1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990 Regents of the University of California. 3 * All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)tcp_input.c 7.33 (Berkeley) 02/12/93 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 struct tcpcb *tcp_newtcpcb(); 44 45 extern u_long sb_max; 46 47 #endif /* TUBA_INCLUDE */ 48 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 49 50 /* for modulo comparisons of timestamps */ 51 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 52 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 53 54 55 /* 56 * Insert segment ti into reassembly queue of tcp with 57 * control block tp. Return TH_FIN if reassembly now includes 58 * a segment with FIN. The macro form does the common case inline 59 * (segment is the next to be received on an established connection, 60 * and the queue is empty), avoiding linkage into and removal 61 * from the queue and repetition of various conversions. 62 * Set DELACK for segments received in order, but ack immediately 63 * when segments are out of order (so fast retransmit can work). 64 */ 65 #define TCP_REASS(tp, ti, m, so, flags) { \ 66 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 67 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 68 (tp)->t_state == TCPS_ESTABLISHED) { \ 69 tp->t_flags |= TF_DELACK; \ 70 (tp)->rcv_nxt += (ti)->ti_len; \ 71 flags = (ti)->ti_flags & TH_FIN; \ 72 tcpstat.tcps_rcvpack++;\ 73 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 74 sbappend(&(so)->so_rcv, (m)); \ 75 sorwakeup(so); \ 76 } else { \ 77 (flags) = tcp_reass((tp), (ti), (m)); \ 78 tp->t_flags |= TF_ACKNOW; \ 79 } \ 80 } 81 #ifndef TUBA_INCLUDE 82 83 tcp_reass(tp, ti, m) 84 register struct tcpcb *tp; 85 register struct tcpiphdr *ti; 86 struct mbuf *m; 87 { 88 register struct tcpiphdr *q; 89 struct socket *so = tp->t_inpcb->inp_socket; 90 int flags; 91 92 /* 93 * Call with ti==0 after become established to 94 * force pre-ESTABLISHED data up to user socket. 95 */ 96 if (ti == 0) 97 goto present; 98 99 /* 100 * Find a segment which begins after this one does. 101 */ 102 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 103 q = (struct tcpiphdr *)q->ti_next) 104 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 105 break; 106 107 /* 108 * If there is a preceding segment, it may provide some of 109 * our data already. If so, drop the data from the incoming 110 * segment. If it provides all of our data, drop us. 111 */ 112 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 113 register int i; 114 q = (struct tcpiphdr *)q->ti_prev; 115 /* conversion to int (in i) handles seq wraparound */ 116 i = q->ti_seq + q->ti_len - ti->ti_seq; 117 if (i > 0) { 118 if (i >= ti->ti_len) { 119 tcpstat.tcps_rcvduppack++; 120 tcpstat.tcps_rcvdupbyte += ti->ti_len; 121 m_freem(m); 122 return (0); 123 } 124 m_adj(m, i); 125 ti->ti_len -= i; 126 ti->ti_seq += i; 127 } 128 q = (struct tcpiphdr *)(q->ti_next); 129 } 130 tcpstat.tcps_rcvoopack++; 131 tcpstat.tcps_rcvoobyte += ti->ti_len; 132 REASS_MBUF(ti) = m; /* XXX */ 133 134 /* 135 * While we overlap succeeding segments trim them or, 136 * if they are completely covered, dequeue them. 137 */ 138 while (q != (struct tcpiphdr *)tp) { 139 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 140 if (i <= 0) 141 break; 142 if (i < q->ti_len) { 143 q->ti_seq += i; 144 q->ti_len -= i; 145 m_adj(REASS_MBUF(q), i); 146 break; 147 } 148 q = (struct tcpiphdr *)q->ti_next; 149 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 150 remque(q->ti_prev); 151 m_freem(m); 152 } 153 154 /* 155 * Stick new segment in its place. 156 */ 157 insque(ti, q->ti_prev); 158 159 present: 160 /* 161 * Present data to user, advancing rcv_nxt through 162 * completed sequence space. 163 */ 164 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 165 return (0); 166 ti = tp->seg_next; 167 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 168 return (0); 169 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 170 return (0); 171 do { 172 tp->rcv_nxt += ti->ti_len; 173 flags = ti->ti_flags & TH_FIN; 174 remque(ti); 175 m = REASS_MBUF(ti); 176 ti = (struct tcpiphdr *)ti->ti_next; 177 if (so->so_state & SS_CANTRCVMORE) 178 m_freem(m); 179 else 180 sbappend(&so->so_rcv, m); 181 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 182 sorwakeup(so); 183 return (flags); 184 } 185 186 /* 187 * TCP input routine, follows pages 65-76 of the 188 * protocol specification dated September, 1981 very closely. 