1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994 3 * The Regents of the University of California. All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)tcp_input.c 8.7 (Berkeley) 10/27/94 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 struct tcpiphdr tcp_saveti; 38 struct inpcb *tcp_last_inpcb = &tcb; 39 40 extern u_long sb_max; 41 42 #endif /* TUBA_INCLUDE */ 43 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 44 45 /* for modulo comparisons of timestamps */ 46 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 47 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 48 49 50 /* 51 * Insert segment ti into reassembly queue of tcp with 52 * control block tp. Return TH_FIN if reassembly now includes 53 * a segment with FIN. The macro form does the common case inline 54 * (segment is the next to be received on an established connection, 55 * and the queue is empty), avoiding linkage into and removal 56 * from the queue and repetition of various conversions. 57 * Set DELACK for segments received in order, but ack immediately 58 * when segments are out of order (so fast retransmit can work). 59 */ 60 #define TCP_REASS(tp, ti, m, so, flags) { \ 61 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 62 (tp)->seg_next == (struct tcpiphdr *)(tp) && \ 63 (tp)->t_state == TCPS_ESTABLISHED) { \ 64 tp->t_flags |= TF_DELACK; \ 65 (tp)->rcv_nxt += (ti)->ti_len; \ 66 flags = (ti)->ti_flags & TH_FIN; \ 67 tcpstat.tcps_rcvpack++;\ 68 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 69 sbappend(&(so)->so_rcv, (m)); \ 70 sorwakeup(so); \ 71 } else { \ 72 (flags) = tcp_reass((tp), (ti), (m)); \ 73 tp->t_flags |= TF_ACKNOW; \ 74 } \ 75 } 76 #ifndef TUBA_INCLUDE 77 78 int 79 tcp_reass(tp, ti, m) 80 register struct tcpcb *tp; 81 register struct tcpiphdr *ti; 82 struct mbuf *m; 83 { 84 register struct tcpiphdr *q; 85 struct socket *so = tp->t_inpcb->inp_socket; 86 int flags; 87 88 /* 89 * Call with ti==0 after become established to 90 * force pre-ESTABLISHED data up to user socket. 91 */ 92 if (ti == 0) 93 goto present; 94 95 /* 96 * Find a segment which begins after this one does. 97 */ 98 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 99 q = (struct tcpiphdr *)q->ti_next) 100 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 101 break; 102 103 /* 104 * If there is a preceding segment, it may provide some of 105 * our data already. If so, drop the data from the incoming 106 * segment. If it provides all of our data, drop us. 107 */ 108 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 109 register int i; 110 q = (struct tcpiphdr *)q->ti_prev; 111 /* conversion to int (in i) handles seq wraparound */ 112 i = q->ti_seq + q->ti_len - ti->ti_seq; 113 if (i > 0) { 114 if (i >= ti->ti_len) { 115 tcpstat.tcps_rcvduppack++; 116 tcpstat.tcps_rcvdupbyte += ti->ti_len; 117 m_freem(m); 118 return (0); 119 } 120 m_adj(m, i); 121 ti->ti_len -= i; 122 ti->ti_seq += i; 123 } 124 q = (struct tcpiphdr *)(q->ti_next); 125 } 126 tcpstat.tcps_rcvoopack++; 127 tcpstat.tcps_rcvoobyte += ti->ti_len; 128 REASS_MBUF(ti) = m; /* XXX */ 129 130 /* 131 * While we overlap succeeding segments trim them or, 132 * if they are completely covered, dequeue them. 133 */ 134 while (q != (struct tcpiphdr *)tp) { 135 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 136 if (i <= 0) 137 break; 138 if (i < q->ti_len) { 139 q->ti_seq += i; 140 q->ti_len -= i; 141 m_adj(REASS_MBUF(q), i); 142 break; 143 } 144 q = (struct tcpiphdr *)q->ti_next; 145 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 146 remque(q->ti_prev); 147 m_freem(m); 148 } 149 150 /* 151 * Stick new segment in its place. 152 */ 153 insque(ti, q->ti_prev); 154 155 present: 156 /* 157 * Present data to user, advancing rcv_nxt through 158 * completed sequence space. 159 */ 160 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 161 return (0); 162 ti = tp->seg_next; 163 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 164 return (0); 165 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 166 return (0); 167 do { 168 tp->rcv_nxt += ti->ti_len; 169 flags = ti->ti_flags & TH_FIN; 170 remque(ti); 171 m = REASS_MBUF(ti); 172 ti = (struct tcpiphdr *)ti->ti_next; 173 if (so->so_state & SS_CANTRCVMORE) 174 m_freem(m); 175 else 176 sbappend(&so->so_rcv, m); 177 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 178 sorwakeup(so); 179 return (flags); 180 } 181 182 /* 183 * TCP input routine, follows pages 65-76 of the 184 * protocol specification dated September, 1981 very closely. 