tcp_input.c revision 1.63
1 /* $NetBSD: tcp_input.c,v 1.63 1998/08/02 00:35:51 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 FREE(tiqe, M_IPQ); 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 FREE(q, M_IPQ); 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 FREE(q, M_IPQ); 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 MALLOC(tiqe, struct ipqent *, sizeof (struct ipqent), M_IPQ, M_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 FREE(q, M_IPQ); 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 * completeed, 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 } 622 } else { 623 /* 624 * Received a SYN. 625 */ 626 if (in_hosteq(ti->ti_src, ti->ti_dst) && 627 ti->ti_sport == ti->ti_dport) { 628 /* 629 * LISTEN socket received a SYN 630 * from itself? This can't possibly 631 * be valid; drop the packet. 632 */ 633 tcpstat.tcps_badsyn++; 634 goto drop; 635 } 636 /* 637 * SYN looks ok; create compressed TCP 638 * state for it. 639 */ 640 if (so->so_qlen <= so->so_qlimit && 641 syn_cache_add(so, m, optp, optlen, &opti)) 642 m = NULL; 643 } 644 goto drop; 645 } 646 } 647 648 after_listen: 649 #ifdef DIAGNOSTIC 650 /* 651 * Should not happen now that all embryonic connections 652 * are handled with compressed state. 653 */ 654 if (tp->t_state == TCPS_LISTEN) 655 panic("tcp_input: TCPS_LISTEN"); 656 #endif 657 658 /* 659 * Segment received on connection. 660 * Reset idle time and keep-alive timer. 661 */ 662 tp->t_idle = 0; 663 if (TCPS_HAVEESTABLISHED(tp->t_state)) 664 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 665 666 /* 667 * Process options. 668 */ 669 if (optp) 670 tcp_dooptions(tp, optp, optlen, ti, &opti); 671 672 /* 673 * Header prediction: check for the two common cases 674 * of a uni-directional data xfer. If the packet has 675 * no control flags, is in-sequence, the window didn't 676 * change and we're not retransmitting, it's a 677 * candidate. If the length is zero and the ack moved 678 * forward, we're the sender side of the xfer. Just 679 * free the data acked & wake any higher level process 680 * that was blocked waiting for space. If the length 681 * is non-zero and the ack didn't move, we're the 682 * receiver side. If we're getting packets in-order 683 * (the reassembly queue is empty), add the data to 684 * the socket buffer and note that we need a delayed ack. 685 */ 686 if (tp->t_state == TCPS_ESTABLISHED && 687 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 688 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 689 ti->ti_seq == tp->rcv_nxt && 690 tiwin && tiwin == tp->snd_wnd && 691 tp->snd_nxt == tp->snd_max) { 692 693 /* 694 * If last ACK falls within this segment's sequence numbers, 695 * record the timestamp. 696 */ 697 if (opti.ts_present && 698 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 699 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) { 700 tp->ts_recent_age = tcp_now; 701 tp->ts_recent = opti.ts_val; 702 } 703 704 if (ti->ti_len == 0) { 705 if (SEQ_GT(ti->ti_ack, tp->snd_una) && 706 SEQ_LEQ(ti->ti_ack, tp->snd_max) && 707 tp->snd_cwnd >= tp->snd_wnd && 708 tp->t_dupacks < tcprexmtthresh) { 709 /* 710 * this is a pure ack for outstanding data. 711 */ 712 ++tcpstat.tcps_predack; 713 if (opti.ts_present) 714 tcp_xmit_timer(tp, 715 tcp_now-opti.ts_ecr+1); 716 else if (tp->t_rtt && 717 SEQ_GT(ti->ti_ack, tp->t_rtseq)) 718 tcp_xmit_timer(tp, tp->t_rtt); 719 acked = ti->ti_ack - tp->snd_una; 720 tcpstat.tcps_rcvackpack++; 721 tcpstat.tcps_rcvackbyte += acked; 722 sbdrop(&so->so_snd, acked); 723 tp->snd_una = ti->ti_ack; 724 m_freem(m); 725 726 /* 727 * If all outstanding data are acked, stop 728 * retransmit timer, otherwise restart timer 729 * using current (possibly backed-off) value. 730 * If process is waiting for space, 731 * wakeup/selwakeup/signal. If data 732 * are ready to send, let tcp_output 733 * decide between more output or persist. 734 */ 735 if (tp->snd_una == tp->snd_max) 736 TCP_TIMER_DISARM(tp, TCPT_REXMT); 737 else if (TCP_TIMER_ISARMED(tp, 738 TCPT_PERSIST) == 0) 739 TCP_TIMER_ARM(tp, TCPT_REXMT, 740 tp->t_rxtcur); 741 742 sowwakeup(so); 743 if (so->so_snd.sb_cc) 744 (void) tcp_output(tp); 745 return; 746 } 747 } else if (ti->ti_ack == tp->snd_una && 748 tp->segq.lh_first == NULL && 749 ti->ti_len <= sbspace(&so->so_rcv)) { 750 /* 751 * this is a pure, in-sequence data packet 752 * with nothing on the reassembly queue and 753 * we have enough buffer space to take it. 754 */ 755 ++tcpstat.tcps_preddat; 756 tp->rcv_nxt += ti->ti_len; 757 tcpstat.tcps_rcvpack++; 758 tcpstat.tcps_rcvbyte += ti->ti_len; 759 /* 760 * Drop TCP, IP headers and TCP options then add data 761 * to socket buffer. 762 */ 763 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 764 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr); 765 sbappend(&so->so_rcv, m); 766 sorwakeup(so); 767 TCP_SETUP_ACK(tp, ti); 768 if (tp->t_flags & TF_ACKNOW) 769 (void) tcp_output(tp); 770 return; 771 } 772 } 773 774 /* 775 * Drop TCP, IP headers and TCP options. 776 */ 777 hdroptlen = sizeof(struct tcpiphdr) + off - sizeof(struct tcphdr); 778 m->m_data += hdroptlen; 779 m->m_len -= hdroptlen; 780 781 /* 782 * Calculate amount of space in receive window, 783 * and then do TCP input processing. 784 * Receive window is amount of space in rcv queue, 785 * but not less than advertised window. 786 */ 787 { int win; 788 789 win = sbspace(&so->so_rcv); 790 if (win < 0) 791 win = 0; 792 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 793 } 794 795 switch (tp->t_state) { 796 797 /* 798 * If the state is SYN_SENT: 799 * if seg contains an ACK, but not for our SYN, drop the input. 800 * if seg contains a RST, then drop the connection. 