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 1.85 1999/05/03 23:57:30 billf 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> 72 static struct tcpiphdr tcp_saveti; 73 #endif 74 75 static int tcprexmtthresh = 3; 76 tcp_seq tcp_iss; 77 tcp_cc tcp_ccgen; 78 79 struct tcpstat tcpstat; 80 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RD, 81 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 82 83 static int log_in_vain = 0; 84 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 85 &log_in_vain, 0, "Log all incoming TCP connections"); 86 87 int tcp_delack_enabled = 1; 88 SYSCTL_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 92 u_long tcp_now; 93 struct inpcbhead tcb; 94 struct inpcbinfo tcbinfo; 95 96 static void tcp_dooptions __P((struct tcpcb *, 97 u_char *, int, struct tcpiphdr *, struct tcpopt *)); 98 static void tcp_pulloutofband __P((struct socket *, 99 struct tcpiphdr *, struct mbuf *)); 100 static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 101 static 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 } 133 134 static int 135 tcp_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 } 225 226 present: 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 */ 258 void 259 tcp_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 */ 356 findpcb: 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 } 853 } else 854 tp->t_flags &= ~TF_RCVD_CC; 855 tcpstat.tcps_connects++; 856 soisconnected(so); 857 /* Do window scaling on this connection? */ 858 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 859 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 860 tp->snd_scale = tp->requested_s_scale; 861 tp->rcv_scale = tp->request_r_scale; 862 } 863 /* Segment is acceptable, update cache if undefined. */ 864 if (taop->tao_ccsent == 0) 865 taop->tao_ccsent = to.to_ccecho; 866 867 tp->rcv_adv += tp->rcv_wnd; 868 tp->snd_una++; /* SYN is acked */ 869 /* 870 * If there's data, delay ACK; if there's also a FIN 871 * ACKNOW will be turned on later. 872 */ 873 if (tcp_delack_enabled && ti->ti_len != 0) 874 tp->t_flags |= TF_DELACK; 875 else 876 tp->t_flags |= TF_ACKNOW; 877 /* 878 * Received <SYN,ACK> in SYN_SENT[*] state. 879 * Transitions: 880 * SYN_SENT --> ESTABLISHED 881 * SYN_SENT* --> FIN_WAIT_1 882 */ 883 if (tp->t_flags & TF_NEEDFIN) { 884 tp->t_state = TCPS_FIN_WAIT_1; 885 tp->t_flags &= ~TF_NEEDFIN; 886 tiflags &= ~TH_SYN; 887 } else { 888 tp->t_state = TCPS_ESTABLISHED; 889 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 890 } 891 } else { 892 /* 893 * Received initial SYN in SYN-SENT[*] state => simul- 894 * taneous open. If segment contains CC option and there is 895 * a cached CC, apply TAO test; if it succeeds, connection is 896 * half-synchronized. Otherwise, do 3-way handshake: 897 * SYN-SENT -> SYN-RECEIVED 898 * SYN-SENT* -> SYN-RECEIVED* 899 * If there was no CC option, clear cached CC value. 900 */ 901 tp->t_flags |= TF_ACKNOW; 902 tp->t_timer[TCPT_REXMT] = 0; 903 if (to.to_flag & TOF_CC) { 904 if (taop->tao_cc != 0 && 905 CC_GT(to.to_cc, taop->tao_cc)) { 906 /* 907 * update cache and make transition: 908 * SYN-SENT -> ESTABLISHED* 909 * SYN-SENT* -> FIN-WAIT-1* 910 */ 911 taop->tao_cc = to.to_cc; 912 if (tp->t_flags & TF_NEEDFIN) { 913 tp->t_state = TCPS_FIN_WAIT_1; 914 tp->t_flags &= ~TF_NEEDFIN; 915 } else { 916 tp->t_state = TCPS_ESTABLISHED; 917 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 918 } 919 tp->t_flags |= TF_NEEDSYN; 920 } else 921 tp->t_state = TCPS_SYN_RECEIVED; 922 } else { 923 /* CC.NEW or no option => invalidate cache */ 924 taop->tao_cc = 0; 925 tp->t_state = TCPS_SYN_RECEIVED; 926 } 927 } 928 929 trimthenstep6: 930 /* 931 * Advance ti->ti_seq to correspond to first data byte. 932 * If data, trim to stay within window, 933 * dropping FIN if necessary. 934 */ 935 ti->ti_seq++; 936 if (ti->ti_len > tp->rcv_wnd) { 937 todrop = ti->ti_len - tp->rcv_wnd; 938 m_adj(m, -todrop); 939 ti->ti_len = tp->rcv_wnd; 940 tiflags &= ~TH_FIN; 941 tcpstat.tcps_rcvpackafterwin++; 942 tcpstat.tcps_rcvbyteafterwin += todrop; 943 } 944 tp->snd_wl1 = ti->ti_seq - 1; 945 tp->rcv_up = ti->ti_seq; 946 /* 947 * Client side of transaction: already sent SYN and data. 948 * If the remote host used T/TCP to validate the SYN, 949 * our data will be ACK'd; if so, enter normal data segment 950 * processing in the middle of step 5, ack processing. 951 * Otherwise, goto step 6. 952 */ 953 if (tiflags & TH_ACK) 954 goto process_ACK; 955 goto step6; 956 /* 957 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 958 * if segment contains a SYN and CC [not CC.NEW] option: 959 * if state == TIME_WAIT and connection duration > MSL, 960 * drop packet and send RST; 961 * 962 * if SEG.CC > CCrecv then is new SYN, and can implicitly 963 * ack the FIN (and data) in retransmission queue. 964 * Complete close and delete TCPCB. Then reprocess 965 * segment, hoping to find new TCPCB in LISTEN state; 966 * 967 * else must be old SYN; drop it. 968 * else do normal processing. 969 */ 970 case TCPS_LAST_ACK: 971 case TCPS_CLOSING: 972 case TCPS_TIME_WAIT: 973 if ((tiflags & TH_SYN) && 974 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 975 if (tp->t_state == TCPS_TIME_WAIT && 976 tp->t_duration > TCPTV_MSL) 977 goto dropwithreset; 978 if (CC_GT(to.to_cc, tp->cc_recv)) { 979 tp = tcp_close(tp); 980 goto findpcb; 981 } 982 else 983 goto drop; 984 } 985 break; /* continue normal processing */ 986 } 987 988 /* 989 * States other than LISTEN or SYN_SENT. 990 * First check the RST flag and sequence number since reset segments 991 * are exempt from the timestamp and connection count tests. This 992 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 993 * below which allowed reset segments in half the sequence space 994 * to fall though and be processed (which gives forged reset 995 * segments with a random sequence number a 50 percent chance of 996 * killing a connection). 