185 */ 186 void 187 tcp_input(m, iphlen) 188 register struct mbuf *m; 189 int iphlen; 190 { 191 register struct tcpiphdr *ti; 192 register struct inpcb *inp; 193 u_char *optp = NULL; 194 int optlen; 195 int len, tlen, off; 196 register struct tcpcb *tp = 0; 197 register int tiflags; 198 struct socket *so; 199 int todrop, acked, ourfinisacked, needoutput = 0; 200 short ostate; 201 struct in_addr laddr; 202 int dropsocket = 0; 203 int iss = 0; 204 u_long tiwin, ts_val, ts_ecr; 205 int ts_present = 0; 206 207 tcpstat.tcps_rcvtotal++; 208 /* 209 * Get IP and TCP header together in first mbuf. 210 * Note: IP leaves IP header in first mbuf. 211 */ 212 ti = mtod(m, struct tcpiphdr *); 213 if (iphlen > sizeof (struct ip)) 214 ip_stripoptions(m, (struct mbuf *)0); 215 if (m->m_len < sizeof (struct tcpiphdr)) { 216 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 217 tcpstat.tcps_rcvshort++; 218 return; 219 } 220 ti = mtod(m, struct tcpiphdr *); 221 } 222 223 /* 224 * Checksum extended TCP header and data. 225 */ 226 tlen = ((struct ip *)ti)->ip_len; 227 len = sizeof (struct ip) + tlen; 228 ti->ti_next = ti->ti_prev = 0; 229 ti->ti_x1 = 0; 230 ti->ti_len = (u_short)tlen; 231 HTONS(ti->ti_len); 232 if (ti->ti_sum = in_cksum(m, len)) { 233 tcpstat.tcps_rcvbadsum++; 234 goto drop; 235 } 236 #endif /* TUBA_INCLUDE */ 237 238 /* 239 * Check that TCP offset makes sense, 240 * pull out TCP options and adjust length. XXX 241 */ 242 off = ti->ti_off << 2; 243 if (off < sizeof (struct tcphdr) || off > tlen) { 244 tcpstat.tcps_rcvbadoff++; 245 goto drop; 246 } 247 tlen -= off; 248 ti->ti_len = tlen; 249 if (off > sizeof (struct tcphdr)) { 250 if (m->m_len < sizeof(struct ip) + off) { 251 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 252 tcpstat.tcps_rcvshort++; 253 return; 254 } 255 ti = mtod(m, struct tcpiphdr *); 256 } 257 optlen = off - sizeof (struct tcphdr); 258 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr); 259 /* 260 * Do quick retrieval of timestamp options ("options 261 * prediction?"). If timestamp is the only option and it's 262 * formatted as recommended in RFC 1323 appendix A, we 263 * quickly get the values now and not bother calling 264 * tcp_dooptions(), etc. 265 */ 266 if ((optlen == TCPOLEN_TSTAMP_APPA || 267 (optlen > TCPOLEN_TSTAMP_APPA && 268 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 269 *(u_long *)optp == htonl(TCPOPT_TSTAMP_HDR) && 270 (ti->ti_flags & TH_SYN) == 0) { 271 ts_present = 1; 272 ts_val = ntohl(*(u_long *)(optp + 4)); 273 ts_ecr = ntohl(*(u_long *)(optp + 8)); 274 optp = NULL; /* we've parsed the options */ 275 } 276 } 277 tiflags = ti->ti_flags; 278 279 /* 280 * Convert TCP protocol specific fields to host format. 281 */ 282 NTOHL(ti->ti_seq); 283 NTOHL(ti->ti_ack); 284 NTOHS(ti->ti_win); 285 NTOHS(ti->ti_urp); 286 287 /* 288 * Locate pcb for segment. 289 */ 290 findpcb: 291 inp = tcp_last_inpcb; 292 if (inp->inp_lport != ti->ti_dport || 293 inp->inp_fport != ti->ti_sport || 294 inp->inp_faddr.s_addr != ti->ti_src.s_addr || 295 inp->inp_laddr.s_addr != ti->ti_dst.s_addr) { 296 inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport, 297 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 298 if (inp) 299 tcp_last_inpcb = inp; 300 ++tcpstat.tcps_pcbcachemiss; 301 } 302 303 /* 304 * If the state is CLOSED (i.e., TCB does not exist) then 305 * all data in the incoming segment is discarded. 306 * If the TCB exists but is in CLOSED state, it is embryonic, 307 * but should either do a listen or a connect soon. 308 */ 309 if (inp == 0) 310 goto dropwithreset; 311 tp = intotcpcb(inp); 312 if (tp == 0) 313 goto dropwithreset; 314 if (tp->t_state == TCPS_CLOSED) 315 goto drop; 316 317 /* Unscale the window into a 32-bit value. */ 318 if ((tiflags & TH_SYN) == 0) 319 tiwin = ti->ti_win << tp->snd_scale; 320 else 321 tiwin = ti->ti_win; 322 323 so = inp->inp_socket; 324 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 325 if (so->so_options & SO_DEBUG) { 326 ostate = tp->t_state; 327 tcp_saveti = *ti; 328 } 329 if (so->so_options & SO_ACCEPTCONN) { 330 so = sonewconn(so, 0); 331 if (so == 0) 332 goto drop; 333 /* 334 * This is ugly, but .... 335 * 336 * Mark socket as temporary until we're 337 * committed to keeping it. The code at 338 * ``drop'' and ``dropwithreset'' check the 339 * flag dropsocket to see if the temporary 340 * socket created here should be discarded. 341 * We mark the socket as discardable until 342 * we're committed to it below in TCPS_LISTEN. 343 */ 344 dropsocket++; 345 inp = (struct inpcb *)so->so_pcb; 346 inp->inp_laddr = ti->ti_dst; 347 inp->inp_lport = ti->ti_dport; 348 #if BSD>=43 349 inp->inp_options = ip_srcroute(); 350 #endif 351 tp = intotcpcb(inp); 352 tp->t_state = TCPS_LISTEN; 353 354 /* Compute proper scaling value from buffer space 355 */ 356 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 357 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 358 tp->request_r_scale++; 359 } 360 } 361 362 /* 363 * Segment received on connection. 364 * Reset idle time and keep-alive timer. 