tcp_input.c revision 30005
155714Skris/* 2296341Sdelphij * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3296341Sdelphij * The Regents of the University of California. All rights reserved. 4296341Sdelphij * 555714Skris * Redistribution and use in source and binary forms, with or without 655714Skris * modification, are permitted provided that the following conditions 755714Skris * are met: 855714Skris * 1. Redistributions of source code must retain the above copyright 955714Skris * notice, this list of conditions and the following disclaimer. 1055714Skris * 2. Redistributions in binary form must reproduce the above copyright 1155714Skris * notice, this list of conditions and the following disclaimer in the 1255714Skris * documentation and/or other materials provided with the distribution. 1355714Skris * 3. All advertising materials mentioning features or use of this software 14296341Sdelphij * must display the following acknowledgement: 1555714Skris * This product includes software developed by the University of 1655714Skris * California, Berkeley and its contributors. 1755714Skris * 4. Neither the name of the University nor the names of its contributors 1855714Skris * may be used to endorse or promote products derived from this software 1955714Skris * without specific prior written permission. 2055714Skris * 2155714Skris * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 2255714Skris * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2355714Skris * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2455714Skris * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 2555714Skris * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2655714Skris * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2755714Skris * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2855714Skris * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2955714Skris * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 3055714Skris * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3155714Skris * SUCH DAMAGE. 3255714Skris * 3355714Skris * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 3455714Skris * $Id: tcp_input.c,v 1.61 1997/09/16 18:36:05 joerg Exp $ 3555714Skris */ 3655714Skris 3755714Skris#include "opt_tcpdebug.h" 3855714Skris 3955714Skris#ifndef TUBA_INCLUDE 4055714Skris#include <sys/param.h> 4155714Skris#include <sys/queue.h> 4255714Skris#include <sys/systm.h> 4355714Skris#include <sys/kernel.h> 4455714Skris#include <sys/sysctl.h> 4555714Skris#include <sys/malloc.h> 4655714Skris#include <sys/mbuf.h> 4755714Skris#include <sys/proc.h> /* for proc0 declaration */ 4855714Skris#include <sys/protosw.h> 4955714Skris#include <sys/socket.h> 5055714Skris#include <sys/socketvar.h> 5155714Skris#include <sys/errno.h> 5255714Skris#include <sys/syslog.h> 5355714Skris 5455714Skris#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 5555714Skris 5655714Skris#include <net/if.h> 5755714Skris#include <net/route.h> 5855714Skris 5955714Skris#include <netinet/in.h> 60109998Smarkm#include <netinet/in_systm.h> 61296341Sdelphij#include <netinet/ip.h> 62296341Sdelphij#include <netinet/in_pcb.h> 63296341Sdelphij#include <netinet/ip_var.h> 64296341Sdelphij#include <netinet/tcp.h> 65296341Sdelphij#include <netinet/tcp_fsm.h> 66296341Sdelphij#include <netinet/tcp_seq.h> 6755714Skris#include <netinet/tcp_timer.h> 6855714Skris#include <netinet/tcp_var.h> 69160814Ssimon#include <netinet/tcpip.h> 70296341Sdelphij#ifdef TCPDEBUG 7155714Skris#include <netinet/tcp_debug.h> 72296341Sdelphijstatic struct tcpiphdr tcp_saveti; 73296341Sdelphij#endif 74296341Sdelphij 75296341Sdelphijstatic int tcprexmtthresh = 3; 76296341Sdelphijtcp_seq tcp_iss; 77109998Smarkmtcp_cc tcp_ccgen; 78296341Sdelphij 79296341Sdelphijstruct tcpstat tcpstat; 80296341SdelphijSYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, 81296341Sdelphij CTLFLAG_RD, &tcpstat , tcpstat, ""); 82160814Ssimon 83296341Sdelphijstatic int log_in_vain = 0; 84296341SdelphijSYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 85296341Sdelphij &log_in_vain, 0, ""); 86296341Sdelphij 87296341Sdelphiju_long tcp_now; 88296341Sdelphijstruct inpcbhead tcb; 89296341Sdelphijstruct inpcbinfo tcbinfo; 90296341Sdelphij 91296341Sdelphijstatic void tcp_dooptions __P((struct tcpcb *, 92296341Sdelphij u_char *, int, struct tcpiphdr *, struct tcpopt *)); 93296341Sdelphijstatic void tcp_pulloutofband __P((struct socket *, 94296341Sdelphij struct tcpiphdr *, struct mbuf *)); 95296341Sdelphijstatic int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 96296341Sdelphijstatic void tcp_xmit_timer __P((struct tcpcb *, int)); 97296341Sdelphij 98296341Sdelphij#endif /* TUBA_INCLUDE */ 99296341Sdelphij 100296341Sdelphij/* 101296341Sdelphij * Insert segment ti into reassembly queue of tcp with 102296341Sdelphij * control block tp. Return TH_FIN if reassembly now includes 103296341Sdelphij * a segment with FIN. The macro form does the common case inline 104296341Sdelphij * (segment is the next to be received on an established connection, 105296341Sdelphij * and the queue is empty), avoiding linkage into and removal 106296341Sdelphij * from the queue and repetition of various conversions. 107296341Sdelphij * Set DELACK for segments received in order, but ack immediately 108296341Sdelphij * when segments are out of order (so fast retransmit can work). 109296341Sdelphij */ 110296341Sdelphij#ifdef TCP_ACK_HACK 11155714Skris#define TCP_REASS(tp, ti, m, so, flags) { \ 11255714Skris if ((ti)->ti_seq == (tp)->rcv_nxt && \ 11355714Skris (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 114160814Ssimon (tp)->t_state == TCPS_ESTABLISHED) { \ 11555714Skris if (ti->ti_flags & TH_PUSH) \ 116296341Sdelphij tp->t_flags |= TF_ACKNOW; \ 117296341Sdelphij else \ 118296341Sdelphij tp->t_flags |= TF_DELACK; \ 119296341Sdelphij (tp)->rcv_nxt += (ti)->ti_len; \ 120296341Sdelphij flags = (ti)->ti_flags & TH_FIN; \ 121296341Sdelphij tcpstat.tcps_rcvpack++;\ 122296341Sdelphij tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 123296341Sdelphij sbappend(&(so)->so_rcv, (m)); \ 124296341Sdelphij sorwakeup(so); \ 125296341Sdelphij } else { \ 126296341Sdelphij (flags) = tcp_reass((tp), (ti), (m)); \ 127284295Sdelphij tp->t_flags |= TF_ACKNOW; \ 128296341Sdelphij } \ 129296341Sdelphij} 13055714Skris#else 13155714Skris#define TCP_REASS(tp, ti, m, so, flags) { \ 13255714Skris if ((ti)->ti_seq == (tp)->rcv_nxt && \ 13355714Skris (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 134160814Ssimon (tp)->t_state == TCPS_ESTABLISHED) { \ 135296341Sdelphij tp->t_flags |= TF_DELACK; \ 136296341Sdelphij (tp)->rcv_nxt += (ti)->ti_len; \ 13755714Skris flags = (ti)->ti_flags & TH_FIN; \ 138296341Sdelphij tcpstat.tcps_rcvpack++;\ 139296341Sdelphij tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 14055714Skris sbappend(&(so)->so_rcv, (m)); \ 141296341Sdelphij sorwakeup(so); \ 142296341Sdelphij } else { \ 143296341Sdelphij (flags) = tcp_reass((tp), (ti), (m)); \ 144296341Sdelphij tp->t_flags |= TF_ACKNOW; \ 145296341Sdelphij } \ 146296341Sdelphij} 147296341Sdelphij#endif 148296341Sdelphij#ifndef TUBA_INCLUDE 149296341Sdelphij 150296341Sdelphijstatic int 151296341Sdelphijtcp_reass(tp, ti, m) 152296341Sdelphij register struct tcpcb *tp; 153296341Sdelphij register struct tcpiphdr *ti; 154296341Sdelphij struct mbuf *m; 155296341Sdelphij{ 15655714Skris register struct tcpiphdr *q; 15755714Skris struct socket *so = tp->t_inpcb->inp_socket; 15855714Skris int flags; 159160814Ssimon 160296341Sdelphij /* 16155714Skris * Call with ti==0 after become established to 162296341Sdelphij * force pre-ESTABLISHED data up to user socket. 163296341Sdelphij */ 164296341Sdelphij if (ti == 0) 165296341Sdelphij goto present; 166296341Sdelphij 167296341Sdelphij /* 168296341Sdelphij * Find a segment which begins after this one does. 169296341Sdelphij */ 170296341Sdelphij for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 171296341Sdelphij q = (struct tcpiphdr *)q->ti_next) 172296341Sdelphij if (SEQ_GT(q->ti_seq, ti->ti_seq)) 173296341Sdelphij break; 174296341Sdelphij 175296341Sdelphij /* 176296341Sdelphij * If there is a preceding segment, it may provide some of 177296341Sdelphij * our data already. If so, drop the data from the incoming 178296341Sdelphij * segment. If it provides all of our data, drop us. 179296341Sdelphij */ 180296341Sdelphij if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 181296341Sdelphij register int i; 182296341Sdelphij q = (struct tcpiphdr *)q->ti_prev; 183296341Sdelphij /* conversion to int (in i) handles seq wraparound */ 184296341Sdelphij i = q->ti_seq + q->ti_len - ti->ti_seq; 185296341Sdelphij if (i > 0) { 186296341Sdelphij if (i >= ti->ti_len) { 187296341Sdelphij tcpstat.tcps_rcvduppack++; 188296341Sdelphij tcpstat.tcps_rcvdupbyte += ti->ti_len; 189296341Sdelphij m_freem(m); 190296341Sdelphij /* 191296341Sdelphij * Try to present any queued data 192296341Sdelphij * at the left window edge to the user. 193296341Sdelphij * This is needed after the 3-WHS 19455714Skris * completes. 