048 do { \ 2049 delayed_dupack = TRUE; \ 2050 th_dupack = th->th_ack; \ 2051 to_flags = to.to_flags; \ 2052 } while (0) 2053 if (maynotdup) { 2054 if (!tcp_do_rfc6675 || 2055 !TCP_DO_SACK(tp) || 2056 (to.to_flags & 2057 (TOF_SACK | TOF_SACK_REDUNDANT)) 2058 != TOF_SACK) { 2059 tp->t_dupacks = 0; 2060 } else { 2061 DELAY_DUPACK; 2062 } 2063 break; 2064 } 2065 if ((thflags & TH_FIN) && !(tp->t_flags & TF_QUEDFIN)) { 2066 /* 2067 * This could happen, if the reassemable 2068 * queue overflew or was drained. Don't 2069 * drop this FIN here; defer the duplicated 2070 * ACK processing until this FIN gets queued. 2071 */ 2072 DELAY_DUPACK; 2073 break; 2074 } 2075 #undef DELAY_DUPACK 2076 2077 if (tcp_recv_dupack(tp, th->th_ack, to.to_flags)) 2078 goto drop; 2079 else 2080 break; 2081 } 2082 2083 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 2084 tp->t_dupacks = 0; 2085 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2086 /* 2087 * Detected optimistic ACK attack. 2088 * Force slow-start to de-synchronize attack. 2089 */ 2090 tp->snd_cwnd = tp->t_maxseg; 2091 tp->snd_wacked = 0; 2092 2093 tcpstat.tcps_rcvacktoomuch++; 2094 goto dropafterack; 2095 } 2096 /* 2097 * If we reach this point, ACK is not a duplicate, 2098 * i.e., it ACKs something we sent. 2099 */ 2100 if (tp->t_flags & TF_NEEDSYN) { 2101 /* 2102 * T/TCP: Connection was half-synchronized, and our 2103 * SYN has been ACK'd (so connection is now fully 2104 * synchronized). Go to non-starred state, 2105 * increment snd_una for ACK of SYN, and check if 2106 * we can do window scaling. 2107 */ 2108 tp->t_flags &= ~TF_NEEDSYN; 2109 tp->snd_una++; 2110 /* Do window scaling? */ 2111 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 2112 (TF_RCVD_SCALE | TF_REQ_SCALE)) 2113 tp->rcv_scale = tp->request_r_scale; 2114 } 2115 2116 process_ACK: 2117 acked = th->th_ack - tp->snd_una; 2118 tcpstat.tcps_rcvackpack++; 2119 tcpstat.tcps_rcvackbyte += acked; 2120 2121 if (tcp_do_eifel_detect && acked > 0 && 2122 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) && 2123 (tp->rxt_flags & TRXT_F_FIRSTACCACK)) { 2124 /* Eifel detection applicable. */ 2125 if (to.to_tsecr < tp->t_rexmtTS) { 2126 ++tcpstat.tcps_eifeldetected; 2127 tcp_revert_congestion_state(tp); 2128 if (tp->t_rxtshift != 1 || 2129 ticks >= tp->t_badrxtwin) 2130 ++tcpstat.tcps_rttcantdetect; 2131 } 2132 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2133 /* 2134 * If we just performed our first retransmit, 2135 * and the ACK arrives within our recovery window, 2136 * then it was a mistake to do the retransmit 2137 * in the first place. Recover our original cwnd 2138 * and ssthresh, and proceed to transmit where we 2139 * left off. 2140 */ 2141 tcp_revert_congestion_state(tp); 2142 ++tcpstat.tcps_rttdetected; 2143 } 2144 2145 /* 2146 * If we have a timestamp reply, update smoothed 2147 * round trip time. If no timestamp is present but 2148 * transmit timer is running and timed sequence 2149 * number was acked, update smoothed round trip time. 2150 * Since we now have an rtt measurement, cancel the 2151 * timer backoff (cf., Phil Karn's retransmit alg.). 2152 * Recompute the initial retransmit timer. 2153 * 2154 * Some machines (certain windows boxes) send broken 2155 * timestamp replies during the SYN+ACK phase, ignore 2156 * timestamps of 0. 2157 */ 2158 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) 2159 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1, 2160 th->th_ack); 2161 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 2162 tcp_xmit_timer(tp, ticks - tp->t_rtttime + 1, 2163 th->th_ack); 2164 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2165 2166 /* 2167 * If no data (only SYN) was ACK'd, 2168 * skip rest of ACK processing. 2169 */ 2170 if (acked == 0) 2171 goto step6; 2172 2173 /* Stop looking for an acceptable ACK since one was received. */ 2174 tp->rxt_flags &= ~(TRXT_F_FIRSTACCACK | 2175 TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT); 2176 2177 if (acked > so->so_snd.ssb_cc) { 2178 tp->snd_wnd -= so->so_snd.ssb_cc; 2179 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc); 2180 ourfinisacked = TRUE; 2181 } else { 2182 sbdrop(&so->so_snd.sb, acked); 2183 tp->snd_wnd -= acked; 2184 ourfinisacked = FALSE; 2185 } 2186 sowwakeup(so); 2187 2188 /* 2189 * Update window information. 2190 */ 2191 if (acceptable_window_update(tp, th, tiwin)) { 2192 /* keep track of pure window updates */ 2193 if (tlen == 0 && tp->snd_wl2 == th->th_ack && 2194 tiwin > tp->snd_wnd) 2195 tcpstat.tcps_rcvwinupd++; 2196 tp->snd_wnd = tiwin; 2197 tp->snd_wl1 = th->th_seq; 2198 tp->snd_wl2 = th->th_ack; 2199 if (tp->snd_wnd > tp->max_sndwnd) 2200 tp->max_sndwnd = tp->snd_wnd; 2201 needoutput = TRUE; 2202 } 2203 2204 tp->snd_una = th->th_ack; 2205 if (TCP_DO_SACK(tp)) 2206 tcp_sack_update_scoreboard(tp, &to); 2207 if (IN_FASTRECOVERY(tp)) { 2208 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2209 EXIT_FASTRECOVERY(tp); 2210 needoutput = TRUE; 2211 /* 2212 * If the congestion window was inflated 2213 * to account for the other side's 2214 * cached packets, retract it. 2215 */ 2216 if (!TCP_DO_SACK(tp)) 2217 tp->snd_cwnd = tp->snd_ssthresh; 2218 2219 /* 2220 * Window inflation should have left us 2221 * with approximately snd_ssthresh outstanding 2222 * data. But, in case we would be inclined 2223 * to send a burst, better do it using 2224 * slow start. 2225 */ 2226 if (SEQ_GT(th->th_ack + tp->snd_cwnd, 2227 tp->snd_max + 2 * tp->t_maxseg)) 2228 tp->snd_cwnd = 2229 (tp->snd_max - tp->snd_una) + 2230 2 * tp->t_maxseg; 2231 2232 tp->snd_wacked = 0; 2233 } else { 2234 if (TCP_DO_SACK(tp)) { 2235 tp->snd_max_rexmt = tp->snd_max; 2236 tcp_sack_rexmt(tp, 2237 tp->snd_una == tp->rexmt_high); 2238 } else { 2239 tcp_newreno_partial_ack(tp, th, acked); 2240 } 2241 needoutput = FALSE; 2242 } 2243 } else { 2244 /* 2245 * Open the congestion window. When in slow-start, 2246 * open exponentially: maxseg per packet. Otherwise, 2247 * open linearly: maxseg per window. 2248 */ 2249 if (tp->snd_cwnd <= tp->snd_ssthresh) { 2250 u_int abc_sslimit = 2251 (SEQ_LT(tp->snd_nxt, tp->snd_max) ? 2252 tp->t_maxseg : 2 * tp->t_maxseg); 2253 2254 /* slow-start */ 2255 tp->snd_cwnd += tcp_do_abc ? 