189 */ 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 /* RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN */ 522 if (m->m_flags & (M_BCAST|M_MCAST) || 523 in_broadcast(ti->ti_dst) || IN_MULTICAST(ti->ti_dst.s_addr)) 524 goto drop; 525 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 526 if (am == NULL) 527 goto drop; 528 am->m_len = sizeof (struct sockaddr_in); 529 sin = mtod(am, struct sockaddr_in *); 530 sin->sin_family = AF_INET; 531 sin->sin_len = sizeof(*sin); 532 sin->sin_addr = ti->ti_src; 533 sin->sin_port = ti->ti_sport; 534 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 535 laddr = inp->inp_laddr; 536 if (inp->inp_laddr.s_addr == INADDR_ANY) 537 inp->inp_laddr = ti->ti_dst; 538 if (in_pcbconnect(inp, am)) { 539 inp->inp_laddr = laddr; 540 (void) m_free(am); 541 goto drop; 542 } 543 (void) m_free(am); 544 tp->t_template = tcp_template(tp); 545 if (tp->t_template == 0) { 546 tp = tcp_drop(tp, ENOBUFS); 547 dropsocket = 0; /* socket is already gone */ 548 goto drop; 549 } 550 if (optp) 551 tcp_dooptions(tp, optp, optlen, ti, 552 &ts_present, &ts_val, &ts_ecr); 553 if (iss) 554 tp->iss = iss; 555 else 556 tp->iss = tcp_iss; 557 tcp_iss += TCP_ISSINCR/2; 558 tp->irs = ti->ti_seq; 559 tcp_sendseqinit(tp); 560 tcp_rcvseqinit(tp); 561 tp->t_flags |= TF_ACKNOW; 562 tp->t_state = TCPS_SYN_RECEIVED; 563 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 564 dropsocket = 0; /* committed to socket */ 565 tcpstat.tcps_accepts++; 566 goto trimthenstep6; 567 } 568 569 /* 570 * If the state is SYN_SENT: 571 * if seg contains an ACK, but not for our SYN, drop the input. 572 * if seg contains a RST, then drop the connection. 573 * if seg does not contain SYN, then drop it. 574 * Otherwise this is an acceptable SYN segment 575 * initialize tp->rcv_nxt and tp->irs 576 * if seg contains ack then advance tp->snd_una 577 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 578 * arrange for segment to be acked (eventually) 579 * continue processing rest of data/controls, beginning with URG 580 */ 581 case TCPS_SYN_SENT: 582 if ((tiflags & TH_ACK) && 583 (SEQ_LEQ(ti->ti_ack, tp->iss) || 584 SEQ_GT(ti->ti_ack, tp->snd_max))) 585 goto dropwithreset; 586 if (tiflags & TH_RST) { 587 if (tiflags & TH_ACK) 588 tp = tcp_drop(tp, ECONNREFUSED); 589 goto drop; 590 } 591 if ((tiflags & TH_SYN) == 0) 592 goto drop; 593 if (tiflags & TH_ACK) { 594 tp->snd_una = ti->ti_ack; 595 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 596 tp->snd_nxt = tp->snd_una; 597 } 598 tp->t_timer[TCPT_REXMT] = 0; 599 tp->irs = ti->ti_seq; 600 tcp_rcvseqinit(tp); 601 tp->t_flags |= TF_ACKNOW; 602 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 603 tcpstat.tcps_connects++; 604 soisconnected(so); 605 tp->t_state = TCPS_ESTABLISHED; 606 /* Do window scaling on this connection? */ 607 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 608 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 609 tp->snd_scale = tp->requested_s_scale; 610 tp->rcv_scale = tp->request_r_scale; 611 } 612 (void) tcp_reass(tp, (struct tcpiphdr *)0, 613 (struct mbuf *)0); 614 /* 615 * if we didn't have to retransmit the SYN, 616 * use its rtt as our initial srtt & rtt var. 617 */ 618 if (tp->t_rtt) 619 tcp_xmit_timer(tp, tp->t_rtt); 620 } else 621 tp->t_state = TCPS_SYN_RECEIVED; 622 623 trimthenstep6: 624 /* 625 * Advance ti->ti_seq to correspond to first data byte. 626 * If data, trim to stay within window, 627 * dropping FIN if necessary. 628 */ 629 ti->ti_seq++; 630 if (ti->ti_len > tp->rcv_wnd) { 631 todrop = ti->ti_len - tp->rcv_wnd; 632 m_adj(m, -todrop); 633 ti->ti_len = tp->rcv_wnd; 634 tiflags &= ~TH_FIN; 635 tcpstat.tcps_rcvpackafterwin++; 636 tcpstat.tcps_rcvbyteafterwin += todrop; 637 } 638 tp->snd_wl1 = ti->ti_seq - 1; 639 tp->rcv_up = ti->ti_seq; 640 goto step6; 641 } 642 643 /* 644 * States other than LISTEN or SYN_SENT. 645 * First check timestamp, if present. 646 * Then check that at least some bytes of segment are within 647 * receive window. If segment begins before rcv_nxt, 648 * drop leading data (and SYN); if nothing left, just ack. 649 * 650 * RFC 1323 PAWS: If we have a timestamp reply on this segment 651 * and it's less than ts_recent, drop it. 652 */ 653 if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 654 TSTMP_LT(ts_val, tp->ts_recent)) { 655 656 /* Check to see if ts_recent is over 24 days old. */ 657 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 658 /* 659 * Invalidate ts_recent. If this segment updates 660 * ts_recent, the age will be reset later and ts_recent 661 * will get a valid value. If it does not, setting 662 * ts_recent to zero will at least satisfy the 663 * requirement that zero be placed in the timestamp 664 * echo reply when ts_recent isn't valid. The 665 * age isn't reset until we get a valid ts_recent 666 * because we don't want out-of-order segments to be 667 * dropped when ts_recent is old. 668 */ 669 tp->ts_recent = 0; 670 } else { 671 tcpstat.tcps_rcvduppack++; 672 tcpstat.tcps_rcvdupbyte += ti->ti_len; 673 tcpstat.tcps_pawsdrop++; 674 goto dropafterack; 675 } 676 } 677 678 todrop = tp->rcv_nxt - ti->ti_seq; 679 if (todrop > 0) { 680 if (tiflags & TH_SYN) { 681 tiflags &= ~TH_SYN; 682 ti->ti_seq++; 683 if (ti->ti_urp > 1) 684 ti->ti_urp--; 685 else 686 tiflags &= ~TH_URG; 687 todrop--; 688 } 689 if (todrop > ti->ti_len || 690 todrop == ti->ti_len && (tiflags&TH_FIN) == 0) { 691 tcpstat.tcps_rcvduppack++; 692 tcpstat.tcps_rcvdupbyte += ti->ti_len; 693 /* 694 * If segment is just one to the left of the window, 695 * check two special cases: 696 * 1. Don't toss RST in response to 4.2-style keepalive. 