1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)tcp_input.c 8.4 (Berkeley) 01/11/94 8 */ 9 10 #ifndef TUBA_INCLUDE 11 #include <sys/param.h> 12 #include <sys/systm.h> 13 #include <sys/malloc.h> 14 #include <sys/mbuf.h> 15 #include <sys/protosw.h> 16 #include <sys/socket.h> 17 #include <sys/socketvar.h> 18 #include <sys/errno.h> 19 20 #include <net/if.h> 21 #include <net/route.h> 22 23 #include <netinet/in.h> 24 #include <netinet/in_systm.h> 25 #include <netinet/ip.h> 26 #include <netinet/in_pcb.h> 27 #include <netinet/ip_var.h> 28 #include <netinet/tcp.h> 29 #include <netinet/tcp_fsm.h> 30 #include <netinet/tcp_seq.h> 31 #include <netinet/tcp_timer.h> 32 #include <netinet/tcp_var.h> 33 #include <netinet/tcpip.h> 34 #include <netinet/tcp_debug.h> 35 36 int tcprexmtthresh = 3; 37 int tcppredack; /* XXX debugging: times hdr predict ok for acks */ 38 int tcppreddat; /* XXX # times header prediction ok for data packets */ 39 int tcppcbcachemiss; 40 struct tcpiphdr tcp_saveti; 41 struct inpcb *tcp_last_inpcb = &tcb; 42 43 extern u_long sb_max; 44 45 #endif /* TUBA_INCLUDE */ 46 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 47 48 /* for modulo comparisons of timestamps */ 49 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 50 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 51 52 53 /* 54 * Insert segment ti into reassembly queue of tcp with 55 * control block tp. Return TH_FIN if reassembly now includes 56 * a segment with FIN. The macro form does the common case inline 57 * (segment is the next to be received on an established connection, 58 * and the queue is empty), avoiding linkage into and removal 59 * from the queue and repetition of various conversions. 60 * Set DELACK for segments received in order, but ack immediately 61 * when segments are out of order (so fast retransmit can work). 62 */ 63 #define TCP_REASS(tp, ti, m, so, flags) { \ 64 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 65 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 66 (tp)->t_state == TCPS_ESTABLISHED) { \ 67 tp->t_flags |= TF_DELACK; \ 68 (tp)->rcv_nxt += (ti)->ti_len; \ 69 flags = (ti)->ti_flags & TH_FIN; \ 70 tcpstat.tcps_rcvpack++;\ 71 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 72 sbappend(&(so)->so_rcv, (m)); \ 73 sorwakeup(so); \ 74 } else { \ 75 (flags) = tcp_reass((tp), (ti), (m)); \ 76 tp->t_flags |= TF_ACKNOW; \ 77 } \ 78 } 79 #ifndef TUBA_INCLUDE 80 81 int 82 tcp_reass(tp, ti, m) 83 register struct tcpcb *tp; 84 register struct tcpiphdr *ti; 85 struct mbuf *m; 86 { 87 register struct tcpiphdr *q; 88 struct socket *so = tp->t_inpcb->inp_socket; 89 int flags; 90 91 /* 92 * Call with ti==0 after become established to 93 * force pre-ESTABLISHED data up to user socket. 94 */ 95 if (ti == 0) 96 goto present; 97 98 /* 99 * Find a segment which begins after this one does. 100 */ 101 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 102 q = (struct tcpiphdr *)q->ti_next) 103 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 104 break; 105 106 /* 107 * If there is a preceding segment, it may provide some of 108 * our data already. If so, drop the data from the incoming 109 * segment. If it provides all of our data, drop us. 110 */ 111 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 112 register int i; 113 q = (struct tcpiphdr *)q->ti_prev; 114 /* conversion to int (in i) handles seq wraparound */ 115 i = q->ti_seq + q->ti_len - ti->ti_seq; 116 if (i > 0) { 117 if (i >= ti->ti_len) { 118 tcpstat.tcps_rcvduppack++; 119 tcpstat.tcps_rcvdupbyte += ti->ti_len; 120 m_freem(m); 121 return (0); 122 } 123 m_adj(m, i); 124 ti->ti_len -= i; 125 ti->ti_seq += i; 126 } 127 q = (struct tcpiphdr *)(q->ti_next); 128 } 129 tcpstat.tcps_rcvoopack++; 130 tcpstat.tcps_rcvoobyte += ti->ti_len; 131 REASS_MBUF(ti) = m; /* XXX */ 132 133 /* 134 * While we overlap succeeding segments trim them or, 135 * if they are completely covered, dequeue them. 136 */ 137 while (q != (struct tcpiphdr *)tp) { 138 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 139 if (i <= 0) 140 break; 141 if (i < q->ti_len) { 142 q->ti_seq += i; 143 q->ti_len -= i; 144 m_adj(REASS_MBUF(q), i); 145 break; 146 } 147 q = (struct tcpiphdr *)q->ti_next; 148 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 149 remque(q->ti_prev); 150 m_freem(m); 151 } 152 153 /* 154 * Stick new segment in its place. 155 */ 156 insque(ti, q->ti_prev); 157 158 present: 159 /* 160 * Present data to user, advancing rcv_nxt through 161 * completed sequence space. 162 */ 163 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 164 return (0); 165 ti = tp->seg_next; 166 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 167 return (0); 168 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 169 return (0); 170 do { 171 tp->rcv_nxt += ti->ti_len; 172 flags = ti->ti_flags & TH_FIN; 173 remque(ti); 174 m = REASS_MBUF(ti); 175 ti = (struct tcpiphdr *)ti->ti_next; 176 if (so->so_state & SS_CANTRCVMORE) 177 m_freem(m); 178 else 179 sbappend(&so->so_rcv, m); 180 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 181 sorwakeup(so); 182 return (flags); 183 } 184 185 /* 186 * TCP input routine, follows pages 65-76 of the 187 * protocol specification dated September, 1981 very closely. 