19555714Skris */ 196 goto present; /* ??? */ 197 } 198 m_adj(m, i); 199 ti->ti_len -= i; 200 ti->ti_seq += i; 201 } 202 q = (struct tcpiphdr *)(q->ti_next); 203 } 204 tcpstat.tcps_rcvoopack++; 205 tcpstat.tcps_rcvoobyte += ti->ti_len; 206 REASS_MBUF(ti) = m; /* XXX */ 207 208 /* 209 * While we overlap succeeding segments trim them or, 210 * if they are completely covered, dequeue them. 211 */ 212 while (q != (struct tcpiphdr *)tp) { 213 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 214 if (i <= 0) 215 break; 216 if (i < q->ti_len) { 217 q->ti_seq += i; 218 q->ti_len -= i; 219 m_adj(REASS_MBUF(q), i); 220 break; 221 } 222 q = (struct tcpiphdr *)q->ti_next; 223 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 224 remque(q->ti_prev); 225 m_freem(m); 226 } 227 228 /* 229 * Stick new segment in its place. 230 */ 231 insque(ti, q->ti_prev); 232 233present: 234 /* 235 * Present data to user, advancing rcv_nxt through 236 * completed sequence space. 237 */ 238 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 239 return (0); 240 ti = tp->seg_next; 241 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 242 return (0); 243 do { 244 tp->rcv_nxt += ti->ti_len; 245 flags = ti->ti_flags & TH_FIN; 246 remque(ti); 247 m = REASS_MBUF(ti); 248 ti = (struct tcpiphdr *)ti->ti_next; 249 if (so->so_state & SS_CANTRCVMORE) 250 m_freem(m); 251 else 252 sbappend(&so->so_rcv, m); 253 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 254 sorwakeup(so); 255 return (flags); 256} 257 258/* 259 * TCP input routine, follows pages 65-76 of the 260 * protocol specification dated September, 1981 very closely. 261 */ 262void 263tcp_input(m, iphlen) 264 register struct mbuf *m; 265 int iphlen; 266{ 267 register struct tcpiphdr *ti; 268 register struct inpcb *inp; 269 u_char *optp = NULL; 270 int optlen = 0; 271 int len, tlen, off; 272 register struct tcpcb *tp = 0; 273 register int tiflags; 274 struct socket *so = 0; 275 int todrop, acked, ourfinisacked, needoutput = 0; 276 struct in_addr laddr; 277 int dropsocket = 0; 278 int iss = 0; 279 u_long tiwin; 280 struct tcpopt to; /* options in this segment */ 281 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 282 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 283#ifdef TCPDEBUG 284 short ostate = 0; 285#endif 286 287 bzero((char *)&to, sizeof(to)); 288 289 tcpstat.tcps_rcvtotal++; 290 /* 291 * Get IP and TCP header together in first mbuf. 292 * Note: IP leaves IP header in first mbuf. 293 */ 294 ti = mtod(m, struct tcpiphdr *); 295 if (iphlen > sizeof (struct ip)) 296 ip_stripoptions(m, (struct mbuf *)0); 297 if (m->m_len < sizeof (struct tcpiphdr)) { 298 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 299 tcpstat.tcps_rcvshort++; 300 return; 301 } 302 ti = mtod(m, struct tcpiphdr *); 303 } 304 305 /* 306 * Checksum extended TCP header and data. 307 */ 308 tlen = ((struct ip *)ti)->ip_len; 309 len = sizeof (struct ip) + tlen; 310 ti->ti_next = ti->ti_prev = 0; 311 ti->ti_x1 = 0; 312 ti->ti_len = (u_short)tlen; 313 HTONS(ti->ti_len); 314 ti->ti_sum = in_cksum(m, len); 315 if (ti->ti_sum) { 316 tcpstat.tcps_rcvbadsum++; 317 goto drop; 318 } 319#endif /* TUBA_INCLUDE */ 320 321 /* 322 * Check that TCP offset makes sense, 323 * pull out TCP options and adjust length. XXX 324 */ 325 off = ti->ti_off << 2; 326 if (off < sizeof (struct tcphdr) || off > tlen) { 327 tcpstat.tcps_rcvbadoff++; 328 goto drop; 329 } 330 tlen -= off; 331 ti->ti_len = tlen; 332 if (off > sizeof (struct tcphdr)) { 333 if (m->m_len < sizeof(struct ip) + off) { 334 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 335 tcpstat.tcps_rcvshort++; 336 return; 337 } 338 ti = mtod(m, struct tcpiphdr *); 339 } 340 optlen = off - sizeof (struct tcphdr); 341 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 342 } 343 tiflags = ti->ti_flags; 344 345 /* 346 * Convert TCP protocol specific fields to host format. 347 */ 348 NTOHL(ti->ti_seq); 349 NTOHL(ti->ti_ack); 350 NTOHS(ti->ti_win); 351 NTOHS(ti->ti_urp); 352 353 /* 354 * Drop TCP, IP headers and TCP options. 355 */ 356 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 357 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 358 359 /* 360 * Locate pcb for segment. 361 */ 362findpcb: 363 inp = in_pcblookuphash(&tcbinfo, ti->ti_src, ti->ti_sport, 364 ti->ti_dst, ti->ti_dport, 1); 365 366 /* 367 * If the state is CLOSED (i.e., TCB does not exist) then 368 * all data in the incoming segment is discarded. 369 * If the TCB exists but is in CLOSED state, it is embryonic, 370 * but should either do a listen or a connect soon. 371 */ 372 if (inp == NULL) { 373 if (log_in_vain && tiflags & TH_SYN) { 374 char buf[4*sizeof "123"]; 375 376 strcpy(buf, inet_ntoa(ti->ti_dst)); 377 log(LOG_INFO, "Connection attempt to TCP %s:%d" 378 " from %s:%d\n", 379 buf, ntohs(ti->ti_dport), 380 inet_ntoa(ti->ti_src), ntohs(ti->ti_sport)); 381 } 382 goto dropwithreset; 383 } 384 tp = intotcpcb(inp); 385 if (tp == 0) 386 goto dropwithreset; 387 if (tp->t_state == TCPS_CLOSED) 388 goto drop; 389 390 /* Unscale the window into a 32-bit value. */ 391 if ((tiflags & TH_SYN) == 0) 392 tiwin = ti->ti_win << tp->snd_scale; 393 else 394 tiwin = ti->ti_win; 395 396 so = inp->inp_socket; 397 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 398#ifdef TCPDEBUG 399 if (so->so_options & SO_DEBUG) { 400 ostate = tp->t_state; 401 tcp_saveti = *ti; 402 } 403#endif 404 if (so->so_options & SO_ACCEPTCONN) { 405 register struct tcpcb *tp0 = tp; 406 struct socket *so2; 407 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 408 /* 409 * Note: dropwithreset makes sure we don't 410 * send a RST in response to a RST. 411 */ 412 if (tiflags & TH_ACK) { 413 tcpstat.tcps_badsyn++; 414 goto dropwithreset; 415 } 416 goto drop; 417 } 418 so2 = sonewconn(so, 0); 419 if (so2 == 0) { 420 tcpstat.tcps_listendrop++; 421 so2 = sodropablereq(so); 422 if (so2) { 423 tcp_drop(sototcpcb(so2), ETIMEDOUT); 424 so2 = sonewconn(so, 0); 425 } 426 if (!so2) 427 goto drop; 428 } 429 so = so2; 430 /* 431 * This is ugly, but .... 432 * 433 * Mark socket as temporary until we're 434 * committed to keeping it. The code at 435 * ``drop'' and ``dropwithreset'' check the 436 * flag dropsocket to see if the temporary 437 * socket created here should be discarded. 438 * We mark the socket as discardable until 439 * we're committed to it below in TCPS_LISTEN. 440 */ 441 dropsocket++; 442 inp = (struct inpcb *)so->so_pcb; 443 inp->inp_laddr = ti->ti_dst; 444 inp->inp_lport = ti->ti_dport; 445 in_pcbrehash(inp); 446#if BSD>=43 447 inp->inp_options = ip_srcroute(); 448#endif 449 tp = intotcpcb(inp); 450 tp->t_state = TCPS_LISTEN; 451 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 452 453 /* Compute proper scaling value from buffer space */ 454 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 455 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 456 tp->request_r_scale++; 457 } 458 } 459 460 /* 461 * Segment received on connection. 462 * Reset idle time and keep-alive timer. 463 */ 464 tp->t_idle = 0; 465 if (TCPS_HAVEESTABLISHED(tp->t_state)) 466 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 467 468 /* 469 * Process options if not in LISTEN state, 470 * else do it below (after getting remote address). 471 */ 472 if (tp->t_state != TCPS_LISTEN) 473 tcp_dooptions(tp, optp, optlen, ti, &to); 474 475 /* 476 * Header prediction: check for the two common cases 477 * of a uni-directional data xfer. If the packet has 478 * no control flags, is in-sequence, the window didn't 479 * change and we're not retransmitting, it's a 480 * candidate. If the length is zero and the ack moved 481 * forward, we're the sender side of the xfer. Just 482 * free the data acked & wake any higher level process 483 * that was blocked waiting for space. If the length 484 * is non-zero and the ack didn't move, we're the 485 * receiver side. If we're getting packets in-order 486 * (the reassembly queue is empty), add the data to 487 * the socket buffer and note that we need a delayed ack. 488 * Make sure that the hidden state-flags are also off. 489 * Since we check for TCPS_ESTABLISHED above, it can only 490 * be TH_NEEDSYN. 