697 * 2. If the only thing to drop is a FIN, we can drop 698 * it, but check the ACK or we will get into FIN 699 * wars if our FINs crossed (both CLOSING). 700 * In either case, send ACK to resynchronize, 701 * but keep on processing for RST or ACK. 702 */ 703 if ((tiflags & TH_FIN && todrop == ti->ti_len + 1) 704 #ifdef TCP_COMPAT_42 705 || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1) 706 #endif 707 ) { 708 todrop = ti->ti_len; 709 tiflags &= ~TH_FIN; 710 tp->t_flags |= TF_ACKNOW; 711 } else { 712 /* 713 * Handle the case when a bound socket connects 714 * to itself. Allow packets with a SYN and 715 * an ACK to continue with the processing. 716 */ 717 if (todrop != 0 || (tiflags & TH_ACK) == 0) 718 goto dropafterack; 719 } 720 } else { 721 tcpstat.tcps_rcvpartduppack++; 722 tcpstat.tcps_rcvpartdupbyte += todrop; 723 } 724 m_adj(m, todrop); 725 ti->ti_seq += todrop; 726 ti->ti_len -= todrop; 727 if (ti->ti_urp > todrop) 728 ti->ti_urp -= todrop; 729 else { 730 tiflags &= ~TH_URG; 731 ti->ti_urp = 0; 732 } 733 } 734 735 /* 736 * If new data are received on a connection after the 737 * user processes are gone, then RST the other end. 738 */ 739 if ((so->so_state & SS_NOFDREF) && 740 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 741 tp = tcp_close(tp); 742 tcpstat.tcps_rcvafterclose++; 743 goto dropwithreset; 744 } 745 746 /* 747 * If segment ends after window, drop trailing data 748 * (and PUSH and FIN); if nothing left, just ACK. 749 */ 750 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 751 if (todrop > 0) { 752 tcpstat.tcps_rcvpackafterwin++; 753 if (todrop >= ti->ti_len) { 754 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 755 /* 756 * If a new connection request is received 757 * while in TIME_WAIT, drop the old connection 758 * and start over if the sequence numbers 759 * are above the previous ones. 760 */ 761 if (tiflags & TH_SYN && 762 tp->t_state == TCPS_TIME_WAIT && 763 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 764 iss = tp->rcv_nxt + TCP_ISSINCR; 765 tp = tcp_close(tp); 766 goto findpcb; 767 } 768 /* 769 * If window is closed can only take segments at 770 * window edge, and have to drop data and PUSH from 771 * incoming segments. Continue processing, but 772 * remember to ack. Otherwise, drop segment 773 * and ack. 774 */ 775 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 776 tp->t_flags |= TF_ACKNOW; 777 tcpstat.tcps_rcvwinprobe++; 778 } else 779 goto dropafterack; 780 } else 781 tcpstat.tcps_rcvbyteafterwin += todrop; 782 m_adj(m, -todrop); 783 ti->ti_len -= todrop; 784 tiflags &= ~(TH_PUSH|TH_FIN); 785 } 786 787 /* 788 * If last ACK falls within this segment's sequence numbers, 789 * record its timestamp. 790 */ 791 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 792 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + 793 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 794 tp->ts_recent_age = tcp_now; 795 tp->ts_recent = ts_val; 796 } 797 798 /* 799 * If the RST bit is set examine the state: 800 * SYN_RECEIVED STATE: 801 * If passive open, return to LISTEN state. 802 * If active open, inform user that connection was refused. 803 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 804 * Inform user that connection was reset, and close tcb. 805 * CLOSING, LAST_ACK, TIME_WAIT STATES 806 * Close the tcb. 807 */ 808 if (tiflags&TH_RST) switch (tp->t_state) { 809 810 case TCPS_SYN_RECEIVED: 811 so->so_error = ECONNREFUSED; 812 goto close; 813 814 case TCPS_ESTABLISHED: 815 case TCPS_FIN_WAIT_1: 816 case TCPS_FIN_WAIT_2: 817 case TCPS_CLOSE_WAIT: 818 so->so_error = ECONNRESET; 819 close: 820 tp->t_state = TCPS_CLOSED; 821 tcpstat.tcps_drops++; 822 tp = tcp_close(tp); 823 goto drop; 824 825 case TCPS_CLOSING: 826 case TCPS_LAST_ACK: 827 case TCPS_TIME_WAIT: 828 tp = tcp_close(tp); 829 goto drop; 830 } 831 832 /* 833 * If a SYN is in the window, then this is an 834 * error and we send an RST and drop the connection. 835 */ 836 if (tiflags & TH_SYN) { 837 tp = tcp_drop(tp, ECONNRESET); 838 goto dropwithreset; 839 } 840 841 /* 842 * If the ACK bit is off we drop the segment and return. 843 */ 844 if ((tiflags & TH_ACK) == 0) 845 goto drop; 846 847 /* 848 * Ack processing. 