tcp_input.c revision 50015
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 1.88 1999/08/17 12:17:52 csgr 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 87static int blackhole = 0; 88SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 89 &blackhole, 0, "Do not send RST when dropping refused connections"); 90 91int tcp_delack_enabled = 1; 92SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 93 &tcp_delack_enabled, 0, 94 "Delay ACK to try and piggyback it onto a data packet"); 95 96u_long tcp_now; 97struct inpcbhead tcb; 98struct inpcbinfo tcbinfo; 99 100static void tcp_dooptions __P((struct tcpcb *, 101 u_char *, int, struct tcpiphdr *, struct tcpopt *)); 102static void tcp_pulloutofband __P((struct socket *, 103 struct tcpiphdr *, struct mbuf *)); 104static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *)); 105static void tcp_xmit_timer __P((struct tcpcb *, int)); 106 107 108/* 109 * Insert segment ti into reassembly queue of tcp with 110 * control block tp. Return TH_FIN if reassembly now includes 111 * a segment with FIN. The macro form does the common case inline 112 * (segment is the next to be received on an established connection, 113 * and the queue is empty), avoiding linkage into and removal 114 * from the queue and repetition of various conversions. 115 * Set DELACK for segments received in order, but ack immediately 116 * when segments are out of order (so fast retransmit can work). 117 */ 118#define TCP_REASS(tp, ti, m, so, flags) { \ 119 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 120 (tp)->t_segq == NULL && \ 121 (tp)->t_state == TCPS_ESTABLISHED) { \ 122 if (tcp_delack_enabled) \ 123 tp->t_flags |= TF_DELACK; \ 124 else \ 125 tp->t_flags |= TF_ACKNOW; \ 126 (tp)->rcv_nxt += (ti)->ti_len; \ 127 flags = (ti)->ti_flags & TH_FIN; \ 128 tcpstat.tcps_rcvpack++;\ 129 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 130 sbappend(&(so)->so_rcv, (m)); \ 131 sorwakeup(so); \ 132 } else { \ 133 (flags) = tcp_reass((tp), (ti), (m)); \ 134 tp->t_flags |= TF_ACKNOW; \ 135 } \ 136} 137 138static int 139tcp_reass(tp, ti, m) 140 register struct tcpcb *tp; 141 register struct tcpiphdr *ti; 142 struct mbuf *m; 143{ 144 struct mbuf *q; 145 struct mbuf *p; 146 struct mbuf *nq; 147 struct socket *so = tp->t_inpcb->inp_socket; 148 int flags; 149 150#define GETTCP(m) ((struct tcpiphdr *)m->m_pkthdr.header) 151 152 /* 153 * Call with ti==0 after become established to 154 * force pre-ESTABLISHED data up to user socket. 155 */ 156 if (ti == 0) 157 goto present; 158 159 m->m_pkthdr.header = ti; 160 161 /* 162 * Find a segment which begins after this one does. 163 */ 164 for (q = tp->t_segq, p = NULL; q; p = q, q = q->m_nextpkt) 165 if (SEQ_GT(GETTCP(q)->ti_seq, ti->ti_seq)) 166 break; 167 168 /* 169 * If there is a preceding segment, it may provide some of 170 * our data already. If so, drop the data from the incoming 171 * segment. If it provides all of our data, drop us. 172 */ 173 if (p != NULL) { 174 register int i; 175 /* conversion to int (in i) handles seq wraparound */ 176 i = GETTCP(p)->ti_seq + GETTCP(p)->ti_len - ti->ti_seq; 177 if (i > 0) { 178 if (i >= ti->ti_len) { 179 tcpstat.tcps_rcvduppack++; 180 tcpstat.tcps_rcvdupbyte += ti->ti_len; 181 m_freem(m); 182 /* 183 * Try to present any queued data 184 * at the left window edge to the user. 185 * This is needed after the 3-WHS 186 * completes. 187 */ 188 goto present; /* ??? */ 189 } 190 m_adj(m, i); 191 ti->ti_len -= i; 192 ti->ti_seq += i; 193 } 194 } 195 tcpstat.tcps_rcvoopack++; 196 tcpstat.tcps_rcvoobyte += ti->ti_len; 197 198 /* 199 * While we overlap succeeding segments trim them or, 200 * if they are completely covered, dequeue them. 201 */ 202 while (q) { 203 register int i = (ti->ti_seq + ti->ti_len) - GETTCP(q)->ti_seq; 204 if (i <= 0) 205 break; 206 if (i < GETTCP(q)->ti_len) { 207 GETTCP(q)->ti_seq += i; 208 GETTCP(q)->ti_len -= i; 209 m_adj(q, i); 210 break; 211 } 212 213 nq = q->m_nextpkt; 214 if (p) 215 p->m_nextpkt = nq; 216 else 217 tp->t_segq = nq; 218 m_freem(q); 219 q = nq; 220 } 221 222 if (p == NULL) { 223 m->m_nextpkt = tp->t_segq; 224 tp->t_segq = m; 225 } else { 226 m->m_nextpkt = p->m_nextpkt; 227 p->m_nextpkt = m; 228 } 229 230present: 231 /* 232 * Present data to user, advancing rcv_nxt through 233 * completed sequence space. 234 */ 235 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 236 return (0); 237 q = tp->t_segq; 238 if (!q || GETTCP(q)->ti_seq != tp->rcv_nxt) 239 return (0); 240 do { 241 tp->rcv_nxt += GETTCP(q)->ti_len; 242 flags = GETTCP(q)->ti_flags & TH_FIN; 243 nq = q->m_nextpkt; 244 tp->t_segq = nq; 245 q->m_nextpkt = NULL; 246 if (so->so_state & SS_CANTRCVMORE) 247 m_freem(q); 248 else 249 sbappend(&so->so_rcv, q); 250 q = nq; 251 } while (q && GETTCP(q)->ti_seq == tp->rcv_nxt); 252 sorwakeup(so); 253 return (flags); 254 255#undef GETTCP 256} 257 258/* 259 * TCP input routine, follows pages 65-76 of the 260 * protocol specification dated September, 1981 very closely. 261 */ 262void 263tcp_input(m, iphlen) 264 register struct mbuf *m; 265 int iphlen; 266{ 267 register struct tcpiphdr *ti; 268 register struct inpcb *inp; 269 u_char *optp = NULL; 270 int optlen = 0; 271 int len, tlen, off; 272 register struct tcpcb *tp = 0; 273 register int tiflags; 274 struct socket *so = 0; 275 int todrop, acked, ourfinisacked, needoutput = 0; 276 struct in_addr laddr; 277 int dropsocket = 0; 278 int iss = 0; 279 u_long tiwin; 280 struct tcpopt to; /* options in this segment */ 281 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 282 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 283#ifdef TCPDEBUG 284 short ostate = 0; 285#endif 286 287 bzero((char *)&to, sizeof(to)); 288 289 tcpstat.tcps_rcvtotal++; 290 /* 291 * Get IP and TCP header together in first mbuf. 292 * Note: IP leaves IP header in first mbuf. 293 */ 294 ti = mtod(m, struct tcpiphdr *); 295 if (iphlen > sizeof (struct ip)) 296 ip_stripoptions(m, (struct mbuf *)0); 297 if (m->m_len < sizeof (struct tcpiphdr)) { 298 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 299 tcpstat.tcps_rcvshort++; 300 return; 301 } 302 ti = mtod(m, struct tcpiphdr *); 303 } 304 305 /* 306 * Checksum extended TCP header and data. 307 */ 308 tlen = ((struct ip *)ti)->ip_len; 309 len = sizeof (struct ip) + tlen; 310 bzero(ti->ti_x1, sizeof(ti->ti_x1)); 311 ti->ti_len = (u_short)tlen; 312 HTONS(ti->ti_len); 313 ti->ti_sum = in_cksum(m, len); 314 if (ti->ti_sum) { 315 tcpstat.tcps_rcvbadsum++; 316 goto drop; 317 } 318 319 /* 320 * Check that TCP offset makes sense, 321 * pull out TCP options and adjust length. XXX 322 */ 323 off = ti->ti_off << 2; 324 if (off < sizeof (struct tcphdr) || off > tlen) { 325 tcpstat.tcps_rcvbadoff++; 326 goto drop; 327 } 328 tlen -= off; 329 ti->ti_len = tlen; 330 if (off > sizeof (struct tcphdr)) { 331 if (m->m_len < sizeof(struct ip) + off) { 332 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 333 tcpstat.tcps_rcvshort++; 334 return; 335 } 336 ti = mtod(m, struct tcpiphdr *); 337 } 338 optlen = off - sizeof (struct tcphdr); 339 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 340 } 341 tiflags = ti->ti_flags; 342 343 /* 344 * Convert TCP protocol specific fields to host format. 345 */ 346 NTOHL(ti->ti_seq); 347 NTOHL(ti->ti_ack); 348 NTOHS(ti->ti_win); 349 NTOHS(ti->ti_urp); 350 351 /* 352 * Drop TCP, IP headers and TCP options. 353 */ 354 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 355 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 356 357 /* 358 * Locate pcb for segment. 359 */ 360findpcb: 361#ifdef IPFIREWALL_FORWARD 362 if (ip_fw_fwd_addr != NULL) { 363 /* 364 * Diverted. Pretend to be the destination. 365 * already got one like this? 366 */ 367 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, ti->ti_sport, 368 ti->ti_dst, ti->ti_dport, 0); 369 if (!inp) { 370 /* 371 * No, then it's new. Try find the ambushing socket 372 */ 373 if (!ip_fw_fwd_addr->sin_port) { 374 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, 375 ti->ti_sport, ip_fw_fwd_addr->sin_addr, 376 ti->ti_dport, 1); 377 } else { 378 inp = in_pcblookup_hash(&tcbinfo, 379 ti->ti_src, ti->ti_sport, 380 ip_fw_fwd_addr->sin_addr, 381 ntohs(ip_fw_fwd_addr->sin_port), 1); 382 } 383 } 384 ip_fw_fwd_addr = NULL; 385 } else 386#endif /* IPFIREWALL_FORWARD */ 387 388 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, ti->ti_sport, 389 ti->ti_dst, ti->ti_dport, 1); 390 391 /* 392 * If the state is CLOSED (i.e., TCB does not exist) then 393 * all data in the incoming segment is discarded. 