2256 min(acked, abc_sslimit) : tp->t_maxseg; 2257 } else { 2258 /* linear increase */ 2259 tp->snd_wacked += tcp_do_abc ? acked : 2260 tp->t_maxseg; 2261 if (tp->snd_wacked >= tp->snd_cwnd) { 2262 tp->snd_wacked -= tp->snd_cwnd; 2263 tp->snd_cwnd += tp->t_maxseg; 2264 } 2265 } 2266 tp->snd_cwnd = min(tp->snd_cwnd, 2267 TCP_MAXWIN << tp->snd_scale); 2268 tp->snd_recover = th->th_ack - 1; 2269 } 2270 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2271 tp->snd_nxt = tp->snd_una; 2272 2273 /* 2274 * If all outstanding data is acked, stop retransmit 2275 * timer and remember to restart (more output or persist). 2276 * If there is more data to be acked, restart retransmit 2277 * timer, using current (possibly backed-off) value. 2278 */ 2279 if (th->th_ack == tp->snd_max) { 2280 tcp_callout_stop(tp, tp->tt_rexmt); 2281 needoutput = TRUE; 2282 } else if (!tcp_callout_active(tp, tp->tt_persist)) { 2283 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur, 2284 tcp_timer_rexmt); 2285 } 2286 2287 switch (tp->t_state) { 2288 /* 2289 * In FIN_WAIT_1 STATE in addition to the processing 2290 * for the ESTABLISHED state if our FIN is now acknowledged 2291 * then enter FIN_WAIT_2. 2292 */ 2293 case TCPS_FIN_WAIT_1: 2294 if (ourfinisacked) { 2295 /* 2296 * If we can't receive any more 2297 * data, then closing user can proceed. 2298 * Starting the timer is contrary to the 2299 * specification, but if we don't get a FIN 2300 * we'll hang forever. 2301 */ 2302 if (so->so_state & SS_CANTRCVMORE) { 2303 soisdisconnected(so); 2304 tcp_callout_reset(tp, tp->tt_2msl, 2305 tp->t_maxidle, tcp_timer_2msl); 2306 } 2307 TCP_STATE_CHANGE(tp, TCPS_FIN_WAIT_2); 2308 } 2309 break; 2310 2311 /* 2312 * In CLOSING STATE in addition to the processing for 2313 * the ESTABLISHED state if the ACK acknowledges our FIN 2314 * then enter the TIME-WAIT state, otherwise ignore 2315 * the segment. 2316 */ 2317 case TCPS_CLOSING: 2318 if (ourfinisacked) { 2319 TCP_STATE_CHANGE(tp, TCPS_TIME_WAIT); 2320 tcp_canceltimers(tp); 2321 tcp_callout_reset(tp, tp->tt_2msl, 2322 2 * tcp_rmx_msl(tp), 2323 tcp_timer_2msl); 2324 soisdisconnected(so); 2325 } 2326 break; 2327 2328 /* 2329 * In LAST_ACK, we may still be waiting for data to drain 2330 * and/or to be acked, as well as for the ack of our FIN. 2331 * If our FIN is now acknowledged, delete the TCB, 2332 * enter the closed state and return. 2333 */ 2334 case TCPS_LAST_ACK: 2335 if (ourfinisacked) { 2336 tp = tcp_close(tp); 2337 goto drop; 2338 } 2339 break; 2340 2341 /* 2342 * In TIME_WAIT state the only thing that should arrive 2343 * is a retransmission of the remote FIN. Acknowledge 2344 * it and restart the finack timer. 2345 */ 2346 case TCPS_TIME_WAIT: 2347 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp), 2348 tcp_timer_2msl); 2349 goto dropafterack; 2350 } 2351 } 2352 2353 step6: 2354 /* 2355 * Update window information. 2356 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2357 */ 2358 if ((thflags & TH_ACK) && 2359 acceptable_window_update(tp, th, tiwin)) { 2360 /* keep track of pure window updates */ 2361 if (tlen == 0 && tp->snd_wl2 == th->th_ack && 2362 tiwin > tp->snd_wnd) 2363 tcpstat.tcps_rcvwinupd++; 2364 tp->snd_wnd = tiwin; 2365 tp->snd_wl1 = th->th_seq; 2366 tp->snd_wl2 = th->th_ack; 2367 if (tp->snd_wnd > tp->max_sndwnd) 2368 tp->max_sndwnd = tp->snd_wnd; 2369 needoutput = TRUE; 2370 } 2371 2372 /* 2373 * Process segments with URG. 2374 */ 2375 if ((thflags & TH_URG) && th->th_urp && 2376 !TCPS_HAVERCVDFIN(tp->t_state)) { 2377 /* 2378 * This is a kludge, but if we receive and accept 2379 * random urgent pointers, we'll crash in 2380 * soreceive. It's hard to imagine someone 2381 * actually wanting to send this much urgent data. 2382 */ 2383 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) { 2384 th->th_urp = 0; /* XXX */ 2385 thflags &= ~TH_URG; /* XXX */ 2386 goto dodata; /* XXX */ 2387 } 2388 /* 2389 * If this segment advances the known urgent pointer, 2390 * then mark the data stream. This should not happen 2391 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2392 * a FIN has been received from the remote side. 2393 * In these states we ignore the URG. 2394 * 2395 * According to RFC961 (Assigned Protocols), 2396 * the urgent pointer points to the last octet 2397 * of urgent data. We continue, however, 2398 * to consider it to indicate the first octet 2399 * of data past the urgent section as the original 2400 * spec states (in one of two places). 2401 */ 2402 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) { 2403 tp->rcv_up = th->th_seq + th->th_urp; 2404 so->so_oobmark = so->so_rcv.ssb_cc + 2405 (tp->rcv_up - tp->rcv_nxt) - 1; 2406 if (so->so_oobmark == 0) 2407 sosetstate(so, SS_RCVATMARK); 2408 sohasoutofband(so); 2409 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2410 } 2411 /* 2412 * Remove out of band data so doesn't get presented to user. 2413 * This can happen independent of advancing the URG pointer, 2414 * but if two URG's are pending at once, some out-of-band 2415 * data may creep in... ick. 2416 */ 2417 if (th->th_urp <= (u_long)tlen && 2418 !(so->so_options & SO_OOBINLINE)) { 2419 /* hdr drop is delayed */ 2420 tcp_pulloutofband(so, th, m, drop_hdrlen); 2421 } 2422 } else { 2423 /* 2424 * If no out of band data is expected, 2425 * pull receive urgent pointer along 2426 * with the receive window. 2427 */ 2428 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2429 tp->rcv_up = tp->rcv_nxt; 2430 } 2431 2432 dodata: /* XXX */ 2433 /* 2434 * Process the segment text, merging it into the TCP sequencing queue, 2435 * and arranging for acknowledgment of receipt if necessary. 2436 * This process logically involves adjusting tp->rcv_wnd as data 2437 * is presented to the user (this happens in tcp_usrreq.c, 2438 * case PRU_RCVD). If a FIN has already been received on this 2439 * connection then we just ignore the text. 2440 */ 2441 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) { 2442 if (thflags & TH_FIN) 2443 tp->t_flags |= TF_SAWFIN; 2444 m_adj(m, drop_hdrlen); /* delayed header drop */ 2445 /* 2446 * Insert segment which includes th into TCP reassembly queue 2447 * with control block tp. Set thflags to whether reassembly now 2448 * includes a segment with FIN. This handles the common case 2449 * inline (segment is the next to be received on an established 2450 * connection, and the queue is empty), avoiding linkage into 2451 * and removal from the queue and repetition of various 2452 * conversions. 