849 */ 850 switch (tp->t_state) { 851 852 /* 853 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 854 * ESTABLISHED state and continue processing, otherwise 855 * send an RST. 856 */ 857 case TCPS_SYN_RECEIVED: 858 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 859 SEQ_GT(ti->ti_ack, tp->snd_max)) 860 goto dropwithreset; 861 tcpstat.tcps_connects++; 862 soisconnected(so); 863 tp->t_state = TCPS_ESTABLISHED; 864 /* Do window scaling? */ 865 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 866 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 867 tp->snd_scale = tp->requested_s_scale; 868 tp->rcv_scale = tp->request_r_scale; 869 } 870 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 871 tp->snd_wl1 = ti->ti_seq - 1; 872 /* fall into ... */ 873 874 /* 875 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 876 * ACKs. If the ack is in the range 877 * tp->snd_una < ti->ti_ack <= tp->snd_max 878 * then advance tp->snd_una to ti->ti_ack and drop 879 * data from the retransmission queue. If this ACK reflects 880 * more up to date window information we update our window information. 881 */ 882 case TCPS_ESTABLISHED: 883 case TCPS_FIN_WAIT_1: 884 case TCPS_FIN_WAIT_2: 885 case TCPS_CLOSE_WAIT: 886 case TCPS_CLOSING: 887 case TCPS_LAST_ACK: 888 case TCPS_TIME_WAIT: 889 890 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 891 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 892 tcpstat.tcps_rcvdupack++; 893 /* 894 * If we have outstanding data (other than 895 * a window probe), this is a completely 896 * duplicate ack (ie, window info didn't 897 * change), the ack is the biggest we've 898 * seen and we've seen exactly our rexmt 899 * threshhold of them, assume a packet 900 * has been dropped and retransmit it. 901 * Kludge snd_nxt & the congestion 902 * window so we send only this one 903 * packet. 904 * 905 * We know we're losing at the current 906 * window size so do congestion avoidance 907 * (set ssthresh to half the current window 908 * and pull our congestion window back to 909 * the new ssthresh). 910 * 911 * Dup acks mean that packets have left the 912 * network (they're now cached at the receiver) 913 * so bump cwnd by the amount in the receiver 914 * to keep a constant cwnd packets in the 915 * network. 916 */ 917 if (tp->t_timer[TCPT_REXMT] == 0 || 918 ti->ti_ack != tp->snd_una) 919 tp->t_dupacks = 0; 920 else if (++tp->t_dupacks == tcprexmtthresh) { 921 tcp_seq onxt = tp->snd_nxt; 922 u_int win = 923 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 924 tp->t_maxseg; 925 926 if (win < 2) 927 win = 2; 928 tp->snd_ssthresh = win * tp->t_maxseg; 929 tp->t_timer[TCPT_REXMT] = 0; 930 tp->t_rtt = 0; 931 tp->snd_nxt = ti->ti_ack; 932 tp->snd_cwnd = tp->t_maxseg; 933 (void) tcp_output(tp); 934 tp->snd_cwnd = tp->snd_ssthresh + 935 tp->t_maxseg * tp->t_dupacks; 936 if (SEQ_GT(onxt, tp->snd_nxt)) 937 tp->snd_nxt = onxt; 938 goto drop; 939 } else if (tp->t_dupacks > tcprexmtthresh) { 940 tp->snd_cwnd += tp->t_maxseg; 941 (void) tcp_output(tp); 942 goto drop; 943 } 944 } else 945 tp->t_dupacks = 0; 946 break; 947 } 948 /* 949 * If the congestion window was inflated to account 950 * for the other side's cached packets, retract it. 951 */ 952 if (tp->t_dupacks > tcprexmtthresh && 953 tp->snd_cwnd > tp->snd_ssthresh) 954 tp->snd_cwnd = tp->snd_ssthresh; 955 tp->t_dupacks = 0; 956 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 957 tcpstat.tcps_rcvacktoomuch++; 958 goto dropafterack; 959 } 960 acked = ti->ti_ack - tp->snd_una; 961 tcpstat.tcps_rcvackpack++; 962 tcpstat.tcps_rcvackbyte += acked; 963 964 /* 965 * If we have a timestamp reply, update smoothed 966 * round trip time. If no timestamp is present but 967 * transmit timer is running and timed sequence 968 * number was acked, update smoothed round trip time. 969 * Since we now have an rtt measurement, cancel the 970 * timer backoff (cf., Phil Karn's retransmit alg.). 971 * Recompute the initial retransmit timer. 972 */ 973 if (ts_present) 974 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 975 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 976 tcp_xmit_timer(tp,tp->t_rtt); 977 978 /* 979 * If all outstanding data is acked, stop retransmit 980 * timer and remember to restart (more output or persist). 981 * If there is more data to be acked, restart retransmit 982 * timer, using current (possibly backed-off) value. 983 */ 984 if (ti->ti_ack == tp->snd_max) { 985 tp->t_timer[TCPT_REXMT] = 0; 986 needoutput = 1; 987 } else if (tp->t_timer[TCPT_PERSIST] == 0) 988 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 989 /* 990 * When new data is acked, open the congestion window. 991 * If the window gives us less than ssthresh packets 992 * in flight, open exponentially (maxseg per packet). 993 * Otherwise open linearly: maxseg per window 994 * (maxseg^2 / cwnd per packet), plus a constant 995 * fraction of a packet (maxseg/8) to help larger windows 996 * open quickly enough. 