365 */ 366 tp->t_idle = 0; 367 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 368 369 /* 370 * Process options if not in LISTEN state, 371 * else do it below (after getting remote address). 372 */ 373 if (optp && tp->t_state != TCPS_LISTEN) 374 tcp_dooptions(tp, optp, optlen, ti, 375 &ts_present, &ts_val, &ts_ecr); 376 377 /* 378 * Header prediction: check for the two common cases 379 * of a uni-directional data xfer. If the packet has 380 * no control flags, is in-sequence, the window didn't 381 * change and we're not retransmitting, it's a 382 * candidate. If the length is zero and the ack moved 383 * forward, we're the sender side of the xfer. Just 384 * free the data acked & wake any higher level process 385 * that was blocked waiting for space. If the length 386 * is non-zero and the ack didn't move, we're the 387 * receiver side. If we're getting packets in-order 388 * (the reassembly queue is empty), add the data to 389 * the socket buffer and note that we need a delayed ack. 390 */ 391 if (tp->t_state == TCPS_ESTABLISHED && 392 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 393 (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && 394 ti->ti_seq == tp->rcv_nxt && 395 tiwin && tiwin == tp->snd_wnd && 396 tp->snd_nxt == tp->snd_max) { 397 398 /* 399 * If last ACK falls within this segment's sequence numbers, 400 * record the timestamp. 401 */ 402 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 403 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) { 404 tp->ts_recent_age = tcp_now; 405 tp->ts_recent = ts_val; 406 } 407 408 if (ti->ti_len == 0) { 409 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 410 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 411 tp->snd_cwnd >= tp->snd_wnd) { 412 /* 413 * this is a pure ack for outstanding data. 414 */ 415 ++tcpstat.tcps_predack; 416 if (ts_present) 417 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 418 else if (tp->t_rtt && 419 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 420 tcp_xmit_timer(tp, tp->t_rtt); 421 acked = ti->ti_ack - tp->snd_una; 422 tcpstat.tcps_rcvackpack++; 423 tcpstat.tcps_rcvackbyte += acked; 424 sbdrop(&so->so_snd, acked); 425 tp->snd_una = ti->ti_ack; 426 m_freem(m); 427 428 /* 429 * If all outstanding data are acked, stop 430 * retransmit timer, otherwise restart timer 431 * using current (possibly backed-off) value. 432 * If process is waiting for space, 433 * wakeup/selwakeup/signal. If data 434 * are ready to send, let tcp_output 435 * decide between more output or persist. 436 */ 437 if (tp->snd_una == tp->snd_max) 438 tp->t_timer[TCPT_REXMT] = 0; 439 else if (tp->t_timer[TCPT_PERSIST] == 0) 440 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 441 442 if (so->so_snd.sb_flags & SB_NOTIFY) 443 sowwakeup(so); 444 if (so->so_snd.sb_cc) 445 (void) tcp_output(tp); 446 return; 447 } 448 } else if (ti->ti_ack == tp->snd_una && 449 tp->seg_next == (struct tcpiphdr *)tp && 450 ti->ti_len <= sbspace(&so->so_rcv)) { 451 /* 452 * this is a pure, in-sequence data packet 453 * with nothing on the reassembly queue and 454 * we have enough buffer space to take it. 455 */ 456 ++tcpstat.tcps_preddat; 457 tp->rcv_nxt += ti->ti_len; 458 tcpstat.tcps_rcvpack++; 459 tcpstat.tcps_rcvbyte += ti->ti_len; 460 /* 461 * Drop TCP, IP headers and TCP options then add data 462 * to socket buffer. 463 */ 464 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 465 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 466 sbappend(&so->so_rcv, m); 467 sorwakeup(so); 468 tp->t_flags |= TF_DELACK; 469 return; 470 } 471 } 472 473 /* 474 * Drop TCP, IP headers and TCP options. 475 */ 476 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 477 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 478 479 /* 480 * Calculate amount of space in receive window, 481 * and then do TCP input processing. 482 * Receive window is amount of space in rcv queue, 483 * but not less than advertised window. 484 */ 485 { int win; 486 487 win = sbspace(&so->so_rcv); 488 if (win < 0) 489 win = 0; 490 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 491 } 492 493 switch (tp->t_state) { 494 495 /* 496 * If the state is LISTEN then ignore segment if it contains an RST. 497 * If the segment contains an ACK then it is bad and send a RST. 498 * If it does not contain a SYN then it is not interesting; drop it. 499 * Don't bother responding if the destination was a broadcast. 500 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 501 * tp->iss, and send a segment: 502 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 503 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 504 * Fill in remote peer address fields if not previously specified. 505 * Enter SYN_RECEIVED state, and process any other fields of this 506 * segment in this state. 