801 * if seg does not contain SYN, then drop it. 802 * Otherwise this is an acceptable SYN segment 803 * initialize tp->rcv_nxt and tp->irs 804 * if seg contains ack then advance tp->snd_una 805 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 806 * arrange for segment to be acked (eventually) 807 * continue processing rest of data/controls, beginning with URG 808 */ 809 case TCPS_SYN_SENT: 810 if ((tiflags & TH_ACK) && 811 (SEQ_LEQ(ti->ti_ack, tp->iss) || 812 SEQ_GT(ti->ti_ack, tp->snd_max))) 813 goto dropwithreset; 814 if (tiflags & TH_RST) { 815 if (tiflags & TH_ACK) 816 tp = tcp_drop(tp, ECONNREFUSED); 817 goto drop; 818 } 819 if ((tiflags & TH_SYN) == 0) 820 goto drop; 821 if (tiflags & TH_ACK) { 822 tp->snd_una = ti->ti_ack; 823 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 824 tp->snd_nxt = tp->snd_una; 825 } 826 TCP_TIMER_DISARM(tp, TCPT_REXMT); 827 tp->irs = ti->ti_seq; 828 tcp_rcvseqinit(tp); 829 tp->t_flags |= TF_ACKNOW; 830 tcp_mss_from_peer(tp, opti.maxseg); 831 832 /* 833 * Initialize the initial congestion window. If we 834 * had to retransmit the SYN, we must initialize cwnd 835 * to 1 segment (i.e. the Loss Window). 836 */ 837 if (tp->t_flags & TF_SYN_REXMT) 838 tp->snd_cwnd = tp->t_peermss; 839 else 840 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, 841 tp->t_peermss); 842 843 tcp_rmx_rtt(tp); 844 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 845 tcpstat.tcps_connects++; 846 soisconnected(so); 847 tcp_established(tp); 848 /* Do window scaling on this connection? */ 849 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 850 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 851 tp->snd_scale = tp->requested_s_scale; 852 tp->rcv_scale = tp->request_r_scale; 853 } 854 (void) tcp_reass(tp, (struct tcpiphdr *)0, 855 (struct mbuf *)0); 856 /* 857 * if we didn't have to retransmit the SYN, 858 * use its rtt as our initial srtt & rtt var. 859 */ 860 if (tp->t_rtt) 861 tcp_xmit_timer(tp, tp->t_rtt); 862 } else 863 tp->t_state = TCPS_SYN_RECEIVED; 864 865 /* 866 * Advance ti->ti_seq to correspond to first data byte. 867 * If data, trim to stay within window, 868 * dropping FIN if necessary. 869 */ 870 ti->ti_seq++; 871 if (ti->ti_len > tp->rcv_wnd) { 872 todrop = ti->ti_len - tp->rcv_wnd; 873 m_adj(m, -todrop); 874 ti->ti_len = tp->rcv_wnd; 875 tiflags &= ~TH_FIN; 876 tcpstat.tcps_rcvpackafterwin++; 877 tcpstat.tcps_rcvbyteafterwin += todrop; 878 } 879 tp->snd_wl1 = ti->ti_seq - 1; 880 tp->rcv_up = ti->ti_seq; 881 goto step6; 882 883 /* 884 * If the state is SYN_RECEIVED: 885 * If seg contains an ACK, but not for our SYN, drop the input 886 * and generate an RST. See page 36, rfc793 887 */ 888 case TCPS_SYN_RECEIVED: 889 if ((tiflags & TH_ACK) && 890 (SEQ_LEQ(ti->ti_ack, tp->iss) || 891 SEQ_GT(ti->ti_ack, tp->snd_max))) 892 goto dropwithreset; 893 break; 894 } 895 896 /* 897 * States other than LISTEN or SYN_SENT. 898 * First check timestamp, if present. 899 * Then check that at least some bytes of segment are within 900 * receive window. If segment begins before rcv_nxt, 901 * drop leading data (and SYN); if nothing left, just ack. 902 * 903 * RFC 1323 PAWS: If we have a timestamp reply on this segment 904 * and it's less than ts_recent, drop it. 905 */ 906 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 907 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 908 909 /* Check to see if ts_recent is over 24 days old. */ 910 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 911 /* 912 * Invalidate ts_recent. If this segment updates 913 * ts_recent, the age will be reset later and ts_recent 914 * will get a valid value. If it does not, setting 915 * ts_recent to zero will at least satisfy the 916 * requirement that zero be placed in the timestamp 917 * echo reply when ts_recent isn't valid. The 918 * age isn't reset until we get a valid ts_recent 919 * because we don't want out-of-order segments to be 920 * dropped when ts_recent is old. 921 */ 922 tp->ts_recent = 0; 923 } else { 924 tcpstat.tcps_rcvduppack++; 925 tcpstat.tcps_rcvdupbyte += ti->ti_len; 926 tcpstat.tcps_pawsdrop++; 927 goto dropafterack; 928 } 929 } 930 931 todrop = tp->rcv_nxt - ti->ti_seq; 932 if (todrop > 0) { 933 if (tiflags & TH_SYN) { 934 tiflags &= ~TH_SYN; 935 ti->ti_seq++; 936 if (ti->ti_urp > 1) 937 ti->ti_urp--; 938 else { 939 tiflags &= ~TH_URG; 940 ti->ti_urp = 0; 941 } 942 todrop--; 943 } 944 if (todrop > ti->ti_len || 945 (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) { 946 /* 947 * Any valid FIN must be to the left of the window. 948 * At this point the FIN must be a duplicate or 949 * out of sequence; drop it. 950 */ 951 tiflags &= ~TH_FIN; 952 /* 953 * Send an ACK to resynchronize and drop any data. 954 * But keep on processing for RST or ACK. 955 */ 956 tp->t_flags |= TF_ACKNOW; 957 todrop = ti->ti_len; 958 tcpstat.tcps_rcvdupbyte += todrop; 959 tcpstat.tcps_rcvduppack++; 960 } else { 961 tcpstat.tcps_rcvpartduppack++; 962 tcpstat.tcps_rcvpartdupbyte += todrop; 963 } 964 m_adj(m, todrop); 965 ti->ti_seq += todrop; 966 ti->ti_len -= todrop; 967 if (ti->ti_urp > todrop) 968 ti->ti_urp -= todrop; 969 else { 970 tiflags &= ~TH_URG; 971 ti->ti_urp = 0; 972 } 973 } 974 975 /* 976 * If new data are received on a connection after the 977 * user processes are gone, then RST the other end. 978 */ 979 if ((so->so_state & SS_NOFDREF) && 980 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) { 981 tp = tcp_close(tp); 982 tcpstat.tcps_rcvafterclose++; 983 goto dropwithreset; 984 } 985 986 /* 987 * If segment ends after window, drop trailing data 988 * (and PUSH and FIN); if nothing left, just ACK. 989 */ 990 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd); 991 if (todrop > 0) { 992 tcpstat.tcps_rcvpackafterwin++; 993 if (todrop >= ti->ti_len) { 994 tcpstat.tcps_rcvbyteafterwin += ti->ti_len; 995 /* 996 * If a new connection request is received 997 * while in TIME_WAIT, drop the old connection 998 * and start over if the sequence numbers 999 * are above the previous ones. 1000 */ 1001 if (tiflags & TH_SYN && 1002 tp->t_state == TCPS_TIME_WAIT && 1003 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) { 1004 iss = tcp_new_iss(tp, sizeof(struct tcpcb), 1005 tp->rcv_nxt); 1006 tp = tcp_close(tp); 1007 /* 1008 * We have already advanced the mbuf 1009 * pointers past the IP+TCP headers and 1010 * options. Restore those pointers before 1011 * attempting to use the TCP header again. 