997 * Then check timestamp, if present. 998 * Then check the connection count, if present. 999 * Then check that at least some bytes of segment are within 1000 * receive window. If segment begins before rcv_nxt, 1001 * drop leading data (and SYN); if nothing left, just ack. 1002 * 1003 * 1004 * If the RST bit is set, check the sequence number to see 1005 * if this is a valid reset segment. 1006 * RFC 793 page 37: 1007 * In all states except SYN-SENT, all reset (RST) segments 1008 * are validated by checking their SEQ-fields. A reset is 1009 * valid if its sequence number is in the window. 1010 * Note: this does not take into account delayed ACKs, so 1011 * we should test against last_ack_sent instead of rcv_nxt. 1012 * Also, it does not make sense to allow reset segments with 1013 * sequence numbers greater than last_ack_sent to be processed 1014 * since these sequence numbers are just the acknowledgement 1015 * numbers in our outgoing packets being echoed back at us, 1016 * and these acknowledgement numbers are monotonically 1017 * increasing. 1018 * If we have multiple segments in flight, the intial reset 1019 * segment sequence numbers will be to the left of last_ack_sent, 1020 * but they will eventually catch up. 1021 * In any case, it never made sense to trim reset segments to 1022 * fit the receive window since RFC 1122 says: 1023 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1024 * 1025 * A TCP SHOULD allow a received RST segment to include data. 1026 * 1027 * DISCUSSION 1028 * It has been suggested that a RST segment could contain 1029 * ASCII text that encoded and explained the cause of the 1030 * RST. No standard has yet been established for such 1031 * data. 1032 * 1033 * If the reset segment passes the sequence number test examine 1034 * the state: 1035 * SYN_RECEIVED STATE: 1036 * If passive open, return to LISTEN state. 1037 * If active open, inform user that connection was refused. 1038 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1039 * Inform user that connection was reset, and close tcb. 1040 * CLOSING, LAST_ACK, TIME_WAIT STATES 1041 * Close the tcb. 1042 * TIME_WAIT state: 1043 * Drop the segment - see Stevens, vol. 2, p. 964 and 1044 * RFC 1337. 1045 */ 1046 if (tiflags & TH_RST) { 1047 if (tp->last_ack_sent == ti->ti_seq) { 1048 switch (tp->t_state) { 1049 1050 case TCPS_SYN_RECEIVED: 1051 so->so_error = ECONNREFUSED; 1052 goto close; 1053 1054 case TCPS_ESTABLISHED: 1055 case TCPS_FIN_WAIT_1: 1056 case TCPS_FIN_WAIT_2: 1057 case TCPS_CLOSE_WAIT: 1058 so->so_error = ECONNRESET; 1059 close: 1060 tp->t_state = TCPS_CLOSED; 1061 tcpstat.tcps_drops++; 1062 tp = tcp_close(tp); 1063 break; 1064 1065 case TCPS_CLOSING: 1066 case TCPS_LAST_ACK: 1067 tp = tcp_close(tp); 1068 break; 1069 1070 case TCPS_TIME_WAIT: 1071 break; 1072 } 1073 } 1074 goto drop; 1075 } 1076 1077 /* 1078 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1079 * and it's less than ts_recent, drop it. 1080 */ 1081 if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent && 1082 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1083 1084 /* Check to see if ts_recent is over 24 days old. */ 1085 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1086 /* 1087 * Invalidate ts_recent. If this segment updates 1088 * ts_recent, the age will be reset later and ts_recent 1089 * will get a valid value. If it does not, setting 1090 * ts_recent to zero will at least satisfy the 1091 * requirement that zero be placed in the timestamp 1092 * echo reply when ts_recent isn't valid. The 1093 * age isn't reset until we get a valid ts_recent 1094 * because we don't want out-of-order segments to be 1095 * dropped when ts_recent is old. 1096 */ 1097 tp->ts_recent = 0; 1098 } else { 1099 tcpstat.tcps_rcvduppack++; 1100 tcpstat.tcps_rcvdupbyte += ti->ti_len; 1101 tcpstat.tcps_pawsdrop++; 1102 goto dropafterack; 1103 } 1104 } 1105 1106 /* 1107 * T/TCP mechanism 1108 * If T/TCP was negotiated and the segment doesn't have CC, 1109 * or if its CC is wrong then drop the segment. 1110 * RST segments do not have to comply with this. 1111 */ 1112 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1113 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1114 goto dropafterack; 1115 1116 /* 1117 * In the SYN-RECEIVED state, validate that the packet belongs to 1118 * this connection before trimming the data to fit the receive 1119 * window. Check the sequence number versus IRS since we know 1120 * the sequence numbers haven't wrapped. This is a partial fix 1121 * for the "LAND" DoS attack. 1122 */ 1123 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(ti->ti_seq, tp->irs)) 1124 goto dropwithreset; 1125 1126 todrop = tp->rcv_nxt - ti->ti_seq; 1127 if (todrop > 0) { 1128 if (tiflags & TH_SYN) { 1129 tiflags &= ~TH_SYN; 1130 ti->ti_seq++; 1131 if (ti->ti_urp > 1) 1132 ti->ti_urp--; 1133 else 1134 tiflags &= ~TH_URG; 1135 todrop--; 1136 } 1137 /* 1138 * Following if statement from Stevens, vol. 2, p. 960. 1139 */ 1140 if (todrop > ti->ti_len 1141 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 1142 /* 1143 * Any valid FIN must be to the left of the window. 