394 * If the TCB exists but is in CLOSED state, it is embryonic, 395 * but should either do a listen or a connect soon. 396 */ 397 if (inp == NULL) { 398 if (log_in_vain) { 399 char buf[4*sizeof "123"]; 400 401 strcpy(buf, inet_ntoa(ti->ti_dst)); 402 switch (log_in_vain) { 403 case 1: 404 if(tiflags & TH_SYN) 405 log(LOG_INFO, 406 "Connection attempt to TCP %s:%d from %s:%d\n", 407 buf, ntohs(ti->ti_dport), inet_ntoa(ti->ti_src), 408 ntohs(ti->ti_sport)); 409 break; 410 case 2: 411 log(LOG_INFO, 412 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", 413 buf, ntohs(ti->ti_dport), inet_ntoa(ti->ti_src), 414 ntohs(ti->ti_sport), tiflags); 415 break; 416 default: 417 break; 418 } 419 } 420#ifdef ICMP_BANDLIM 421 if (badport_bandlim(1) < 0) 422 goto drop; 423#endif 424 if(blackhole) { 425 switch (blackhole) { 426 case 1: if(tiflags & TH_SYN) goto drop; 427 case 2: goto drop ; 428 default : goto drop ; 429 } 430 } else 431 goto dropwithreset; 432 } 433 tp = intotcpcb(inp); 434 if (tp == 0) 435 goto dropwithreset; 436 if (tp->t_state == TCPS_CLOSED) 437 goto drop; 438 439 /* Unscale the window into a 32-bit value. */ 440 if ((tiflags & TH_SYN) == 0) 441 tiwin = ti->ti_win << tp->snd_scale; 442 else 443 tiwin = ti->ti_win; 444 445 so = inp->inp_socket; 446 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 447#ifdef TCPDEBUG 448 if (so->so_options & SO_DEBUG) { 449 ostate = tp->t_state; 450 tcp_saveti = *ti; 451 } 452#endif 453 if (so->so_options & SO_ACCEPTCONN) { 454 register struct tcpcb *tp0 = tp; 455 struct socket *so2; 456 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 457 /* 458 * Note: dropwithreset makes sure we don't 459 * send a RST in response to a RST. 460 */ 461 if (tiflags & TH_ACK) { 462 tcpstat.tcps_badsyn++; 463 goto dropwithreset; 464 } 465 goto drop; 466 } 467 so2 = sonewconn(so, 0); 468 if (so2 == 0) { 469 tcpstat.tcps_listendrop++; 470 so2 = sodropablereq(so); 471 if (so2) { 472 tcp_drop(sototcpcb(so2), ETIMEDOUT); 473 so2 = sonewconn(so, 0); 474 } 475 if (!so2) 476 goto drop; 477 } 478 so = so2; 479 /* 480 * This is ugly, but .... 481 * 482 * Mark socket as temporary until we're 483 * committed to keeping it. The code at 484 * ``drop'' and ``dropwithreset'' check the 485 * flag dropsocket to see if the temporary 486 * socket created here should be discarded. 487 * We mark the socket as discardable until 488 * we're committed to it below in TCPS_LISTEN. 489 */ 490 dropsocket++; 491 inp = (struct inpcb *)so->so_pcb; 492 inp->inp_laddr = ti->ti_dst; 493 inp->inp_lport = ti->ti_dport; 494 if (in_pcbinshash(inp) != 0) { 495 /* 496 * Undo the assignments above if we failed to put 497 * the PCB on the hash lists. 498 */ 499 inp->inp_laddr.s_addr = INADDR_ANY; 500 inp->inp_lport = 0; 501 goto drop; 502 } 503 inp->inp_options = ip_srcroute(); 504 tp = intotcpcb(inp); 505 tp->t_state = TCPS_LISTEN; 506 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 507 508 /* Compute proper scaling value from buffer space */ 509 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 510 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 511 tp->request_r_scale++; 512 } 513 } 514 515 /* 516 * Segment received on connection. 517 * Reset idle time and keep-alive timer. 518 */ 519 tp->t_idle = 0; 520 if (TCPS_HAVEESTABLISHED(tp->t_state)) 521 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 522 523 /* 524 * Process options if not in LISTEN state, 525 * else do it below (after getting remote address). 526 */ 527 if (tp->t_state != TCPS_LISTEN) 528 tcp_dooptions(tp, optp, optlen, ti, &to); 529 530 /* 531 * Header prediction: check for the two common cases 532 * of a uni-directional data xfer. If the packet has 533 * no control flags, is in-sequence, the window didn't 534 * change and we're not retransmitting, it's a 535 * candidate. If the length is zero and the ack moved 536 * forward, we're the sender side of the xfer. Just 537 * free the data acked & wake any higher level process 538 * that was blocked waiting for space. If the length 539 * is non-zero and the ack didn't move, we're the 540 * receiver side. If we're getting packets in-order 541 * (the reassembly queue is empty), add the data to 542 * the socket buffer and note that we need a delayed ack. 543 * Make sure that the hidden state-flags are also off. 544 * Since we check for TCPS_ESTABLISHED above, it can only 545 * be TH_NEEDSYN. 546 */ 547 if (tp->t_state == TCPS_ESTABLISHED && 548 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 549 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 550 ((to.to_flag & TOF_TS) == 0 || 551 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 552 /* 553 * Using the CC option is compulsory if once started: 554 * the segment is OK if no T/TCP was negotiated or 555 * if the segment has a CC option equal to CCrecv 556 */ 557 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 558 ((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 559 ti->ti_seq == tp->rcv_nxt && 560 tiwin && tiwin == tp->snd_wnd && 561 tp->snd_nxt == tp->snd_max) { 562 563 /* 564 * If last ACK falls within this segment's sequence numbers, 565 * record the timestamp. 566 * NOTE that the test is modified according to the latest 567 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 568 */ 569 if ((to.to_flag & TOF_TS) != 0 && 570 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 571 tp->ts_recent_age = tcp_now; 572 tp->ts_recent = to.to_tsval; 573 } 574 575 if (ti->ti_len == 0) { 576 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 577 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 578 tp->snd_cwnd >= tp->snd_wnd && 579 tp->t_dupacks < tcprexmtthresh) { 580 /* 581 * this is a pure ack for outstanding data. 582 */ 583 ++tcpstat.tcps_predack; 584 if ((to.to_flag & TOF_TS) != 0) 585 tcp_xmit_timer(tp, 586 tcp_now - to.to_tsecr + 1); 587 else if (tp->t_rtt && 588 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 589 tcp_xmit_timer(tp, tp->t_rtt); 590 acked = ti->ti_ack - tp->snd_una; 591 tcpstat.tcps_rcvackpack++; 592 tcpstat.tcps_rcvackbyte += acked; 593 sbdrop(&so->so_snd, acked); 594 tp->snd_una = ti->ti_ack; 595 m_freem(m); 596 597 /* 598 * If all outstanding data are acked, stop 599 * retransmit timer, otherwise restart timer 600 * using current (possibly backed-off) value. 601 * If process is waiting for space, 602 * wakeup/selwakeup/signal. If data 603 * are ready to send, let tcp_output 604 * decide between more output or persist. 605 */ 606 if (tp->snd_una == tp->snd_max) 607 tp->t_timer[TCPT_REXMT] = 0; 608 else if (tp->t_timer[TCPT_PERSIST] == 0) 609 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 610 611 sowwakeup(so); 612 if (so->so_snd.sb_cc) 613 (void) tcp_output(tp); 614 return; 615 } 616 } else if (ti->ti_ack == tp->snd_una && 617 tp->t_segq == NULL && 618 ti->ti_len <= sbspace(&so->so_rcv)) { 619 /* 620 * this is a pure, in-sequence data packet 621 * with nothing on the reassembly queue and 622 * we have enough buffer space to take it. 623 */ 624 ++tcpstat.tcps_preddat; 625 tp->rcv_nxt += ti->ti_len; 626 tcpstat.tcps_rcvpack++; 627 tcpstat.tcps_rcvbyte += ti->ti_len; 628 /* 629 * Add data to socket buffer. 630 */ 631 sbappend(&so->so_rcv, m); 632 sorwakeup(so); 633 if (tcp_delack_enabled) { 634 tp->t_flags |= TF_DELACK; 635 } else { 636 tp->t_flags |= TF_ACKNOW; 637 tcp_output(tp); 638 } 639 return; 640 } 641 } 642 643 /* 644 * Calculate amount of space in receive window, 645 * and then do TCP input processing. 646 * Receive window is amount of space in rcv queue, 647 * but not less than advertised window. 648 */ 649 { int win; 650 651 win = sbspace(&so->so_rcv); 652 if (win < 0) 653 win = 0; 654 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 655 } 656 657 switch (tp->t_state) { 658 659 /* 660 * If the state is LISTEN then ignore segment if it contains an RST. 661 * If the segment contains an ACK then it is bad and send a RST. 662 * If it does not contain a SYN then it is not interesting; drop it. 663 * If it is from this socket, drop it, it must be forged. 664 * Don't bother responding if the destination was a broadcast. 