1012 */ 1013 m->m_data -= hdroptlen; 1014 m->m_len += hdroptlen; 1015 goto findpcb; 1016 } 1017 /* 1018 * If window is closed can only take segments at 1019 * window edge, and have to drop data and PUSH from 1020 * incoming segments. Continue processing, but 1021 * remember to ack. Otherwise, drop segment 1022 * and ack. 1023 */ 1024 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) { 1025 tp->t_flags |= TF_ACKNOW; 1026 tcpstat.tcps_rcvwinprobe++; 1027 } else 1028 goto dropafterack; 1029 } else 1030 tcpstat.tcps_rcvbyteafterwin += todrop; 1031 m_adj(m, -todrop); 1032 ti->ti_len -= todrop; 1033 tiflags &= ~(TH_PUSH|TH_FIN); 1034 } 1035 1036 /* 1037 * If last ACK falls within this segment's sequence numbers, 1038 * and the timestamp is newer, record it. 1039 */ 1040 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1041 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) && 1042 SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len + 1043 ((tiflags & (TH_SYN|TH_FIN)) != 0))) { 1044 tp->ts_recent_age = tcp_now; 1045 tp->ts_recent = opti.ts_val; 1046 } 1047 1048 /* 1049 * If the RST bit is set examine the state: 1050 * SYN_RECEIVED STATE: 1051 * If passive open, return to LISTEN state. 1052 * If active open, inform user that connection was refused. 1053 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1054 * Inform user that connection was reset, and close tcb. 1055 * CLOSING, LAST_ACK, TIME_WAIT STATES 1056 * Close the tcb. 1057 */ 1058 if (tiflags&TH_RST) switch (tp->t_state) { 1059 1060 case TCPS_SYN_RECEIVED: 1061 so->so_error = ECONNREFUSED; 1062 goto close; 1063 1064 case TCPS_ESTABLISHED: 1065 case TCPS_FIN_WAIT_1: 1066 case TCPS_FIN_WAIT_2: 1067 case TCPS_CLOSE_WAIT: 1068 so->so_error = ECONNRESET; 1069 close: 1070 tp->t_state = TCPS_CLOSED; 1071 tcpstat.tcps_drops++; 1072 tp = tcp_close(tp); 1073 goto drop; 1074 1075 case TCPS_CLOSING: 1076 case TCPS_LAST_ACK: 1077 case TCPS_TIME_WAIT: 1078 tp = tcp_close(tp); 1079 goto drop; 1080 } 1081 1082 /* 1083 * If a SYN is in the window, then this is an 1084 * error and we send an RST and drop the connection. 1085 */ 1086 if (tiflags & TH_SYN) { 1087 tp = tcp_drop(tp, ECONNRESET); 1088 goto dropwithreset; 1089 } 1090 1091 /* 1092 * If the ACK bit is off we drop the segment and return. 1093 */ 1094 if ((tiflags & TH_ACK) == 0) 1095 goto drop; 1096 1097 /* 1098 * Ack processing. 1099 */ 1100 switch (tp->t_state) { 1101 1102 /* 1103 * In SYN_RECEIVED state if the ack ACKs our SYN then enter 1104 * ESTABLISHED state and continue processing, otherwise 1105 * send an RST. 1106 */ 1107 case TCPS_SYN_RECEIVED: 1108 if (SEQ_GT(tp->snd_una, ti->ti_ack) || 1109 SEQ_GT(ti->ti_ack, tp->snd_max)) 1110 goto dropwithreset; 1111 tcpstat.tcps_connects++; 1112 soisconnected(so); 1113 tcp_established(tp); 1114 /* Do window scaling? */ 1115 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1116 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1117 tp->snd_scale = tp->requested_s_scale; 1118 tp->rcv_scale = tp->request_r_scale; 1119 } 1120 (void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0); 1121 tp->snd_wl1 = ti->ti_seq - 1; 1122 /* fall into ... */ 1123 1124 /* 1125 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1126 * ACKs. If the ack is in the range 1127 * tp->snd_una < ti->ti_ack <= tp->snd_max 1128 * then advance tp->snd_una to ti->ti_ack and drop 1129 * data from the retransmission queue. If this ACK reflects 1130 * more up to date window information we update our window information. 1131 */ 1132 case TCPS_ESTABLISHED: 1133 case TCPS_FIN_WAIT_1: 1134 case TCPS_FIN_WAIT_2: 1135 case TCPS_CLOSE_WAIT: 1136 case TCPS_CLOSING: 1137 case TCPS_LAST_ACK: 1138 case TCPS_TIME_WAIT: 1139 1140 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) { 1141 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) { 1142 tcpstat.tcps_rcvdupack++; 1143 /* 1144 * If we have outstanding data (other than 1145 * a window probe), this is a completely 1146 * duplicate ack (ie, window info didn't 1147 * change), the ack is the biggest we've 1148 * seen and we've seen exactly our rexmt 1149 * threshhold of them, assume a packet 1150 * has been dropped and retransmit it. 1151 * Kludge snd_nxt & the congestion 1152 * window so we send only this one 1153 * packet. 1154 * 1155 * We know we're losing at the current 1156 * window size so do congestion avoidance 1157 * (set ssthresh to half the current window 1158 * and pull our congestion window back to 1159 * the new ssthresh). 1160 * 1161 * Dup acks mean that packets have left the 1162 * network (they're now cached at the receiver) 1163 * so bump cwnd by the amount in the receiver 1164 * to keep a constant cwnd packets in the 1165 * network. 