491 */ 492 if (tp->t_state == TCPS_ESTABLISHED && 493 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 494 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 495 ((to.to_flag & TOF_TS) == 0 || 496 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 497 /* 498 * Using the CC option is compulsory if once started: 499 * the segment is OK if no T/TCP was negotiated or 500 * if the segment has a CC option equal to CCrecv 501 */ 502 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 503 (to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv) && 504 ti->ti_seq == tp->rcv_nxt && 505 tiwin && tiwin == tp->snd_wnd && 506 tp->snd_nxt == tp->snd_max) { 507 508 /* 509 * If last ACK falls within this segment's sequence numbers, 510 * record the timestamp. 511 * NOTE that the test is modified according to the latest 512 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 513 */ 514 if ((to.to_flag & TOF_TS) != 0 && 515 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 516 tp->ts_recent_age = tcp_now; 517 tp->ts_recent = to.to_tsval; 518 } 519 520 if (ti->ti_len == 0) { 521 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 522 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 523 tp->snd_cwnd >= tp->snd_wnd && 524 tp->t_dupacks < tcprexmtthresh) { 525 /* 526 * this is a pure ack for outstanding data. 527 */ 528 ++tcpstat.tcps_predack; 529 if ((to.to_flag & TOF_TS) != 0) 530 tcp_xmit_timer(tp, 531 tcp_now - to.to_tsecr + 1); 532 else if (tp->t_rtt && 533 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 534 tcp_xmit_timer(tp, tp->t_rtt); 535 acked = ti->ti_ack - tp->snd_una; 536 tcpstat.tcps_rcvackpack++; 537 tcpstat.tcps_rcvackbyte += acked; 538 sbdrop(&so->so_snd, acked); 539 tp->snd_una = ti->ti_ack; 540 m_freem(m); 541 542 /* 543 * If all outstanding data are acked, stop 544 * retransmit timer, otherwise restart timer 545 * using current (possibly backed-off) value. 546 * If process is waiting for space, 547 * wakeup/selwakeup/signal. If data 548 * are ready to send, let tcp_output 549 * decide between more output or persist. 550 */ 551 if (tp->snd_una == tp->snd_max) 552 tp->t_timer[TCPT_REXMT] = 0; 553 else if (tp->t_timer[TCPT_PERSIST] == 0) 554 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 555 556 if (so->so_snd.sb_flags & SB_NOTIFY) 557 sowwakeup(so); 558 if (so->so_snd.sb_cc) 559 (void) tcp_output(tp); 560 return; 561 } 562 } else if (ti->ti_ack == tp->snd_una && 563 tp->seg_next == (struct tcpiphdr *)tp && 564 ti->ti_len <= sbspace(&so->so_rcv)) { 565 /* 566 * this is a pure, in-sequence data packet 567 * with nothing on the reassembly queue and 568 * we have enough buffer space to take it. 569 */ 570 ++tcpstat.tcps_preddat; 571 tp->rcv_nxt += ti->ti_len; 572 tcpstat.tcps_rcvpack++; 573 tcpstat.tcps_rcvbyte += ti->ti_len; 574 /* 575 * Add data to socket buffer. 576 */ 577 sbappend(&so->so_rcv, m); 578 sorwakeup(so); 579#ifdef TCP_ACK_HACK 580 /* 581 * If this is a short packet, then ACK now - with Nagel 582 * congestion avoidance sender won't send more until 583 * he gets an ACK. 584 */ 585 if (tiflags & TH_PUSH) { 586 tp->t_flags |= TF_ACKNOW; 587 tcp_output(tp); 588 } else { 589 tp->t_flags |= TF_DELACK; 590 } 591#else 592 tp->t_flags |= TF_DELACK; 593#endif 594 return; 595 } 596 } 597 598 /* 599 * Calculate amount of space in receive window, 600 * and then do TCP input processing. 601 * Receive window is amount of space in rcv queue, 602 * but not less than advertised window. 603 */ 604 { int win; 605 606 win = sbspace(&so->so_rcv); 607 if (win < 0) 608 win = 0; 609 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 610 } 611 612 switch (tp->t_state) { 613 614 /* 615 * If the state is LISTEN then ignore segment if it contains an RST. 616 * If the segment contains an ACK then it is bad and send a RST. 617 * If it does not contain a SYN then it is not interesting; drop it. 618 * Don't bother responding if the destination was a broadcast. 619 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 620 * tp->iss, and send a segment: 621 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 622 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 623 * Fill in remote peer address fields if not previously specified. 624 * Enter SYN_RECEIVED state, and process any other fields of this 625 * segment in this state. 626 */ 627 case TCPS_LISTEN: { 628 struct mbuf *am; 629 register struct sockaddr_in *sin; 630 631 if (tiflags & TH_RST) 632 goto drop; 633 if (tiflags & TH_ACK) 634 goto dropwithreset; 635 if ((tiflags & TH_SYN) == 0) 636 goto drop; 637 /* 638 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 639 * in_broadcast() should never return true on a received 640 * packet with M_BCAST not set. 641 */ 642 if (m->m_flags & (M_BCAST|M_MCAST) || 643 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 644 goto drop; 645 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 646 M_NOWAIT); 647 if (sin == NULL) 648 goto drop; 649 sin->sin_family = AF_INET; 650 sin->sin_len = sizeof(*sin); 651 sin->sin_addr = ti->ti_src; 652 sin->sin_port = ti->ti_sport; 653 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 654 laddr = inp->inp_laddr; 655 if (inp->inp_laddr.s_addr == INADDR_ANY) 656 inp->inp_laddr = ti->ti_dst; 657 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 658 inp->inp_laddr = laddr; 659 FREE(sin, M_SONAME); 660 goto drop; 661 } 662 FREE(sin, M_SONAME); 663 tp->t_template = tcp_template(tp); 664 if (tp->t_template == 0) { 665 tp = tcp_drop(tp, ENOBUFS); 666 dropsocket = 0; /* socket is already gone */ 667 goto drop; 668 } 669 if ((taop = tcp_gettaocache(inp)) == NULL) { 670 taop = &tao_noncached; 671 bzero(taop, sizeof(*taop)); 672 } 673 tcp_dooptions(tp, optp, optlen, ti, &to); 674 if (iss) 675 tp->iss = iss; 676 else 677 tp->iss = tcp_iss; 678 tcp_iss += TCP_ISSINCR/4; 679 tp->irs = ti->ti_seq; 680 tcp_sendseqinit(tp); 681 tcp_rcvseqinit(tp); 682 /* 683 * Initialization of the tcpcb for transaction; 684 * set SND.WND = SEG.WND, 685 * initialize CCsend and CCrecv. 686 */ 687 tp->snd_wnd = tiwin; /* initial send-window */ 688 tp->cc_send = CC_INC(tcp_ccgen); 689 tp->cc_recv = to.to_cc; 690 /* 691 * Perform TAO test on incoming CC (SEG.CC) option, if any. 692 * - compare SEG.CC against cached CC from the same host, 693 * if any. 694 * - if SEG.CC > chached value, SYN must be new and is accepted 695 * immediately: save new CC in the cache, mark the socket 696 * connected, enter ESTABLISHED state, turn on flag to 697 * send a SYN in the next segment. 698 * A virtual advertised window is set in rcv_adv to 699 * initialize SWS prevention. Then enter normal segment 700 * processing: drop SYN, process data and FIN. 701 * - otherwise do a normal 3-way handshake. 702 */ 703 if ((to.to_flag & TOF_CC) != 0) { 704 if (taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 705 taop->tao_cc = to.to_cc; 706 tp->t_state = TCPS_ESTABLISHED; 707 708 /* 709 * If there is a FIN, or if there is data and the 710 * connection is local, then delay SYN,ACK(SYN) in 711 * the hope of piggy-backing it on a response 712 * segment. Otherwise must send ACK now in case 713 * the other side is slow starting. 714 */ 715 if ((tiflags & TH_FIN) || (ti->ti_len != 0 && 716 in_localaddr(inp->inp_faddr))) 717 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 718 else 719 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 720 721 /* 722 * Limit the `virtual advertised window' to TCP_MAXWIN 723 * here. Even if we requested window scaling, it will 724 * become effective only later when our SYN is acked. 725 */ 726 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 727 tcpstat.tcps_connects++; 728 soisconnected(so); 729 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 730 dropsocket = 0; /* committed to socket */ 731 tcpstat.tcps_accepts++; 732 goto trimthenstep6; 733 } 734 /* else do standard 3-way handshake */ 735 } else { 736 /* 737 * No CC option, but maybe CC.NEW: 738 * invalidate cached value. 739 */ 740 taop->tao_cc = 0; 741 } 742 /* 743 * TAO test failed or there was no CC option, 744 * do a standard 3-way handshake. 745 */ 746 tp->t_flags |= TF_ACKNOW; 747 tp->t_state = TCPS_SYN_RECEIVED; 748 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 749 dropsocket = 0; /* committed to socket */ 750 tcpstat.tcps_accepts++; 751 goto trimthenstep6; 752 } 753 754 /* 755 * If the state is SYN_RECEIVED: 756 * do just the ack and RST checks from SYN_SENT state. 