1 /* $NetBSD: tcp_input.c,v 1.71 1998/10/08 01:19:26 thorpej Exp $ */ 2 3 /*- 4 * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation 9 * Facility, NASA Ames Research Center. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the NetBSD 22 * Foundation, Inc. and its contributors. 23 * 4. Neither the name of The NetBSD Foundation nor the names of its 24 * contributors may be used to endorse or promote products derived 25 * from this software without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 37 * POSSIBILITY OF SUCH DAMAGE. 38 */ 39 40 /* 41 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 42 * The Regents of the University of California. All rights reserved. 43 * 44 * Redistribution and use in source and binary forms, with or without 45 * modification, are permitted provided that the following conditions 46 * are met: 47 * 1. Redistributions of source code must retain the above copyright 48 * notice, this list of conditions and the following disclaimer. 49 * 2. Redistributions in binary form must reproduce the above copyright 50 * notice, this list of conditions and the following disclaimer in the 51 * documentation and/or other materials provided with the distribution. 52 * 3. All advertising materials mentioning features or use of this software 53 * must display the following acknowledgement: 54 * This product includes software developed by the University of 55 * California, Berkeley and its contributors. 56 * 4. Neither the name of the University nor the names of its contributors 57 * may be used to endorse or promote products derived from this software 58 * without specific prior written permission. 59 * 60 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 61 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 62 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 63 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 64 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 65 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 66 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 67 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 68 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 69 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 70 * SUCH DAMAGE. 71 * 72 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 73 */ 74 75 /* 76 * TODO list for SYN cache stuff: 77 * 78 * Find room for a "state" field, which is needed to keep a 79 * compressed state for TIME_WAIT TCBs. It's been noted already 80 * that this is fairly important for very high-volume web and 81 * mail servers, which use a large number of short-lived 82 * connections. 83 */ 84 85 #include <sys/param.h> 86 #include <sys/systm.h> 87 #include <sys/malloc.h> 88 #include <sys/mbuf.h> 89 #include <sys/protosw.h> 90 #include <sys/socket.h> 91 #include <sys/socketvar.h> 92 #include <sys/errno.h> 93 #include <sys/syslog.h> 94 #include <sys/pool.h> 95 96 #include <net/if.h> 97 #include <net/route.h> 98 99 #include <netinet/in.h> 100 #include <netinet/in_systm.h> 101 #include <netinet/ip.h> 102 #include <netinet/in_pcb.h> 103 #include <netinet/ip_var.h> 104 #include <netinet/tcp.h> 105 #include <netinet/tcp_fsm.h> 106 #include <netinet/tcp_seq.h> 107 #include <netinet/tcp_timer.h> 108 #include <netinet/tcp_var.h> 109 #include <netinet/tcpip.h> 110 #include <netinet/tcp_debug.h> 111 112 #include <machine/stdarg.h> 113 114 int tcprexmtthresh = 3; 115 struct tcpiphdr tcp_saveti; 116 117 extern u_long sb_max; 118 119 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 120 121 /* for modulo comparisons of timestamps */ 122 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 123 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 124 125 /* 126 * Macro to compute ACK transmission behavior. Delay the ACK unless 127 * we have already delayed an ACK (must send an ACK every two segments). 128 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 129 * option is enabled. 130 */ 131 #define TCP_SETUP_ACK(tp, ti) \ 132 do { \ 133 if ((tp)->t_flags & TF_DELACK || \ 134 (tcp_ack_on_push && (ti)->ti_flags & TH_PUSH)) \ 135 tp->t_flags |= TF_ACKNOW; \ 136 else \ 137 TCP_SET_DELACK(tp); \ 138 } while (0) 139 140 /* 141 * Insert segment ti into reassembly queue of tcp with 142 * control block tp. Return TH_FIN if reassembly now includes 143 * a segment with FIN. The macro form does the common case inline 144 * (segment is the next to be received on an established connection, 145 * and the queue is empty), avoiding linkage into and removal 146 * from the queue and repetition of various conversions. 147 * Set DELACK for segments received in order, but ack immediately 148 * when segments are out of order (so fast retransmit can work). 149 */ 150 #define TCP_REASS(tp, ti, m, so, flags) { \ 151 if ((ti)->ti_seq == (tp)->rcv_nxt && \ 152 (tp)->segq.lh_first == NULL && \ 153 (tp)->t_state == TCPS_ESTABLISHED) { \ 154 TCP_SETUP_ACK(tp, ti); \ 155 (tp)->rcv_nxt += (ti)->ti_len; \ 156 flags = (ti)->ti_flags & TH_FIN; \ 157 tcpstat.tcps_rcvpack++;\ 158 tcpstat.tcps_rcvbyte += (ti)->ti_len;\ 159 sbappend(&(so)->so_rcv, (m)); \ 160 sorwakeup(so); \ 161 } else { \ 162 (flags) = tcp_reass((tp), (ti), (m)); \ 163 tp->t_flags |= TF_ACKNOW; \ 164 } \ 165 } 166 167 int 168 tcp_reass(tp, ti, m) 169 register struct tcpcb *tp; 170 register struct tcpiphdr *ti; 171 struct mbuf *m; 172 { 173 register struct ipqent *p, *q, *nq, *tiqe = NULL; 174 struct socket *so = tp->t_inpcb->inp_socket; 175 int pkt_flags; 176 tcp_seq pkt_seq; 177 unsigned pkt_len; 178 u_long rcvpartdupbyte = 0; 179 u_long rcvoobyte; 180 181 /* 182 * Call with ti==0 after become established to 183 * force pre-ESTABLISHED data up to user socket. 184 */ 185 if (ti == 0) 186 goto present; 187 188 rcvoobyte = ti->ti_len; 189 /* 190 * Copy these to local variables because the tcpiphdr 191 * gets munged while we are collapsing mbufs. 192 */ 193 pkt_seq = ti->ti_seq; 194 pkt_len = ti->ti_len; 195 pkt_flags = ti->ti_flags; 196 /* 197 * Find a segment which begins after this one does. 198 */ 199 for (p = NULL, q = tp->segq.lh_first; q != NULL; q = nq) { 200 nq = q->ipqe_q.le_next; 201 /* 202 * If the received segment is just right after this 203 * fragment, merge the two together and then check 204 * for further overlaps. 205 */ 206 if (q->ipqe_seq + q->ipqe_len == pkt_seq) { 207 #ifdef TCPREASS_DEBUG 208 printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n", 209 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 210 q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len); 211 #endif 212 pkt_len += q->ipqe_len; 213 pkt_flags |= q->ipqe_flags; 214 pkt_seq = q->ipqe_seq; 215 m_cat(q->ipqe_m, m); 216 m = q->ipqe_m; 217 goto free_ipqe; 218 } 219 /* 220 * If the received segment is completely past this 221 * fragment, we need to go the next fragment. 222 */ 223 if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 224 p = q; 225 continue; 226 } 227 /* 228 * If the fragment is past the received segment, 229 * it (or any following) can't be concatenated. 230 */ 231 if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) 232 break; 233 /* 234 * We've received all the data in this segment before. 235 * mark it as a duplicate and return. 236 */ 237 if (SEQ_LEQ(q->ipqe_seq, pkt_seq) && 238 SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 239 tcpstat.tcps_rcvduppack++; 240 tcpstat.tcps_rcvdupbyte += pkt_len; 241 m_freem(m); 242 if (tiqe != NULL) 243 pool_put(&ipqent_pool, tiqe); 244 return (0); 245 } 246 /* 247 * Received segment completely overlaps this fragment 248 * so we drop the fragment (this keeps the temporal 249 * ordering of segments correct). 250 */ 251 if (SEQ_GEQ(q->ipqe_seq, pkt_seq) && 252 SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) { 253 rcvpartdupbyte += q->ipqe_len; 254 m_freem(q->ipqe_m); 255 goto free_ipqe; 256 } 257 /* 258 * RX'ed segment extends past the end of the 259 * fragment. Drop the overlapping bytes. Then 260 * merge the fragment and segment then treat as 261 * a longer received packet. 262 */ 263 if (SEQ_LT(q->ipqe_seq, pkt_seq) 264 && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq)) { 265 int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq; 266 #ifdef TCPREASS_DEBUG 267 printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n", 268 tp, overlap, 269 pkt_seq, pkt_seq + pkt_len, pkt_len); 270 #endif 271 m_adj(m, overlap); 272 rcvpartdupbyte += overlap; 273 m_cat(q->ipqe_m, m); 274 m = q->ipqe_m; 275 pkt_seq = q->ipqe_seq; 276 pkt_len += q->ipqe_len - overlap; 277 rcvoobyte -= overlap; 278 goto free_ipqe; 279 } 280 /* 281 * RX'ed segment extends past the front of the 282 * fragment. Drop the overlapping bytes on the 283 * received packet. The packet will then be 284 * contatentated with this fragment a bit later. 