1 /* $NetBSD: tcp_input.c,v 1.20 1995/11/21 01:07:39 cgd Exp $ */ 2 3 /* 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by the University of 18 * California, Berkeley and its contributors. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)tcp_input.c 8.5 (Berkeley) 4/10/94 36 */ 37 38 #ifndef TUBA_INCLUDE 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/malloc.h> 42 #include <sys/mbuf.h> 43 #include <sys/protosw.h> 44 #include <sys/socket.h> 45 #include <sys/socketvar.h> 46 #include <sys/errno.h> 47 48 #include <net/if.h> 49 #include <net/route.h> 50 51 #include <netinet/in.h> 52 #include <netinet/in_systm.h> 53 #include <netinet/ip.h> 54 #include <netinet/in_pcb.h> 55 #include <netinet/ip_var.h> 56 #include <netinet/tcp.h> 57 #include <netinet/tcp_fsm.h> 58 #include <netinet/tcp_seq.h> 59 #include <netinet/tcp_timer.h> 60 #include <netinet/tcp_var.h> 61 #include <netinet/tcpip.h> 62 #include <netinet/tcp_debug.h> 63 64 int tcprexmtthresh = 3; 65 struct tcpiphdr tcp_saveti; 66 struct inpcb *tcp_last_inpcb = 0; 67 68 extern u_long sb_max; 69 70 #endif /* TUBA_INCLUDE */ 71 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 72 73 /* for modulo comparisons of timestamps */ 74 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 75 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 76 77 78 /* 79 * Insert segment ti into reassembly queue of tcp with 80 * control block tp. Return TH_FIN if reassembly now includes 81 * a segment with FIN. The macro form does the common case inline 82 * (segment is the next to be received on an established connection, 83 * and the queue is empty), avoiding linkage into and removal 84 * from the queue and repetition of various conversions. 85 * Set DELACK for segments received in order, but ack immediately 86 * when segments are out of order (so fast retransmit can work). 87 */ 88 #define TCP_REASS(tp, ti, m, so, flags) { \ 89 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 90 (tp)->segq.lh_first == NULL && \ 91 (tp)->t_state == TCPS_ESTABLISHED) { \ 92 if ((ti)->ti_flags & TH_PUSH) \ 93 tp->t_flags |= TF_ACKNOW; \ 94 else \ 95 tp->t_flags |= TF_DELACK; \ 96 (tp)->rcv_nxt += (ti)->ti_len; \ 97 flags = (ti)->ti_flags & TH_FIN; \ 98 tcpstat.tcps_rcvpack++;\ 99 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 100 sbappend(&(so)->so_rcv, (m)); \ 101 sorwakeup(so); \ 102 } else { \ 103 (flags) = tcp_reass((tp), (ti), (m)); \ 104 tp->t_flags |= TF_ACKNOW; \ 105 } \ 106 } 107 #ifndef TUBA_INCLUDE 108 109 int 110 tcp_reass(tp, ti, m) 111 register struct tcpcb *tp; 112 register struct tcpiphdr *ti; 113 struct mbuf *m; 114 { 115 register struct ipqent *p, *q, *nq, *tiqe; 116 struct socket *so = tp->t_inpcb->inp_socket; 117 int flags; 118 119 /* 120 * Call with ti==0 after become established to 121 * force pre-ESTABLISHED data up to user socket. 122 */ 123 if (ti == 0) 124 goto present; 125 126 /* 127 * Allocate a new queue entry, before we throw away any data. 128 * If we can't, just drop the packet. XXX 129 */ 130 MALLOC(tiqe, struct ipqent *, sizeof (struct ipqent), M_IPQ, M_NOWAIT); 131 if (tiqe == NULL) { 132 tcpstat.tcps_rcvmemdrop++; 133 m_freem(m); 134 return (0); 135 } 136 137 /* 138 * Find a segment which begins after this one does. 139 */ 140 for (p = NULL, q = tp->segq.lh_first; q != NULL; 141 p = q, q = q->ipqe_q.le_next) 142 if (SEQ_GT(q->ipqe_tcp->ti_seq, ti->ti_seq)) 143 break; 144 145 /* 146 * If there is a preceding segment, it may provide some of 147 * our data already. If so, drop the data from the incoming 148 * segment. If it provides all of our data, drop us. 149 */ 150 if (p != NULL) { 151 register struct tcpiphdr *phdr = p->ipqe_tcp; 152 register int i; 153 154 /* conversion to int (in i) handles seq wraparound */ 155 i = phdr->ti_seq + phdr->ti_len - ti->ti_seq; 156 if (i > 0) { 157 if (i >= ti->ti_len) { 158 tcpstat.tcps_rcvduppack++; 159 tcpstat.tcps_rcvdupbyte += ti->ti_len; 160 m_freem(m); 161 FREE(tiqe, M_IPQ); 162 return (0); 163 } 164 m_adj(m, i); 165 ti->ti_len -= i; 166 ti->ti_seq += i; 167 } 168 } 169 tcpstat.tcps_rcvoopack++; 170 tcpstat.tcps_rcvoobyte += ti->ti_len; 171 172 /* 173 * While we overlap succeeding segments trim them or, 174 * if they are completely covered, dequeue them. 175 */ 176 for (; q != NULL; q = nq) { 177 register struct tcpiphdr *qhdr = q->ipqe_tcp; 178 register int i = (ti->ti_seq + ti->ti_len) - qhdr->ti_seq; 179 180 if (i <= 0) 181 break; 182 if (i < qhdr->ti_len) { 183 qhdr->ti_seq += i; 184 qhdr->ti_len -= i; 185 m_adj(q->ipqe_m, i); 186 break; 187 } 188 nq = q->ipqe_q.le_next; 189 m_freem(q->ipqe_m); 190 LIST_REMOVE(q, ipqe_q); 191 FREE(q, M_IPQ); 192 } 193 194 /* Insert the new fragment queue entry into place. */ 195 tiqe->ipqe_m = m; 196 tiqe->ipqe_tcp = ti; 197 if (p == NULL) { 198 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 199 } else { 200 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 201 } 202 203 present: 204 /* 205 * Present data to user, advancing rcv_nxt through 206 * completed sequence space. 207 */ 208 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 209 return (0); 210 q = tp->segq.lh_first; 211 if (q == NULL || q->ipqe_tcp->ti_seq != tp->rcv_nxt) 212 return (0); 213 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_tcp->ti_len) 214 return (0); 215 do { 216 tp->rcv_nxt += q->ipqe_tcp->ti_len; 217 flags = q->ipqe_tcp->ti_flags & TH_FIN; 218 219 nq = q->ipqe_q.le_next; 220 LIST_REMOVE(q, ipqe_q); 221 if (so->so_state & SS_CANTRCVMORE) 222 m_freem(q->ipqe_m); 223 else 224 sbappend(&so->so_rcv, q->ipqe_m); 225 FREE(q, M_IPQ); 226 q = nq; 227 } while (q != NULL && q->ipqe_tcp->ti_seq == tp->rcv_nxt); 228 sorwakeup(so); 229 return (flags); 230 } 231 232 /* 233 * TCP input routine, follows pages 65-76 of the 234 * protocol specification dated September, 1981 very closely. 