757 * If the state is SYN_SENT: 758 * if seg contains an ACK, but not for our SYN, drop the input. 759 * if seg contains a RST, then drop the connection. 760 * if seg does not contain SYN, then drop it. 761 * Otherwise this is an acceptable SYN segment 762 * initialize tp->rcv_nxt and tp->irs 763 * if seg contains ack then advance tp->snd_una 764 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 765 * arrange for segment to be acked (eventually) 766 * continue processing rest of data/controls, beginning with URG 767 */ 768 case TCPS_SYN_RECEIVED: 769 case TCPS_SYN_SENT: 770 if ((taop = tcp_gettaocache(inp)) == NULL) { 771 taop = &tao_noncached; 772 bzero(taop, sizeof(*taop)); 773 } 774 775 if ((tiflags & TH_ACK) && 776 (SEQ_LEQ(ti->ti_ack, tp->iss) || 777 SEQ_GT(ti->ti_ack, tp->snd_max))) { 778 /* 779 * If we have a cached CCsent for the remote host, 780 * hence we haven't just crashed and restarted, 781 * do not send a RST. This may be a retransmission 782 * from the other side after our earlier ACK was lost. 783 * Our new SYN, when it arrives, will serve as the 784 * needed ACK. 785 */ 786 if (taop->tao_ccsent != 0) 787 goto drop; 788 else 789 goto dropwithreset; 790 } 791 if (tiflags & TH_RST) { 792 if (tiflags & TH_ACK) 793 tp = tcp_drop(tp, ECONNREFUSED); 794 goto drop; 795 } 796 if (tp->t_state == TCPS_SYN_RECEIVED) 797 break; 798 if ((tiflags & TH_SYN) == 0) 799 goto drop; 800 tp->snd_wnd = ti->ti_win; /* initial send window */ 801 tp->cc_recv = to.to_cc; /* foreign CC */ 802 803 tp->irs = ti->ti_seq; 804 tcp_rcvseqinit(tp); 805 if (tiflags & TH_ACK) { 806 /* 807 * Our SYN was acked. If segment contains CC.ECHO 808 * option, check it to make sure this segment really 809 * matches our SYN. If not, just drop it as old 810 * duplicate, but send an RST if we're still playing 811 * by the old rules. If no CC.ECHO option, make sure 812 * we don't get fooled into using T/TCP. 813 */ 814 if (to.to_flag & TOF_CCECHO) { 815 if (tp->cc_send != to.to_ccecho) 816 if (taop->tao_ccsent != 0) 817 goto drop; 818 else 819 goto dropwithreset; 820 } else 821 tp->t_flags &= ~TF_RCVD_CC; 822 tcpstat.tcps_connects++; 823 soisconnected(so); 824 /* Do window scaling on this connection? */ 825 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 826 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 827 tp->snd_scale = tp->requested_s_scale; 828 tp->rcv_scale = tp->request_r_scale; 829 } 830 /* Segment is acceptable, update cache if undefined. */ 831 if (taop->tao_ccsent == 0) 832 taop->tao_ccsent = to.to_ccecho; 833 834 tp->rcv_adv += tp->rcv_wnd; 835 tp->snd_una++; /* SYN is acked */ 836 /* 837 * If there's data, delay ACK; if there's also a FIN 838 * ACKNOW will be turned on later. 839 */ 840 if (ti->ti_len != 0) 841 tp->t_flags |= TF_DELACK; 842 else 843 tp->t_flags |= TF_ACKNOW; 844 /* 845 * Received <SYN,ACK> in SYN_SENT[*] state. 846 * Transitions: 847 * SYN_SENT --> ESTABLISHED 848 * SYN_SENT* --> FIN_WAIT_1 849 */ 850 if (tp->t_flags & TF_NEEDFIN) { 851 tp->t_state = TCPS_FIN_WAIT_1; 852 tp->t_flags &= ~TF_NEEDFIN; 853 tiflags &= ~TH_SYN; 854 } else { 855 tp->t_state = TCPS_ESTABLISHED; 856 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 857 } 858 } else { 859 /* 860 * Received initial SYN in SYN-SENT[*] state => simul- 861 * taneous open. If segment contains CC option and there is 862 * a cached CC, apply TAO test; if it succeeds, connection is 863 * half-synchronized. Otherwise, do 3-way handshake: 864 * SYN-SENT -> SYN-RECEIVED 865 * SYN-SENT* -> SYN-RECEIVED* 866 * If there was no CC option, clear cached CC value. 867 */ 868 tp->t_flags |= TF_ACKNOW; 869 tp->t_timer[TCPT_REXMT] = 0; 870 if (to.to_flag & TOF_CC) { 871 if (taop->tao_cc != 0 && 872 CC_GT(to.to_cc, taop->tao_cc)) { 873 /* 874 * update cache and make transition: 875 * SYN-SENT -> ESTABLISHED* 876 * SYN-SENT* -> FIN-WAIT-1* 877 */ 878 taop->tao_cc = to.to_cc; 879 if (tp->t_flags & TF_NEEDFIN) { 880 tp->t_state = TCPS_FIN_WAIT_1; 881 tp->t_flags &= ~TF_NEEDFIN; 882 } else { 883 tp->t_state = TCPS_ESTABLISHED; 884 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 885 } 886 tp->t_flags |= TF_NEEDSYN; 887 } else 888 tp->t_state = TCPS_SYN_RECEIVED; 889 } else { 890 /* CC.NEW or no option => invalidate cache */ 891 taop->tao_cc = 0; 892 tp->t_state = TCPS_SYN_RECEIVED; 893 } 894 } 895 896trimthenstep6: 897 /* 898 * Advance ti->ti_seq to correspond to first data byte. 899 * If data, trim to stay within window, 900 * dropping FIN if necessary. 901 */ 902 ti->ti_seq++; 903 if (ti->ti_len > tp->rcv_wnd) { 904 todrop = ti->ti_len - tp->rcv_wnd; 905 m_adj(m, -todrop); 906 ti->ti_len = tp->rcv_wnd; 907 tiflags &= ~TH_FIN; 908 tcpstat.tcps_rcvpackafterwin++; 909 tcpstat.tcps_rcvbyteafterwin += todrop; 910 } 911 tp->snd_wl1 = ti->ti_seq - 1; 912 tp->rcv_up = ti->ti_seq; 913 /* 914 * Client side of transaction: already sent SYN and data. 915 * If the remote host used T/TCP to validate the SYN, 916 * our data will be ACK'd; if so, enter normal data segment 917 * processing in the middle of step 5, ack processing. 918 * Otherwise, goto step 6. 919 */ 920 if (tiflags & TH_ACK) 921 goto process_ACK; 922 goto step6; 923 /* 924 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 925 * if segment contains a SYN and CC [not CC.NEW] option: 926 * if state == TIME_WAIT and connection duration > MSL, 927 * drop packet and send RST; 928 * 929 * if SEG.CC > CCrecv then is new SYN, and can implicitly 930 * ack the FIN (and data) in retransmission queue. 931 * Complete close and delete TCPCB. Then reprocess 932 * segment, hoping to find new TCPCB in LISTEN state; 933 * 934 * else must be old SYN; drop it. 935 * else do normal processing. 936 */ 937 case TCPS_LAST_ACK: 938 case TCPS_CLOSING: 939 case TCPS_TIME_WAIT: 940 if ((tiflags & TH_SYN) && 941 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 942 if (tp->t_state == TCPS_TIME_WAIT && 943 tp->t_duration > TCPTV_MSL) 944 goto dropwithreset; 945 if (CC_GT(to.to_cc, tp->cc_recv)) { 946 tp = tcp_close(tp); 947 goto findpcb; 948 } 949 else 950 goto drop; 951 } 952 break; /* continue normal processing */ 953 } 954 955 /* 956 * States other than LISTEN or SYN_SENT. 957 * First check timestamp, if present. 958 * Then check the connection count, if present. 959 * Then check that at least some bytes of segment are within 960 * receive window. If segment begins before rcv_nxt, 961 * drop leading data (and SYN); if nothing left, just ack. 962 * 963 * RFC 1323 PAWS: If we have a timestamp reply on this segment 964 * and it's less than ts_recent, drop it. 965 */ 966 if ((to.to_flag & TOF_TS) != 0 && (tiflags & TH_RST) == 0 && 967 tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { 968 969 /* Check to see if ts_recent is over 24 days old. */ 970 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 971 /* 972 * Invalidate ts_recent. If this segment updates 973 * ts_recent, the age will be reset later and ts_recent 974 * will get a valid value. If it does not, setting 975 * ts_recent to zero will at least satisfy the 976 * requirement that zero be placed in the timestamp 977 * echo reply when ts_recent isn't valid. The 978 * age isn't reset until we get a valid ts_recent 979 * because we don't want out-of-order segments to be 980 * dropped when ts_recent is old. 981 */ 982 tp->ts_recent = 0; 983 } else { 984 tcpstat.tcps_rcvduppack++; 985 tcpstat.tcps_rcvdupbyte += ti->ti_len; 986 tcpstat.tcps_pawsdrop++; 987 goto dropafterack; 988 } 989 } 990 991 /* 992 * T/TCP mechanism 993 * If T/TCP was negotiated and the segment doesn't have CC, 994 * or if it's CC is wrong then drop the segment. 995 * RST segments do not have to comply with this. 996 */ 997 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 998 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc) && 999 (tiflags & TH_RST) == 0) 1000 goto dropafterack; 1001 1002 todrop = tp->rcv_nxt - ti->ti_seq; 1003 if (todrop > 0) { 1004 if (tiflags & TH_SYN) { 1005 tiflags &= ~TH_SYN; 1006 ti->ti_seq++; 1007 if (ti->ti_urp > 1) 1008 ti->ti_urp--; 1009 else 1010 tiflags &= ~TH_URG; 1011 todrop--; 1012 } 1013 /* 1014 * Following if statement from Stevens, vol. 2, p. 960. 1015 */ 1016 if (todrop > ti->ti_len 1017 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 1018 /* 1019 * Any valid FIN must be to the left of the window. 