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 8.9 (Berkeley) 05/01/95 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 if ((ti)->ti_flags & TH_PUSH) \ 65 tp->t_flags |= TF_ACKNOW; \ 66 else \ 67 tp->t_flags |= TF_DELACK; \ 68 (tp)->rcv_nxt += (ti)->ti_len; \ 69 flags = (ti)->ti_flags & TH_FIN; \ 70 tcpstat.tcps_rcvpack++;\ 71 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 72 sbappend(&(so)->so_rcv, (m)); \ 73 sorwakeup(so); \ 74 } else { \ 75 (flags) = tcp_reass((tp), (ti), (m)); \ 76 tp->t_flags |= TF_ACKNOW; \ 77 } \ 78 } 79 #ifndef TUBA_INCLUDE 80 81 int 82 tcp_reass(tp, ti, m) 83 register struct tcpcb *tp; 84 register struct tcpiphdr *ti; 85 struct mbuf *m; 86 { 87 register struct tcpiphdr *q; 88 struct socket *so = tp->t_inpcb->inp_socket; 89 int flags; 90 91 /* 92 * Call with ti==0 after become established to 93 * force pre-ESTABLISHED data up to user socket. 94 */ 95 if (ti == 0) 96 goto present; 97 98 /* 99 * Find a segment which begins after this one does. 100 */ 101 for (q = tp->seg_next; q != (struct tcpiphdr *)tp; 102 q = (struct tcpiphdr *)q->ti_next) 103 if (SEQ_GT(q->ti_seq, ti->ti_seq)) 104 break; 105 106 /* 107 * If there is a preceding segment, it may provide some of 108 * our data already. If so, drop the data from the incoming 109 * segment. If it provides all of our data, drop us. 110 */ 111 if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) { 112 register int i; 113 q = (struct tcpiphdr *)q->ti_prev; 114 /* conversion to int (in i) handles seq wraparound */ 115 i = q->ti_seq + q->ti_len - ti->ti_seq; 116 if (i > 0) { 117 if (i >= ti->ti_len) { 118 tcpstat.tcps_rcvduppack++; 119 tcpstat.tcps_rcvdupbyte += ti->ti_len; 120 m_freem(m); 121 return (0); 122 } 123 m_adj(m, i); 124 ti->ti_len -= i; 125 ti->ti_seq += i; 126 } 127 q = (struct tcpiphdr *)(q->ti_next); 128 } 129 tcpstat.tcps_rcvoopack++; 130 tcpstat.tcps_rcvoobyte += ti->ti_len; 131 REASS_MBUF(ti) = m; /* XXX */ 132 133 /* 134 * While we overlap succeeding segments trim them or, 135 * if they are completely covered, dequeue them. 136 */ 137 while (q != (struct tcpiphdr *)tp) { 138 register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq; 139 if (i <= 0) 140 break; 141 if (i < q->ti_len) { 142 q->ti_seq += i; 143 q->ti_len -= i; 144 m_adj(REASS_MBUF(q), i); 145 break; 146 } 147 q = (struct tcpiphdr *)q->ti_next; 148 m = REASS_MBUF((struct tcpiphdr *)q->ti_prev); 149 remque(q->ti_prev); 150 m_freem(m); 151 } 152 153 /* 154 * Stick new segment in its place. 155 */ 156 insque(ti, q->ti_prev); 157 158 present: 159 /* 160 * Present data to user, advancing rcv_nxt through 161 * completed sequence space. 162 */ 163 if (TCPS_HAVERCVDSYN(tp->t_state) == 0) 164 return (0); 165 ti = tp->seg_next; 166 if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt) 167 return (0); 168 if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len) 169 return (0); 170 do { 171 tp->rcv_nxt += ti->ti_len; 172 flags = ti->ti_flags & TH_FIN; 173 remque(ti); 174 m = REASS_MBUF(ti); 175 ti = (struct tcpiphdr *)ti->ti_next; 176 if (so->so_state & SS_CANTRCVMORE) 177 m_freem(m); 178 else 179 sbappend(&so->so_rcv, m); 180 } while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt); 181 sorwakeup(so); 182 return (flags); 183 } 184 185 /* 186 * TCP input routine, follows pages 65-76 of the 187 * protocol specification dated September, 1981 very closely. 188 */ 189 void 190 tcp_input(m, iphlen) 191 register struct mbuf *m; 192 int iphlen; 193 { 194 register struct tcpiphdr *ti; 195 register struct inpcb *inp; 196 u_char *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, u_char *) + 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 ++tcpstat.tcps_pcbcachemiss; 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 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 334 /* 335 * Note: dropwithreset makes sure we don't 336 * send a reset in response to a RST. 337 */ 338 if (tiflags & (TH_ACK|TH_RST)) 339 goto dropwithreset; 340 goto drop; 341 } 342 so = sonewconn(so, 0); 343 if (so == 0) 344 goto drop; 345 /* 346 * This is ugly, but .... 347 * 348 * Mark socket as temporary until we're 349 * committed to keeping it. The code at 350 * ``drop'' and ``dropwithreset'' check the 351 * flag dropsocket to see if the temporary 352 * socket created here should be discarded. 353 * We mark the socket as discardable until 354 * we're committed to it below in TCPS_LISTEN. 355 */ 356 dropsocket++; 357 inp = (struct inpcb *)so->so_pcb; 358 inp->inp_laddr = ti->ti_dst; 359 inp->inp_lport = ti->ti_dport; 360 #if BSD>=43 361 inp->inp_options = ip_srcroute(); 362 #endif 363 tp = intotcpcb(inp); 364 tp->t_state = TCPS_LISTEN; 365 366 /* Compute proper scaling value from buffer space 367 */ 368 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 369 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 370 tp->request_r_scale++; 371 } 372 } 373 374 /* 375 * Segment received on connection. 376 * Reset idle time and keep-alive timer. 377 */ 378 tp->t_idle = 0; 379 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 380 381 /* 382 * Process options if not in LISTEN state, 383 * else do it below (after getting remote address). 