188 */ 189 void 190 tcp_input(m, iphlen) 191 register struct mbuf *m; 192 int iphlen; 193 { 194 register struct tcpiphdr *ti; 195 register struct inpcb *inp; 196 caddr_t optp = NULL; 197 int optlen; 198 int len, tlen, off; 199 register struct tcpcb *tp = 0; 200 register int tiflags; 201 struct socket *so; 202 int todrop, acked, ourfinisacked, needoutput = 0; 203 short ostate; 204 struct in_addr laddr; 205 int dropsocket = 0; 206 int iss = 0; 207 u_long tiwin, ts_val, ts_ecr; 208 int ts_present = 0; 209 210 tcpstat.tcps_rcvtotal++; 211 /* 212 * Get IP and TCP header together in first mbuf. 213 * Note: IP leaves IP header in first mbuf. 214 */ 215 ti = mtod(m, struct tcpiphdr *); 216 if (iphlen > sizeof (struct ip)) 217 ip_stripoptions(m, (struct mbuf *)0); 218 if (m->m_len < sizeof (struct tcpiphdr)) { 219 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 220 tcpstat.tcps_rcvshort++; 221 return; 222 } 223 ti = mtod(m, struct tcpiphdr *); 224 } 225 226 /* 227 * Checksum extended TCP header and data. 228 */ 229 tlen = ((struct ip *)ti)->ip_len; 230 len = sizeof (struct ip) + tlen; 231 ti->ti_next = ti->ti_prev = 0; 232 ti->ti_x1 = 0; 233 ti->ti_len = (u_short)tlen; 234 HTONS(ti->ti_len); 235 if (ti->ti_sum = in_cksum(m, len)) { 236 tcpstat.tcps_rcvbadsum++; 237 goto drop; 238 } 239 #endif /* TUBA_INCLUDE */ 240 241 /* 242 * Check that TCP offset makes sense, 243 * pull out TCP options and adjust length. XXX 244 */ 245 off = ti->ti_off << 2; 246 if (off < sizeof (struct tcphdr) || off > tlen) { 247 tcpstat.tcps_rcvbadoff++; 248 goto drop; 249 } 250 tlen -= off; 251 ti->ti_len = tlen; 252 if (off > sizeof (struct tcphdr)) { 253 if (m->m_len < sizeof(struct ip) + off) { 254 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 255 tcpstat.tcps_rcvshort++; 256 return; 257 } 258 ti = mtod(m, struct tcpiphdr *); 259 } 260 optlen = off - sizeof (struct tcphdr); 261 optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 262 /* 263 * Do quick retrieval of timestamp options ("options 264 * prediction?"). If timestamp is the only option and it's 265 * formatted as recommended in RFC 1323 appendix A, we 266 * quickly get the values now and not bother calling 267 * tcp_dooptions(), etc. 268 */ 269 if ((optlen == TCPOLEN_TSTAMP_APPA || 270 (optlen > TCPOLEN_TSTAMP_APPA && 271 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 272 *(u_long *)optp == htonl(TCPOPT_TSTAMP_HDR) && 273 (ti->ti_flags & TH_SYN) == 0) { 274 ts_present = 1; 275 ts_val = ntohl(*(u_long *)(optp + 4)); 276 ts_ecr = ntohl(*(u_long *)(optp + 8)); 277 optp = NULL; /* we've parsed the options */ 278 } 279 } 280 tiflags = ti->ti_flags; 281 282 /* 283 * Convert TCP protocol specific fields to host format. 284 */ 285 NTOHL(ti->ti_seq); 286 NTOHL(ti->ti_ack); 287 NTOHS(ti->ti_win); 288 NTOHS(ti->ti_urp); 289 290 /* 291 * Locate pcb for segment. 292 */ 293 findpcb: 294 inp = tcp_last_inpcb; 295 if (inp->inp_lport != ti->ti_dport || 296 inp->inp_fport != ti->ti_sport || 297 inp->inp_faddr.s_addr != ti->ti_src.s_addr || 298 inp->inp_laddr.s_addr != ti->ti_dst.s_addr) { 299 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 300 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 301 if (inp) 302 tcp_last_inpcb = inp; 303 ++tcppcbcachemiss; 304 } 305 306 /* 307 * If the state is CLOSED (i.e., TCB does not exist) then 308 * all data in the incoming segment is discarded. 309 * If the TCB exists but is in CLOSED state, it is embryonic, 310 * but should either do a listen or a connect soon. 311 */ 312 if (inp == 0) 313 goto dropwithreset; 314 tp = intotcpcb(inp); 315 if (tp == 0) 316 goto dropwithreset; 317 if (tp->t_state == TCPS_CLOSED) 318 goto drop; 319 320 /* Unscale the window into a 32-bit value. */ 321 if ((tiflags & TH_SYN) == 0) 322 tiwin = ti->ti_win << tp->snd_scale; 323 else 324 tiwin = ti->ti_win; 325 326 so = inp->inp_socket; 327 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 328 if (so->so_options & SO_DEBUG) { 329 ostate = tp->t_state; 330 tcp_saveti = *ti; 331 } 332 if (so->so_options & SO_ACCEPTCONN) { 333 so = sonewconn(so, 0); 334 if (so == 0) 335 goto drop; 336 /* 337 * This is ugly, but .... 338 * 339 * Mark socket as temporary until we're 340 * committed to keeping it. The code at 341 * ``drop'' and ``dropwithreset'' check the 342 * flag dropsocket to see if the temporary 343 * socket created here should be discarded. 344 * We mark the socket as discardable until 345 * we're committed to it below in TCPS_LISTEN. 346 */ 347 dropsocket++; 348 inp = (struct inpcb *)so->so_pcb; 349 inp->inp_laddr = ti->ti_dst; 350 inp->inp_lport = ti->ti_dport; 351 #if BSD>=43 352 inp->inp_options = ip_srcroute(); 353 #endif 354 tp = intotcpcb(inp); 355 tp->t_state = TCPS_LISTEN; 356 357 /* Compute proper scaling value from buffer space 358 */ 359 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 360 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 361 tp->request_r_scale++; 362 } 363 } 364 365 /* 366 * Segment received on connection. 367 * Reset idle time and keep-alive timer. 368 */ 369 tp->t_idle = 0; 370 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 371 372 /* 373 * Process options if not in LISTEN state, 374 * else do it below (after getting remote address). 375 */ 376 if (optp && tp->t_state != TCPS_LISTEN) 377 tcp_dooptions(tp, optp, optlen, ti, 378 &ts_present, &ts_val, &ts_ecr); 379 380 /* 381 * Header prediction: check for the two common cases 382 * of a uni-directional data xfer. If the packet has 383 * no control flags, is in-sequence, the window didn't 384 * change and we're not retransmitting, it's a 385 * candidate. If the length is zero and the ack moved 386 * forward, we're the sender side of the xfer. Just 387 * free the data acked & wake any higher level process 388 * that was blocked waiting for space. If the length 389 * is non-zero and the ack didn't move, we're the 390 * receiver side. If we're getting packets in-order 391 * (the reassembly queue is empty), add the data to 392 * the socket buffer and note that we need a delayed ack. 393 */ 394 if (tp->t_state == TCPS_ESTABLISHED && 395 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 396 (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && 397 ti->ti_seq == tp->rcv_nxt && 398 tiwin && tiwin == tp->snd_wnd && 399 tp->snd_nxt == tp->snd_max) { 400 401 /* 402 * If last ACK falls within this segment's sequence numbers, 403 * record the timestamp. 404 */ 405 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 406 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) { 407 tp->ts_recent_age = tcp_now; 408 tp->ts_recent = ts_val; 409 } 410 411 if (ti->ti_len == 0) { 412 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 413 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 414 tp->snd_cwnd >= tp->snd_wnd) { 415 /* 416 * this is a pure ack for outstanding data. 417 */ 418 ++tcppredack; 419 if (ts_present) 420 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 421 else if (tp->t_rtt && 422 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 423 tcp_xmit_timer(tp, tp->t_rtt); 424 acked = ti->ti_ack - tp->snd_una; 425 tcpstat.tcps_rcvackpack++; 426 tcpstat.tcps_rcvackbyte += acked; 427 sbdrop(&so->so_snd, acked); 428 tp->snd_una = ti->ti_ack; 429 m_freem(m); 430 431 /* 432 * If all outstanding data are acked, stop 433 * retransmit timer, otherwise restart timer 434 * using current (possibly backed-off) value. 