235 */ 236 void 237 tcp_input(m, iphlen) 238 register struct mbuf *m; 239 int iphlen; 240 { 241 register struct tcpiphdr *ti; 242 register struct inpcb *inp; 243 caddr_t optp = NULL; 244 int optlen; 245 int len, tlen, off; 246 register struct tcpcb *tp = 0; 247 register int tiflags; 248 struct socket *so; 249 int todrop, acked, ourfinisacked, needoutput = 0; 250 short ostate; 251 struct in_addr laddr; 252 int dropsocket = 0; 253 int iss = 0; 254 u_long tiwin; 255 u_int32_t ts_val, ts_ecr; 256 int ts_present = 0; 257 258 tcpstat.tcps_rcvtotal++; 259 /* 260 * Get IP and TCP header together in first mbuf. 261 * Note: IP leaves IP header in first mbuf. 262 */ 263 ti = mtod(m, struct tcpiphdr *); 264 if (iphlen > sizeof (struct ip)) 265 ip_stripoptions(m, (struct mbuf *)0); 266 if (m->m_len < sizeof (struct tcpiphdr)) { 267 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 268 tcpstat.tcps_rcvshort++; 269 return; 270 } 271 ti = mtod(m, struct tcpiphdr *); 272 } 273 274 /* 275 * Checksum extended TCP header and data. 276 */ 277 tlen = ((struct ip *)ti)->ip_len; 278 len = sizeof (struct ip) + tlen; 279 bzero(ti->ti_x1, sizeof ti->ti_x1); 280 ti->ti_len = (u_int16_t)tlen; 281 HTONS(ti->ti_len); 282 if (ti->ti_sum = in_cksum(m, len)) { 283 tcpstat.tcps_rcvbadsum++; 284 goto drop; 285 } 286 #endif /* TUBA_INCLUDE */ 287 288 /* 289 * Check that TCP offset makes sense, 290 * pull out TCP options and adjust length. XXX 291 */ 292 off = ti->ti_off << 2; 293 if (off < sizeof (struct tcphdr) || off > tlen) { 294 tcpstat.tcps_rcvbadoff++; 295 goto drop; 296 } 297 tlen -= off; 298 ti->ti_len = tlen; 299 if (off > sizeof (struct tcphdr)) { 300 if (m->m_len < sizeof(struct ip) + off) { 301 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 302 tcpstat.tcps_rcvshort++; 303 return; 304 } 305 ti = mtod(m, struct tcpiphdr *); 306 } 307 optlen = off - sizeof (struct tcphdr); 308 optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 309 /* 310 * Do quick retrieval of timestamp options ("options 311 * prediction?"). If timestamp is the only option and it's 312 * formatted as recommended in RFC 1323 appendix A, we 313 * quickly get the values now and not bother calling 314 * tcp_dooptions(), etc. 315 */ 316 if ((optlen == TCPOLEN_TSTAMP_APPA || 317 (optlen > TCPOLEN_TSTAMP_APPA && 318 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 319 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 320 (ti->ti_flags & TH_SYN) == 0) { 321 ts_present = 1; 322 ts_val = ntohl(*(u_int32_t *)(optp + 4)); 323 ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 324 optp = NULL; /* we've parsed the options */ 325 } 326 } 327 tiflags = ti->ti_flags; 328 329 /* 330 * Convert TCP protocol specific fields to host format. 331 */ 332 NTOHL(ti->ti_seq); 333 NTOHL(ti->ti_ack); 334 NTOHS(ti->ti_win); 335 NTOHS(ti->ti_urp); 336 337 /* 338 * Locate pcb for segment. 339 */ 340 findpcb: 341 inp = tcp_last_inpcb; 342 if (inp == 0 || 343 inp->inp_lport != ti->ti_dport || 344 inp->inp_fport != ti->ti_sport || 345 inp->inp_faddr.s_addr != ti->ti_src.s_addr || 346 inp->inp_laddr.s_addr != ti->ti_dst.s_addr) { 347 ++tcpstat.tcps_pcbcachemiss; 348 inp = in_pcblookup(&tcbtable, ti->ti_src, ti->ti_sport, 349 ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD); 350 /* 351 * If the state is CLOSED (i.e., TCB does not exist) then 352 * all data in the incoming segment is discarded. 353 * If the TCB exists but is in CLOSED state, it is embryonic, 354 * but should either do a listen or a connect soon. 355 */ 356 if (inp == 0) 357 goto dropwithreset; 358 tcp_last_inpcb = inp; 359 } 360 361 tp = intotcpcb(inp); 362 if (tp == 0) 363 goto dropwithreset; 364 if (tp->t_state == TCPS_CLOSED) 365 goto drop; 366 367 /* Unscale the window into a 32-bit value. */ 368 if ((tiflags & TH_SYN) == 0) 369 tiwin = ti->ti_win << tp->snd_scale; 370 else 371 tiwin = ti->ti_win; 372 373 so = inp->inp_socket; 374 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 375 if (so->so_options & SO_DEBUG) { 376 ostate = tp->t_state; 377 tcp_saveti = *ti; 378 } 379 if (so->so_options & SO_ACCEPTCONN) { 380 so = sonewconn(so, 0); 381 if (so == 0) 382 goto drop; 383 /* 384 * This is ugly, but .... 385 * 386 * Mark socket as temporary until we're 387 * committed to keeping it. The code at 388 * ``drop'' and ``dropwithreset'' check the 389 * flag dropsocket to see if the temporary 390 * socket created here should be discarded. 391 * We mark the socket as discardable until 392 * we're committed to it below in TCPS_LISTEN. 393 */ 394 dropsocket++; 395 inp = (struct inpcb *)so->so_pcb; 396 inp->inp_laddr = ti->ti_dst; 397 inp->inp_lport = ti->ti_dport; 398 #if BSD>=43 399 inp->inp_options = ip_srcroute(); 400 #endif 401 tp = intotcpcb(inp); 402 tp->t_state = TCPS_LISTEN; 403 404 /* Compute proper scaling value from buffer space 405 */ 406 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 407 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat) 408 tp->request_r_scale++; 409 } 410 } 411 412 /* 413 * Segment received on connection. 414 * Reset idle time and keep-alive timer. 415 */ 416 tp->t_idle = 0; 417 tp->t_timer[TCPT_KEEP] = tcp_keepidle; 418 419 /* 420 * Process options if not in LISTEN state, 421 * else do it below (after getting remote address). 422 */ 423 if (optp && tp->t_state != TCPS_LISTEN) 424 tcp_dooptions(tp, optp, optlen, ti, 425 &ts_present, &ts_val, &ts_ecr); 426 427 /* 428 * Header prediction: check for the two common cases 429 * of a uni-directional data xfer. If the packet has 430 * no control flags, is in-sequence, the window didn't 431 * change and we're not retransmitting, it's a 432 * candidate. If the length is zero and the ack moved 433 * forward, we're the sender side of the xfer. Just 434 * free the data acked & wake any higher level process 435 * that was blocked waiting for space. If the length 436 * is non-zero and the ack didn't move, we're the 437 * receiver side. If we're getting packets in-order 438 * (the reassembly queue is empty), add the data to 439 * the socket buffer and note that we need a delayed ack. 440 */ 441 if (tp->t_state == TCPS_ESTABLISHED && 442 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 443 (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && 444 ti->ti_seq == tp->rcv_nxt && 445 tiwin && tiwin == tp->snd_wnd && 446 tp->snd_nxt == tp->snd_max) { 447 448 /* 449 * If last ACK falls within this segment's sequence numbers, 450 * record the timestamp. 