384 */ 385 if (optp && tp->t_state != TCPS_LISTEN) 386 tcp_dooptions(tp, optp, optlen, ti, 387 &ts_present, &ts_val, &ts_ecr); 388 389 /* 390 * Header prediction: check for the two common cases 391 * of a uni-directional data xfer. If the packet has 392 * no control flags, is in-sequence, the window didn't 393 * change and we're not retransmitting, it's a 394 * candidate. If the length is zero and the ack moved 395 * forward, we're the sender side of the xfer. Just 396 * free the data acked & wake any higher level process 397 * that was blocked waiting for space. If the length 398 * is non-zero and the ack didn't move, we're the 399 * receiver side. If we're getting packets in-order 400 * (the reassembly queue is empty), add the data to 401 * the socket buffer and note that we need a delayed ack. 402 */ 403 if (tp->t_state == TCPS_ESTABLISHED && 404 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 405 (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && 406 ti->ti_seq == tp->rcv_nxt && 407 tiwin && tiwin == tp->snd_wnd && 408 tp->snd_nxt == tp->snd_max) { 409 410 /* 411 * If last ACK falls within this segment's sequence numbers, 412 * record the timestamp. 413 */ 414 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 415 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) { 416 tp->ts_recent_age = tcp_now; 417 tp->ts_recent = ts_val; 418 } 419 420 if (ti->ti_len == 0) { 421 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 422 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 423 tp->snd_cwnd >= tp->snd_wnd) { 424 /* 425 * this is a pure ack for outstanding data. 426 */ 427 ++tcpstat.tcps_predack; 428 if (ts_present) 429 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 430 else if (tp->t_rtt && 431 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 432 tcp_xmit_timer(tp, tp->t_rtt); 433 acked = ti->ti_ack - tp->snd_una; 434 tcpstat.tcps_rcvackpack++; 435 tcpstat.tcps_rcvackbyte += acked; 436 sbdrop(&so->so_snd, acked); 437 tp->snd_una = ti->ti_ack; 438 m_freem(m); 439 440 /* 441 * If all outstanding data are acked, stop 442 * retransmit timer, otherwise restart timer 443 * using current (possibly backed-off) value. 444 * If process is waiting for space, 445 * wakeup/selwakeup/signal. If data 446 * are ready to send, let tcp_output 447 * decide between more output or persist. 448 */ 449 if (tp->snd_una == tp->snd_max) 450 tp->t_timer[TCPT_REXMT] = 0; 451 else if (tp->t_timer[TCPT_PERSIST] == 0) 452 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 453 454 if (so->so_snd.sb_flags & SB_NOTIFY) 455 sowwakeup(so); 456 if (so->so_snd.sb_cc) 457 (void) tcp_output(tp); 458 return; 459 } 460 } else if (ti->ti_ack == tp->snd_una && 461 tp->seg_next == (struct tcpiphdr *)tp && 462 ti->ti_len <= sbspace(&so->so_rcv)) { 463 /* 464 * this is a pure, in-sequence data packet 465 * with nothing on the reassembly queue and 466 * we have enough buffer space to take it. 467 */ 468 ++tcpstat.tcps_preddat; 469 tp->rcv_nxt += ti->ti_len; 470 tcpstat.tcps_rcvpack++; 471 tcpstat.tcps_rcvbyte += ti->ti_len; 472 /* 473 * Drop TCP, IP headers and TCP options then add data 474 * to socket buffer. 475 */ 476 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 477 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 478 sbappend(&so->so_rcv, m); 479 sorwakeup(so); 480 if (ti->ti_flags & TH_PUSH) 481 tp->t_flags |= TF_ACKNOW; 482 else 483 tp->t_flags |= TF_DELACK; 484 return; 485 } 486 } 487 488 /* 489 * Drop TCP, IP headers and TCP options. 490 */ 491 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 492 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 493 494 /* 495 * Calculate amount of space in receive window, 496 * and then do TCP input processing. 497 * Receive window is amount of space in rcv queue, 498 * but not less than advertised window. 499 */ 500 { int win; 501 502 win = sbspace(&so->so_rcv); 503 if (win < 0) 504 win = 0; 505 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 506 } 507 508 switch (tp->t_state) { 509 510 /* 511 * If the state is LISTEN then ignore segment if it contains an RST. 512 * If the segment contains an ACK then it is bad and send a RST. 513 * If it does not contain a SYN then it is not interesting; drop it. 514 * Don't bother responding if the destination was a broadcast. 515 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 516 * tp->iss, and send a segment: 517 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 518 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 519 * Fill in remote peer address fields if not previously specified. 520 * Enter SYN_RECEIVED state, and process any other fields of this 521 * segment in this state. 522 */ 523 case TCPS_LISTEN: { 524 struct mbuf *am; 525 register struct sockaddr_in *sin; 526 527 if (tiflags & TH_RST) 528 goto drop; 529 if (tiflags & TH_ACK) 530 goto dropwithreset; 531 if ((tiflags & TH_SYN) == 0) 532 goto drop; 533 /* 534 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 535 * in_broadcast() should never return true on a received 536 * packet with M_BCAST not set. 537 */ 538 if (m->m_flags & (M_BCAST|M_MCAST) || 539 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 540 goto drop; 541 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 542 if (am == NULL) 543 goto drop; 544 am->m_len = sizeof (struct sockaddr_in); 545 sin = mtod(am, struct sockaddr_in *); 546 sin->sin_family = AF_INET; 547 sin->sin_len = sizeof(*sin); 548 sin->sin_addr = ti->ti_src; 549 sin->sin_port = ti->ti_sport; 550 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 551 laddr = inp->inp_laddr; 552 if (inp->inp_laddr.s_addr == INADDR_ANY) 553 inp->inp_laddr = ti->ti_dst; 554 if (in_pcbconnect(inp, am)) { 555 inp->inp_laddr = laddr; 556 (void) m_free(am); 557 goto drop; 558 } 559 (void) m_free(am); 560 tp->t_template = tcp_template(tp); 561 if (tp->t_template == 0) { 562 tp = tcp_drop(tp, ENOBUFS); 563 dropsocket = 0; /* socket is already gone */ 564 goto drop; 565 } 566 if (optp) 567 tcp_dooptions(tp, optp, optlen, ti, 568 &ts_present, &ts_val, &ts_ecr); 569 if (iss) 570 tp->iss = iss; 571 else 572 tp->iss = tcp_iss; 573 tcp_iss += TCP_ISSINCR/2; 574 tp->irs = ti->ti_seq; 575 tcp_sendseqinit(tp); 576 tcp_rcvseqinit(tp); 577 tp->t_flags |= TF_ACKNOW; 578 tp->t_state = TCPS_SYN_RECEIVED; 579 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 580 dropsocket = 0; /* committed to socket */ 581 tcpstat.tcps_accepts++; 582 goto trimthenstep6; 583 } 584 585 /* 586 * If the state is SYN_SENT: 587 * if seg contains an ACK, but not for our SYN, drop the input. 