435 * If process is waiting for space, 436 * wakeup/selwakeup/signal. If data 437 * are ready to send, let tcp_output 438 * decide between more output or persist. 439 */ 440 if (tp->snd_una == tp->snd_max) 441 tp->t_timer[TCPT_REXMT] = 0; 442 else if (tp->t_timer[TCPT_PERSIST] == 0) 443 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 444 445 if (so->so_snd.sb_flags & SB_NOTIFY) 446 sowwakeup(so); 447 if (so->so_snd.sb_cc) 448 (void) tcp_output(tp); 449 return; 450 } 451 } else if (ti->ti_ack == tp->snd_una && 452 tp->seg_next == (struct tcpiphdr *)tp && 453 ti->ti_len <= sbspace(&so->so_rcv)) { 454 /* 455 * this is a pure, in-sequence data packet 456 * with nothing on the reassembly queue and 457 * we have enough buffer space to take it. 458 */ 459 ++tcppreddat; 460 tp->rcv_nxt += ti->ti_len; 461 tcpstat.tcps_rcvpack++; 462 tcpstat.tcps_rcvbyte += ti->ti_len; 463 /* 464 * Drop TCP, IP headers and TCP options then add data 465 * to socket buffer. 466 */ 467 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 468 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 469 sbappend(&so->so_rcv, m); 470 sorwakeup(so); 471 tp->t_flags |= TF_DELACK; 472 return; 473 } 474 } 475 476 /* 477 * Drop TCP, IP headers and TCP options. 478 */ 479 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 480 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 481 482 /* 483 * Calculate amount of space in receive window, 484 * and then do TCP input processing. 485 * Receive window is amount of space in rcv queue, 486 * but not less than advertised window. 487 */ 488 { int win; 489 490 win = sbspace(&so->so_rcv); 491 if (win < 0) 492 win = 0; 493 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 494 } 495 496 switch (tp->t_state) { 497 498 /* 499 * If the state is LISTEN then ignore segment if it contains an RST. 500 * If the segment contains an ACK then it is bad and send a RST. 501 * If it does not contain a SYN then it is not interesting; drop it. 502 * Don't bother responding if the destination was a broadcast. 503 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 504 * tp->iss, and send a segment: 505 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 506 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 507 * Fill in remote peer address fields if not previously specified. 508 * Enter SYN_RECEIVED state, and process any other fields of this 509 * segment in this state. 510 */ 511 case TCPS_LISTEN: { 512 struct mbuf *am; 513 register struct sockaddr_in *sin; 514 515 if (tiflags & TH_RST) 516 goto drop; 517 if (tiflags & TH_ACK) 518 goto dropwithreset; 519 if ((tiflags & TH_SYN) == 0) 520 goto drop; 521 /* 522 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 523 * in_broadcast() should never return true on a received 524 * packet with M_BCAST not set. 525 */ 526 if (m->m_flags & (M_BCAST|M_MCAST) || 527 IN_MULTICAST(ti->ti_dst.s_addr)) 528 goto drop; 529 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 530 if (am == NULL) 531 goto drop; 532 am->m_len = sizeof (struct sockaddr_in); 533 sin = mtod(am, struct sockaddr_in *); 534 sin->sin_family = AF_INET; 535 sin->sin_len = sizeof(*sin); 536 sin->sin_addr = ti->ti_src; 537 sin->sin_port = ti->ti_sport; 538 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 539 laddr = inp->inp_laddr; 540 if (inp->inp_laddr.s_addr == INADDR_ANY) 541 inp->inp_laddr = ti->ti_dst; 542 if (in_pcbconnect(inp, am)) { 543 inp->inp_laddr = laddr; 544 (void) m_free(am); 545 goto drop; 546 } 547 (void) m_free(am); 548 tp->t_template = tcp_template(tp); 549 if (tp->t_template == 0) { 550 tp = tcp_drop(tp, ENOBUFS); 551 dropsocket = 0; /* socket is already gone */ 552 goto drop; 553 } 554 if (optp) 555 tcp_dooptions(tp, optp, optlen, ti, 556 &ts_present, &ts_val, &ts_ecr); 557 if (iss) 558 tp->iss = iss; 559 else 560 tp->iss = tcp_iss; 561 tcp_iss += TCP_ISSINCR/2; 562 tp->irs = ti->ti_seq; 563 tcp_sendseqinit(tp); 564 tcp_rcvseqinit(tp); 565 tp->t_flags |= TF_ACKNOW; 566 tp->t_state = TCPS_SYN_RECEIVED; 567 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 568 dropsocket = 0; /* committed to socket */ 569 tcpstat.tcps_accepts++; 570 goto trimthenstep6; 571 } 572 573 /* 574 * If the state is SYN_SENT: 575 * if seg contains an ACK, but not for our SYN, drop the input. 576 * if seg contains a RST, then drop the connection. 577 * if seg does not contain SYN, then drop it. 578 * Otherwise this is an acceptable SYN segment 579 * initialize tp->rcv_nxt and tp->irs 580 * if seg contains ack then advance tp->snd_una 581 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 582 * arrange for segment to be acked (eventually) 583 * continue processing rest of data/controls, beginning with URG 584 */ 585 case TCPS_SYN_SENT: 586 if ((tiflags & TH_ACK) && 587 (SEQ_LEQ(ti->ti_ack, tp->iss) || 588 SEQ_GT(ti->ti_ack, tp->snd_max))) 589 goto dropwithreset; 590 if (tiflags & TH_RST) { 591 if (tiflags & TH_ACK) 592 tp = tcp_drop(tp, ECONNREFUSED); 593 goto drop; 594 } 595 if ((tiflags & TH_SYN) == 0) 596 goto drop; 597 if (tiflags & TH_ACK) { 598 tp->snd_una = ti->ti_ack; 599 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 600 tp->snd_nxt = tp->snd_una; 601 } 602 tp->t_timer[TCPT_REXMT] = 0; 603 tp->irs = ti->ti_seq; 604 tcp_rcvseqinit(tp); 605 tp->t_flags |= TF_ACKNOW; 606 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 607 tcpstat.tcps_connects++; 608 soisconnected(so); 609 tp->t_state = TCPS_ESTABLISHED; 610 /* Do window scaling on this connection? */ 611 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 612 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 613 tp->snd_scale = tp->requested_s_scale; 614 tp->rcv_scale = tp->request_r_scale; 615 } 616 (void) tcp_reass(tp, (struct tcpiphdr *)0, 617 (struct mbuf *)0); 618 /* 619 * if we didn't have to retransmit the SYN, 620 * use its rtt as our initial srtt & rtt var. 621 */ 622 if (tp->t_rtt) 623 tcp_xmit_timer(tp, tp->t_rtt); 624 } else 625 tp->t_state = TCPS_SYN_RECEIVED; 626 627 trimthenstep6: 628 /* 629 * Advance ti->ti_seq to correspond to first data byte. 630 * If data, trim to stay within window, 631 * dropping FIN if necessary. 