tcp_input.c revision 46381
1/* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 34 * $Id: tcp_input.c,v 1.84 1999/02/06 00:47:45 fenner Exp $ 35 */ 36 37#include "opt_ipfw.h" /* for ipfw_fwd */ 38#include "opt_tcpdebug.h" 39 40#include <sys/param.h> 41#include <sys/systm.h> 42#include <sys/kernel.h> 43#include <sys/sysctl.h> 44#include <sys/malloc.h> 45#include <sys/mbuf.h> 46#include <sys/proc.h> /* for proc0 declaration */ 47#include <sys/protosw.h> 48#include <sys/socket.h> 49#include <sys/socketvar.h> 50#include <sys/syslog.h> 51 52#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 53 54#include <net/if.h> 55#include <net/route.h> 56 57#include <netinet/in.h> 58#include <netinet/in_systm.h> 59#include <netinet/ip.h> 60#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 61#include <netinet/in_pcb.h> 62#include <netinet/ip_var.h> 63#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 64#include <netinet/tcp.h> 65#include <netinet/tcp_fsm.h> 66#include <netinet/tcp_seq.h> 67#include <netinet/tcp_timer.h> 68#include <netinet/tcp_var.h> 69#include <netinet/tcpip.h> 70#ifdef TCPDEBUG 71#include <netinet/tcp_debug.h> 72static struct tcpiphdr tcp_saveti; 73#endif 74 75static int tcprexmtthresh = 3; 76tcp_seq tcp_iss; 77tcp_cc tcp_ccgen; 78 79struct tcpstat tcpstat; 80SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RD, 81 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 82 83static int log_in_vain = 0; 84SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 85 &log_in_vain, 0, "Log all incoming TCP connections"); 86 87int tcp_delack_enabled = 1; 88SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 89 &tcp_delack_enabled, 0, 90 "Delay ACK to try and piggyback it onto a data packet"); 91 92u_long tcp_now; 93struct inpcbhead tcb; 94struct inpcbinfo tcbinfo; 95 96static void tcp_dooptions __P((struct tcpcb *, 97 u_char *, int, struct tcpiphdr *, struct tcpopt *)); 98static void tcp_pulloutofband __P((struct socket *, 99 struct tcpiphdr *, struct mbuf *)); 100static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 101static void tcp_xmit_timer __P((struct tcpcb *, int)); 102 103 104/* 105 * Insert segment ti into reassembly queue of tcp with 106 * control block tp. Return TH_FIN if reassembly now includes 107 * a segment with FIN. The macro form does the common case inline 108 * (segment is the next to be received on an established connection, 109 * and the queue is empty), avoiding linkage into and removal 110 * from the queue and repetition of various conversions. 111 * Set DELACK for segments received in order, but ack immediately 112 * when segments are out of order (so fast retransmit can work). 113 */ 114#define TCP_REASS(tp, ti, m, so, flags) { \ 115 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 116 (tp)->t_segq == NULL && \ 117 (tp)->t_state == TCPS_ESTABLISHED) { \ 118 if (tcp_delack_enabled) \ 119 tp->t_flags |= TF_DELACK; \ 120 else \ 121 tp->t_flags |= TF_ACKNOW; \ 122 (tp)->rcv_nxt += (ti)->ti_len; \ 123 flags = (ti)->ti_flags & TH_FIN; \ 124 tcpstat.tcps_rcvpack++;\ 125 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 126 sbappend(&(so)->so_rcv, (m)); \ 127 sorwakeup(so); \ 128 } else { \ 129 (flags) = tcp_reass((tp), (ti), (m)); \ 130 tp->t_flags |= TF_ACKNOW; \ 131 } \ 132} 133 134static int 135tcp_reass(tp, ti, m) 136 register struct tcpcb *tp; 137 register struct tcpiphdr *ti; 138 struct mbuf *m; 139{ 140 struct mbuf *q; 141 struct mbuf *p; 142 struct mbuf *nq; 143 struct socket *so = tp->t_inpcb->inp_socket; 144 int flags; 145 146#define GETTCP(m) ((struct tcpiphdr *)m->m_pkthdr.header) 147 148 /* 149 * Call with ti==0 after become established to 150 * force pre-ESTABLISHED data up to user socket. 151 */ 152 if (ti == 0) 153 goto present; 154 155 m->m_pkthdr.header = ti; 156 157 /* 158 * Find a segment which begins after this one does. 159 */ 160 for (q = tp->t_segq, p = NULL; q; p = q, q = q->m_nextpkt) 161 if (SEQ_GT(GETTCP(q)->ti_seq, ti->ti_seq)) 162 break; 163 164 /* 165 * If there is a preceding segment, it may provide some of 166 * our data already. If so, drop the data from the incoming 167 * segment. If it provides all of our data, drop us. 168 */ 169 if (p != NULL) { 170 register int i; 171 /* conversion to int (in i) handles seq wraparound */ 172 i = GETTCP(p)->ti_seq + GETTCP(p)->ti_len - ti->ti_seq; 173 if (i > 0) { 174 if (i >= ti->ti_len) { 175 tcpstat.tcps_rcvduppack++; 176 tcpstat.tcps_rcvdupbyte += ti->ti_len; 177 m_freem(m); 178 /* 179 * Try to present any queued data 180 * at the left window edge to the user. 181 * This is needed after the 3-WHS 182 * completes. 183 */ 184 goto present; /* ??? */ 185 } 186 m_adj(m, i); 187 ti->ti_len -= i; 188 ti->ti_seq += i; 189 } 190 } 191 tcpstat.tcps_rcvoopack++; 192 tcpstat.tcps_rcvoobyte += ti->ti_len; 193 194 /* 195 * While we overlap succeeding segments trim them or, 196 * if they are completely covered, dequeue them. 197 */ 198 while (q) { 199 register int i = (ti->ti_seq + ti->ti_len) - GETTCP(q)->ti_seq; 200 if (i <= 0) 201 break; 202 if (i < GETTCP(q)->ti_len) { 203 GETTCP(q)->ti_seq += i; 204 GETTCP(q)->ti_len -= i; 205 m_adj(q, i); 206 break; 207 } 208 209 nq = q->m_nextpkt; 210 if (p) 211 p->m_nextpkt = nq; 212 else 213 tp->t_segq = nq; 214 m_freem(q); 215 q = nq; 216 } 217 218 if (p == NULL) { 219 m->m_nextpkt = tp->t_segq; 220 tp->t_segq = m; 221 } else { 222 m->m_nextpkt = p->m_nextpkt; 223 p->m_nextpkt = m; 224 } 225 226present: 227 /* 228 * Present data to user, advancing rcv_nxt through 229 * completed sequence space. 230 */ 231 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 232 return (0); 233 q = tp->t_segq; 234 if (!q || GETTCP(q)->ti_seq != tp->rcv_nxt) 235 return (0); 236 do { 237 tp->rcv_nxt += GETTCP(q)->ti_len; 238 flags = GETTCP(q)->ti_flags & TH_FIN; 239 nq = q->m_nextpkt; 240 tp->t_segq = nq; 241 q->m_nextpkt = NULL; 242 if (so->so_state & SS_CANTRCVMORE) 243 m_freem(q); 244 else 245 sbappend(&so->so_rcv, q); 246 q = nq; 247 } while (q && GETTCP(q)->ti_seq == tp->rcv_nxt); 248 sorwakeup(so); 249 return (flags); 250 251#undef GETTCP 252} 253 254/* 255 * TCP input routine, follows pages 65-76 of the 256 * protocol specification dated September, 1981 very closely. 257 */ 258void 259tcp_input(m, iphlen) 260 register struct mbuf *m; 261 int iphlen; 262{ 263 register struct tcpiphdr *ti; 264 register struct inpcb *inp; 265 u_char *optp = NULL; 266 int optlen = 0; 267 int len, tlen, off; 268 register struct tcpcb *tp = 0; 269 register int tiflags; 270 struct socket *so = 0; 271 int todrop, acked, ourfinisacked, needoutput = 0; 272 struct in_addr laddr; 273 int dropsocket = 0; 274 int iss = 0; 275 u_long tiwin; 276 struct tcpopt to; /* options in this segment */ 277 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 278 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 279#ifdef TCPDEBUG 280 short ostate = 0; 281#endif 282 283 bzero((char *)&to, sizeof(to)); 284 285 tcpstat.tcps_rcvtotal++; 286 /* 287 * Get IP and TCP header together in first mbuf. 288 * Note: IP leaves IP header in first mbuf. 289 */ 290 ti = mtod(m, struct tcpiphdr *); 291 if (iphlen > sizeof (struct ip)) 292 ip_stripoptions(m, (struct mbuf *)0); 293 if (m->m_len < sizeof (struct tcpiphdr)) { 294 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 295 tcpstat.tcps_rcvshort++; 296 return; 297 } 298 ti = mtod(m, struct tcpiphdr *); 299 } 300 301 /* 302 * Checksum extended TCP header and data. 303 */ 304 tlen = ((struct ip *)ti)->ip_len; 305 len = sizeof (struct ip) + tlen; 306 bzero(ti->ti_x1, sizeof(ti->ti_x1)); 307 ti->ti_len = (u_short)tlen; 308 HTONS(ti->ti_len); 309 ti->ti_sum = in_cksum(m, len); 310 if (ti->ti_sum) { 311 tcpstat.tcps_rcvbadsum++; 312 goto drop; 313 } 314 315 /* 316 * Check that TCP offset makes sense, 317 * pull out TCP options and adjust length. XXX 318 */ 319 off = ti->ti_off << 2; 320 if (off < sizeof (struct tcphdr) || off > tlen) { 321 tcpstat.tcps_rcvbadoff++; 322 goto drop; 323 } 324 tlen -= off; 325 ti->ti_len = tlen; 326 if (off > sizeof (struct tcphdr)) { 327 if (m->m_len < sizeof(struct ip) + off) { 328 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 329 tcpstat.tcps_rcvshort++; 330 return; 331 } 332 ti = mtod(m, struct tcpiphdr *); 333 } 334 optlen = off - sizeof (struct tcphdr); 335 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 336 } 337 tiflags = ti->ti_flags; 338 339 /* 340 * Convert TCP protocol specific fields to host format. 