588 * if seg contains a RST, then drop the connection. 589 * if seg does not contain SYN, then drop it. 590 * Otherwise this is an acceptable SYN segment 591 * initialize tp->rcv_nxt and tp->irs 592 * if seg contains ack then advance tp->snd_una 593 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 594 * arrange for segment to be acked (eventually) 595 * continue processing rest of data/controls, beginning with URG 596 */ 597 case TCPS_SYN_SENT: 598 if ((tiflags & TH_ACK) && 599 (SEQ_LEQ(ti->ti_ack, tp->iss) || 600 SEQ_GT(ti->ti_ack, tp->snd_max))) 601 goto dropwithreset; 602 if (tiflags & TH_RST) { 603 if (tiflags & TH_ACK) 604 tp = tcp_drop(tp, ECONNREFUSED); 605 goto drop; 606 } 607 if ((tiflags & TH_SYN) == 0) 608 goto drop; 609 if (tiflags & TH_ACK) { 610 tp->snd_una = ti->ti_ack; 611 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 612 tp->snd_nxt = tp->snd_una; 613 } 614 tp->t_timer[TCPT_REXMT] = 0; 615 tp->irs = ti->ti_seq; 616 tcp_rcvseqinit(tp); 617 tp->t_flags |= TF_ACKNOW; 618 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 619 tcpstat.tcps_connects++; 620 soisconnected(so); 621 tp->t_state = TCPS_ESTABLISHED; 622 /* Do window scaling on this connection? */ 623 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 624 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 625 tp->snd_scale = tp->requested_s_scale; 626 tp->rcv_scale = tp->request_r_scale; 627 } 628 (void) tcp_reass(tp, (struct tcpiphdr *)0, 629 (struct mbuf *)0); 630 /* 631 * if we didn't have to retransmit the SYN, 632 * use its rtt as our initial srtt & rtt var. 633 */ 634 if (tp->t_rtt) 635 tcp_xmit_timer(tp, tp->t_rtt); 636 } else 637 tp->t_state = TCPS_SYN_RECEIVED; 638 639 trimthenstep6: 640 /* 641 * Advance ti->ti_seq to correspond to first data byte. 642 * If data, trim to stay within window, 643 * dropping FIN if necessary. 644 */ 645 ti->ti_seq++; 646 if (ti->ti_len > tp->rcv_wnd) { 647 todrop = ti->ti_len - tp->rcv_wnd; 648 m_adj(m, -todrop); 649 ti->ti_len = tp->rcv_wnd; 650 tiflags &= ~TH_FIN; 651 tcpstat.tcps_rcvpackafterwin++; 652 tcpstat.tcps_rcvbyteafterwin += todrop; 653 } 654 tp->snd_wl1 = ti->ti_seq - 1; 655 tp->rcv_up = ti->ti_seq; 656 goto step6; 657 } 658 659 /* 660 * States other than LISTEN or SYN_SENT. 661 * First check timestamp, if present. 662 * Then check that at least some bytes of segment are within 663 * receive window. If segment begins before rcv_nxt, 664 * drop leading data (and SYN); if nothing left, just ack. 665 * 666 * RFC 1323 PAWS: If we have a timestamp reply on this segment 667 * and it's less than ts_recent, drop it. 668 */ 669 if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 670 TSTMP_LT(ts_val, tp->ts_recent)) { 671 672 /* Check to see if ts_recent is over 24 days old. */ 673 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 674 /* 675 * Invalidate ts_recent. If this segment updates 676 * ts_recent, the age will be reset later and ts_recent 677 * will get a valid value. If it does not, setting 678 * ts_recent to zero will at least satisfy the 679 * requirement that zero be placed in the timestamp 680 * echo reply when ts_recent isn't valid. The 681 * age isn't reset until we get a valid ts_recent 682 * because we don't want out-of-order segments to be 683 * dropped when ts_recent is old. 684 */ 685 tp->ts_recent = 0; 686 } else { 687 tcpstat.tcps_rcvduppack++; 688 tcpstat.tcps_rcvdupbyte += ti->ti_len; 689 tcpstat.tcps_pawsdrop++; 690 goto dropafterack; 691 } 692 } 693 694 todrop = tp->rcv_nxt - ti->ti_seq; 695 if (todrop > 0) { 696 if (tiflags & TH_SYN) { 697 tiflags &= ~TH_SYN; 698 ti->ti_seq++; 699 if (ti->ti_urp > 1) 700 ti->ti_urp--; 701 else 702 tiflags &= ~TH_URG; 703 todrop--; 704 } 705 if (todrop >= ti->ti_len) { 706 tcpstat.tcps_rcvduppack++; 707 tcpstat.tcps_rcvdupbyte += ti->ti_len; 708 /* 709 * If segment is just one to the left of the window, 710 * check two special cases: 711 * 1. Don't toss RST in response to 4.2-style keepalive. 712 * 2. If the only thing to drop is a FIN, we can drop 713 * it, but check the ACK or we will get into FIN 714 * wars if our FINs crossed (both CLOSING). 715 * In either case, send ACK to resynchronize, 716 * but keep on processing for RST or ACK. 717 */ 718 if ((tiflags & TH_FIN && todrop == ti->ti_len + 1) 719 #ifdef TCP_COMPAT_42 720 || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1) 721 #endif 722 ) { 723 todrop = ti->ti_len; 724 tiflags &= ~TH_FIN; 725 } else { 726 /* 727 * Handle the case when a bound socket connects 728 * to itself. Allow packets with a SYN and 729 * an ACK to continue with the processing. 