2453 * Set DELACK for segments received in order, but ack 2454 * immediately when segments are out of order (so 2455 * fast retransmit can work). 2456 */ 2457 if (th->th_seq == tp->rcv_nxt && 2458 TAILQ_EMPTY(&tp->t_segq) && 2459 TCPS_HAVEESTABLISHED(tp->t_state)) { 2460 if (thflags & TH_FIN) 2461 tp->t_flags |= TF_QUEDFIN; 2462 if (DELAY_ACK(tp)) { 2463 tcp_callout_reset(tp, tp->tt_delack, 2464 tcp_delacktime, tcp_timer_delack); 2465 } else { 2466 tp->t_flags |= TF_ACKNOW; 2467 } 2468 tp->rcv_nxt += tlen; 2469 thflags = th->th_flags & TH_FIN; 2470 tcpstat.tcps_rcvpack++; 2471 tcpstat.tcps_rcvbyte += tlen; 2472 ND6_HINT(tp); 2473 if (so->so_state & SS_CANTRCVMORE) { 2474 m_freem(m); 2475 } else { 2476 lwkt_gettoken(&so->so_rcv.ssb_token); 2477 ssb_appendstream(&so->so_rcv, m); 2478 lwkt_reltoken(&so->so_rcv.ssb_token); 2479 } 2480 sorwakeup(so); 2481 } else { 2482 if (!(tp->sack_flags & TSACK_F_DUPSEG)) { 2483 /* Initialize SACK report block. */ 2484 tp->reportblk.rblk_start = th->th_seq; 2485 tp->reportblk.rblk_end = TCP_SACK_BLKEND( 2486 th->th_seq + tlen, thflags); 2487 } 2488 thflags = tcp_reass(tp, th, &tlen, m); 2489 tp->t_flags |= TF_ACKNOW; 2490 } 2491 2492 /* 2493 * Note the amount of data that peer has sent into 2494 * our window, in order to estimate the sender's 2495 * buffer size. 2496 */ 2497 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2498 } else { 2499 m_freem(m); 2500 thflags &= ~TH_FIN; 2501 } 2502 2503 /* 2504 * If FIN is received ACK the FIN and let the user know 2505 * that the connection is closing. 2506 */ 2507 if (thflags & TH_FIN) { 2508 if (!TCPS_HAVERCVDFIN(tp->t_state)) { 2509 socantrcvmore(so); 2510 /* 2511 * If connection is half-synchronized 2512 * (ie NEEDSYN flag on) then delay ACK, 2513 * so it may be piggybacked when SYN is sent. 2514 * Otherwise, since we received a FIN then no 2515 * more input can be expected, send ACK now. 2516 */ 2517 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) { 2518 tcp_callout_reset(tp, tp->tt_delack, 2519 tcp_delacktime, tcp_timer_delack); 2520 } else { 2521 tp->t_flags |= TF_ACKNOW; 2522 } 2523 tp->rcv_nxt++; 2524 } 2525 2526 switch (tp->t_state) { 2527 /* 2528 * In SYN_RECEIVED and ESTABLISHED STATES 2529 * enter the CLOSE_WAIT state. 2530 */ 2531 case TCPS_SYN_RECEIVED: 2532 tp->t_starttime = ticks; 2533 /*FALLTHROUGH*/ 2534 case TCPS_ESTABLISHED: 2535 TCP_STATE_CHANGE(tp, TCPS_CLOSE_WAIT); 2536 break; 2537 2538 /* 2539 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2540 * enter the CLOSING state. 2541 */ 2542 case TCPS_FIN_WAIT_1: 2543 TCP_STATE_CHANGE(tp, TCPS_CLOSING); 2544 break; 2545 2546 /* 2547 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2548 * starting the time-wait timer, turning off the other 2549 * standard timers. 2550 */ 2551 case TCPS_FIN_WAIT_2: 2552 TCP_STATE_CHANGE(tp, TCPS_TIME_WAIT); 2553 tcp_canceltimers(tp); 2554 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp), 2555 tcp_timer_2msl); 2556 soisdisconnected(so); 2557 break; 2558 2559 /* 2560 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2561 */ 2562 case TCPS_TIME_WAIT: 2563 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp), 2564 tcp_timer_2msl); 2565 break; 2566 } 2567 } 2568 2569 #ifdef TCPDEBUG 2570 if (so->so_options & SO_DEBUG) 2571 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2572 #endif 2573 2574 /* 2575 * Delayed duplicated ACK processing 2576 */ 2577 if (delayed_dupack && tcp_recv_dupack(tp, th_dupack, to_flags)) 2578 needoutput = FALSE; 2579 2580 /* 2581 * Return any desired output. 2582 */ 2583 if ((tp->t_flags & TF_ACKNOW) || 2584 (needoutput && tcp_sack_report_needed(tp))) { 2585 tcp_output_cancel(tp); 2586 tcp_output_fair(tp); 2587 } else if (needoutput && !tcp_output_pending(tp)) { 2588 tcp_output_fair(tp); 2589 } 2590 tcp_sack_report_cleanup(tp); 2591 return(IPPROTO_DONE); 2592 2593 dropafterack: 2594 /* 2595 * Generate an ACK dropping incoming segment if it occupies 2596 * sequence space, where the ACK reflects our state. 2597 * 2598 * We can now skip the test for the RST flag since all 2599 * paths to this code happen after packets containing 2600 * RST have been dropped. 2601 * 2602 * In the SYN-RECEIVED state, don't send an ACK unless the 2603 * segment we received passes the SYN-RECEIVED ACK test. 2604 * If it fails send a RST. This breaks the loop in the 2605 * "LAND" DoS attack, and also prevents an ACK storm 2606 * between two listening ports that have been sent forged 2607 * SYN segments, each with the source address of the other. 2608 */ 2609 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2610 (SEQ_GT(tp->snd_una, th->th_ack) || 2611 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2612 rstreason = BANDLIM_RST_OPENPORT; 2613 goto dropwithreset; 2614 } 2615 #ifdef TCPDEBUG 2616 if (so->so_options & SO_DEBUG) 2617 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2618 #endif 2619 m_freem(m); 2620 tp->t_flags |= TF_ACKNOW; 2621 tcp_output(tp); 2622 tcp_sack_report_cleanup(tp); 2623 return(IPPROTO_DONE); 2624 2625 dropwithreset: 2626 /* 2627 * Generate a RST, dropping incoming segment. 2628 * Make ACK acceptable to originator of segment. 2629 * Don't bother to respond if destination was broadcast/multicast. 2630 */ 2631 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST)) 2632 goto drop; 2633 if (isipv6) { 2634 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2635 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2636 goto drop; 2637 } else { 2638 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2639 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2640 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2641 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2642 goto drop; 2643 } 2644 /* IPv6 anycast check is done at tcp6_input() */ 2645 2646 /* 2647 * Perform bandwidth limiting. 