285 */ 286 if (SEQ_GT(q->ipqe_seq, pkt_seq) 287 && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len)) { 288 int overlap = pkt_seq + pkt_len - q->ipqe_seq; 289 #ifdef TCPREASS_DEBUG 290 printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n", 291 tp, overlap, 292 pkt_seq, pkt_seq + pkt_len, pkt_len); 293 #endif 294 m_adj(m, -overlap); 295 pkt_len -= overlap; 296 rcvpartdupbyte += overlap; 297 rcvoobyte -= overlap; 298 } 299 /* 300 * If the received segment immediates precedes this 301 * fragment then tack the fragment onto this segment 302 * and reinsert the data. 303 */ 304 if (q->ipqe_seq == pkt_seq + pkt_len) { 305 #ifdef TCPREASS_DEBUG 306 printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n", 307 tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len, 308 pkt_seq, pkt_seq + pkt_len, pkt_len); 309 #endif 310 pkt_len += q->ipqe_len; 311 pkt_flags |= q->ipqe_flags; 312 m_cat(m, q->ipqe_m); 313 LIST_REMOVE(q, ipqe_q); 314 LIST_REMOVE(q, ipqe_timeq); 315 if (tiqe == NULL) { 316 tiqe = q; 317 } else { 318 pool_put(&ipqent_pool, q); 319 } 320 break; 321 } 322 /* 323 * If the fragment is before the segment, remember it. 324 * When this loop is terminated, p will contain the 325 * pointer to fragment that is right before the received 326 * segment. 327 */ 328 if (SEQ_LEQ(q->ipqe_seq, pkt_seq)) 329 p = q; 330 331 continue; 332 333 /* 334 * This is a common operation. It also will allow 335 * to save doing a malloc/free in most instances. 336 */ 337 free_ipqe: 338 LIST_REMOVE(q, ipqe_q); 339 LIST_REMOVE(q, ipqe_timeq); 340 if (tiqe == NULL) { 341 tiqe = q; 342 } else { 343 pool_put(&ipqent_pool, q); 344 } 345 } 346 347 /* 348 * Allocate a new queue entry since the received segment did not 349 * collapse onto any other out-of-order block; thus we are allocating 350 * a new block. If it had collapsed, tiqe would not be NULL and 351 * we would be reusing it. 352 * XXX If we can't, just drop the packet. XXX 353 */ 354 if (tiqe == NULL) { 355 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 356 if (tiqe == NULL) { 357 tcpstat.tcps_rcvmemdrop++; 358 m_freem(m); 359 return (0); 360 } 361 } 362 363 /* 364 * Update the counters. 365 */ 366 tcpstat.tcps_rcvoopack++; 367 tcpstat.tcps_rcvoobyte += rcvoobyte; 368 if (rcvpartdupbyte) { 369 tcpstat.tcps_rcvpartduppack++; 370 tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte; 371 } 372 373 /* 374 * Insert the new fragment queue entry into both queues. 375 */ 376 tiqe->ipqe_m = m; 377 tiqe->ipqe_seq = pkt_seq; 378 tiqe->ipqe_len = pkt_len; 379 tiqe->ipqe_flags = pkt_flags; 380 if (p == NULL) { 381 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 382 #ifdef TCPREASS_DEBUG 383 if (tiqe->ipqe_seq != tp->rcv_nxt) 384 printf("tcp_reass[%p]: insert %u:%u(%u) at front\n", 385 tp, pkt_seq, pkt_seq + pkt_len, pkt_len); 386 #endif 387 } else { 388 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 389 #ifdef TCPREASS_DEBUG 390 printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n", 391 tp, pkt_seq, pkt_seq + pkt_len, pkt_len, 392 p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len); 393 #endif 394 } 395 396 LIST_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq); 397 398 present: 399 /* 400 * Present data to user, advancing rcv_nxt through 401 * completed sequence space. 402 */ 403 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 404 return (0); 405 q = tp->segq.lh_first; 406 if (q == NULL || q->ipqe_seq != tp->rcv_nxt) 407 return (0); 408 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len) 409 return (0); 410 411 tp->rcv_nxt += q->ipqe_len; 412 pkt_flags = q->ipqe_flags & TH_FIN; 413 414 LIST_REMOVE(q, ipqe_q); 415 LIST_REMOVE(q, ipqe_timeq); 416 if (so->so_state & SS_CANTRCVMORE) 417 m_freem(q->ipqe_m); 418 else 419 sbappend(&so->so_rcv, q->ipqe_m); 420 pool_put(&ipqent_pool, q); 421 sorwakeup(so); 422 return (pkt_flags); 423 } 424 425 /* 426 * TCP input routine, follows pages 65-76 of the 427 * protocol specification dated September, 1981 very closely. 428 */ 429 void 430 #if __STDC__ 431 tcp_input(struct mbuf *m, ...) 432 #else 433 tcp_input(m, va_alist) 434 register struct mbuf *m; 435 #endif 436 { 437 register struct tcpiphdr *ti; 438 register struct inpcb *inp; 439 caddr_t optp = NULL; 440 int optlen = 0; 441 int len, tlen, off, hdroptlen; 442 register struct tcpcb *tp = 0; 443 register int tiflags; 444 struct socket *so = NULL; 445 int todrop, acked, ourfinisacked, needoutput = 0; 446 short ostate = 0; 447 int iss = 0; 448 u_long tiwin; 449 struct tcp_opt_info opti; 450 int iphlen; 451 va_list ap; 452 453 va_start(ap, m); 454 iphlen = va_arg(ap, int); 455 va_end(ap); 456 457 tcpstat.tcps_rcvtotal++; 458 459 opti.ts_present = 0; 460 opti.maxseg = 0; 461 462 /* 463 * Get IP and TCP header together in first mbuf. 464 * Note: IP leaves IP header in first mbuf. 465 */ 466 ti = mtod(m, struct tcpiphdr *); 467 if (iphlen > sizeof (struct ip)) 468 ip_stripoptions(m, (struct mbuf *)0); 469 if (m->m_len < sizeof (struct tcpiphdr)) { 470 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 471 tcpstat.tcps_rcvshort++; 472 return; 473 } 474 ti = mtod(m, struct tcpiphdr *); 475 } 476 477 /* 478 * Checksum extended TCP header and data. 479 */ 480 tlen = ((struct ip *)ti)->ip_len; 481 len = sizeof (struct ip) + tlen; 482 bzero(ti->ti_x1, sizeof ti->ti_x1); 483 ti->ti_len = (u_int16_t)tlen; 484 HTONS(ti->ti_len); 485 if ((ti->ti_sum = in_cksum(m, len)) != 0) { 486 tcpstat.tcps_rcvbadsum++; 487 goto drop; 488 } 489 490 /* 491 * Check that TCP offset makes sense, 492 * pull out TCP options and adjust length. XXX 493 */ 494 off = ti->ti_off << 2; 495 if (off < sizeof (struct tcphdr) || off > tlen) { 496 tcpstat.tcps_rcvbadoff++; 497 goto drop; 498 } 499 tlen -= off; 500 ti->ti_len = tlen; 501 if (off > sizeof (struct tcphdr)) { 502 if (m->m_len < sizeof(struct ip) + off) { 503 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 504 tcpstat.tcps_rcvshort++; 505 return; 506 } 507 ti = mtod(m, struct tcpiphdr *); 508 } 509 optlen = off - sizeof (struct tcphdr); 510 optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr); 511 /* 512 * Do quick retrieval of timestamp options ("options 513 * prediction?"). If timestamp is the only option and it's 514 * formatted as recommended in RFC 1323 appendix A, we 515 * quickly get the values now and not bother calling 516 * tcp_dooptions(), etc. 517 */ 518 if ((optlen == TCPOLEN_TSTAMP_APPA || 519 (optlen > TCPOLEN_TSTAMP_APPA && 520 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 521 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 522 (ti->ti_flags & TH_SYN) == 0) { 523 opti.ts_present = 1; 524 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 525 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 526 optp = NULL; /* we've parsed the options */ 527 } 528 } 529 tiflags = ti->ti_flags; 530 531 /* 532 * Convert TCP protocol specific fields to host format. 533 */ 534 NTOHL(ti->ti_seq); 535 NTOHL(ti->ti_ack); 536 NTOHS(ti->ti_win); 537 NTOHS(ti->ti_urp); 538 539 /* 540 * Locate pcb for segment. 541 */ 542 findpcb: 543 inp = in_pcblookup_connect(&tcbtable, ti->ti_src, ti->ti_sport, 544 ti->ti_dst, ti->ti_dport); 545 if (inp == 0) { 546 ++tcpstat.tcps_pcbhashmiss; 547 inp = in_pcblookup_bind(&tcbtable, ti->ti_dst, ti->ti_dport); 548 if (inp == 0) { 549 ++tcpstat.tcps_noport; 550 goto dropwithreset; 551 } 552 } 553 554 /* 555 * If the state is CLOSED (i.e., TCB does not exist) then 556 * all data in the incoming segment is discarded. 557 * If the TCB exists but is in CLOSED state, it is embryonic, 558 * but should either do a listen or a connect soon. 559 */ 560 tp = intotcpcb(inp); 561 if (tp == 0) 562 goto dropwithreset; 563 if (tp->t_state == TCPS_CLOSED) 564 goto drop; 565 566 /* Unscale the window into a 32-bit value. */ 567 if ((tiflags & TH_SYN) == 0) 568 tiwin = ti->ti_win << tp->snd_scale; 569 else 570 tiwin = ti->ti_win; 571 572 so = inp->inp_socket; 573 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 574 if (so->so_options & SO_DEBUG) { 575 ostate = tp->t_state; 576 tcp_saveti = *ti; 577 } 578 if (so->so_options & SO_ACCEPTCONN) { 579 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 580 if (tiflags & TH_RST) { 581 syn_cache_reset(ti); 582 } else if ((tiflags & (TH_ACK|TH_SYN)) == 583 (TH_ACK|TH_SYN)) { 584 /* 585 * Received a SYN,ACK. This should 586 * never happen while we are in 587 * LISTEN. Send an RST. 588 */ 589 goto badsyn; 590 } else if (tiflags & TH_ACK) { 591 so = syn_cache_get(so, m); 592 if (so == NULL) { 593 /* 594 * We don't have a SYN for 595 * this ACK; send an RST. 596 */ 597 goto badsyn; 598 } else if (so == 599 (struct socket *)(-1)) { 600 /* 601 * We were unable to create 602 * the connection. If the 603 * 3-way handshake was 604 * completed, and RST has 605 * been sent to the peer. 606 * Since the mbuf might be 607 * in use for the reply, 608 * do not free it. 609 */ 610 m = NULL; 611 } else { 612 /* 613 * We have created a 614 * full-blown connection. 615 */ 616 inp = sotoinpcb(so); 617 tp = intotcpcb(inp); 618 tiwin <<= tp->snd_scale; 619 goto after_listen; 620 } 621 } else { 622 /* 623 * None of RST, SYN or ACK was set. 624 * This is an invalid packet for a 625 * TCB in LISTEN state. Send a RST. 626 */ 627 goto badsyn; 628 } 629 } else { 630 /* 631 * Received a SYN. 632 */ 633 if (in_hosteq(ti->ti_src, ti->ti_dst) && 634 ti->ti_sport == ti->ti_dport) { 635 /* 636 * LISTEN socket received a SYN 637 * from itself? This can't possibly 638 * be valid; drop the packet. 