1144 * At this point the FIN must be a duplicate or out 1145 * of sequence; drop it. 1146 */ 1147 tiflags &= ~TH_FIN; 1148 1149 /* 1150 * Send an ACK to resynchronize and drop any data. 1151 * But keep on processing for RST or ACK. 1152 */ 1153 tp->t_flags |= TF_ACKNOW; 1154 todrop = ti->ti_len; 1155 tcpstat.tcps_rcvduppack++; 1156 tcpstat.tcps_rcvdupbyte += todrop; 1157 } else { 1158 tcpstat.tcps_rcvpartduppack++; 1159 tcpstat.tcps_rcvpartdupbyte += todrop; 1160 } 1161 m_adj(m, todrop); 1162 ti->ti_seq += todrop; 1163 ti->ti_len -= todrop; 1164 if (ti->ti_urp > todrop) 1165 ti->ti_urp -= todrop; 1166 else { 1167 tiflags &= ~TH_URG; 1168 ti->ti_urp = 0; 1169 } 1170 } 1171 1172 /* 1173 * If new data are received on a connection after the 1174 * user processes are gone, then RST the other end. 1175 */ 1176 if ((so->so_state & SS_NOFDREF) && 1177 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1178 tp = tcp_close(tp); 1179 tcpstat.tcps_rcvafterclose++; 1180 goto dropwithreset; 1181 } 1182 1183 /* 1184 * If segment ends after window, drop trailing data 1185 * (and PUSH and FIN); if nothing left, just ACK. 1186 */ 1187 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1188 if (todrop > 0) { 1189 tcpstat.tcps_rcvpackafterwin++; 1190 if (todrop >= ti->ti_len) { 1191 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1192 /* 1193 * If a new connection request is received 1194 * while in TIME_WAIT, drop the old connection 1195 * and start over if the sequence numbers 1196 * are above the previous ones. 1197 */ 1198 if (tiflags & TH_SYN && 1199 tp->t_state == TCPS_TIME_WAIT && 1200 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1201 iss = tp->snd_nxt + TCP_ISSINCR; 1202 tp = tcp_close(tp); 1203 goto findpcb; 1204 } 1205 /* 1206 * If window is closed can only take segments at 1207 * window edge, and have to drop data and PUSH from 1208 * incoming segments. Continue processing, but 1209 * remember to ack. Otherwise, drop segment 1210 * and ack. 1211 */ 1212 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1213 tp->t_flags |= TF_ACKNOW; 1214 tcpstat.tcps_rcvwinprobe++; 1215 } else 1216 goto dropafterack; 1217 } else 1218 tcpstat.tcps_rcvbyteafterwin += todrop; 1219 m_adj(m, -todrop); 1220 ti->ti_len -= todrop; 1221 tiflags &= ~(TH_PUSH|TH_FIN); 1222 } 1223 1224 /* 1225 * If last ACK falls within this segment's sequence numbers, 1226 * record its timestamp. 1227 * NOTE that the test is modified according to the latest 1228 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1229 */ 1230 if ((to.to_flag & TOF_TS) != 0 && 1231 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1232 tp->ts_recent_age = tcp_now; 1233 tp->ts_recent = to.to_tsval; 1234 } 1235 1236 /* 1237 * If a SYN is in the window, then this is an 1238 * error and we send an RST and drop the connection. 1239 */ 1240 if (tiflags & TH_SYN) { 1241 tp = tcp_drop(tp, ECONNRESET); 1242 goto dropwithreset; 1243 } 1244 1245 /* 1246 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1247 * flag is on (half-synchronized state), then queue data for 1248 * later processing; else drop segment and return. 1249 */ 1250 if ((tiflags & TH_ACK) == 0) { 1251 if (tp->t_state == TCPS_SYN_RECEIVED || 1252 (tp->t_flags & TF_NEEDSYN)) 1253 goto step6; 1254 else 1255 goto drop; 1256 } 1257 1258 /* 1259 * Ack processing. 1260 */ 1261 switch (tp->t_state) { 1262 1263 /* 1264 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1265 * ESTABLISHED state and continue processing. 1266 * The ACK was checked above. 1267 */ 1268 case TCPS_SYN_RECEIVED: 1269 1270 tcpstat.tcps_connects++; 1271 soisconnected(so); 1272 /* Do window scaling? */ 1273 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1274 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1275 tp->snd_scale = tp->requested_s_scale; 1276 tp->rcv_scale = tp->request_r_scale; 1277 } 1278 /* 1279 * Upon successful completion of 3-way handshake, 1280 * update cache.CC if it was undefined, pass any queued 1281 * data to the user, and advance state appropriately. 1282 */ 1283 if ((taop = tcp_gettaocache(inp)) != NULL && 1284 taop->tao_cc == 0) 1285 taop->tao_cc = tp->cc_recv; 1286 1287 /* 1288 * Make transitions: 1289 * SYN-RECEIVED -> ESTABLISHED 1290 * SYN-RECEIVED* -> FIN-WAIT-1 1291 */ 1292 if (tp->t_flags & TF_NEEDFIN) { 1293 tp->t_state = TCPS_FIN_WAIT_1; 1294 tp->t_flags &= ~TF_NEEDFIN; 1295 } else { 1296 tp->t_state = TCPS_ESTABLISHED; 1297 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 1298 } 1299 /* 1300 * If segment contains data or ACK, will call tcp_reass() 1301 * later; if not, do so now to pass queued data to user. 1302 */ 1303 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1304 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1305 (struct mbuf *)0); 1306 tp->snd_wl1 = ti->ti_seq - 1; 1307 /* fall into ... */ 1308 1309 /* 1310 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1311 * ACKs. If the ack is in the range 1312 * tp->snd_una < ti->ti_ack <= tp->snd_max 1313 * then advance tp->snd_una to ti->ti_ack and drop 1314 * data from the retransmission queue. If this ACK reflects 1315 * more up to date window information we update our window information. 1316 */ 1317 case TCPS_ESTABLISHED: 1318 case TCPS_FIN_WAIT_1: 1319 case TCPS_FIN_WAIT_2: 1320 case TCPS_CLOSE_WAIT: 1321 case TCPS_CLOSING: 1322 case TCPS_LAST_ACK: 1323 case TCPS_TIME_WAIT: 1324 1325 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1326 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1327 tcpstat.tcps_rcvdupack++; 1328 /* 1329 * If we have outstanding data (other than 1330 * a window probe), this is a completely 1331 * duplicate ack (ie, window info didn't 1332 * change), the ack is the biggest we've 1333 * seen and we've seen exactly our rexmt 1334 * threshhold of them, assume a packet 1335 * has been dropped and retransmit it. 1336 * Kludge snd_nxt & the congestion 1337 * window so we send only this one 1338 * packet. 