2648 */ 2649 #ifdef ICMP_BANDLIM 2650 if (badport_bandlim(rstreason) < 0) 2651 goto drop; 2652 #endif 2653 2654 #ifdef TCPDEBUG 2655 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2656 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2657 #endif 2658 if (thflags & TH_ACK) 2659 /* mtod() below is safe as long as hdr dropping is delayed */ 2660 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2661 TH_RST); 2662 else { 2663 if (thflags & TH_SYN) 2664 tlen++; 2665 /* mtod() below is safe as long as hdr dropping is delayed */ 2666 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen, 2667 (tcp_seq)0, TH_RST | TH_ACK); 2668 } 2669 if (tp != NULL) 2670 tcp_sack_report_cleanup(tp); 2671 return(IPPROTO_DONE); 2672 2673 drop: 2674 /* 2675 * Drop space held by incoming segment and return. 2676 */ 2677 #ifdef TCPDEBUG 2678 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2679 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0); 2680 #endif 2681 m_freem(m); 2682 if (tp != NULL) 2683 tcp_sack_report_cleanup(tp); 2684 return(IPPROTO_DONE); 2685 } 2686 2687 /* 2688 * Parse TCP options and place in tcpopt. 2689 */ 2690 static void 2691 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn, 2692 tcp_seq ack) 2693 { 2694 int opt, optlen, i; 2695 2696 to->to_flags = 0; 2697 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2698 opt = cp[0]; 2699 if (opt == TCPOPT_EOL) 2700 break; 2701 if (opt == TCPOPT_NOP) 2702 optlen = 1; 2703 else { 2704 if (cnt < 2) 2705 break; 2706 optlen = cp[1]; 2707 if (optlen < 2 || optlen > cnt) 2708 break; 2709 } 2710 switch (opt) { 2711 case TCPOPT_MAXSEG: 2712 if (optlen != TCPOLEN_MAXSEG) 2713 continue; 2714 if (!is_syn) 2715 continue; 2716 to->to_flags |= TOF_MSS; 2717 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss); 2718 to->to_mss = ntohs(to->to_mss); 2719 break; 2720 case TCPOPT_WINDOW: 2721 if (optlen != TCPOLEN_WINDOW) 2722 continue; 2723 if (!is_syn) 2724 continue; 2725 to->to_flags |= TOF_SCALE; 2726 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2727 break; 2728 case TCPOPT_TIMESTAMP: 2729 if (optlen != TCPOLEN_TIMESTAMP) 2730 continue; 2731 to->to_flags |= TOF_TS; 2732 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval); 2733 to->to_tsval = ntohl(to->to_tsval); 2734 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr); 2735 to->to_tsecr = ntohl(to->to_tsecr); 2736 /* 2737 * If echoed timestamp is later than the current time, 2738 * fall back to non RFC1323 RTT calculation. 2739 */ 2740 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks)) 2741 to->to_tsecr = 0; 2742 break; 2743 case TCPOPT_SACK_PERMITTED: 2744 if (optlen != TCPOLEN_SACK_PERMITTED) 2745 continue; 2746 if (!is_syn) 2747 continue; 2748 to->to_flags |= TOF_SACK_PERMITTED; 2749 break; 2750 case TCPOPT_SACK: 2751 if ((optlen - 2) & 0x07) /* not multiple of 8 */ 2752 continue; 2753 to->to_nsackblocks = (optlen - 2) / 8; 2754 to->to_sackblocks = (struct raw_sackblock *) (cp + 2); 2755 to->to_flags |= TOF_SACK; 2756 for (i = 0; i < to->to_nsackblocks; i++) { 2757 struct raw_sackblock *r = &to->to_sackblocks[i]; 2758 2759 r->rblk_start = ntohl(r->rblk_start); 2760 r->rblk_end = ntohl(r->rblk_end); 2761 2762 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) { 2763 /* 2764 * Invalid SACK block; discard all 2765 * SACK blocks 2766 */ 2767 tcpstat.tcps_rcvbadsackopt++; 2768 to->to_nsackblocks = 0; 2769 to->to_sackblocks = NULL; 2770 to->to_flags &= ~TOF_SACK; 2771 break; 2772 } 2773 } 2774 if ((to->to_flags & TOF_SACK) && 2775 tcp_sack_ndsack_blocks(to->to_sackblocks, 2776 to->to_nsackblocks, ack)) 2777 to->to_flags |= TOF_DSACK; 2778 break; 2779 #ifdef TCP_SIGNATURE 2780 /* 2781 * XXX In order to reply to a host which has set the 2782 * TCP_SIGNATURE option in its initial SYN, we have to 2783 * record the fact that the option was observed here 2784 * for the syncache code to perform the correct response. 2785 */ 2786 case TCPOPT_SIGNATURE: 2787 if (optlen != TCPOLEN_SIGNATURE) 2788 continue; 2789 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2790 break; 2791 #endif /* TCP_SIGNATURE */ 2792 default: 2793 continue; 2794 } 2795 } 2796 } 2797 2798 /* 2799 * Pull out of band byte out of a segment so 2800 * it doesn't appear in the user's data queue. 2801 * It is still reflected in the segment length for 2802 * sequencing purposes. 2803 * "off" is the delayed to be dropped hdrlen. 2804 */ 2805 static void 2806 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off) 2807 { 2808 int cnt = off + th->th_urp - 1; 2809 2810 while (cnt >= 0) { 2811 if (m->m_len > cnt) { 2812 char *cp = mtod(m, caddr_t) + cnt; 2813 struct tcpcb *tp = sototcpcb(so); 2814 2815 tp->t_iobc = *cp; 2816 tp->t_oobflags |= TCPOOB_HAVEDATA; 2817 bcopy(cp + 1, cp, m->m_len - cnt - 1); 2818 m->m_len--; 2819 if (m->m_flags & M_PKTHDR) 2820 m->m_pkthdr.len--; 2821 return; 2822 } 2823 cnt -= m->m_len; 2824 m = m->m_next; 2825 if (m == NULL) 2826 break; 2827 } 2828 panic("tcp_pulloutofband"); 2829 } 2830 2831 /* 2832 * Collect new round-trip time estimate and update averages and current 2833 * timeout. 2834 */ 2835 static void 2836 tcp_xmit_timer(struct tcpcb *tp, int rtt, tcp_seq ack) 2837 { 2838 int rebaserto = 0; 2839 2840 tcpstat.tcps_rttupdated++; 2841 tp->t_rttupdated++; 2842 if ((tp->rxt_flags & TRXT_F_REBASERTO) && 2843 SEQ_GT(ack, tp->snd_max_prev)) { 2844 #ifdef DEBUG_EIFEL_RESPONSE 2845 kprintf("srtt/rttvar, prev %d/%d, cur %d/%d, ", 2846 tp->t_srtt_prev, tp->t_rttvar_prev, 2847 tp->t_srtt, tp->t_rttvar); 2848 #endif 2849 2850 tcpstat.tcps_eifelresponse++; 2851 rebaserto = 1; 2852 tp->rxt_flags &= ~TRXT_F_REBASERTO; 2853 tp->t_srtt = max(tp->t_srtt_prev, (rtt << TCP_RTT_SHIFT)); 2854 tp->t_rttvar = max(tp->t_rttvar_prev, 2855 (rtt << (TCP_RTTVAR_SHIFT - 1))); 2856 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2857 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2858 2859 #ifdef DEBUG_EIFEL_RESPONSE 2860 kprintf("new %d/%d ", tp->t_srtt, tp->t_rttvar); 2861 #endif 2862 } else if (tp->t_srtt != 0) { 2863 int delta; 2864 2865 /* 2866 * srtt is stored as fixed point with 5 bits after the 2867 * binary point (i.e., scaled by 32). The following magic 2868 * is equivalent to the smoothing algorithm in rfc793 with 2869 * an alpha of .875 (srtt = rtt/32 + srtt*31/32 in fixed 2870 * point). Adjust rtt to origin 0. 2871 */ 2872 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2873 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2874 2875 if ((tp->t_srtt += delta) <= 0) 2876 tp->t_srtt = 1; 2877 2878 /* 2879 * We accumulate a smoothed rtt variance (actually, a 2880 * smoothed mean difference), then set the retransmit 2881 * timer to smoothed rtt + 4 times the smoothed variance. 