639 */ 640 tcpstat.tcps_badsyn++; 641 goto drop; 642 } 643 /* 644 * SYN looks ok; create compressed TCP 645 * state for it. 646 */ 647 if (so->so_qlen <= so->so_qlimit && 648 syn_cache_add(so, m, optp, optlen, &opti)) 649 m = NULL; 650 } 651 goto drop; 652 } 653 } 654 655 after_listen: 656 #ifdef DIAGNOSTIC 657 /* 658 * Should not happen now that all embryonic connections 659 * are handled with compressed state. 660 */ 661 if (tp->t_state == TCPS_LISTEN) 662 panic("tcp_input: TCPS_LISTEN"); 663 #endif 664 665 /* 666 * Segment received on connection. 667 * Reset idle time and keep-alive timer. 668 */ 669 tp->t_idle = 0; 670 if (TCPS_HAVEESTABLISHED(tp->t_state)) 671 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 672 673 /* 674 * Process options. 675 */ 676 if (optp) 677 tcp_dooptions(tp, optp, optlen, ti, &opti); 678 679 /* 680 * Header prediction: check for the two common cases 681 * of a uni-directional data xfer. If the packet has 682 * no control flags, is in-sequence, the window didn't 683 * change and we're not retransmitting, it's a 684 * candidate. If the length is zero and the ack moved 685 * forward, we're the sender side of the xfer. Just 686 * free the data acked & wake any higher level process 687 * that was blocked waiting for space. If the length 688 * is non-zero and the ack didn't move, we're the 689 * receiver side. If we're getting packets in-order 690 * (the reassembly queue is empty), add the data to 691 * the socket buffer and note that we need a delayed ack. 692 */ 693 if (tp->t_state == TCPS_ESTABLISHED && 694 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 695 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 696 ti->ti_seq == tp->rcv_nxt && 697 tiwin && tiwin == tp->snd_wnd && 698 tp->snd_nxt == tp->snd_max) { 699 700 /* 701 * If last ACK falls within this segment's sequence numbers, 702 * record the timestamp. 703 */ 704 if (opti.ts_present && 705 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 706 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) { 707 tp->ts_recent_age = tcp_now; 708 tp->ts_recent = opti.ts_val; 709 } 710 711 if (ti->ti_len == 0) { 712 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 713 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 714 tp->snd_cwnd >= tp->snd_wnd && 715 tp->t_dupacks < tcprexmtthresh) { 716 /* 717 * this is a pure ack for outstanding data. 718 */ 719 ++tcpstat.tcps_predack; 720 if (opti.ts_present) 721 tcp_xmit_timer(tp, 722 tcp_now-opti.ts_ecr+1); 723 else if (tp->t_rtt && 724 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 725 tcp_xmit_timer(tp, tp->t_rtt); 726 acked = ti->ti_ack - tp->snd_una; 727 tcpstat.tcps_rcvackpack++; 728 tcpstat.tcps_rcvackbyte += acked; 729 sbdrop(&so->so_snd, acked); 730 tp->snd_una = ti->ti_ack; 731 m_freem(m); 732 733 /* 734 * If all outstanding data are acked, stop 735 * retransmit timer, otherwise restart timer 736 * using current (possibly backed-off) value. 737 * If process is waiting for space, 738 * wakeup/selwakeup/signal. If data 739 * are ready to send, let tcp_output 740 * decide between more output or persist. 741 */ 742 if (tp->snd_una == tp->snd_max) 743 TCP_TIMER_DISARM(tp, TCPT_REXMT); 744 else if (TCP_TIMER_ISARMED(tp, 745 TCPT_PERSIST) == 0) 746 TCP_TIMER_ARM(tp, TCPT_REXMT, 747 tp->t_rxtcur); 748 749 sowwakeup(so); 750 if (so->so_snd.sb_cc) 751 (void) tcp_output(tp); 752 return; 753 } 754 } else if (ti->ti_ack == tp->snd_una && 755 tp->segq.lh_first == NULL && 756 ti->ti_len <= sbspace(&so->so_rcv)) { 757 /* 758 * this is a pure, in-sequence data packet 759 * with nothing on the reassembly queue and 760 * we have enough buffer space to take it. 761 */ 762 ++tcpstat.tcps_preddat; 763 tp->rcv_nxt += ti->ti_len; 764 tcpstat.tcps_rcvpack++; 765 tcpstat.tcps_rcvbyte += ti->ti_len; 766 /* 767 * Drop TCP, IP headers and TCP options then add data 768 * to socket buffer. 769 */ 770 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 771 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 772 sbappend(&so->so_rcv, m); 773 sorwakeup(so); 774 TCP_SETUP_ACK(tp, ti); 775 if (tp->t_flags & TF_ACKNOW) 776 (void) tcp_output(tp); 777 return; 778 } 779 } 780 781 /* 782 * Drop TCP, IP headers and TCP options. 783 */ 784 hdroptlen = sizeof(struct tcpiphdr) + off - sizeof(struct tcphdr); 785 m->m_data += hdroptlen; 786 m->m_len -= hdroptlen; 787 788 /* 789 * Calculate amount of space in receive window, 790 * and then do TCP input processing. 791 * Receive window is amount of space in rcv queue, 792 * but not less than advertised window. 793 */ 794 { int win; 795 796 win = sbspace(&so->so_rcv); 797 if (win < 0) 798 win = 0; 799 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 800 } 801 802 switch (tp->t_state) { 803 804 /* 805 * If the state is SYN_SENT: 806 * if seg contains an ACK, but not for our SYN, drop the input. 807 * if seg contains a RST, then drop the connection. 808 * if seg does not contain SYN, then drop it. 809 * Otherwise this is an acceptable SYN segment 810 * initialize tp->rcv_nxt and tp->irs 811 * if seg contains ack then advance tp->snd_una 812 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 813 * arrange for segment to be acked (eventually) 814 * continue processing rest of data/controls, beginning with URG 815 */ 816 case TCPS_SYN_SENT: 817 if ((tiflags & TH_ACK) && 818 (SEQ_LEQ(ti->ti_ack, tp->iss) || 819 SEQ_GT(ti->ti_ack, tp->snd_max))) 820 goto dropwithreset; 821 if (tiflags & TH_RST) { 822 if (tiflags & TH_ACK) 823 tp = tcp_drop(tp, ECONNREFUSED); 824 goto drop; 825 } 826 if ((tiflags & TH_SYN) == 0) 827 goto drop; 828 if (tiflags & TH_ACK) { 829 tp->snd_una = ti->ti_ack; 830 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 831 tp->snd_nxt = tp->snd_una; 832 } 833 TCP_TIMER_DISARM(tp, TCPT_REXMT); 834 tp->irs = ti->ti_seq; 835 tcp_rcvseqinit(tp); 836 tp->t_flags |= TF_ACKNOW; 837 tcp_mss_from_peer(tp, opti.maxseg); 838 839 /* 840 * Initialize the initial congestion window. If we 841 * had to retransmit the SYN, we must initialize cwnd 842 * to 1 segment (i.e. the Loss Window). 843 */ 844 if (tp->t_flags & TF_SYN_REXMT) 845 tp->snd_cwnd = tp->t_peermss; 846 else 847 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 848 tp->t_peermss); 849 850 tcp_rmx_rtt(tp); 851 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 852 tcpstat.tcps_connects++; 853 soisconnected(so); 854 tcp_established(tp); 855 /* Do window scaling on this connection? */ 856 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 857 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 858 tp->snd_scale = tp->requested_s_scale; 859 tp->rcv_scale = tp->request_r_scale; 860 } 861 (void) tcp_reass(tp, (struct tcpiphdr *)0, 862 (struct mbuf *)0); 863 /* 864 * if we didn't have to retransmit the SYN, 865 * use its rtt as our initial srtt & rtt var. 866 */ 867 if (tp->t_rtt) 868 tcp_xmit_timer(tp, tp->t_rtt); 869 } else 870 tp->t_state = TCPS_SYN_RECEIVED; 871 872 /* 873 * Advance ti->ti_seq to correspond to first data byte. 874 * If data, trim to stay within window, 875 * dropping FIN if necessary. 876 */ 877 ti->ti_seq++; 878 if (ti->ti_len > tp->rcv_wnd) { 879 todrop = ti->ti_len - tp->rcv_wnd; 880 m_adj(m, -todrop); 881 ti->ti_len = tp->rcv_wnd; 882 tiflags &= ~TH_FIN; 883 tcpstat.tcps_rcvpackafterwin++; 884 tcpstat.tcps_rcvbyteafterwin += todrop; 885 } 886 tp->snd_wl1 = ti->ti_seq - 1; 887 tp->rcv_up = ti->ti_seq; 888 goto step6; 889 890 /* 891 * If the state is SYN_RECEIVED: 892 * If seg contains an ACK, but not for our SYN, drop the input 893 * and generate an RST. See page 36, rfc793 894 */ 895 case TCPS_SYN_RECEIVED: 896 if ((tiflags & TH_ACK) && 897 (SEQ_LEQ(ti->ti_ack, tp->iss) || 898 SEQ_GT(ti->ti_ack, tp->snd_max))) 899 goto dropwithreset; 900 break; 901 } 902 903 /* 904 * States other than LISTEN or SYN_SENT. 905 * First check timestamp, if present. 906 * Then check that at least some bytes of segment are within 907 * receive window. If segment begins before rcv_nxt, 908 * drop leading data (and SYN); if nothing left, just ack. 909 * 910 * RFC 1323 PAWS: If we have a timestamp reply on this segment 911 * and it's less than ts_recent, drop it. 912 */ 913 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 914 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 915 916 /* Check to see if ts_recent is over 24 days old. */ 917 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 918 /* 919 * Invalidate ts_recent. If this segment updates 920 * ts_recent, the age will be reset later and ts_recent 921 * will get a valid value. If it does not, setting 922 * ts_recent to zero will at least satisfy the 923 * requirement that zero be placed in the timestamp 924 * echo reply when ts_recent isn't valid. The 925 * age isn't reset until we get a valid ts_recent 926 * because we don't want out-of-order segments to be 927 * dropped when ts_recent is old. 928 */ 929 tp->ts_recent = 0; 930 } else { 931 tcpstat.tcps_rcvduppack++; 932 tcpstat.tcps_rcvdupbyte += ti->ti_len; 933 tcpstat.tcps_pawsdrop++; 934 goto dropafterack; 935 } 936 } 937 938 todrop = tp->rcv_nxt - ti->ti_seq; 939 if (todrop > 0) { 940 if (tiflags & TH_SYN) { 941 tiflags &= ~TH_SYN; 942 ti->ti_seq++; 943 if (ti->ti_urp > 1) 944 ti->ti_urp--; 945 else { 946 tiflags &= ~TH_URG; 947 ti->ti_urp = 0; 948 } 949 todrop--; 950 } 951 if (todrop > ti->ti_len || 952 (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 953 /* 954 * Any valid FIN must be to the left of the window. 