632 */ 633 ti->ti_seq++; 634 if (ti->ti_len > tp->rcv_wnd) { 635 todrop = ti->ti_len - tp->rcv_wnd; 636 m_adj(m, -todrop); 637 ti->ti_len = tp->rcv_wnd; 638 tiflags &= ~TH_FIN; 639 tcpstat.tcps_rcvpackafterwin++; 640 tcpstat.tcps_rcvbyteafterwin += todrop; 641 } 642 tp->snd_wl1 = ti->ti_seq - 1; 643 tp->rcv_up = ti->ti_seq; 644 goto step6; 645 } 646 647 /* 648 * States other than LISTEN or SYN_SENT. 649 * First check timestamp, if present. 650 * Then check that at least some bytes of segment are within 651 * receive window. If segment begins before rcv_nxt, 652 * drop leading data (and SYN); if nothing left, just ack. 653 * 654 * RFC 1323 PAWS: If we have a timestamp reply on this segment 655 * and it's less than ts_recent, drop it. 656 */ 657 if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 658 TSTMP_LT(ts_val, tp->ts_recent)) { 659 660 /* Check to see if ts_recent is over 24 days old. */ 661 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 662 /* 663 * Invalidate ts_recent. If this segment updates 664 * ts_recent, the age will be reset later and ts_recent 665 * will get a valid value. If it does not, setting 666 * ts_recent to zero will at least satisfy the 667 * requirement that zero be placed in the timestamp 668 * echo reply when ts_recent isn't valid. The 669 * age isn't reset until we get a valid ts_recent 670 * because we don't want out-of-order segments to be 671 * dropped when ts_recent is old. 672 */ 673 tp->ts_recent = 0; 674 } else { 675 tcpstat.tcps_rcvduppack++; 676 tcpstat.tcps_rcvdupbyte += ti->ti_len; 677 tcpstat.tcps_pawsdrop++; 678 goto dropafterack; 679 } 680 } 681 682 todrop = tp->rcv_nxt - ti->ti_seq; 683 if (todrop > 0) { 684 if (tiflags & TH_SYN) { 685 tiflags &= ~TH_SYN; 686 ti->ti_seq++; 687 if (ti->ti_urp > 1) 688 ti->ti_urp--; 689 else 690 tiflags &= ~TH_URG; 691 todrop--; 692 } 693 if (todrop >= ti->ti_len) { 694 tcpstat.tcps_rcvduppack++; 695 tcpstat.tcps_rcvdupbyte += ti->ti_len; 696 /* 697 * If segment is just one to the left of the window, 698 * check two special cases: 699 * 1. Don't toss RST in response to 4.2-style keepalive. 700 * 2. If the only thing to drop is a FIN, we can drop 701 * it, but check the ACK or we will get into FIN 702 * wars if our FINs crossed (both CLOSING). 703 * In either case, send ACK to resynchronize, 704 * but keep on processing for RST or ACK. 705 */ 706 if ((tiflags & TH_FIN && todrop == ti->ti_len + 1) 707 #ifdef TCP_COMPAT_42 708 || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1) 709 #endif 710 ) { 711 todrop = ti->ti_len; 712 tiflags &= ~TH_FIN; 713 tp->t_flags |= TF_ACKNOW; 714 } else { 715 /* 716 * Handle the case when a bound socket connects 717 * to itself. Allow packets with a SYN and 718 * an ACK to continue with the processing. 719 */ 720 if (todrop != 0 || (tiflags & TH_ACK) == 0) 721 goto dropafterack; 722 } 723 } else { 724 tcpstat.tcps_rcvpartduppack++; 725 tcpstat.tcps_rcvpartdupbyte += todrop; 726 } 727 m_adj(m, todrop); 728 ti->ti_seq += todrop; 729 ti->ti_len -= todrop; 730 if (ti->ti_urp > todrop) 731 ti->ti_urp -= todrop; 732 else { 733 tiflags &= ~TH_URG; 734 ti->ti_urp = 0; 735 } 736 } 737 738 /* 739 * If new data are received on a connection after the 740 * user processes are gone, then RST the other end. 741 */ 742 if ((so->so_state & SS_NOFDREF) && 743 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 744 tp = tcp_close(tp); 745 tcpstat.tcps_rcvafterclose++; 746 goto dropwithreset; 747 } 748 749 /* 750 * If segment ends after window, drop trailing data 751 * (and PUSH and FIN); if nothing left, just ACK. 752 */ 753 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 754 if (todrop > 0) { 755 tcpstat.tcps_rcvpackafterwin++; 756 if (todrop >= ti->ti_len) { 757 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 758 /* 759 * If a new connection request is received 760 * while in TIME_WAIT, drop the old connection 761 * and start over if the sequence numbers 762 * are above the previous ones. 763 */ 764 if (tiflags & TH_SYN && 765 tp->t_state == TCPS_TIME_WAIT && 766 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 767 iss = tp->rcv_nxt + TCP_ISSINCR; 768 tp = tcp_close(tp); 769 goto findpcb; 770 } 771 /* 772 * If window is closed can only take segments at 773 * window edge, and have to drop data and PUSH from 774 * incoming segments. Continue processing, but 775 * remember to ack. Otherwise, drop segment 776 * and ack. 777 */ 778 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 779 tp->t_flags |= TF_ACKNOW; 780 tcpstat.tcps_rcvwinprobe++; 781 } else 782 goto dropafterack; 783 } else 784 tcpstat.tcps_rcvbyteafterwin += todrop; 785 m_adj(m, -todrop); 786 ti->ti_len -= todrop; 787 tiflags &= ~(TH_PUSH|TH_FIN); 788 } 789 790 /* 791 * If last ACK falls within this segment's sequence numbers, 792 * record its timestamp. 793 */ 794 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 795 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + 796 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 797 tp->ts_recent_age = tcp_now; 798 tp->ts_recent = ts_val; 799 } 800 801 /* 802 * If the RST bit is set examine the state: 803 * SYN_RECEIVED STATE: 804 * If passive open, return to LISTEN state. 805 * If active open, inform user that connection was refused. 806 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 807 * Inform user that connection was reset, and close tcb. 808 * CLOSING, LAST_ACK, TIME_WAIT STATES 809 * Close the tcb. 810 */ 811 if (tiflags&TH_RST) switch (tp->t_state) { 812 813 case TCPS_SYN_RECEIVED: 814 so->so_error = ECONNREFUSED; 815 goto close; 816 817 case TCPS_ESTABLISHED: 818 case TCPS_FIN_WAIT_1: 819 case TCPS_FIN_WAIT_2: 820 case TCPS_CLOSE_WAIT: 821 so->so_error = ECONNRESET; 822 close: 823 tp->t_state = TCPS_CLOSED; 824 tcpstat.tcps_drops++; 825 tp = tcp_close(tp); 826 goto drop; 827 828 case TCPS_CLOSING: 829 case TCPS_LAST_ACK: 830 case TCPS_TIME_WAIT: 831 tp = tcp_close(tp); 832 goto drop; 833 } 834 835 /* 836 * If a SYN is in the window, then this is an 837 * error and we send an RST and drop the connection. 