341 */ 342 NTOHL(ti->ti_seq); 343 NTOHL(ti->ti_ack); 344 NTOHS(ti->ti_win); 345 NTOHS(ti->ti_urp); 346 347 /* 348 * Drop TCP, IP headers and TCP options. 349 */ 350 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 351 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 352 353 /* 354 * Locate pcb for segment. 355 */ 356findpcb: 357#ifdef IPFIREWALL_FORWARD 358 if (ip_fw_fwd_addr != NULL) { 359 /* 360 * Diverted. Pretend to be the destination. 361 * already got one like this? 362 */ 363 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, ti->ti_sport, 364 ti->ti_dst, ti->ti_dport, 0); 365 if (!inp) { 366 /* 367 * No, then it's new. Try find the ambushing socket 368 */ 369 if (!ip_fw_fwd_addr->sin_port) { 370 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, 371 ti->ti_sport, ip_fw_fwd_addr->sin_addr, 372 ti->ti_dport, 1); 373 } else { 374 inp = in_pcblookup_hash(&tcbinfo, 375 ti->ti_src, ti->ti_sport, 376 ip_fw_fwd_addr->sin_addr, 377 ntohs(ip_fw_fwd_addr->sin_port), 1); 378 } 379 } 380 ip_fw_fwd_addr = NULL; 381 } else 382#endif /* IPFIREWALL_FORWARD */ 383 384 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, ti->ti_sport, 385 ti->ti_dst, ti->ti_dport, 1); 386 387 /* 388 * If the state is CLOSED (i.e., TCB does not exist) then 389 * all data in the incoming segment is discarded. 390 * If the TCB exists but is in CLOSED state, it is embryonic, 391 * but should either do a listen or a connect soon. 392 */ 393 if (inp == NULL) { 394 if (log_in_vain && tiflags & TH_SYN) { 395 char buf[4*sizeof "123"]; 396 397 strcpy(buf, inet_ntoa(ti->ti_dst)); 398 log(LOG_INFO, 399 "Connection attempt to TCP %s:%d from %s:%d\n", 400 buf, ntohs(ti->ti_dport), inet_ntoa(ti->ti_src), 401 ntohs(ti->ti_sport)); 402 } 403#ifdef ICMP_BANDLIM 404 if (badport_bandlim(1) < 0) 405 goto drop; 406#endif 407 goto dropwithreset; 408 } 409 tp = intotcpcb(inp); 410 if (tp == 0) 411 goto dropwithreset; 412 if (tp->t_state == TCPS_CLOSED) 413 goto drop; 414 415 /* Unscale the window into a 32-bit value. */ 416 if ((tiflags & TH_SYN) == 0) 417 tiwin = ti->ti_win << tp->snd_scale; 418 else 419 tiwin = ti->ti_win; 420 421 so = inp->inp_socket; 422 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 423#ifdef TCPDEBUG 424 if (so->so_options & SO_DEBUG) { 425 ostate = tp->t_state; 426 tcp_saveti = *ti; 427 } 428#endif 429 if (so->so_options & SO_ACCEPTCONN) { 430 register struct tcpcb *tp0 = tp; 431 struct socket *so2; 432 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 433 /* 434 * Note: dropwithreset makes sure we don't 435 * send a RST in response to a RST. 436 */ 437 if (tiflags & TH_ACK) { 438 tcpstat.tcps_badsyn++; 439 goto dropwithreset; 440 } 441 goto drop; 442 } 443 so2 = sonewconn(so, 0); 444 if (so2 == 0) { 445 tcpstat.tcps_listendrop++; 446 so2 = sodropablereq(so); 447 if (so2) { 448 tcp_drop(sototcpcb(so2), ETIMEDOUT); 449 so2 = sonewconn(so, 0); 450 } 451 if (!so2) 452 goto drop; 453 } 454 so = so2; 455 /* 456 * This is ugly, but .... 457 * 458 * Mark socket as temporary until we're 459 * committed to keeping it. The code at 460 * ``drop'' and ``dropwithreset'' check the 461 * flag dropsocket to see if the temporary 462 * socket created here should be discarded. 463 * We mark the socket as discardable until 464 * we're committed to it below in TCPS_LISTEN. 465 */ 466 dropsocket++; 467 inp = (struct inpcb *)so->so_pcb; 468 inp->inp_laddr = ti->ti_dst; 469 inp->inp_lport = ti->ti_dport; 470 if (in_pcbinshash(inp) != 0) { 471 /* 472 * Undo the assignments above if we failed to put 473 * the PCB on the hash lists. 474 */ 475 inp->inp_laddr.s_addr = INADDR_ANY; 476 inp->inp_lport = 0; 477 goto drop; 478 } 479 inp->inp_options = ip_srcroute(); 480 tp = intotcpcb(inp); 481 tp->t_state = TCPS_LISTEN; 482 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 483 484 /* Compute proper scaling value from buffer space */ 485 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 486 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 487 tp->request_r_scale++; 488 } 489 } 490 491 /* 492 * Segment received on connection. 493 * Reset idle time and keep-alive timer. 494 */ 495 tp->t_idle = 0; 496 if (TCPS_HAVEESTABLISHED(tp->t_state)) 497 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 498 499 /* 500 * Process options if not in LISTEN state, 501 * else do it below (after getting remote address). 502 */ 503 if (tp->t_state != TCPS_LISTEN) 504 tcp_dooptions(tp, optp, optlen, ti, &to); 505 506 /* 507 * Header prediction: check for the two common cases 508 * of a uni-directional data xfer. If the packet has 509 * no control flags, is in-sequence, the window didn't 510 * change and we're not retransmitting, it's a 511 * candidate. If the length is zero and the ack moved 512 * forward, we're the sender side of the xfer. Just 513 * free the data acked & wake any higher level process 514 * that was blocked waiting for space. If the length 515 * is non-zero and the ack didn't move, we're the 516 * receiver side. If we're getting packets in-order 517 * (the reassembly queue is empty), add the data to 518 * the socket buffer and note that we need a delayed ack. 519 * Make sure that the hidden state-flags are also off. 520 * Since we check for TCPS_ESTABLISHED above, it can only 521 * be TH_NEEDSYN. 522 */ 523 if (tp->t_state == TCPS_ESTABLISHED && 524 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 525 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 526 ((to.to_flag & TOF_TS) == 0 || 527 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 528 /* 529 * Using the CC option is compulsory if once started: 530 * the segment is OK if no T/TCP was negotiated or 531 * if the segment has a CC option equal to CCrecv 532 */ 533 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 534 ((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 535 ti->ti_seq == tp->rcv_nxt && 536 tiwin && tiwin == tp->snd_wnd && 537 tp->snd_nxt == tp->snd_max) { 538 539 /* 540 * If last ACK falls within this segment's sequence numbers, 541 * record the timestamp. 542 * NOTE that the test is modified according to the latest 543 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 544 */ 545 if ((to.to_flag & TOF_TS) != 0 && 546 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 547 tp->ts_recent_age = tcp_now; 548 tp->ts_recent = to.to_tsval; 549 } 550 551 if (ti->ti_len == 0) { 552 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 553 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 554 tp->snd_cwnd >= tp->snd_wnd && 555 tp->t_dupacks < tcprexmtthresh) { 556 /* 557 * this is a pure ack for outstanding data. 558 */ 559 ++tcpstat.tcps_predack; 560 if ((to.to_flag & TOF_TS) != 0) 561 tcp_xmit_timer(tp, 562 tcp_now - to.to_tsecr + 1); 563 else if (tp->t_rtt && 564 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 565 tcp_xmit_timer(tp, tp->t_rtt); 566 acked = ti->ti_ack - tp->snd_una; 567 tcpstat.tcps_rcvackpack++; 568 tcpstat.tcps_rcvackbyte += acked; 569 sbdrop(&so->so_snd, acked); 570 tp->snd_una = ti->ti_ack; 571 m_freem(m); 572 573 /* 574 * If all outstanding data are acked, stop 575 * retransmit timer, otherwise restart timer 576 * using current (possibly backed-off) value. 577 * If process is waiting for space, 578 * wakeup/selwakeup/signal. If data 579 * are ready to send, let tcp_output 580 * decide between more output or persist. 581 */ 582 if (tp->snd_una == tp->snd_max) 583 tp->t_timer[TCPT_REXMT] = 0; 584 else if (tp->t_timer[TCPT_PERSIST] == 0) 585 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 586 587 sowwakeup(so); 588 if (so->so_snd.sb_cc) 589 (void) tcp_output(tp); 590 return; 591 } 592 } else if (ti->ti_ack == tp->snd_una && 593 tp->t_segq == NULL && 594 ti->ti_len <= sbspace(&so->so_rcv)) { 595 /* 596 * this is a pure, in-sequence data packet 597 * with nothing on the reassembly queue and 598 * we have enough buffer space to take it. 599 */ 600 ++tcpstat.tcps_preddat; 601 tp->rcv_nxt += ti->ti_len; 602 tcpstat.tcps_rcvpack++; 603 tcpstat.tcps_rcvbyte += ti->ti_len; 604 /* 605 * Add data to socket buffer. 606 */ 607 sbappend(&so->so_rcv, m); 608 sorwakeup(so); 609 if (tcp_delack_enabled) { 610 tp->t_flags |= TF_DELACK; 611 } else { 612 tp->t_flags |= TF_ACKNOW; 613 tcp_output(tp); 614 } 615 return; 616 } 617 } 618 619 /* 620 * Calculate amount of space in receive window, 621 * and then do TCP input processing. 622 * Receive window is amount of space in rcv queue, 623 * but not less than advertised window. 