955 * At this point the FIN must be a duplicate or 956 * out of sequence; drop it. 957 */ 958 tiflags &= ~TH_FIN; 959 /* 960 * Send an ACK to resynchronize and drop any data. 961 * But keep on processing for RST or ACK. 962 */ 963 tp->t_flags |= TF_ACKNOW; 964 todrop = ti->ti_len; 965 tcpstat.tcps_rcvdupbyte += todrop; 966 tcpstat.tcps_rcvduppack++; 967 } else { 968 tcpstat.tcps_rcvpartduppack++; 969 tcpstat.tcps_rcvpartdupbyte += todrop; 970 } 971 m_adj(m, todrop); 972 ti->ti_seq += todrop; 973 ti->ti_len -= todrop; 974 if (ti->ti_urp > todrop) 975 ti->ti_urp -= todrop; 976 else { 977 tiflags &= ~TH_URG; 978 ti->ti_urp = 0; 979 } 980 } 981 982 /* 983 * If new data are received on a connection after the 984 * user processes are gone, then RST the other end. 985 */ 986 if ((so->so_state & SS_NOFDREF) && 987 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 988 tp = tcp_close(tp); 989 tcpstat.tcps_rcvafterclose++; 990 goto dropwithreset; 991 } 992 993 /* 994 * If segment ends after window, drop trailing data 995 * (and PUSH and FIN); if nothing left, just ACK. 996 */ 997 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 998 if (todrop > 0) { 999 tcpstat.tcps_rcvpackafterwin++; 1000 if (todrop >= ti->ti_len) { 1001 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 1002 /* 1003 * If a new connection request is received 1004 * while in TIME_WAIT, drop the old connection 1005 * and start over if the sequence numbers 1006 * are above the previous ones. 1007 */ 1008 if (tiflags & TH_SYN && 1009 tp->t_state == TCPS_TIME_WAIT && 1010 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1011 iss = tcp_new_iss(tp, sizeof(struct tcpcb), 1012 tp->rcv_nxt); 1013 tp = tcp_close(tp); 1014 /* 1015 * We have already advanced the mbuf 1016 * pointers past the IP+TCP headers and 1017 * options. Restore those pointers before 1018 * attempting to use the TCP header again. 1019 */ 1020 m->m_data -= hdroptlen; 1021 m->m_len += hdroptlen; 1022 goto findpcb; 1023 } 1024 /* 1025 * If window is closed can only take segments at 1026 * window edge, and have to drop data and PUSH from 1027 * incoming segments. Continue processing, but 1028 * remember to ack. Otherwise, drop segment 1029 * and ack. 1030 */ 1031 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1032 tp->t_flags |= TF_ACKNOW; 1033 tcpstat.tcps_rcvwinprobe++; 1034 } else 1035 goto dropafterack; 1036 } else 1037 tcpstat.tcps_rcvbyteafterwin += todrop; 1038 m_adj(m, -todrop); 1039 ti->ti_len -= todrop; 1040 tiflags &= ~(TH_PUSH|TH_FIN); 1041 } 1042 1043 /* 1044 * If last ACK falls within this segment's sequence numbers, 1045 * and the timestamp is newer, record it. 1046 */ 1047 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1048 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 1049 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + 1050 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1051 tp->ts_recent_age = tcp_now; 1052 tp->ts_recent = opti.ts_val; 1053 } 1054 1055 /* 1056 * If the RST bit is set examine the state: 1057 * SYN_RECEIVED STATE: 1058 * If passive open, return to LISTEN state. 1059 * If active open, inform user that connection was refused. 1060 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1061 * Inform user that connection was reset, and close tcb. 1062 * CLOSING, LAST_ACK, TIME_WAIT STATES 1063 * Close the tcb. 1064 */ 1065 if (tiflags&TH_RST) switch (tp->t_state) { 1066 1067 case TCPS_SYN_RECEIVED: 1068 so->so_error = ECONNREFUSED; 1069 goto close; 1070 1071 case TCPS_ESTABLISHED: 1072 case TCPS_FIN_WAIT_1: 1073 case TCPS_FIN_WAIT_2: 1074 case TCPS_CLOSE_WAIT: 1075 so->so_error = ECONNRESET; 1076 close: 1077 tp->t_state = TCPS_CLOSED; 1078 tcpstat.tcps_drops++; 1079 tp = tcp_close(tp); 1080 goto drop; 1081 1082 case TCPS_CLOSING: 1083 case TCPS_LAST_ACK: 1084 case TCPS_TIME_WAIT: 1085 tp = tcp_close(tp); 1086 goto drop; 1087 } 1088 1089 /* 1090 * If a SYN is in the window, then this is an 1091 * error and we send an RST and drop the connection. 1092 */ 1093 if (tiflags & TH_SYN) { 1094 tp = tcp_drop(tp, ECONNRESET); 1095 goto dropwithreset; 1096 } 1097 1098 /* 1099 * If the ACK bit is off we drop the segment and return. 1100 */ 1101 if ((tiflags & TH_ACK) == 0) 1102 goto drop; 1103 1104 /* 1105 * Ack processing. 1106 */ 1107 switch (tp->t_state) { 1108 1109 /* 1110 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1111 * ESTABLISHED state and continue processing, otherwise 1112 * send an RST. 1113 */ 1114 case TCPS_SYN_RECEIVED: 1115 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1116 SEQ_GT(ti->ti_ack, tp->snd_max)) 1117 goto dropwithreset; 1118 tcpstat.tcps_connects++; 1119 soisconnected(so); 1120 tcp_established(tp); 1121 /* Do window scaling? */ 1122 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1123 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1124 tp->snd_scale = tp->requested_s_scale; 1125 tp->rcv_scale = tp->request_r_scale; 1126 } 1127 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 1128 tp->snd_wl1 = ti->ti_seq - 1; 1129 /* fall into ... */ 1130 1131 /* 1132 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1133 * ACKs. If the ack is in the range 1134 * tp->snd_una < ti->ti_ack <= tp->snd_max 1135 * then advance tp->snd_una to ti->ti_ack and drop 1136 * data from the retransmission queue. If this ACK reflects 1137 * more up to date window information we update our window information. 1138 */ 1139 case TCPS_ESTABLISHED: 1140 case TCPS_FIN_WAIT_1: 1141 case TCPS_FIN_WAIT_2: 1142 case TCPS_CLOSE_WAIT: 1143 case TCPS_CLOSING: 1144 case TCPS_LAST_ACK: 1145 case TCPS_TIME_WAIT: 1146 1147 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1148 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1149 tcpstat.tcps_rcvdupack++; 1150 /* 1151 * If we have outstanding data (other than 1152 * a window probe), this is a completely 1153 * duplicate ack (ie, window info didn't 1154 * change), the ack is the biggest we've 1155 * seen and we've seen exactly our rexmt 1156 * threshhold of them, assume a packet 1157 * has been dropped and retransmit it. 1158 * Kludge snd_nxt & the congestion 1159 * window so we send only this one 1160 * packet. 1161 * 1162 * We know we're losing at the current 1163 * window size so do congestion avoidance 1164 * (set ssthresh to half the current window 1165 * and pull our congestion window back to 1166 * the new ssthresh). 1167 * 1168 * Dup acks mean that packets have left the 1169 * network (they're now cached at the receiver) 1170 * so bump cwnd by the amount in the receiver 1171 * to keep a constant cwnd packets in the 1172 * network. 1173 */ 1174 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1175 ti->ti_ack != tp->snd_una) 1176 tp->t_dupacks = 0; 1177 else if (++tp->t_dupacks == tcprexmtthresh) { 1178 tcp_seq onxt = tp->snd_nxt; 1179 u_int win = 1180 min(tp->snd_wnd, tp->snd_cwnd) / 1181 2 / tp->t_segsz; 1182 if (SEQ_LT(ti->ti_ack, tp->snd_recover)) { 1183 /* 1184 * False fast retransmit after 1185 * timeout. Do not cut window. 1186 */ 1187 tp->snd_cwnd += tp->t_segsz; 1188 tp->t_dupacks = 0; 1189 (void) tcp_output(tp); 1190 goto drop; 1191 } 1192 1193 if (win < 2) 1194 win = 2; 1195 tp->snd_ssthresh = win * tp->t_segsz; 1196 tp->snd_recover = tp->snd_max; 1197 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1198 tp->t_rtt = 0; 1199 tp->snd_nxt = ti->ti_ack; 1200 tp->snd_cwnd = tp->t_segsz; 1201 (void) tcp_output(tp); 1202 tp->snd_cwnd = tp->snd_ssthresh + 1203 tp->t_segsz * tp->t_dupacks; 1204 if (SEQ_GT(onxt, tp->snd_nxt)) 1205 tp->snd_nxt = onxt; 1206 goto drop; 1207 } else if (tp->t_dupacks > tcprexmtthresh) { 1208 tp->snd_cwnd += tp->t_segsz; 1209 (void) tcp_output(tp); 1210 goto drop; 1211 } 1212 } else 1213 tp->t_dupacks = 0; 1214 break; 1215 } 1216 /* 1217 * If the congestion window was inflated to account 1218 * for the other side's cached packets, retract it. 1219 */ 1220 if (!tcp_do_newreno) { 1221 if (tp->t_dupacks >= tcprexmtthresh && 1222 tp->snd_cwnd > tp->snd_ssthresh) 1223 tp->snd_cwnd = tp->snd_ssthresh; 1224 tp->t_dupacks = 0; 1225 } else if (tp->t_dupacks >= tcprexmtthresh 1226 && !tcp_newreno(tp, ti)) { 1227 tp->snd_cwnd = tp->snd_ssthresh; 1228 /* 1229 * Window inflation should have left us with approx. 1230 * snd_ssthresh outstanding data. But in case we 1231 * would be inclined to send a burst, better to do 1232 * it via the slow start mechanism. 