1166 */ 1167 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 || 1168 ti->ti_ack != tp->snd_una) 1169 tp->t_dupacks = 0; 1170 else if (++tp->t_dupacks == tcprexmtthresh) { 1171 tcp_seq onxt = tp->snd_nxt; 1172 u_int win = 1173 min(tp->snd_wnd, tp->snd_cwnd) / 1174 2 / tp->t_segsz; 1175 1176 if (win < 2) 1177 win = 2; 1178 tp->snd_ssthresh = win * tp->t_segsz; 1179 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1180 tp->t_rtt = 0; 1181 tp->snd_nxt = ti->ti_ack; 1182 tp->snd_cwnd = tp->t_segsz; 1183 (void) tcp_output(tp); 1184 tp->snd_cwnd = tp->snd_ssthresh + 1185 tp->t_segsz * tp->t_dupacks; 1186 if (SEQ_GT(onxt, tp->snd_nxt)) 1187 tp->snd_nxt = onxt; 1188 goto drop; 1189 } else if (tp->t_dupacks > tcprexmtthresh) { 1190 tp->snd_cwnd += tp->t_segsz; 1191 (void) tcp_output(tp); 1192 goto drop; 1193 } 1194 } else 1195 tp->t_dupacks = 0; 1196 break; 1197 } 1198 /* 1199 * If the congestion window was inflated to account 1200 * for the other side's cached packets, retract it. 1201 */ 1202 if (tp->t_dupacks >= tcprexmtthresh && 1203 tp->snd_cwnd > tp->snd_ssthresh) 1204 tp->snd_cwnd = tp->snd_ssthresh; 1205 tp->t_dupacks = 0; 1206 if (SEQ_GT(ti->ti_ack, tp->snd_max)) { 1207 tcpstat.tcps_rcvacktoomuch++; 1208 goto dropafterack; 1209 } 1210 acked = ti->ti_ack - tp->snd_una; 1211 tcpstat.tcps_rcvackpack++; 1212 tcpstat.tcps_rcvackbyte += acked; 1213 1214 /* 1215 * If we have a timestamp reply, update smoothed 1216 * round trip time. If no timestamp is present but 1217 * transmit timer is running and timed sequence 1218 * number was acked, update smoothed round trip time. 1219 * Since we now have an rtt measurement, cancel the 1220 * timer backoff (cf., Phil Karn's retransmit alg.). 1221 * Recompute the initial retransmit timer. 1222 */ 1223 if (opti.ts_present) 1224 tcp_xmit_timer(tp, tcp_now - opti.ts_ecr + 1); 1225 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) 1226 tcp_xmit_timer(tp,tp->t_rtt); 1227 1228 /* 1229 * If all outstanding data is acked, stop retransmit 1230 * timer and remember to restart (more output or persist). 1231 * If there is more data to be acked, restart retransmit 1232 * timer, using current (possibly backed-off) value. 1233 */ 1234 if (ti->ti_ack == tp->snd_max) { 1235 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1236 needoutput = 1; 1237 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1238 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1239 /* 1240 * When new data is acked, open the congestion window. 1241 * If the window gives us less than ssthresh packets 1242 * in flight, open exponentially (segsz per packet). 1243 * Otherwise open linearly: segsz per window 1244 * (segsz^2 / cwnd per packet), plus a constant 1245 * fraction of a packet (segsz/8) to help larger windows 1246 * open quickly enough. 1247 */ 1248 { 1249 register u_int cw = tp->snd_cwnd; 1250 register u_int incr = tp->t_segsz; 1251 1252 if (cw > tp->snd_ssthresh) 1253 incr = incr * incr / cw; 1254 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1255 } 1256 if (acked > so->so_snd.sb_cc) { 1257 tp->snd_wnd -= so->so_snd.sb_cc; 1258 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1259 ourfinisacked = 1; 1260 } else { 1261 sbdrop(&so->so_snd, acked); 1262 tp->snd_wnd -= acked; 1263 ourfinisacked = 0; 1264 } 1265 sowwakeup(so); 1266 tp->snd_una = ti->ti_ack; 1267 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1268 tp->snd_nxt = tp->snd_una; 1269 1270 switch (tp->t_state) { 1271 1272 /* 1273 * In FIN_WAIT_1 STATE in addition to the processing 1274 * for the ESTABLISHED state if our FIN is now acknowledged 1275 * then enter FIN_WAIT_2. 1276 */ 1277 case TCPS_FIN_WAIT_1: 1278 if (ourfinisacked) { 1279 /* 1280 * If we can't receive any more 1281 * data, then closing user can proceed. 1282 * Starting the timer is contrary to the 1283 * specification, but if we don't get a FIN 1284 * we'll hang forever. 1285 */ 1286 if (so->so_state & SS_CANTRCVMORE) { 1287 soisdisconnected(so); 1288 TCP_TIMER_ARM(tp, TCPT_2MSL, 1289 tcp_maxidle); 1290 } 1291 tp->t_state = TCPS_FIN_WAIT_2; 1292 } 1293 break; 1294 1295 /* 1296 * In CLOSING STATE in addition to the processing for 1297 * the ESTABLISHED state if the ACK acknowledges our FIN 1298 * then enter the TIME-WAIT state, otherwise ignore 1299 * the segment. 1300 */ 1301 case TCPS_CLOSING: 1302 if (ourfinisacked) { 1303 tp->t_state = TCPS_TIME_WAIT; 1304 tcp_canceltimers(tp); 1305 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1306 soisdisconnected(so); 1307 } 1308 break; 1309 1310 /* 1311 * In LAST_ACK, we may still be waiting for data to drain 1312 * and/or to be acked, as well as for the ack of our FIN. 1313 * If our FIN is now acknowledged, delete the TCB, 1314 * enter the closed state and return. 1315 */ 1316 case TCPS_LAST_ACK: 1317 if (ourfinisacked) { 1318 tp = tcp_close(tp); 1319 goto drop; 1320 } 1321 break; 1322 1323 /* 1324 * In TIME_WAIT state the only thing that should arrive 1325 * is a retransmission of the remote FIN. Acknowledge 1326 * it and restart the finack timer. 1327 */ 1328 case TCPS_TIME_WAIT: 1329 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1330 goto dropafterack; 1331 } 1332 } 1333 1334 step6: 1335 /* 1336 * Update window information. 1337 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1338 */ 1339 if (((tiflags & TH_ACK) && SEQ_LT(tp->snd_wl1, ti->ti_seq)) || 1340 (tp->snd_wl1 == ti->ti_seq && SEQ_LT(tp->snd_wl2, ti->ti_ack)) || 1341 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)) { 1342 /* keep track of pure window updates */ 1343 if (ti->ti_len == 0 && 1344 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) 1345 tcpstat.tcps_rcvwinupd++; 1346 tp->snd_wnd = tiwin; 1347 tp->snd_wl1 = ti->ti_seq; 1348 tp->snd_wl2 = ti->ti_ack; 1349 if (tp->snd_wnd > tp->max_sndwnd) 1350 tp->max_sndwnd = tp->snd_wnd; 1351 needoutput = 1; 1352 } 1353 1354 /* 1355 * Process segments with URG. 1356 */ 1357 if ((tiflags & TH_URG) && ti->ti_urp && 1358 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1359 /* 1360 * This is a kludge, but if we receive and accept 1361 * random urgent pointers, we'll crash in 1362 * soreceive. It's hard to imagine someone 1363 * actually wanting to send this much urgent data. 1364 */ 1365 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) { 1366 ti->ti_urp = 0; /* XXX */ 1367 tiflags &= ~TH_URG; /* XXX */ 1368 goto dodata; /* XXX */ 1369 } 1370 /* 1371 * If this segment advances the known urgent pointer, 1372 * then mark the data stream. This should not happen 1373 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1374 * a FIN has been received from the remote side. 1375 * In these states we ignore the URG. 1376 * 1377 * According to RFC961 (Assigned Protocols), 1378 * the urgent pointer points to the last octet 1379 * of urgent data. We continue, however, 1380 * to consider it to indicate the first octet 1381 * of data past the urgent section as the original 1382 * spec states (in one of two places). 