1339 * 1340 * We know we're losing at the current 1341 * window size so do congestion avoidance 1342 * (set ssthresh to half the current window 1343 * and pull our congestion window back to 1344 * the new ssthresh). 1345 * 1346 * Dup acks mean that packets have left the 1347 * network (they're now cached at the receiver) 1348 * so bump cwnd by the amount in the receiver 1349 * to keep a constant cwnd packets in the 1350 * network. 1351 */ 1352 if (tp->t_timer[TCPT_REXMT] == 0 || 1353 ti->ti_ack != tp->snd_una) 1354 tp->t_dupacks = 0; 1355 else if (++tp->t_dupacks == tcprexmtthresh) { 1356 tcp_seq onxt = tp->snd_nxt; 1357 u_int win = 1358 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1359 tp->t_maxseg; 1360 1361 if (win < 2) 1362 win = 2; 1363 tp->snd_ssthresh = win * tp->t_maxseg; 1364 tp->t_timer[TCPT_REXMT] = 0; 1365 tp->t_rtt = 0; 1366 tp->snd_nxt = ti->ti_ack; 1367 tp->snd_cwnd = tp->t_maxseg; 1368 (void) tcp_output(tp); 1369 tp->snd_cwnd = tp->snd_ssthresh + 1370 tp->t_maxseg * tp->t_dupacks; 1371 if (SEQ_GT(onxt, tp->snd_nxt)) 1372 tp->snd_nxt = onxt; 1373 goto drop; 1374 } else if (tp->t_dupacks > tcprexmtthresh) { 1375 tp->snd_cwnd += tp->t_maxseg; 1376 (void) tcp_output(tp); 1377 goto drop; 1378 } 1379 } else 1380 tp->t_dupacks = 0; 1381 break; 1382 } 1383 /* 1384 * If the congestion window was inflated to account 1385 * for the other side's cached packets, retract it. 1386 */ 1387 if (tp->t_dupacks >= tcprexmtthresh && 1388 tp->snd_cwnd > tp->snd_ssthresh) 1389 tp->snd_cwnd = tp->snd_ssthresh; 1390 tp->t_dupacks = 0; 1391 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1392 tcpstat.tcps_rcvacktoomuch++; 1393 goto dropafterack; 1394 } 1395 /* 1396 * If we reach this point, ACK is not a duplicate, 1397 * i.e., it ACKs something we sent. 1398 */ 1399 if (tp->t_flags & TF_NEEDSYN) { 1400 /* 1401 * T/TCP: Connection was half-synchronized, and our 1402 * SYN has been ACK'd (so connection is now fully 1403 * synchronized). Go to non-starred state, 1404 * increment snd_una for ACK of SYN, and check if 1405 * we can do window scaling. 1406 */ 1407 tp->t_flags &= ~TF_NEEDSYN; 1408 tp->snd_una++; 1409 /* Do window scaling? */ 1410 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1411 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1412 tp->snd_scale = tp->requested_s_scale; 1413 tp->rcv_scale = tp->request_r_scale; 1414 } 1415 } 1416 1417 process_ACK: 1418 acked = ti->ti_ack - tp->snd_una; 1419 tcpstat.tcps_rcvackpack++; 1420 tcpstat.tcps_rcvackbyte += acked; 1421 1422 /* 1423 * If we have a timestamp reply, update smoothed 1424 * round trip time. If no timestamp is present but 1425 * transmit timer is running and timed sequence 1426 * number was acked, update smoothed round trip time. 1427 * Since we now have an rtt measurement, cancel the 1428 * timer backoff (cf., Phil Karn's retransmit alg.). 1429 * Recompute the initial retransmit timer. 1430 */ 1431 if (to.to_flag & TOF_TS) 1432 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1433 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1434 tcp_xmit_timer(tp,tp->t_rtt); 1435 1436 /* 1437 * If all outstanding data is acked, stop retransmit 1438 * timer and remember to restart (more output or persist). 1439 * If there is more data to be acked, restart retransmit 1440 * timer, using current (possibly backed-off) value. 1441 */ 1442 if (ti->ti_ack == tp->snd_max) { 1443 tp->t_timer[TCPT_REXMT] = 0; 1444 needoutput = 1; 1445 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1446 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1447 1448 /* 1449 * If no data (only SYN) was ACK'd, 1450 * skip rest of ACK processing. 1451 */ 1452 if (acked == 0) 1453 goto step6; 1454 1455 /* 1456 * When new data is acked, open the congestion window. 1457 * If the window gives us less than ssthresh packets 1458 * in flight, open exponentially (maxseg per packet). 1459 * Otherwise open linearly: maxseg per window 1460 * (maxseg^2 / cwnd per packet). 1461 */ 1462 { 1463 register u_int cw = tp->snd_cwnd; 1464 register u_int incr = tp->t_maxseg; 1465 1466 if (cw > tp->snd_ssthresh) 1467 incr = incr * incr / cw; 1468 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1469 } 1470 if (acked > so->so_snd.sb_cc) { 1471 tp->snd_wnd -= so->so_snd.sb_cc; 1472 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1473 ourfinisacked = 1; 1474 } else { 1475 sbdrop(&so->so_snd, acked); 1476 tp->snd_wnd -= acked; 1477 ourfinisacked = 0; 1478 } 1479 sowwakeup(so); 1480 tp->snd_una = ti->ti_ack; 1481 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1482 tp->snd_nxt = tp->snd_una; 1483 1484 switch (tp->t_state) { 1485 1486 /* 1487 * In FIN_WAIT_1 STATE in addition to the processing 1488 * for the ESTABLISHED state if our FIN is now acknowledged 1489 * then enter FIN_WAIT_2. 1490 */ 1491 case TCPS_FIN_WAIT_1: 1492 if (ourfinisacked) { 1493 /* 1494 * If we can't receive any more 1495 * data, then closing user can proceed. 1496 * Starting the timer is contrary to the 1497 * specification, but if we don't get a FIN 1498 * we'll hang forever. 1499 */ 1500 if (so->so_state & SS_CANTRCVMORE) { 1501 soisdisconnected(so); 1502 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1503 } 1504 tp->t_state = TCPS_FIN_WAIT_2; 1505 } 1506 break; 1507 1508 /* 1509 * In CLOSING STATE in addition to the processing for 1510 * the ESTABLISHED state if the ACK acknowledges our FIN 1511 * then enter the TIME-WAIT state, otherwise ignore 1512 * the segment. 