451 */ 452 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 453 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) { 454 tp->ts_recent_age = tcp_now; 455 tp->ts_recent = ts_val; 456 } 457 458 if (ti->ti_len == 0) { 459 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 460 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 461 tp->snd_cwnd >= tp->snd_wnd && 462 tp->t_dupacks < tcprexmtthresh) { 463 /* 464 * this is a pure ack for outstanding data. 465 */ 466 ++tcpstat.tcps_predack; 467 if (ts_present) 468 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 469 else if (tp->t_rtt && 470 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 471 tcp_xmit_timer(tp, tp->t_rtt); 472 acked = ti->ti_ack - tp->snd_una; 473 tcpstat.tcps_rcvackpack++; 474 tcpstat.tcps_rcvackbyte += acked; 475 sbdrop(&so->so_snd, acked); 476 tp->snd_una = ti->ti_ack; 477 m_freem(m); 478 479 /* 480 * If all outstanding data are acked, stop 481 * retransmit timer, otherwise restart timer 482 * using current (possibly backed-off) value. 483 * If process is waiting for space, 484 * wakeup/selwakeup/signal. If data 485 * are ready to send, let tcp_output 486 * decide between more output or persist. 487 */ 488 if (tp->snd_una == tp->snd_max) 489 tp->t_timer[TCPT_REXMT] = 0; 490 else if (tp->t_timer[TCPT_PERSIST] == 0) 491 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 492 493 if (sb_notify(&so->so_snd)) 494 sowwakeup(so); 495 if (so->so_snd.sb_cc) 496 (void) tcp_output(tp); 497 return; 498 } 499 } else if (ti->ti_ack == tp->snd_una && 500 tp->segq.lh_first == NULL && 501 ti->ti_len <= sbspace(&so->so_rcv)) { 502 /* 503 * this is a pure, in-sequence data packet 504 * with nothing on the reassembly queue and 505 * we have enough buffer space to take it. 506 */ 507 ++tcpstat.tcps_preddat; 508 tp->rcv_nxt += ti->ti_len; 509 tcpstat.tcps_rcvpack++; 510 tcpstat.tcps_rcvbyte += ti->ti_len; 511 /* 512 * Drop TCP, IP headers and TCP options then add data 513 * to socket buffer. 514 */ 515 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 516 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 517 sbappend(&so->so_rcv, m); 518 sorwakeup(so); 519 if (ti->ti_flags & TH_PUSH) 520 tp->t_flags |= TF_ACKNOW; 521 else 522 tp->t_flags |= TF_DELACK; 523 return; 524 } 525 } 526 527 /* 528 * Drop TCP, IP headers and TCP options. 529 */ 530 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 531 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 532 533 /* 534 * Calculate amount of space in receive window, 535 * and then do TCP input processing. 536 * Receive window is amount of space in rcv queue, 537 * but not less than advertised window. 538 */ 539 { int win; 540 541 win = sbspace(&so->so_rcv); 542 if (win < 0) 543 win = 0; 544 tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 545 } 546 547 switch (tp->t_state) { 548 549 /* 550 * If the state is LISTEN then ignore segment if it contains an RST. 551 * If the segment contains an ACK then it is bad and send a RST. 552 * If it does not contain a SYN then it is not interesting; drop it. 553 * Don't bother responding if the destination was a broadcast. 554 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 555 * tp->iss, and send a segment: 556 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 557 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 558 * Fill in remote peer address fields if not previously specified. 559 * Enter SYN_RECEIVED state, and process any other fields of this 560 * segment in this state. 561 */ 562 case TCPS_LISTEN: { 563 struct mbuf *am; 564 register struct sockaddr_in *sin; 565 566 if (tiflags & TH_RST) 567 goto drop; 568 if (tiflags & TH_ACK) 569 goto dropwithreset; 570 if ((tiflags & TH_SYN) == 0) 571 goto drop; 572 /* 573 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 574 * in_broadcast() should never return true on a received 575 * packet with M_BCAST not set. 576 */ 577 if (m->m_flags & (M_BCAST|M_MCAST) || 578 IN_MULTICAST(ti->ti_dst.s_addr)) 579 goto drop; 580 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 581 if (am == NULL) 582 goto drop; 583 am->m_len = sizeof (struct sockaddr_in); 584 sin = mtod(am, struct sockaddr_in *); 585 sin->sin_family = AF_INET; 586 sin->sin_len = sizeof(*sin); 587 sin->sin_addr = ti->ti_src; 588 sin->sin_port = ti->ti_sport; 589 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 590 laddr = inp->inp_laddr; 591 if (inp->inp_laddr.s_addr == INADDR_ANY) 592 inp->inp_laddr = ti->ti_dst; 593 if (in_pcbconnect(inp, am)) { 594 inp->inp_laddr = laddr; 595 (void) m_free(am); 596 goto drop; 597 } 598 (void) m_free(am); 599 tp->t_template = tcp_template(tp); 600 if (tp->t_template == 0) { 601 tp = tcp_drop(tp, ENOBUFS); 602 dropsocket = 0; /* socket is already gone */ 603 goto drop; 604 } 605 if (optp) 606 tcp_dooptions(tp, optp, optlen, ti, 607 &ts_present, &ts_val, &ts_ecr); 608 if (iss) 609 tp->iss = iss; 610 else 611 tp->iss = tcp_iss; 612 tcp_iss += TCP_ISSINCR/2; 613 tp->irs = ti->ti_seq; 614 tcp_sendseqinit(tp); 615 tcp_rcvseqinit(tp); 616 tp->t_flags |= TF_ACKNOW; 617 tp->t_state = TCPS_SYN_RECEIVED; 618 tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT; 619 dropsocket = 0; /* committed to socket */ 620 tcpstat.tcps_accepts++; 621 goto trimthenstep6; 622 } 623 624 /* 625 * If the state is SYN_SENT: 626 * if seg contains an ACK, but not for our SYN, drop the input. 627 * if seg contains a RST, then drop the connection. 