1383 */ 1384 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) { 1385 tp->rcv_up = ti->ti_seq + ti->ti_urp; 1386 so->so_oobmark = so->so_rcv.sb_cc + 1387 (tp->rcv_up - tp->rcv_nxt) - 1; 1388 if (so->so_oobmark == 0) 1389 so->so_state |= SS_RCVATMARK; 1390 sohasoutofband(so); 1391 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1392 } 1393 /* 1394 * Remove out of band data so doesn't get presented to user. 1395 * This can happen independent of advancing the URG pointer, 1396 * but if two URG's are pending at once, some out-of-band 1397 * data may creep in... ick. 1398 */ 1399 if (ti->ti_urp <= (u_int16_t) ti->ti_len 1400 #ifdef SO_OOBINLINE 1401 && (so->so_options & SO_OOBINLINE) == 0 1402 #endif 1403 ) 1404 tcp_pulloutofband(so, ti, m); 1405 } else 1406 /* 1407 * If no out of band data is expected, 1408 * pull receive urgent pointer along 1409 * with the receive window. 1410 */ 1411 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1412 tp->rcv_up = tp->rcv_nxt; 1413 dodata: /* XXX */ 1414 1415 /* 1416 * Process the segment text, merging it into the TCP sequencing queue, 1417 * and arranging for acknowledgment of receipt if necessary. 1418 * This process logically involves adjusting tp->rcv_wnd as data 1419 * is presented to the user (this happens in tcp_usrreq.c, 1420 * case PRU_RCVD). If a FIN has already been received on this 1421 * connection then we just ignore the text. 1422 */ 1423 if ((ti->ti_len || (tiflags & TH_FIN)) && 1424 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1425 TCP_REASS(tp, ti, m, so, tiflags); 1426 /* 1427 * Note the amount of data that peer has sent into 1428 * our window, in order to estimate the sender's 1429 * buffer size. 1430 */ 1431 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 1432 } else { 1433 m_freem(m); 1434 tiflags &= ~TH_FIN; 1435 } 1436 1437 /* 1438 * If FIN is received ACK the FIN and let the user know 1439 * that the connection is closing. Ignore a FIN received before 1440 * the connection is fully established. 1441 */ 1442 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 1443 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1444 socantrcvmore(so); 1445 tp->t_flags |= TF_ACKNOW; 1446 tp->rcv_nxt++; 1447 } 1448 switch (tp->t_state) { 1449 1450 /* 1451 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 1452 */ 1453 case TCPS_ESTABLISHED: 1454 tp->t_state = TCPS_CLOSE_WAIT; 1455 break; 1456 1457 /* 1458 * If still in FIN_WAIT_1 STATE FIN has not been acked so 1459 * enter the CLOSING state. 1460 */ 1461 case TCPS_FIN_WAIT_1: 1462 tp->t_state = TCPS_CLOSING; 1463 break; 1464 1465 /* 1466 * In FIN_WAIT_2 state enter the TIME_WAIT state, 1467 * starting the time-wait timer, turning off the other 1468 * standard timers. 1469 */ 1470 case TCPS_FIN_WAIT_2: 1471 tp->t_state = TCPS_TIME_WAIT; 1472 tcp_canceltimers(tp); 1473 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1474 soisdisconnected(so); 1475 break; 1476 1477 /* 1478 * In TIME_WAIT state restart the 2 MSL time_wait timer. 1479 */ 1480 case TCPS_TIME_WAIT: 1481 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1482 break; 1483 } 1484 } 1485 if (so->so_options & SO_DEBUG) 1486 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0); 1487 1488 /* 1489 * Return any desired output. 1490 */ 1491 if (needoutput || (tp->t_flags & TF_ACKNOW)) 1492 (void) tcp_output(tp); 1493 return; 1494 1495 badsyn: 1496 /* 1497 * Received a bad SYN. Increment counters and dropwithreset. 1498 */ 1499 tcpstat.tcps_badsyn++; 1500 tp = NULL; 1501 goto dropwithreset; 1502 1503 dropafterack: 1504 /* 1505 * Generate an ACK dropping incoming segment if it occupies 1506 * sequence space, where the ACK reflects our state. 1507 */ 1508 if (tiflags & TH_RST) 1509 goto drop; 1510 m_freem(m); 1511 tp->t_flags |= TF_ACKNOW; 1512 (void) tcp_output(tp); 1513 return; 1514 1515 dropwithreset: 1516 /* 1517 * Generate a RST, dropping incoming segment. 1518 * Make ACK acceptable to originator of segment. 1519 * Don't bother to respond if destination was broadcast/multicast. 1520 */ 1521 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) || 1522 IN_MULTICAST(ti->ti_dst.s_addr)) 1523 goto drop; 1524 if (tiflags & TH_ACK) 1525 (void)tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST); 1526 else { 1527 if (tiflags & TH_SYN) 1528 ti->ti_len++; 1529 (void)tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0, 1530 TH_RST|TH_ACK); 1531 } 1532 return; 1533 1534 drop: 1535 /* 1536 * Drop space held by incoming segment and return. 1537 */ 1538 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 1539 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0); 1540 m_freem(m); 1541 return; 1542 } 1543 1544 void 1545 tcp_dooptions(tp, cp, cnt, ti, oi) 1546 struct tcpcb *tp; 1547 u_char *cp; 1548 int cnt; 1549 struct tcpiphdr *ti; 1550 struct tcp_opt_info *oi; 1551 { 1552 u_int16_t mss; 1553 int opt, optlen; 1554 1555 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1556 opt = cp[0]; 1557 if (opt == TCPOPT_EOL) 1558 break; 1559 if (opt == TCPOPT_NOP) 1560 optlen = 1; 1561 else { 1562 optlen = cp[1]; 1563 if (optlen <= 0) 1564 break; 1565 } 1566 switch (opt) { 1567 1568 default: 1569 continue; 1570 1571 case TCPOPT_MAXSEG: 1572 if (optlen != TCPOLEN_MAXSEG) 1573 continue; 1574 if (!(ti->ti_flags & TH_SYN)) 1575 continue; 1576 bcopy(cp + 2, &mss, sizeof(mss)); 1577 oi->maxseg = ntohs(mss); 1578 break; 1579 1580 case TCPOPT_WINDOW: 1581 if (optlen != TCPOLEN_WINDOW) 1582 continue; 1583 if (!