624 */ 625 { int win; 626 627 win = sbspace(&so->so_rcv); 628 if (win < 0) 629 win = 0; 630 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 631 } 632 633 switch (tp->t_state) { 634 635 /* 636 * If the state is LISTEN then ignore segment if it contains an RST. 637 * If the segment contains an ACK then it is bad and send a RST. 638 * If it does not contain a SYN then it is not interesting; drop it. 639 * If it is from this socket, drop it, it must be forged. 640 * Don't bother responding if the destination was a broadcast. 641 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 642 * tp->iss, and send a segment: 643 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 644 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 645 * Fill in remote peer address fields if not previously specified. 646 * Enter SYN_RECEIVED state, and process any other fields of this 647 * segment in this state. 648 */ 649 case TCPS_LISTEN: { 650 register struct sockaddr_in *sin; 651 652 if (tiflags & TH_RST) 653 goto drop; 654 if (tiflags & TH_ACK) 655 goto dropwithreset; 656 if ((tiflags & TH_SYN) == 0) 657 goto drop; 658 if ((ti->ti_dport == ti->ti_sport) && 659 (ti->ti_dst.s_addr == ti->ti_src.s_addr)) 660 goto drop; 661 /* 662 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 663 * in_broadcast() should never return true on a received 664 * packet with M_BCAST not set. 665 */ 666 if (m->m_flags & (M_BCAST|M_MCAST) || 667 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 668 goto drop; 669 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 670 M_NOWAIT); 671 if (sin == NULL) 672 goto drop; 673 sin->sin_family = AF_INET; 674 sin->sin_len = sizeof(*sin); 675 sin->sin_addr = ti->ti_src; 676 sin->sin_port = ti->ti_sport; 677 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 678 laddr = inp->inp_laddr; 679 if (inp->inp_laddr.s_addr == INADDR_ANY) 680 inp->inp_laddr = ti->ti_dst; 681 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 682 inp->inp_laddr = laddr; 683 FREE(sin, M_SONAME); 684 goto drop; 685 } 686 FREE(sin, M_SONAME); 687 tp->t_template = tcp_template(tp); 688 if (tp->t_template == 0) { 689 tp = tcp_drop(tp, ENOBUFS); 690 dropsocket = 0; /* socket is already gone */ 691 goto drop; 692 } 693 if ((taop = tcp_gettaocache(inp)) == NULL) { 694 taop = &tao_noncached; 695 bzero(taop, sizeof(*taop)); 696 } 697 tcp_dooptions(tp, optp, optlen, ti, &to); 698 if (iss) 699 tp->iss = iss; 700 else 701 tp->iss = tcp_iss; 702 tcp_iss += TCP_ISSINCR/4; 703 tp->irs = ti->ti_seq; 704 tcp_sendseqinit(tp); 705 tcp_rcvseqinit(tp); 706 /* 707 * Initialization of the tcpcb for transaction; 708 * set SND.WND = SEG.WND, 709 * initialize CCsend and CCrecv. 710 */ 711 tp->snd_wnd = tiwin; /* initial send-window */ 712 tp->cc_send = CC_INC(tcp_ccgen); 713 tp->cc_recv = to.to_cc; 714 /* 715 * Perform TAO test on incoming CC (SEG.CC) option, if any. 716 * - compare SEG.CC against cached CC from the same host, 717 * if any. 718 * - if SEG.CC > chached value, SYN must be new and is accepted 719 * immediately: save new CC in the cache, mark the socket 720 * connected, enter ESTABLISHED state, turn on flag to 721 * send a SYN in the next segment. 722 * A virtual advertised window is set in rcv_adv to 723 * initialize SWS prevention. Then enter normal segment 724 * processing: drop SYN, process data and FIN. 725 * - otherwise do a normal 3-way handshake. 726 */ 727 if ((to.to_flag & TOF_CC) != 0) { 728 if (((tp->t_flags & TF_NOPUSH) != 0) && 729 taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 730 731 taop->tao_cc = to.to_cc; 732 tp->t_state = TCPS_ESTABLISHED; 733 734 /* 735 * If there is a FIN, or if there is data and the 736 * connection is local, then delay SYN,ACK(SYN) in 737 * the hope of piggy-backing it on a response 738 * segment. Otherwise must send ACK now in case 739 * the other side is slow starting. 740 */ 741 if (tcp_delack_enabled && ((tiflags & TH_FIN) || (ti->ti_len != 0 && 742 in_localaddr(inp->inp_faddr)))) 743 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 744 else 745 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 746 747 /* 748 * Limit the `virtual advertised window' to TCP_MAXWIN 749 * here. Even if we requested window scaling, it will 750 * become effective only later when our SYN is acked. 751 */ 752 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 753 tcpstat.tcps_connects++; 754 soisconnected(so); 755 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 756 dropsocket = 0; /* committed to socket */ 757 tcpstat.tcps_accepts++; 758 goto trimthenstep6; 759 } 760 /* else do standard 3-way handshake */ 761 } else { 762 /* 763 * No CC option, but maybe CC.NEW: 764 * invalidate cached value. 765 */ 766 taop->tao_cc = 0; 767 } 768 /* 769 * TAO test failed or there was no CC option, 770 * do a standard 3-way handshake. 771 */ 772 tp->t_flags |= TF_ACKNOW; 773 tp->t_state = TCPS_SYN_RECEIVED; 774 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 775 dropsocket = 0; /* committed to socket */ 776 tcpstat.tcps_accepts++; 777 goto trimthenstep6; 778 } 779 780 /* 781 * If the state is SYN_RECEIVED: 782 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 783 */ 784 case TCPS_SYN_RECEIVED: 785 if ((tiflags & TH_ACK) && 786 (SEQ_LEQ(ti->ti_ack, tp->snd_una) || 787 SEQ_GT(ti->ti_ack, tp->snd_max))) 788 goto dropwithreset; 789 break; 790 791 /* 792 * If the state is SYN_SENT: 793 * if seg contains an ACK, but not for our SYN, drop the input. 794 * if seg contains a RST, then drop the connection. 795 * if seg does not contain SYN, then drop it. 796 * Otherwise this is an acceptable SYN segment 797 * initialize tp->rcv_nxt and tp->irs 798 * if seg contains ack then advance tp->snd_una 799 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 800 * arrange for segment to be acked (eventually) 801 * continue processing rest of data/controls, beginning with URG 802 */ 803 case TCPS_SYN_SENT: 804 if ((taop = tcp_gettaocache(inp)) == NULL) { 805 taop = &tao_noncached; 806 bzero(taop, sizeof(*taop)); 807 } 808 809 if ((tiflags & TH_ACK) && 810 (SEQ_LEQ(ti->ti_ack, tp->iss) || 811 SEQ_GT(ti->ti_ack, tp->snd_max))) { 812 /* 813 * If we have a cached CCsent for the remote host, 814 * hence we haven't just crashed and restarted, 815 * do not send a RST. This may be a retransmission 816 * from the other side after our earlier ACK was lost. 817 * Our new SYN, when it arrives, will serve as the 818 * needed ACK. 819 */ 820 if (taop->tao_ccsent != 0) 821 goto drop; 822 else 823 goto dropwithreset; 824 } 825 if (tiflags & TH_RST) { 826 if (tiflags & TH_ACK) 827 tp = tcp_drop(tp, ECONNREFUSED); 828 goto drop; 829 } 830 if ((tiflags & TH_SYN) == 0) 831 goto drop; 832 tp->snd_wnd = ti->ti_win; /* initial send window */ 833 tp->cc_recv = to.to_cc; /* foreign CC */ 834 835 tp->irs = ti->ti_seq; 836 tcp_rcvseqinit(tp); 837 if (tiflags & TH_ACK) { 838 /* 839 * Our SYN was acked. If segment contains CC.ECHO 840 * option, check it to make sure this segment really 841 * matches our SYN. If not, just drop it as old 842 * duplicate, but send an RST if we're still playing 843 * by the old rules. If no CC.ECHO option, make sure 844 * we don't get fooled into using T/TCP. 845 */ 846 if (to.to_flag & TOF_CCECHO) { 847 if (tp->cc_send != to.to_ccecho) 848 if (taop->tao_ccsent != 0) 849 goto drop; 850 else 851 goto dropwithreset; 852 } else 853 tp->t_flags &= ~TF_RCVD_CC; 854 tcpstat.tcps_connects++; 855 soisconnected(so); 856 /* Do window scaling on this connection? */ 857 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 858 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 859 tp->snd_scale = tp->requested_s_scale; 860 tp->rcv_scale = tp->request_r_scale; 861 } 862 /* Segment is acceptable, update cache if undefined. */ 863 if (taop->tao_ccsent == 0) 864 taop->tao_ccsent = to.to_ccecho; 865 866 tp->rcv_adv += tp->rcv_wnd; 867 tp->snd_una++; /* SYN is acked */ 868 /* 869 * If there's data, delay ACK; if there's also a FIN 870 * ACKNOW will be turned on later. 871 */ 872 if (tcp_delack_enabled && ti->ti_len != 0) 873 tp->t_flags |= TF_DELACK; 874 else 875 tp->t_flags |= TF_ACKNOW; 876 /* 877 * Received <SYN,ACK> in SYN_SENT[*] state. 