665 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 666 * tp->iss, and send a segment: 667 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 668 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 669 * Fill in remote peer address fields if not previously specified. 670 * Enter SYN_RECEIVED state, and process any other fields of this 671 * segment in this state. 672 */ 673 case TCPS_LISTEN: { 674 register struct sockaddr_in *sin; 675 676 if (tiflags & TH_RST) 677 goto drop; 678 if (tiflags & TH_ACK) 679 goto dropwithreset; 680 if ((tiflags & TH_SYN) == 0) 681 goto drop; 682 if ((ti->ti_dport == ti->ti_sport) && 683 (ti->ti_dst.s_addr == ti->ti_src.s_addr)) 684 goto drop; 685 /* 686 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 687 * in_broadcast() should never return true on a received 688 * packet with M_BCAST not set. 689 */ 690 if (m->m_flags & (M_BCAST|M_MCAST) || 691 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 692 goto drop; 693 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 694 M_NOWAIT); 695 if (sin == NULL) 696 goto drop; 697 sin->sin_family = AF_INET; 698 sin->sin_len = sizeof(*sin); 699 sin->sin_addr = ti->ti_src; 700 sin->sin_port = ti->ti_sport; 701 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 702 laddr = inp->inp_laddr; 703 if (inp->inp_laddr.s_addr == INADDR_ANY) 704 inp->inp_laddr = ti->ti_dst; 705 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 706 inp->inp_laddr = laddr; 707 FREE(sin, M_SONAME); 708 goto drop; 709 } 710 FREE(sin, M_SONAME); 711 tp->t_template = tcp_template(tp); 712 if (tp->t_template == 0) { 713 tp = tcp_drop(tp, ENOBUFS); 714 dropsocket = 0; /* socket is already gone */ 715 goto drop; 716 } 717 if ((taop = tcp_gettaocache(inp)) == NULL) { 718 taop = &tao_noncached; 719 bzero(taop, sizeof(*taop)); 720 } 721 tcp_dooptions(tp, optp, optlen, ti, &to); 722 if (iss) 723 tp->iss = iss; 724 else 725 tp->iss = tcp_iss; 726 tcp_iss += TCP_ISSINCR/4; 727 tp->irs = ti->ti_seq; 728 tcp_sendseqinit(tp); 729 tcp_rcvseqinit(tp); 730 /* 731 * Initialization of the tcpcb for transaction; 732 * set SND.WND = SEG.WND, 733 * initialize CCsend and CCrecv. 734 */ 735 tp->snd_wnd = tiwin; /* initial send-window */ 736 tp->cc_send = CC_INC(tcp_ccgen); 737 tp->cc_recv = to.to_cc; 738 /* 739 * Perform TAO test on incoming CC (SEG.CC) option, if any. 740 * - compare SEG.CC against cached CC from the same host, 741 * if any. 742 * - if SEG.CC > chached value, SYN must be new and is accepted 743 * immediately: save new CC in the cache, mark the socket 744 * connected, enter ESTABLISHED state, turn on flag to 745 * send a SYN in the next segment. 746 * A virtual advertised window is set in rcv_adv to 747 * initialize SWS prevention. Then enter normal segment 748 * processing: drop SYN, process data and FIN. 749 * - otherwise do a normal 3-way handshake. 750 */ 751 if ((to.to_flag & TOF_CC) != 0) { 752 if (((tp->t_flags & TF_NOPUSH) != 0) && 753 taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 754 755 taop->tao_cc = to.to_cc; 756 tp->t_state = TCPS_ESTABLISHED; 757 758 /* 759 * If there is a FIN, or if there is data and the 760 * connection is local, then delay SYN,ACK(SYN) in 761 * the hope of piggy-backing it on a response 762 * segment. Otherwise must send ACK now in case 763 * the other side is slow starting. 764 */ 765 if (tcp_delack_enabled && ((tiflags & TH_FIN) || (ti->ti_len != 0 && 766 in_localaddr(inp->inp_faddr)))) 767 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 768 else 769 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 770 771 /* 772 * Limit the `virtual advertised window' to TCP_MAXWIN 773 * here. Even if we requested window scaling, it will 774 * become effective only later when our SYN is acked. 775 */ 776 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 777 tcpstat.tcps_connects++; 778 soisconnected(so); 779 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 780 dropsocket = 0; /* committed to socket */ 781 tcpstat.tcps_accepts++; 782 goto trimthenstep6; 783 } 784 /* else do standard 3-way handshake */ 785 } else { 786 /* 787 * No CC option, but maybe CC.NEW: 788 * invalidate cached value. 789 */ 790 taop->tao_cc = 0; 791 } 792 /* 793 * TAO test failed or there was no CC option, 794 * do a standard 3-way handshake. 795 */ 796 tp->t_flags |= TF_ACKNOW; 797 tp->t_state = TCPS_SYN_RECEIVED; 798 tp->t_timer[TCPT_KEEP] = tcp_keepinit; 799 dropsocket = 0; /* committed to socket */ 800 tcpstat.tcps_accepts++; 801 goto trimthenstep6; 802 } 803 804 /* 805 * If the state is SYN_RECEIVED: 806 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 807 */ 808 case TCPS_SYN_RECEIVED: 809 if ((tiflags & TH_ACK) && 810 (SEQ_LEQ(ti->ti_ack, tp->snd_una) || 811 SEQ_GT(ti->ti_ack, tp->snd_max))) 812 goto dropwithreset; 813 break; 814 815 /* 816 * If the state is SYN_SENT: 817 * if seg contains an ACK, but not for our SYN, drop the input. 818 * if seg contains a RST, then drop the connection. 819 * if seg does not contain SYN, then drop it. 820 * Otherwise this is an acceptable SYN segment 821 * initialize tp->rcv_nxt and tp->irs 822 * if seg contains ack then advance tp->snd_una 823 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 824 * arrange for segment to be acked (eventually) 825 * continue processing rest of data/controls, beginning with URG 826 */ 827 case TCPS_SYN_SENT: 828 if ((taop = tcp_gettaocache(inp)) == NULL) { 829 taop = &tao_noncached; 830 bzero(taop, sizeof(*taop)); 831 } 832 833 if ((tiflags & TH_ACK) && 834 (SEQ_LEQ(ti->ti_ack, tp->iss) || 835 SEQ_GT(ti->ti_ack, tp->snd_max))) { 836 /* 837 * If we have a cached CCsent for the remote host, 838 * hence we haven't just crashed and restarted, 839 * do not send a RST. This may be a retransmission 840 * from the other side after our earlier ACK was lost. 841 * Our new SYN, when it arrives, will serve as the 842 * needed ACK. 843 */ 844 if (taop->tao_ccsent != 0) 845 goto drop; 846 else 847 goto dropwithreset; 848 } 849 if (tiflags & TH_RST) { 850 if (tiflags & TH_ACK) 851 tp = tcp_drop(tp, ECONNREFUSED); 852 goto drop; 853 } 854 if ((tiflags & TH_SYN) == 0) 855 goto drop; 856 tp->snd_wnd = ti->ti_win; /* initial send window */ 857 tp->cc_recv = to.to_cc; /* foreign CC */ 858 859 tp->irs = ti->ti_seq; 860 tcp_rcvseqinit(tp); 861 if (tiflags & TH_ACK) { 862 /* 863 * Our SYN was acked. If segment contains CC.ECHO 864 * option, check it to make sure this segment really 865 * matches our SYN. If not, just drop it as old 866 * duplicate, but send an RST if we're still playing 867 * by the old rules. If no CC.ECHO option, make sure 868 * we don't get fooled into using T/TCP. 869 */ 870 if (to.to_flag & TOF_CCECHO) { 871 if (tp->cc_send != to.to_ccecho) { 872 if (taop->tao_ccsent != 0) 873 goto drop; 874 else 875 goto dropwithreset; 876 } 877 } else 878 tp->t_flags &= ~TF_RCVD_CC; 879 tcpstat.tcps_connects++; 880 soisconnected(so); 881 /* Do window scaling on this connection? */ 882 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 883 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 884 tp->snd_scale = tp->requested_s_scale; 885 tp->rcv_scale = tp->request_r_scale; 886 } 887 /* Segment is acceptable, update cache if undefined. */ 888 if (taop->tao_ccsent == 0) 889 taop->tao_ccsent = to.to_ccecho; 890 891 tp->rcv_adv += tp->rcv_wnd; 892 tp->snd_una++; /* SYN is acked */ 893 /* 894 * If there's data, delay ACK; if there's also a FIN 895 * ACKNOW will be turned on later. 896 */ 897 if (tcp_delack_enabled && ti->ti_len != 0) 898 tp->t_flags |= TF_DELACK; 899 else 900 tp->t_flags |= TF_ACKNOW; 901 /* 902 * Received <SYN,ACK> in SYN_SENT[*] state. 903 * Transitions: 904 * SYN_SENT --> ESTABLISHED 905 * SYN_SENT* --> FIN_WAIT_1 906 */ 907 if (tp->t_flags & TF_NEEDFIN) { 908 tp->t_state = TCPS_FIN_WAIT_1; 909 tp->t_flags &= ~TF_NEEDFIN; 910 tiflags &= ~TH_SYN; 911 } else { 912 tp->t_state = TCPS_ESTABLISHED; 913 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 914 } 915 } else { 916 /* 917 * Received initial SYN in SYN-SENT[*] state => simul- 918 * taneous open. If segment contains CC option and there is 919 * a cached CC, apply TAO test; if it succeeds, connection is 920 * half-synchronized. Otherwise, do 3-way handshake: 921 * SYN-SENT -> SYN-RECEIVED 922 * SYN-SENT* -> SYN-RECEIVED* 923 * If there was no CC option, clear cached CC value. 