1513 */ 1514 case TCPS_CLOSING: 1515 if (ourfinisacked) { 1516 tp->t_state = TCPS_TIME_WAIT; 1517 tcp_canceltimers(tp); 1518 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1519 if (tp->cc_recv != 0 && 1520 tp->t_duration < TCPTV_MSL) 1521 tp->t_timer[TCPT_2MSL] = 1522 tp->t_rxtcur * TCPTV_TWTRUNC; 1523 else 1524 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1525 soisdisconnected(so); 1526 } 1527 break; 1528 1529 /* 1530 * In LAST_ACK, we may still be waiting for data to drain 1531 * and/or to be acked, as well as for the ack of our FIN. 1532 * If our FIN is now acknowledged, delete the TCB, 1533 * enter the closed state and return. 1534 */ 1535 case TCPS_LAST_ACK: 1536 if (ourfinisacked) { 1537 tp = tcp_close(tp); 1538 goto drop; 1539 } 1540 break; 1541 1542 /* 1543 * In TIME_WAIT state the only thing that should arrive 1544 * is a retransmission of the remote FIN. Acknowledge 1545 * it and restart the finack timer. 1546 */ 1547 case TCPS_TIME_WAIT: 1548 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1549 goto dropafterack; 1550 } 1551 } 1552 1553 step6: 1554 /* 1555 * Update window information. 1556 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1557 */ 1558 if ((tiflags & TH_ACK) && 1559 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1560 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1561 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1562 /* keep track of pure window updates */ 1563 if (ti->ti_len == 0 && 1564 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1565 tcpstat.tcps_rcvwinupd++; 1566 tp->snd_wnd = tiwin; 1567 tp->snd_wl1 = ti->ti_seq; 1568 tp->snd_wl2 = ti->ti_ack; 1569 if (tp->snd_wnd > tp->max_sndwnd) 1570 tp->max_sndwnd = tp->snd_wnd; 1571 needoutput = 1; 1572 } 1573 1574 /* 1575 * Process segments with URG. 1576 */ 1577 if ((tiflags & TH_URG) && ti->ti_urp && 1578 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1579 /* 1580 * This is a kludge, but if we receive and accept 1581 * random urgent pointers, we'll crash in 1582 * soreceive. It's hard to imagine someone 1583 * actually wanting to send this much urgent data. 1584 */ 1585 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1586 ti->ti_urp = 0; /* XXX */ 1587 tiflags &= ~TH_URG; /* XXX */ 1588 goto dodata; /* XXX */ 1589 } 1590 /* 1591 * If this segment advances the known urgent pointer, 1592 * then mark the data stream. This should not happen 1593 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1594 * a FIN has been received from the remote side. 1595 * In these states we ignore the URG. 1596 * 1597 * According to RFC961 (Assigned Protocols), 1598 * the urgent pointer points to the last octet 1599 * of urgent data. We continue, however, 1600 * to consider it to indicate the first octet 1601 * of data past the urgent section as the original 1602 * spec states (in one of two places). 1603 */ 1604 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1605 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1606 so->so_oobmark = so->so_rcv.sb_cc + 1607 (tp->rcv_up - tp->rcv_nxt) - 1; 1608 if (so->so_oobmark == 0) 1609 so->so_state |= SS_RCVATMARK; 1610 sohasoutofband(so); 1611 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1612 } 1613 /* 1614 * Remove out of band data so doesn't get presented to user. 1615 * This can happen independent of advancing the URG pointer, 1616 * but if two URG's are pending at once, some out-of-band 1617 * data may creep in... ick. 1618 */ 1619 if (ti->ti_urp <= (u_long)ti->ti_len 1620 #ifdef SO_OOBINLINE 1621 && (so->so_options & SO_OOBINLINE) == 0 1622 #endif 1623 ) 1624 tcp_pulloutofband(so, ti, m); 1625 } else 1626 /* 1627 * If no out of band data is expected, 1628 * pull receive urgent pointer along 1629 * with the receive window. 1630 */ 1631 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1632 tp->rcv_up = tp->rcv_nxt; 1633 dodata: /* XXX */ 1634 1635 /* 1636 * Process the segment text, merging it into the TCP sequencing queue, 1637 * and arranging for acknowledgment of receipt if necessary. 1638 * This process logically involves adjusting tp->rcv_wnd as data 1639 * is presented to the user (this happens in tcp_usrreq.c, 1640 * case PRU_RCVD). If a FIN has already been received on this 1641 * connection then we just ignore the text. 1642 */ 1643 if ((ti->ti_len || (tiflags&TH_FIN)) && 1644 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1645 TCP_REASS(tp, ti, m, so, tiflags); 1646 /* 1647 * Note the amount of data that peer has sent into 1648 * our window, in order to estimate the sender's 1649 * buffer size. 1650 */ 1651 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1652 } else { 1653 m_freem(m); 1654 tiflags &= ~TH_FIN; 1655 } 1656 1657 /* 1658 * If FIN is received ACK the FIN and let the user know 1659 * that the connection is closing. 1660 */ 1661 if (tiflags & TH_FIN) { 1662 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1663 socantrcvmore(so); 1664 /* 1665 * If connection is half-synchronized 1666 * (ie NEEDSYN flag on) then delay ACK, 1667 * so it may be piggybacked when SYN is sent. 1668 * Otherwise, since we received a FIN then no 1669 * more input can be expected, send ACK now. 1670 */ 1671 if (tcp_delack_enabled && (tp->t_flags & TF_NEEDSYN)) 1672 tp->t_flags |= TF_DELACK; 1673 else 1674 tp->t_flags |= TF_ACKNOW; 1675 tp->rcv_nxt++; 1676 } 1677 switch (tp->t_state) { 1678 1679 /* 1680 * In SYN_RECEIVED and ESTABLISHED STATES 1681 * enter the CLOSE_WAIT state. 