878 * Transitions: 879 * SYN_SENT --> ESTABLISHED 880 * SYN_SENT* --> FIN_WAIT_1 881 */ 882 if (tp->t_flags & TF_NEEDFIN) { 883 tp->t_state = TCPS_FIN_WAIT_1; 884 tp->t_flags &= ~TF_NEEDFIN; 885 tiflags &= ~TH_SYN; 886 } else { 887 tp->t_state = TCPS_ESTABLISHED; 888 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 889 } 890 } else { 891 /* 892 * Received initial SYN in SYN-SENT[*] state => simul- 893 * taneous open. If segment contains CC option and there is 894 * a cached CC, apply TAO test; if it succeeds, connection is 895 * half-synchronized. Otherwise, do 3-way handshake: 896 * SYN-SENT -> SYN-RECEIVED 897 * SYN-SENT* -> SYN-RECEIVED* 898 * If there was no CC option, clear cached CC value. 899 */ 900 tp->t_flags |= TF_ACKNOW; 901 tp->t_timer[TCPT_REXMT] = 0; 902 if (to.to_flag & TOF_CC) { 903 if (taop->tao_cc != 0 && 904 CC_GT(to.to_cc, taop->tao_cc)) { 905 /* 906 * update cache and make transition: 907 * SYN-SENT -> ESTABLISHED* 908 * SYN-SENT* -> FIN-WAIT-1* 909 */ 910 taop->tao_cc = to.to_cc; 911 if (tp->t_flags & TF_NEEDFIN) { 912 tp->t_state = TCPS_FIN_WAIT_1; 913 tp->t_flags &= ~TF_NEEDFIN; 914 } else { 915 tp->t_state = TCPS_ESTABLISHED; 916 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 917 } 918 tp->t_flags |= TF_NEEDSYN; 919 } else 920 tp->t_state = TCPS_SYN_RECEIVED; 921 } else { 922 /* CC.NEW or no option => invalidate cache */ 923 taop->tao_cc = 0; 924 tp->t_state = TCPS_SYN_RECEIVED; 925 } 926 } 927 928trimthenstep6: 929 /* 930 * Advance ti->ti_seq to correspond to first data byte. 931 * If data, trim to stay within window, 932 * dropping FIN if necessary. 933 */ 934 ti->ti_seq++; 935 if (ti->ti_len > tp->rcv_wnd) { 936 todrop = ti->ti_len - tp->rcv_wnd; 937 m_adj(m, -todrop); 938 ti->ti_len = tp->rcv_wnd; 939 tiflags &= ~TH_FIN; 940 tcpstat.tcps_rcvpackafterwin++; 941 tcpstat.tcps_rcvbyteafterwin += todrop; 942 } 943 tp->snd_wl1 = ti->ti_seq - 1; 944 tp->rcv_up = ti->ti_seq; 945 /* 946 * Client side of transaction: already sent SYN and data. 947 * If the remote host used T/TCP to validate the SYN, 948 * our data will be ACK'd; if so, enter normal data segment 949 * processing in the middle of step 5, ack processing. 950 * Otherwise, goto step 6. 951 */ 952 if (tiflags & TH_ACK) 953 goto process_ACK; 954 goto step6; 955 /* 956 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 957 * if segment contains a SYN and CC [not CC.NEW] option: 958 * if state == TIME_WAIT and connection duration > MSL, 959 * drop packet and send RST; 960 * 961 * if SEG.CC > CCrecv then is new SYN, and can implicitly 962 * ack the FIN (and data) in retransmission queue. 963 * Complete close and delete TCPCB. Then reprocess 964 * segment, hoping to find new TCPCB in LISTEN state; 965 * 966 * else must be old SYN; drop it. 967 * else do normal processing. 968 */ 969 case TCPS_LAST_ACK: 970 case TCPS_CLOSING: 971 case TCPS_TIME_WAIT: 972 if ((tiflags & TH_SYN) && 973 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 974 if (tp->t_state == TCPS_TIME_WAIT && 975 tp->t_duration > TCPTV_MSL) 976 goto dropwithreset; 977 if (CC_GT(to.to_cc, tp->cc_recv)) { 978 tp = tcp_close(tp); 979 goto findpcb; 980 } 981 else 982 goto drop; 983 } 984 break; /* continue normal processing */ 985 } 986 987 /* 988 * States other than LISTEN or SYN_SENT. 989 * First check the RST flag and sequence number since reset segments 990 * are exempt from the timestamp and connection count tests. This 991 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 992 * below which allowed reset segments in half the sequence space 993 * to fall though and be processed (which gives forged reset 994 * segments with a random sequence number a 50 percent chance of 995 * killing a connection). 996 * Then check timestamp, if present. 997 * Then check the connection count, if present. 998 * Then check that at least some bytes of segment are within 999 * receive window. If segment begins before rcv_nxt, 1000 * drop leading data (and SYN); if nothing left, just ack. 1001 * 1002 * 1003 * If the RST bit is set, check the sequence number to see 1004 * if this is a valid reset segment. 1005 * RFC 793 page 37: 1006 * In all states except SYN-SENT, all reset (RST) segments 1007 * are validated by checking their SEQ-fields. A reset is 1008 * valid if its sequence number is in the window. 1009 * Note: this does not take into account delayed ACKs, so 1010 * we should test against last_ack_sent instead of rcv_nxt. 1011 * Also, it does not make sense to allow reset segments with 1012 * sequence numbers greater than last_ack_sent to be processed 1013 * since these sequence numbers are just the acknowledgement 1014 * numbers in our outgoing packets being echoed back at us, 1015 * and these acknowledgement numbers are monotonically 1016 * increasing. 1017 * If we have multiple segments in flight, the intial reset 1018 * segment sequence numbers will be to the left of last_ack_sent, 1019 * but they will eventually catch up. 1020 * In any case, it never made sense to trim reset segments to 1021 * fit the receive window since RFC 1122 says: 1022 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1023 * 1024 * A TCP SHOULD allow a received RST segment to include data. 1025 * 1026 * DISCUSSION 1027 * It has been suggested that a RST segment could contain 1028 * ASCII text that encoded and explained the cause of the 1029 * RST. No standard has yet been established for such 1030 * data. 1031 * 1032 * If the reset segment passes the sequence number test examine 1033 * the state: 1034 * SYN_RECEIVED STATE: 1035 * If passive open, return to LISTEN state. 1036 * If active open, inform user that connection was refused. 1037 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1038 * Inform user that connection was reset, and close tcb. 1039 * CLOSING, LAST_ACK, TIME_WAIT STATES 1040 * Close the tcb. 1041 * TIME_WAIT state: 1042 * Drop the segment - see Stevens, vol. 2, p. 964 and 1043 * RFC 1337. 1044 */ 1045 if (tiflags & TH_RST) { 1046 if (tp->last_ack_sent == ti->ti_seq) { 1047 switch (tp->t_state) { 1048 1049 case TCPS_SYN_RECEIVED: 1050 so->so_error = ECONNREFUSED; 1051 goto close; 1052 1053 case TCPS_ESTABLISHED: 1054 case TCPS_FIN_WAIT_1: 1055 case TCPS_FIN_WAIT_2: 1056 case TCPS_CLOSE_WAIT: 1057 so->so_error = ECONNRESET; 1058 close: 1059 tp->t_state = TCPS_CLOSED; 1060 tcpstat.tcps_drops++; 1061 tp = tcp_close(tp); 1062 break; 1063 1064 case TCPS_CLOSING: 1065 case TCPS_LAST_ACK: 1066 tp = tcp_close(tp); 1067 break; 1068 1069 case TCPS_TIME_WAIT: 1070 break; 1071 } 1072 } 1073 goto drop; 1074 } 1075 1076 /* 1077 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1078 * and it's less than ts_recent, drop it. 1079 */ 1080 if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent && 1081 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1082 1083 /* Check to see if ts_recent is over 24 days old. */ 1084 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1085 /* 1086 * Invalidate ts_recent. If this segment updates 1087 * ts_recent, the age will be reset later and ts_recent 1088 * will get a valid value. If it does not, setting 1089 * ts_recent to zero will at least satisfy the 1090 * requirement that zero be placed in the timestamp 1091 * echo reply when ts_recent isn't valid. The 1092 * age isn't reset until we get a valid ts_recent 1093 * because we don't want out-of-order segments to be 1094 * dropped when ts_recent is old. 1095 */ 1096 tp->ts_recent = 0; 1097 } else { 1098 tcpstat.tcps_rcvduppack++; 1099 tcpstat.tcps_rcvdupbyte += ti->ti_len; 1100 tcpstat.tcps_pawsdrop++; 1101 goto dropafterack; 1102 } 1103 } 1104 1105 /* 1106 * T/TCP mechanism 1107 * If T/TCP was negotiated and the segment doesn't have CC, 1108 * or if its CC is wrong then drop the segment. 1109 * RST segments do not have to comply with this. 1110 */ 1111 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1112 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1113 goto dropafterack; 1114 1115 /* 1116 * In the SYN-RECEIVED state, validate that the packet belongs to 1117 * this connection before trimming the data to fit the receive 1118 * window. Check the sequence number versus IRS since we know 1119 * the sequence numbers haven't wrapped. This is a partial fix 1120 * for the "LAND" DoS attack. 1121 */ 1122 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(ti->ti_seq, tp->irs)) 1123 goto dropwithreset; 1124 1125 todrop = tp->rcv_nxt - ti->ti_seq; 1126 if (todrop > 0) { 1127 if (tiflags & TH_SYN) { 1128 tiflags &= ~TH_SYN; 1129 ti->ti_seq++; 1130 if (ti->ti_urp > 1) 1131 ti->ti_urp--; 1132 else 1133 tiflags &= ~TH_URG; 1134 todrop--; 1135 } 1136 /* 1137 * Following if statement from Stevens, vol. 2, p. 960. 