628 * if seg does not contain SYN, then drop it. 629 * Otherwise this is an acceptable SYN segment 630 * initialize tp->rcv_nxt and tp->irs 631 * if seg contains ack then advance tp->snd_una 632 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 633 * arrange for segment to be acked (eventually) 634 * continue processing rest of data/controls, beginning with URG 635 */ 636 case TCPS_SYN_SENT: 637 if ((tiflags & TH_ACK) && 638 (SEQ_LEQ(ti->ti_ack, tp->iss) || 639 SEQ_GT(ti->ti_ack, tp->snd_max))) 640 goto dropwithreset; 641 if (tiflags & TH_RST) { 642 if (tiflags & TH_ACK) 643 tp = tcp_drop(tp, ECONNREFUSED); 644 goto drop; 645 } 646 if ((tiflags & TH_SYN) == 0) 647 goto drop; 648 if (tiflags & TH_ACK) { 649 tp->snd_una = ti->ti_ack; 650 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 651 tp->snd_nxt = tp->snd_una; 652 } 653 tp->t_timer[TCPT_REXMT] = 0; 654 tp->irs = ti->ti_seq; 655 tcp_rcvseqinit(tp); 656 tp->t_flags |= TF_ACKNOW; 657 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 658 tcpstat.tcps_connects++; 659 soisconnected(so); 660 tp->t_state = TCPS_ESTABLISHED; 661 /* Do window scaling on this connection? */ 662 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 663 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 664 tp->snd_scale = tp->requested_s_scale; 665 tp->rcv_scale = tp->request_r_scale; 666 } 667 (void) tcp_reass(tp, (struct tcpiphdr *)0, 668 (struct mbuf *)0); 669 /* 670 * if we didn't have to retransmit the SYN, 671 * use its rtt as our initial srtt & rtt var. 672 */ 673 if (tp->t_rtt) 674 tcp_xmit_timer(tp, tp->t_rtt); 675 } else 676 tp->t_state = TCPS_SYN_RECEIVED; 677 678 trimthenstep6: 679 /* 680 * Advance ti->ti_seq to correspond to first data byte. 681 * If data, trim to stay within window, 682 * dropping FIN if necessary. 683 */ 684 ti->ti_seq++; 685 if (ti->ti_len > tp->rcv_wnd) { 686 todrop = ti->ti_len - tp->rcv_wnd; 687 m_adj(m, -todrop); 688 ti->ti_len = tp->rcv_wnd; 689 tiflags &= ~TH_FIN; 690 tcpstat.tcps_rcvpackafterwin++; 691 tcpstat.tcps_rcvbyteafterwin += todrop; 692 } 693 tp->snd_wl1 = ti->ti_seq - 1; 694 tp->rcv_up = ti->ti_seq; 695 goto step6; 696 } 697 698 /* 699 * States other than LISTEN or SYN_SENT. 700 * First check timestamp, if present. 701 * Then check that at least some bytes of segment are within 702 * receive window. If segment begins before rcv_nxt, 703 * drop leading data (and SYN); if nothing left, just ack. 704 * 705 * RFC 1323 PAWS: If we have a timestamp reply on this segment 706 * and it's less than ts_recent, drop it. 707 */ 708 if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 709 TSTMP_LT(ts_val, tp->ts_recent)) { 710 711 /* Check to see if ts_recent is over 24 days old. */ 712 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 713 /* 714 * Invalidate ts_recent. If this segment updates 715 * ts_recent, the age will be reset later and ts_recent 716 * will get a valid value. If it does not, setting 717 * ts_recent to zero will at least satisfy the 718 * requirement that zero be placed in the timestamp 719 * echo reply when ts_recent isn't valid. The 720 * age isn't reset until we get a valid ts_recent 721 * because we don't want out-of-order segments to be 722 * dropped when ts_recent is old. 723 */ 724 tp->ts_recent = 0; 725 } else { 726 tcpstat.tcps_rcvduppack++; 727 tcpstat.tcps_rcvdupbyte += ti->ti_len; 728 tcpstat.tcps_pawsdrop++; 729 goto dropafterack; 730 } 731 } 732 733 todrop = tp->rcv_nxt - ti->ti_seq; 734 if (todrop > 0) { 735 if (tiflags & TH_SYN) { 736 tiflags &= ~TH_SYN; 737 ti->ti_seq++; 738 if (ti->ti_urp > 1) 739 ti->ti_urp--; 740 else 741 tiflags &= ~TH_URG; 742 todrop--; 743 } 744 if (todrop >= ti->ti_len) { 745 /* 746 * Any valid FIN must be to the left of the 747 * window. At this point, FIN must be a 748 * duplicate or out-of-sequence, so drop it. 749 */ 750 tiflags &= ~TH_FIN; 751 /* 752 * Send ACK to resynchronize, and drop any data, 753 * but keep on processing for RST or ACK. 754 */ 755 tp->t_flags |= TF_ACKNOW; 756 tcpstat.tcps_rcvdupbyte += todrop = ti->ti_len; 757 tcpstat.tcps_rcvduppack++; 758 } else { 759 tcpstat.tcps_rcvpartduppack++; 760 tcpstat.tcps_rcvpartdupbyte += todrop; 761 } 762 m_adj(m, todrop); 763 ti->ti_seq += todrop; 764 ti->ti_len -= todrop; 765 if (ti->ti_urp > todrop) 766 ti->ti_urp -= todrop; 767 else { 768 tiflags &= ~TH_URG; 769 ti->ti_urp = 0; 770 } 771 } 772 773 /* 774 * If new data are received on a connection after the 775 * user processes are gone, then RST the other end. 776 */ 777 if ((so->so_state & SS_NOFDREF) && 778 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 779 tp = tcp_close(tp); 780 tcpstat.tcps_rcvafterclose++; 781 goto dropwithreset; 782 } 783 784 /* 785 * If segment ends after window, drop trailing data 786 * (and PUSH and FIN); if nothing left, just ACK. 787 */ 788 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 789 if (todrop > 0) { 790 tcpstat.tcps_rcvpackafterwin++; 791 if (todrop >= ti->ti_len) { 792 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 793 /* 794 * If a new connection request is received 795 * while in TIME_WAIT, drop the old connection 796 * and start over if the sequence numbers 797 * are above the previous ones. 798 */ 799 if (tiflags & TH_SYN && 800 tp->t_state == TCPS_TIME_WAIT && 801 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 802 iss = tp->rcv_nxt + TCP_ISSINCR; 803 tp = tcp_close(tp); 804 goto findpcb; 805 } 806 /* 807 * If window is closed can only take segments at 808 * window edge, and have to drop data and PUSH from 809 * incoming segments. Continue processing, but 810 * remember to ack. Otherwise, drop segment 811 * and ack. 812 */ 813 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 814 tp->t_flags |= TF_ACKNOW; 815 tcpstat.tcps_rcvwinprobe++; 816 } else 817 goto dropafterack; 818 } else 819 tcpstat.tcps_rcvbyteafterwin += todrop; 820 m_adj(m, -todrop); 821 ti->ti_len -= todrop; 822 tiflags &= ~(TH_PUSH|TH_FIN); 823 } 824 825 /* 826 * If last ACK falls within this segment's sequence numbers, 827 * record its timestamp. 828 */ 829 if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 830 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + 831 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 832 tp->ts_recent_age = tcp_now; 833 tp->ts_recent = ts_val; 834 } 835 836 /* 837 * If the RST bit is set examine the state: 838 * SYN_RECEIVED STATE: 839 * If passive open, return to LISTEN state. 840 * If active open, inform user that connection was refused. 841 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 842 * Inform user that connection was reset, and close tcb. 843 * CLOSING, LAST_ACK, TIME_WAIT STATES 844 * Close the tcb. 845 */ 846 if (tiflags&TH_RST) switch (tp->t_state) { 847 848 case TCPS_SYN_RECEIVED: 849 so->so_error = ECONNREFUSED; 850 goto close; 851 852 case TCPS_ESTABLISHED: 853 case TCPS_FIN_WAIT_1: 854 case TCPS_FIN_WAIT_2: 855 case TCPS_CLOSE_WAIT: 856 so->so_error = ECONNRESET; 857 close: 858 tp->t_state = TCPS_CLOSED; 859 tcpstat.tcps_drops++; 860 tp = tcp_close(tp); 861 goto drop; 862 863 case TCPS_CLOSING: 864 case TCPS_LAST_ACK: 865 case TCPS_TIME_WAIT: 866 tp = tcp_close(tp); 867 goto drop; 868 } 869 870 /* 871 * If a SYN is in the window, then this is an 872 * error and we send an RST and drop the connection. 