1233 */ 1234 if (SEQ_SUB(tp->snd_max, ti->ti_ack) < tp->snd_ssthresh) 1235 tp->snd_cwnd = SEQ_SUB(tp->snd_max, ti->ti_ack) 1236 + tp->t_segsz; 1237 tp->t_dupacks = 0; 1238 } 1239 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1240 tcpstat.tcps_rcvacktoomuch++; 1241 goto dropafterack; 1242 } 1243 acked = ti->ti_ack - tp->snd_una; 1244 tcpstat.tcps_rcvackpack++; 1245 tcpstat.tcps_rcvackbyte += acked; 1246 1247 /* 1248 * If we have a timestamp reply, update smoothed 1249 * round trip time. If no timestamp is present but 1250 * transmit timer is running and timed sequence 1251 * number was acked, update smoothed round trip time. 1252 * Since we now have an rtt measurement, cancel the 1253 * timer backoff (cf., Phil Karn's retransmit alg.). 1254 * Recompute the initial retransmit timer. 1255 */ 1256 if (opti.ts_present) 1257 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr + 1); 1258 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1259 tcp_xmit_timer(tp,tp->t_rtt); 1260 1261 /* 1262 * If all outstanding data is acked, stop retransmit 1263 * timer and remember to restart (more output or persist). 1264 * If there is more data to be acked, restart retransmit 1265 * timer, using current (possibly backed-off) value. 1266 */ 1267 if (ti->ti_ack == tp->snd_max) { 1268 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1269 needoutput = 1; 1270 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1271 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1272 /* 1273 * When new data is acked, open the congestion window. 1274 * If the window gives us less than ssthresh packets 1275 * in flight, open exponentially (segsz per packet). 1276 * Otherwise open linearly: segsz per window 1277 * (segsz^2 / cwnd per packet), plus a constant 1278 * fraction of a packet (segsz/8) to help larger windows 1279 * open quickly enough. 1280 */ 1281 { 1282 register u_int cw = tp->snd_cwnd; 1283 register u_int incr = tp->t_segsz; 1284 1285 if (cw > tp->snd_ssthresh) 1286 incr = incr * incr / cw; 1287 if (!tcp_do_newreno || SEQ_GEQ(ti->ti_ack, tp->snd_recover)) 1288 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1289 } 1290 if (acked > so->so_snd.sb_cc) { 1291 tp->snd_wnd -= so->so_snd.sb_cc; 1292 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1293 ourfinisacked = 1; 1294 } else { 1295 sbdrop(&so->so_snd, acked); 1296 tp->snd_wnd -= acked; 1297 ourfinisacked = 0; 1298 } 1299 sowwakeup(so); 1300 tp->snd_una = ti->ti_ack; 1301 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1302 tp->snd_nxt = tp->snd_una; 1303 1304 switch (tp->t_state) { 1305 1306 /* 1307 * In FIN_WAIT_1 STATE in addition to the processing 1308 * for the ESTABLISHED state if our FIN is now acknowledged 1309 * then enter FIN_WAIT_2. 1310 */ 1311 case TCPS_FIN_WAIT_1: 1312 if (ourfinisacked) { 1313 /* 1314 * If we can't receive any more 1315 * data, then closing user can proceed. 1316 * Starting the timer is contrary to the 1317 * specification, but if we don't get a FIN 1318 * we'll hang forever. 1319 */ 1320 if (so->so_state & SS_CANTRCVMORE) { 1321 soisdisconnected(so); 1322 if (tcp_maxidle > 0) 1323 TCP_TIMER_ARM(tp, TCPT_2MSL, 1324 tcp_maxidle); 1325 } 1326 tp->t_state = TCPS_FIN_WAIT_2; 1327 } 1328 break; 1329 1330 /* 1331 * In CLOSING STATE in addition to the processing for 1332 * the ESTABLISHED state if the ACK acknowledges our FIN 1333 * then enter the TIME-WAIT state, otherwise ignore 1334 * the segment. 1335 */ 1336 case TCPS_CLOSING: 1337 if (ourfinisacked) { 1338 tp->t_state = TCPS_TIME_WAIT; 1339 tcp_canceltimers(tp); 1340 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1341 soisdisconnected(so); 1342 } 1343 break; 1344 1345 /* 1346 * In LAST_ACK, we may still be waiting for data to drain 1347 * and/or to be acked, as well as for the ack of our FIN. 1348 * If our FIN is now acknowledged, delete the TCB, 1349 * enter the closed state and return. 1350 */ 1351 case TCPS_LAST_ACK: 1352 if (ourfinisacked) { 1353 tp = tcp_close(tp); 1354 goto drop; 1355 } 1356 break; 1357 1358 /* 1359 * In TIME_WAIT state the only thing that should arrive 1360 * is a retransmission of the remote FIN. Acknowledge 1361 * it and restart the finack timer. 1362 */ 1363 case TCPS_TIME_WAIT: 1364 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1365 goto dropafterack; 1366 } 1367 } 1368 1369 step6: 1370 /* 1371 * Update window information. 1372 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1373 */ 1374 if (((tiflags & TH_ACK) && SEQ_LT(tp->snd_wl1, ti->ti_seq)) || 1375 (tp->snd_wl1 == ti->ti_seq && SEQ_LT(tp->snd_wl2, ti->ti_ack)) || 1376 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)) { 1377 /* keep track of pure window updates */ 1378 if (ti->ti_len == 0 && 1379 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1380 tcpstat.tcps_rcvwinupd++; 1381 tp->snd_wnd = tiwin; 1382 tp->snd_wl1 = ti->ti_seq; 1383 tp->snd_wl2 = ti->ti_ack; 1384 if (tp->snd_wnd > tp->max_sndwnd) 1385 tp->max_sndwnd = tp->snd_wnd; 1386 needoutput = 1; 1387 } 1388 1389 /* 1390 * Process segments with URG. 1391 */ 1392 if ((tiflags & TH_URG) && ti->ti_urp && 1393 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1394 /* 1395 * This is a kludge, but if we receive and accept 1396 * random urgent pointers, we'll crash in 1397 * soreceive. It's hard to imagine someone 1398 * actually wanting to send this much urgent data. 1399 */ 1400 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1401 ti->ti_urp = 0; /* XXX */ 1402 tiflags &= ~TH_URG; /* XXX */ 1403 goto dodata; /* XXX */ 1404 } 1405 /* 1406 * If this segment advances the known urgent pointer, 1407 * then mark the data stream. This should not happen 1408 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1409 * a FIN has been received from the remote side. 1410 * In these states we ignore the URG. 1411 * 1412 * According to RFC961 (Assigned Protocols), 1413 * the urgent pointer points to the last octet 1414 * of urgent data. We continue, however, 1415 * to consider it to indicate the first octet 1416 * of data past the urgent section as the original 1417 * spec states (in one of two places). 1418 */ 1419 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1420 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1421 so->so_oobmark = so->so_rcv.sb_cc + 1422 (tp->rcv_up - tp->rcv_nxt) - 1; 1423 if (so->so_oobmark == 0) 1424 so->so_state |= SS_RCVATMARK; 1425 sohasoutofband(so); 1426 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1427 } 1428 /* 1429 * Remove out of band data so doesn't get presented to user. 1430 * This can happen independent of advancing the URG pointer, 1431 * but if two URG's are pending at once, some out-of-band 1432 * data may creep in... ick. 1433 */ 1434 if (ti->ti_urp <= (u_int16_t) ti->ti_len 1435 #ifdef SO_OOBINLINE 1436 && (so->so_options & SO_OOBINLINE) == 0 1437 #endif 1438 ) 1439 tcp_pulloutofband(so, ti, m); 1440 } else 1441 /* 1442 * If no out of band data is expected, 1443 * pull receive urgent pointer along 1444 * with the receive window. 1445 */ 1446 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1447 tp->rcv_up = tp->rcv_nxt; 1448 dodata: /* XXX */ 1449 1450 /* 1451 * Process the segment text, merging it into the TCP sequencing queue, 1452 * and arranging for acknowledgment of receipt if necessary. 1453 * This process logically involves adjusting tp->rcv_wnd as data 1454 * is presented to the user (this happens in tcp_usrreq.c, 1455 * case PRU_RCVD). If a FIN has already been received on this 1456 * connection then we just ignore the text. 1457 */ 1458 if ((ti->ti_len || (tiflags & TH_FIN)) && 1459 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1460 TCP_REASS(tp, ti, m, so, tiflags); 1461 /* 1462 * Note the amount of data that peer has sent into 1463 * our window, in order to estimate the sender's 1464 * buffer size. 1465 */ 1466 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1467 } else { 1468 m_freem(m); 1469 tiflags &= ~TH_FIN; 1470 } 1471 1472 /* 1473 * If FIN is received ACK the FIN and let the user know 1474 * that the connection is closing. Ignore a FIN received before 1475 * the connection is fully established. 1476 */ 1477 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 1478 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1479 socantrcvmore(so); 1480 tp->t_flags |= TF_ACKNOW; 1481 tp->rcv_nxt++; 1482 } 1483 switch (tp->t_state) { 1484 1485 /* 1486 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 1487 */ 1488 case TCPS_ESTABLISHED: 1489 tp->t_state = TCPS_CLOSE_WAIT; 1490 break; 1491 1492 /* 1493 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1494 * enter the CLOSING state. 1495 */ 1496 case TCPS_FIN_WAIT_1: 1497 tp->t_state = TCPS_CLOSING; 1498 break; 1499 1500 /* 1501 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1502 * starting the time-wait timer, turning off the other 1503 * standard timers. 1504 */ 1505 case TCPS_FIN_WAIT_2: 1506 tp->t_state = TCPS_TIME_WAIT; 1507 tcp_canceltimers(tp); 1508 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1509 soisdisconnected(so); 1510 break; 1511 1512 /* 1513 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1514 */ 1515 case TCPS_TIME_WAIT: 1516 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1517 break; 1518 } 1519 } 1520 if (so->so_options & SO_DEBUG) 1521 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1522 1523 /* 1524 * Return any desired output. 