507 */ 508 case TCPS_LISTEN: { 509 struct mbuf *am; 510 register struct sockaddr_in *sin; 511 512 if (tiflags & TH_RST) 513 goto drop; 514 if (tiflags & TH_ACK) 515 goto dropwithreset; 516 if ((tiflags & TH_SYN) == 0) 517 goto drop; 518 /* 519 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 520 * in_broadcast() should never return true on a received 521 * packet with M_BCAST not set. 522 */ 523 if (m->m_flags & (M_BCAST|M_MCAST) || 524 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 525 goto drop; 526 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 527 if (am == NULL) 528 goto drop; 529 am->m_len = sizeof (struct sockaddr_in); 530 sin = mtod(am, struct sockaddr_in *); 531 sin->sin_family = AF_INET; 532 sin->sin_len = sizeof(*sin); 533 sin->sin_addr = ti->ti_src; 534 sin->sin_port = ti->ti_sport; 535 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 536 laddr = inp->inp_laddr; 537 if (inp->inp_laddr.s_addr == INADDR_ANY) 538 inp->inp_laddr = ti->ti_dst; 539 if (in_pcbconnect(inp, am)) { 540 inp->inp_laddr = laddr; 541 (void) m_free(am); 542 goto drop; 543 } 544 (void) m_free(am); 545 tp->t_template = tcp_template(tp); 546 if (tp->t_template == 0) { 547 tp = tcp_drop(tp, ENOBUFS); 548 dropsocket = 0; /* socket is already gone */ 549 goto drop; 550 } 551 if (optp) 552 tcp_dooptions(tp, optp, optlen, ti, 553 &ts_present, &ts_val, &ts_ecr); 554 if (iss) 555 tp->iss = iss; 556 else 557 tp->iss = tcp_iss; 558 tcp_iss += TCP_ISSINCR/2; 559 tp->irs = ti->ti_seq; 560 tcp_sendseqinit(tp); 561 tcp_rcvseqinit(tp); 562 tp->t_flags |= TF_ACKNOW; 563 tp->t_state = TCPS_SYN_RECEIVED; 564 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 565 dropsocket = 0; /* committed to socket */ 566 tcpstat.tcps_accepts++; 567 goto trimthenstep6; 568 } 569 570 /* 571 * If the state is SYN_SENT: 572 * if seg contains an ACK, but not for our SYN, drop the input. 573 * if seg contains a RST, then drop the connection. 574 * if seg does not contain SYN, then drop it. 575 * Otherwise this is an acceptable SYN segment 576 * initialize tp->rcv_nxt and tp->irs 577 * if seg contains ack then advance tp->snd_una 578 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 579 * arrange for segment to be acked (eventually) 580 * continue processing rest of data/controls, beginning with URG 581 */ 582 case TCPS_SYN_SENT: 583 if ((tiflags & TH_ACK) && 584 (SEQ_LEQ(ti->ti_ack, tp->iss) || 585 SEQ_GT(ti->ti_ack, tp->snd_max))) 586 goto dropwithreset; 587 if (tiflags & TH_RST) { 588 if (tiflags & TH_ACK) 589 tp = tcp_drop(tp, ECONNREFUSED); 590 goto drop; 591 } 592 if ((tiflags & TH_SYN) == 0) 593 goto drop; 594 if (tiflags & TH_ACK) { 595 tp->snd_una = ti->ti_ack; 596 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 597 tp->snd_nxt = tp->snd_una; 598 } 599 tp->t_timer[TCPT_REXMT] = 0; 600 tp->irs = ti->ti_seq; 601 tcp_rcvseqinit(tp); 602 tp->t_flags |= TF_ACKNOW; 603 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 604 tcpstat.tcps_connects++; 605 soisconnected(so); 606 tp->t_state = TCPS_ESTABLISHED; 607 /* Do window scaling on this connection? */ 608 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 609 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 610 tp->snd_scale = tp->requested_s_scale; 611 tp->rcv_scale = tp->request_r_scale; 612 } 613 (void) tcp_reass(tp, (struct tcpiphdr *)0, 614 (struct mbuf *)0); 615 /* 616 * if we didn't have to retransmit the SYN, 617 * use its rtt as our initial srtt & rtt var. 618 */ 619 if (tp->t_rtt) 620 tcp_xmit_timer(tp, tp->t_rtt); 621 } else 622 tp->t_state = TCPS_SYN_RECEIVED; 623 624 trimthenstep6: 625 /* 626 * Advance ti->ti_seq to correspond to first data byte. 627 * If data, trim to stay within window, 628 * dropping FIN if necessary. 629 */ 630 ti->ti_seq++; 631 if (ti->ti_len > tp->rcv_wnd) { 632 todrop = ti->ti_len - tp->rcv_wnd; 633 m_adj(m, -todrop); 634 ti->ti_len = tp->rcv_wnd; 635 tiflags &= ~TH_FIN; 636 tcpstat.tcps_rcvpackafterwin++; 637 tcpstat.tcps_rcvbyteafterwin += todrop; 638 } 639 tp->snd_wl1 = ti->ti_seq - 1; 640 tp->rcv_up = ti->ti_seq; 641 goto step6; 642 } 643 644 /* 645 * States other than LISTEN or SYN_SENT. 646 * First check timestamp, if present. 647 * Then check that at least some bytes of segment are within 648 * receive window. If segment begins before rcv_nxt, 649 * drop leading data (and SYN); if nothing left, just ack. 650 * 651 * RFC 1323 PAWS: If we have a timestamp reply on this segment 652 * and it's less than ts_recent, drop it. 653 */ 654 if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 655 TSTMP_LT(ts_val, tp->ts_recent)) { 656 657 /* Check to see if ts_recent is over 24 days old. */ 658 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 659 /* 660 * Invalidate ts_recent. If this segment updates 661 * ts_recent, the age will be reset later and ts_recent 662 * will get a valid value. If it does not, setting 663 * ts_recent to zero will