730 */ 731 if (todrop != 0 || (tiflags & TH_ACK) == 0) 732 goto dropafterack; 733 } 734 tp->t_flags |= TF_ACKNOW; 735 } else { 736 tcpstat.tcps_rcvpartduppack++; 737 tcpstat.tcps_rcvpartdupbyte += todrop; 738 } 739 m_adj(m, todrop); 740 ti->ti_seq += todrop; 741 ti->ti_len -= todrop; 742 if (ti->ti_urp > todrop) 743 ti->ti_urp -= todrop; 744 else { 745 tiflags &= ~TH_URG; 746 ti->ti_urp = 0; 747 } 748 } 749 750 /* 751 * If new data are received on a connection after the 752 * user processes are gone, then RST the other end. 753 */ 754 if ((so->so_state & SS_NOFDREF) && 755 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 756 tp = tcp_close(tp); 757 tcpstat.tcps_rcvafterclose++; 758 goto dropwithreset; 759 } 760 761 /* 762 * If segment ends after window, drop trailing data 763 * (and PUSH and FIN); if nothing left, just ACK. 764 */ 765 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 766 if (todrop > 0) { 767 tcpstat.tcps_rcvpackafterwin++; 768 if (todrop >= ti->ti_len) { 769 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 770 /* 771 * If a new connection request is received 772 * while in TIME_WAIT, drop the old connection 773 * and start over if the sequence numbers 774 * are above the previous ones. 775 */ 776 if (tiflags & TH_SYN && 777 tp->t_state == TCPS_TIME_WAIT && 778 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 779 iss = tp->rcv_nxt + TCP_ISSINCR; 780 tp = tcp_close(tp); 781 goto findpcb; 782 } 783 /* 784 * If window is closed can only take segments at 785 * window edge, and have to drop data and PUSH from 786 * incoming segments. Continue processing, but 787 * remember to ack. Otherwise, drop segment 788 * and ack. 789 */ 790 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 791 tp->t_flags |= TF_ACKNOW; 792 tcpstat.tcps_rcvwinprobe++; 793 } else 794 goto dropafterack; 795 } else 796 tcpstat.tcps_rcvbyteafterwin += todrop; 797 m_adj(m, -todrop); 798 ti->ti_len -= todrop; 799 tiflags &= ~(TH_PUSH|TH_FIN); 800 } 801 802 /* 803 * If last ACK falls within this segment's sequence numbers, 804 * record its timestamp. 805 */ 806 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 807 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + 808 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 809 tp->ts_recent_age = tcp_now; 810 tp->ts_recent = ts_val; 811 } 812 813 /* 814 * If the RST bit is set examine the state: 815 * SYN_RECEIVED STATE: 816 * If passive open, return to LISTEN state. 817 * If active open, inform user that connection was refused. 818 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 819 * Inform user that connection was reset, and close tcb. 820 * CLOSING, LAST_ACK, TIME_WAIT STATES 821 * Close the tcb. 822 */ 823 if (tiflags&TH_RST) switch (tp->t_state) { 824 825 case TCPS_SYN_RECEIVED: 826 so->so_error = ECONNREFUSED; 827 goto close; 828 829 case TCPS_ESTABLISHED: 830 case TCPS_FIN_WAIT_1: 831 case TCPS_FIN_WAIT_2: 832 case TCPS_CLOSE_WAIT: 833 so->so_error = ECONNRESET; 834 close: 835 tp->t_state = TCPS_CLOSED; 836 tcpstat.tcps_drops++; 837 tp = tcp_close(tp); 838 goto drop; 839 840 case TCPS_CLOSING: 841 case TCPS_LAST_ACK: 842 case TCPS_TIME_WAIT: 843 tp = tcp_close(tp); 844 goto drop; 845 } 846 847 /* 848 * If a SYN is in the window, then this is an 849 * error and we send an RST and drop the connection. 850 */ 851 if (tiflags & TH_SYN) { 852 tp = tcp_drop(tp, ECONNRESET); 853 goto dropwithreset; 854 } 855 856 /* 857 * If the ACK bit is off we drop the segment and return. 858 */ 859 if ((tiflags & TH_ACK) == 0) 860 goto drop; 861 862 /* 863 * Ack processing. 864 */ 865 switch (tp->t_state) { 866 867 /* 868 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 869 * ESTABLISHED state and continue processing, otherwise 870 * send an RST. 871 */ 872 case TCPS_SYN_RECEIVED: 873 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 874 SEQ_GT(ti->ti_ack, tp->snd_max)) 875 goto dropwithreset; 876 tcpstat.tcps_connects++; 877 soisconnected(so); 878 tp->t_state = TCPS_ESTABLISHED; 879 /* Do window scaling? */ 880 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 881 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 882 tp->snd_scale = tp->requested_s_scale; 883 tp->rcv_scale = tp->request_r_scale; 884 } 885 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 886 tp->snd_wl1 = ti->ti_seq - 1; 887 /* fall into ... */ 888 889 /* 890 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 891 * ACKs. If the ack is in the range 892 * tp->snd_una < ti->ti_ack <= tp->snd_max 893 * then advance tp->snd_una to ti->ti_ack and drop 894 * data from the retransmission queue. If this ACK reflects 895 * more up to date window information we update our window information. 896 */ 897 case TCPS_ESTABLISHED: 898 case TCPS_FIN_WAIT_1: 899 case TCPS_FIN_WAIT_2: 900 case TCPS_CLOSE_WAIT: 901 case TCPS_CLOSING: 902 case TCPS_LAST_ACK: 903 case TCPS_TIME_WAIT: 904 905 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 906 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 907 tcpstat.tcps_rcvdupack++; 908 /* 909 * If we have outstanding data (other than 910 * a window probe), this is a completely 911 * duplicate ack (ie, window info didn't 912 * change), the ack is the biggest we've 913 * seen and we've seen exactly our rexmt 914 * threshhold of them, assume a packet 915 * has been dropped and retransmit it. 916 * Kludge snd_nxt & the congestion 917 * window so we send only this one 918 * packet. 919 * 920 * We know we're losing at the current 921 * window size so do congestion avoidance 922 * (set ssthresh to half the current window 923 * and pull our congestion window back to 924 * the new ssthresh). 