997 */ 998 { 999 register u_int cw = tp->snd_cwnd; 1000 register u_int incr = tp->t_maxseg; 1001 1002 if (cw > tp->snd_ssthresh) 1003 incr = incr * incr / cw + incr / 8; 1004 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1005 } 1006 if (acked > so->so_snd.sb_cc) { 1007 tp->snd_wnd -= so->so_snd.sb_cc; 1008 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1009 ourfinisacked = 1; 1010 } else { 1011 sbdrop(&so->so_snd, acked); 1012 tp->snd_wnd -= acked; 1013 ourfinisacked = 0; 1014 } 1015 if (so->so_snd.sb_flags & SB_NOTIFY) 1016 sowwakeup(so); 1017 tp->snd_una = ti->ti_ack; 1018 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1019 tp->snd_nxt = tp->snd_una; 1020 1021 switch (tp->t_state) { 1022 1023 /* 1024 * In FIN_WAIT_1 STATE in addition to the processing 1025 * for the ESTABLISHED state if our FIN is now acknowledged 1026 * then enter FIN_WAIT_2. 1027 */ 1028 case TCPS_FIN_WAIT_1: 1029 if (ourfinisacked) { 1030 /* 1031 * If we can't receive any more 1032 * data, then closing user can proceed. 1033 * Starting the timer is contrary to the 1034 * specification, but if we don't get a FIN 1035 * we'll hang forever. 1036 */ 1037 if (so->so_state & SS_CANTRCVMORE) { 1038 soisdisconnected(so); 1039 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1040 } 1041 tp->t_state = TCPS_FIN_WAIT_2; 1042 } 1043 break; 1044 1045 /* 1046 * In CLOSING STATE in addition to the processing for 1047 * the ESTABLISHED state if the ACK acknowledges our FIN 1048 * then enter the TIME-WAIT state, otherwise ignore 1049 * the segment. 1050 */ 1051 case TCPS_CLOSING: 1052 if (ourfinisacked) { 1053 tp->t_state = TCPS_TIME_WAIT; 1054 tcp_canceltimers(tp); 1055 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1056 soisdisconnected(so); 1057 } 1058 break; 1059 1060 /* 1061 * In LAST_ACK, we may still be waiting for data to drain 1062 * and/or to be acked, as well as for the ack of our FIN. 1063 * If our FIN is now acknowledged, delete the TCB, 1064 * enter the closed state and return. 1065 */ 1066 case TCPS_LAST_ACK: 1067 if (ourfinisacked) { 1068 tp = tcp_close(tp); 1069 goto drop; 1070 } 1071 break; 1072 1073 /* 1074 * In TIME_WAIT state the only thing that should arrive 1075 * is a retransmission of the remote FIN. Acknowledge 1076 * it and restart the finack timer. 1077 */ 1078 case TCPS_TIME_WAIT: 1079 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1080 goto dropafterack; 1081 } 1082 } 1083 1084 step6: 1085 /* 1086 * Update window information. 1087 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1088 */ 1089 if ((tiflags & TH_ACK) && 1090 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 1091 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1092 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))) { 1093 /* keep track of pure window updates */ 1094 if (ti->ti_len == 0 && 1095 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1096 tcpstat.tcps_rcvwinupd++; 1097 tp->snd_wnd = tiwin; 1098 tp->snd_wl1 = ti->ti_seq; 1099 tp->snd_wl2 = ti->ti_ack; 1100 if (tp->snd_wnd > tp->max_sndwnd) 1101 tp->max_sndwnd = tp->snd_wnd; 1102 needoutput = 1; 1103 } 1104 1105 /* 1106 * Process segments with URG. 1107 */ 1108 if ((tiflags & TH_URG) && ti->ti_urp && 1109 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1110 /* 1111 * This is a kludge, but if we receive and accept 1112 * random urgent pointers, we'll crash in 1113 * soreceive. It's hard to imagine someone 1114 * actually wanting to send this much urgent data. 1115 */ 1116 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1117 ti->ti_urp = 0; /* XXX */ 1118 tiflags &= ~TH_URG; /* XXX */ 1119 goto dodata; /* XXX */ 1120 } 1121 /* 1122 * If this segment advances the known urgent pointer, 1123 * then mark the data stream. This should not happen 1124 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1125 * a FIN has been received from the remote side. 1126 * In these states we ignore the URG. 1127 * 1128 * According to RFC961 (Assigned Protocols), 1129 * the urgent pointer points to the last octet 1130 * of urgent data. We continue, however, 1131 * to consider it to indicate the first octet 1132 * of data past the urgent section as the original 1133 * spec states (in one of two places). 1134 */ 1135 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1136 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1137 so->so_oobmark = so->so_rcv.sb_cc + 1138 (tp->rcv_up - tp->rcv_nxt) - 1; 1139 if (so->so_oobmark == 0) 1140 so->so_state |= SS_RCVATMARK; 1141 sohasoutofband(so); 1142 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1143 } 1144 /* 1145 * Remove out of band data so doesn't get presented to user. 1146 * This can happen independent of advancing the URG pointer, 1147 * but if two URG's are pending at once, some out-of-band 1148 * data may creep in... ick. 1149 */ 1150 if (ti->ti_urp <= ti->ti_len 1151 #ifdef SO_OOBINLINE 1152 && (so->so_options & SO_OOBINLINE) == 0 1153 #endif 1154 ) 1155 tcp_pulloutofband(so, ti, m); 1156 } else 1157 /* 1158 * If no out of band data is expected, 1159 * pull receive urgent pointer along 1160 * with the receive window. 