1020 * At this point the FIN must be a duplicate or out 1021 * of sequence; drop it. 1022 */ 1023 tiflags &= ~TH_FIN; 1024 1025 /* 1026 * Send an ACK to resynchronize and drop any data. 1027 * But keep on processing for RST or ACK. 1028 */ 1029 tp->t_flags |= TF_ACKNOW; 1030 todrop = ti->ti_len; 1031 tcpstat.tcps_rcvduppack++; 1032 tcpstat.tcps_rcvdupbyte += todrop; 1033 } else { 1034 tcpstat.tcps_rcvpartduppack++; 1035 tcpstat.tcps_rcvpartdupbyte += todrop; 1036 } 1037 m_adj(m, todrop); 1038 ti->ti_seq += todrop; 1039 ti->ti_len -= todrop; 1040 if (ti->ti_urp > todrop) 1041 ti->ti_urp -= todrop; 1042 else { 1043 tiflags &= ~TH_URG; 1044 ti->ti_urp = 0; 1045 } 1046 } 1047 1048 /* 1049 * If new data are received on a connection after the 1050 * user processes are gone, then RST the other end. 1051 */ 1052 if ((so->so_state & SS_NOFDREF) && 1053 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1054 tp = tcp_close(tp); 1055 tcpstat.tcps_rcvafterclose++; 1056 goto dropwithreset; 1057 } 1058 1059 /* 1060 * If segment ends after window, drop trailing data 1061 * (and PUSH and FIN); if nothing left, just ACK. 1062 */ 1063 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1064 if (todrop > 0) { 1065 tcpstat.tcps_rcvpackafterwin++; 1066 if (todrop >= ti->ti_len) { 1067 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1068 /* 1069 * If a new connection request is received 1070 * while in TIME_WAIT, drop the old connection 1071 * and start over if the sequence numbers 1072 * are above the previous ones. 1073 */ 1074 if (tiflags & TH_SYN && 1075 tp->t_state == TCPS_TIME_WAIT && 1076 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1077 iss = tp->rcv_nxt + TCP_ISSINCR; 1078 tp = tcp_close(tp); 1079 goto findpcb; 1080 } 1081 /* 1082 * If window is closed can only take segments at 1083 * window edge, and have to drop data and PUSH from 1084 * incoming segments. Continue processing, but 1085 * remember to ack. Otherwise, drop segment 1086 * and ack. 1087 */ 1088 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1089 tp->t_flags |= TF_ACKNOW; 1090 tcpstat.tcps_rcvwinprobe++; 1091 } else 1092 goto dropafterack; 1093 } else 1094 tcpstat.tcps_rcvbyteafterwin += todrop; 1095 m_adj(m, -todrop); 1096 ti->ti_len -= todrop; 1097 tiflags &= ~(TH_PUSH|TH_FIN); 1098 } 1099 1100 /* 1101 * If last ACK falls within this segment's sequence numbers, 1102 * record its timestamp. 1103 * NOTE that the test is modified according to the latest 1104 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1105 */ 1106 if ((to.to_flag & TOF_TS) != 0 && 1107 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1108 tp->ts_recent_age = tcp_now; 1109 tp->ts_recent = to.to_tsval; 1110 } 1111 1112 /* 1113 * If the RST bit is set examine the state: 1114 * SYN_RECEIVED STATE: 1115 * If passive open, return to LISTEN state. 1116 * If active open, inform user that connection was refused. 1117 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1118 * Inform user that connection was reset, and close tcb. 1119 * CLOSING, LAST_ACK, TIME_WAIT STATES 1120 * Close the tcb. 1121 */ 1122 if (tiflags&TH_RST) switch (tp->t_state) { 1123 1124 case TCPS_SYN_RECEIVED: 1125 so->so_error = ECONNREFUSED; 1126 goto close; 1127 1128 case TCPS_ESTABLISHED: 1129 case TCPS_FIN_WAIT_1: 1130 case TCPS_FIN_WAIT_2: 1131 case TCPS_CLOSE_WAIT: 1132 so->so_error = ECONNRESET; 1133 close: 1134 tp->t_state = TCPS_CLOSED; 1135 tcpstat.tcps_drops++; 1136 tp = tcp_close(tp); 1137 goto drop; 1138 1139 case TCPS_CLOSING: 1140 case TCPS_LAST_ACK: 1141 case TCPS_TIME_WAIT: 1142 tp = tcp_close(tp); 1143 goto drop; 1144 } 1145 1146 /* 1147 * If a SYN is in the window, then this is an 1148 * error and we send an RST and drop the connection. 1149 */ 1150 if (tiflags & TH_SYN) { 1151 tp = tcp_drop(tp, ECONNRESET); 1152 goto dropwithreset; 1153 } 1154 1155 /* 1156 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1157 * flag is on (half-synchronized state), then queue data for 1158 * later processing; else drop segment and return. 1159 */ 1160 if ((tiflags & TH_ACK) == 0) { 1161 if (tp->t_state == TCPS_SYN_RECEIVED || 1162 (tp->t_flags & TF_NEEDSYN)) 1163 goto step6; 1164 else 1165 goto drop; 1166 } 1167 1168 /* 1169 * Ack processing. 1170 */ 1171 switch (tp->t_state) { 1172 1173 /* 1174 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1175 * ESTABLISHED state and continue processing, otherwise 1176 * send an RST. 1177 */ 1178 case TCPS_SYN_RECEIVED: 1179 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1180 SEQ_GT(ti->ti_ack, tp->snd_max)) 1181 goto dropwithreset; 1182 1183 tcpstat.tcps_connects++; 1184 soisconnected(so); 1185 /* Do window scaling? */ 1186 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1187 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1188 tp->snd_scale = tp->requested_s_scale; 1189 tp->rcv_scale = tp->request_r_scale; 1190 } 1191 /* 1192 * Upon successful completion of 3-way handshake, 1193 * update cache.CC if it was undefined, pass any queued 1194 * data to the user, and advance state appropriately. 1195 */ 1196 if ((taop = tcp_gettaocache(inp)) != NULL && 1197 taop->tao_cc == 0) 1198 taop->tao_cc = tp->cc_recv; 1199 1200 /* 1201 * Make transitions: 1202 * SYN-RECEIVED -> ESTABLISHED 1203 * SYN-RECEIVED* -> FIN-WAIT-1 1204 */ 1205 if (tp->t_flags & TF_NEEDFIN) { 1206 tp->t_state = TCPS_FIN_WAIT_1; 1207 tp->t_flags &= ~TF_NEEDFIN; 1208 } else { 1209 tp->t_state = TCPS_ESTABLISHED; 1210 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 1211 } 1212 /* 1213 * If segment contains data or ACK, will call tcp_reass() 1214 * later; if not, do so now to pass queued data to user. 1215 */ 1216 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1217 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1218 (struct mbuf *)0); 1219 tp->snd_wl1 = ti->ti_seq - 1; 1220 /* fall into ... */ 1221 1222 /* 1223 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1224 * ACKs. If the ack is in the range 1225 * tp->snd_una < ti->ti_ack <= tp->snd_max 1226 * then advance tp->snd_una to ti->ti_ack and drop 1227 * data from the retransmission queue. If this ACK reflects 1228 * more up to date window information we update our window information. 1229 */ 1230 case TCPS_ESTABLISHED: 1231 case TCPS_FIN_WAIT_1: 1232 case TCPS_FIN_WAIT_2: 1233 case TCPS_CLOSE_WAIT: 1234 case TCPS_CLOSING: 1235 case TCPS_LAST_ACK: 1236 case TCPS_TIME_WAIT: 1237 1238 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1239 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1240 tcpstat.tcps_rcvdupack++; 1241 /* 1242 * If we have outstanding data (other than 1243 * a window probe), this is a completely 1244 * duplicate ack (ie, window info didn't 1245 * change), the ack is the biggest we've 1246 * seen and we've seen exactly our rexmt 1247 * threshhold of them, assume a packet 1248 * has been dropped and retransmit it. 1249 * Kludge snd_nxt & the congestion 1250 * window so we send only this one 1251 * packet. 1252 * 1253 * We know we're losing at the current 1254 * window size so do congestion avoidance 1255 * (set ssthresh to half the current window 1256 * and pull our congestion window back to 1257 * the new ssthresh). 1258 * 1259 * Dup acks mean that packets have left the 1260 * network (they're now cached at the receiver) 1261 * so bump cwnd by the amount in the receiver 1262 * to keep a constant cwnd packets in the 1263 * network. 1264 */ 1265 if (tp->t_timer[TCPT_REXMT] == 0 || 1266 ti->ti_ack != tp->snd_una) 1267 tp->t_dupacks = 0; 1268 else if (++tp->t_dupacks == tcprexmtthresh) { 1269 tcp_seq onxt = tp->snd_nxt; 1270 u_int win = 1271 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1272 tp->t_maxseg; 1273 1274 if (win < 2) 1275 win = 2; 1276 tp->snd_ssthresh = win * tp->t_maxseg; 1277 tp->t_timer[TCPT_REXMT] = 0; 1278 tp->t_rtt = 0; 1279 tp->snd_nxt = ti->ti_ack; 1280 tp->snd_cwnd = tp->t_maxseg; 1281 (void) tcp_output(tp); 1282 tp->snd_cwnd = tp->snd_ssthresh + 1283 tp->t_maxseg * tp->t_dupacks; 1284 if (SEQ_GT(onxt, tp->snd_nxt)) 1285 tp->snd_nxt = onxt; 1286 goto drop; 1287 } else if (tp->t_dupacks > tcprexmtthresh) { 1288 tp->snd_cwnd += tp->t_maxseg; 1289 (void) tcp_output(tp); 1290 goto drop; 1291 } 1292 } else 1293 tp->t_dupacks = 0; 1294 break; 1295 } 1296 /* 1297 * If the congestion window was inflated to account 1298 * for the other side's cached packets, retract it. 1299 */ 1300 if (tp->t_dupacks >= tcprexmtthresh && 1301 tp->snd_cwnd > tp->snd_ssthresh) 1302 tp->snd_cwnd = tp->snd_ssthresh; 1303 tp->t_dupacks = 0; 1304 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1305 tcpstat.tcps_rcvacktoomuch++; 1306 goto dropafterack; 1307 } 1308 /* 1309 * If we reach this point, ACK is not a duplicate, 1310 * i.e., it ACKs something we sent. 1311 */ 1312 if (tp->t_flags & TF_NEEDSYN) { 1313 /* 1314 * T/TCP: Connection was half-synchronized, and our 1315 * SYN has been ACK'd (so connection is now fully 1316 * synchronized). Go to non-starred state, 1317 * increment snd_una for ACK of SYN, and check if 1318 * we can do window scaling. 