838 */ 839 if (tiflags & TH_SYN) { 840 tp = tcp_drop(tp, ECONNRESET); 841 goto dropwithreset; 842 } 843 844 /* 845 * If the ACK bit is off we drop the segment and return. 846 */ 847 if ((tiflags & TH_ACK) == 0) 848 goto drop; 849 850 /* 851 * Ack processing. 852 */ 853 switch (tp->t_state) { 854 855 /* 856 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 857 * ESTABLISHED state and continue processing, otherwise 858 * send an RST. 859 */ 860 case TCPS_SYN_RECEIVED: 861 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 862 SEQ_GT(ti->ti_ack, tp->snd_max)) 863 goto dropwithreset; 864 tcpstat.tcps_connects++; 865 soisconnected(so); 866 tp->t_state = TCPS_ESTABLISHED; 867 /* Do window scaling? */ 868 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 869 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 870 tp->snd_scale = tp->requested_s_scale; 871 tp->rcv_scale = tp->request_r_scale; 872 } 873 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 874 tp->snd_wl1 = ti->ti_seq - 1; 875 /* fall into ... */ 876 877 /* 878 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 879 * ACKs. If the ack is in the range 880 * tp->snd_una < ti->ti_ack <= tp->snd_max 881 * then advance tp->snd_una to ti->ti_ack and drop 882 * data from the retransmission queue. If this ACK reflects 883 * more up to date window information we update our window information. 884 */ 885 case TCPS_ESTABLISHED: 886 case TCPS_FIN_WAIT_1: 887 case TCPS_FIN_WAIT_2: 888 case TCPS_CLOSE_WAIT: 889 case TCPS_CLOSING: 890 case TCPS_LAST_ACK: 891 case TCPS_TIME_WAIT: 892 893 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 894 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 895 tcpstat.tcps_rcvdupack++; 896 /* 897 * If we have outstanding data (other than 898 * a window probe), this is a completely 899 * duplicate ack (ie, window info didn't 900 * change), the ack is the biggest we've 901 * seen and we've seen exactly our rexmt 902 * threshhold of them, assume a packet 903 * has been dropped and retransmit it. 904 * Kludge snd_nxt & the congestion 905 * window so we send only this one 906 * packet. 907 * 908 * We know we're losing at the current 909 * window size so do congestion avoidance 910 * (set ssthresh to half the current window 911 * and pull our congestion window back to 912 * the new ssthresh). 913 * 914 * Dup acks mean that packets have left the 915 * network (they're now cached at the receiver) 916 * so bump cwnd by the amount in the receiver 917 * to keep a constant cwnd packets in the 918 * network. 919 */ 920 if (tp->t_timer[TCPT_REXMT] == 0 || 921 ti->ti_ack != tp->snd_una) 922 tp->t_dupacks = 0; 923 else if (++tp->t_dupacks == tcprexmtthresh) { 924 tcp_seq onxt = tp->snd_nxt; 925 u_int win = 926 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 927 tp->t_maxseg; 928 929 if (win < 2) 930 win = 2; 931 tp->snd_ssthresh = win * tp->t_maxseg; 932 tp->t_timer[TCPT_REXMT] = 0; 933 tp->t_rtt = 0; 934 tp->snd_nxt = ti->ti_ack; 935 tp->snd_cwnd = tp->t_maxseg; 936 (void) tcp_output(tp); 937 tp->snd_cwnd = tp->snd_ssthresh + 938 tp->t_maxseg * tp->t_dupacks; 939 if (SEQ_GT(onxt, tp->snd_nxt)) 940 tp->snd_nxt = onxt; 941 goto drop; 942 } else if (tp->t_dupacks > tcprexmtthresh) { 943 tp->snd_cwnd += tp->t_maxseg; 944 (void) tcp_output(tp); 945 goto drop; 946 } 947 } else 948 tp->t_dupacks = 0; 949 break; 950 } 951 /* 952 * If the congestion window was inflated to account 953 * for the other side's cached packets, retract it. 954 */ 955 if (tp->t_dupacks > tcprexmtthresh && 956 tp->snd_cwnd > tp->snd_ssthresh) 957 tp->snd_cwnd = tp->snd_ssthresh; 958 tp->t_dupacks = 0; 959 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 960 tcpstat.tcps_rcvacktoomuch++; 961 goto dropafterack; 962 } 963 acked = ti->ti_ack - tp->snd_una; 964 tcpstat.tcps_rcvackpack++; 965 tcpstat.tcps_rcvackbyte += acked; 966 967 /* 968 * If we have a timestamp reply, update smoothed 969 * round trip time. If no timestamp is present but 970 * transmit timer is running and timed sequence 971 * number was acked, update smoothed round trip time. 972 * Since we now have an rtt measurement, cancel the 973 * timer backoff (cf., Phil Karn's retransmit alg.). 974 * Recompute the initial retransmit timer. 975 */ 976 if (ts_present) 977 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 978 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 979 tcp_xmit_timer(tp,tp->t_rtt); 980 981 /* 982 * If all outstanding data is acked, stop retransmit 983 * timer and remember to restart (more output or persist). 984 * If there is more data to be acked, restart retransmit 985 * timer, using current (possibly backed-off) value. 986 */ 987 if (ti->ti_ack == tp->snd_max) { 988 tp->t_timer[TCPT_REXMT] = 0; 989 needoutput = 1; 990 } else if (tp->t_timer[TCPT_PERSIST] == 0) 991 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 992 /* 993 * When new data is acked, open the congestion window. 994 * If the window gives us less than ssthresh packets 995 * in flight, open exponentially (maxseg per packet). 996 * Otherwise open linearly: maxseg per window 997 * (maxseg^2 / cwnd per packet), plus a constant 998 * fraction of a packet (maxseg/8) to help larger windows 999 * open quickly enough. 1000 */ 1001 { 1002 register u_int cw = tp->snd_cwnd; 1003 register u_int incr = tp->t_maxseg; 1004 1005 if (cw > tp->snd_ssthresh) 1006 incr = incr * incr / cw + incr / 8; 1007 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1008 } 1009 if (acked > so->so_snd.sb_cc) { 1010 tp->snd_wnd -= so->so_snd.sb_cc; 1011 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1012 ourfinisacked = 1; 1013 } else { 1014 sbdrop(&so->so_snd, acked); 1015 tp->snd_wnd -= acked; 1016 ourfinisacked = 0; 1017 } 1018 if (so->so_snd.sb_flags & SB_NOTIFY) 1019 sowwakeup(so); 1020 tp->snd_una = ti->ti_ack; 1021 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1022 tp->snd_nxt = tp->snd_una; 1023 1024 switch (tp->t_state) { 1025 1026 /* 1027 * In FIN_WAIT_1 STATE in addition to the processing 1028 * for the ESTABLISHED state if our FIN is now acknowledged 1029 * then enter FIN_WAIT_2. 