1161 */ 1162 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1163 tp->rcv_up = tp->rcv_nxt; 1164 dodata: /* XXX */ 1165 1166 /* 1167 * Process the segment text, merging it into the TCP sequencing queue, 1168 * and arranging for acknowledgment of receipt if necessary. 1169 * This process logically involves adjusting tp->rcv_wnd as data 1170 * is presented to the user (this happens in tcp_usrreq.c, 1171 * case PRU_RCVD). If a FIN has already been received on this 1172 * connection then we just ignore the text. 1173 */ 1174 if ((ti->ti_len || (tiflags&TH_FIN)) && 1175 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1176 TCP_REASS(tp, ti, m, so, tiflags); 1177 /* 1178 * Note the amount of data that peer has sent into 1179 * our window, in order to estimate the sender's 1180 * buffer size. 1181 */ 1182 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1183 } else { 1184 m_freem(m); 1185 tiflags &= ~TH_FIN; 1186 } 1187 1188 /* 1189 * If FIN is received ACK the FIN and let the user know 1190 * that the connection is closing. 1191 */ 1192 if (tiflags & TH_FIN) { 1193 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1194 socantrcvmore(so); 1195 tp->t_flags |= TF_ACKNOW; 1196 tp->rcv_nxt++; 1197 } 1198 switch (tp->t_state) { 1199 1200 /* 1201 * In SYN_RECEIVED and ESTABLISHED STATES 1202 * enter the CLOSE_WAIT state. 1203 */ 1204 case TCPS_SYN_RECEIVED: 1205 case TCPS_ESTABLISHED: 1206 tp->t_state = TCPS_CLOSE_WAIT; 1207 break; 1208 1209 /* 1210 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1211 * enter the CLOSING state. 1212 */ 1213 case TCPS_FIN_WAIT_1: 1214 tp->t_state = TCPS_CLOSING; 1215 break; 1216 1217 /* 1218 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1219 * starting the time-wait timer, turning off the other 1220 * standard timers. 1221 */ 1222 case TCPS_FIN_WAIT_2: 1223 tp->t_state = TCPS_TIME_WAIT; 1224 tcp_canceltimers(tp); 1225 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1226 soisdisconnected(so); 1227 break; 1228 1229 /* 1230 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1231 */ 1232 case TCPS_TIME_WAIT: 1233 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1234 break; 1235 } 1236 } 1237 if (so->so_options & SO_DEBUG) 1238 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1239 1240 /* 1241 * Return any desired output. 1242 */ 1243 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1244 (void) tcp_output(tp); 1245 return; 1246 1247 dropafterack: 1248 /* 1249 * Generate an ACK dropping incoming segment if it occupies 1250 * sequence space, where the ACK reflects our state. 1251 */ 1252 if (tiflags & TH_RST) 1253 goto drop; 1254 m_freem(m); 1255 tp->t_flags |= TF_ACKNOW; 1256 (void) tcp_output(tp); 1257 return; 1258 1259 dropwithreset: 1260 /* 1261 * Generate a RST, dropping incoming segment. 1262 * Make ACK acceptable to originator of segment. 1263 * Don't bother to respond if destination was broadcast/multicast. 1264 */ 1265 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1266 in_broadcast(ti->ti_dst) || IN_MULTICAST(ti->ti_dst.s_addr)) 1267 goto drop; 1268 if (tiflags & TH_ACK) 1269 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1270 else { 1271 if (tiflags & TH_SYN) 1272 ti->ti_len++; 1273 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1274 TH_RST|TH_ACK); 1275 } 1276 /* destroy temporarily created socket */ 1277 if (dropsocket) 1278 (void) soabort(so); 1279 return; 1280 1281 drop: 1282 /* 1283 * Drop space held by incoming segment and return. 1284 */ 1285 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1286 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1287 m_freem(m); 1288 /* destroy temporarily created socket */ 1289 if (dropsocket) 1290 (void) soabort(so); 1291 return; 1292 #ifndef TUBA_INCLUDE 1293 } 1294 1295 tcp_dooptions(tp, cp, cnt, ti, ts_present, ts_val, ts_ecr) 1296 struct tcpcb *tp; 1297 u_char *cp; 1298 int cnt; 1299 struct tcpiphdr *ti; 1300 int *ts_present; 1301 u_long *ts_val, *ts_ecr; 1302 { 1303 u_short mss; 1304 int opt, optlen; 1305 1306 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1307 opt = cp[0]; 1308 if (opt == TCPOPT_EOL) 1309 break; 1310 if (opt == TCPOPT_NOP) 1311 optlen = 1; 1312 else { 1313 optlen = cp[1]; 1314 if (optlen <= 0) 1315 break; 1316 } 1317 switch (opt) { 1318 1319 default: 1320 continue; 1321 1322 case TCPOPT_MAXSEG: 1323 if (optlen != TCPOLEN_MAXSEG) 1324 continue; 1325 if (!