1682 */ 1683 case TCPS_SYN_RECEIVED: 1684 case TCPS_ESTABLISHED: 1685 tp->t_state = TCPS_CLOSE_WAIT; 1686 break; 1687 1688 /* 1689 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1690 * enter the CLOSING state. 1691 */ 1692 case TCPS_FIN_WAIT_1: 1693 tp->t_state = TCPS_CLOSING; 1694 break; 1695 1696 /* 1697 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1698 * starting the time-wait timer, turning off the other 1699 * standard timers. 1700 */ 1701 case TCPS_FIN_WAIT_2: 1702 tp->t_state = TCPS_TIME_WAIT; 1703 tcp_canceltimers(tp); 1704 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1705 if (tp->cc_recv != 0 && 1706 tp->t_duration < TCPTV_MSL) { 1707 tp->t_timer[TCPT_2MSL] = 1708 tp->t_rxtcur * TCPTV_TWTRUNC; 1709 /* For transaction client, force ACK now. */ 1710 tp->t_flags |= TF_ACKNOW; 1711 } 1712 else 1713 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1714 soisdisconnected(so); 1715 break; 1716 1717 /* 1718 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1719 */ 1720 case TCPS_TIME_WAIT: 1721 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1722 break; 1723 } 1724 } 1725 #ifdef TCPDEBUG 1726 if (so->so_options & SO_DEBUG) 1727 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1728 #endif 1729 1730 /* 1731 * Return any desired output. 1732 */ 1733 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1734 (void) tcp_output(tp); 1735 return; 1736 1737 dropafterack: 1738 /* 1739 * Generate an ACK dropping incoming segment if it occupies 1740 * sequence space, where the ACK reflects our state. 1741 * 1742 * We can now skip the test for the RST flag since all 1743 * paths to this code happen after packets containing 1744 * RST have been dropped. 1745 * 1746 * In the SYN-RECEIVED state, don't send an ACK unless the 1747 * segment we received passes the SYN-RECEIVED ACK test. 1748 * If it fails send a RST. This breaks the loop in the 1749 * "LAND" DoS attack, and also prevents an ACK storm 1750 * between two listening ports that have been sent forged 1751 * SYN segments, each with the source address of the other. 1752 */ 1753 if (tp->t_state == TCPS_SYN_RECEIVED && (tiflags & TH_ACK) && 1754 (SEQ_GT(tp->snd_una, ti->ti_ack) || 1755 SEQ_GT(ti->ti_ack, tp->snd_max)) ) 1756 goto dropwithreset; 1757 #ifdef TCPDEBUG 1758 if (so->so_options & SO_DEBUG) 1759 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1760 #endif 1761 m_freem(m); 1762 tp->t_flags |= TF_ACKNOW; 1763 (void) tcp_output(tp); 1764 return; 1765 1766 dropwithreset: 1767 /* 1768 * Generate a RST, dropping incoming segment. 1769 * Make ACK acceptable to originator of segment. 1770 * Don't bother to respond if destination was broadcast/multicast. 1771 */ 1772 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1773 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1774 goto drop; 1775 #ifdef TCPDEBUG 1776 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1777 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1778 #endif 1779 if (tiflags & TH_ACK) 1780 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1781 else { 1782 if (tiflags & TH_SYN) 1783 ti->ti_len++; 1784 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1785 TH_RST|TH_ACK); 1786 } 1787 /* destroy temporarily created socket */ 1788 if (dropsocket) 1789 (void) soabort(so); 1790 return; 1791 1792 drop: 1793 /* 1794 * Drop space held by incoming segment and return. 1795 */ 1796 #ifdef TCPDEBUG 1797 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1798 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1799 #endif 1800 m_freem(m); 1801 /* destroy temporarily created socket */ 1802 if (dropsocket) 1803 (void) soabort(so); 1804 return; 1805 } 1806 1807 static void 1808 tcp_dooptions(tp, cp, cnt, ti, to) 1809 struct tcpcb *tp; 1810 u_char *cp; 1811 int cnt; 1812 struct tcpiphdr *ti; 1813 struct tcpopt *to; 1814 { 1815 u_short mss = 0; 1816 int opt, optlen; 1817 1818 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1819 opt = cp[0]; 1820 if (opt == TCPOPT_EOL) 1821 break; 1822 if (opt == TCPOPT_NOP) 1823 optlen = 1; 1824 else { 1825 optlen = cp[1]; 1826 if (optlen <= 0) 1827 break; 1828 } 1829 switch (opt) { 1830 1831 default: 1832 continue; 1833 1834 case TCPOPT_MAXSEG: 1835 if (optlen != TCPOLEN_MAXSEG) 1836 continue; 1837 if (!(ti->ti_flags & TH_SYN)) 1838 continue; 1839 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1840 NTOHS(mss); 1841 break; 1842 1843 case TCPOPT_WINDOW: 1844 if (optlen != TCPOLEN_WINDOW) 1845 continue; 1846 if (!(ti->ti_flags & TH_SYN)) 1847 continue; 1848 tp->t_flags |= TF_RCVD_SCALE; 1849 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1850 break; 1851 1852 case TCPOPT_TIMESTAMP: 1853 if (optlen != TCPOLEN_TIMESTAMP) 1854 continue; 1855 to->to_flag |= TOF_TS; 1856 bcopy((char *)cp + 2, 1857 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1858 NTOHL(to->to_tsval); 1859 bcopy((char *)cp + 6, 1860 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1861 NTOHL(to->to_tsecr); 1862 1863 /* 1864 * A timestamp received in a SYN makes 1865 * it ok to send timestamp requests and replies. 1866 */ 1867 if (ti->ti_flags & TH_SYN) { 1868 tp->t_flags |= TF_RCVD_TSTMP; 1869 tp->ts_recent = to->to_tsval; 1870 tp->ts_recent_age = tcp_now; 1871 } 1872 break; 1873 case TCPOPT_CC: 1874 if (optlen != TCPOLEN_CC) 1875 continue; 1876 to->to_flag |= TOF_CC; 1877 bcopy((char *)cp + 2, 1878 (char *)&to->to_cc, sizeof(to->to_cc)); 1879 NTOHL(to->to_cc); 1880 /* 1881 * A CC or CC.new option received in a SYN makes 1882 * it ok to send CC in subsequent segments. 1883 */ 1884 if (ti->ti_flags & TH_SYN) 1885 tp->t_flags |= TF_RCVD_CC; 1886 break; 1887 case TCPOPT_CCNEW: 1888 if (optlen != TCPOLEN_CC) 1889 continue; 1890 if (!