925 * 926 * Dup acks mean that packets have left the 927 * network (they're now cached at the receiver) 928 * so bump cwnd by the amount in the receiver 929 * to keep a constant cwnd packets in the 930 * network. 931 */ 932 if (tp->t_timer[TCPT_REXMT] == 0 || 933 ti->ti_ack != tp->snd_una) 934 tp->t_dupacks = 0; 935 else if (++tp->t_dupacks == tcprexmtthresh) { 936 tcp_seq onxt = tp->snd_nxt; 937 u_int win = 938 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 939 tp->t_maxseg; 940 941 if (win < 2) 942 win = 2; 943 tp->snd_ssthresh = win * tp->t_maxseg; 944 tp->t_timer[TCPT_REXMT] = 0; 945 tp->t_rtt = 0; 946 tp->snd_nxt = ti->ti_ack; 947 tp->snd_cwnd = tp->t_maxseg; 948 (void) tcp_output(tp); 949 tp->snd_cwnd = tp->snd_ssthresh + 950 tp->t_maxseg * tp->t_dupacks; 951 if (SEQ_GT(onxt, tp->snd_nxt)) 952 tp->snd_nxt = onxt; 953 goto drop; 954 } else if (tp->t_dupacks > tcprexmtthresh) { 955 tp->snd_cwnd += tp->t_maxseg; 956 (void) tcp_output(tp); 957 goto drop; 958 } 959 } else 960 tp->t_dupacks = 0; 961 break; 962 } 963 /* 964 * If the congestion window was inflated to account 965 * for the other side's cached packets, retract it. 966 */ 967 if (tp->t_dupacks > tcprexmtthresh && 968 tp->snd_cwnd > tp->snd_ssthresh) 969 tp->snd_cwnd = tp->snd_ssthresh; 970 tp->t_dupacks = 0; 971 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 972 tcpstat.tcps_rcvacktoomuch++; 973 goto dropafterack; 974 } 975 acked = ti->ti_ack - tp->snd_una; 976 tcpstat.tcps_rcvackpack++; 977 tcpstat.tcps_rcvackbyte += acked; 978 979 /* 980 * If we have a timestamp reply, update smoothed 981 * round trip time. If no timestamp is present but 982 * transmit timer is running and timed sequence 983 * number was acked, update smoothed round trip time. 984 * Since we now have an rtt measurement, cancel the 985 * timer backoff (cf., Phil Karn's retransmit alg.). 986 * Recompute the initial retransmit timer. 987 */ 988 if (ts_present) 989 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 990 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 991 tcp_xmit_timer(tp,tp->t_rtt); 992 993 /* 994 * If all outstanding data is acked, stop retransmit 995 * timer and remember to restart (more output or persist). 996 * If there is more data to be acked, restart retransmit 997 * timer, using current (possibly backed-off) value. 998 */ 999 if (ti->ti_ack == tp->snd_max) { 1000 tp->t_timer[TCPT_REXMT] = 0; 1001 needoutput = 1; 1002 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1003 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1004 /* 1005 * When new data is acked, open the congestion window. 1006 * If the window gives us less than ssthresh packets 1007 * in flight, open exponentially (maxseg per packet). 1008 * Otherwise open linearly: maxseg per window 1009 * (maxseg * (maxseg / cwnd) per packet). 1010 */ 1011 { 1012 register u_int cw = tp->snd_cwnd; 1013 register u_int incr = tp->t_maxseg; 1014 1015 if (cw > tp->snd_ssthresh) 1016 incr = incr * incr / cw; 1017 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1018 } 1019 if (acked > so->so_snd.sb_cc) { 1020 tp->snd_wnd -= so->so_snd.sb_cc; 1021 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1022 ourfinisacked = 1; 1023 } else { 1024 sbdrop(&so->so_snd, acked); 1025 tp->snd_wnd -= acked; 1026 ourfinisacked = 0; 1027 } 1028 if (so->so_snd.sb_flags & SB_NOTIFY) 1029 sowwakeup(so); 1030 tp->snd_una = ti->ti_ack; 1031 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1032 tp->snd_nxt = tp->snd_una; 1033 1034 switch (tp->t_state) { 1035 1036 /* 1037 * In FIN_WAIT_1 STATE in addition to the processing 1038 * for the ESTABLISHED state if our FIN is now acknowledged 1039 * then enter FIN_WAIT_2. 1040 */ 1041 case TCPS_FIN_WAIT_1: 1042 if (ourfinisacked) { 1043 /* 1044 * If we can't receive any more 1045 * data, then closing user can proceed. 1046 * Starting the timer is contrary to the 1047 * specification, but if we don't get a FIN 1048 * we'll hang forever. 1049 */ 1050 if (so->so_state & SS_CANTRCVMORE) { 1051 soisdisconnected(so); 1052 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1053 } 1054 tp->t_state = TCPS_FIN_WAIT_2; 1055 } 1056 break; 1057 1058 /* 1059 * In CLOSING STATE in addition to the processing for 1060 * the ESTABLISHED state if the ACK acknowledges our FIN 1061 * then enter the TIME-WAIT state, otherwise ignore 1062 * the segment. 1063 */ 1064 case TCPS_CLOSING: 1065 if (ourfinisacked) { 1066 tp->t_state = TCPS_TIME_WAIT; 1067 tcp_canceltimers(tp); 1068 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1069 soisdisconnected(so); 1070 } 1071 break; 1072 1073 /* 1074 * In LAST_ACK, we may still be waiting for data to drain 1075 * and/or to be acked, as well as for the ack of our FIN. 1076 * If our FIN is now acknowledged, delete the TCB, 1077 * enter the closed state and return. 1078 */ 1079 case TCPS_LAST_ACK: 1080 if (ourfinisacked) { 1081 tp = tcp_close(tp); 1082 goto drop; 1083 } 1084 break; 1085 1086 /* 1087 * In TIME_WAIT state the only thing that should arrive 1088 * is a retransmission of the remote FIN. Acknowledge 1089 * it and restart the finack timer. 1090 */ 1091 case TCPS_TIME_WAIT: 1092 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1093 goto dropafterack; 1094 } 1095 } 1096 1097 step6: 1098 /* 1099 * Update window information. 