873 */ 874 if (tiflags & TH_SYN) { 875 tp = tcp_drop(tp, ECONNRESET); 876 goto dropwithreset; 877 } 878 879 /* 880 * If the ACK bit is off we drop the segment and return. 881 */ 882 if ((tiflags & TH_ACK) == 0) 883 goto drop; 884 885 /* 886 * Ack processing. 887 */ 888 switch (tp->t_state) { 889 890 /* 891 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 892 * ESTABLISHED state and continue processing, otherwise 893 * send an RST. 894 */ 895 case TCPS_SYN_RECEIVED: 896 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 897 SEQ_GT(ti->ti_ack, tp->snd_max)) 898 goto dropwithreset; 899 tcpstat.tcps_connects++; 900 soisconnected(so); 901 tp->t_state = TCPS_ESTABLISHED; 902 /* Do window scaling? */ 903 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 904 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 905 tp->snd_scale = tp->requested_s_scale; 906 tp->rcv_scale = tp->request_r_scale; 907 } 908 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 909 tp->snd_wl1 = ti->ti_seq - 1; 910 /* fall into ... */ 911 912 /* 913 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 914 * ACKs. If the ack is in the range 915 * tp->snd_una < ti->ti_ack <= tp->snd_max 916 * then advance tp->snd_una to ti->ti_ack and drop 917 * data from the retransmission queue. If this ACK reflects 918 * more up to date window information we update our window information. 919 */ 920 case TCPS_ESTABLISHED: 921 case TCPS_FIN_WAIT_1: 922 case TCPS_FIN_WAIT_2: 923 case TCPS_CLOSE_WAIT: 924 case TCPS_CLOSING: 925 case TCPS_LAST_ACK: 926 case TCPS_TIME_WAIT: 927 928 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 929 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 930 tcpstat.tcps_rcvdupack++; 931 /* 932 * If we have outstanding data (other than 933 * a window probe), this is a completely 934 * duplicate ack (ie, window info didn't 935 * change), the ack is the biggest we've 936 * seen and we've seen exactly our rexmt 937 * threshhold of them, assume a packet 938 * has been dropped and retransmit it. 939 * Kludge snd_nxt & the congestion 940 * window so we send only this one 941 * packet. 942 * 943 * We know we're losing at the current 944 * window size so do congestion avoidance 945 * (set ssthresh to half the current window 946 * and pull our congestion window back to 947 * the new ssthresh). 948 * 949 * Dup acks mean that packets have left the 950 * network (they're now cached at the receiver) 951 * so bump cwnd by the amount in the receiver 952 * to keep a constant cwnd packets in the 953 * network. 954 */ 955 if (tp->t_timer[TCPT_REXMT] == 0 || 956 ti->ti_ack != tp->snd_una) 957 tp->t_dupacks = 0; 958 else if (++tp->t_dupacks == tcprexmtthresh) { 959 tcp_seq onxt = tp->snd_nxt; 960 u_int win = 961 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 962 tp->t_maxseg; 963 964 if (win < 2) 965 win = 2; 966 tp->snd_ssthresh = win * tp->t_maxseg; 967 tp->t_timer[TCPT_REXMT] = 0; 968 tp->t_rtt = 0; 969 tp->snd_nxt = ti->ti_ack; 970 tp->snd_cwnd = tp->t_maxseg; 971 (void) tcp_output(tp); 972 tp->snd_cwnd = tp->snd_ssthresh + 973 tp->t_maxseg * tp->t_dupacks; 974 if (SEQ_GT(onxt, tp->snd_nxt)) 975 tp->snd_nxt = onxt; 976 goto drop; 977 } else if (tp->t_dupacks > tcprexmtthresh) { 978 tp->snd_cwnd += tp->t_maxseg; 979 (void) tcp_output(tp); 980 goto drop; 981 } 982 } else 983 tp->t_dupacks = 0; 984 break; 985 } 986 /* 987 * If the congestion window was inflated to account 988 * for the other side's cached packets, retract it. 989 */ 990 if (tp->t_dupacks >= tcprexmtthresh && 991 tp->snd_cwnd > tp->snd_ssthresh) 992 tp->snd_cwnd = tp->snd_ssthresh; 993 tp->t_dupacks = 0; 994 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 995 tcpstat.tcps_rcvacktoomuch++; 996 goto dropafterack; 997 } 998 acked = ti->ti_ack - tp->snd_una; 999 tcpstat.tcps_rcvackpack++; 1000 tcpstat.tcps_rcvackbyte += acked; 1001 1002 /* 1003 * If we have a timestamp reply, update smoothed 1004 * round trip time. If no timestamp is present but 1005 * transmit timer is running and timed sequence 1006 * number was acked, update smoothed round trip time. 1007 * Since we now have an rtt measurement, cancel the 1008 * timer backoff (cf., Phil Karn's retransmit alg.). 1009 * Recompute the initial retransmit timer. 1010 */ 1011 if (ts_present) 1012 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 1013 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1014 tcp_xmit_timer(tp,tp->t_rtt); 1015 1016 /* 1017 * If all outstanding data is acked, stop retransmit 1018 * timer and remember to restart (more output or persist). 1019 * If there is more data to be acked, restart retransmit 1020 * timer, using current (possibly backed-off) value. 1021 */ 1022 if (ti->ti_ack == tp->snd_max) { 1023 tp->t_timer[TCPT_REXMT] = 0; 1024 needoutput = 1; 1025 } else if (tp->t_timer[TCPT_PERSIST] == 0) 1026 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur; 1027 /* 1028 * When new data is acked, open the congestion window. 1029 * If the window gives us less than ssthresh packets 1030 * in flight, open exponentially (maxseg per packet). 1031 * Otherwise open linearly: maxseg per window 1032 * (maxseg^2 / cwnd per packet), plus a constant 1033 * fraction of a packet (maxseg/8) to help larger windows 1034 * open quickly enough. 1035 */ 1036 { 1037 register u_int cw = tp->snd_cwnd; 1038 register u_int incr = tp->t_maxseg; 1039 1040 if (cw > tp->snd_ssthresh) 1041 incr = incr * incr / cw; 1042 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1043 } 1044 if (acked > so->so_snd.sb_cc) { 1045 tp->snd_wnd -= so->so_snd.sb_cc; 1046 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1047 ourfinisacked = 1; 1048 } else { 1049 sbdrop(&so->so_snd, acked); 1050 tp->snd_wnd -= acked; 1051 ourfinisacked = 0; 1052 } 1053 if (sb_notify(&so->so_snd)) 1054 sowwakeup(so); 1055 tp->snd_una = ti->ti_ack; 1056 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1057 tp->snd_nxt = tp->snd_una; 1058 1059 switch (tp->t_state) { 1060 1061 /* 1062 * In FIN_WAIT_1 STATE in addition to the processing 1063 * for the ESTABLISHED state if our FIN is now acknowledged 1064 * then enter FIN_WAIT_2. 