at least satisfy the 664 * requirement that zero be placed in the timestamp 665 * echo reply when ts_recent isn't valid. The 666 * age isn't reset until we get a valid ts_recent 667 * because we don't want out-of-order segments to be 668 * dropped when ts_recent is old. 669 */ 670 tp->ts_recent = 0; 671 } else { 672 tcpstat.tcps_rcvduppack++; 673 tcpstat.tcps_rcvdupbyte += ti->ti_len; 674 tcpstat.tcps_pawsdrop++; 675 goto dropafterack; 676 } 677 } 678 679 todrop = tp->rcv_nxt - ti->ti_seq; 680 if (todrop > 0) { 681 if (tiflags & TH_SYN) { 682 tiflags &= ~TH_SYN; 683 ti->ti_seq++; 684 if (ti->ti_urp > 1) 685 ti->ti_urp--; 686 else 687 tiflags &= ~TH_URG; 688 todrop--; 689 } 690 if (todrop >= ti->ti_len) { 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 * (maxseg / cwnd) per packet). 995 */ 996 { 997 register u_int cw = tp->snd_cwnd; 998 register u_int incr = tp->t_maxseg; 999 1000 if (cw > tp->snd_ssthresh) 1001 incr = incr * incr / cw; 1002 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1003 } 1004 if (acked > so->so_snd.sb_cc) { 1005 tp->snd_wnd -= so->so_snd.sb_cc; 1006 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1007 ourfinisacked = 1; 1008 } else { 1009 sbdrop(&so->so_snd, acked); 1010 tp->snd_wnd -= acked; 1011 ourfinisacked = 0; 1012 } 1013 if (so->so_snd.sb_flags & SB_NOTIFY) 1014 sowwakeup(so); 1015 tp->snd_una = ti->ti_ack; 1016 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1017 tp->snd_nxt = tp->snd_una; 1018 1019 switch (tp->t_state) { 1020 1021 /* 1022 * In FIN_WAIT_1 STATE in addition to the processing 1023 * for the ESTABLISHED state if our FIN is now acknowledged 1024 * then enter FIN_WAIT_2. 1025 */ 1026 case TCPS_FIN_WAIT_1: 1027 if (ourfinisacked) { 1028 /* 1029 * If we can't receive any more 1030 * data, then closing user can proceed. 1031 * Starting the timer is contrary to the 1032 * specification, but if we don't get a FIN 1033 * we'll hang forever. 1034 */ 1035 if (so->so_state & SS_CANTRCVMORE) { 1036 soisdisconnected(so); 1037 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1038 } 1039 tp->t_state = TCPS_FIN_WAIT_2; 1040 } 1041 break; 1042 1043 /* 1044 * In CLOSING STATE in addition to the processing for 1045 * the ESTABLISHED state if the ACK acknowledges our FIN 1046 * then enter the TIME-WAIT state, otherwise ignore 1047 * the segment. 1048 */ 1049 case TCPS_CLOSING: 1050 if (ourfinisacked) { 1051 tp->t_state = TCPS_TIME_WAIT; 1052 tcp_canceltimers(tp); 1053 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1054 soisdisconnected(so); 1055 } 1056 break; 1057 1058 /* 1059 * In LAST_ACK, we may still be waiting for data to drain 1060 * and/or to be acked, as well as for the ack of our FIN. 1061 * If our FIN is now acknowledged, delete the TCB, 1062 * enter the closed state and return. 1063 */ 1064 case TCPS_LAST_ACK: 1065 if (ourfinisacked) { 1066 tp = tcp_close(tp); 1067 goto drop; 1068 } 1069 break; 1070 1071 /* 1072 * In TIME_WAIT state the only thing that should arrive 1073 * is a retransmission of the remote FIN. Acknowledge 1074 * it and restart the finack timer. 1075 */ 1076 case TCPS_TIME_WAIT: 1077 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1078 goto dropafterack; 1079 } 1080 } 1081 1082 step6: 1083 /* 1084 * Update window information. 1085 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1086 */ 1087 if ((tiflags & TH_ACK) && 1088 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 1089 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1090 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))) { 1091 /* keep track of pure window updates */ 1092 if (ti->ti_len == 0 && 1093 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1094 tcpstat.tcps_rcvwinupd++; 1095 tp->snd_wnd = tiwin; 1096 tp->snd_wl1 = ti->ti_seq; 1097 tp->snd_wl2 = ti->ti_ack; 1098 if (tp->snd_wnd > tp->max_sndwnd) 1099 tp->max_sndwnd = tp->snd_wnd; 1100 needoutput = 1; 1101 } 1102 1103 /* 1104 * Process segments with URG. 1105 */ 1106 if ((tiflags & TH_URG) && ti->ti_urp && 1107 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1108 /* 1109 * This is a kludge, but if we receive and accept 1110 * random urgent pointers, we'll crash in 1111 * soreceive. It's hard to imagine someone 1112 * actually wanting to send this much urgent data. 1113 */ 1114 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1115 ti->ti_urp = 0; /* XXX */ 1116 tiflags &= ~TH_URG; /* XXX */ 1117 goto dodata; /* XXX */ 1118 } 1119 /* 1120 * If this segment advances the known urgent pointer, 1121 * then mark the data stream. This should not happen 1122 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1123 * a FIN has been received from the remote side. 1124 * In these states we ignore the URG. 1125 * 1126 * According to RFC961 (Assigned Protocols), 1127 * the urgent pointer points to the last octet 1128 * of urgent data. We continue, however, 1129 * to consider it to indicate the first octet 1130 * of data past the urgent section as the original 1131 * spec states (in one of two places). 