1100 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1101 */ 1102 if ((tiflags & TH_ACK) && 1103 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 1104 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1105 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))) { 1106 /* keep track of pure window updates */ 1107 if (ti->ti_len == 0 && 1108 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1109 tcpstat.tcps_rcvwinupd++; 1110 tp->snd_wnd = tiwin; 1111 tp->snd_wl1 = ti->ti_seq; 1112 tp->snd_wl2 = ti->ti_ack; 1113 if (tp->snd_wnd > tp->max_sndwnd) 1114 tp->max_sndwnd = tp->snd_wnd; 1115 needoutput = 1; 1116 } 1117 1118 /* 1119 * Process segments with URG. 1120 */ 1121 if ((tiflags & TH_URG) && ti->ti_urp && 1122 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1123 /* 1124 * This is a kludge, but if we receive and accept 1125 * random urgent pointers, we'll crash in 1126 * soreceive. It's hard to imagine someone 1127 * actually wanting to send this much urgent data. 1128 */ 1129 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1130 ti->ti_urp = 0; /* XXX */ 1131 tiflags &= ~TH_URG; /* XXX */ 1132 goto dodata; /* XXX */ 1133 } 1134 /* 1135 * If this segment advances the known urgent pointer, 1136 * then mark the data stream. This should not happen 1137 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1138 * a FIN has been received from the remote side. 1139 * In these states we ignore the URG. 1140 * 1141 * According to RFC961 (Assigned Protocols), 1142 * the urgent pointer points to the last octet 1143 * of urgent data. We continue, however, 1144 * to consider it to indicate the first octet 1145 * of data past the urgent section as the original 1146 * spec states (in one of two places). 1147 */ 1148 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1149 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1150 so->so_oobmark = so->so_rcv.sb_cc + 1151 (tp->rcv_up - tp->rcv_nxt) - 1; 1152 if (so->so_oobmark == 0) 1153 so->so_state |= SS_RCVATMARK; 1154 sohasoutofband(so); 1155 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1156 } 1157 /* 1158 * Remove out of band data so doesn't get presented to user. 1159 * This can happen independent of advancing the URG pointer, 1160 * but if two URG's are pending at once, some out-of-band 1161 * data may creep in... ick. 1162 */ 1163 if (ti->ti_urp <= ti->ti_len 1164 #ifdef SO_OOBINLINE 1165 && (so->so_options & SO_OOBINLINE) == 0 1166 #endif 1167 ) 1168 tcp_pulloutofband(so, ti, m); 1169 } else 1170 /* 1171 * If no out of band data is expected, 1172 * pull receive urgent pointer along 1173 * with the receive window. 1174 */ 1175 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1176 tp->rcv_up = tp->rcv_nxt; 1177 dodata: /* XXX */ 1178 1179 /* 1180 * Process the segment text, merging it into the TCP sequencing queue, 1181 * and arranging for acknowledgment of receipt if necessary. 1182 * This process logically involves adjusting tp->rcv_wnd as data 1183 * is presented to the user (this happens in tcp_usrreq.c, 1184 * case PRU_RCVD). If a FIN has already been received on this 1185 * connection then we just ignore the text. 1186 */ 1187 if ((ti->ti_len || (tiflags&TH_FIN)) && 1188 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1189 TCP_REASS(tp, ti, m, so, tiflags); 1190 /* 1191 * Note the amount of data that peer has sent into 1192 * our window, in order to estimate the sender's 1193 * buffer size. 1194 */ 1195 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1196 } else { 1197 m_freem(m); 1198 tiflags &= ~TH_FIN; 1199 } 1200 1201 /* 1202 * If FIN is received ACK the FIN and let the user know 1203 * that the connection is closing. 1204 */ 1205 if (tiflags & TH_FIN) { 1206 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1207 socantrcvmore(so); 1208 tp->t_flags |= TF_ACKNOW; 1209 tp->rcv_nxt++; 1210 } 1211 switch (tp->t_state) { 1212 1213 /* 1214 * In SYN_RECEIVED and ESTABLISHED STATES 1215 * enter the CLOSE_WAIT state. 1216 */ 1217 case TCPS_SYN_RECEIVED: 1218 case TCPS_ESTABLISHED: 1219 tp->t_state = TCPS_CLOSE_WAIT; 1220 break; 1221 1222 /* 1223 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1224 * enter the CLOSING state. 1225 */ 1226 case TCPS_FIN_WAIT_1: 1227 tp->t_state = TCPS_CLOSING; 1228 break; 1229 1230 /* 1231 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1232 * starting the time-wait timer, turning off the other 1233 * standard timers. 1234 */ 1235 case TCPS_FIN_WAIT_2: 1236 tp->t_state = TCPS_TIME_WAIT; 1237 tcp_canceltimers(tp); 1238 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1239 soisdisconnected(so); 1240 break; 1241 1242 /* 1243 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1244 */ 1245 case TCPS_TIME_WAIT: 1246 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1247 break; 1248 } 1249 } 1250 if (so->so_options & SO_DEBUG) 1251 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1252 1253 /* 1254 * Return any desired output. 1255 */ 1256 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1257 (void) tcp_output(tp); 1258 return; 1259 1260 dropafterack: 1261 /* 1262 * Generate an ACK dropping incoming segment if it occupies 1263 * sequence space, where the ACK reflects our state. 1264 */ 1265 if (tiflags & TH_RST) 1266 goto drop; 1267 m_freem(m); 1268 tp->t_flags |= TF_ACKNOW; 1269 (void) tcp_output(tp); 1270 return; 1271 1272 dropwithreset: 1273 /* 1274 * Generate a RST, dropping incoming segment. 1275 * Make ACK acceptable to originator of segment. 1276 * Don't bother to respond if destination was broadcast/multicast. 