(ti->ti_flags & TH_SYN)) 1326 continue; 1327 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1328 NTOHS(mss); 1329 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1330 break; 1331 1332 case TCPOPT_WINDOW: 1333 if (optlen != TCPOLEN_WINDOW) 1334 continue; 1335 if (!(ti->ti_flags & TH_SYN)) 1336 continue; 1337 tp->t_flags |= TF_RCVD_SCALE; 1338 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1339 break; 1340 1341 case TCPOPT_TIMESTAMP: 1342 if (optlen != TCPOLEN_TIMESTAMP) 1343 continue; 1344 *ts_present = 1; 1345 bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val)); 1346 NTOHL(*ts_val); 1347 bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr)); 1348 NTOHL(*ts_ecr); 1349 1350 /* 1351 * A timestamp received in a SYN makes 1352 * it ok to send timestamp requests and replies. 1353 */ 1354 if (ti->ti_flags & TH_SYN) { 1355 tp->t_flags |= TF_RCVD_TSTMP; 1356 tp->ts_recent = *ts_val; 1357 tp->ts_recent_age = tcp_now; 1358 } 1359 break; 1360 } 1361 } 1362 } 1363 1364 /* 1365 * Pull out of band byte out of a segment so 1366 * it doesn't appear in the user's data queue. 1367 * It is still reflected in the segment length for 1368 * sequencing purposes. 1369 */ 1370 tcp_pulloutofband(so, ti, m) 1371 struct socket *so; 1372 struct tcpiphdr *ti; 1373 register struct mbuf *m; 1374 { 1375 int cnt = ti->ti_urp - 1; 1376 1377 while (cnt >= 0) { 1378 if (m->m_len > cnt) { 1379 char *cp = mtod(m, caddr_t) + cnt; 1380 struct tcpcb *tp = sototcpcb(so); 1381 1382 tp->t_iobc = *cp; 1383 tp->t_oobflags |= TCPOOB_HAVEDATA; 1384 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1385 m->m_len--; 1386 return; 1387 } 1388 cnt -= m->m_len; 1389 m = m->m_next; 1390 if (m == 0) 1391 break; 1392 } 1393 panic("tcp_pulloutofband"); 1394 } 1395 1396 /* 1397 * Collect new round-trip time estimate 1398 * and update averages and current timeout. 1399 */ 1400 tcp_xmit_timer(tp, rtt) 1401 register struct tcpcb *tp; 1402 short rtt; 1403 { 1404 register short delta; 1405 1406 tcpstat.tcps_rttupdated++; 1407 if (tp->t_srtt != 0) { 1408 /* 1409 * srtt is stored as fixed point with 3 bits after the 1410 * binary point (i.e., scaled by 8). The following magic 1411 * is equivalent to the smoothing algorithm in rfc793 with 1412 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1413 * point). Adjust rtt to origin 0. 1414 */ 1415 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1416 if ((tp->t_srtt += delta) <= 0) 1417 tp->t_srtt = 1; 1418 /* 1419 * We accumulate a smoothed rtt variance (actually, a 1420 * smoothed mean difference), then set the retransmit 1421 * timer to smoothed rtt + 4 times the smoothed variance. 1422 * rttvar is stored as fixed point with 2 bits after the 1423 * binary point (scaled by 4). The following is 1424 * equivalent to rfc793 smoothing with an alpha of .75 1425 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1426 * rfc793's wired-in beta. 1427 */ 1428 if (delta < 0) 1429 delta = -delta; 1430 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1431 if ((tp->t_rttvar += delta) <= 0) 1432 tp->t_rttvar = 1; 1433 } else { 1434 /* 1435 * No rtt measurement yet - use the unsmoothed rtt. 1436 * Set the variance to half the rtt (so our first 1437 * retransmit happens at 3*rtt). 1438 */ 1439 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1440 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1441 } 1442 tp->t_rtt = 0; 1443 tp->t_rxtshift = 0; 1444 1445 /* 1446 * the retransmit should happen at rtt + 4 * rttvar. 1447 * Because of the way we do the smoothing, srtt and rttvar 1448 * will each average +1/2 tick of bias. When we compute 1449 * the retransmit timer, we want 1/2 tick of rounding and 1450 * 1 extra tick because of +-1/2 tick uncertainty in the 1451 * firing of the timer. The bias will give us exactly the 1452 * 1.5 tick we need. But, because the bias is 1453 * statistical, we have to test that we don't drop below 1454 * the minimum feasible timer (which is 2 ticks). 1455 */ 1456 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1457 tp->t_rttmin, TCPTV_REXMTMAX); 1458 1459 /* 1460 * We received an ack for a packet that wasn't retransmitted; 1461 * it is probably safe to discard any error indications we've 1462 * received recently. This isn't quite right, but close enough 1463 * for now (a route might have failed after we sent a segment, 1464 * and the return path might not be symmetrical). 1465 */ 1466 tp->t_softerror = 0; 1467 } 1468 1469 /* 1470 * Determine a reasonable value for maxseg size. 1471 * If the route is known, check route for mtu. 1472 * If none, use an mss that can be handled on the outgoing 1473 * interface without forcing IP to fragment; if bigger than 1474 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1475 * to utilize large mbufs. If no route is found, route has no mtu, 1476 * or the destination isn't local, use a default, hopefully conservative 1477 * size (usually 512 or the default IP max size, but no more than the mtu 1478 * of the interface), as we can't discover anything about intervening 1479 * gateways or networks. We also initialize the congestion/slow start 1480 * window to be a single segment if the destination isn't local. 