1319 */ 1320 tp->t_flags &= ~TF_NEEDSYN; 1321 tp->snd_una++; 1322 /* Do window scaling? */ 1323 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1324 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1325 tp->snd_scale = tp->requested_s_scale; 1326 tp->rcv_scale = tp->request_r_scale; 1327 } 1328 } 1329 1330process_ACK: 1331 acked = ti->ti_ack - tp->snd_una; 1332 tcpstat.tcps_rcvackpack++; 1333 tcpstat.tcps_rcvackbyte += acked; 1334 1335 /* 1336 * If we have a timestamp reply, update smoothed 1337 * round trip time. If no timestamp is present but 1338 * transmit timer is running and timed sequence 1339 * number was acked, update smoothed round trip time. 1340 * Since we now have an rtt measurement, cancel the 1341 * timer backoff (cf., Phil Karn's retransmit alg.). 1342 * Recompute the initial retransmit timer. 1343 */ 1344 if (to.to_flag & TOF_TS) 1345 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1346 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1347 tcp_xmit_timer(tp,tp->t_rtt); 1348 1349 /* 1350 * If all outstanding data is acked, stop retransmit 1351 * timer and remember to restart (more output or persist). 1352 * If there is more data to be acked, restart retransmit 1353 * timer, using current (possibly backed-off) value. 1354 */ 1355 if (ti->ti_ack == tp->snd_max) { 1356 tp->t_timer[TCPT_REXMT] = 0; 1357 needoutput = 1; 1358 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1359 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1360 1361 /* 1362 * If no data (only SYN) was ACK'd, 1363 * skip rest of ACK processing. 1364 */ 1365 if (acked == 0) 1366 goto step6; 1367 1368 /* 1369 * When new data is acked, open the congestion window. 1370 * If the window gives us less than ssthresh packets 1371 * in flight, open exponentially (maxseg per packet). 1372 * Otherwise open linearly: maxseg per window 1373 * (maxseg^2 / cwnd per packet). 1374 */ 1375 { 1376 register u_int cw = tp->snd_cwnd; 1377 register u_int incr = tp->t_maxseg; 1378 1379 if (cw > tp->snd_ssthresh) 1380 incr = incr * incr / cw; 1381 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1382 } 1383 if (acked > so->so_snd.sb_cc) { 1384 tp->snd_wnd -= so->so_snd.sb_cc; 1385 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1386 ourfinisacked = 1; 1387 } else { 1388 sbdrop(&so->so_snd, acked); 1389 tp->snd_wnd -= acked; 1390 ourfinisacked = 0; 1391 } 1392 if (so->so_snd.sb_flags & SB_NOTIFY) 1393 sowwakeup(so); 1394 tp->snd_una = ti->ti_ack; 1395 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1396 tp->snd_nxt = tp->snd_una; 1397 1398 switch (tp->t_state) { 1399 1400 /* 1401 * In FIN_WAIT_1 STATE in addition to the processing 1402 * for the ESTABLISHED state if our FIN is now acknowledged 1403 * then enter FIN_WAIT_2. 1404 */ 1405 case TCPS_FIN_WAIT_1: 1406 if (ourfinisacked) { 1407 /* 1408 * If we can't receive any more 1409 * data, then closing user can proceed. 1410 * Starting the timer is contrary to the 1411 * specification, but if we don't get a FIN 1412 * we'll hang forever. 1413 */ 1414 if (so->so_state & SS_CANTRCVMORE) { 1415 soisdisconnected(so); 1416 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1417 } 1418 tp->t_state = TCPS_FIN_WAIT_2; 1419 } 1420 break; 1421 1422 /* 1423 * In CLOSING STATE in addition to the processing for 1424 * the ESTABLISHED state if the ACK acknowledges our FIN 1425 * then enter the TIME-WAIT state, otherwise ignore 1426 * the segment. 1427 */ 1428 case TCPS_CLOSING: 1429 if (ourfinisacked) { 1430 tp->t_state = TCPS_TIME_WAIT; 1431 tcp_canceltimers(tp); 1432 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1433 if (tp->cc_recv != 0 && 1434 tp->t_duration < TCPTV_MSL) 1435 tp->t_timer[TCPT_2MSL] = 1436 tp->t_rxtcur * TCPTV_TWTRUNC; 1437 else 1438 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1439 soisdisconnected(so); 1440 } 1441 break; 1442 1443 /* 1444 * In LAST_ACK, we may still be waiting for data to drain 1445 * and/or to be acked, as well as for the ack of our FIN. 1446 * If our FIN is now acknowledged, delete the TCB, 1447 * enter the closed state and return. 1448 */ 1449 case TCPS_LAST_ACK: 1450 if (ourfinisacked) { 1451 tp = tcp_close(tp); 1452 goto drop; 1453 } 1454 break; 1455 1456 /* 1457 * In TIME_WAIT state the only thing that should arrive 1458 * is a retransmission of the remote FIN. Acknowledge 1459 * it and restart the finack timer. 1460 */ 1461 case TCPS_TIME_WAIT: 1462 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1463 goto dropafterack; 1464 } 1465 } 1466 1467step6: 1468 /* 1469 * Update window information. 1470 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1471 */ 1472 if ((tiflags & TH_ACK) && 1473 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1474 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1475 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1476 /* keep track of pure window updates */ 1477 if (ti->ti_len == 0 && 1478 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1479 tcpstat.tcps_rcvwinupd++; 1480 tp->snd_wnd = tiwin; 1481 tp->snd_wl1 = ti->ti_seq; 1482 tp->snd_wl2 = ti->ti_ack; 1483 if (tp->snd_wnd > tp->max_sndwnd) 1484 tp->max_sndwnd = tp->snd_wnd; 1485 needoutput = 1; 1486 } 1487 1488 /* 1489 * Process segments with URG. 1490 */ 1491 if ((tiflags & TH_URG) && ti->ti_urp && 1492 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1493 /* 1494 * This is a kludge, but if we receive and accept 1495 * random urgent pointers, we'll crash in 1496 * soreceive. It's hard to imagine someone 1497 * actually wanting to send this much urgent data. 1498 */ 1499 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1500 ti->ti_urp = 0; /* XXX */ 1501 tiflags &= ~TH_URG; /* XXX */ 1502 goto dodata; /* XXX */ 1503 } 1504 /* 1505 * If this segment advances the known urgent pointer, 1506 * then mark the data stream. This should not happen 1507 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1508 * a FIN has been received from the remote side. 1509 * In these states we ignore the URG. 1510 * 1511 * According to RFC961 (Assigned Protocols), 1512 * the urgent pointer points to the last octet 1513 * of urgent data. We continue, however, 1514 * to consider it to indicate the first octet 1515 * of data past the urgent section as the original 1516 * spec states (in one of two places). 1517 */ 1518 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1519 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1520 so->so_oobmark = so->so_rcv.sb_cc + 1521 (tp->rcv_up - tp->rcv_nxt) - 1; 1522 if (so->so_oobmark == 0) 1523 so->so_state |= SS_RCVATMARK; 1524 sohasoutofband(so); 1525 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1526 } 1527 /* 1528 * Remove out of band data so doesn't get presented to user. 1529 * This can happen independent of advancing the URG pointer, 1530 * but if two URG's are pending at once, some out-of-band 1531 * data may creep in... ick. 1532 */ 1533 if (ti->ti_urp <= (u_long)ti->ti_len 1534#ifdef SO_OOBINLINE 1535 && (so->so_options & SO_OOBINLINE) == 0 1536#endif 1537 ) 1538 tcp_pulloutofband(so, ti, m); 1539 } else 1540 /* 1541 * If no out of band data is expected, 1542 * pull receive urgent pointer along 1543 * with the receive window. 1544 */ 1545 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1546 tp->rcv_up = tp->rcv_nxt; 1547dodata: /* XXX */ 1548 1549 /* 1550 * Process the segment text, merging it into the TCP sequencing queue, 1551 * and arranging for acknowledgment of receipt if necessary. 1552 * This process logically involves adjusting tp->rcv_wnd as data 1553 * is presented to the user (this happens in tcp_usrreq.c, 1554 * case PRU_RCVD). If a FIN has already been received on this 1555 * connection then we just ignore the text. 1556 */ 1557 if ((ti->ti_len || (tiflags&TH_FIN)) && 1558 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1559 TCP_REASS(tp, ti, m, so, tiflags); 1560 /* 1561 * Note the amount of data that peer has sent into 1562 * our window, in order to estimate the sender's 1563 * buffer size. 1564 */ 1565 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1566 } else { 1567 m_freem(m); 1568 tiflags &= ~TH_FIN; 1569 } 1570 1571 /* 1572 * If FIN is received ACK the FIN and let the user know 1573 * that the connection is closing. 