1277 */ 1278 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1279 IN_MULTICAST(ntohl(ti->ti_dst.s_addr))) 1280 goto drop; 1281 if (tiflags & TH_ACK) 1282 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1283 else { 1284 if (tiflags & TH_SYN) 1285 ti->ti_len++; 1286 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1287 TH_RST|TH_ACK); 1288 } 1289 /* destroy temporarily created socket */ 1290 if (dropsocket) 1291 (void) soabort(so); 1292 return; 1293 1294 drop: 1295 /* 1296 * Drop space held by incoming segment and return. 1297 */ 1298 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1299 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1300 m_freem(m); 1301 /* destroy temporarily created socket */ 1302 if (dropsocket) 1303 (void) soabort(so); 1304 return; 1305 #ifndef TUBA_INCLUDE 1306 } 1307 1308 void 1309 tcp_dooptions(tp, cp, cnt, ti, ts_present, ts_val, ts_ecr) 1310 struct tcpcb *tp; 1311 u_char *cp; 1312 int cnt; 1313 struct tcpiphdr *ti; 1314 int *ts_present; 1315 u_long *ts_val, *ts_ecr; 1316 { 1317 u_short mss; 1318 int opt, optlen; 1319 1320 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1321 opt = cp[0]; 1322 if (opt == TCPOPT_EOL) 1323 break; 1324 if (opt == TCPOPT_NOP) 1325 optlen = 1; 1326 else { 1327 optlen = cp[1]; 1328 if (optlen <= 0) 1329 break; 1330 } 1331 switch (opt) { 1332 1333 default: 1334 continue; 1335 1336 case TCPOPT_MAXSEG: 1337 if (optlen != TCPOLEN_MAXSEG) 1338 continue; 1339 if (!(ti->ti_flags & TH_SYN)) 1340 continue; 1341 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1342 NTOHS(mss); 1343 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1344 break; 1345 1346 case TCPOPT_WINDOW: 1347 if (optlen != TCPOLEN_WINDOW) 1348 continue; 1349 if (!(ti->ti_flags & TH_SYN)) 1350 continue; 1351 tp->t_flags |= TF_RCVD_SCALE; 1352 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1353 break; 1354 1355 case TCPOPT_TIMESTAMP: 1356 if (optlen != TCPOLEN_TIMESTAMP) 1357 continue; 1358 *ts_present = 1; 1359 bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val)); 1360 NTOHL(*ts_val); 1361 bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr)); 1362 NTOHL(*ts_ecr); 1363 1364 /* 1365 * A timestamp received in a SYN makes 1366 * it ok to send timestamp requests and replies. 1367 */ 1368 if (ti->ti_flags & TH_SYN) { 1369 tp->t_flags |= TF_RCVD_TSTMP; 1370 tp->ts_recent = *ts_val; 1371 tp->ts_recent_age = tcp_now; 1372 } 1373 break; 1374 } 1375 } 1376 } 1377 1378 /* 1379 * Pull out of band byte out of a segment so 1380 * it doesn't appear in the user's data queue. 1381 * It is still reflected in the segment length for 1382 * sequencing purposes. 1383 */ 1384 void 1385 tcp_pulloutofband(so, ti, m) 1386 struct socket *so; 1387 struct tcpiphdr *ti; 1388 register struct mbuf *m; 1389 { 1390 int cnt = ti->ti_urp - 1; 1391 1392 while (cnt >= 0) { 1393 if (m->m_len > cnt) { 1394 char *cp = mtod(m, caddr_t) + cnt; 1395 struct tcpcb *tp = sototcpcb(so); 1396 1397 tp->t_iobc = *cp; 1398 tp->t_oobflags |= TCPOOB_HAVEDATA; 1399 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1400 m->m_len--; 1401 return; 1402 } 1403 cnt -= m->m_len; 1404 m = m->m_next; 1405 if (m == 0) 1406 break; 1407 } 1408 panic("tcp_pulloutofband"); 1409 } 1410 1411 /* 1412 * Collect new round-trip time estimate 1413 * and update averages and current timeout. 1414 */ 1415 void 1416 tcp_xmit_timer(tp, rtt) 1417 register struct tcpcb *tp; 1418 short rtt; 1419 { 1420 register short delta; 1421 1422 tcpstat.tcps_rttupdated++; 1423 if (tp->t_srtt != 0) { 1424 /* 1425 * srtt is stored as fixed point with 3 bits after the 1426 * binary point (i.e., scaled by 8). The following magic 1427 * is equivalent to the smoothing algorithm in rfc793 with 1428 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1429 * point). Adjust rtt to origin 0. 1430 */ 1431 delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT); 1432 if ((tp->t_srtt += delta) <= 0) 1433 tp->t_srtt = 1; 1434 /* 1435 * We accumulate a smoothed rtt variance (actually, a 1436 * smoothed mean difference), then set the retransmit 1437 * timer to smoothed rtt + 4 times the smoothed variance. 1438 * rttvar is stored as fixed point with 2 bits after the 1439 * binary point (scaled by 4). The following is 1440 * equivalent to rfc793 smoothing with an alpha of .75 1441 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1442 * rfc793's wired-in beta. 1443 */ 1444 if (delta < 0) 1445 delta = -delta; 1446 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1447 if ((tp->t_rttvar += delta) <= 0) 1448 tp->t_rttvar = 1; 1449 } else { 1450 /* 1451 * No rtt measurement yet - use the unsmoothed rtt. 1452 * Set the variance to half the rtt (so our first 1453 * retransmit happens at 3*rtt). 1454 */ 1455 tp->t_srtt = rtt << TCP_RTT_SHIFT; 1456 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 1457 } 1458 tp->t_rtt = 0; 1459 tp->t_rxtshift = 0; 1460 1461 /* 1462 * the retransmit should happen at rtt + 4 * rttvar. 1463 * Because of the way we do the smoothing, srtt and rttvar 1464 * will each average +1/2 tick of bias. When we compute 1465 * the retransmit timer, we want 1/2 tick of rounding and 1466 * 1 extra tick because of +-1/2 tick uncertainty in the 1467 * firing of the timer. The bias will give us exactly the 1468 * 1.5 tick we need. But, because the bias is 1469 * statistical, we have to test that we don't drop below 1470 * the minimum feasible timer (which is 2 ticks). 