1481 * While looking at the routing entry, we also initialize other path-dependent 1482 * parameters from pre-set or cached values in the routing entry. 1483 */ 1484 1485 tcp_mss(tp, offer) 1486 register struct tcpcb *tp; 1487 u_short offer; 1488 { 1489 struct route *ro; 1490 register struct rtentry *rt; 1491 struct ifnet *ifp; 1492 register int rtt, mss; 1493 u_long bufsize; 1494 struct inpcb *inp; 1495 struct socket *so; 1496 extern int tcp_mssdflt, tcp_rttdflt; 1497 1498 inp = tp->t_inpcb; 1499 ro = &inp->inp_route; 1500 1501 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1502 /* No route yet, so try to acquire one */ 1503 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1504 ro->ro_dst.sa_family = AF_INET; 1505 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1506 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1507 inp->inp_faddr; 1508 rtalloc(ro); 1509 } 1510 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1511 return (tcp_mssdflt); 1512 } 1513 ifp = rt->rt_ifp; 1514 so = inp->inp_socket; 1515 1516 #ifdef RTV_MTU /* if route characteristics exist ... */ 1517 /* 1518 * While we're here, check if there's an initial rtt 1519 * or rttvar. Convert from the route-table units 1520 * to scaled multiples of the slow timeout timer. 1521 */ 1522 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1523 /* 1524 * XXX the lock bit for MTU indicates that the value 1525 * is also a minimum value; this is subject to time. 1526 */ 1527 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1528 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1529 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1530 if (rt->rt_rmx.rmx_rttvar) 1531 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1532 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1533 else 1534 /* default variation is +- 1 rtt */ 1535 tp->t_rttvar = 1536 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1537 TCPT_RANGESET(tp->t_rxtcur, 1538 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1539 tp->t_rttmin, TCPTV_REXMTMAX); 1540 } 1541 /* 1542 * if there's an mtu associated with the route, use it 1543 */ 1544 if (rt->rt_rmx.rmx_mtu) 1545 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1546 else 1547 #endif /* RTV_MTU */ 1548 { 1549 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1550 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1551 if (mss > MCLBYTES) 1552 mss &= ~(MCLBYTES-1); 1553 #else 1554 if (mss > MCLBYTES) 1555 mss = mss / MCLBYTES * MCLBYTES; 1556 #endif 1557 if (!in_localaddr(inp->inp_faddr)) 1558 mss = min(mss, tcp_mssdflt); 1559 } 1560 /* 1561 * The current mss, t_maxseg, is initialized to the default value. 1562 * If we compute a smaller value, reduce the current mss. 1563 * If we compute a larger value, return it for use in sending 1564 * a max seg size option, but don't store it for use 1565 * unless we received an offer at least that large from peer. 1566 * However, do not accept offers under 32 bytes. 1567 */ 1568 if (offer) 1569 mss = min(mss, offer); 1570 mss = max(mss, 32); /* sanity */ 1571 if (mss < tp->t_maxseg || offer != 0) { 1572 /* 1573 * If there's a pipesize, change the socket buffer 1574 * to that size. Make the socket buffers an integral 1575 * number of mss units; if the mss is larger than 1576 * the socket buffer, decrease the mss. 1577 */ 1578 #ifdef RTV_SPIPE 1579 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1580 #endif 1581 bufsize = so->so_snd.sb_hiwat; 1582 if (bufsize < mss) 1583 mss = bufsize; 1584 else { 1585 bufsize = roundup(bufsize, mss); 1586 if (bufsize > sb_max) 1587 bufsize = sb_max; 1588 (void)sbreserve(&so->so_snd, bufsize); 1589 } 1590 tp->t_maxseg = mss; 1591 1592 #ifdef RTV_RPIPE 1593 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1594 #endif 1595 bufsize = so->so_rcv.sb_hiwat; 1596 if (bufsize > mss) { 1597 bufsize = roundup(bufsize, mss); 1598 if (bufsize > sb_max) 1599 bufsize = sb_max; 1600 (void)sbreserve(&so->so_rcv, bufsize); 1601 } 1602 } 1603 tp->snd_cwnd = mss; 1604 1605 #ifdef RTV_SSTHRESH 1606 if (rt->rt_rmx.rmx_ssthresh) { 1607 /* 1608 * There's some sort of gateway or interface 1609 * buffer limit on the path. Use this to set 1610 * the slow start threshhold, but set the 1611 * threshold to no less than 2*mss. 1612 */ 1613 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1614 } 1615 #endif /* RTV_MTU */ 1616 return (mss); 1617 } 1618 #endif /* TUBA_INCLUDE */ 1619