(ti->ti_flags & TH_SYN)) 1584 continue; 1585 tp->t_flags |= TF_RCVD_SCALE; 1586 tp->requested_s_scale = cp[2]; 1587 if (tp->requested_s_scale > TCP_MAX_WINSHIFT) { 1588 log(LOG_ERR, "TCP: invalid wscale %d from " 1589 "0x%08x, assuming %d\n", 1590 tp->requested_s_scale, 1591 ntohl(ti->ti_src.s_addr), 1592 TCP_MAX_WINSHIFT); 1593 tp->requested_s_scale = TCP_MAX_WINSHIFT; 1594 } 1595 break; 1596 1597 case TCPOPT_TIMESTAMP: 1598 if (optlen != TCPOLEN_TIMESTAMP) 1599 continue; 1600 oi->ts_present = 1; 1601 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 1602 NTOHL(oi->ts_val); 1603 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 1604 NTOHL(oi->ts_ecr); 1605 1606 /* 1607 * A timestamp received in a SYN makes 1608 * it ok to send timestamp requests and replies. 1609 */ 1610 if (ti->ti_flags & TH_SYN) { 1611 tp->t_flags |= TF_RCVD_TSTMP; 1612 tp->ts_recent = oi->ts_val; 1613 tp->ts_recent_age = tcp_now; 1614 } 1615 break; 1616 case TCPOPT_SACK_PERMITTED: 1617 if (optlen != TCPOLEN_SACK_PERMITTED) 1618 continue; 1619 if (!(ti->ti_flags & TH_SYN)) 1620 continue; 1621 tp->t_flags &= ~TF_CANT_TXSACK; 1622 break; 1623 1624 case TCPOPT_SACK: 1625 if (tp->t_flags & TF_IGNR_RXSACK) 1626 continue; 1627 if (optlen % 8 != 2 || optlen < 10) 1628 continue; 1629 cp += 2; 1630 optlen -= 2; 1631 for (; optlen > 0; cp -= 8, optlen -= 8) { 1632 tcp_seq lwe, rwe; 1633 bcopy((char *)cp, (char *) &lwe, sizeof(lwe)); 1634 NTOHL(lwe); 1635 bcopy((char *)cp, (char *) &rwe, sizeof(rwe)); 1636 NTOHL(rwe); 1637 /* tcp_mark_sacked(tp, lwe, rwe); */ 1638 } 1639 break; 1640 } 1641 } 1642 } 1643 1644 /* 1645 * Pull out of band byte out of a segment so 1646 * it doesn't appear in the user's data queue. 1647 * It is still reflected in the segment length for 1648 * sequencing purposes. 1649 */ 1650 void 1651 tcp_pulloutofband(so, ti, m) 1652 struct socket *so; 1653 struct tcpiphdr *ti; 1654 register struct mbuf *m; 1655 { 1656 int cnt = ti->ti_urp - 1; 1657 1658 while (cnt >= 0) { 1659 if (m->m_len > cnt) { 1660 char *cp = mtod(m, caddr_t) + cnt; 1661 struct tcpcb *tp = sototcpcb(so); 1662 1663 tp->t_iobc = *cp; 1664 tp->t_oobflags |= TCPOOB_HAVEDATA; 1665 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 1666 m->m_len--; 1667 return; 1668 } 1669 cnt -= m->m_len; 1670 m = m->m_next; 1671 if (m == 0) 1672 break; 1673 } 1674 panic("tcp_pulloutofband"); 1675 } 1676 1677 /* 1678 * Collect new round-trip time estimate 1679 * and update averages and current timeout. 1680 */ 1681 void 1682 tcp_xmit_timer(tp, rtt) 1683 register struct tcpcb *tp; 1684 short rtt; 1685 { 1686 register short delta; 1687 short rttmin; 1688 1689 tcpstat.tcps_rttupdated++; 1690 --rtt; 1691 if (tp->t_srtt != 0) { 1692 /* 1693 * srtt is stored as fixed point with 3 bits after the 1694 * binary point (i.e., scaled by 8). The following magic 1695 * is equivalent to the smoothing algorithm in rfc793 with 1696 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 1697 * point). Adjust rtt to origin 0. 1698 */ 1699 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 1700 if ((tp->t_srtt += delta) <= 0) 1701 tp->t_srtt = 1 << 2; 1702 /* 1703 * We accumulate a smoothed rtt variance (actually, a 1704 * smoothed mean difference), then set the retransmit 1705 * timer to smoothed rtt + 4 times the smoothed variance. 1706 * rttvar is stored as fixed point with 2 bits after the 1707 * binary point (scaled by 4). The following is 1708 * equivalent to rfc793 smoothing with an alpha of .75 1709 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 1710 * rfc793's wired-in beta. 1711 */ 1712 if (delta < 0) 1713 delta = -delta; 1714 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 1715 if ((tp->t_rttvar += delta) <= 0) 1716 tp->t_rttvar = 1 << 2; 1717 } else { 1718 /* 1719 * No rtt measurement yet - use the unsmoothed rtt. 1720 * Set the variance to half the rtt (so our first 1721 * retransmit happens at 3*rtt). 1722 */ 1723 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 1724 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 1725 } 1726 tp->t_rtt = 0; 1727 tp->t_rxtshift = 0; 1728 1729 /* 1730 * the retransmit should happen at rtt + 4 * rttvar. 1731 * Because of the way we do the smoothing, srtt and rttvar 1732 * will each average +1/2 tick of bias. When we compute 1733 * the retransmit timer, we want 1/2 tick of rounding and 1734 * 1 extra tick because of +-1/2 tick uncertainty in the 1735 * firing of the timer. The bias will give us exactly the 1736 * 1.5 tick we need. But, because the bias is 1737 * statistical, we have to test that we don't drop below 1738 * the minimum feasible timer (which is 2 ticks). 1739 */ 1740 if (tp->t_rttmin > rtt + 2) 1741 rttmin = tp->t_rttmin; 1742 else 1743 rttmin = rtt + 2; 1744 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 1745 1746 /* 1747 * We received an ack for a packet that wasn't retransmitted; 1748 * it is probably safe to discard any error indications we've 1749 * received recently. This isn't quite right, but close enough 1750 * for now (a route might have failed after we sent a segment, 1751 * and the return path might not be symmetrical). 1752 */ 1753 tp->t_softerror = 0; 1754 } 1755 1756 /* 1757 * TCP compressed state engine. Currently used to hold compressed 1758 * state for SYN_RECEIVED. 1759 */ 1760 1761 u_long syn_cache_count; 1762 u_int32_t syn_hash1, syn_hash2; 1763 1764 #define SYN_HASH(sa, sp, dp) \ 1765 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 1766 ((u_int32_t)(sp)))^syn_hash2))) 1767 1768 LIST_HEAD(, syn_cache_head) tcp_syn_cache_queue; 1769 1770 #define SYN_CACHE_RM(sc, scp) \ 1771 do { \ 1772 TAILQ_REMOVE(&(scp)->sch_queue, (sc), sc_queue); \ 1773 if (--(scp)->sch_length == 0) \ 1774 LIST_REMOVE((scp), sch_headq); \ 1775 syn_cache_count--; \ 1776 } while (0) 1777 1778 struct pool syn_cache_pool; 1779 1780 void 1781 syn_cache_init() 1782 { 1783 int i; 1784 1785 /* Initialize the hash bucket queues. */ 1786 for (i = 0; i < tcp_syn_cache_size; i++) 1787 TAILQ_INIT(&tcp_syn_cache[i].sch_queue); 1788 1789 /* Initialize the active hash bucket cache. */ 1790 LIST_INIT(&tcp_syn_cache_queue); 1791 1792 /* Initialize the syn cache pool. */ 1793 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 1794 "synpl", 0, NULL, NULL, M_PCB); 1795 } 1796 1797 void 1798 syn_cache_insert(sc) 1799 struct syn_cache *sc; 1800 { 1801 struct syn_cache_head *scp, *scp2, *sce; 1802 struct syn_cache *sc2; 1803 int s; 1804 1805 /* 1806 * If there are no entries in the hash table, reinitialize 1807 * the hash secrets. 