1065 */ 1066 case TCPS_FIN_WAIT_1: 1067 if (ourfinisacked) { 1068 /* 1069 * If we can't receive any more 1070 * data, then closing user can proceed. 1071 * Starting the timer is contrary to the 1072 * specification, but if we don't get a FIN 1073 * we'll hang forever. 1074 */ 1075 if (so->so_state & SS_CANTRCVMORE) { 1076 soisdisconnected(so); 1077 tp->t_timer[TCPT_2MSL] = tcp_maxidle; 1078 } 1079 tp->t_state = TCPS_FIN_WAIT_2; 1080 } 1081 break; 1082 1083 /* 1084 * In CLOSING STATE in addition to the processing for 1085 * the ESTABLISHED state if the ACK acknowledges our FIN 1086 * then enter the TIME-WAIT state, otherwise ignore 1087 * the segment. 1088 */ 1089 case TCPS_CLOSING: 1090 if (ourfinisacked) { 1091 tp->t_state = TCPS_TIME_WAIT; 1092 tcp_canceltimers(tp); 1093 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1094 soisdisconnected(so); 1095 } 1096 break; 1097 1098 /* 1099 * In LAST_ACK, we may still be waiting for data to drain 1100 * and/or to be acked, as well as for the ack of our FIN. 1101 * If our FIN is now acknowledged, delete the TCB, 1102 * enter the closed state and return. 1103 */ 1104 case TCPS_LAST_ACK: 1105 if (ourfinisacked) { 1106 tp = tcp_close(tp); 1107 goto drop; 1108 } 1109 break; 1110 1111 /* 1112 * In TIME_WAIT state the only thing that should arrive 1113 * is a retransmission of the remote FIN. Acknowledge 1114 * it and restart the finack timer. 1115 */ 1116 case TCPS_TIME_WAIT: 1117 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1118 goto dropafterack; 1119 } 1120 } 1121 1122 step6: 1123 /* 1124 * Update window information. 1125 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1126 */ 1127 if ((tiflags & TH_ACK) && 1128 (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq && 1129 (SEQ_LT(tp->snd_wl2, ti->ti_ack) || 1130 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))) { 1131 /* keep track of pure window updates */ 1132 if (ti->ti_len == 0 && 1133 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1134 tcpstat.tcps_rcvwinupd++; 1135 tp->snd_wnd = tiwin; 1136 tp->snd_wl1 = ti->ti_seq; 1137 tp->snd_wl2 = ti->ti_ack; 1138 if (tp->snd_wnd > tp->max_sndwnd) 1139 tp->max_sndwnd = tp->snd_wnd; 1140 needoutput = 1; 1141 } 1142 1143 /* 1144 * Process segments with URG. 1145 */ 1146 if ((tiflags & TH_URG) && ti->ti_urp && 1147 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1148 /* 1149 * This is a kludge, but if we receive and accept 1150 * random urgent pointers, we'll crash in 1151 * soreceive. It's hard to imagine someone 1152 * actually wanting to send this much urgent data. 1153 */ 1154 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1155 ti->ti_urp = 0; /* XXX */ 1156 tiflags &= ~TH_URG; /* XXX */ 1157 goto dodata; /* XXX */ 1158 } 1159 /* 1160 * If this segment advances the known urgent pointer, 1161 * then mark the data stream. This should not happen 1162 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1163 * a FIN has been received from the remote side. 1164 * In these states we ignore the URG. 1165 * 1166 * According to RFC961 (Assigned Protocols), 1167 * the urgent pointer points to the last octet 1168 * of urgent data. We continue, however, 1169 * to consider it to indicate the first octet 1170 * of data past the urgent section as the original 1171 * spec states (in one of two places). 1172 */ 1173 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1174 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1175 so->so_oobmark = so->so_rcv.sb_cc + 1176 (tp->rcv_up - tp->rcv_nxt) - 1; 1177 if (so->so_oobmark == 0) 1178 so->so_state |= SS_RCVATMARK; 1179 sohasoutofband(so); 1180 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1181 } 1182 /* 1183 * Remove out of band data so doesn't get presented to user. 1184 * This can happen independent of advancing the URG pointer, 1185 * but if two URG's are pending at once, some out-of-band 1186 * data may creep in... ick. 1187 */ 1188 if (ti->ti_urp <= ti->ti_len 1189 #ifdef SO_OOBINLINE 1190 && (so->so_options & SO_OOBINLINE) == 0 1191 #endif 1192 ) 1193 tcp_pulloutofband(so, ti, m); 1194 } else 1195 /* 1196 * If no out of band data is expected, 1197 * pull receive urgent pointer along 1198 * with the receive window. 1199 */ 1200 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1201 tp->rcv_up = tp->rcv_nxt; 1202 dodata: /* XXX */ 1203 1204 /* 1205 * Process the segment text, merging it into the TCP sequencing queue, 1206 * and arranging for acknowledgment of receipt if necessary. 1207 * This process logically involves adjusting tp->rcv_wnd as data 1208 * is presented to the user (this happens in tcp_usrreq.c, 1209 * case PRU_RCVD). If a FIN has already been received on this 1210 * connection then we just ignore the text. 1211 */ 1212 if ((ti->ti_len || (tiflags&TH_FIN)) && 1213 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1214 TCP_REASS(tp, ti, m, so, tiflags); 1215 /* 1216 * Note the amount of data that peer has sent into 1217 * our window, in order to estimate the sender's 1218 * buffer size. 1219 */ 1220 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1221 } else { 1222 m_freem(m); 1223 tiflags &= ~TH_FIN; 1224 } 1225 1226 /* 1227 * If FIN is received ACK the FIN and let the user know 1228 * that the connection is closing. 1229 */ 1230 if (tiflags & TH_FIN) { 1231 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1232 socantrcvmore(so); 1233 tp->t_flags |= TF_ACKNOW; 1234 tp->rcv_nxt++; 1235 } 1236 switch (tp->t_state) { 1237 1238 /* 1239 * In SYN_RECEIVED and ESTABLISHED STATES 1240 * enter the CLOSE_WAIT state. 1241 */ 1242 case TCPS_SYN_RECEIVED: 1243 case TCPS_ESTABLISHED: 1244 tp->t_state = TCPS_CLOSE_WAIT; 1245 break; 1246 1247 /* 1248 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1249 * enter the CLOSING state. 