1471 */ 1472 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1473 tp->t_rttmin, TCPTV_REXMTMAX); 1474 1475 /* 1476 * We received an ack for a packet that wasn't retransmitted; 1477 * it is probably safe to discard any error indications we've 1478 * received recently. This isn't quite right, but close enough 1479 * for now (a route might have failed after we sent a segment, 1480 * and the return path might not be symmetrical). 1481 */ 1482 tp->t_softerror = 0; 1483 } 1484 1485 /* 1486 * Determine a reasonable value for maxseg size. 1487 * If the route is known, check route for mtu. 1488 * If none, use an mss that can be handled on the outgoing 1489 * interface without forcing IP to fragment; if bigger than 1490 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1491 * to utilize large mbufs. If no route is found, route has no mtu, 1492 * or the destination isn't local, use a default, hopefully conservative 1493 * size (usually 512 or the default IP max size, but no more than the mtu 1494 * of the interface), as we can't discover anything about intervening 1495 * gateways or networks. We also initialize the congestion/slow start 1496 * window to be a single segment if the destination isn't local. 1497 * While looking at the routing entry, we also initialize other path-dependent 1498 * parameters from pre-set or cached values in the routing entry. 1499 */ 1500 int 1501 tcp_mss(tp, offer) 1502 register struct tcpcb *tp; 1503 u_int offer; 1504 { 1505 struct route *ro; 1506 register struct rtentry *rt; 1507 struct ifnet *ifp; 1508 register int rtt, mss; 1509 u_long bufsize; 1510 struct inpcb *inp; 1511 struct socket *so; 1512 extern int tcp_mssdflt; 1513 1514 inp = tp->t_inpcb; 1515 ro = &inp->inp_route; 1516 1517 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1518 /* No route yet, so try to acquire one */ 1519 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1520 ro->ro_dst.sa_family = AF_INET; 1521 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1522 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1523 inp->inp_faddr; 1524 rtalloc(ro); 1525 } 1526 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1527 return (tcp_mssdflt); 1528 } 1529 ifp = rt->rt_ifp; 1530 so = inp->inp_socket; 1531 1532 #ifdef RTV_MTU /* if route characteristics exist ... */ 1533 /* 1534 * While we're here, check if there's an initial rtt 1535 * or rttvar. Convert from the route-table units 1536 * to scaled multiples of the slow timeout timer. 1537 */ 1538 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1539 /* 1540 * XXX the lock bit for MTU indicates that the value 1541 * is also a minimum value; this is subject to time. 1542 */ 1543 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1544 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1545 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1546 if (rt->rt_rmx.rmx_rttvar) 1547 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1548 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1549 else 1550 /* default variation is +- 1 rtt */ 1551 tp->t_rttvar = 1552 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1553 TCPT_RANGESET(tp->t_rxtcur, 1554 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1555 tp->t_rttmin, TCPTV_REXMTMAX); 1556 } 1557 /* 1558 * if there's an mtu associated with the route, use it 1559 */ 1560 if (rt->rt_rmx.rmx_mtu) 1561 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1562 else 1563 #endif /* RTV_MTU */ 1564 { 1565 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1566 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1567 if (mss > MCLBYTES) 1568 mss &= ~(MCLBYTES-1); 1569 #else 1570 if (mss > MCLBYTES) 1571 mss = mss / MCLBYTES * MCLBYTES; 1572 #endif 1573 if (!in_localaddr(inp->inp_faddr)) 1574 mss = min(mss, tcp_mssdflt); 1575 } 1576 /* 1577 * The current mss, t_maxseg, is initialized to the default value. 1578 * If we compute a smaller value, reduce the current mss. 1579 * If we compute a larger value, return it for use in sending 1580 * a max seg size option, but don't store it for use 1581 * unless we received an offer at least that large from peer. 1582 * However, do not accept offers under 32 bytes. 1583 */ 1584 if (offer) 1585 mss = min(mss, offer); 1586 mss = max(mss, 32); /* sanity */ 1587 if (mss < tp->t_maxseg || offer != 0) { 1588 /* 1589 * If there's a pipesize, change the socket buffer 1590 * to that size. Make the socket buffers an integral 1591 * number of mss units; if the mss is larger than 1592 * the socket buffer, decrease the mss. 1593 */ 1594 #ifdef RTV_SPIPE 1595 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1596 #endif 1597 bufsize = so->so_snd.sb_hiwat; 1598 if (bufsize < mss) 1599 mss = bufsize; 1600 else { 1601 bufsize = roundup(bufsize, mss); 1602 if (bufsize > sb_max) 1603 bufsize = sb_max; 1604 (void)sbreserve(&so->so_snd, bufsize); 1605 } 1606 tp->t_maxseg = mss; 1607 1608 #ifdef RTV_RPIPE 1609 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1610 #endif 1611 bufsize = so->so_rcv.sb_hiwat; 1612 if (bufsize > mss) { 1613 bufsize = roundup(bufsize, mss); 1614 if (bufsize > sb_max) 1615 bufsize = sb_max; 1616 (void)sbreserve(&so->so_rcv, bufsize); 1617 } 1618 } 1619 tp->snd_cwnd = mss; 1620 1621 #ifdef RTV_SSTHRESH 1622 if (rt->rt_rmx.rmx_ssthresh) { 1623 /* 1624 * There's some sort of gateway or interface 1625 * buffer limit on the path. Use this to set 1626 * the slow start threshhold, but set the 1627 * threshold to no less than 2*mss. 1628 */ 1629 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1630 } 1631 #endif /* RTV_MTU */ 1632 return (mss); 1633 } 1634 #endif /* TUBA_INCLUDE */ 1635