1132 */ 1133 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1134 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1135 so->so_oobmark = so->so_rcv.sb_cc + 1136 (tp->rcv_up - tp->rcv_nxt) - 1; 1137 if (so->so_oobmark == 0) 1138 so->so_state |= SS_RCVATMARK; 1139 sohasoutofband(so); 1140 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1141 } 1142 /* 1143 * Remove out of band data so doesn't get presented to user. 1144 * This can happen independent of advancing the URG pointer, 1145 * but if two URG's are pending at once, some out-of-band 1146 * data may creep in... ick. 1147 */ 1148 if (ti->ti_urp <= ti->ti_len 1149 #ifdef SO_OOBINLINE 1150 && (so->so_options & SO_OOBINLINE) == 0 1151 #endif 1152 ) 1153 tcp_pulloutofband(so, ti, m); 1154 } else 1155 /* 1156 * If no out of band data is expected, 1157 * pull receive urgent pointer along 1158 * with the receive window. 1159 */ 1160 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1161 tp->rcv_up = tp->rcv_nxt; 1162 dodata: /* XXX */ 1163 1164 /* 1165 * Process the segment text, merging it into the TCP sequencing queue, 1166 * and arranging for acknowledgment of receipt if necessary. 1167 * This process logically involves adjusting tp->rcv_wnd as data 1168 * is presented to the user (this happens in tcp_usrreq.c, 1169 * case PRU_RCVD). If a FIN has already been received on this 1170 * connection then we just ignore the text. 1171 */ 1172 if ((ti->ti_len || (tiflags&TH_FIN)) && 1173 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1174 TCP_REASS(tp, ti, m, so, tiflags); 1175 /* 1176 * Note the amount of data that peer has sent into 1177 * our window, in order to estimate the sender's 1178 * buffer size. 1179 */ 1180 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1181 } else { 1182 m_freem(m); 1183 tiflags &= ~TH_FIN; 1184 } 1185 1186 /* 1187 * If FIN is received ACK the FIN and let the user know 1188 * that the connection is closing. 1189 */ 1190 if (tiflags & TH_FIN) { 1191 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1192 socantrcvmore(so); 1193 tp->t_flags |= TF_ACKNOW; 1194 tp->rcv_nxt++; 1195 } 1196 switch (tp->t_state) { 1197 1198 /* 1199 * In SYN_RECEIVED and ESTABLISHED STATES 1200 * enter the CLOSE_WAIT state. 1201 */ 1202 case TCPS_SYN_RECEIVED: 1203 case TCPS_ESTABLISHED: 1204 tp->t_state = TCPS_CLOSE_WAIT; 1205 break; 1206 1207 /* 1208 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1209 * enter the CLOSING state. 1210 */ 1211 case TCPS_FIN_WAIT_1: 1212 tp->t_state = TCPS_CLOSING; 1213 break; 1214 1215 /* 1216 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1217 * starting the time-wait timer, turning off the other 1218 * standard timers. 1219 */ 1220 case TCPS_FIN_WAIT_2: 1221 tp->t_state = TCPS_TIME_WAIT; 1222 tcp_canceltimers(tp); 1223 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1224 soisdisconnected(so); 1225 break; 1226 1227 /* 1228 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1229 */ 1230 case TCPS_TIME_WAIT: 1231 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1232 break; 1233 } 1234 } 1235 if (so->so_options & SO_DEBUG) 1236 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1237 1238 /* 1239 * Return any desired output. 1240 */ 1241 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1242 (void) tcp_output(tp); 1243 return; 1244 1245 dropafterack: 1246 /* 1247 * Generate an ACK dropping incoming segment if it occupies 1248 * sequence space, where the ACK reflects our state. 1249 */ 1250 if (tiflags & TH_RST) 1251 goto drop; 1252 m_freem(m); 1253 tp->t_flags |= TF_ACKNOW; 1254 (void) tcp_output(tp); 1255 return; 1256 1257 dropwithreset: 1258 /* 1259 * Generate a RST, dropping incoming segment. 1260 * Make ACK acceptable to originator of segment. 1261 * Don't bother to respond if destination was broadcast/multicast. 1262 */ 1263 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1264 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1265 goto drop; 1266 if (tiflags & TH_ACK) 1267 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1268 else { 1269 if (tiflags & TH_SYN) 1270 ti->ti_len++; 1271 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1272 TH_RST|TH_ACK); 1273 } 1274 /* destroy temporarily created socket */ 1275 if (dropsocket) 1276 (void) soabort(so); 1277 return; 1278 1279 drop: 1280 /* 1281 * Drop space held by incoming segment and return. 