2882 * rttvar is stored as fixed point with 4 bits after the 2883 * binary point (scaled by 16). The following is 2884 * equivalent to rfc793 smoothing with an alpha of .75 2885 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2886 * rfc793's wired-in beta. 2887 */ 2888 if (delta < 0) 2889 delta = -delta; 2890 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2891 if ((tp->t_rttvar += delta) <= 0) 2892 tp->t_rttvar = 1; 2893 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2894 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2895 } else { 2896 /* 2897 * No rtt measurement yet - use the unsmoothed rtt. 2898 * Set the variance to half the rtt (so our first 2899 * retransmit happens at 3*rtt). 2900 */ 2901 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2902 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2903 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2904 } 2905 tp->t_rtttime = 0; 2906 tp->t_rxtshift = 0; 2907 2908 #ifdef DEBUG_EIFEL_RESPONSE 2909 if (rebaserto) { 2910 kprintf("| rxtcur prev %d, old %d, ", 2911 tp->t_rxtcur_prev, tp->t_rxtcur); 2912 } 2913 #endif 2914 2915 /* 2916 * the retransmit should happen at rtt + 4 * rttvar. 2917 * Because of the way we do the smoothing, srtt and rttvar 2918 * will each average +1/2 tick of bias. When we compute 2919 * the retransmit timer, we want 1/2 tick of rounding and 2920 * 1 extra tick because of +-1/2 tick uncertainty in the 2921 * firing of the timer. The bias will give us exactly the 2922 * 1.5 tick we need. But, because the bias is 2923 * statistical, we have to test that we don't drop below 2924 * the minimum feasible timer (which is 2 ticks). 2925 */ 2926 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2927 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2928 2929 if (rebaserto) { 2930 if (tp->t_rxtcur < tp->t_rxtcur_prev + tcp_eifel_rtoinc) { 2931 /* 2932 * RFC4015 requires that the new RTO is at least 2933 * 2*G (tcp_eifel_rtoinc) greater then the RTO 2934 * (t_rxtcur_prev) when the spurious retransmit 2935 * timeout happens. 2936 * 2937 * The above condition could be true, if the SRTT 2938 * and RTTVAR used to calculate t_rxtcur_prev 2939 * resulted in a value less than t_rttmin. So 2940 * simply increasing SRTT by tcp_eifel_rtoinc when 2941 * preparing for the Eifel response could not ensure 2942 * that the new RTO will be tcp_eifel_rtoinc greater 2943 * t_rxtcur_prev. 2944 */ 2945 tp->t_rxtcur = tp->t_rxtcur_prev + tcp_eifel_rtoinc; 2946 } 2947 #ifdef DEBUG_EIFEL_RESPONSE 2948 kprintf("new %d\n", tp->t_rxtcur); 2949 #endif 2950 } 2951 2952 /* 2953 * We received an ack for a packet that wasn't retransmitted; 2954 * it is probably safe to discard any error indications we've 2955 * received recently. This isn't quite right, but close enough 2956 * for now (a route might have failed after we sent a segment, 2957 * and the return path might not be symmetrical). 2958 */ 2959 tp->t_softerror = 0; 2960 } 2961 2962 /* 2963 * Determine a reasonable value for maxseg size. 2964 * If the route is known, check route for mtu. 2965 * If none, use an mss that can be handled on the outgoing 2966 * interface without forcing IP to fragment; if bigger than 2967 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2968 * to utilize large mbufs. If no route is found, route has no mtu, 2969 * or the destination isn't local, use a default, hopefully conservative 2970 * size (usually 512 or the default IP max size, but no more than the mtu 2971 * of the interface), as we can't discover anything about intervening 2972 * gateways or networks. We also initialize the congestion/slow start 2973 * window to be a single segment if the destination isn't local. 2974 * 2975 * Also take into account the space needed for options that we 2976 * send regularly. Make maxseg shorter by that amount to assure 2977 * that we can send maxseg amount of data even when the options 2978 * are present. Store the upper limit of the length of options plus 2979 * data in maxopd. 2980 * 2981 * NOTE that this routine is only called when we process an incoming 2982 * segment, for outgoing segments only tcp_mssopt is called. 2983 */ 2984 static void 2985 tcp_rmx_mss(struct tcpcb *tp, struct rtentry *rt, int offer) 2986 { 2987 struct ifnet *ifp; 2988 int mss; 2989 u_long bufsize; 2990 struct inpcb *inp = tp->t_inpcb; 2991 struct socket *so; 2992 #ifdef INET6 2993 boolean_t isipv6 = INP_ISIPV6(inp); 2994 size_t min_protoh = isipv6 ? 2995 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 2996 sizeof(struct tcpiphdr); 2997 #else 2998 const boolean_t isipv6 = FALSE; 2999 const size_t min_protoh = sizeof(struct tcpiphdr); 3000 #endif 3001 3002 if (rt == NULL) { 3003 tp->t_maxopd = tp->t_maxseg = 3004 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 3005 return; 3006 } 3007 ifp = rt->rt_ifp; 3008 so = inp->inp_socket; 3009 3010 /* 3011 * Offer == 0 means that there was no MSS on the SYN segment, 3012 * in this case we use either the interface mtu or tcp_mssdflt. 3013 * 3014 * An offer which is too large will be cut down later. 3015 */ 3016 if (offer == 0) { 3017 if (isipv6) { 3018 if (in6_localaddr(&inp->in6p_faddr)) 3019 offer = IN6_LINKMTU(rt->rt_ifp) - min_protoh; 3020 else 3021 offer = tcp_v6mssdflt; 3022 } else { 3023 if (in_localaddr(inp->inp_faddr)) 3024 offer = ifp->if_mtu - min_protoh; 3025 else 3026 offer = tcp_mssdflt; 3027 } 3028 } 3029 3030 /* 3031 * Prevent DoS attack with too small MSS. Round up 3032 * to at least minmss. 3033 * 3034 * Sanity check: make sure that maxopd will be large 3035 * enough to allow some data on segments even is the 3036 * all the option space is used (40bytes). Otherwise 3037 * funny things may happen in tcp_output. 3038 */ 3039 offer = max(offer, tcp_minmss); 3040 offer = max(offer, 64); 3041 3042 rt->rt_rmx.rmx_mssopt = offer; 3043 3044 /* 3045 * if there's an mtu associated with the route, use it 3046 * else, use the link mtu. Take the smaller of mss or offer 3047 * as our final mss. 3048 */ 3049 if (rt->rt_rmx.rmx_mtu) { 3050 mss = rt->rt_rmx.rmx_mtu; 3051 } else { 3052 if (isipv6) 3053 mss = IN6_LINKMTU(rt->rt_ifp); 3054 else 3055 mss = ifp->if_mtu; 3056 } 3057 mss -= min_protoh; 3058 mss = min(mss, offer); 3059 3060 /* 3061 * maxopd stores the maximum length of data AND options 3062 * in a segment; maxseg is the amount of data in a normal 3063 * segment. We need to store this value (maxopd) apart 3064 * from maxseg, because now every segment carries options 3065 * and thus we normally have somewhat less data in segments. 