1030 */ 1031 case TCPS_FIN_WAIT_1: 1032 if (ourfinisacked) { 1033 /* 1034 * If we can't receive any more 1035 * data, then closing user can proceed. 1036 * Starting the timer is contrary to the 1037 * specification, but if we don't get a FIN 1038 * we'll hang forever. 1039 */ 1040 if (so->so_state & SS_CANTRCVMORE) { 1041 soisdisconnected(so); 1042 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1043 } 1044 tp->t_state = TCPS_FIN_WAIT_2; 1045 } 1046 break; 1047 1048 /* 1049 * In CLOSING STATE in addition to the processing for 1050 * the ESTABLISHED state if the ACK acknowledges our FIN 1051 * then enter the TIME-WAIT state, otherwise ignore 1052 * the segment. 1053 */ 1054 case TCPS_CLOSING: 1055 if (ourfinisacked) { 1056 tp->t_state = TCPS_TIME_WAIT; 1057 tcp_canceltimers(tp); 1058 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1059 soisdisconnected(so); 1060 } 1061 break; 1062 1063 /* 1064 * In LAST_ACK, we may still be waiting for data to drain 1065 * and/or to be acked, as well as for the ack of our FIN. 1066 * If our FIN is now acknowledged, delete the TCB, 1067 * enter the closed state and return. 1068 */ 1069 case TCPS_LAST_ACK: 1070 if (ourfinisacked) { 1071 tp = tcp_close(tp); 1072 goto drop; 1073 } 1074 break; 1075 1076 /* 1077 * In TIME_WAIT state the only thing that should arrive 1078 * is a retransmission of the remote FIN. Acknowledge 1079 * it and restart the finack timer. 1080 */ 1081 case TCPS_TIME_WAIT: 1082 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1083 goto dropafterack; 1084 } 1085 } 1086 1087 step6: 1088 /* 1089 * Update window information. 1090 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1091 */ 1092 if ((tiflags & TH_ACK) && 1093 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 1094 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1095 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))) { 1096 /* keep track of pure window updates */ 1097 if (ti->ti_len == 0 && 1098 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1099 tcpstat.tcps_rcvwinupd++; 1100 tp->snd_wnd = tiwin; 1101 tp->snd_wl1 = ti->ti_seq; 1102 tp->snd_wl2 = ti->ti_ack; 1103 if (tp->snd_wnd > tp->max_sndwnd) 1104 tp->max_sndwnd = tp->snd_wnd; 1105 needoutput = 1; 1106 } 1107 1108 /* 1109 * Process segments with URG. 1110 */ 1111 if ((tiflags & TH_URG) && ti->ti_urp && 1112 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1113 /* 1114 * This is a kludge, but if we receive and accept 1115 * random urgent pointers, we'll crash in 1116 * soreceive. It's hard to imagine someone 1117 * actually wanting to send this much urgent data. 1118 */ 1119 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1120 ti->ti_urp = 0; /* XXX */ 1121 tiflags &= ~TH_URG; /* XXX */ 1122 goto dodata; /* XXX */ 1123 } 1124 /* 1125 * If this segment advances the known urgent pointer, 1126 * then mark the data stream. This should not happen 1127 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1128 * a FIN has been received from the remote side. 1129 * In these states we ignore the URG. 1130 * 1131 * According to RFC961 (Assigned Protocols), 1132 * the urgent pointer points to the last octet 1133 * of urgent data. We continue, however, 1134 * to consider it to indicate the first octet 1135 * of data past the urgent section as the original 1136 * spec states (in one of two places). 1137 */ 1138 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1139 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1140 so->so_oobmark = so->so_rcv.sb_cc + 1141 (tp->rcv_up - tp->rcv_nxt) - 1; 1142 if (so->so_oobmark == 0) 1143 so->so_state |= SS_RCVATMARK; 1144 sohasoutofband(so); 1145 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1146 } 1147 /* 1148 * Remove out of band data so doesn't get presented to user. 1149 * This can happen independent of advancing the URG pointer, 1150 * but if two URG's are pending at once, some out-of-band 1151 * data may creep in... ick. 1152 */ 1153 if (ti->ti_urp <= ti->ti_len 1154 #ifdef SO_OOBINLINE 1155 && (so->so_options & SO_OOBINLINE) == 0 1156 #endif 1157 ) 1158 tcp_pulloutofband(so, ti, m); 1159 } else 1160 /* 1161 * If no out of band data is expected, 1162 * pull receive urgent pointer along 1163 * with the receive window. 1164 */ 1165 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1166 tp->rcv_up = tp->rcv_nxt; 1167 dodata: /* XXX */ 1168 1169 /* 1170 * Process the segment text, merging it into the TCP sequencing queue, 1171 * and arranging for acknowledgment of receipt if necessary. 1172 * This process logically involves adjusting tp->rcv_wnd as data 1173 * is presented to the user (this happens in tcp_usrreq.c, 1174 * case PRU_RCVD). If a FIN has already been received on this 1175 * connection then we just ignore the text. 1176 */ 1177 if ((ti->ti_len || (tiflags&TH_FIN)) && 1178 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1179 TCP_REASS(tp, ti, m, so, tiflags); 1180 /* 1181 * Note the amount of data that peer has sent into 1182 * our window, in order to estimate the sender's 1183 * buffer size. 1184 */ 1185 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1186 } else { 1187 m_freem(m); 1188 tiflags &= ~TH_FIN; 1189 } 1190 1191 /* 1192 * If FIN is received ACK the FIN and let the user know 1193 * that the connection is closing. 1194 */ 1195 if (tiflags & TH_FIN) { 1196 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1197 socantrcvmore(so); 1198 tp->t_flags |= TF_ACKNOW; 1199 tp->rcv_nxt++; 1200 } 1201 switch (tp->t_state) { 1202 1203 /* 1204 * In SYN_RECEIVED and ESTABLISHED STATES 1205 * enter the CLOSE_WAIT state. 1206 */ 1207 case TCPS_SYN_RECEIVED: 1208 case TCPS_ESTABLISHED: 1209 tp->t_state = TCPS_CLOSE_WAIT; 1210 break; 1211 1212 /* 1213 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1214 * enter the CLOSING state. 1215 */ 1216 case TCPS_FIN_WAIT_1: 1217 tp->t_state = TCPS_CLOSING; 1218 break; 1219 1220 /* 1221 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1222 * starting the time-wait timer, turning off the other 1223 * standard timers. 