924 */ 925 tp->t_flags |= TF_ACKNOW; 926 tp->t_timer[TCPT_REXMT] = 0; 927 if (to.to_flag & TOF_CC) { 928 if (taop->tao_cc != 0 && 929 CC_GT(to.to_cc, taop->tao_cc)) { 930 /* 931 * update cache and make transition: 932 * SYN-SENT -> ESTABLISHED* 933 * SYN-SENT* -> FIN-WAIT-1* 934 */ 935 taop->tao_cc = to.to_cc; 936 if (tp->t_flags & TF_NEEDFIN) { 937 tp->t_state = TCPS_FIN_WAIT_1; 938 tp->t_flags &= ~TF_NEEDFIN; 939 } else { 940 tp->t_state = TCPS_ESTABLISHED; 941 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 942 } 943 tp->t_flags |= TF_NEEDSYN; 944 } else 945 tp->t_state = TCPS_SYN_RECEIVED; 946 } else { 947 /* CC.NEW or no option => invalidate cache */ 948 taop->tao_cc = 0; 949 tp->t_state = TCPS_SYN_RECEIVED; 950 } 951 } 952 953trimthenstep6: 954 /* 955 * Advance ti->ti_seq to correspond to first data byte. 956 * If data, trim to stay within window, 957 * dropping FIN if necessary. 958 */ 959 ti->ti_seq++; 960 if (ti->ti_len > tp->rcv_wnd) { 961 todrop = ti->ti_len - tp->rcv_wnd; 962 m_adj(m, -todrop); 963 ti->ti_len = tp->rcv_wnd; 964 tiflags &= ~TH_FIN; 965 tcpstat.tcps_rcvpackafterwin++; 966 tcpstat.tcps_rcvbyteafterwin += todrop; 967 } 968 tp->snd_wl1 = ti->ti_seq - 1; 969 tp->rcv_up = ti->ti_seq; 970 /* 971 * Client side of transaction: already sent SYN and data. 972 * If the remote host used T/TCP to validate the SYN, 973 * our data will be ACK'd; if so, enter normal data segment 974 * processing in the middle of step 5, ack processing. 975 * Otherwise, goto step 6. 976 */ 977 if (tiflags & TH_ACK) 978 goto process_ACK; 979 goto step6; 980 /* 981 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 982 * if segment contains a SYN and CC [not CC.NEW] option: 983 * if state == TIME_WAIT and connection duration > MSL, 984 * drop packet and send RST; 985 * 986 * if SEG.CC > CCrecv then is new SYN, and can implicitly 987 * ack the FIN (and data) in retransmission queue. 988 * Complete close and delete TCPCB. Then reprocess 989 * segment, hoping to find new TCPCB in LISTEN state; 990 * 991 * else must be old SYN; drop it. 992 * else do normal processing. 993 */ 994 case TCPS_LAST_ACK: 995 case TCPS_CLOSING: 996 case TCPS_TIME_WAIT: 997 if ((tiflags & TH_SYN) && 998 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 999 if (tp->t_state == TCPS_TIME_WAIT && 1000 tp->t_duration > TCPTV_MSL) 1001 goto dropwithreset; 1002 if (CC_GT(to.to_cc, tp->cc_recv)) { 1003 tp = tcp_close(tp); 1004 goto findpcb; 1005 } 1006 else 1007 goto drop; 1008 } 1009 break; /* continue normal processing */ 1010 } 1011 1012 /* 1013 * States other than LISTEN or SYN_SENT. 1014 * First check the RST flag and sequence number since reset segments 1015 * are exempt from the timestamp and connection count tests. This 1016 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1017 * below which allowed reset segments in half the sequence space 1018 * to fall though and be processed (which gives forged reset 1019 * segments with a random sequence number a 50 percent chance of 1020 * killing a connection). 1021 * Then check timestamp, if present. 1022 * Then check the connection count, if present. 1023 * Then check that at least some bytes of segment are within 1024 * receive window. If segment begins before rcv_nxt, 1025 * drop leading data (and SYN); if nothing left, just ack. 1026 * 1027 * 1028 * If the RST bit is set, check the sequence number to see 1029 * if this is a valid reset segment. 1030 * RFC 793 page 37: 1031 * In all states except SYN-SENT, all reset (RST) segments 1032 * are validated by checking their SEQ-fields. A reset is 1033 * valid if its sequence number is in the window. 1034 * Note: this does not take into account delayed ACKs, so 1035 * we should test against last_ack_sent instead of rcv_nxt. 1036 * Also, it does not make sense to allow reset segments with 1037 * sequence numbers greater than last_ack_sent to be processed 1038 * since these sequence numbers are just the acknowledgement 1039 * numbers in our outgoing packets being echoed back at us, 1040 * and these acknowledgement numbers are monotonically 1041 * increasing. 1042 * If we have multiple segments in flight, the intial reset 1043 * segment sequence numbers will be to the left of last_ack_sent, 1044 * but they will eventually catch up. 1045 * In any case, it never made sense to trim reset segments to 1046 * fit the receive window since RFC 1122 says: 1047 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1048 * 1049 * A TCP SHOULD allow a received RST segment to include data. 1050 * 1051 * DISCUSSION 1052 * It has been suggested that a RST segment could contain 1053 * ASCII text that encoded and explained the cause of the 1054 * RST. No standard has yet been established for such 1055 * data. 1056 * 1057 * If the reset segment passes the sequence number test examine 1058 * the state: 1059 * SYN_RECEIVED STATE: 1060 * If passive open, return to LISTEN state. 1061 * If active open, inform user that connection was refused. 1062 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1063 * Inform user that connection was reset, and close tcb. 1064 * CLOSING, LAST_ACK STATES: 1065 * Close the tcb. 1066 * TIME_WAIT STATE: 1067 * Drop the segment - see Stevens, vol. 2, p. 964 and 1068 * RFC 1337. 1069 */ 1070 if (tiflags & TH_RST) { 1071 if (tp->last_ack_sent == ti->ti_seq) { 1072 switch (tp->t_state) { 1073 1074 case TCPS_SYN_RECEIVED: 1075 so->so_error = ECONNREFUSED; 1076 goto close; 1077 1078 case TCPS_ESTABLISHED: 1079 case TCPS_FIN_WAIT_1: 1080 case TCPS_FIN_WAIT_2: 1081 case TCPS_CLOSE_WAIT: 1082 so->so_error = ECONNRESET; 1083 close: 1084 tp->t_state = TCPS_CLOSED; 1085 tcpstat.tcps_drops++; 1086 tp = tcp_close(tp); 1087 break; 1088 1089 case TCPS_CLOSING: 1090 case TCPS_LAST_ACK: 1091 tp = tcp_close(tp); 1092 break; 1093 1094 case TCPS_TIME_WAIT: 1095 break; 1096 } 1097 } 1098 goto drop; 1099 } 1100 1101 /* 1102 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1103 * and it's less than ts_recent, drop it. 1104 */ 1105 if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent && 1106 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1107 1108 /* Check to see if ts_recent is over 24 days old. */ 1109 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1110 /* 1111 * Invalidate ts_recent. If this segment updates 1112 * ts_recent, the age will be reset later and ts_recent 1113 * will get a valid value. If it does not, setting 1114 * ts_recent to zero will at least satisfy the 1115 * requirement that zero be placed in the timestamp 1116 * echo reply when ts_recent isn't valid. The 1117 * age isn't reset until we get a valid ts_recent 1118 * because we don't want out-of-order segments to be 1119 * dropped when ts_recent is old. 1120 */ 1121 tp->ts_recent = 0; 1122 } else { 1123 tcpstat.tcps_rcvduppack++; 1124 tcpstat.tcps_rcvdupbyte += ti->ti_len; 1125 tcpstat.tcps_pawsdrop++; 1126 goto dropafterack; 1127 } 1128 } 1129 1130 /* 1131 * T/TCP mechanism 1132 * If T/TCP was negotiated and the segment doesn't have CC, 1133 * or if its CC is wrong then drop the segment. 1134 * RST segments do not have to comply with this. 1135 */ 1136 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1137 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1138 goto dropafterack; 1139 1140 /* 1141 * In the SYN-RECEIVED state, validate that the packet belongs to 1142 * this connection before trimming the data to fit the receive 1143 * window. Check the sequence number versus IRS since we know 1144 * the sequence numbers haven't wrapped. This is a partial fix 1145 * for the "LAND" DoS attack. 1146 */ 1147 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(ti->ti_seq, tp->irs)) 1148 goto dropwithreset; 1149 1150 todrop = tp->rcv_nxt - ti->ti_seq; 1151 if (todrop > 0) { 1152 if (tiflags & TH_SYN) { 1153 tiflags &= ~TH_SYN; 1154 ti->ti_seq++; 1155 if (ti->ti_urp > 1) 1156 ti->ti_urp--; 1157 else 1158 tiflags &= ~TH_URG; 1159 todrop--; 1160 } 1161 /* 1162 * Following if statement from Stevens, vol. 2, p. 960. 1163 */ 1164 if (todrop > ti->ti_len 1165 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 1166 /* 1167 * Any valid FIN must be to the left of the window. 