3066 */ 3067 tp->t_maxopd = mss; 3068 3069 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP && 3070 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3071 mss -= TCPOLEN_TSTAMP_APPA; 3072 3073 #if (MCLBYTES & (MCLBYTES - 1)) == 0 3074 if (mss > MCLBYTES) 3075 mss &= ~(MCLBYTES-1); 3076 #else 3077 if (mss > MCLBYTES) 3078 mss = mss / MCLBYTES * MCLBYTES; 3079 #endif 3080 /* 3081 * If there's a pipesize, change the socket buffer 3082 * to that size. Make the socket buffers an integral 3083 * number of mss units; if the mss is larger than 3084 * the socket buffer, decrease the mss. 3085 */ 3086 #ifdef RTV_SPIPE 3087 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 3088 #endif 3089 bufsize = so->so_snd.ssb_hiwat; 3090 if (bufsize < mss) 3091 mss = bufsize; 3092 else { 3093 bufsize = roundup(bufsize, mss); 3094 if (bufsize > sb_max) 3095 bufsize = sb_max; 3096 if (bufsize > so->so_snd.ssb_hiwat) 3097 ssb_reserve(&so->so_snd, bufsize, so, NULL); 3098 } 3099 tp->t_maxseg = mss; 3100 3101 #ifdef RTV_RPIPE 3102 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 3103 #endif 3104 bufsize = so->so_rcv.ssb_hiwat; 3105 if (bufsize > mss) { 3106 bufsize = roundup(bufsize, mss); 3107 if (bufsize > sb_max) 3108 bufsize = sb_max; 3109 if (bufsize > so->so_rcv.ssb_hiwat) { 3110 lwkt_gettoken(&so->so_rcv.ssb_token); 3111 ssb_reserve(&so->so_rcv, bufsize, so, NULL); 3112 lwkt_reltoken(&so->so_rcv.ssb_token); 3113 } 3114 } 3115 3116 /* 3117 * Set the slow-start flight size 3118 * 3119 * NOTE: t_maxseg must have been configured! 3120 */ 3121 tp->snd_cwnd = tcp_initial_window(tp); 3122 3123 if (rt->rt_rmx.rmx_ssthresh) { 3124 /* 3125 * There's some sort of gateway or interface 3126 * buffer limit on the path. Use this to set 3127 * the slow start threshhold, but set the 3128 * threshold to no less than 2*mss. 3129 */ 3130 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 3131 tcpstat.tcps_usedssthresh++; 3132 } 3133 } 3134 3135 static void 3136 tcp_rmx_rtt(struct tcpcb *tp, struct rtentry *rt) 3137 { 3138 int rtt; 3139 3140 if (rt == NULL) 3141 return; 3142 3143 /* 3144 * Check if there's an initial rtt or rttvar. Convert 3145 * from the route-table units to scaled multiples of 3146 * the slow timeout timer. 3147 */ 3148 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 3149 /* 3150 * XXX the lock bit for RTT indicates that the value 3151 * is also a minimum value; this is subject to time. 3152 */ 3153 if (rt->rt_rmx.rmx_locks & RTV_RTT) 3154 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 3155 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 3156 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3157 tcpstat.tcps_usedrtt++; 3158 if (rt->rt_rmx.rmx_rttvar) { 3159 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 3160 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 3161 tcpstat.tcps_usedrttvar++; 3162 } else { 3163 /* default variation is +- 1 rtt */ 3164 tp->t_rttvar = 3165 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3166 } 3167 TCPT_RANGESET(tp->t_rxtcur, 3168 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3169 tp->t_rttmin, TCPTV_REXMTMAX); 3170 } 3171 } 3172 3173 void 3174 tcp_rmx_init(struct tcpcb *tp, int offer) 3175 { 3176 struct inpcb *inp = tp->t_inpcb; 3177 #ifdef INET6 3178 boolean_t isipv6 = INP_ISIPV6(inp); 3179 #else 3180 const boolean_t isipv6 = FALSE; 3181 #endif 3182 struct rtentry *rt; 3183 3184 if (isipv6) 3185 rt = tcp_rtlookup6(&inp->inp_inc); 3186 else 3187 rt = tcp_rtlookup(&inp->inp_inc); 3188 3189 tcp_rmx_mss(tp, rt, offer); 3190 tcp_rmx_rtt(tp, rt); 3191 3192 if (rt != NULL && !tcp_ncr_linklocal && (rt->rt_flags & RTF_LLINFO)) { 3193 /* Don't enable NCR on link-local network. */ 3194 tp->t_flags &= ~TF_NCR; 3195 } 3196 } 3197 3198 /* 3199 * Determine the MSS option to send on an outgoing SYN. 3200 */ 3201 int 3202 tcp_mssopt(struct tcpcb *tp) 3203 { 3204 struct rtentry *rt; 3205 #ifdef INET6 3206 boolean_t isipv6 = INP_ISIPV6(tp->t_inpcb); 3207 int min_protoh = isipv6 ? 3208 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 3209 sizeof(struct tcpiphdr); 3210 #else 3211 const boolean_t isipv6 = FALSE; 3212 const size_t min_protoh = sizeof(struct tcpiphdr); 3213 #endif 3214 3215 if (isipv6) 3216 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 3217 else 3218 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 3219 if (rt == NULL) 3220 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 3221 3222 #ifdef INET6 3223 return ((isipv6 ? IN6_LINKMTU(rt->rt_ifp) : rt->rt_ifp->if_mtu) - 3224 min_protoh); 3225 #else 3226 return (rt->rt_ifp->if_mtu - min_protoh); 3227 #endif 3228 } 3229 3230 /* 3231 * When a partial ack arrives, force the retransmission of the 3232 * next unacknowledged segment. Do not exit Fast Recovery. 3233 * 3234 * Implement the Slow-but-Steady variant of NewReno by restarting the 3235 * the retransmission timer. Turn it off here so it can be restarted 3236 * later in tcp_output(). 3237 */ 3238 static void 3239 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked) 3240 { 3241 tcp_seq old_snd_nxt = tp->snd_nxt; 3242 u_long ocwnd = tp->snd_cwnd; 3243 3244 tcp_callout_stop(tp, tp->tt_rexmt); 3245 tp->t_rtttime = 0; 3246 tp->snd_nxt = th->th_ack; 3247 /* Set snd_cwnd to one segment beyond acknowledged offset. */ 3248 tp->snd_cwnd = tp->t_maxseg; 3249 tp->t_flags |= TF_ACKNOW; 3250 tcp_output(tp); 3251 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 3252 tp->snd_nxt = old_snd_nxt; 3253 /* partial window deflation */ 3254 if (ocwnd > acked) 3255 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg; 3256 else 3257 tp->snd_cwnd = tp->t_maxseg; 3258 } 3259 3260 /* 3261 * In contrast to the Slow-but-Steady NewReno variant, 3262 * we do not reset the retransmission timer for SACK retransmissions, 3263 * except when retransmitting snd_una. 3264 */ 3265 static void 3266 tcp_sack_rexmt(struct tcpcb *tp, boolean_t force) 3267 { 3268 tcp_seq old_snd_nxt = tp->snd_nxt; 3269 u_long ocwnd = tp->snd_cwnd; 3270 uint32_t pipe; 3271 int nseg = 0; /* consecutive new segments */ 3272 int nseg_rexmt = 0; /* retransmitted segments */ 3273 int maxrexmt = 0; 3274 3275 if (force) { 3276 uint32_t unsacked = tcp_sack_first_unsacked_len(tp); 3277 3278 /* 3279 * Try to fill the first hole in the receiver's 3280 * reassemble queue. 