1138 */ 1139 if (todrop > ti->ti_len 1140 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 1141 /* 1142 * Any valid FIN must be to the left of the window. 1143 * At this point the FIN must be a duplicate or out 1144 * of sequence; drop it. 1145 */ 1146 tiflags &= ~TH_FIN; 1147 1148 /* 1149 * Send an ACK to resynchronize and drop any data. 1150 * But keep on processing for RST or ACK. 1151 */ 1152 tp->t_flags |= TF_ACKNOW; 1153 todrop = ti->ti_len; 1154 tcpstat.tcps_rcvduppack++; 1155 tcpstat.tcps_rcvdupbyte += todrop; 1156 } else { 1157 tcpstat.tcps_rcvpartduppack++; 1158 tcpstat.tcps_rcvpartdupbyte += todrop; 1159 } 1160 m_adj(m, todrop); 1161 ti->ti_seq += todrop; 1162 ti->ti_len -= todrop; 1163 if (ti->ti_urp > todrop) 1164 ti->ti_urp -= todrop; 1165 else { 1166 tiflags &= ~TH_URG; 1167 ti->ti_urp = 0; 1168 } 1169 } 1170 1171 /* 1172 * If new data are received on a connection after the 1173 * user processes are gone, then RST the other end. 1174 */ 1175 if ((so->so_state & SS_NOFDREF) && 1176 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1177 tp = tcp_close(tp); 1178 tcpstat.tcps_rcvafterclose++; 1179 goto dropwithreset; 1180 } 1181 1182 /* 1183 * If segment ends after window, drop trailing data 1184 * (and PUSH and FIN); if nothing left, just ACK. 1185 */ 1186 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1187 if (todrop > 0) { 1188 tcpstat.tcps_rcvpackafterwin++; 1189 if (todrop >= ti->ti_len) { 1190 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1191 /* 1192 * If a new connection request is received 1193 * while in TIME_WAIT, drop the old connection 1194 * and start over if the sequence numbers 1195 * are above the previous ones. 1196 */ 1197 if (tiflags & TH_SYN && 1198 tp->t_state == TCPS_TIME_WAIT && 1199 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1200 iss = tp->snd_nxt + TCP_ISSINCR; 1201 tp = tcp_close(tp); 1202 goto findpcb; 1203 } 1204 /* 1205 * If window is closed can only take segments at 1206 * window edge, and have to drop data and PUSH from 1207 * incoming segments. Continue processing, but 1208 * remember to ack. Otherwise, drop segment 1209 * and ack. 1210 */ 1211 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1212 tp->t_flags |= TF_ACKNOW; 1213 tcpstat.tcps_rcvwinprobe++; 1214 } else 1215 goto dropafterack; 1216 } else 1217 tcpstat.tcps_rcvbyteafterwin += todrop; 1218 m_adj(m, -todrop); 1219 ti->ti_len -= todrop; 1220 tiflags &= ~(TH_PUSH|TH_FIN); 1221 } 1222 1223 /* 1224 * If last ACK falls within this segment's sequence numbers, 1225 * record its timestamp. 1226 * NOTE that the test is modified according to the latest 1227 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1228 */ 1229 if ((to.to_flag & TOF_TS) != 0 && 1230 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1231 tp->ts_recent_age = tcp_now; 1232 tp->ts_recent = to.to_tsval; 1233 } 1234 1235 /* 1236 * If a SYN is in the window, then this is an 1237 * error and we send an RST and drop the connection. 1238 */ 1239 if (tiflags & TH_SYN) { 1240 tp = tcp_drop(tp, ECONNRESET); 1241 goto dropwithreset; 1242 } 1243 1244 /* 1245 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1246 * flag is on (half-synchronized state), then queue data for 1247 * later processing; else drop segment and return. 1248 */ 1249 if ((tiflags & TH_ACK) == 0) { 1250 if (tp->t_state == TCPS_SYN_RECEIVED || 1251 (tp->t_flags & TF_NEEDSYN)) 1252 goto step6; 1253 else 1254 goto drop; 1255 } 1256 1257 /* 1258 * Ack processing. 1259 */ 1260 switch (tp->t_state) { 1261 1262 /* 1263 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1264 * ESTABLISHED state and continue processing. 1265 * The ACK was checked above. 1266 */ 1267 case TCPS_SYN_RECEIVED: 1268 1269 tcpstat.tcps_connects++; 1270 soisconnected(so); 1271 /* Do window scaling? */ 1272 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1273 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1274 tp->snd_scale = tp->requested_s_scale; 1275 tp->rcv_scale = tp->request_r_scale; 1276 } 1277 /* 1278 * Upon successful completion of 3-way handshake, 1279 * update cache.CC if it was undefined, pass any queued 1280 * data to the user, and advance state appropriately. 1281 */ 1282 if ((taop = tcp_gettaocache(inp)) != NULL && 1283 taop->tao_cc == 0) 1284 taop->tao_cc = tp->cc_recv; 1285 1286 /* 1287 * Make transitions: 1288 * SYN-RECEIVED -> ESTABLISHED 1289 * SYN-RECEIVED* -> FIN-WAIT-1 1290 */ 1291 if (tp->t_flags & TF_NEEDFIN) { 1292 tp->t_state = TCPS_FIN_WAIT_1; 1293 tp->t_flags &= ~TF_NEEDFIN; 1294 } else { 1295 tp->t_state = TCPS_ESTABLISHED; 1296 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 1297 } 1298 /* 1299 * If segment contains data or ACK, will call tcp_reass() 1300 * later; if not, do so now to pass queued data to user. 1301 */ 1302 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1303 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1304 (struct mbuf *)0); 1305 tp->snd_wl1 = ti->ti_seq - 1; 1306 /* fall into ... */ 1307 1308 /* 1309 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1310 * ACKs. If the ack is in the range 1311 * tp->snd_una < ti->ti_ack <= tp->snd_max 1312 * then advance tp->snd_una to ti->ti_ack and drop 1313 * data from the retransmission queue. If this ACK reflects 1314 * more up to date window information we update our window information. 1315 */ 1316 case TCPS_ESTABLISHED: 1317 case TCPS_FIN_WAIT_1: 1318 case TCPS_FIN_WAIT_2: 1319 case TCPS_CLOSE_WAIT: 1320 case TCPS_CLOSING: 1321 case TCPS_LAST_ACK: 1322 case TCPS_TIME_WAIT: 1323 1324 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1325 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1326 tcpstat.tcps_rcvdupack++; 1327 /* 1328 * If we have outstanding data (other than 1329 * a window probe), this is a completely 1330 * duplicate ack (ie, window info didn't 1331 * change), the ack is the biggest we've 1332 * seen and we've seen exactly our rexmt 1333 * threshhold of them, assume a packet 1334 * has been dropped and retransmit it. 1335 * Kludge snd_nxt & the congestion 1336 * window so we send only this one 1337 * packet. 1338 * 1339 * We know we're losing at the current 1340 * window size so do congestion avoidance 1341 * (set ssthresh to half the current window 1342 * and pull our congestion window back to 1343 * the new ssthresh). 1344 * 1345 * Dup acks mean that packets have left the 1346 * network (they're now cached at the receiver) 1347 * so bump cwnd by the amount in the receiver 1348 * to keep a constant cwnd packets in the 1349 * network. 1350 */ 1351 if (tp->t_timer[TCPT_REXMT] == 0 || 1352 ti->ti_ack != tp->snd_una) 1353 tp->t_dupacks = 0; 1354 else if (++tp->t_dupacks == tcprexmtthresh) { 1355 tcp_seq onxt = tp->snd_nxt; 1356 u_int win = 1357 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1358 tp->t_maxseg; 1359 1360 if (win < 2) 1361 win = 2; 1362 tp->snd_ssthresh = win * tp->t_maxseg; 1363 tp->t_timer[TCPT_REXMT] = 0; 1364 tp->t_rtt = 0; 1365 tp->snd_nxt = ti->ti_ack; 1366 tp->snd_cwnd = tp->t_maxseg; 1367 (void) tcp_output(tp); 1368 tp->snd_cwnd = tp->snd_ssthresh + 1369 tp->t_maxseg * tp->t_dupacks; 1370 if (SEQ_GT(onxt, tp->snd_nxt)) 1371 tp->snd_nxt = onxt; 1372 goto drop; 1373 } else if (tp->t_dupacks > tcprexmtthresh) { 1374 tp->snd_cwnd += tp->t_maxseg; 1375 (void) tcp_output(tp); 1376 goto drop; 1377 } 1378 } else 1379 tp->t_dupacks = 0; 1380 break; 1381 } 1382 /* 1383 * If the congestion window was inflated to account 1384 * for the other side's cached packets, retract it. 1385 */ 1386 if (tp->t_dupacks >= tcprexmtthresh && 1387 tp->snd_cwnd > tp->snd_ssthresh) 1388 tp->snd_cwnd = tp->snd_ssthresh; 1389 tp->t_dupacks = 0; 1390 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1391 tcpstat.tcps_rcvacktoomuch++; 1392 goto dropafterack; 1393 } 1394 /* 1395 * If we reach this point, ACK is not a duplicate, 1396 * i.e., it ACKs something we sent. 1397 */ 1398 if (tp->t_flags & TF_NEEDSYN) { 1399 /* 1400 * T/TCP: Connection was half-synchronized, and our 1401 * SYN has been ACK'd (so connection is now fully 1402 * synchronized). Go to non-starred state, 1403 * increment snd_una for ACK of SYN, and check if 1404 * we can do window scaling. 