1168 * At this point the FIN must be a duplicate or out 1169 * of sequence; drop it. 1170 */ 1171 tiflags &= ~TH_FIN; 1172 1173 /* 1174 * Send an ACK to resynchronize and drop any data. 1175 * But keep on processing for RST or ACK. 1176 */ 1177 tp->t_flags |= TF_ACKNOW; 1178 todrop = ti->ti_len; 1179 tcpstat.tcps_rcvduppack++; 1180 tcpstat.tcps_rcvdupbyte += todrop; 1181 } else { 1182 tcpstat.tcps_rcvpartduppack++; 1183 tcpstat.tcps_rcvpartdupbyte += todrop; 1184 } 1185 m_adj(m, todrop); 1186 ti->ti_seq += todrop; 1187 ti->ti_len -= todrop; 1188 if (ti->ti_urp > todrop) 1189 ti->ti_urp -= todrop; 1190 else { 1191 tiflags &= ~TH_URG; 1192 ti->ti_urp = 0; 1193 } 1194 } 1195 1196 /* 1197 * If new data are received on a connection after the 1198 * user processes are gone, then RST the other end. 1199 */ 1200 if ((so->so_state & SS_NOFDREF) && 1201 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 1202 tp = tcp_close(tp); 1203 tcpstat.tcps_rcvafterclose++; 1204 goto dropwithreset; 1205 } 1206 1207 /* 1208 * If segment ends after window, drop trailing data 1209 * (and PUSH and FIN); if nothing left, just ACK. 1210 */ 1211 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 1212 if (todrop > 0) { 1213 tcpstat.tcps_rcvpackafterwin++; 1214 if (todrop >= ti->ti_len) { 1215 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1216 /* 1217 * If a new connection request is received 1218 * while in TIME_WAIT, drop the old connection 1219 * and start over if the sequence numbers 1220 * are above the previous ones. 1221 */ 1222 if (tiflags & TH_SYN && 1223 tp->t_state == TCPS_TIME_WAIT && 1224 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1225 iss = tp->snd_nxt + TCP_ISSINCR; 1226 tp = tcp_close(tp); 1227 goto findpcb; 1228 } 1229 /* 1230 * If window is closed can only take segments at 1231 * window edge, and have to drop data and PUSH from 1232 * incoming segments. Continue processing, but 1233 * remember to ack. Otherwise, drop segment 1234 * and ack. 1235 */ 1236 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1237 tp->t_flags |= TF_ACKNOW; 1238 tcpstat.tcps_rcvwinprobe++; 1239 } else 1240 goto dropafterack; 1241 } else 1242 tcpstat.tcps_rcvbyteafterwin += todrop; 1243 m_adj(m, -todrop); 1244 ti->ti_len -= todrop; 1245 tiflags &= ~(TH_PUSH|TH_FIN); 1246 } 1247 1248 /* 1249 * If last ACK falls within this segment's sequence numbers, 1250 * record its timestamp. 1251 * NOTE that the test is modified according to the latest 1252 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1253 */ 1254 if ((to.to_flag & TOF_TS) != 0 && 1255 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) { 1256 tp->ts_recent_age = tcp_now; 1257 tp->ts_recent = to.to_tsval; 1258 } 1259 1260 /* 1261 * If a SYN is in the window, then this is an 1262 * error and we send an RST and drop the connection. 1263 */ 1264 if (tiflags & TH_SYN) { 1265 tp = tcp_drop(tp, ECONNRESET); 1266 goto dropwithreset; 1267 } 1268 1269 /* 1270 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1271 * flag is on (half-synchronized state), then queue data for 1272 * later processing; else drop segment and return. 1273 */ 1274 if ((tiflags & TH_ACK) == 0) { 1275 if (tp->t_state == TCPS_SYN_RECEIVED || 1276 (tp->t_flags & TF_NEEDSYN)) 1277 goto step6; 1278 else 1279 goto drop; 1280 } 1281 1282 /* 1283 * Ack processing. 1284 */ 1285 switch (tp->t_state) { 1286 1287 /* 1288 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1289 * ESTABLISHED state and continue processing. 1290 * The ACK was checked above. 1291 */ 1292 case TCPS_SYN_RECEIVED: 1293 1294 tcpstat.tcps_connects++; 1295 soisconnected(so); 1296 /* Do window scaling? */ 1297 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1298 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1299 tp->snd_scale = tp->requested_s_scale; 1300 tp->rcv_scale = tp->request_r_scale; 1301 } 1302 /* 1303 * Upon successful completion of 3-way handshake, 1304 * update cache.CC if it was undefined, pass any queued 1305 * data to the user, and advance state appropriately. 1306 */ 1307 if ((taop = tcp_gettaocache(inp)) != NULL && 1308 taop->tao_cc == 0) 1309 taop->tao_cc = tp->cc_recv; 1310 1311 /* 1312 * Make transitions: 1313 * SYN-RECEIVED -> ESTABLISHED 1314 * SYN-RECEIVED* -> FIN-WAIT-1 1315 */ 1316 if (tp->t_flags & TF_NEEDFIN) { 1317 tp->t_state = TCPS_FIN_WAIT_1; 1318 tp->t_flags &= ~TF_NEEDFIN; 1319 } else { 1320 tp->t_state = TCPS_ESTABLISHED; 1321 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 1322 } 1323 /* 1324 * If segment contains data or ACK, will call tcp_reass() 1325 * later; if not, do so now to pass queued data to user. 1326 */ 1327 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0) 1328 (void) tcp_reass(tp, (struct tcpiphdr *)0, 1329 (struct mbuf *)0); 1330 tp->snd_wl1 = ti->ti_seq - 1; 1331 /* fall into ... */ 1332 1333 /* 1334 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1335 * ACKs. If the ack is in the range 1336 * tp->snd_una < ti->ti_ack <= tp->snd_max 1337 * then advance tp->snd_una to ti->ti_ack and drop 1338 * data from the retransmission queue. If this ACK reflects 1339 * more up to date window information we update our window information. 1340 */ 1341 case TCPS_ESTABLISHED: 1342 case TCPS_FIN_WAIT_1: 1343 case TCPS_FIN_WAIT_2: 1344 case TCPS_CLOSE_WAIT: 1345 case TCPS_CLOSING: 1346 case TCPS_LAST_ACK: 1347 case TCPS_TIME_WAIT: 1348 1349 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1350 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1351 tcpstat.tcps_rcvdupack++; 1352 /* 1353 * If we have outstanding data (other than 1354 * a window probe), this is a completely 1355 * duplicate ack (ie, window info didn't 1356 * change), the ack is the biggest we've 1357 * seen and we've seen exactly our rexmt 1358 * threshhold of them, assume a packet 1359 * has been dropped and retransmit it. 1360 * Kludge snd_nxt & the congestion 1361 * window so we send only this one 1362 * packet. 1363 * 1364 * We know we're losing at the current 1365 * window size so do congestion avoidance 1366 * (set ssthresh to half the current window 1367 * and pull our congestion window back to 1368 * the new ssthresh). 1369 * 1370 * Dup acks mean that packets have left the 1371 * network (they're now cached at the receiver) 1372 * so bump cwnd by the amount in the receiver 1373 * to keep a constant cwnd packets in the 1374 * network. 1375 */ 1376 if (tp->t_timer[TCPT_REXMT] == 0 || 1377 ti->ti_ack != tp->snd_una) 1378 tp->t_dupacks = 0; 1379 else if (++tp->t_dupacks == tcprexmtthresh) { 1380 tcp_seq onxt = tp->snd_nxt; 1381 u_int win = 1382 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1383 tp->t_maxseg; 1384 1385 if (win < 2) 1386 win = 2; 1387 tp->snd_ssthresh = win * tp->t_maxseg; 1388 tp->t_timer[TCPT_REXMT] = 0; 1389 tp->t_rtt = 0; 1390 tp->snd_nxt = ti->ti_ack; 1391 tp->snd_cwnd = tp->t_maxseg; 1392 (void) tcp_output(tp); 1393 tp->snd_cwnd = tp->snd_ssthresh + 1394 tp->t_maxseg * tp->t_dupacks; 1395 if (SEQ_GT(onxt, tp->snd_nxt)) 1396 tp->snd_nxt = onxt; 1397 goto drop; 1398 } else if (tp->t_dupacks > tcprexmtthresh) { 1399 tp->snd_cwnd += tp->t_maxseg; 1400 (void) tcp_output(tp); 1401 goto drop; 1402 } 1403 } else 1404 tp->t_dupacks = 0; 1405 break; 1406 } 1407 /* 1408 * If the congestion window was inflated to account 1409 * for the other side's cached packets, retract it. 1410 */ 1411 if (tp->t_dupacks >= tcprexmtthresh && 1412 tp->snd_cwnd > tp->snd_ssthresh) 1413 tp->snd_cwnd = tp->snd_ssthresh; 1414 tp->t_dupacks = 0; 1415 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1416 tcpstat.tcps_rcvacktoomuch++; 1417 goto dropafterack; 1418 } 1419 /* 1420 * If we reach this point, ACK is not a duplicate, 1421 * i.e., it ACKs something we sent. 1422 */ 1423 if (tp->t_flags & TF_NEEDSYN) { 1424 /* 1425 * T/TCP: Connection was half-synchronized, and our 1426 * SYN has been ACK'd (so connection is now fully 1427 * synchronized). Go to non-starred state, 1428 * increment snd_una for ACK of SYN, and check if 1429 * we can do window scaling. 1430 */ 1431 tp->t_flags &= ~TF_NEEDSYN; 1432 tp->snd_una++; 1433 /* Do window scaling? */ 1434 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1435 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1436 tp->snd_scale = tp->requested_s_scale; 1437 tp->rcv_scale = tp->request_r_scale; 1438 } 1439 } 1440 1441process_ACK: 1442 acked = ti->ti_ack - tp->snd_una; 1443 tcpstat.tcps_rcvackpack++; 1444 tcpstat.tcps_rcvackbyte += acked; 1445 1446 /* 1447 * If we have a timestamp reply, update smoothed 1448 * round trip time. If no timestamp is present but 1449 * transmit timer is running and timed sequence 1450 * number was acked, update smoothed round trip time. 1451 * Since we now have an rtt measurement, cancel the 1452 * timer backoff (cf., Phil Karn's retransmit alg.). 