3281 */ 3282 maxrexmt = howmany(unsacked, tp->t_maxseg); 3283 if (maxrexmt > tcp_force_sackrxt) 3284 maxrexmt = tcp_force_sackrxt; 3285 } 3286 3287 tp->t_rtttime = 0; 3288 pipe = tcp_sack_compute_pipe(tp); 3289 while (((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg 3290 || (force && nseg_rexmt < maxrexmt && nseg == 0)) && 3291 (!tcp_do_smartsack || nseg < TCP_SACK_MAXBURST)) { 3292 tcp_seq old_snd_max, old_rexmt_high, nextrexmt; 3293 uint32_t sent, seglen; 3294 boolean_t rescue; 3295 int error; 3296 3297 old_rexmt_high = tp->rexmt_high; 3298 if (!tcp_sack_nextseg(tp, &nextrexmt, &seglen, &rescue)) { 3299 tp->rexmt_high = old_rexmt_high; 3300 break; 3301 } 3302 3303 /* 3304 * If the next tranmission is a rescue retranmission, 3305 * we check whether we have already sent some data 3306 * (either new segments or retransmitted segments) 3307 * into the the network or not. Since the idea of rescue 3308 * retransmission is to sustain ACK clock, as long as 3309 * some segments are in the network, ACK clock will be 3310 * kept ticking. 3311 */ 3312 if (rescue && (nseg_rexmt > 0 || nseg > 0)) { 3313 tp->rexmt_high = old_rexmt_high; 3314 break; 3315 } 3316 3317 if (nextrexmt == tp->snd_max) 3318 ++nseg; 3319 else 3320 ++nseg_rexmt; 3321 tp->snd_nxt = nextrexmt; 3322 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen; 3323 old_snd_max = tp->snd_max; 3324 if (nextrexmt == tp->snd_una) 3325 tcp_callout_stop(tp, tp->tt_rexmt); 3326 tp->t_flags |= TF_XMITNOW; 3327 error = tcp_output(tp); 3328 if (error != 0) { 3329 tp->rexmt_high = old_rexmt_high; 3330 break; 3331 } 3332 sent = tp->snd_nxt - nextrexmt; 3333 if (sent <= 0) { 3334 tp->rexmt_high = old_rexmt_high; 3335 break; 3336 } 3337 pipe += sent; 3338 tcpstat.tcps_sndsackpack++; 3339 tcpstat.tcps_sndsackbyte += sent; 3340 3341 if (rescue) { 3342 tcpstat.tcps_sackrescue++; 3343 tp->rexmt_rescue = tp->snd_nxt; 3344 tp->sack_flags |= TSACK_F_SACKRESCUED; 3345 break; 3346 } 3347 if (SEQ_LT(nextrexmt, old_snd_max) && 3348 SEQ_LT(tp->rexmt_high, tp->snd_nxt)) { 3349 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max); 3350 if (tcp_aggressive_rescuesack && 3351 (tp->sack_flags & TSACK_F_SACKRESCUED) && 3352 SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) { 3353 /* Drag RescueRxt along with HighRxt */ 3354 tp->rexmt_rescue = tp->rexmt_high; 3355 } 3356 } 3357 } 3358 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 3359 tp->snd_nxt = old_snd_nxt; 3360 tp->snd_cwnd = ocwnd; 3361 } 3362 3363 /* 3364 * Return TRUE, if some new segments are sent 3365 */ 3366 static boolean_t 3367 tcp_sack_limitedxmit(struct tcpcb *tp) 3368 { 3369 tcp_seq oldsndnxt = tp->snd_nxt; 3370 tcp_seq oldsndmax = tp->snd_max; 3371 u_long ocwnd = tp->snd_cwnd; 3372 uint32_t pipe, sent; 3373 boolean_t ret = FALSE; 3374 tcp_seq_diff_t cwnd_left; 3375 tcp_seq next; 3376 3377 tp->rexmt_high = tp->snd_una - 1; 3378 pipe = tcp_sack_compute_pipe(tp); 3379 cwnd_left = (tcp_seq_diff_t)(ocwnd - pipe); 3380 if (cwnd_left < (tcp_seq_diff_t)tp->t_maxseg) 3381 return FALSE; 3382 3383 if (tcp_do_smartsack) 3384 cwnd_left = ulmin(cwnd_left, tp->t_maxseg * TCP_SACK_MAXBURST); 3385 3386 next = tp->snd_nxt = tp->snd_max; 3387 tp->snd_cwnd = tp->snd_nxt - tp->snd_una + 3388 rounddown(cwnd_left, tp->t_maxseg); 3389 3390 tp->t_flags |= TF_XMITNOW; 3391 tcp_output(tp); 3392 3393 sent = tp->snd_nxt - next; 3394 if (sent > 0) { 3395 tcpstat.tcps_sndlimited += howmany(sent, tp->t_maxseg); 3396 ret = TRUE; 3397 } 3398 3399 if (SEQ_LT(oldsndnxt, oldsndmax)) { 3400 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una), 3401 ("snd_una moved in other threads")); 3402 tp->snd_nxt = oldsndnxt; 3403 } 3404 tp->snd_cwnd = ocwnd; 3405 3406 if (ret && TCP_DO_NCR(tp)) 3407 tcp_ncr_update_rxtthresh(tp); 3408 3409 return ret; 3410 } 3411 3412 /* 3413 * Reset idle time and keep-alive timer, typically called when a valid 3414 * tcp packet is received but may also be called when FASTKEEP is set 3415 * to prevent the previous long-timeout from calculating to a drop. 3416 * 3417 * Only update t_rcvtime for non-SYN packets. 3418 * 3419 * Handle the case where one side thinks the connection is established 3420 * but the other side has, say, rebooted without cleaning out the 3421 * connection. The SYNs could be construed as an attack and wind 3422 * up ignored, but in case it isn't an attack we can validate the 3423 * connection by forcing a keepalive. 3424 */ 3425 void 3426 tcp_timer_keep_activity(struct tcpcb *tp, int thflags) 3427 { 3428 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3429 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) { 3430 tp->t_flags |= TF_KEEPALIVE; 3431 tcp_callout_reset(tp, tp->tt_keep, hz / 2, 3432 tcp_timer_keep); 3433 } else { 3434 tp->t_rcvtime = ticks; 3435 tp->t_flags &= ~TF_KEEPALIVE; 3436 tcp_callout_reset(tp, tp->tt_keep, 3437 tp->t_keepidle, 3438 tcp_timer_keep); 3439 } 3440 } 3441 } 3442 3443 static int 3444 tcp_rmx_msl(const struct tcpcb *tp) 3445 { 3446 struct rtentry *rt; 3447 struct inpcb *inp = tp->t_inpcb; 3448 int msl; 3449 #ifdef INET6 3450 boolean_t isipv6 = INP_ISIPV6(inp); 3451 #else 3452 const boolean_t isipv6 = FALSE; 3453 #endif 3454 3455 if (isipv6) 3456 rt = tcp_rtlookup6(&inp->inp_inc); 3457 else 3458 rt = tcp_rtlookup(&inp->inp_inc); 3459 if (rt == NULL || rt->rt_rmx.rmx_msl == 0) 3460 return tcp_msl; 3461 3462 msl = (rt->rt_rmx.rmx_msl * hz) / 1000; 3463 if (msl == 0) 3464 msl = 1; 3465 3466 return msl; 3467 } 3468 3469 static void 3470 tcp_established(struct tcpcb *tp) 3471 { 3472 TCP_STATE_CHANGE(tp, TCPS_ESTABLISHED); 3473 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle, tcp_timer_keep); 3474 3475 if (tp->t_rxtsyn > 0) { 3476 /* 3477 * RFC6298: 3478 * "If the timer expires awaiting the ACK of a SYN segment 3479 * and the TCP implementation is using an RTO less than 3 3480 * seconds, the RTO MUST be re-initialized to 3 seconds 3481 * when data transmission begins" 3482 */ 3483 if (tp->t_rxtcur < TCPTV_RTOBASE3) 3484 tp->t_rxtcur = TCPTV_RTOBASE3; 3485 } 3486 } 3487 3488 /* 3489 * Returns TRUE, if the ACK should be dropped 3490 */ 3491 static boolean_t 3492 tcp_recv_dupack(struct tcpcb *tp, tcp_seq th_ack, u_int to_flags) 3493 { 3494 boolean_t fast_sack_rexmt = TRUE; 3495 3496 tcpstat.tcps_rcvdupack++; 3497 3498 /* 3499 * We have outstanding data (other than a window probe), 3500 * this is a completely duplicate ack (ie, window info 3501 * didn't change), the ack is the biggest we've seen and 3502 * we've seen exactly our rexmt threshhold of them, so 3503 * assume a packet has been dropped and retransmit it. 3504 * Kludge snd_nxt & the congestion window so we send only 3505 * this one packet. 