1250 */ 1251 case TCPS_FIN_WAIT_1: 1252 tp->t_state = TCPS_CLOSING; 1253 break; 1254 1255 /* 1256 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1257 * starting the time-wait timer, turning off the other 1258 * standard timers. 1259 */ 1260 case TCPS_FIN_WAIT_2: 1261 tp->t_state = TCPS_TIME_WAIT; 1262 tcp_canceltimers(tp); 1263 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1264 soisdisconnected(so); 1265 break; 1266 1267 /* 1268 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1269 */ 1270 case TCPS_TIME_WAIT: 1271 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL; 1272 break; 1273 } 1274 } 1275 if (so->so_options & SO_DEBUG) 1276 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1277 1278 /* 1279 * Return any desired output. 1280 */ 1281 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1282 (void) tcp_output(tp); 1283 return; 1284 1285 dropafterack: 1286 /* 1287 * Generate an ACK dropping incoming segment if it occupies 1288 * sequence space, where the ACK reflects our state. 1289 */ 1290 if (tiflags & TH_RST) 1291 goto drop; 1292 m_freem(m); 1293 tp->t_flags |= TF_ACKNOW; 1294 (void) tcp_output(tp); 1295 return; 1296 1297 dropwithreset: 1298 /* 1299 * Generate a RST, dropping incoming segment. 1300 * Make ACK acceptable to originator of segment. 1301 * Don't bother to respond if destination was broadcast/multicast. 1302 */ 1303 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1304 IN_MULTICAST(ti->ti_dst.s_addr)) 1305 goto drop; 1306 if (tiflags & TH_ACK) 1307 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1308 else { 1309 if (tiflags & TH_SYN) 1310 ti->ti_len++; 1311 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1312 TH_RST|TH_ACK); 1313 } 1314 /* destroy temporarily created socket */ 1315 if (dropsocket) 1316 (void) soabort(so); 1317 return; 1318 1319 drop: 1320 /* 1321 * Drop space held by incoming segment and return. 1322 */ 1323 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1324 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1325 m_freem(m); 1326 /* destroy temporarily created socket */ 1327 if (dropsocket) 1328 (void) soabort(so); 1329 return; 1330 #ifndef TUBA_INCLUDE 1331 } 1332 1333 void 1334 tcp_dooptions(tp, cp, cnt, ti, ts_present, ts_val, ts_ecr) 1335 struct tcpcb *tp; 1336 u_char *cp; 1337 int cnt; 1338 struct tcpiphdr *ti; 1339 int *ts_present; 1340 u_int32_t *ts_val, *ts_ecr; 1341 { 1342 u_int16_t mss; 1343 int opt, optlen; 1344 1345 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1346 opt = cp[0]; 1347 if (opt == TCPOPT_EOL) 1348 break; 1349 if (opt == TCPOPT_NOP) 1350 optlen = 1; 1351 else { 1352 optlen = cp[1]; 1353 if (optlen <= 0) 1354 break; 1355 } 1356 switch (opt) { 1357 1358 default: 1359 continue; 1360 1361 case TCPOPT_MAXSEG: 1362 if (optlen != TCPOLEN_MAXSEG) 1363 continue; 1364 if (!(ti->ti_flags & TH_SYN)) 1365 continue; 1366 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 1367 NTOHS(mss); 1368 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 1369 break; 1370 1371 case TCPOPT_WINDOW: 1372 if (optlen != TCPOLEN_WINDOW) 1373 continue; 1374 if (!(ti->ti_flags & TH_SYN)) 1375 continue; 1376 tp->t_flags |= TF_RCVD_SCALE; 1377 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 1378 break; 1379 1380 case TCPOPT_TIMESTAMP: 1381 if (optlen != TCPOLEN_TIMESTAMP) 1382 continue; 1383 *ts_present = 1; 1384 bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val)); 1385 NTOHL(*ts_val); 1386 bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr)); 1387 NTOHL(*ts_ecr); 1388 1389 /* 1390 * A timestamp received in a SYN makes 1391 * it ok to send timestamp requests and replies. 1392 */ 1393 if (ti->ti_flags & TH_SYN) { 1394 tp->t_flags |= TF_RCVD_TSTMP; 1395 tp->ts_recent = *ts_val; 1396 tp->ts_recent_age = tcp_now; 1397 } 1398 break; 1399 } 1400 } 1401 } 1402 1403 /* 1404 * Pull out of band byte out of a segment so 1405 * it doesn't appear in the user's data queue. 1406 * It is still reflected in the segment length for 1407 * sequencing purposes. 1408 */ 1409 void 1410 tcp_pulloutofband(so, ti, m) 1411 struct socket *so; 1412 struct tcpiphdr *ti; 1413 register struct mbuf *m; 1414 { 1415 int cnt = ti->ti_urp - 1; 1416 1417 while (cnt >= 0) { 1418 if (m->m_len > cnt) { 1419 char *cp = mtod(m, caddr_t) + cnt; 1420 struct tcpcb *tp = sototcpcb(so); 1421 1422 tp->t_iobc = *cp; 1423 tp->t_oobflags |= TCPOOB_HAVEDATA; 1424 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1425 m->m_len--; 1426 return; 1427 } 1428 cnt -= m->m_len; 1429 m = m->m_next; 1430 if (m == 0) 1431 break; 1432 } 1433 panic("tcp_pulloutofband"); 1434 } 1435 1436 /* 1437 * Collect new round-trip time estimate 1438 * and update averages and current timeout. 1439 */ 1440 void 1441 tcp_xmit_timer(tp, rtt) 1442 register struct tcpcb *tp; 1443 short rtt; 1444 { 1445 register short delta; 1446 1447 tcpstat.tcps_rttupdated++; 1448 --rtt; 1449 if (tp->t_srtt != 0) { 1450 /* 1451 * srtt is stored as fixed point with 3 bits after the 1452 * binary point (i.e., scaled by 8). The following magic 1453 * is equivalent to the smoothing algorithm in rfc793 with 1454 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1455 * point). Adjust rtt to origin 0. 1456 */ 1457 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 1458 if ((tp->t_srtt += delta) <= 0) 1459 tp->t_srtt = 1; 1460 /* 1461 * We accumulate a smoothed rtt variance (actually, a 1462 * smoothed mean difference), then set the retransmit 1463 * timer to smoothed rtt + 4 times the smoothed variance. 1464 * rttvar is stored as fixed point with 2 bits after the 1465 * binary point (scaled by 4). The following is 1466 * equivalent to rfc793 smoothing with an alpha of .75 1467 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1468 * rfc793's wired-in beta. 1469 */ 1470 if (delta < 0) 1471 delta = -delta; 1472 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1473 if ((tp->t_rttvar += delta) <= 0) 1474 tp->t_rttvar = 1; 1475 } else { 1476 /* 1477 * No rtt measurement yet - use the unsmoothed rtt. 1478 * Set the variance to half the rtt (so our first 1479 * retransmit happens at 3*rtt). 1480 */ 1481 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 1482 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 1483 } 1484 tp->t_rtt = 0; 1485 tp->t_rxtshift = 0; 1486 1487 /* 1488 * the retransmit should happen at rtt + 4 * rttvar. 