(ti->ti_flags & TH_SYN)) 1891 continue; 1892 to->to_flag |= TOF_CCNEW; 1893 bcopy((char *)cp + 2, 1894 (char *)&to->to_cc, sizeof(to->to_cc)); 1895 NTOHL(to->to_cc); 1896 /* 1897 * A CC or CC.new option received in a SYN makes 1898 * it ok to send CC in subsequent segments. 1899 */ 1900 tp->t_flags |= TF_RCVD_CC; 1901 break; 1902 case TCPOPT_CCECHO: 1903 if (optlen != TCPOLEN_CC) 1904 continue; 1905 if (!(ti->ti_flags & TH_SYN)) 1906 continue; 1907 to->to_flag |= TOF_CCECHO; 1908 bcopy((char *)cp + 2, 1909 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1910 NTOHL(to->to_ccecho); 1911 break; 1912 } 1913 } 1914 if (ti->ti_flags & TH_SYN) 1915 tcp_mss(tp, mss); /* sets t_maxseg */ 1916 } 1917 1918 /* 1919 * Pull out of band byte out of a segment so 1920 * it doesn't appear in the user's data queue. 1921 * It is still reflected in the segment length for 1922 * sequencing purposes. 1923 */ 1924 static void 1925 tcp_pulloutofband(so, ti, m) 1926 struct socket *so; 1927 struct tcpiphdr *ti; 1928 register struct mbuf *m; 1929 { 1930 int cnt = ti->ti_urp - 1; 1931 1932 while (cnt >= 0) { 1933 if (m->m_len > cnt) { 1934 char *cp = mtod(m, caddr_t) + cnt; 1935 struct tcpcb *tp = sototcpcb(so); 1936 1937 tp->t_iobc = *cp; 1938 tp->t_oobflags |= TCPOOB_HAVEDATA; 1939 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1940 m->m_len--; 1941 return; 1942 } 1943 cnt -= m->m_len; 1944 m = m->m_next; 1945 if (m == 0) 1946 break; 1947 } 1948 panic("tcp_pulloutofband"); 1949 } 1950 1951 /* 1952 * Collect new round-trip time estimate 1953 * and update averages and current timeout. 1954 */ 1955 static void 1956 tcp_xmit_timer(tp, rtt) 1957 register struct tcpcb *tp; 1958 short rtt; 1959 { 1960 register int delta; 1961 1962 tcpstat.tcps_rttupdated++; 1963 tp->t_rttupdated++; 1964 if (tp->t_srtt != 0) { 1965 /* 1966 * srtt is stored as fixed point with 5 bits after the 1967 * binary point (i.e., scaled by 8). The following magic 1968 * is equivalent to the smoothing algorithm in rfc793 with 1969 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1970 * point). Adjust rtt to origin 0. 1971 */ 1972 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 1973 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 1974 1975 if ((tp->t_srtt += delta) <= 0) 1976 tp->t_srtt = 1; 1977 1978 /* 1979 * We accumulate a smoothed rtt variance (actually, a 1980 * smoothed mean difference), then set the retransmit 1981 * timer to smoothed rtt + 4 times the smoothed variance. 1982 * rttvar is stored as fixed point with 4 bits after the 1983 * binary point (scaled by 16). The following is 1984 * equivalent to rfc793 smoothing with an alpha of .75 1985 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1986 * rfc793's wired-in beta. 1987 */ 1988 if (delta < 0) 1989 delta = -delta; 1990 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 1991 if ((tp->t_rttvar += delta) <= 0) 1992 tp->t_rttvar = 1; 1993 } else { 1994 /* 1995 * No rtt measurement yet - use the unsmoothed rtt. 1996 * Set the variance to half the rtt (so our first 1997 * retransmit happens at 3*rtt). 1998 */ 1999 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2000 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2001 } 2002 tp->t_rtt = 0; 2003 tp->t_rxtshift = 0; 2004 2005 /* 2006 * the retransmit should happen at rtt + 4 * rttvar. 2007 * Because of the way we do the smoothing, srtt and rttvar 2008 * will each average +1/2 tick of bias. When we compute 2009 * the retransmit timer, we want 1/2 tick of rounding and 2010 * 1 extra tick because of +-1/2 tick uncertainty in the 2011 * firing of the timer. The bias will give us exactly the 2012 * 1.5 tick we need. But, because the bias is 2013 * statistical, we have to test that we don't drop below 2014 * the minimum feasible timer (which is 2 ticks). 2015 */ 2016 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2017 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2018 2019 /* 2020 * We received an ack for a packet that wasn't retransmitted; 2021 * it is probably safe to discard any error indications we've 2022 * received recently. This isn't quite right, but close enough 2023 * for now (a route might have failed after we sent a segment, 2024 * and the return path might not be symmetrical). 2025 */ 2026 tp->t_softerror = 0; 2027 } 2028 2029 /* 2030 * Determine a reasonable value for maxseg size. 2031 * If the route is known, check route for mtu. 2032 * If none, use an mss that can be handled on the outgoing 2033 * interface without forcing IP to fragment; if bigger than 2034 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2035 * to utilize large mbufs. If no route is found, route has no mtu, 2036 * or the destination isn't local, use a default, hopefully conservative 2037 * size (usually 512 or the default IP max size, but no more than the mtu 2038 * of the interface), as we can't discover anything about intervening 2039 * gateways or networks. We also initialize the congestion/slow start 2040 * window to be a single segment if the destination isn't local. 2041 * While looking at the routing entry, we also initialize other path-dependent 2042 * parameters from pre-set or cached values in the routing entry. 2043 * 2044 * Also take into account the space needed for options that we 2045 * send regularly. Make maxseg shorter by that amount to assure 2046 * that we can send maxseg amount of data even when the options 2047 * are present. Store the upper limit of the length of options plus 2048 * data in maxopd. 