3506 */ 3507 if (IN_FASTRECOVERY(tp)) { 3508 if (TCP_DO_SACK(tp)) { 3509 boolean_t force = FALSE; 3510 3511 if (tp->snd_una == tp->rexmt_high && 3512 (to_flags & (TOF_SACK | TOF_SACK_REDUNDANT)) == 3513 TOF_SACK) { 3514 /* 3515 * New segments got SACKed and 3516 * no retransmit yet. 3517 */ 3518 force = TRUE; 3519 } 3520 3521 /* No artifical cwnd inflation. */ 3522 tcp_sack_rexmt(tp, force); 3523 } else { 3524 /* 3525 * Dup acks mean that packets have left 3526 * the network (they're now cached at the 3527 * receiver) so bump cwnd by the amount in 3528 * the receiver to keep a constant cwnd 3529 * packets in the network. 3530 */ 3531 tp->snd_cwnd += tp->t_maxseg; 3532 tcp_output(tp); 3533 } 3534 return TRUE; 3535 } else if (SEQ_LT(th_ack, tp->snd_recover)) { 3536 tp->t_dupacks = 0; 3537 return FALSE; 3538 } else if (tcp_ignore_redun_dsack && TCP_DO_SACK(tp) && 3539 (to_flags & (TOF_DSACK | TOF_SACK_REDUNDANT)) == 3540 (TOF_DSACK | TOF_SACK_REDUNDANT)) { 3541 /* 3542 * If the ACK carries DSACK and other SACK blocks 3543 * carry information that we have already known, 3544 * don't count this ACK as duplicate ACK. This 3545 * prevents spurious early retransmit and fast 3546 * retransmit. This also meets the requirement of 3547 * RFC3042 that new segments should not be sent if 3548 * the SACK blocks do not contain new information 3549 * (XXX we actually loosen the requirment that only 3550 * DSACK is checked here). 3551 * 3552 * This kind of ACKs are usually sent after spurious 3553 * retransmit. 3554 */ 3555 /* Do nothing; don't change t_dupacks */ 3556 return TRUE; 3557 } else if (tp->t_dupacks == 0 && TCP_DO_NCR(tp)) { 3558 tcp_ncr_update_rxtthresh(tp); 3559 } 3560 3561 if (++tp->t_dupacks == tp->t_rxtthresh) { 3562 tcp_seq old_snd_nxt; 3563 u_int win; 3564 3565 fastretransmit: 3566 if (tcp_do_eifel_detect && (tp->t_flags & TF_RCVD_TSTMP)) { 3567 tcp_save_congestion_state(tp); 3568 tp->rxt_flags |= TRXT_F_FASTREXMT; 3569 } 3570 /* 3571 * We know we're losing at the current window size, 3572 * so do congestion avoidance: set ssthresh to half 3573 * the current window and pull our congestion window 3574 * back to the new ssthresh. 3575 */ 3576 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg; 3577 if (win < 2) 3578 win = 2; 3579 tp->snd_ssthresh = win * tp->t_maxseg; 3580 ENTER_FASTRECOVERY(tp); 3581 tp->snd_recover = tp->snd_max; 3582 tcp_callout_stop(tp, tp->tt_rexmt); 3583 tp->t_rtttime = 0; 3584 old_snd_nxt = tp->snd_nxt; 3585 tp->snd_nxt = th_ack; 3586 if (TCP_DO_SACK(tp)) { 3587 uint32_t rxtlen; 3588 3589 rxtlen = tcp_sack_first_unsacked_len(tp); 3590 if (rxtlen > tp->t_maxseg) 3591 rxtlen = tp->t_maxseg; 3592 tp->snd_cwnd = rxtlen; 3593 } else { 3594 tp->snd_cwnd = tp->t_maxseg; 3595 } 3596 tcp_output(tp); 3597 ++tcpstat.tcps_sndfastrexmit; 3598 tp->snd_cwnd = tp->snd_ssthresh; 3599 tp->rexmt_high = tp->snd_nxt; 3600 tp->sack_flags &= ~TSACK_F_SACKRESCUED; 3601 if (SEQ_GT(old_snd_nxt, tp->snd_nxt)) 3602 tp->snd_nxt = old_snd_nxt; 3603 KASSERT(tp->snd_limited <= 2, ("tp->snd_limited too big")); 3604 if (TCP_DO_SACK(tp)) { 3605 if (fast_sack_rexmt) 3606 tcp_sack_rexmt(tp, FALSE); 3607 } else { 3608 tp->snd_cwnd += tp->t_maxseg * 3609 (tp->t_dupacks - tp->snd_limited); 3610 } 3611 } else if ((tcp_do_rfc6675 && TCP_DO_SACK(tp)) || TCP_DO_NCR(tp)) { 3612 /* 3613 * The RFC6675 recommends to reduce the byte threshold, 3614 * and enter fast retransmit if IsLost(snd_una). However, 3615 * if we use IsLost(snd_una) based fast retransmit here, 3616 * segments reordering will cause spurious retransmit. So 3617 * we defer the IsLost(snd_una) based fast retransmit until 3618 * the extended limited transmit can't send any segments and 3619 * early retransmit can't be done. 3620 */ 3621 if (tcp_rfc6675_rxt && tcp_do_rfc6675 && 3622 tcp_sack_islost(&tp->scb, tp->snd_una)) 3623 goto fastretransmit; 3624 3625 if (tcp_do_limitedtransmit || TCP_DO_NCR(tp)) { 3626 if (!tcp_sack_limitedxmit(tp)) { 3627 /* outstanding data */ 3628 uint32_t ownd = tp->snd_max - tp->snd_una; 3629 3630 if (need_early_retransmit(tp, ownd)) { 3631 ++tcpstat.tcps_sndearlyrexmit; 3632 tp->rxt_flags |= TRXT_F_EARLYREXMT; 3633 goto fastretransmit; 3634 } else if (tcp_do_rfc6675 && 3635 tcp_sack_islost(&tp->scb, tp->snd_una)) { 3636 fast_sack_rexmt = FALSE; 3637 goto fastretransmit; 3638 } 3639 } 3640 } 3641 } else if (tcp_do_limitedtransmit) { 3642 u_long oldcwnd = tp->snd_cwnd; 3643 tcp_seq oldsndmax = tp->snd_max; 3644 tcp_seq oldsndnxt = tp->snd_nxt; 3645 /* outstanding data */ 3646 uint32_t ownd = tp->snd_max - tp->snd_una; 3647 u_int sent; 3648 3649 KASSERT(tp->t_dupacks == 1 || tp->t_dupacks == 2, 3650 ("dupacks not 1 or 2")); 3651 if (tp->t_dupacks == 1) 3652 tp->snd_limited = 0; 3653 tp->snd_nxt = tp->snd_max; 3654 tp->snd_cwnd = ownd + 3655 (tp->t_dupacks - tp->snd_limited) * tp->t_maxseg; 3656 tp->t_flags |= TF_XMITNOW; 3657 tcp_output(tp); 3658 3659 if (SEQ_LT(oldsndnxt, oldsndmax)) { 3660 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una), 3661 ("snd_una moved in other threads")); 3662 tp->snd_nxt = oldsndnxt; 3663 } 3664 tp->snd_cwnd = oldcwnd; 3665 sent = tp->snd_max - oldsndmax; 3666 if (sent > tp->t_maxseg) { 3667 KASSERT((tp->t_dupacks == 2 && tp->snd_limited == 0) || 3668 (sent == tp->t_maxseg + 1 && 3669 (tp->t_flags & TF_SENTFIN)), 3670 ("sent too much")); 3671 KASSERT(sent <= tp->t_maxseg * 2, 3672 ("sent too many segments")); 3673 tp->snd_limited = 2; 3674 tcpstat.tcps_sndlimited += 2; 3675 } else if (sent > 0) { 3676 ++tp->snd_limited; 3677 ++tcpstat.tcps_sndlimited; 3678 } else if (need_early_retransmit(tp, ownd)) { 3679 ++tcpstat.tcps_sndearlyrexmit; 3680 tp->rxt_flags |= TRXT_F_EARLYREXMT; 3681 goto fastretransmit; 3682 } 3683 } 3684 return TRUE; 3685 } 3686