1282 */ 1283 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1284 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1285 m_freem(m); 1286 /* destroy temporarily created socket */ 1287 if (dropsocket) 1288 (void) soabort(so); 1289 return; 1290 #ifndef TUBA_INCLUDE 1291 } 1292 1293 void 1294 tcp_dooptions(tp, cp, cnt, ti, ts_present, ts_val, ts_ecr) 1295 struct tcpcb *tp; 1296 u_char *cp; 1297 int cnt; 1298 struct tcpiphdr *ti; 1299 int *ts_present; 1300 u_long *ts_val, *ts_ecr; 1301 { 1302 u_short mss; 1303 int opt, optlen; 1304 1305 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1306 opt = cp[0]; 1307 if (opt == TCPOPT_EOL) 1308 break; 1309 if (opt == TCPOPT_NOP) 1310 optlen = 1; 1311 else { 1312 optlen = cp[1]; 1313 if (optlen <= 0) 1314 break; 1315 } 1316 switch (opt) { 1317 1318 default: 1319 continue; 1320 1321 case TCPOPT_MAXSEG: 1322 if (optlen != TCPOLEN_MAXSEG) 1323 continue; 1324 if (!(ti->ti_flags & TH_SYN)) 1325 continue; 1326 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1327 NTOHS(mss); 1328 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1329 break; 1330 1331 case TCPOPT_WINDOW: 1332 if (optlen != TCPOLEN_WINDOW) 1333 continue; 1334 if (!(ti->ti_flags & TH_SYN)) 1335 continue; 1336 tp->t_flags |= TF_RCVD_SCALE; 1337 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1338 break; 1339 1340 case TCPOPT_TIMESTAMP: 1341 if (optlen != TCPOLEN_TIMESTAMP) 1342 continue; 1343 *ts_present = 1; 1344 bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val)); 1345 NTOHL(*ts_val); 1346 bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr)); 1347 NTOHL(*ts_ecr); 1348 1349 /* 1350 * A timestamp received in a SYN makes 1351 * it ok to send timestamp requests and replies. 1352 */ 1353 if (ti->ti_flags & TH_SYN) { 1354 tp->t_flags |= TF_RCVD_TSTMP; 1355 tp->ts_recent = *ts_val; 1356 tp->ts_recent_age = tcp_now; 1357 } 1358 break; 1359 } 1360 } 1361 } 1362 1363 /* 1364 * Pull out of band byte out of a segment so 1365 * it doesn't appear in the user's data queue. 1366 * It is still reflected in the segment length for 1367 * sequencing purposes. 1368 */ 1369 void 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 void 1401 tcp_xmit_timer(tp, rtt) 1402 register struct tcpcb *tp; 1403 short rtt; 1404 { 1405 register short delta; 1406 1407 tcpstat.tcps_rttupdated++; 1408 if (tp->t_srtt != 0) { 1409 /* 1410 * srtt is stored as fixed point with 3 bits after the 1411 * binary point (i.e., scaled by 8). The following magic 1412 * is equivalent to the smoothing algorithm in rfc793 with 1413 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1414 * point). Adjust rtt to origin 0. 1415 */ 1416 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1417 if ((tp->t_srtt += delta) <= 0) 1418 tp->t_srtt = 1; 1419 /* 1420 * We accumulate a smoothed rtt variance (actually, a 1421 * smoothed mean difference), then set the retransmit 1422 * timer to smoothed rtt + 4 times the smoothed variance. 1423 * rttvar is stored as fixed point with 2 bits after the 1424 * binary point (scaled by 4). The following is 1425 * equivalent to rfc793 smoothing with an alpha of .75 1426 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1427 * rfc793's wired-in beta. 1428 */ 1429 if (delta < 0) 1430 delta = -delta; 1431 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1432 if ((tp->t_rttvar += delta) <= 0) 1433 tp->t_rttvar = 1; 1434 } else { 1435 /* 1436 * No rtt measurement yet - use the unsmoothed rtt. 1437 * Set the variance to half the rtt (so our first 1438 * retransmit happens at 3*rtt). 1439 */ 1440 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1441 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1442 } 1443 tp->t_rtt = 0; 1444 tp->t_rxtshift = 0; 1445 1446 /* 1447 * the retransmit should happen at rtt + 4 * rttvar. 1448 * Because of the way we do the smoothing, srtt and rttvar 1449 * will each average +1/2 tick of bias. When we compute 1450 * the retransmit timer, we want 1/2 tick of rounding and 1451 * 1 extra tick because of +-1/2 tick uncertainty in the 1452 * firing of the timer. The bias will give us exactly the 1453 * 1.5 tick we need. But, because the bias is 1454 * statistical, we have to test that we don't drop below 1455 * the minimum feasible timer (which is 2 ticks). 1456 */ 1457 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1458 tp->t_rttmin, TCPTV_REXMTMAX); 1459 1460 /* 1461 * We received an ack for a packet that wasn't retransmitted; 1462 * it is probably safe to discard any error indications we've 1463 * received recently. This isn't quite right, but close enough 1464 * for now (a route might have failed after we sent a segment, 1465 * and the return path might not be symmetrical). 