1405 */ 1406 tp->t_flags &= ~TF_NEEDSYN; 1407 tp->snd_una++; 1408 /* Do window scaling? */ 1409 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1410 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1411 tp->snd_scale = tp->requested_s_scale; 1412 tp->rcv_scale = tp->request_r_scale; 1413 } 1414 } 1415 1416process_ACK: 1417 acked = ti->ti_ack - tp->snd_una; 1418 tcpstat.tcps_rcvackpack++; 1419 tcpstat.tcps_rcvackbyte += acked; 1420 1421 /* 1422 * If we have a timestamp reply, update smoothed 1423 * round trip time. If no timestamp is present but 1424 * transmit timer is running and timed sequence 1425 * number was acked, update smoothed round trip time. 1426 * Since we now have an rtt measurement, cancel the 1427 * timer backoff (cf., Phil Karn's retransmit alg.). 1428 * Recompute the initial retransmit timer. 1429 */ 1430 if (to.to_flag & TOF_TS) 1431 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1432 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1433 tcp_xmit_timer(tp,tp->t_rtt); 1434 1435 /* 1436 * If all outstanding data is acked, stop retransmit 1437 * timer and remember to restart (more output or persist). 1438 * If there is more data to be acked, restart retransmit 1439 * timer, using current (possibly backed-off) value. 1440 */ 1441 if (ti->ti_ack == tp->snd_max) { 1442 tp->t_timer[TCPT_REXMT] = 0; 1443 needoutput = 1; 1444 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1445 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1446 1447 /* 1448 * If no data (only SYN) was ACK'd, 1449 * skip rest of ACK processing. 1450 */ 1451 if (acked == 0) 1452 goto step6; 1453 1454 /* 1455 * When new data is acked, open the congestion window. 1456 * If the window gives us less than ssthresh packets 1457 * in flight, open exponentially (maxseg per packet). 1458 * Otherwise open linearly: maxseg per window 1459 * (maxseg^2 / cwnd per packet). 1460 */ 1461 { 1462 register u_int cw = tp->snd_cwnd; 1463 register u_int incr = tp->t_maxseg; 1464 1465 if (cw > tp->snd_ssthresh) 1466 incr = incr * incr / cw; 1467 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1468 } 1469 if (acked > so->so_snd.sb_cc) { 1470 tp->snd_wnd -= so->so_snd.sb_cc; 1471 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1472 ourfinisacked = 1; 1473 } else { 1474 sbdrop(&so->so_snd, acked); 1475 tp->snd_wnd -= acked; 1476 ourfinisacked = 0; 1477 } 1478 sowwakeup(so); 1479 tp->snd_una = ti->ti_ack; 1480 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1481 tp->snd_nxt = tp->snd_una; 1482 1483 switch (tp->t_state) { 1484 1485 /* 1486 * In FIN_WAIT_1 STATE in addition to the processing 1487 * for the ESTABLISHED state if our FIN is now acknowledged 1488 * then enter FIN_WAIT_2. 1489 */ 1490 case TCPS_FIN_WAIT_1: 1491 if (ourfinisacked) { 1492 /* 1493 * If we can't receive any more 1494 * data, then closing user can proceed. 1495 * Starting the timer is contrary to the 1496 * specification, but if we don't get a FIN 1497 * we'll hang forever. 1498 */ 1499 if (so->so_state & SS_CANTRCVMORE) { 1500 soisdisconnected(so); 1501 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1502 } 1503 tp->t_state = TCPS_FIN_WAIT_2; 1504 } 1505 break; 1506 1507 /* 1508 * In CLOSING STATE in addition to the processing for 1509 * the ESTABLISHED state if the ACK acknowledges our FIN 1510 * then enter the TIME-WAIT state, otherwise ignore 1511 * the segment. 1512 */ 1513 case TCPS_CLOSING: 1514 if (ourfinisacked) { 1515 tp->t_state = TCPS_TIME_WAIT; 1516 tcp_canceltimers(tp); 1517 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1518 if (tp->cc_recv != 0 && 1519 tp->t_duration < TCPTV_MSL) 1520 tp->t_timer[TCPT_2MSL] = 1521 tp->t_rxtcur * TCPTV_TWTRUNC; 1522 else 1523 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1524 soisdisconnected(so); 1525 } 1526 break; 1527 1528 /* 1529 * In LAST_ACK, we may still be waiting for data to drain 1530 * and/or to be acked, as well as for the ack of our FIN. 1531 * If our FIN is now acknowledged, delete the TCB, 1532 * enter the closed state and return. 1533 */ 1534 case TCPS_LAST_ACK: 1535 if (ourfinisacked) { 1536 tp = tcp_close(tp); 1537 goto drop; 1538 } 1539 break; 1540 1541 /* 1542 * In TIME_WAIT state the only thing that should arrive 1543 * is a retransmission of the remote FIN. Acknowledge 1544 * it and restart the finack timer. 1545 */ 1546 case TCPS_TIME_WAIT: 1547 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1548 goto dropafterack; 1549 } 1550 } 1551 1552step6: 1553 /* 1554 * Update window information. 1555 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1556 */ 1557 if ((tiflags & TH_ACK) && 1558 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1559 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1560 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1561 /* keep track of pure window updates */ 1562 if (ti->ti_len == 0 && 1563 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1564 tcpstat.tcps_rcvwinupd++; 1565 tp->snd_wnd = tiwin; 1566 tp->snd_wl1 = ti->ti_seq; 1567 tp->snd_wl2 = ti->ti_ack; 1568 if (tp->snd_wnd > tp->max_sndwnd) 1569 tp->max_sndwnd = tp->snd_wnd; 1570 needoutput = 1; 1571 } 1572 1573 /* 1574 * Process segments with URG. 1575 */ 1576 if ((tiflags & TH_URG) && ti->ti_urp && 1577 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1578 /* 1579 * This is a kludge, but if we receive and accept 1580 * random urgent pointers, we'll crash in 1581 * soreceive. It's hard to imagine someone 1582 * actually wanting to send this much urgent data. 1583 */ 1584 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1585 ti->ti_urp = 0; /* XXX */ 1586 tiflags &= ~TH_URG; /* XXX */ 1587 goto dodata; /* XXX */ 1588 } 1589 /* 1590 * If this segment advances the known urgent pointer, 1591 * then mark the data stream. This should not happen 1592 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1593 * a FIN has been received from the remote side. 1594 * In these states we ignore the URG. 1595 * 1596 * According to RFC961 (Assigned Protocols), 1597 * the urgent pointer points to the last octet 1598 * of urgent data. We continue, however, 1599 * to consider it to indicate the first octet 1600 * of data past the urgent section as the original 1601 * spec states (in one of two places). 1602 */ 1603 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1604 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1605 so->so_oobmark = so->so_rcv.sb_cc + 1606 (tp->rcv_up - tp->rcv_nxt) - 1; 1607 if (so->so_oobmark == 0) 1608 so->so_state |= SS_RCVATMARK; 1609 sohasoutofband(so); 1610 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1611 } 1612 /* 1613 * Remove out of band data so doesn't get presented to user. 1614 * This can happen independent of advancing the URG pointer, 1615 * but if two URG's are pending at once, some out-of-band 1616 * data may creep in... ick. 1617 */ 1618 if (ti->ti_urp <= (u_long)ti->ti_len 1619#ifdef SO_OOBINLINE 1620 && (so->so_options & SO_OOBINLINE) == 0 1621#endif 1622 ) 1623 tcp_pulloutofband(so, ti, m); 1624 } else 1625 /* 1626 * If no out of band data is expected, 1627 * pull receive urgent pointer along 1628 * with the receive window. 1629 */ 1630 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1631 tp->rcv_up = tp->rcv_nxt; 1632dodata: /* XXX */ 1633 1634 /* 1635 * Process the segment text, merging it into the TCP sequencing queue, 1636 * and arranging for acknowledgment of receipt if necessary. 1637 * This process logically involves adjusting tp->rcv_wnd as data 1638 * is presented to the user (this happens in tcp_usrreq.c, 1639 * case PRU_RCVD). If a FIN has already been received on this 1640 * connection then we just ignore the text. 1641 */ 1642 if ((ti->ti_len || (tiflags&TH_FIN)) && 1643 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1644 TCP_REASS(tp, ti, m, so, tiflags); 1645 /* 1646 * Note the amount of data that peer has sent into 1647 * our window, in order to estimate the sender's 1648 * buffer size. 1649 */ 1650 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1651 } else { 1652 m_freem(m); 1653 tiflags &= ~TH_FIN; 1654 } 1655 1656 /* 1657 * If FIN is received ACK the FIN and let the user know 1658 * that the connection is closing. 1659 */ 1660 if (tiflags & TH_FIN) { 1661 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1662 socantrcvmore(so); 1663 /* 1664 * If connection is half-synchronized 1665 * (ie NEEDSYN flag on) then delay ACK, 1666 * so it may be piggybacked when SYN is sent. 1667 * Otherwise, since we received a FIN then no 1668 * more input can be expected, send ACK now. 1669 */ 1670 if (tcp_delack_enabled && (tp->t_flags & TF_NEEDSYN)) 1671 tp->t_flags |= TF_DELACK; 1672 else 1673 tp->t_flags |= TF_ACKNOW; 1674 tp->rcv_nxt++; 1675 } 1676 switch (tp->t_state) { 1677 1678 /* 1679 * In SYN_RECEIVED and ESTABLISHED STATES 1680 * enter the CLOSE_WAIT state. 