1808 */ 1809 if (syn_cache_count == 0) { 1810 struct timeval tv; 1811 microtime(&tv); 1812 syn_hash1 = random() ^ (u_long)≻ 1813 syn_hash2 = random() ^ tv.tv_usec; 1814 } 1815 1816 sc->sc_hash = SYN_HASH(&sc->sc_src, sc->sc_sport, sc->sc_dport); 1817 scp = &tcp_syn_cache[sc->sc_hash % tcp_syn_cache_size]; 1818 1819 /* 1820 * Make sure that we don't overflow the per-bucket 1821 * limit or the total cache size limit. 1822 */ 1823 s = splsoftnet(); 1824 if (scp->sch_length >= tcp_syn_bucket_limit) { 1825 tcpstat.tcps_sc_bucketoverflow++; 1826 /* 1827 * The bucket is full. Toss the first (i.e. oldest) 1828 * element in this bucket. 1829 */ 1830 sc2 = TAILQ_FIRST(&scp->sch_queue); 1831 SYN_CACHE_RM(sc2, scp); 1832 if (sc2->sc_ipopts) 1833 (void) m_free(sc2->sc_ipopts); 1834 pool_put(&syn_cache_pool, sc2); 1835 } else if (syn_cache_count >= tcp_syn_cache_limit) { 1836 tcpstat.tcps_sc_overflowed++; 1837 /* 1838 * The cache is full. Toss the first (i.e. oldest) 1839 * element in the first non-empty bucket we can find. 1840 */ 1841 scp2 = scp; 1842 if (TAILQ_FIRST(&scp2->sch_queue) == NULL) { 1843 sce = &tcp_syn_cache[tcp_syn_cache_size]; 1844 for (++scp2; scp2 != scp; scp2++) { 1845 if (scp2 >= sce) 1846 scp2 = &tcp_syn_cache[0]; 1847 if (TAILQ_FIRST(&scp2->sch_queue) != NULL) 1848 break; 1849 } 1850 } 1851 sc2 = TAILQ_FIRST(&scp2->sch_queue); 1852 if (sc2 == NULL) { 1853 if (sc->sc_ipopts) 1854 (void) m_free(sc->sc_ipopts); 1855 pool_put(&syn_cache_pool, sc); 1856 return; 1857 } 1858 SYN_CACHE_RM(sc2, scp2); 1859 if (sc2->sc_ipopts) 1860 (void) m_free(sc2->sc_ipopts); 1861 pool_put(&syn_cache_pool, sc2); 1862 } 1863 1864 /* Set entry's timer. */ 1865 PRT_SLOW_ARM(sc->sc_timer, tcp_syn_cache_timeo); 1866 1867 /* Put it into the bucket. */ 1868 TAILQ_INSERT_TAIL(&scp->sch_queue, sc, sc_queue); 1869 if (++scp->sch_length == 1) 1870 LIST_INSERT_HEAD(&tcp_syn_cache_queue, scp, sch_headq); 1871 syn_cache_count++; 1872 1873 tcpstat.tcps_sc_added++; 1874 splx(s); 1875 } 1876 1877 /* 1878 * Walk down the cache list, looking for expired entries in each bucket. 1879 */ 1880 void 1881 syn_cache_timer() 1882 { 1883 struct syn_cache_head *scp, *nscp; 1884 struct syn_cache *sc, *nsc; 1885 int s; 1886 1887 s = splsoftnet(); 1888 for (scp = LIST_FIRST(&tcp_syn_cache_queue); scp != NULL; scp = nscp) { 1889 #ifdef DIAGNOSTIC 1890 if (TAILQ_FIRST(&scp->sch_queue) == NULL) 1891 panic("syn_cache_timer: queue inconsistency"); 1892 #endif 1893 nscp = LIST_NEXT(scp, sch_headq); 1894 for (sc = TAILQ_FIRST(&scp->sch_queue); 1895 sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_timer); 1896 sc = nsc) { 1897 nsc = TAILQ_NEXT(sc, sc_queue); 1898 tcpstat.tcps_sc_timed_out++; 1899 SYN_CACHE_RM(sc, scp); 1900 if (sc->sc_ipopts) 1901 (void) m_free(sc->sc_ipopts); 1902 pool_put(&syn_cache_pool, sc); 1903 } 1904 } 1905 splx(s); 1906 } 1907 1908 /* 1909 * Find an entry in the syn cache. 1910 */ 1911 struct syn_cache * 1912 syn_cache_lookup(ti, headp) 1913 struct tcpiphdr *ti; 1914 struct syn_cache_head **headp; 1915 { 1916 struct syn_cache *sc; 1917 struct syn_cache_head *scp; 1918 u_int32_t hash; 1919 int s; 1920 1921 hash = SYN_HASH(&ti->ti_src, ti->ti_sport, ti->ti_dport); 1922 1923 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 1924 *headp = scp; 1925 s = splsoftnet(); 1926 for (sc = TAILQ_FIRST(&scp->sch_queue); sc != NULL; 1927 sc = TAILQ_NEXT(sc, sc_queue)) { 1928 if (sc->sc_hash != hash) 1929 continue; 1930 if (sc->sc_src.s_addr == ti->ti_src.s_addr && 1931 sc->sc_sport == ti->ti_sport && 1932 sc->sc_dport == ti->ti_dport && 1933 sc->sc_dst.s_addr == ti->ti_dst.s_addr) { 1934 splx(s); 1935 return (sc); 1936 } 1937 } 1938 splx(s); 1939 return (NULL); 1940 } 1941 1942 /* 1943 * This function gets called when we receive an ACK for a 1944 * socket in the LISTEN state. We look up the connection 1945 * in the syn cache, and if its there, we pull it out of 1946 * the cache and turn it into a full-blown connection in 1947 * the SYN-RECEIVED state. 1948 * 1949 * The return values may not be immediately obvious, and their effects 1950 * can be subtle, so here they are: 1951 * 1952 * NULL SYN was not found in cache; caller should drop the 1953 * packet and send an RST. 1954 * 1955 * -1 We were unable to create the new connection, and are 1956 * aborting it. An ACK,RST is being sent to the peer 1957 * (unless we got screwey sequence numbners; see below), 1958 * because the 3-way handshake has been completed. Caller 1959 * should not free the mbuf, since we may be using it. If 1960 * we are not, we will free it. 1961 * 1962 * Otherwise, the return value is a pointer to the new socket 1963 * associated with the connection. 1964 */ 1965 struct socket * 1966 syn_cache_get(so, m) 1967 struct socket *so; 1968 struct mbuf *m; 1969 { 1970 struct syn_cache *sc; 1971 struct syn_cache_head *scp; 1972 register struct inpcb *inp; 1973 register struct tcpcb *tp = 0; 1974 register struct tcpiphdr *ti; 1975 struct sockaddr_in *sin; 1976 struct mbuf *am; 1977 long win; 1978 int s; 1979 1980 ti = mtod(m, struct tcpiphdr *); 1981 s = splsoftnet(); 1982 if ((sc = syn_cache_lookup(ti, &scp)) == NULL) { 1983 splx(s); 1984 return (NULL); 1985 } 1986 1987 win = sbspace(&so->so_rcv); 1988 if (win > TCP_MAXWIN) 1989 win = TCP_MAXWIN; 1990 1991 /* 1992 * Verify the sequence and ack numbers. 