1466 */ 1467 tp->t_softerror = 0; 1468 } 1469 1470 /* 1471 * Determine a reasonable value for maxseg size. 1472 * If the route is known, check route for mtu. 1473 * If none, use an mss that can be handled on the outgoing 1474 * interface without forcing IP to fragment; if bigger than 1475 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1476 * to utilize large mbufs. If no route is found, route has no mtu, 1477 * or the destination isn't local, use a default, hopefully conservative 1478 * size (usually 512 or the default IP max size, but no more than the mtu 1479 * of the interface), as we can't discover anything about intervening 1480 * gateways or networks. We also initialize the congestion/slow start 1481 * window to be a single segment if the destination isn't local. 1482 * While looking at the routing entry, we also initialize other path-dependent 1483 * parameters from pre-set or cached values in the routing entry. 1484 */ 1485 int 1486 tcp_mss(tp, offer) 1487 register struct tcpcb *tp; 1488 u_int offer; 1489 { 1490 struct route *ro; 1491 register struct rtentry *rt; 1492 struct ifnet *ifp; 1493 register int rtt, mss; 1494 u_long bufsize; 1495 struct inpcb *inp; 1496 struct socket *so; 1497 extern int tcp_mssdflt; 1498 1499 inp = tp->t_inpcb; 1500 ro = &inp->inp_route; 1501 1502 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1503 /* No route yet, so try to acquire one */ 1504 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1505 ro->ro_dst.sa_family = AF_INET; 1506 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1507 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1508 inp->inp_faddr; 1509 rtalloc(ro); 1510 } 1511 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1512 return (tcp_mssdflt); 1513 } 1514 ifp = rt->rt_ifp; 1515 so = inp->inp_socket; 1516 1517 #ifdef RTV_MTU /* if route characteristics exist ... */ 1518 /* 1519 * While we're here, check if there's an initial rtt 1520 * or rttvar. Convert from the route-table units 1521 * to scaled multiples of the slow timeout timer. 1522 */ 1523 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1524 /* 1525 * XXX the lock bit for MTU indicates that the value 1526 * is also a minimum value; this is subject to time. 1527 */ 1528 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1529 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1530 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1531 if (rt->rt_rmx.rmx_rttvar) 1532 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1533 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1534 else 1535 /* default variation is +- 1 rtt */ 1536 tp->t_rttvar = 1537 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1538 TCPT_RANGESET(tp->t_rxtcur, 1539 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1540 tp->t_rttmin, TCPTV_REXMTMAX); 1541 } 1542 /* 1543 * if there's an mtu associated with the route, use it 1544 */ 1545 if (rt->rt_rmx.rmx_mtu) 1546 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1547 else 1548 #endif /* RTV_MTU */ 1549 { 1550 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1551 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1552 if (mss > MCLBYTES) 1553 mss &= ~(MCLBYTES-1); 1554 #else 1555 if (mss > MCLBYTES) 1556 mss = mss / MCLBYTES * MCLBYTES; 1557 #endif 1558 if (!in_localaddr(inp->inp_faddr)) 1559 mss = min(mss, tcp_mssdflt); 1560 } 1561 /* 1562 * The current mss, t_maxseg, is initialized to the default value. 1563 * If we compute a smaller value, reduce the current mss. 1564 * If we compute a larger value, return it for use in sending 1565 * a max seg size option, but don't store it for use 1566 * unless we received an offer at least that large from peer. 1567 * However, do not accept offers under 32 bytes. 1568 */ 1569 if (offer) 1570 mss = min(mss, offer); 1571 mss = max(mss, 32); /* sanity */ 1572 if (mss < tp->t_maxseg || offer != 0) { 1573 /* 1574 * If there's a pipesize, change the socket buffer 1575 * to that size. Make the socket buffers an integral 1576 * number of mss units; if the mss is larger than 1577 * the socket buffer, decrease the mss. 1578 */ 1579 #ifdef RTV_SPIPE 1580 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1581 #endif 1582 bufsize = so->so_snd.sb_hiwat; 1583 if (bufsize < mss) 1584 mss = bufsize; 1585 else { 1586 bufsize = roundup(bufsize, mss); 1587 if (bufsize > sb_max) 1588 bufsize = sb_max; 1589 (void)sbreserve(&so->so_snd, bufsize); 1590 } 1591 tp->t_maxseg = mss; 1592 1593 #ifdef RTV_RPIPE 1594 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1595 #endif 1596 bufsize = so->so_rcv.sb_hiwat; 1597 if (bufsize > mss) { 1598 bufsize = roundup(bufsize, mss); 1599 if (bufsize > sb_max) 1600 bufsize = sb_max; 1601 (void)sbreserve(&so->so_rcv, bufsize); 1602 } 1603 } 1604 tp->snd_cwnd = mss; 1605 1606 #ifdef RTV_SSTHRESH 1607 if (rt->rt_rmx.rmx_ssthresh) { 1608 /* 1609 * There's some sort of gateway or interface 1610 * buffer limit on the path. Use this to set 1611 * the slow start threshhold, but set the 1612 * threshold to no less than 2*mss. 1613 */ 1614 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1615 } 1616 #endif /* RTV_MTU */ 1617 return (mss); 1618 } 1619 #endif /* TUBA_INCLUDE */ 1620