1574 */ 1575 if (tiflags & TH_FIN) { 1576 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1577 socantrcvmore(so); 1578 /* 1579 * If connection is half-synchronized 1580 * (ie NEEDSYN flag on) then delay ACK, 1581 * so it may be piggybacked when SYN is sent. 1582 * Otherwise, since we received a FIN then no 1583 * more input can be expected, send ACK now. 1584 */ 1585 if (tp->t_flags & TF_NEEDSYN) 1586 tp->t_flags |= TF_DELACK; 1587 else 1588 tp->t_flags |= TF_ACKNOW; 1589 tp->rcv_nxt++; 1590 } 1591 switch (tp->t_state) { 1592 1593 /* 1594 * In SYN_RECEIVED and ESTABLISHED STATES 1595 * enter the CLOSE_WAIT state. 1596 */ 1597 case TCPS_SYN_RECEIVED: 1598 case TCPS_ESTABLISHED: 1599 tp->t_state = TCPS_CLOSE_WAIT; 1600 break; 1601 1602 /* 1603 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1604 * enter the CLOSING state. 1605 */ 1606 case TCPS_FIN_WAIT_1: 1607 tp->t_state = TCPS_CLOSING; 1608 break; 1609 1610 /* 1611 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1612 * starting the time-wait timer, turning off the other 1613 * standard timers. 1614 */ 1615 case TCPS_FIN_WAIT_2: 1616 tp->t_state = TCPS_TIME_WAIT; 1617 tcp_canceltimers(tp); 1618 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1619 if (tp->cc_recv != 0 && 1620 tp->t_duration < TCPTV_MSL) { 1621 tp->t_timer[TCPT_2MSL] = 1622 tp->t_rxtcur * TCPTV_TWTRUNC; 1623 /* For transaction client, force ACK now. */ 1624 tp->t_flags |= TF_ACKNOW; 1625 } 1626 else 1627 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1628 soisdisconnected(so); 1629 break; 1630 1631 /* 1632 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1633 */ 1634 case TCPS_TIME_WAIT: 1635 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1636 break; 1637 } 1638 } 1639#ifdef TCPDEBUG 1640 if (so->so_options & SO_DEBUG) 1641 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1642#endif 1643 1644 /* 1645 * Return any desired output. 1646 */ 1647 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1648 (void) tcp_output(tp); 1649 return; 1650 1651dropafterack: 1652 /* 1653 * Generate an ACK dropping incoming segment if it occupies 1654 * sequence space, where the ACK reflects our state. 1655 */ 1656 if (tiflags & TH_RST) 1657 goto drop; 1658#ifdef TCPDEBUG 1659 if (so->so_options & SO_DEBUG) 1660 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1661#endif 1662 m_freem(m); 1663 tp->t_flags |= TF_ACKNOW; 1664 (void) tcp_output(tp); 1665 return; 1666 1667dropwithreset: 1668 /* 1669 * Generate a RST, dropping incoming segment. 1670 * Make ACK acceptable to originator of segment. 1671 * Don't bother to respond if destination was broadcast/multicast. 1672 */ 1673 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1674 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1675 goto drop; 1676#ifdef TCPDEBUG 1677 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1678 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1679#endif 1680 if (tiflags & TH_ACK) 1681 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1682 else { 1683 if (tiflags & TH_SYN) 1684 ti->ti_len++; 1685 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1686 TH_RST|TH_ACK); 1687 } 1688 /* destroy temporarily created socket */ 1689 if (dropsocket) 1690 (void) soabort(so); 1691 return; 1692 1693drop: 1694 /* 1695 * Drop space held by incoming segment and return. 1696 */ 1697#ifdef TCPDEBUG 1698 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1699 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1700#endif 1701 m_freem(m); 1702 /* destroy temporarily created socket */ 1703 if (dropsocket) 1704 (void) soabort(so); 1705 return; 1706#ifndef TUBA_INCLUDE 1707} 1708 1709static void 1710tcp_dooptions(tp, cp, cnt, ti, to) 1711 struct tcpcb *tp; 1712 u_char *cp; 1713 int cnt; 1714 struct tcpiphdr *ti; 1715 struct tcpopt *to; 1716{ 1717 u_short mss = 0; 1718 int opt, optlen; 1719 1720 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1721 opt = cp[0]; 1722 if (opt == TCPOPT_EOL) 1723 break; 1724 if (opt == TCPOPT_NOP) 1725 optlen = 1; 1726 else { 1727 optlen = cp[1]; 1728 if (optlen <= 0) 1729 break; 1730 } 1731 switch (opt) { 1732 1733 default: 1734 continue; 1735 1736 case TCPOPT_MAXSEG: 1737 if (optlen != TCPOLEN_MAXSEG) 1738 continue; 1739 if (!(ti->ti_flags & TH_SYN)) 1740 continue; 1741 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1742 NTOHS(mss); 1743 break; 1744 1745 case TCPOPT_WINDOW: 1746 if (optlen != TCPOLEN_WINDOW) 1747 continue; 1748 if (!(ti->ti_flags & TH_SYN)) 1749 continue; 1750 tp->t_flags |= TF_RCVD_SCALE; 1751 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1752 break; 1753 1754 case TCPOPT_TIMESTAMP: 1755 if (optlen != TCPOLEN_TIMESTAMP) 1756 continue; 1757 to->to_flag |= TOF_TS; 1758 bcopy((char *)cp + 2, 1759 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1760 NTOHL(to->to_tsval); 1761 bcopy((char *)cp + 6, 1762 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1763 NTOHL(to->to_tsecr); 1764 1765 /* 1766 * A timestamp received in a SYN makes 1767 * it ok to send timestamp requests and replies. 1768 */ 1769 if (ti->ti_flags & TH_SYN) { 1770 tp->t_flags |= TF_RCVD_TSTMP; 1771 tp->ts_recent = to->to_tsval; 1772 tp->ts_recent_age = tcp_now; 1773 } 1774 break; 1775 case TCPOPT_CC: 1776 if (optlen != TCPOLEN_CC) 1777 continue; 1778 to->to_flag |= TOF_CC; 1779 bcopy((char *)cp + 2, 1780 (char *)&to->to_cc, sizeof(to->to_cc)); 1781 NTOHL(to->to_cc); 1782 /* 1783 * A CC or CC.new option received in a SYN makes 1784 * it ok to send CC in subsequent segments. 1785 */ 1786 if (ti->ti_flags & TH_SYN) 1787 tp->t_flags |= TF_RCVD_CC; 1788 break; 1789 case TCPOPT_CCNEW: 1790 if (optlen != TCPOLEN_CC) 1791 continue; 1792 if (!(ti->ti_flags & TH_SYN)) 1793 continue; 1794 to->to_flag |= TOF_CCNEW; 1795 bcopy((char *)cp + 2, 1796 (char *)&to->to_cc, sizeof(to->to_cc)); 1797 NTOHL(to->to_cc); 1798 /* 1799 * A CC or CC.new option received in a SYN makes 1800 * it ok to send CC in subsequent segments. 1801 */ 1802 tp->t_flags |= TF_RCVD_CC; 1803 break; 1804 case TCPOPT_CCECHO: 1805 if (optlen != TCPOLEN_CC) 1806 continue; 1807 if (!(ti->ti_flags & TH_SYN)) 1808 continue; 1809 to->to_flag |= TOF_CCECHO; 1810 bcopy((char *)cp + 2, 1811 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1812 NTOHL(to->to_ccecho); 1813 break; 1814 } 1815 } 1816 if (ti->ti_flags & TH_SYN) 1817 tcp_mss(tp, mss); /* sets t_maxseg */ 1818} 1819 1820/* 1821 * Pull out of band byte out of a segment so 1822 * it doesn't appear in the user's data queue. 1823 * It is still reflected in the segment length for 1824 * sequencing purposes. 1825 */ 1826static void 1827tcp_pulloutofband(so, ti, m) 1828 struct socket *so; 1829 struct tcpiphdr *ti; 1830 register struct mbuf *m; 1831{ 1832 int cnt = ti->ti_urp - 1; 1833 1834 while (cnt >= 0) { 1835 if (m->m_len > cnt) { 1836 char *cp = mtod(m, caddr_t) + cnt; 1837 struct tcpcb *tp = sototcpcb(so); 1838 1839 tp->t_iobc = *cp; 1840 tp->t_oobflags |= TCPOOB_HAVEDATA; 1841 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1842 m->m_len--; 1843 return; 1844 } 1845 cnt -= m->m_len; 1846 m = m->m_next; 1847 if (m == 0) 1848 break; 1849 } 1850 panic("tcp_pulloutofband"); 1851} 1852 1853/* 1854 * Collect new round-trip time estimate 1855 * and update averages and current timeout. 1856 */ 1857static void 1858tcp_xmit_timer(tp, rtt) 1859 register struct tcpcb *tp; 1860 short rtt; 1861{ 1862 register int delta; 1863 1864 tcpstat.tcps_rttupdated++; 1865 tp->t_rttupdated++; 1866 if (tp->t_srtt != 0) { 1867 /* 1868 * srtt is stored as fixed point with 5 bits after the 1869 * binary point (i.e., scaled by 8). The following magic 1870 * is equivalent to the smoothing algorithm in rfc793 with 1871 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1872 * point). Adjust rtt to origin 0. 1873 */ 1874 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 1875 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 1876 1877 if ((tp->t_srtt += delta) <= 0) 1878 tp->t_srtt = 1; 1879 1880 /* 1881 * We accumulate a smoothed rtt variance (actually, a 1882 * smoothed mean difference), then set the retransmit 1883 * timer to smoothed rtt + 4 times the smoothed variance. 