1453 * Recompute the initial retransmit timer. 1454 */ 1455 if (to.to_flag & TOF_TS) 1456 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1); 1457 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1458 tcp_xmit_timer(tp,tp->t_rtt); 1459 1460 /* 1461 * If all outstanding data is acked, stop retransmit 1462 * timer and remember to restart (more output or persist). 1463 * If there is more data to be acked, restart retransmit 1464 * timer, using current (possibly backed-off) value. 1465 */ 1466 if (ti->ti_ack == tp->snd_max) { 1467 tp->t_timer[TCPT_REXMT] = 0; 1468 needoutput = 1; 1469 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1470 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1471 1472 /* 1473 * If no data (only SYN) was ACK'd, 1474 * skip rest of ACK processing. 1475 */ 1476 if (acked == 0) 1477 goto step6; 1478 1479 /* 1480 * When new data is acked, open the congestion window. 1481 * If the window gives us less than ssthresh packets 1482 * in flight, open exponentially (maxseg per packet). 1483 * Otherwise open linearly: maxseg per window 1484 * (maxseg^2 / cwnd per packet). 1485 */ 1486 { 1487 register u_int cw = tp->snd_cwnd; 1488 register u_int incr = tp->t_maxseg; 1489 1490 if (cw > tp->snd_ssthresh) 1491 incr = incr * incr / cw; 1492 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1493 } 1494 if (acked > so->so_snd.sb_cc) { 1495 tp->snd_wnd -= so->so_snd.sb_cc; 1496 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1497 ourfinisacked = 1; 1498 } else { 1499 sbdrop(&so->so_snd, acked); 1500 tp->snd_wnd -= acked; 1501 ourfinisacked = 0; 1502 } 1503 sowwakeup(so); 1504 tp->snd_una = ti->ti_ack; 1505 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1506 tp->snd_nxt = tp->snd_una; 1507 1508 switch (tp->t_state) { 1509 1510 /* 1511 * In FIN_WAIT_1 STATE in addition to the processing 1512 * for the ESTABLISHED state if our FIN is now acknowledged 1513 * then enter FIN_WAIT_2. 1514 */ 1515 case TCPS_FIN_WAIT_1: 1516 if (ourfinisacked) { 1517 /* 1518 * If we can't receive any more 1519 * data, then closing user can proceed. 1520 * Starting the timer is contrary to the 1521 * specification, but if we don't get a FIN 1522 * we'll hang forever. 1523 */ 1524 if (so->so_state & SS_CANTRCVMORE) { 1525 soisdisconnected(so); 1526 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1527 } 1528 tp->t_state = TCPS_FIN_WAIT_2; 1529 } 1530 break; 1531 1532 /* 1533 * In CLOSING STATE in addition to the processing for 1534 * the ESTABLISHED state if the ACK acknowledges our FIN 1535 * then enter the TIME-WAIT state, otherwise ignore 1536 * the segment. 1537 */ 1538 case TCPS_CLOSING: 1539 if (ourfinisacked) { 1540 tp->t_state = TCPS_TIME_WAIT; 1541 tcp_canceltimers(tp); 1542 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1543 if (tp->cc_recv != 0 && 1544 tp->t_duration < TCPTV_MSL) 1545 tp->t_timer[TCPT_2MSL] = 1546 tp->t_rxtcur * TCPTV_TWTRUNC; 1547 else 1548 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1549 soisdisconnected(so); 1550 } 1551 break; 1552 1553 /* 1554 * In LAST_ACK, we may still be waiting for data to drain 1555 * and/or to be acked, as well as for the ack of our FIN. 1556 * If our FIN is now acknowledged, delete the TCB, 1557 * enter the closed state and return. 1558 */ 1559 case TCPS_LAST_ACK: 1560 if (ourfinisacked) { 1561 tp = tcp_close(tp); 1562 goto drop; 1563 } 1564 break; 1565 1566 /* 1567 * In TIME_WAIT state the only thing that should arrive 1568 * is a retransmission of the remote FIN. Acknowledge 1569 * it and restart the finack timer. 1570 */ 1571 case TCPS_TIME_WAIT: 1572 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1573 goto dropafterack; 1574 } 1575 } 1576 1577step6: 1578 /* 1579 * Update window information. 1580 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1581 */ 1582 if ((tiflags & TH_ACK) && 1583 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || 1584 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1585 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) { 1586 /* keep track of pure window updates */ 1587 if (ti->ti_len == 0 && 1588 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1589 tcpstat.tcps_rcvwinupd++; 1590 tp->snd_wnd = tiwin; 1591 tp->snd_wl1 = ti->ti_seq; 1592 tp->snd_wl2 = ti->ti_ack; 1593 if (tp->snd_wnd > tp->max_sndwnd) 1594 tp->max_sndwnd = tp->snd_wnd; 1595 needoutput = 1; 1596 } 1597 1598 /* 1599 * Process segments with URG. 1600 */ 1601 if ((tiflags & TH_URG) && ti->ti_urp && 1602 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1603 /* 1604 * This is a kludge, but if we receive and accept 1605 * random urgent pointers, we'll crash in 1606 * soreceive. It's hard to imagine someone 1607 * actually wanting to send this much urgent data. 1608 */ 1609 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1610 ti->ti_urp = 0; /* XXX */ 1611 tiflags &= ~TH_URG; /* XXX */ 1612 goto dodata; /* XXX */ 1613 } 1614 /* 1615 * If this segment advances the known urgent pointer, 1616 * then mark the data stream. This should not happen 1617 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1618 * a FIN has been received from the remote side. 1619 * In these states we ignore the URG. 1620 * 1621 * According to RFC961 (Assigned Protocols), 1622 * the urgent pointer points to the last octet 1623 * of urgent data. We continue, however, 1624 * to consider it to indicate the first octet 1625 * of data past the urgent section as the original 1626 * spec states (in one of two places). 1627 */ 1628 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1629 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1630 so->so_oobmark = so->so_rcv.sb_cc + 1631 (tp->rcv_up - tp->rcv_nxt) - 1; 1632 if (so->so_oobmark == 0) 1633 so->so_state |= SS_RCVATMARK; 1634 sohasoutofband(so); 1635 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1636 } 1637 /* 1638 * Remove out of band data so doesn't get presented to user. 1639 * This can happen independent of advancing the URG pointer, 1640 * but if two URG's are pending at once, some out-of-band 1641 * data may creep in... ick. 1642 */ 1643 if (ti->ti_urp <= (u_long)ti->ti_len 1644#ifdef SO_OOBINLINE 1645 && (so->so_options & SO_OOBINLINE) == 0 1646#endif 1647 ) 1648 tcp_pulloutofband(so, ti, m); 1649 } else 1650 /* 1651 * If no out of band data is expected, 1652 * pull receive urgent pointer along 1653 * with the receive window. 1654 */ 1655 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1656 tp->rcv_up = tp->rcv_nxt; 1657dodata: /* XXX */ 1658 1659 /* 1660 * Process the segment text, merging it into the TCP sequencing queue, 1661 * and arranging for acknowledgment of receipt if necessary. 1662 * This process logically involves adjusting tp->rcv_wnd as data 1663 * is presented to the user (this happens in tcp_usrreq.c, 1664 * case PRU_RCVD). If a FIN has already been received on this 1665 * connection then we just ignore the text. 1666 */ 1667 if ((ti->ti_len || (tiflags&TH_FIN)) && 1668 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1669 TCP_REASS(tp, ti, m, so, tiflags); 1670 /* 1671 * Note the amount of data that peer has sent into 1672 * our window, in order to estimate the sender's 1673 * buffer size. 1674 */ 1675 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1676 } else { 1677 m_freem(m); 1678 tiflags &= ~TH_FIN; 1679 } 1680 1681 /* 1682 * If FIN is received ACK the FIN and let the user know 1683 * that the connection is closing. 1684 */ 1685 if (tiflags & TH_FIN) { 1686 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1687 socantrcvmore(so); 1688 /* 1689 * If connection is half-synchronized 1690 * (ie NEEDSYN flag on) then delay ACK, 1691 * so it may be piggybacked when SYN is sent. 1692 * Otherwise, since we received a FIN then no 1693 * more input can be expected, send ACK now. 1694 */ 1695 if (tcp_delack_enabled && (tp->t_flags & TF_NEEDSYN)) 1696 tp->t_flags |= TF_DELACK; 1697 else 1698 tp->t_flags |= TF_ACKNOW; 1699 tp->rcv_nxt++; 1700 } 1701 switch (tp->t_state) { 1702 1703 /* 1704 * In SYN_RECEIVED and ESTABLISHED STATES 1705 * enter the CLOSE_WAIT state. 