1681 */ 1682 case TCPS_SYN_RECEIVED: 1683 case TCPS_ESTABLISHED: 1684 tp->t_state = TCPS_CLOSE_WAIT; 1685 break; 1686 1687 /* 1688 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1689 * enter the CLOSING state. 1690 */ 1691 case TCPS_FIN_WAIT_1: 1692 tp->t_state = TCPS_CLOSING; 1693 break; 1694 1695 /* 1696 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1697 * starting the time-wait timer, turning off the other 1698 * standard timers. 1699 */ 1700 case TCPS_FIN_WAIT_2: 1701 tp->t_state = TCPS_TIME_WAIT; 1702 tcp_canceltimers(tp); 1703 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1704 if (tp->cc_recv != 0 && 1705 tp->t_duration < TCPTV_MSL) { 1706 tp->t_timer[TCPT_2MSL] = 1707 tp->t_rxtcur * TCPTV_TWTRUNC; 1708 /* For transaction client, force ACK now. */ 1709 tp->t_flags |= TF_ACKNOW; 1710 } 1711 else 1712 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1713 soisdisconnected(so); 1714 break; 1715 1716 /* 1717 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1718 */ 1719 case TCPS_TIME_WAIT: 1720 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1721 break; 1722 } 1723 } 1724#ifdef TCPDEBUG 1725 if (so->so_options & SO_DEBUG) 1726 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1727#endif 1728 1729 /* 1730 * Return any desired output. 1731 */ 1732 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1733 (void) tcp_output(tp); 1734 return; 1735 1736dropafterack: 1737 /* 1738 * Generate an ACK dropping incoming segment if it occupies 1739 * sequence space, where the ACK reflects our state. 1740 * 1741 * We can now skip the test for the RST flag since all 1742 * paths to this code happen after packets containing 1743 * RST have been dropped. 1744 * 1745 * In the SYN-RECEIVED state, don't send an ACK unless the 1746 * segment we received passes the SYN-RECEIVED ACK test. 1747 * If it fails send a RST. This breaks the loop in the 1748 * "LAND" DoS attack, and also prevents an ACK storm 1749 * between two listening ports that have been sent forged 1750 * SYN segments, each with the source address of the other. 1751 */ 1752 if (tp->t_state == TCPS_SYN_RECEIVED && (tiflags & TH_ACK) && 1753 (SEQ_GT(tp->snd_una, ti->ti_ack) || 1754 SEQ_GT(ti->ti_ack, tp->snd_max)) ) 1755 goto dropwithreset; 1756#ifdef TCPDEBUG 1757 if (so->so_options & SO_DEBUG) 1758 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1759#endif 1760 m_freem(m); 1761 tp->t_flags |= TF_ACKNOW; 1762 (void) tcp_output(tp); 1763 return; 1764 1765dropwithreset: 1766 /* 1767 * Generate a RST, dropping incoming segment. 1768 * Make ACK acceptable to originator of segment. 1769 * Don't bother to respond if destination was broadcast/multicast. 1770 */ 1771 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1772 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1773 goto drop; 1774#ifdef TCPDEBUG 1775 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1776 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1777#endif 1778 if (tiflags & TH_ACK) 1779 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1780 else { 1781 if (tiflags & TH_SYN) 1782 ti->ti_len++; 1783 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1784 TH_RST|TH_ACK); 1785 } 1786 /* destroy temporarily created socket */ 1787 if (dropsocket) 1788 (void) soabort(so); 1789 return; 1790 1791drop: 1792 /* 1793 * Drop space held by incoming segment and return. 1794 */ 1795#ifdef TCPDEBUG 1796 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1797 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1798#endif 1799 m_freem(m); 1800 /* destroy temporarily created socket */ 1801 if (dropsocket) 1802 (void) soabort(so); 1803 return; 1804} 1805 1806static void 1807tcp_dooptions(tp, cp, cnt, ti, to) 1808 struct tcpcb *tp; 1809 u_char *cp; 1810 int cnt; 1811 struct tcpiphdr *ti; 1812 struct tcpopt *to; 1813{ 1814 u_short mss = 0; 1815 int opt, optlen; 1816 1817 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1818 opt = cp[0]; 1819 if (opt == TCPOPT_EOL) 1820 break; 1821 if (opt == TCPOPT_NOP) 1822 optlen = 1; 1823 else { 1824 optlen = cp[1]; 1825 if (optlen <= 0) 1826 break; 1827 } 1828 switch (opt) { 1829 1830 default: 1831 continue; 1832 1833 case TCPOPT_MAXSEG: 1834 if (optlen != TCPOLEN_MAXSEG) 1835 continue; 1836 if (!(ti->ti_flags & TH_SYN)) 1837 continue; 1838 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1839 NTOHS(mss); 1840 break; 1841 1842 case TCPOPT_WINDOW: 1843 if (optlen != TCPOLEN_WINDOW) 1844 continue; 1845 if (!(ti->ti_flags & TH_SYN)) 1846 continue; 1847 tp->t_flags |= TF_RCVD_SCALE; 1848 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1849 break; 1850 1851 case TCPOPT_TIMESTAMP: 1852 if (optlen != TCPOLEN_TIMESTAMP) 1853 continue; 1854 to->to_flag |= TOF_TS; 1855 bcopy((char *)cp + 2, 1856 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1857 NTOHL(to->to_tsval); 1858 bcopy((char *)cp + 6, 1859 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1860 NTOHL(to->to_tsecr); 1861 1862 /* 1863 * A timestamp received in a SYN makes 1864 * it ok to send timestamp requests and replies. 1865 */ 1866 if (ti->ti_flags & TH_SYN) { 1867 tp->t_flags |= TF_RCVD_TSTMP; 1868 tp->ts_recent = to->to_tsval; 1869 tp->ts_recent_age = tcp_now; 1870 } 1871 break; 1872 case TCPOPT_CC: 1873 if (optlen != TCPOLEN_CC) 1874 continue; 1875 to->to_flag |= TOF_CC; 1876 bcopy((char *)cp + 2, 1877 (char *)&to->to_cc, sizeof(to->to_cc)); 1878 NTOHL(to->to_cc); 1879 /* 1880 * A CC or CC.new option received in a SYN makes 1881 * it ok to send CC in subsequent segments. 1882 */ 1883 if (ti->ti_flags & TH_SYN) 1884 tp->t_flags |= TF_RCVD_CC; 1885 break; 1886 case TCPOPT_CCNEW: 1887 if (optlen != TCPOLEN_CC) 1888 continue; 1889 if (!(ti->ti_flags & TH_SYN)) 1890 continue; 1891 to->to_flag |= TOF_CCNEW; 1892 bcopy((char *)cp + 2, 1893 (char *)&to->to_cc, sizeof(to->to_cc)); 1894 NTOHL(to->to_cc); 1895 /* 1896 * A CC or CC.new option received in a SYN makes 1897 * it ok to send CC in subsequent segments. 1898 */ 1899 tp->t_flags |= TF_RCVD_CC; 1900 break; 1901 case TCPOPT_CCECHO: 1902 if (optlen != TCPOLEN_CC) 1903 continue; 1904 if (!(ti->ti_flags & TH_SYN)) 1905 continue; 1906 to->to_flag |= TOF_CCECHO; 1907 bcopy((char *)cp + 2, 1908 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1909 NTOHL(to->to_ccecho); 1910 break; 1911 } 1912 } 1913 if (ti->ti_flags & TH_SYN) 1914 tcp_mss(tp, mss); /* sets t_maxseg */ 1915} 1916 1917/* 1918 * Pull out of band byte out of a segment so 1919 * it doesn't appear in the user's data queue. 1920 * It is still reflected in the segment length for 1921 * sequencing purposes. 1922 */ 1923static void 1924tcp_pulloutofband(so, ti, m) 1925 struct socket *so; 1926 struct tcpiphdr *ti; 1927 register struct mbuf *m; 1928{ 1929 int cnt = ti->ti_urp - 1; 1930 1931 while (cnt >= 0) { 1932 if (m->m_len > cnt) { 1933 char *cp = mtod(m, caddr_t) + cnt; 1934 struct tcpcb *tp = sototcpcb(so); 1935 1936 tp->t_iobc = *cp; 1937 tp->t_oobflags |= TCPOOB_HAVEDATA; 1938 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1939 m->m_len--; 1940 return; 1941 } 1942 cnt -= m->m_len; 1943 m = m->m_next; 1944 if (m == 0) 1945 break; 1946 } 1947 panic("tcp_pulloutofband"); 1948} 1949 1950/* 1951 * Collect new round-trip time estimate 1952 * and update averages and current timeout. 1953 */ 1954static void 1955tcp_xmit_timer(tp, rtt) 1956 register struct tcpcb *tp; 1957 short rtt; 1958{ 1959 register int delta; 1960 1961 tcpstat.tcps_rttupdated++; 1962 tp->t_rttupdated++; 1963 if (tp->t_srtt != 0) { 1964 /* 1965 * srtt is stored as fixed point with 5 bits after the 1966 * binary point (i.e., scaled by 8). The following magic 1967 * is equivalent to the smoothing algorithm in rfc793 with 1968 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1969 * point). Adjust rtt to origin 0. 1970 */ 1971 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 1972 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 1973 1974 if ((tp->t_srtt += delta) <= 0) 1975 tp->t_srtt = 1; 1976 1977 /* 1978 * We accumulate a smoothed rtt variance (actually, a 1979 * smoothed mean difference), then set the retransmit 1980 * timer to smoothed rtt + 4 times the smoothed variance. 