1525 */ 1526 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1527 (void) tcp_output(tp); 1528 return; 1529 1530 badsyn: 1531 /* 1532 * Received a bad SYN. Increment counters and dropwithreset. 1533 */ 1534 tcpstat.tcps_badsyn++; 1535 tp = NULL; 1536 goto dropwithreset; 1537 1538 dropafterack: 1539 /* 1540 * Generate an ACK dropping incoming segment if it occupies 1541 * sequence space, where the ACK reflects our state. 1542 */ 1543 if (tiflags & TH_RST) 1544 goto drop; 1545 m_freem(m); 1546 tp->t_flags |= TF_ACKNOW; 1547 (void) tcp_output(tp); 1548 return; 1549 1550 dropwithreset: 1551 /* 1552 * Generate a RST, dropping incoming segment. 1553 * Make ACK acceptable to originator of segment. 1554 * Don't bother to respond if destination was broadcast/multicast. 1555 */ 1556 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1557 IN_MULTICAST(ti->ti_dst.s_addr)) 1558 goto drop; 1559 if (tiflags & TH_ACK) 1560 (void)tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1561 else { 1562 if (tiflags & TH_SYN) 1563 ti->ti_len++; 1564 (void)tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1565 TH_RST|TH_ACK); 1566 } 1567 return; 1568 1569 drop: 1570 /* 1571 * Drop space held by incoming segment and return. 1572 */ 1573 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1574 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1575 m_freem(m); 1576 return; 1577 } 1578 1579 void 1580 tcp_dooptions(tp, cp, cnt, ti, oi) 1581 struct tcpcb *tp; 1582 u_char *cp; 1583 int cnt; 1584 struct tcpiphdr *ti; 1585 struct tcp_opt_info *oi; 1586 { 1587 u_int16_t mss; 1588 int opt, optlen; 1589 1590 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1591 opt = cp[0]; 1592 if (opt == TCPOPT_EOL) 1593 break; 1594 if (opt == TCPOPT_NOP) 1595 optlen = 1; 1596 else { 1597 optlen = cp[1]; 1598 if (optlen <= 0) 1599 break; 1600 } 1601 switch (opt) { 1602 1603 default: 1604 continue; 1605 1606 case TCPOPT_MAXSEG: 1607 if (optlen != TCPOLEN_MAXSEG) 1608 continue; 1609 if (!(ti->ti_flags & TH_SYN)) 1610 continue; 1611 bcopy(cp + 2, &mss, sizeof(mss)); 1612 oi->maxseg = ntohs(mss); 1613 break; 1614 1615 case TCPOPT_WINDOW: 1616 if (optlen != TCPOLEN_WINDOW) 1617 continue; 1618 if (!(ti->ti_flags & TH_SYN)) 1619 continue; 1620 tp->t_flags |= TF_RCVD_SCALE; 1621 tp->requested_s_scale = cp[2]; 1622 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 1623 log(LOG_ERR, "TCP: invalid wscale %d from " 1624 "0x%08x, assuming %d\n", 1625 tp->requested_s_scale, 1626 ntohl(ti->ti_src.s_addr), 1627 TCP_MAX_WINSHIFT); 1628 tp->requested_s_scale = TCP_MAX_WINSHIFT; 1629 } 1630 break; 1631 1632 case TCPOPT_TIMESTAMP: 1633 if (optlen != TCPOLEN_TIMESTAMP) 1634 continue; 1635 oi->ts_present = 1; 1636 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 1637 NTOHL(oi->ts_val); 1638 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 1639 NTOHL(oi->ts_ecr); 1640 1641 /* 1642 * A timestamp received in a SYN makes 1643 * it ok to send timestamp requests and replies. 1644 */ 1645 if (ti->ti_flags & TH_SYN) { 1646 tp->t_flags |= TF_RCVD_TSTMP; 1647 tp->ts_recent = oi->ts_val; 1648 tp->ts_recent_age = tcp_now; 1649 } 1650 break; 1651 case TCPOPT_SACK_PERMITTED: 1652 if (optlen != TCPOLEN_SACK_PERMITTED) 1653 continue; 1654 if (!(ti->ti_flags & TH_SYN)) 1655 continue; 1656 tp->t_flags &= ~TF_CANT_TXSACK; 1657 break; 1658 1659 case TCPOPT_SACK: 1660 if (tp->t_flags & TF_IGNR_RXSACK) 1661 continue; 1662 if (optlen % 8 != 2 || optlen < 10) 1663 continue; 1664 cp += 2; 1665 optlen -= 2; 1666 for (; optlen > 0; cp -= 8, optlen -= 8) { 1667 tcp_seq lwe, rwe; 1668 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 1669 NTOHL(lwe); 1670 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 1671 NTOHL(rwe); 1672 /* tcp_mark_sacked(tp, lwe, rwe); */ 1673 } 1674 break; 1675 } 1676 } 1677 } 1678 1679 /* 1680 * Pull out of band byte out of a segment so 1681 * it doesn't appear in the user's data queue. 1682 * It is still reflected in the segment length for 1683 * sequencing purposes. 1684 */ 1685 void 1686 tcp_pulloutofband(so, ti, m) 1687 struct socket *so; 1688 struct tcpiphdr *ti; 1689 register struct mbuf *m; 1690 { 1691 int cnt = ti->ti_urp - 1; 1692 1693 while (cnt >= 0) { 1694 if (m->m_len > cnt) { 1695 char *cp = mtod(m, caddr_t) + cnt; 1696 struct tcpcb *tp = sototcpcb(so); 1697 1698 tp->t_iobc = *cp; 1699 tp->t_oobflags |= TCPOOB_HAVEDATA; 1700 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1701 m->m_len--; 1702 return; 1703 } 1704 cnt -= m->m_len; 1705 m = m->m_next; 1706 if (m == 0) 1707 break; 1708 } 1709 panic("tcp_pulloutofband"); 1710 } 1711 1712 /* 1713 * Collect new round-trip time estimate 1714 * and update averages and current timeout. 1715 */ 1716 void 1717 tcp_xmit_timer(tp, rtt) 1718 register struct tcpcb *tp; 1719 short rtt; 1720 { 1721 register short delta; 1722 short rttmin; 1723 1724 tcpstat.tcps_rttupdated++; 1725 --rtt; 1726 if (tp->t_srtt != 0) { 1727 /* 1728 * srtt is stored as fixed point with 3 bits after the 1729 * binary point (i.e., scaled by 8). The following magic 1730 * is equivalent to the smoothing algorithm in rfc793 with 1731 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1732 * point). Adjust rtt to origin 0. 1733 */ 1734 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 1735 if ((tp->t_srtt += delta) <= 0) 1736 tp->t_srtt = 1 << 2; 1737 /* 1738 * We accumulate a smoothed rtt variance (actually, a 1739 * smoothed mean difference), then set the retransmit 1740 * timer to smoothed rtt + 4 times the smoothed variance. 1741 * rttvar is stored as fixed point with 2 bits after the 1742 * binary point (scaled by 4). The following is 1743 * equivalent to rfc793 smoothing with an alpha of .75 1744 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1745 * rfc793's wired-in beta. 1746 */ 1747 if (delta < 0) 1748 delta = -delta; 1749 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1750 if ((tp->t_rttvar += delta) <= 0) 1751 tp->t_rttvar = 1 << 2; 1752 } else { 1753 /* 1754 * No rtt measurement yet - use the unsmoothed rtt. 1755 * Set the variance to half the rtt (so our first 1756 * retransmit happens at 3*rtt). 1757 */ 1758 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 1759 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 1760 } 1761 tp->t_rtt = 0; 1762 tp->t_rxtshift = 0; 1763 1764 /* 1765 * the retransmit should happen at rtt + 4 * rttvar. 1766 * Because of the way we do the smoothing, srtt and rttvar 1767 * will each average +1/2 tick of bias. When we compute 1768 * the retransmit timer, we want 1/2 tick of rounding and 1769 * 1 extra tick because of +-1/2 tick uncertainty in the 1770 * firing of the timer. The bias will give us exactly the 1771 * 1.5 tick we need. But, because the bias is 1772 * statistical, we have to test that we don't drop below 1773 * the minimum feasible timer (which is 2 ticks). 1774 */ 1775 if (tp->t_rttmin > rtt + 2) 1776 rttmin = tp->t_rttmin; 1777 else 1778 rttmin = rtt + 2; 1779 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 1780 1781 /* 1782 * We received an ack for a packet that wasn't retransmitted; 1783 * it is probably safe to discard any error indications we've 1784 * received recently. This isn't quite right, but close enough 1785 * for now (a route might have failed after we sent a segment, 1786 * and the return path might not be symmetrical). 1787 */ 1788 tp->t_softerror = 0; 1789 } 1790 1791 /* 1792 * Checks for partial ack. If partial ack arrives, force the retransmission 1793 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 1794 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 1795 * be started again. If the ack advances at least to tp->snd_recover, return 0. 1796 */ 1797 int 1798 tcp_newreno(tp, ti) 1799 struct tcpcb *tp; 1800 struct tcpiphdr *ti; 1801 { 1802 if (SEQ_LT(ti->ti_ack, tp->snd_recover)) { 1803 tcp_seq onxt = tp->snd_nxt; 1804 tcp_seq ouna = tp->snd_una; /* Haven't updated snd_una yet*/ 1805 u_long ocwnd = tp->snd_cwnd; 1806 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1807 tp->t_rtt = 0; 1808 tp->snd_nxt = ti->ti_ack; 1809 tp->snd_cwnd = tp->t_segsz; 1810 tp->snd_una = ti->ti_ack; 1811 (void) tcp_output(tp); 1812 tp->snd_cwnd = ocwnd; 1813 tp->snd_una = ouna; 1814 if (SEQ_GT(onxt, tp->snd_nxt)) 1815 tp->snd_nxt = onxt; 1816 /* 1817 * Partial window deflation. Relies on fact that tp->snd_una 1818 * not updated yet. 1819 */ 1820 tp->snd_cwnd -= (ti->ti_ack - tp->snd_una - tp->t_segsz); 1821 return 1; 1822 } 1823 return 0; 1824 } 1825 1826 1827 /* 1828 * TCP compressed state engine. Currently used to hold compressed 1829 * state for SYN_RECEIVED. 