1224 */ 1225 case TCPS_FIN_WAIT_2: 1226 tp->t_state = TCPS_TIME_WAIT; 1227 tcp_canceltimers(tp); 1228 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1229 soisdisconnected(so); 1230 break; 1231 1232 /* 1233 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1234 */ 1235 case TCPS_TIME_WAIT: 1236 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1237 break; 1238 } 1239 } 1240 if (so->so_options & SO_DEBUG) 1241 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1242 1243 /* 1244 * Return any desired output. 1245 */ 1246 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1247 (void) tcp_output(tp); 1248 return; 1249 1250 dropafterack: 1251 /* 1252 * Generate an ACK dropping incoming segment if it occupies 1253 * sequence space, where the ACK reflects our state. 1254 */ 1255 if (tiflags & TH_RST) 1256 goto drop; 1257 m_freem(m); 1258 tp->t_flags |= TF_ACKNOW; 1259 (void) tcp_output(tp); 1260 return; 1261 1262 dropwithreset: 1263 /* 1264 * Generate a RST, dropping incoming segment. 1265 * Make ACK acceptable to originator of segment. 1266 * Don't bother to respond if destination was broadcast/multicast. 1267 */ 1268 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1269 IN_MULTICAST(ti->ti_dst.s_addr)) 1270 goto drop; 1271 if (tiflags & TH_ACK) 1272 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1273 else { 1274 if (tiflags & TH_SYN) 1275 ti->ti_len++; 1276 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1277 TH_RST|TH_ACK); 1278 } 1279 /* destroy temporarily created socket */ 1280 if (dropsocket) 1281 (void) soabort(so); 1282 return; 1283 1284 drop: 1285 /* 1286 * Drop space held by incoming segment and return. 1287 */ 1288 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1289 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1290 m_freem(m); 1291 /* destroy temporarily created socket */ 1292 if (dropsocket) 1293 (void) soabort(so); 1294 return; 1295 #ifndef TUBA_INCLUDE 1296 } 1297 1298 void 1299 tcp_dooptions(tp, cp, cnt, ti, ts_present, ts_val, ts_ecr) 1300 struct tcpcb *tp; 1301 u_char *cp; 1302 int cnt; 1303 struct tcpiphdr *ti; 1304 int *ts_present; 1305 u_long *ts_val, *ts_ecr; 1306 { 1307 u_short mss; 1308 int opt, optlen; 1309 1310 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1311 opt = cp[0]; 1312 if (opt == TCPOPT_EOL) 1313 break; 1314 if (opt == TCPOPT_NOP) 1315 optlen = 1; 1316 else { 1317 optlen = cp[1]; 1318 if (optlen <= 0) 1319 break; 1320 } 1321 switch (opt) { 1322 1323 default: 1324 continue; 1325 1326 case TCPOPT_MAXSEG: 1327 if (optlen != TCPOLEN_MAXSEG) 1328 continue; 1329 if (!(ti->ti_flags & TH_SYN)) 1330 continue; 1331 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1332 NTOHS(mss); 1333 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1334 break; 1335 1336 case TCPOPT_WINDOW: 1337 if (optlen != TCPOLEN_WINDOW) 1338 continue; 1339 if (!(ti->ti_flags & TH_SYN)) 1340 continue; 1341 tp->t_flags |= TF_RCVD_SCALE; 1342 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1343 break; 1344 1345 case TCPOPT_TIMESTAMP: 1346 if (optlen != TCPOLEN_TIMESTAMP) 1347 continue; 1348 *ts_present = 1; 1349 bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val)); 1350 NTOHL(*ts_val); 1351 bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr)); 1352 NTOHL(*ts_ecr); 1353 1354 /* 1355 * A timestamp received in a SYN makes 1356 * it ok to send timestamp requests and replies. 1357 */ 1358 if (ti->ti_flags & TH_SYN) { 1359 tp->t_flags |= TF_RCVD_TSTMP; 1360 tp->ts_recent = *ts_val; 1361 tp->ts_recent_age = tcp_now; 1362 } 1363 break; 1364 } 1365 } 1366 } 1367 1368 /* 1369 * Pull out of band byte out of a segment so 1370 * it doesn't appear in the user's data queue. 1371 * It is still reflected in the segment length for 1372 * sequencing purposes. 1373 */ 1374 void 1375 tcp_pulloutofband(so, ti, m) 1376 struct socket *so; 1377 struct tcpiphdr *ti; 1378 register struct mbuf *m; 1379 { 1380 int cnt = ti->ti_urp - 1; 1381 1382 while (cnt >= 0) { 1383 if (m->m_len > cnt) { 1384 char *cp = mtod(m, caddr_t) + cnt; 1385 struct tcpcb *tp = sototcpcb(so); 1386 1387 tp->t_iobc = *cp; 1388 tp->t_oobflags |= TCPOOB_HAVEDATA; 1389 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1390 m->m_len--; 1391 return; 1392 } 1393 cnt -= m->m_len; 1394 m = m->m_next; 1395 if (m == 0) 1396 break; 1397 } 1398 panic("tcp_pulloutofband"); 1399 } 1400 1401 /* 1402 * Collect new round-trip time estimate 1403 * and update averages and current timeout. 1404 */ 1405 void 1406 tcp_xmit_timer(tp, rtt) 1407 register struct tcpcb *tp; 1408 short rtt; 1409 { 1410 register short delta; 1411 1412 tcpstat.tcps_rttupdated++; 1413 if (tp->t_srtt != 0) { 1414 /* 1415 * srtt is stored as fixed point with 3 bits after the 1416 * binary point (i.e., scaled by 8). The following magic 1417 * is equivalent to the smoothing algorithm in rfc793 with 1418 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1419 * point). Adjust rtt to origin 0. 1420 */ 1421 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1422 if ((tp->t_srtt += delta) <= 0) 1423 tp->t_srtt = 1; 1424 /* 1425 * We accumulate a smoothed rtt variance (actually, a 1426 * smoothed mean difference), then set the retransmit 1427 * timer to smoothed rtt + 4 times the smoothed variance. 1428 * rttvar is stored as fixed point with 2 bits after the 1429 * binary point (scaled by 4). The following is 1430 * equivalent to rfc793 smoothing with an alpha of .75 1431 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1432 * rfc793's wired-in beta. 1433 */ 1434 if (delta < 0) 1435 delta = -delta; 1436 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1437 if ((tp->t_rttvar += delta) <= 0) 1438 tp->t_rttvar = 1; 1439 } else { 1440 /* 1441 * No rtt measurement yet - use the unsmoothed rtt. 1442 * Set the variance to half the rtt (so our first 1443 * retransmit happens at 3*rtt). 1444 */ 1445 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1446 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1447 } 1448 tp->t_rtt = 0; 1449 tp->t_rxtshift = 0; 1450 1451 /* 1452 * the retransmit should happen at rtt + 4 * rttvar. 