1981 * rttvar is stored as fixed point with 4 bits after the 1982 * binary point (scaled by 16). The following is 1983 * equivalent to rfc793 smoothing with an alpha of .75 1984 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1985 * rfc793's wired-in beta. 1986 */ 1987 if (delta < 0) 1988 delta = -delta; 1989 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 1990 if ((tp->t_rttvar += delta) <= 0) 1991 tp->t_rttvar = 1; 1992 } else { 1993 /* 1994 * No rtt measurement yet - use the unsmoothed rtt. 1995 * Set the variance to half the rtt (so our first 1996 * retransmit happens at 3*rtt). 1997 */ 1998 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1999 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2000 } 2001 tp->t_rtt = 0; 2002 tp->t_rxtshift = 0; 2003 2004 /* 2005 * the retransmit should happen at rtt + 4 * rttvar. 2006 * Because of the way we do the smoothing, srtt and rttvar 2007 * will each average +1/2 tick of bias. When we compute 2008 * the retransmit timer, we want 1/2 tick of rounding and 2009 * 1 extra tick because of +-1/2 tick uncertainty in the 2010 * firing of the timer. The bias will give us exactly the 2011 * 1.5 tick we need. But, because the bias is 2012 * statistical, we have to test that we don't drop below 2013 * the minimum feasible timer (which is 2 ticks). 2014 */ 2015 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2016 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2017 2018 /* 2019 * We received an ack for a packet that wasn't retransmitted; 2020 * it is probably safe to discard any error indications we've 2021 * received recently. This isn't quite right, but close enough 2022 * for now (a route might have failed after we sent a segment, 2023 * and the return path might not be symmetrical). 2024 */ 2025 tp->t_softerror = 0; 2026} 2027 2028/* 2029 * Determine a reasonable value for maxseg size. 2030 * If the route is known, check route for mtu. 2031 * If none, use an mss that can be handled on the outgoing 2032 * interface without forcing IP to fragment; if bigger than 2033 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2034 * to utilize large mbufs. If no route is found, route has no mtu, 2035 * or the destination isn't local, use a default, hopefully conservative 2036 * size (usually 512 or the default IP max size, but no more than the mtu 2037 * of the interface), as we can't discover anything about intervening 2038 * gateways or networks. We also initialize the congestion/slow start 2039 * window to be a single segment if the destination isn't local. 2040 * While looking at the routing entry, we also initialize other path-dependent 2041 * parameters from pre-set or cached values in the routing entry. 2042 * 2043 * Also take into account the space needed for options that we 2044 * send regularly. Make maxseg shorter by that amount to assure 2045 * that we can send maxseg amount of data even when the options 2046 * are present. Store the upper limit of the length of options plus 2047 * data in maxopd. 2048 * 2049 * NOTE that this routine is only called when we process an incoming 2050 * segment, for outgoing segments only tcp_mssopt is called. 2051 * 2052 * In case of T/TCP, we call this routine during implicit connection 2053 * setup as well (offer = -1), to initialize maxseg from the cached 2054 * MSS of our peer. 2055 */ 2056void 2057tcp_mss(tp, offer) 2058 struct tcpcb *tp; 2059 int offer; 2060{ 2061 register struct rtentry *rt; 2062 struct ifnet *ifp; 2063 register int rtt, mss; 2064 u_long bufsize; 2065 struct inpcb *inp; 2066 struct socket *so; 2067 struct rmxp_tao *taop; 2068 int origoffer = offer; 2069 2070 inp = tp->t_inpcb; 2071 if ((rt = tcp_rtlookup(inp)) == NULL) { 2072 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2073 return; 2074 } 2075 ifp = rt->rt_ifp; 2076 so = inp->inp_socket; 2077 2078 taop = rmx_taop(rt->rt_rmx); 2079 /* 2080 * Offer == -1 means that we didn't receive SYN yet, 2081 * use cached value in that case; 2082 */ 2083 if (offer == -1) 2084 offer = taop->tao_mssopt; 2085 /* 2086 * Offer == 0 means that there was no MSS on the SYN segment, 2087 * in this case we use tcp_mssdflt. 2088 */ 2089 if (offer == 0) 2090 offer = tcp_mssdflt; 2091 else 2092 /* 2093 * Sanity check: make sure that maxopd will be large 2094 * enough to allow some data on segments even is the 2095 * all the option space is used (40bytes). Otherwise 2096 * funny things may happen in tcp_output. 2097 */ 2098 offer = max(offer, 64); 2099 taop->tao_mssopt = offer; 2100 2101 /* 2102 * While we're here, check if there's an initial rtt 2103 * or rttvar. Convert from the route-table units 2104 * to scaled multiples of the slow timeout timer. 2105 */ 2106 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2107 /* 2108 * XXX the lock bit for RTT indicates that the value 2109 * is also a minimum value; this is subject to time. 2110 */ 2111 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2112 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 2113 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 2114 tcpstat.tcps_usedrtt++; 2115 if (rt->rt_rmx.rmx_rttvar) { 2116 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2117 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 2118 tcpstat.tcps_usedrttvar++; 2119 } else { 2120 /* default variation is +- 1 rtt */ 2121 tp->t_rttvar = 2122 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2123 } 2124 TCPT_RANGESET(tp->t_rxtcur, 2125 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2126 tp->t_rttmin, TCPTV_REXMTMAX); 2127 } 2128 /* 2129 * if there's an mtu associated with the route, use it 2130 */ 2131 if (rt->rt_rmx.rmx_mtu) 2132 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2133 else 2134 { 2135 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 2136 if (!in_localaddr(inp->inp_faddr)) 2137 mss = min(mss, tcp_mssdflt); 2138 } 2139 mss = min(mss, offer); 2140 /* 2141 * maxopd stores the maximum length of data AND options 2142 * in a segment; maxseg is the amount of data in a normal 2143 * segment. We need to store this value (maxopd) apart 2144 * from maxseg, because now every segment carries options 2145 * and thus we normally have somewhat less data in segments. 2146 */ 2147 tp->t_maxopd = mss; 2148 2149 /* 2150 * In case of T/TCP, origoffer==-1 indicates, that no segments 2151 * were received yet. In this case we just guess, otherwise 2152 * we do the same as before T/TCP. 2153 */ 2154 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2155 (origoffer == -1 || 2156 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2157 mss -= TCPOLEN_TSTAMP_APPA; 2158 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2159 (origoffer == -1 || 2160 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2161 mss -= TCPOLEN_CC_APPA; 2162 2163#if (MCLBYTES & (MCLBYTES - 1)) == 0 2164 if (mss > MCLBYTES) 2165 mss &= ~(MCLBYTES-1); 2166#else 2167 if (mss > MCLBYTES) 2168 mss = mss / MCLBYTES * MCLBYTES; 2169#endif 2170 /* 2171 * If there's a pipesize, change the socket buffer 2172 * to that size. Make the socket buffers an integral 2173 * number of mss units; if the mss is larger than 2174 * the socket buffer, decrease the mss. 2175 */ 2176#ifdef RTV_SPIPE 2177 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2178#endif 2179 bufsize = so->so_snd.sb_hiwat; 2180 if (bufsize < mss) 2181 mss = bufsize; 2182 else { 2183 bufsize = roundup(bufsize, mss); 2184 if (bufsize > sb_max) 2185 bufsize = sb_max; 2186 (void)sbreserve(&so->so_snd, bufsize); 2187 } 2188 tp->t_maxseg = mss; 2189 2190#ifdef RTV_RPIPE 2191 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2192#endif 2193 bufsize = so->so_rcv.sb_hiwat; 2194 if (bufsize > mss) { 2195 bufsize = roundup(bufsize, mss); 2196 if (bufsize > sb_max) 2197 bufsize = sb_max; 2198 (void)sbreserve(&so->so_rcv, bufsize); 2199 } 2200 /* 2201 * Don't force slow-start on local network. 2202 */ 2203 if (!in_localaddr(inp->inp_faddr)) 2204 tp->snd_cwnd = mss; 2205 2206 if (rt->rt_rmx.rmx_ssthresh) { 2207 /* 2208 * There's some sort of gateway or interface 2209 * buffer limit on the path. Use this to set 2210 * the slow start threshhold, but set the 2211 * threshold to no less than 2*mss. 2212 */ 2213 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2214 tcpstat.tcps_usedssthresh++; 2215 } 2216} 2217 2218/* 2219 * Determine the MSS option to send on an outgoing SYN. 2220 */ 2221int 2222tcp_mssopt(tp) 2223 struct tcpcb *tp; 2224{ 2225 struct rtentry *rt; 2226 2227 rt = tcp_rtlookup(tp->t_inpcb); 2228 if (rt == NULL) 2229 return tcp_mssdflt; 2230 2231 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2232} 2233