1884 * rttvar is stored as fixed point with 4 bits after the 1885 * binary point (scaled by 16). The following is 1886 * equivalent to rfc793 smoothing with an alpha of .75 1887 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1888 * rfc793's wired-in beta. 1889 */ 1890 if (delta < 0) 1891 delta = -delta; 1892 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 1893 if ((tp->t_rttvar += delta) <= 0) 1894 tp->t_rttvar = 1; 1895 } else { 1896 /* 1897 * No rtt measurement yet - use the unsmoothed rtt. 1898 * Set the variance to half the rtt (so our first 1899 * retransmit happens at 3*rtt). 1900 */ 1901 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1902 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1903 } 1904 tp->t_rtt = 0; 1905 tp->t_rxtshift = 0; 1906 1907 /* 1908 * the retransmit should happen at rtt + 4 * rttvar. 1909 * Because of the way we do the smoothing, srtt and rttvar 1910 * will each average +1/2 tick of bias. When we compute 1911 * the retransmit timer, we want 1/2 tick of rounding and 1912 * 1 extra tick because of +-1/2 tick uncertainty in the 1913 * firing of the timer. The bias will give us exactly the 1914 * 1.5 tick we need. But, because the bias is 1915 * statistical, we have to test that we don't drop below 1916 * the minimum feasible timer (which is 2 ticks). 1917 */ 1918 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1919 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 1920 1921 /* 1922 * We received an ack for a packet that wasn't retransmitted; 1923 * it is probably safe to discard any error indications we've 1924 * received recently. This isn't quite right, but close enough 1925 * for now (a route might have failed after we sent a segment, 1926 * and the return path might not be symmetrical). 1927 */ 1928 tp->t_softerror = 0; 1929} 1930 1931/* 1932 * Determine a reasonable value for maxseg size. 1933 * If the route is known, check route for mtu. 1934 * If none, use an mss that can be handled on the outgoing 1935 * interface without forcing IP to fragment; if bigger than 1936 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1937 * to utilize large mbufs. If no route is found, route has no mtu, 1938 * or the destination isn't local, use a default, hopefully conservative 1939 * size (usually 512 or the default IP max size, but no more than the mtu 1940 * of the interface), as we can't discover anything about intervening 1941 * gateways or networks. We also initialize the congestion/slow start 1942 * window to be a single segment if the destination isn't local. 1943 * While looking at the routing entry, we also initialize other path-dependent 1944 * parameters from pre-set or cached values in the routing entry. 1945 * 1946 * Also take into account the space needed for options that we 1947 * send regularly. Make maxseg shorter by that amount to assure 1948 * that we can send maxseg amount of data even when the options 1949 * are present. Store the upper limit of the length of options plus 1950 * data in maxopd. 1951 * 1952 * NOTE that this routine is only called when we process an incoming 1953 * segment, for outgoing segments only tcp_mssopt is called. 1954 * 1955 * In case of T/TCP, we call this routine during implicit connection 1956 * setup as well (offer = -1), to initialize maxseg from the cached 1957 * MSS of our peer. 1958 */ 1959void 1960tcp_mss(tp, offer) 1961 struct tcpcb *tp; 1962 int offer; 1963{ 1964 register struct rtentry *rt; 1965 struct ifnet *ifp; 1966 register int rtt, mss; 1967 u_long bufsize; 1968 struct inpcb *inp; 1969 struct socket *so; 1970 struct rmxp_tao *taop; 1971 int origoffer = offer; 1972 1973 inp = tp->t_inpcb; 1974 if ((rt = tcp_rtlookup(inp)) == NULL) { 1975 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 1976 return; 1977 } 1978 ifp = rt->rt_ifp; 1979 so = inp->inp_socket; 1980 1981 taop = rmx_taop(rt->rt_rmx); 1982 /* 1983 * Offer == -1 means that we didn't receive SYN yet, 1984 * use cached value in that case; 1985 */ 1986 if (offer == -1) 1987 offer = taop->tao_mssopt; 1988 /* 1989 * Offer == 0 means that there was no MSS on the SYN segment, 1990 * in this case we use tcp_mssdflt. 1991 */ 1992 if (offer == 0) 1993 offer = tcp_mssdflt; 1994 else 1995 /* 1996 * Sanity check: make sure that maxopd will be large 1997 * enough to allow some data on segments even is the 1998 * all the option space is used (40bytes). Otherwise 1999 * funny things may happen in tcp_output. 2000 */ 2001 offer = max(offer, 64); 2002 taop->tao_mssopt = offer; 2003 2004 /* 2005 * While we're here, check if there's an initial rtt 2006 * or rttvar. Convert from the route-table units 2007 * to scaled multiples of the slow timeout timer. 2008 */ 2009 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2010 /* 2011 * XXX the lock bit for RTT indicates that the value 2012 * is also a minimum value; this is subject to time. 2013 */ 2014 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2015 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 2016 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 2017 tcpstat.tcps_usedrtt++; 2018 if (rt->rt_rmx.rmx_rttvar) { 2019 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2020 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 2021 tcpstat.tcps_usedrttvar++; 2022 } else { 2023 /* default variation is +- 1 rtt */ 2024 tp->t_rttvar = 2025 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2026 } 2027 TCPT_RANGESET(tp->t_rxtcur, 2028 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2029 tp->t_rttmin, TCPTV_REXMTMAX); 2030 } 2031 /* 2032 * if there's an mtu associated with the route, use it 2033 */ 2034 if (rt->rt_rmx.rmx_mtu) 2035 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2036 else 2037 { 2038 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 2039 if (!in_localaddr(inp->inp_faddr)) 2040 mss = min(mss, tcp_mssdflt); 2041 } 2042 mss = min(mss, offer); 2043 /* 2044 * maxopd stores the maximum length of data AND options 2045 * in a segment; maxseg is the amount of data in a normal 2046 * segment. We need to store this value (maxopd) apart 2047 * from maxseg, because now every segment carries options 2048 * and thus we normally have somewhat less data in segments. 2049 */ 2050 tp->t_maxopd = mss; 2051 2052 /* 2053 * In case of T/TCP, origoffer==-1 indicates, that no segments 2054 * were received yet. In this case we just guess, otherwise 2055 * we do the same as before T/TCP. 2056 */ 2057 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2058 (origoffer == -1 || 2059 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2060 mss -= TCPOLEN_TSTAMP_APPA; 2061 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2062 (origoffer == -1 || 2063 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2064 mss -= TCPOLEN_CC_APPA; 2065 2066#if (MCLBYTES & (MCLBYTES - 1)) == 0 2067 if (mss > MCLBYTES) 2068 mss &= ~(MCLBYTES-1); 2069#else 2070 if (mss > MCLBYTES) 2071 mss = mss / MCLBYTES * MCLBYTES; 2072#endif 2073 /* 2074 * If there's a pipesize, change the socket buffer 2075 * to that size. Make the socket buffers an integral 2076 * number of mss units; if the mss is larger than 2077 * the socket buffer, decrease the mss. 2078 */ 2079#ifdef RTV_SPIPE 2080 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2081#endif 2082 bufsize = so->so_snd.sb_hiwat; 2083 if (bufsize < mss) 2084 mss = bufsize; 2085 else { 2086 bufsize = roundup(bufsize, mss); 2087 if (bufsize > sb_max) 2088 bufsize = sb_max; 2089 (void)sbreserve(&so->so_snd, bufsize); 2090 } 2091 tp->t_maxseg = mss; 2092 2093#ifdef RTV_RPIPE 2094 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2095#endif 2096 bufsize = so->so_rcv.sb_hiwat; 2097 if (bufsize > mss) { 2098 bufsize = roundup(bufsize, mss); 2099 if (bufsize > sb_max) 2100 bufsize = sb_max; 2101 (void)sbreserve(&so->so_rcv, bufsize); 2102 } 2103 /* 2104 * Don't force slow-start on local network. 2105 */ 2106 if (!in_localaddr(inp->inp_faddr)) 2107 tp->snd_cwnd = mss; 2108 2109 if (rt->rt_rmx.rmx_ssthresh) { 2110 /* 2111 * There's some sort of gateway or interface 2112 * buffer limit on the path. Use this to set 2113 * the slow start threshhold, but set the 2114 * threshold to no less than 2*mss. 2115 */ 2116 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2117 tcpstat.tcps_usedssthresh++; 2118 } 2119} 2120 2121/* 2122 * Determine the MSS option to send on an outgoing SYN. 2123 */ 2124int 2125tcp_mssopt(tp) 2126 struct tcpcb *tp; 2127{ 2128 struct rtentry *rt; 2129 2130 rt = tcp_rtlookup(tp->t_inpcb); 2131 if (rt == NULL) 2132 return tcp_mssdflt; 2133 2134 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2135} 2136#endif /* TUBA_INCLUDE */ 2137