1706 */ 1707 case TCPS_SYN_RECEIVED: 1708 case TCPS_ESTABLISHED: 1709 tp->t_state = TCPS_CLOSE_WAIT; 1710 break; 1711 1712 /* 1713 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1714 * enter the CLOSING state. 1715 */ 1716 case TCPS_FIN_WAIT_1: 1717 tp->t_state = TCPS_CLOSING; 1718 break; 1719 1720 /* 1721 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1722 * starting the time-wait timer, turning off the other 1723 * standard timers. 1724 */ 1725 case TCPS_FIN_WAIT_2: 1726 tp->t_state = TCPS_TIME_WAIT; 1727 tcp_canceltimers(tp); 1728 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1729 if (tp->cc_recv != 0 && 1730 tp->t_duration < TCPTV_MSL) { 1731 tp->t_timer[TCPT_2MSL] = 1732 tp->t_rxtcur * TCPTV_TWTRUNC; 1733 /* For transaction client, force ACK now. */ 1734 tp->t_flags |= TF_ACKNOW; 1735 } 1736 else 1737 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1738 soisdisconnected(so); 1739 break; 1740 1741 /* 1742 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1743 */ 1744 case TCPS_TIME_WAIT: 1745 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1746 break; 1747 } 1748 } 1749#ifdef TCPDEBUG 1750 if (so->so_options & SO_DEBUG) 1751 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1752#endif 1753 1754 /* 1755 * Return any desired output. 1756 */ 1757 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1758 (void) tcp_output(tp); 1759 return; 1760 1761dropafterack: 1762 /* 1763 * Generate an ACK dropping incoming segment if it occupies 1764 * sequence space, where the ACK reflects our state. 1765 * 1766 * We can now skip the test for the RST flag since all 1767 * paths to this code happen after packets containing 1768 * RST have been dropped. 1769 * 1770 * In the SYN-RECEIVED state, don't send an ACK unless the 1771 * segment we received passes the SYN-RECEIVED ACK test. 1772 * If it fails send a RST. This breaks the loop in the 1773 * "LAND" DoS attack, and also prevents an ACK storm 1774 * between two listening ports that have been sent forged 1775 * SYN segments, each with the source address of the other. 1776 */ 1777 if (tp->t_state == TCPS_SYN_RECEIVED && (tiflags & TH_ACK) && 1778 (SEQ_GT(tp->snd_una, ti->ti_ack) || 1779 SEQ_GT(ti->ti_ack, tp->snd_max)) ) 1780 goto dropwithreset; 1781#ifdef TCPDEBUG 1782 if (so->so_options & SO_DEBUG) 1783 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1784#endif 1785 m_freem(m); 1786 tp->t_flags |= TF_ACKNOW; 1787 (void) tcp_output(tp); 1788 return; 1789 1790dropwithreset: 1791 /* 1792 * Generate a RST, dropping incoming segment. 1793 * Make ACK acceptable to originator of segment. 1794 * Don't bother to respond if destination was broadcast/multicast. 1795 */ 1796 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1797 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1798 goto drop; 1799#ifdef TCPDEBUG 1800 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1801 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1802#endif 1803 if (tiflags & TH_ACK) 1804 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1805 else { 1806 if (tiflags & TH_SYN) 1807 ti->ti_len++; 1808 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1809 TH_RST|TH_ACK); 1810 } 1811 /* destroy temporarily created socket */ 1812 if (dropsocket) 1813 (void) soabort(so); 1814 return; 1815 1816drop: 1817 /* 1818 * Drop space held by incoming segment and return. 1819 */ 1820#ifdef TCPDEBUG 1821 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1822 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1823#endif 1824 m_freem(m); 1825 /* destroy temporarily created socket */ 1826 if (dropsocket) 1827 (void) soabort(so); 1828 return; 1829} 1830 1831static void 1832tcp_dooptions(tp, cp, cnt, ti, to) 1833 struct tcpcb *tp; 1834 u_char *cp; 1835 int cnt; 1836 struct tcpiphdr *ti; 1837 struct tcpopt *to; 1838{ 1839 u_short mss = 0; 1840 int opt, optlen; 1841 1842 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1843 opt = cp[0]; 1844 if (opt == TCPOPT_EOL) 1845 break; 1846 if (opt == TCPOPT_NOP) 1847 optlen = 1; 1848 else { 1849 optlen = cp[1]; 1850 if (optlen <= 0) 1851 break; 1852 } 1853 switch (opt) { 1854 1855 default: 1856 continue; 1857 1858 case TCPOPT_MAXSEG: 1859 if (optlen != TCPOLEN_MAXSEG) 1860 continue; 1861 if (!(ti->ti_flags & TH_SYN)) 1862 continue; 1863 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1864 NTOHS(mss); 1865 break; 1866 1867 case TCPOPT_WINDOW: 1868 if (optlen != TCPOLEN_WINDOW) 1869 continue; 1870 if (!(ti->ti_flags & TH_SYN)) 1871 continue; 1872 tp->t_flags |= TF_RCVD_SCALE; 1873 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1874 break; 1875 1876 case TCPOPT_TIMESTAMP: 1877 if (optlen != TCPOLEN_TIMESTAMP) 1878 continue; 1879 to->to_flag |= TOF_TS; 1880 bcopy((char *)cp + 2, 1881 (char *)&to->to_tsval, sizeof(to->to_tsval)); 1882 NTOHL(to->to_tsval); 1883 bcopy((char *)cp + 6, 1884 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 1885 NTOHL(to->to_tsecr); 1886 1887 /* 1888 * A timestamp received in a SYN makes 1889 * it ok to send timestamp requests and replies. 1890 */ 1891 if (ti->ti_flags & TH_SYN) { 1892 tp->t_flags |= TF_RCVD_TSTMP; 1893 tp->ts_recent = to->to_tsval; 1894 tp->ts_recent_age = tcp_now; 1895 } 1896 break; 1897 case TCPOPT_CC: 1898 if (optlen != TCPOLEN_CC) 1899 continue; 1900 to->to_flag |= TOF_CC; 1901 bcopy((char *)cp + 2, 1902 (char *)&to->to_cc, sizeof(to->to_cc)); 1903 NTOHL(to->to_cc); 1904 /* 1905 * A CC or CC.new option received in a SYN makes 1906 * it ok to send CC in subsequent segments. 1907 */ 1908 if (ti->ti_flags & TH_SYN) 1909 tp->t_flags |= TF_RCVD_CC; 1910 break; 1911 case TCPOPT_CCNEW: 1912 if (optlen != TCPOLEN_CC) 1913 continue; 1914 if (!(ti->ti_flags & TH_SYN)) 1915 continue; 1916 to->to_flag |= TOF_CCNEW; 1917 bcopy((char *)cp + 2, 1918 (char *)&to->to_cc, sizeof(to->to_cc)); 1919 NTOHL(to->to_cc); 1920 /* 1921 * A CC or CC.new option received in a SYN makes 1922 * it ok to send CC in subsequent segments. 1923 */ 1924 tp->t_flags |= TF_RCVD_CC; 1925 break; 1926 case TCPOPT_CCECHO: 1927 if (optlen != TCPOLEN_CC) 1928 continue; 1929 if (!(ti->ti_flags & TH_SYN)) 1930 continue; 1931 to->to_flag |= TOF_CCECHO; 1932 bcopy((char *)cp + 2, 1933 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 1934 NTOHL(to->to_ccecho); 1935 break; 1936 } 1937 } 1938 if (ti->ti_flags & TH_SYN) 1939 tcp_mss(tp, mss); /* sets t_maxseg */ 1940} 1941 1942/* 1943 * Pull out of band byte out of a segment so 1944 * it doesn't appear in the user's data queue. 1945 * It is still reflected in the segment length for 1946 * sequencing purposes. 1947 */ 1948static void 1949tcp_pulloutofband(so, ti, m) 1950 struct socket *so; 1951 struct tcpiphdr *ti; 1952 register struct mbuf *m; 1953{ 1954 int cnt = ti->ti_urp - 1; 1955 1956 while (cnt >= 0) { 1957 if (m->m_len > cnt) { 1958 char *cp = mtod(m, caddr_t) + cnt; 1959 struct tcpcb *tp = sototcpcb(so); 1960 1961 tp->t_iobc = *cp; 1962 tp->t_oobflags |= TCPOOB_HAVEDATA; 1963 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1964 m->m_len--; 1965 return; 1966 } 1967 cnt -= m->m_len; 1968 m = m->m_next; 1969 if (m == 0) 1970 break; 1971 } 1972 panic("tcp_pulloutofband"); 1973} 1974 1975/* 1976 * Collect new round-trip time estimate 1977 * and update averages and current timeout. 1978 */ 1979static void 1980tcp_xmit_timer(tp, rtt) 1981 register struct tcpcb *tp; 1982 short rtt; 1983{ 1984 register int delta; 1985 1986 tcpstat.tcps_rttupdated++; 1987 tp->t_rttupdated++; 1988 if (tp->t_srtt != 0) { 1989 /* 1990 * srtt is stored as fixed point with 5 bits after the 1991 * binary point (i.e., scaled by 8). The following magic 1992 * is equivalent to the smoothing algorithm in rfc793 with 1993 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1994 * point). Adjust rtt to origin 0. 1995 */ 1996 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 1997 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 1998 1999 if ((tp->t_srtt += delta) <= 0) 2000 tp->t_srtt = 1; 2001 2002 /* 2003 * We accumulate a smoothed rtt variance (actually, a 2004 * smoothed mean difference), then set the retransmit 2005 * timer to smoothed rtt + 4 times the smoothed variance. 