1830 */ 1831 1832 u_long syn_cache_count; 1833 u_int32_t syn_hash1, syn_hash2; 1834 1835 #define SYN_HASH(sa, sp, dp) \ 1836 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 1837 ((u_int32_t)(sp)))^syn_hash2))) 1838 1839 LIST_HEAD(, syn_cache_head) tcp_syn_cache_queue; 1840 1841 #define SYN_CACHE_RM(sc, scp) \ 1842 do { \ 1843 TAILQ_REMOVE(&(scp)->sch_queue, (sc), sc_queue); \ 1844 if (--(scp)->sch_length == 0) \ 1845 LIST_REMOVE((scp), sch_headq); \ 1846 syn_cache_count--; \ 1847 } while (0) 1848 1849 struct pool syn_cache_pool; 1850 1851 void 1852 syn_cache_init() 1853 { 1854 int i; 1855 1856 /* Initialize the hash bucket queues. */ 1857 for (i = 0; i < tcp_syn_cache_size; i++) 1858 TAILQ_INIT(&tcp_syn_cache[i].sch_queue); 1859 1860 /* Initialize the active hash bucket cache. */ 1861 LIST_INIT(&tcp_syn_cache_queue); 1862 1863 /* Initialize the syn cache pool. */ 1864 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 1865 "synpl", 0, NULL, NULL, M_PCB); 1866 } 1867 1868 void 1869 syn_cache_insert(sc) 1870 struct syn_cache *sc; 1871 { 1872 struct syn_cache_head *scp, *scp2, *sce; 1873 struct syn_cache *sc2; 1874 int s; 1875 1876 /* 1877 * If there are no entries in the hash table, reinitialize 1878 * the hash secrets. 1879 */ 1880 if (syn_cache_count == 0) { 1881 struct timeval tv; 1882 microtime(&tv); 1883 syn_hash1 = random() ^ (u_long)≻ 1884 syn_hash2 = random() ^ tv.tv_usec; 1885 } 1886 1887 sc->sc_hash = SYN_HASH(&sc->sc_src, sc->sc_sport, sc->sc_dport); 1888 scp = &tcp_syn_cache[sc->sc_hash % tcp_syn_cache_size]; 1889 1890 /* 1891 * Make sure that we don't overflow the per-bucket 1892 * limit or the total cache size limit. 1893 */ 1894 s = splsoftnet(); 1895 if (scp->sch_length >= tcp_syn_bucket_limit) { 1896 tcpstat.tcps_sc_bucketoverflow++; 1897 /* 1898 * The bucket is full. Toss the first (i.e. oldest) 1899 * element in this bucket. 1900 */ 1901 sc2 = TAILQ_FIRST(&scp->sch_queue); 1902 SYN_CACHE_RM(sc2, scp); 1903 if (sc2->sc_ipopts) 1904 (void) m_free(sc2->sc_ipopts); 1905 pool_put(&syn_cache_pool, sc2); 1906 } else if (syn_cache_count >= tcp_syn_cache_limit) { 1907 tcpstat.tcps_sc_overflowed++; 1908 /* 1909 * The cache is full. Toss the first (i.e. oldest) 1910 * element in the first non-empty bucket we can find. 1911 */ 1912 scp2 = scp; 1913 if (TAILQ_FIRST(&scp2->sch_queue) == NULL) { 1914 sce = &tcp_syn_cache[tcp_syn_cache_size]; 1915 for (++scp2; scp2 != scp; scp2++) { 1916 if (scp2 >= sce) 1917 scp2 = &tcp_syn_cache[0]; 1918 if (TAILQ_FIRST(&scp2->sch_queue) != NULL) 1919 break; 1920 } 1921 } 1922 sc2 = TAILQ_FIRST(&scp2->sch_queue); 1923 if (sc2 == NULL) { 1924 if (sc->sc_ipopts) 1925 (void) m_free(sc->sc_ipopts); 1926 pool_put(&syn_cache_pool, sc); 1927 return; 1928 } 1929 SYN_CACHE_RM(sc2, scp2); 1930 if (sc2->sc_ipopts) 1931 (void) m_free(sc2->sc_ipopts); 1932 pool_put(&syn_cache_pool, sc2); 1933 } 1934 1935 /* Set entry's timer. */ 1936 PRT_SLOW_ARM(sc->sc_timer, tcp_syn_cache_timeo); 1937 1938 /* Put it into the bucket. */ 1939 TAILQ_INSERT_TAIL(&scp->sch_queue, sc, sc_queue); 1940 if (++scp->sch_length == 1) 1941 LIST_INSERT_HEAD(&tcp_syn_cache_queue, scp, sch_headq); 1942 syn_cache_count++; 1943 1944 tcpstat.tcps_sc_added++; 1945 splx(s); 1946 } 1947 1948 /* 1949 * Walk down the cache list, looking for expired entries in each bucket. 1950 */ 1951 void 1952 syn_cache_timer() 1953 { 1954 struct syn_cache_head *scp, *nscp; 1955 struct syn_cache *sc, *nsc; 1956 int s; 1957 1958 s = splsoftnet(); 1959 for (scp = LIST_FIRST(&tcp_syn_cache_queue); scp != NULL; scp = nscp) { 1960 #ifdef DIAGNOSTIC 1961 if (TAILQ_FIRST(&scp->sch_queue) == NULL) 1962 panic("syn_cache_timer: queue inconsistency"); 1963 #endif 1964 nscp = LIST_NEXT(scp, sch_headq); 1965 for (sc = TAILQ_FIRST(&scp->sch_queue); 1966 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_timer); 1967 sc = nsc) { 1968 nsc = TAILQ_NEXT(sc, sc_queue); 1969 tcpstat.tcps_sc_timed_out++; 1970 SYN_CACHE_RM(sc, scp); 1971 if (sc->sc_ipopts) 1972 (void) m_free(sc->sc_ipopts); 1973 pool_put(&syn_cache_pool, sc); 1974 } 1975 } 1976 splx(s); 1977 } 1978 1979 /* 1980 * Find an entry in the syn cache. 1981 */ 1982 struct syn_cache * 1983 syn_cache_lookup(ti, headp) 1984 struct tcpiphdr *ti; 1985 struct syn_cache_head **headp; 1986 { 1987 struct syn_cache *sc; 1988 struct syn_cache_head *scp; 1989 u_int32_t hash; 1990 int s; 1991 1992 hash = SYN_HASH(&ti->ti_src, ti->ti_sport, ti->ti_dport); 1993 1994 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 1995 *headp = scp; 1996 s = splsoftnet(); 1997 for (sc = TAILQ_FIRST(&scp->sch_queue); sc != NULL; 1998 sc = TAILQ_NEXT(sc, sc_queue)) { 1999 if (sc->sc_hash != hash) 2000 continue; 2001 if (sc->sc_src.s_addr == ti->ti_src.s_addr && 2002 sc->sc_sport == ti->ti_sport && 2003 sc->sc_dport == ti->ti_dport && 2004 sc->sc_dst.s_addr == ti->ti_dst.s_addr) { 2005 splx(s); 2006 return (sc); 2007 } 2008 } 2009 splx(s); 2010 return (NULL); 2011 } 2012 2013 /* 2014 * This function gets called when we receive an ACK for a 2015 * socket in the LISTEN state. We look up the connection 2016 * in the syn cache, and if its there, we pull it out of 2017 * the cache and turn it into a full-blown connection in 2018 * the SYN-RECEIVED state. 2019 * 2020 * The return values may not be immediately obvious, and their effects 2021 * can be subtle, so here they are: 2022 * 2023 * NULL SYN was not found in cache; caller should drop the 2024 * packet and send an RST. 2025 * 2026 * -1 We were unable to create the new connection, and are 2027 * aborting it. An ACK,RST is being sent to the peer 2028 * (unless we got screwey sequence numbners; see below), 2029 * because the 3-way handshake has been completed. Caller 2030 * should not free the mbuf, since we may be using it. If 2031 * we are not, we will free it. 2032 * 2033 * Otherwise, the return value is a pointer to the new socket 2034 * associated with the connection. 2035 */ 2036 struct socket * 2037 syn_cache_get(so, m) 2038 struct socket *so; 2039 struct mbuf *m; 2040 { 2041 struct syn_cache *sc; 2042 struct syn_cache_head *scp; 2043 register struct inpcb *inp; 2044 register struct tcpcb *tp = 0; 2045 register struct tcpiphdr *ti; 2046 struct sockaddr_in *sin; 2047 struct mbuf *am; 2048 long win; 2049 int s; 2050 2051 ti = mtod(m, struct tcpiphdr *); 2052 s = splsoftnet(); 2053 if ((sc = syn_cache_lookup(ti, &scp)) == NULL) { 2054 splx(s); 2055 return (NULL); 2056 } 2057 2058 win = sbspace(&so->so_rcv); 2059 if (win > TCP_MAXWIN) 2060 win = TCP_MAXWIN; 2061 2062 /* 2063 * Verify the sequence and ack numbers. 2064 */ 2065 if ((ti->ti_ack != sc->sc_iss + 1) || 2066 SEQ_LEQ(ti->ti_seq, sc->sc_irs) || 2067 SEQ_GT(ti->ti_seq, sc->sc_irs + 1 + win)) { 2068 (void) syn_cache_respond(sc, m, ti, win, 0); 2069 splx(s); 2070 return ((struct socket *)(-1)); 2071 } 2072 2073 /* Remove this cache entry */ 2074 SYN_CACHE_RM(sc, scp); 2075 splx(s); 2076 2077 /* 2078 * Ok, create the full blown connection, and set things up 2079 * as they would have been set up if we had created the 2080 * connection when the SYN arrived. If we can't create 2081 * the connection, abort it. 2082 */ 2083 so = sonewconn(so, SS_ISCONNECTED); 2084 if (so == NULL) 2085 goto resetandabort; 2086 2087 inp = sotoinpcb(so); 2088 inp->inp_laddr = sc->sc_dst; 2089 inp->inp_lport = sc->sc_dport; 2090 in_pcbstate(inp, INP_BOUND); 2091 inp->inp_options = ip_srcroute(); 2092 if (inp->inp_options == NULL) { 2093 inp->inp_options = sc->sc_ipopts; 2094 sc->sc_ipopts = NULL; 2095 } 2096 2097 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 2098 if (am == NULL) 2099 goto resetandabort; 2100 am->m_len = sizeof(struct sockaddr_in); 2101 sin = mtod(am, struct sockaddr_in *); 2102 sin->sin_family = AF_INET; 2103 sin->sin_len = sizeof(*sin); 2104 sin->sin_addr = sc->sc_src; 2105 sin->sin_port = sc->sc_sport; 2106 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 2107 if (in_pcbconnect(inp, am)) { 2108 (void) m_free(am); 2109 goto resetandabort; 2110 } 2111 (void) m_free(am); 2112 2113 tp = intotcpcb(inp); 2114 if (sc->sc_request_r_scale != 15) { 2115 tp->requested_s_scale = sc->sc_requested_s_scale; 2116 tp->request_r_scale = sc->sc_request_r_scale; 2117 tp->snd_scale = sc->sc_requested_s_scale; 2118 tp->rcv_scale = sc->sc_request_r_scale; 2119 tp->t_flags |= TF_RCVD_SCALE; 2120 } 2121 if (sc->sc_flags & SCF_TIMESTAMP) 2122 tp->t_flags |= TF_RCVD_TSTMP; 2123 2124 tp->t_template = tcp_template(tp); 2125 if (tp->t_template == 0) { 2126 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 2127 so = NULL; 2128 m_freem(m); 2129 goto abort; 2130 } 2131 2132 tp->iss = sc->sc_iss; 2133 tp->irs = sc->sc_irs; 2134 tcp_sendseqinit(tp); 2135 tcp_rcvseqinit(tp); 2136 tp->t_state = TCPS_SYN_RECEIVED; 2137 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 2138 tcpstat.tcps_accepts++; 2139 2140 /* Initialize tp->t_ourmss before we deal with the peer's! */ 2141 tp->t_ourmss = sc->sc_ourmaxseg; 2142 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 2143 2144 /* 2145 * Initialize the initial congestion window. If we 2146 * had to retransmit the SYN,ACK, we must initialize cwnd 2147 * to 1 segment (i.e. the Loss Window). 