1453 * Because of the way we do the smoothing, srtt and rttvar 1454 * will each average +1/2 tick of bias. When we compute 1455 * the retransmit timer, we want 1/2 tick of rounding and 1456 * 1 extra tick because of +-1/2 tick uncertainty in the 1457 * firing of the timer. The bias will give us exactly the 1458 * 1.5 tick we need. But, because the bias is 1459 * statistical, we have to test that we don't drop below 1460 * the minimum feasible timer (which is 2 ticks). 1461 */ 1462 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1463 tp->t_rttmin, TCPTV_REXMTMAX); 1464 1465 /* 1466 * We received an ack for a packet that wasn't retransmitted; 1467 * it is probably safe to discard any error indications we've 1468 * received recently. This isn't quite right, but close enough 1469 * for now (a route might have failed after we sent a segment, 1470 * and the return path might not be symmetrical). 1471 */ 1472 tp->t_softerror = 0; 1473 } 1474 1475 /* 1476 * Determine a reasonable value for maxseg size. 1477 * If the route is known, check route for mtu. 1478 * If none, use an mss that can be handled on the outgoing 1479 * interface without forcing IP to fragment; if bigger than 1480 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1481 * to utilize large mbufs. If no route is found, route has no mtu, 1482 * or the destination isn't local, use a default, hopefully conservative 1483 * size (usually 512 or the default IP max size, but no more than the mtu 1484 * of the interface), as we can't discover anything about intervening 1485 * gateways or networks. We also initialize the congestion/slow start 1486 * window to be a single segment if the destination isn't local. 1487 * While looking at the routing entry, we also initialize other path-dependent 1488 * parameters from pre-set or cached values in the routing entry. 1489 */ 1490 int 1491 tcp_mss(tp, offer) 1492 register struct tcpcb *tp; 1493 u_int offer; 1494 { 1495 struct route *ro; 1496 register struct rtentry *rt; 1497 struct ifnet *ifp; 1498 register int rtt, mss; 1499 u_long bufsize; 1500 struct inpcb *inp; 1501 struct socket *so; 1502 extern int tcp_mssdflt; 1503 1504 inp = tp->t_inpcb; 1505 ro = &inp->inp_route; 1506 1507 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1508 /* No route yet, so try to acquire one */ 1509 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1510 ro->ro_dst.sa_family = AF_INET; 1511 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1512 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1513 inp->inp_faddr; 1514 rtalloc(ro); 1515 } 1516 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1517 return (tcp_mssdflt); 1518 } 1519 ifp = rt->rt_ifp; 1520 so = inp->inp_socket; 1521 1522 #ifdef RTV_MTU /* if route characteristics exist ... */ 1523 /* 1524 * While we're here, check if there's an initial rtt 1525 * or rttvar. Convert from the route-table units 1526 * to scaled multiples of the slow timeout timer. 1527 */ 1528 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1529 /* 1530 * XXX the lock bit for MTU indicates that the value 1531 * is also a minimum value; this is subject to time. 1532 */ 1533 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1534 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1535 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1536 if (rt->rt_rmx.rmx_rttvar) 1537 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1538 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1539 else 1540 /* default variation is +- 1 rtt */ 1541 tp->t_rttvar = 1542 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1543 TCPT_RANGESET(tp->t_rxtcur, 1544 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1545 tp->t_rttmin, TCPTV_REXMTMAX); 1546 } 1547 /* 1548 * if there's an mtu associated with the route, use it 1549 */ 1550 if (rt->rt_rmx.rmx_mtu) 1551 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1552 else 1553 #endif /* RTV_MTU */ 1554 { 1555 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1556 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1557 if (mss > MCLBYTES) 1558 mss &= ~(MCLBYTES-1); 1559 #else 1560 if (mss > MCLBYTES) 1561 mss = mss / MCLBYTES * MCLBYTES; 1562 #endif 1563 if (!in_localaddr(inp->inp_faddr)) 1564 mss = min(mss, tcp_mssdflt); 1565 } 1566 /* 1567 * The current mss, t_maxseg, is initialized to the default value. 1568 * If we compute a smaller value, reduce the current mss. 1569 * If we compute a larger value, return it for use in sending 1570 * a max seg size option, but don't store it for use 1571 * unless we received an offer at least that large from peer. 1572 * However, do not accept offers under 32 bytes. 1573 */ 1574 if (offer) 1575 mss = min(mss, offer); 1576 mss = max(mss, 32); /* sanity */ 1577 if (mss < tp->t_maxseg || offer != 0) { 1578 /* 1579 * If there's a pipesize, change the socket buffer 1580 * to that size. Make the socket buffers an integral 1581 * number of mss units; if the mss is larger than 1582 * the socket buffer, decrease the mss. 1583 */ 1584 #ifdef RTV_SPIPE 1585 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1586 #endif 1587 bufsize = so->so_snd.sb_hiwat; 1588 if (bufsize < mss) 1589 mss = bufsize; 1590 else { 1591 bufsize = roundup(bufsize, mss); 1592 if (bufsize > sb_max) 1593 bufsize = sb_max; 1594 (void)sbreserve(&so->so_snd, bufsize); 1595 } 1596 tp->t_maxseg = mss; 1597 1598 #ifdef RTV_RPIPE 1599 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1600 #endif 1601 bufsize = so->so_rcv.sb_hiwat; 1602 if (bufsize > mss) { 1603 bufsize = roundup(bufsize, mss); 1604 if (bufsize > sb_max) 1605 bufsize = sb_max; 1606 (void)sbreserve(&so->so_rcv, bufsize); 1607 } 1608 } 1609 tp->snd_cwnd = mss; 1610 1611 #ifdef RTV_SSTHRESH 1612 if (rt->rt_rmx.rmx_ssthresh) { 1613 /* 1614 * There's some sort of gateway or interface 1615 * buffer limit on the path. Use this to set 1616 * the slow start threshhold, but set the 1617 * threshold to no less than 2*mss. 1618 */ 1619 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1620 } 1621 #endif /* RTV_MTU */ 1622 return (mss); 1623 } 1624 #endif /* TUBA_INCLUDE */ 1625