1489 * Because of the way we do the smoothing, srtt and rttvar 1490 * will each average +1/2 tick of bias. When we compute 1491 * the retransmit timer, we want 1/2 tick of rounding and 1492 * 1 extra tick because of +-1/2 tick uncertainty in the 1493 * firing of the timer. The bias will give us exactly the 1494 * 1.5 tick we need. But, because the bias is 1495 * statistical, we have to test that we don't drop below 1496 * the minimum feasible timer (which is 2 ticks). 1497 */ 1498 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 1499 rtt + 2, TCPTV_REXMTMAX); 1500 1501 /* 1502 * We received an ack for a packet that wasn't retransmitted; 1503 * it is probably safe to discard any error indications we've 1504 * received recently. This isn't quite right, but close enough 1505 * for now (a route might have failed after we sent a segment, 1506 * and the return path might not be symmetrical). 1507 */ 1508 tp->t_softerror = 0; 1509 } 1510 1511 /* 1512 * Determine a reasonable value for maxseg size. 1513 * If the route is known, check route for mtu. 1514 * If none, use an mss that can be handled on the outgoing 1515 * interface without forcing IP to fragment; if bigger than 1516 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 1517 * to utilize large mbufs. If no route is found, route has no mtu, 1518 * or the destination isn't local, use a default, hopefully conservative 1519 * size (usually 512 or the default IP max size, but no more than the mtu 1520 * of the interface), as we can't discover anything about intervening 1521 * gateways or networks. We also initialize the congestion/slow start 1522 * window to be a single segment if the destination isn't local. 1523 * While looking at the routing entry, we also initialize other path-dependent 1524 * parameters from pre-set or cached values in the routing entry. 1525 */ 1526 int 1527 tcp_mss(tp, offer) 1528 register struct tcpcb *tp; 1529 u_int offer; 1530 { 1531 struct route *ro; 1532 register struct rtentry *rt; 1533 struct ifnet *ifp; 1534 register int rtt, mss; 1535 u_long bufsize; 1536 struct inpcb *inp; 1537 struct socket *so; 1538 extern int tcp_mssdflt; 1539 1540 inp = tp->t_inpcb; 1541 ro = &inp->inp_route; 1542 1543 if ((rt = ro->ro_rt) == (struct rtentry *)0) { 1544 /* No route yet, so try to acquire one */ 1545 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1546 ro->ro_dst.sa_family = AF_INET; 1547 ro->ro_dst.sa_len = sizeof(ro->ro_dst); 1548 satosin(&ro->ro_dst)->sin_addr = inp->inp_faddr; 1549 rtalloc(ro); 1550 } 1551 if ((rt = ro->ro_rt) == (struct rtentry *)0) 1552 return (tcp_mssdflt); 1553 } 1554 ifp = rt->rt_ifp; 1555 so = inp->inp_socket; 1556 1557 #ifdef RTV_MTU /* if route characteristics exist ... */ 1558 /* 1559 * While we're here, check if there's an initial rtt 1560 * or rttvar. Convert from the route-table units 1561 * to scaled multiples of the slow timeout timer. 1562 */ 1563 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 1564 /* 1565 * XXX the lock bit for MTU indicates that the value 1566 * is also a minimum value; this is subject to time. 1567 */ 1568 if (rt->rt_rmx.rmx_locks & RTV_RTT) 1569 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ); 1570 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 1571 if (rt->rt_rmx.rmx_rttvar) 1572 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 1573 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 1574 else 1575 /* default variation is +- 1 rtt */ 1576 tp->t_rttvar = 1577 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 1578 TCPT_RANGESET(tp->t_rxtcur, 1579 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 1580 tp->t_rttmin, TCPTV_REXMTMAX); 1581 } 1582 /* 1583 * if there's an mtu associated with the route, use it 1584 */ 1585 if (rt->rt_rmx.rmx_mtu) 1586 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr); 1587 else 1588 #endif /* RTV_MTU */ 1589 { 1590 mss = ifp->if_mtu - sizeof(struct tcpiphdr); 1591 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1592 if (mss > MCLBYTES) 1593 mss &= ~(MCLBYTES-1); 1594 #else 1595 if (mss > MCLBYTES) 1596 mss = mss / MCLBYTES * MCLBYTES; 1597 #endif 1598 if (!in_localaddr(inp->inp_faddr)) 1599 mss = min(mss, tcp_mssdflt); 1600 } 1601 /* 1602 * The current mss, t_maxseg, is initialized to the default value. 1603 * If we compute a smaller value, reduce the current mss. 1604 * If we compute a larger value, return it for use in sending 1605 * a max seg size option, but don't store it for use 1606 * unless we received an offer at least that large from peer. 1607 * However, do not accept offers under 32 bytes. 1608 */ 1609 if (offer) 1610 mss = min(mss, offer); 1611 mss = max(mss, 32); /* sanity */ 1612 if (mss < tp->t_maxseg || offer != 0) { 1613 /* 1614 * If there's a pipesize, change the socket buffer 1615 * to that size. Make the socket buffers an integral 1616 * number of mss units; if the mss is larger than 1617 * the socket buffer, decrease the mss. 1618 */ 1619 #ifdef RTV_SPIPE 1620 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 1621 #endif 1622 bufsize = so->so_snd.sb_hiwat; 1623 if (bufsize < mss) 1624 mss = bufsize; 1625 else { 1626 bufsize = roundup(bufsize, mss); 1627 if (bufsize > sb_max) 1628 bufsize = sb_max; 1629 (void)sbreserve(&so->so_snd, bufsize); 1630 } 1631 tp->t_maxseg = mss; 1632 1633 #ifdef RTV_RPIPE 1634 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 1635 #endif 1636 bufsize = so->so_rcv.sb_hiwat; 1637 if (bufsize > mss) { 1638 bufsize = roundup(bufsize, mss); 1639 if (bufsize > sb_max) 1640 bufsize = sb_max; 1641 (void)sbreserve(&so->so_rcv, bufsize); 1642 } 1643 } 1644 tp->snd_cwnd = mss; 1645 1646 #ifdef RTV_SSTHRESH 1647 if (rt->rt_rmx.rmx_ssthresh) { 1648 /* 1649 * There's some sort of gateway or interface 1650 * buffer limit on the path. Use this to set 1651 * the slow start threshhold, but set the 1652 * threshold to no less than 2*mss. 1653 */ 1654 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 1655 } 1656 #endif /* RTV_MTU */ 1657 return (mss); 1658 } 1659 #endif /* TUBA_INCLUDE */ 1660