2049 * 2050 * NOTE that this routine is only called when we process an incoming 2051 * segment, for outgoing segments only tcp_mssopt is called. 2052 * 2053 * In case of T/TCP, we call this routine during implicit connection 2054 * setup as well (offer = -1), to initialize maxseg from the cached 2055 * MSS of our peer. 2056 */ 2057 void 2058 tcp_mss(tp, offer) 2059 struct tcpcb *tp; 2060 int offer; 2061 { 2062 register struct rtentry *rt; 2063 struct ifnet *ifp; 2064 register int rtt, mss; 2065 u_long bufsize; 2066 struct inpcb *inp; 2067 struct socket *so; 2068 struct rmxp_tao *taop; 2069 int origoffer = offer; 2070 2071 inp = tp->t_inpcb; 2072 if ((rt = tcp_rtlookup(inp)) == NULL) { 2073 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2074 return; 2075 } 2076 ifp = rt->rt_ifp; 2077 so = inp->inp_socket; 2078 2079 taop = rmx_taop(rt->rt_rmx); 2080 /* 2081 * Offer == -1 means that we didn't receive SYN yet, 2082 * use cached value in that case; 2083 */ 2084 if (offer == -1) 2085 offer = taop->tao_mssopt; 2086 /* 2087 * Offer == 0 means that there was no MSS on the SYN segment, 2088 * in this case we use tcp_mssdflt. 2089 */ 2090 if (offer == 0) 2091 offer = tcp_mssdflt; 2092 else 2093 /* 2094 * Sanity check: make sure that maxopd will be large 2095 * enough to allow some data on segments even is the 2096 * all the option space is used (40bytes). Otherwise 2097 * funny things may happen in tcp_output. 2098 */ 2099 offer = max(offer, 64); 2100 taop->tao_mssopt = offer; 2101 2102 /* 2103 * While we're here, check if there's an initial rtt 2104 * or rttvar. Convert from the route-table units 2105 * to scaled multiples of the slow timeout timer. 2106 */ 2107 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2108 /* 2109 * XXX the lock bit for RTT indicates that the value 2110 * is also a minimum value; this is subject to time. 2111 */ 2112 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2113 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 2114 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 2115 tcpstat.tcps_usedrtt++; 2116 if (rt->rt_rmx.rmx_rttvar) { 2117 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2118 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 2119 tcpstat.tcps_usedrttvar++; 2120 } else { 2121 /* default variation is +- 1 rtt */ 2122 tp->t_rttvar = 2123 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2124 } 2125 TCPT_RANGESET(tp->t_rxtcur, 2126 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2127 tp->t_rttmin, TCPTV_REXMTMAX); 2128 } 2129 /* 2130 * if there's an mtu associated with the route, use it 2131 */ 2132 if (rt->rt_rmx.rmx_mtu) 2133 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2134 else 2135 { 2136 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 2137 if (!in_localaddr(inp->inp_faddr)) 2138 mss = min(mss, tcp_mssdflt); 2139 } 2140 mss = min(mss, offer); 2141 /* 2142 * maxopd stores the maximum length of data AND options 2143 * in a segment; maxseg is the amount of data in a normal 2144 * segment. We need to store this value (maxopd) apart 2145 * from maxseg, because now every segment carries options 2146 * and thus we normally have somewhat less data in segments. 2147 */ 2148 tp->t_maxopd = mss; 2149 2150 /* 2151 * In case of T/TCP, origoffer==-1 indicates, that no segments 2152 * were received yet. In this case we just guess, otherwise 2153 * we do the same as before T/TCP. 2154 */ 2155 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2156 (origoffer == -1 || 2157 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2158 mss -= TCPOLEN_TSTAMP_APPA; 2159 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2160 (origoffer == -1 || 2161 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2162 mss -= TCPOLEN_CC_APPA; 2163 2164 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2165 if (mss > MCLBYTES) 2166 mss &= ~(MCLBYTES-1); 2167 #else 2168 if (mss > MCLBYTES) 2169 mss = mss / MCLBYTES * MCLBYTES; 2170 #endif 2171 /* 2172 * If there's a pipesize, change the socket buffer 2173 * to that size. Make the socket buffers an integral 2174 * number of mss units; if the mss is larger than 2175 * the socket buffer, decrease the mss. 2176 */ 2177 #ifdef RTV_SPIPE 2178 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2179 #endif 2180 bufsize = so->so_snd.sb_hiwat; 2181 if (bufsize < mss) 2182 mss = bufsize; 2183 else { 2184 bufsize = roundup(bufsize, mss); 2185 if (bufsize > sb_max) 2186 bufsize = sb_max; 2187 (void)sbreserve(&so->so_snd, bufsize); 2188 } 2189 tp->t_maxseg = mss; 2190 2191 #ifdef RTV_RPIPE 2192 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2193 #endif 2194 bufsize = so->so_rcv.sb_hiwat; 2195 if (bufsize > mss) { 2196 bufsize = roundup(bufsize, mss); 2197 if (bufsize > sb_max) 2198 bufsize = sb_max; 2199 (void)sbreserve(&so->so_rcv, bufsize); 2200 } 2201 /* 2202 * Don't force slow-start on local network. 2203 */ 2204 if (!in_localaddr(inp->inp_faddr)) 2205 tp->snd_cwnd = mss; 2206 2207 if (rt->rt_rmx.rmx_ssthresh) { 2208 /* 2209 * There's some sort of gateway or interface 2210 * buffer limit on the path. Use this to set 2211 * the slow start threshhold, but set the 2212 * threshold to no less than 2*mss. 2213 */ 2214 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2215 tcpstat.tcps_usedssthresh++; 2216 } 2217 } 2218 2219 /* 2220 * Determine the MSS option to send on an outgoing SYN. 2221 */ 2222 int 2223 tcp_mssopt(tp) 2224 struct tcpcb *tp; 2225 { 2226 struct rtentry *rt; 2227 2228 rt = tcp_rtlookup(tp->t_inpcb); 2229 if (rt == NULL) 2230 return tcp_mssdflt; 2231 2232 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2233 } 2234