2006 * rttvar is stored as fixed point with 4 bits after the 2007 * binary point (scaled by 16). The following is 2008 * equivalent to rfc793 smoothing with an alpha of .75 2009 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2010 * rfc793's wired-in beta. 2011 */ 2012 if (delta < 0) 2013 delta = -delta; 2014 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2015 if ((tp->t_rttvar += delta) <= 0) 2016 tp->t_rttvar = 1; 2017 } else { 2018 /* 2019 * No rtt measurement yet - use the unsmoothed rtt. 2020 * Set the variance to half the rtt (so our first 2021 * retransmit happens at 3*rtt). 2022 */ 2023 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2024 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2025 } 2026 tp->t_rtt = 0; 2027 tp->t_rxtshift = 0; 2028 2029 /* 2030 * the retransmit should happen at rtt + 4 * rttvar. 2031 * Because of the way we do the smoothing, srtt and rttvar 2032 * will each average +1/2 tick of bias. When we compute 2033 * the retransmit timer, we want 1/2 tick of rounding and 2034 * 1 extra tick because of +-1/2 tick uncertainty in the 2035 * firing of the timer. The bias will give us exactly the 2036 * 1.5 tick we need. But, because the bias is 2037 * statistical, we have to test that we don't drop below 2038 * the minimum feasible timer (which is 2 ticks). 2039 */ 2040 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2041 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2042 2043 /* 2044 * We received an ack for a packet that wasn't retransmitted; 2045 * it is probably safe to discard any error indications we've 2046 * received recently. This isn't quite right, but close enough 2047 * for now (a route might have failed after we sent a segment, 2048 * and the return path might not be symmetrical). 2049 */ 2050 tp->t_softerror = 0; 2051} 2052 2053/* 2054 * Determine a reasonable value for maxseg size. 2055 * If the route is known, check route for mtu. 2056 * If none, use an mss that can be handled on the outgoing 2057 * interface without forcing IP to fragment; if bigger than 2058 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2059 * to utilize large mbufs. If no route is found, route has no mtu, 2060 * or the destination isn't local, use a default, hopefully conservative 2061 * size (usually 512 or the default IP max size, but no more than the mtu 2062 * of the interface), as we can't discover anything about intervening 2063 * gateways or networks. We also initialize the congestion/slow start 2064 * window to be a single segment if the destination isn't local. 2065 * While looking at the routing entry, we also initialize other path-dependent 2066 * parameters from pre-set or cached values in the routing entry. 2067 * 2068 * Also take into account the space needed for options that we 2069 * send regularly. Make maxseg shorter by that amount to assure 2070 * that we can send maxseg amount of data even when the options 2071 * are present. Store the upper limit of the length of options plus 2072 * data in maxopd. 2073 * 2074 * NOTE that this routine is only called when we process an incoming 2075 * segment, for outgoing segments only tcp_mssopt is called. 2076 * 2077 * In case of T/TCP, we call this routine during implicit connection 2078 * setup as well (offer = -1), to initialize maxseg from the cached 2079 * MSS of our peer. 2080 */ 2081void 2082tcp_mss(tp, offer) 2083 struct tcpcb *tp; 2084 int offer; 2085{ 2086 register struct rtentry *rt; 2087 struct ifnet *ifp; 2088 register int rtt, mss; 2089 u_long bufsize; 2090 struct inpcb *inp; 2091 struct socket *so; 2092 struct rmxp_tao *taop; 2093 int origoffer = offer; 2094 2095 inp = tp->t_inpcb; 2096 if ((rt = tcp_rtlookup(inp)) == NULL) { 2097 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2098 return; 2099 } 2100 ifp = rt->rt_ifp; 2101 so = inp->inp_socket; 2102 2103 taop = rmx_taop(rt->rt_rmx); 2104 /* 2105 * Offer == -1 means that we didn't receive SYN yet, 2106 * use cached value in that case; 2107 */ 2108 if (offer == -1) 2109 offer = taop->tao_mssopt; 2110 /* 2111 * Offer == 0 means that there was no MSS on the SYN segment, 2112 * in this case we use tcp_mssdflt. 2113 */ 2114 if (offer == 0) 2115 offer = tcp_mssdflt; 2116 else 2117 /* 2118 * Sanity check: make sure that maxopd will be large 2119 * enough to allow some data on segments even is the 2120 * all the option space is used (40bytes). Otherwise 2121 * funny things may happen in tcp_output. 2122 */ 2123 offer = max(offer, 64); 2124 taop->tao_mssopt = offer; 2125 2126 /* 2127 * While we're here, check if there's an initial rtt 2128 * or rttvar. Convert from the route-table units 2129 * to scaled multiples of the slow timeout timer. 2130 */ 2131 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2132 /* 2133 * XXX the lock bit for RTT indicates that the value 2134 * is also a minimum value; this is subject to time. 2135 */ 2136 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2137 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 2138 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 2139 tcpstat.tcps_usedrtt++; 2140 if (rt->rt_rmx.rmx_rttvar) { 2141 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2142 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 2143 tcpstat.tcps_usedrttvar++; 2144 } else { 2145 /* default variation is +- 1 rtt */ 2146 tp->t_rttvar = 2147 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2148 } 2149 TCPT_RANGESET(tp->t_rxtcur, 2150 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2151 tp->t_rttmin, TCPTV_REXMTMAX); 2152 } 2153 /* 2154 * if there's an mtu associated with the route, use it 2155 */ 2156 if (rt->rt_rmx.rmx_mtu) 2157 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 2158 else 2159 { 2160 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 2161 if (!in_localaddr(inp->inp_faddr)) 2162 mss = min(mss, tcp_mssdflt); 2163 } 2164 mss = min(mss, offer); 2165 /* 2166 * maxopd stores the maximum length of data AND options 2167 * in a segment; maxseg is the amount of data in a normal 2168 * segment. We need to store this value (maxopd) apart 2169 * from maxseg, because now every segment carries options 2170 * and thus we normally have somewhat less data in segments. 2171 */ 2172 tp->t_maxopd = mss; 2173 2174 /* 2175 * In case of T/TCP, origoffer==-1 indicates, that no segments 2176 * were received yet. In this case we just guess, otherwise 2177 * we do the same as before T/TCP. 2178 */ 2179 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2180 (origoffer == -1 || 2181 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2182 mss -= TCPOLEN_TSTAMP_APPA; 2183 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2184 (origoffer == -1 || 2185 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2186 mss -= TCPOLEN_CC_APPA; 2187 2188#if (MCLBYTES & (MCLBYTES - 1)) == 0 2189 if (mss > MCLBYTES) 2190 mss &= ~(MCLBYTES-1); 2191#else 2192 if (mss > MCLBYTES) 2193 mss = mss / MCLBYTES * MCLBYTES; 2194#endif 2195 /* 2196 * If there's a pipesize, change the socket buffer 2197 * to that size. Make the socket buffers an integral 2198 * number of mss units; if the mss is larger than 2199 * the socket buffer, decrease the mss. 2200 */ 2201#ifdef RTV_SPIPE 2202 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2203#endif 2204 bufsize = so->so_snd.sb_hiwat; 2205 if (bufsize < mss) 2206 mss = bufsize; 2207 else { 2208 bufsize = roundup(bufsize, mss); 2209 if (bufsize > sb_max) 2210 bufsize = sb_max; 2211 (void)sbreserve(&so->so_snd, bufsize); 2212 } 2213 tp->t_maxseg = mss; 2214 2215#ifdef RTV_RPIPE 2216 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2217#endif 2218 bufsize = so->so_rcv.sb_hiwat; 2219 if (bufsize > mss) { 2220 bufsize = roundup(bufsize, mss); 2221 if (bufsize > sb_max) 2222 bufsize = sb_max; 2223 (void)sbreserve(&so->so_rcv, bufsize); 2224 } 2225 /* 2226 * Don't force slow-start on local network. 2227 */ 2228 if (!in_localaddr(inp->inp_faddr)) 2229 tp->snd_cwnd = mss; 2230 2231 if (rt->rt_rmx.rmx_ssthresh) { 2232 /* 2233 * There's some sort of gateway or interface 2234 * buffer limit on the path. Use this to set 2235 * the slow start threshhold, but set the 2236 * threshold to no less than 2*mss. 2237 */ 2238 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2239 tcpstat.tcps_usedssthresh++; 2240 } 2241} 2242 2243/* 2244 * Determine the MSS option to send on an outgoing SYN. 2245 */ 2246int 2247tcp_mssopt(tp) 2248 struct tcpcb *tp; 2249{ 2250 struct rtentry *rt; 2251 2252 rt = tcp_rtlookup(tp->t_inpcb); 2253 if (rt == NULL) 2254 return tcp_mssdflt; 2255 2256 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr); 2257} 2258