2148 */ 2149 if (sc->sc_rexmt_count) 2150 tp->snd_cwnd = tp->t_peermss; 2151 else 2152 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 2153 2154 tcp_rmx_rtt(tp); 2155 tp->snd_wl1 = sc->sc_irs; 2156 tp->rcv_up = sc->sc_irs + 1; 2157 2158 /* 2159 * This is what whould have happened in tcp_ouput() when 2160 * the SYN,ACK was sent. 2161 */ 2162 tp->snd_up = tp->snd_una; 2163 tp->snd_max = tp->snd_nxt = tp->iss+1; 2164 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2165 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 2166 tp->rcv_adv = tp->rcv_nxt + win; 2167 tp->last_ack_sent = tp->rcv_nxt; 2168 2169 tcpstat.tcps_sc_completed++; 2170 if (sc->sc_ipopts) 2171 (void) m_free(sc->sc_ipopts); 2172 pool_put(&syn_cache_pool, sc); 2173 return (so); 2174 2175 resetandabort: 2176 (void) tcp_respond(NULL, ti, m, ti->ti_seq+ti->ti_len, 2177 (tcp_seq)0, TH_RST|TH_ACK); 2178 abort: 2179 if (so != NULL) 2180 (void) soabort(so); 2181 if (sc->sc_ipopts) 2182 (void) m_free(sc->sc_ipopts); 2183 pool_put(&syn_cache_pool, sc); 2184 tcpstat.tcps_sc_aborted++; 2185 return ((struct socket *)(-1)); 2186 } 2187 2188 /* 2189 * This function is called when we get a RST for a 2190 * non-existant connection, so that we can see if the 2191 * connection is in the syn cache. If it is, zap it. 2192 */ 2193 2194 void 2195 syn_cache_reset(ti) 2196 register struct tcpiphdr *ti; 2197 { 2198 struct syn_cache *sc; 2199 struct syn_cache_head *scp; 2200 int s = splsoftnet(); 2201 2202 if ((sc = syn_cache_lookup(ti, &scp)) == NULL) { 2203 splx(s); 2204 return; 2205 } 2206 if (SEQ_LT(ti->ti_seq,sc->sc_irs) || 2207 SEQ_GT(ti->ti_seq, sc->sc_irs+1)) { 2208 splx(s); 2209 return; 2210 } 2211 SYN_CACHE_RM(sc, scp); 2212 splx(s); 2213 tcpstat.tcps_sc_reset++; 2214 if (sc->sc_ipopts) 2215 (void) m_free(sc->sc_ipopts); 2216 pool_put(&syn_cache_pool, sc); 2217 } 2218 2219 void 2220 syn_cache_unreach(ip, th) 2221 struct ip *ip; 2222 struct tcphdr *th; 2223 { 2224 struct syn_cache *sc; 2225 struct syn_cache_head *scp; 2226 struct tcpiphdr ti2; 2227 int s; 2228 2229 ti2.ti_src.s_addr = ip->ip_dst.s_addr; 2230 ti2.ti_dst.s_addr = ip->ip_src.s_addr; 2231 ti2.ti_sport = th->th_dport; 2232 ti2.ti_dport = th->th_sport; 2233 2234 s = splsoftnet(); 2235 if ((sc = syn_cache_lookup(&ti2, &scp)) == NULL) { 2236 splx(s); 2237 return; 2238 } 2239 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 2240 if (ntohl (th->th_seq) != sc->sc_iss) { 2241 splx(s); 2242 return; 2243 } 2244 2245 /* 2246 * If we've rertransmitted 3 times and this is our second error, 2247 * we remove the entry. Otherwise, we allow it to continue on. 2248 * This prevents us from incorrectly nuking an entry during a 2249 * spurious network outage. 2250 * 2251 * See tcp_notify(). 2252 */ 2253 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rexmt_count < 3) { 2254 sc->sc_flags |= SCF_UNREACH; 2255 splx(s); 2256 return; 2257 } 2258 2259 SYN_CACHE_RM(sc, scp); 2260 splx(s); 2261 tcpstat.tcps_sc_unreach++; 2262 if (sc->sc_ipopts) 2263 (void) m_free(sc->sc_ipopts); 2264 pool_put(&syn_cache_pool, sc); 2265 } 2266 2267 /* 2268 * Given a LISTEN socket and an inbound SYN request, add 2269 * this to the syn cache, and send back a segment: 2270 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 2271 * to the source. 2272 * 2273 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 2274 * Doing so would require that we hold onto the data and deliver it 2275 * to the application. However, if we are the target of a SYN-flood 2276 * DoS attack, an attacker could send data which would eventually 2277 * consume all available buffer space if it were ACKed. By not ACKing 2278 * the data, we avoid this DoS scenario. 2279 */ 2280 2281 int 2282 syn_cache_add(so, m, optp, optlen, oi) 2283 struct socket *so; 2284 struct mbuf *m; 2285 u_char *optp; 2286 int optlen; 2287 struct tcp_opt_info *oi; 2288 { 2289 register struct tcpiphdr *ti; 2290 struct tcpcb tb, *tp; 2291 long win; 2292 struct syn_cache *sc; 2293 struct syn_cache_head *scp; 2294 struct mbuf *ipopts; 2295 2296 tp = sototcpcb(so); 2297 ti = mtod(m, struct tcpiphdr *); 2298 2299 /* 2300 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 2301 * in_broadcast() should never return true on a received 2302 * packet with M_BCAST not set. 2303 */ 2304 if (m->m_flags & (M_BCAST|M_MCAST) || 2305 IN_MULTICAST(ti->ti_src.s_addr) || 2306 IN_MULTICAST(ti->ti_dst.s_addr)) 2307 return (0); 2308 2309 /* 2310 * Initialize some local state. 2311 */ 2312 win = sbspace(&so->so_rcv); 2313 if (win > TCP_MAXWIN) 2314 win = TCP_MAXWIN; 2315 2316 /* 2317 * Remember the IP options, if any. 2318 */ 2319 ipopts = ip_srcroute(); 2320 2321 if (optp) { 2322 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 2323 tcp_dooptions(&tb, optp, optlen, ti, oi); 2324 } else 2325 tb.t_flags = 0; 2326 2327 /* 2328 * See if we already have an entry for this connection. 2329 * If we do, resend the SYN,ACK, and remember since the 2330 * initial congestion window must be initialized to 1 2331 * segment when the connection completes. 2332 */ 2333 if ((sc = syn_cache_lookup(ti, &scp)) != NULL) { 2334 tcpstat.tcps_sc_dupesyn++; 2335 sc->sc_rexmt_count++; 2336 if (sc->sc_rexmt_count == 0) { 2337 /* 2338 * Eeek! We rolled the counter. Just set it 2339 * to the max value. This shouldn't ever happen, 2340 * but there's no real reason to panic here, since 2341 * the count doesn't have to be very precise. 2342 */ 2343 sc->sc_rexmt_count = USHRT_MAX; 2344 } 2345 2346 if (ipopts) { 2347 /* 2348 * If we were remembering a previous source route, 2349 * forget it and use the new one we've been given. 2350 */ 2351 if (sc->sc_ipopts) 2352 (void) m_free(sc->sc_ipopts); 2353 sc->sc_ipopts = ipopts; 2354 } 2355 2356 if (syn_cache_respond(sc, m, ti, win, tb.ts_recent) == 0) { 2357 tcpstat.tcps_sndacks++; 2358 tcpstat.tcps_sndtotal++; 2359 } 2360 return (1); 2361 } 2362 2363 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 2364 if (sc == NULL) { 2365 if (ipopts) 2366 (void) m_free(ipopts); 2367 return (0); 2368 } 2369 2370 /* 2371 * Fill in the cache, and put the necessary IP and TCP 2372 * options into the reply. 2373 */ 2374 sc->sc_src.s_addr = ti->ti_src.s_addr; 2375 sc->sc_dst.s_addr = ti->ti_dst.s_addr; 2376 sc->sc_sport = ti->ti_sport; 2377 sc->sc_dport = ti->ti_dport; 2378 sc->sc_flags = 0; 2379 sc->sc_ipopts = ipopts; 2380 sc->sc_irs = ti->ti_seq; 2381 sc->sc_iss = tcp_new_iss(sc, sizeof(struct syn_cache), 0); 2382 sc->sc_peermaxseg = oi->maxseg; 2383 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 2384 m->m_pkthdr.rcvif : NULL); 2385 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 2386 sc->sc_flags |= SCF_TIMESTAMP; 2387 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2388 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2389 sc->sc_requested_s_scale = tb.requested_s_scale; 2390 sc->sc_request_r_scale = 0; 2391 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 2392 TCP_MAXWIN << sc->sc_request_r_scale < 2393 so->so_rcv.sb_hiwat) 2394 sc->sc_request_r_scale++; 2395 } else { 2396 sc->sc_requested_s_scale = 15; 2397 sc->sc_request_r_scale = 15; 2398 } 2399 if (syn_cache_respond(sc, m, ti, win, tb.ts_recent) == 0) { 2400 syn_cache_insert(sc); 2401 tcpstat.tcps_sndacks++; 2402 tcpstat.tcps_sndtotal++; 2403 } else { 2404 if (sc->sc_ipopts) 2405 (void) m_free(sc->sc_ipopts); 2406 pool_put(&syn_cache_pool, sc); 2407 tcpstat.tcps_sc_dropped++; 2408 } 2409 return (1); 2410 } 2411 2412 int 2413 syn_cache_respond(sc, m, ti, win, ts) 2414 struct syn_cache *sc; 2415 struct mbuf *m; 2416 register struct tcpiphdr *ti; 2417 long win; 2418 u_long ts; 2419 { 2420 u_int8_t *optp; 2421 int optlen; 2422 2423 /* 2424 * Tack on the TCP options. If there isn't enough trailing 2425 * space for them, move up the fixed header to make space. 2426 */ 2427 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 2428 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 2429 if (optlen > M_TRAILINGSPACE(m)) { 2430 if (M_LEADINGSPACE(m) >= optlen) { 2431 m->m_data -= optlen; 2432 m->m_len += optlen; 2433 } else { 2434 struct mbuf *m0 = m; 2435 if ((m = m_gethdr(M_DONTWAIT, MT_HEADER)) == NULL) { 2436 m_freem(m0); 2437 return (ENOBUFS); 2438 } 2439 MH_ALIGN(m, sizeof(*ti) + optlen); 2440 m->m_next = m0; /* this gets freed below */ 2441 } 2442 bcopy((caddr_t)ti, mtod(m, caddr_t), sizeof(*ti)); 2443 ti = mtod(m, struct tcpiphdr *); 2444 } 2445 2446 optp = (u_int8_t *)(ti + 1); 2447 optp[0] = TCPOPT_MAXSEG; 2448 optp[1] = 4; 2449 optp[2] = (sc->sc_ourmaxseg >> 8) & 0xff; 2450 optp[3] = sc->sc_ourmaxseg & 0xff; 2451 optlen = 4; 2452 2453 if (sc->sc_request_r_scale != 15) { 2454 *((u_int32_t *)(optp + optlen)) = htonl(TCPOPT_NOP << 24 | 2455 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 2456 sc->sc_request_r_scale); 2457 optlen += 4; 2458 } 2459 2460 if (sc->sc_flags & SCF_TIMESTAMP) { 2461 u_int32_t *lp = (u_int32_t *)(optp + optlen); 2462 /* Form timestamp option as shown in appendix A of RFC 1323. */ 2463 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 2464 *lp++ = htonl(tcp_now); 2465 *lp = htonl(ts); 2466 optlen += TCPOLEN_TSTAMP_APPA; 2467 } 2468 2469 /* 2470 * Toss any trailing mbufs. No need to worry about 2471 * m_len and m_pkthdr.len, since tcp_respond() will 2472 * unconditionally set them. 2473 */ 2474 if (m->m_next) { 2475 m_freem(m->m_next); 2476 m->m_next = NULL; 2477 } 2478 2479 /* 2480 * Fill in the fields that tcp_respond() will not touch, and 2481 * then send the response. 2482 */ 2483 ti->ti_off = (sizeof(struct tcphdr) + optlen) >> 2; 2484 ti->ti_win = htons(win); 2485 return (tcp_respond(NULL, ti, m, sc->sc_irs + 1, sc->sc_iss, 2486 TH_SYN|TH_ACK)); 2487 } 2488