1993 */ 1994 if ((ti->ti_ack != sc->sc_iss + 1) || 1995 SEQ_LEQ(ti->ti_seq, sc->sc_irs) || 1996 SEQ_GT(ti->ti_seq, sc->sc_irs + 1 + win)) { 1997 (void) syn_cache_respond(sc, m, ti, win, 0); 1998 splx(s); 1999 return ((struct socket *)(-1)); 2000 } 2001 2002 /* Remove this cache entry */ 2003 SYN_CACHE_RM(sc, scp); 2004 splx(s); 2005 2006 /* 2007 * Ok, create the full blown connection, and set things up 2008 * as they would have been set up if we had created the 2009 * connection when the SYN arrived. If we can't create 2010 * the connection, abort it. 2011 */ 2012 so = sonewconn(so, SS_ISCONNECTED); 2013 if (so == NULL) 2014 goto resetandabort; 2015 2016 inp = sotoinpcb(so); 2017 inp->inp_laddr = sc->sc_dst; 2018 inp->inp_lport = sc->sc_dport; 2019 in_pcbstate(inp, INP_BOUND); 2020 inp->inp_options = ip_srcroute(); 2021 if (inp->inp_options == NULL) { 2022 inp->inp_options = sc->sc_ipopts; 2023 sc->sc_ipopts = NULL; 2024 } 2025 2026 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 2027 if (am == NULL) 2028 goto resetandabort; 2029 am->m_len = sizeof(struct sockaddr_in); 2030 sin = mtod(am, struct sockaddr_in *); 2031 sin->sin_family = AF_INET; 2032 sin->sin_len = sizeof(*sin); 2033 sin->sin_addr = sc->sc_src; 2034 sin->sin_port = sc->sc_sport; 2035 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 2036 if (in_pcbconnect(inp, am)) { 2037 (void) m_free(am); 2038 goto resetandabort; 2039 } 2040 (void) m_free(am); 2041 2042 tp = intotcpcb(inp); 2043 if (sc->sc_request_r_scale != 15) { 2044 tp->requested_s_scale = sc->sc_requested_s_scale; 2045 tp->request_r_scale = sc->sc_request_r_scale; 2046 tp->snd_scale = sc->sc_requested_s_scale; 2047 tp->rcv_scale = sc->sc_request_r_scale; 2048 tp->t_flags |= TF_RCVD_SCALE; 2049 } 2050 if (sc->sc_flags & SCF_TIMESTAMP) 2051 tp->t_flags |= TF_RCVD_TSTMP; 2052 2053 tp->t_template = tcp_template(tp); 2054 if (tp->t_template == 0) { 2055 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 2056 so = NULL; 2057 m_freem(m); 2058 goto abort; 2059 } 2060 2061 tp->iss = sc->sc_iss; 2062 tp->irs = sc->sc_irs; 2063 tcp_sendseqinit(tp); 2064 tcp_rcvseqinit(tp); 2065 tp->t_state = TCPS_SYN_RECEIVED; 2066 TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT); 2067 tcpstat.tcps_accepts++; 2068 2069 /* Initialize tp->t_ourmss before we deal with the peer's! */ 2070 tp->t_ourmss = sc->sc_ourmaxseg; 2071 tcp_mss_from_peer(tp, sc->sc_peermaxseg); 2072 2073 /* 2074 * Initialize the initial congestion window. If we 2075 * had to retransmit the SYN,ACK, we must initialize cwnd 2076 * to 1 segment (i.e. the Loss Window). 2077 */ 2078 if (sc->sc_flags & SCF_SYNACK_REXMT) 2079 tp->snd_cwnd = tp->t_peermss; 2080 else 2081 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss); 2082 2083 tcp_rmx_rtt(tp); 2084 tp->snd_wl1 = sc->sc_irs; 2085 tp->rcv_up = sc->sc_irs + 1; 2086 2087 /* 2088 * This is what whould have happened in tcp_ouput() when 2089 * the SYN,ACK was sent. 2090 */ 2091 tp->snd_up = tp->snd_una; 2092 tp->snd_max = tp->snd_nxt = tp->iss+1; 2093 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 2094 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv)) 2095 tp->rcv_adv = tp->rcv_nxt + win; 2096 tp->last_ack_sent = tp->rcv_nxt; 2097 2098 tcpstat.tcps_sc_completed++; 2099 if (sc->sc_ipopts) 2100 (void) m_free(sc->sc_ipopts); 2101 pool_put(&syn_cache_pool, sc); 2102 return (so); 2103 2104 resetandabort: 2105 (void) tcp_respond(NULL, ti, m, ti->ti_seq+ti->ti_len, 2106 (tcp_seq)0, TH_RST|TH_ACK); 2107 abort: 2108 if (so != NULL) 2109 (void) soabort(so); 2110 if (sc->sc_ipopts) 2111 (void) m_free(sc->sc_ipopts); 2112 pool_put(&syn_cache_pool, sc); 2113 tcpstat.tcps_sc_aborted++; 2114 return ((struct socket *)(-1)); 2115 } 2116 2117 /* 2118 * This function is called when we get a RST for a 2119 * non-existant connection, so that we can see if the 2120 * connection is in the syn cache. If it is, zap it. 2121 */ 2122 2123 void 2124 syn_cache_reset(ti) 2125 register struct tcpiphdr *ti; 2126 { 2127 struct syn_cache *sc; 2128 struct syn_cache_head *scp; 2129 int s = splsoftnet(); 2130 2131 if ((sc = syn_cache_lookup(ti, &scp)) == NULL) { 2132 splx(s); 2133 return; 2134 } 2135 if (SEQ_LT(ti->ti_seq,sc->sc_irs) || 2136 SEQ_GT(ti->ti_seq, sc->sc_irs+1)) { 2137 splx(s); 2138 return; 2139 } 2140 SYN_CACHE_RM(sc, scp); 2141 splx(s); 2142 tcpstat.tcps_sc_reset++; 2143 if (sc->sc_ipopts) 2144 (void) m_free(sc->sc_ipopts); 2145 pool_put(&syn_cache_pool, sc); 2146 } 2147 2148 void 2149 syn_cache_unreach(ip, th) 2150 struct ip *ip; 2151 struct tcphdr *th; 2152 { 2153 struct syn_cache *sc; 2154 struct syn_cache_head *scp; 2155 struct tcpiphdr ti2; 2156 int s; 2157 2158 ti2.ti_src.s_addr = ip->ip_dst.s_addr; 2159 ti2.ti_dst.s_addr = ip->ip_src.s_addr; 2160 ti2.ti_sport = th->th_dport; 2161 ti2.ti_dport = th->th_sport; 2162 2163 s = splsoftnet(); 2164 if ((sc = syn_cache_lookup(&ti2, &scp)) == NULL) { 2165 splx(s); 2166 return; 2167 } 2168 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 2169 if (ntohl (th->th_seq) != sc->sc_iss) { 2170 splx(s); 2171 return; 2172 } 2173 SYN_CACHE_RM(sc, scp); 2174 splx(s); 2175 tcpstat.tcps_sc_unreach++; 2176 if (sc->sc_ipopts) 2177 (void) m_free(sc->sc_ipopts); 2178 pool_put(&syn_cache_pool, sc); 2179 } 2180 2181 /* 2182 * Given a LISTEN socket and an inbound SYN request, add 2183 * this to the syn cache, and send back a segment: 2184 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 2185 * to the source. 2186 * 2187 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 2188 * Doing so would require that we hold onto the data and deliver it 2189 * to the application. However, if we are the target of a SYN-flood 2190 * DoS attack, an attacker could send data which would eventually 2191 * consume all available buffer space if it were ACKed. By not ACKing 2192 * the data, we avoid this DoS scenario. 2193 */ 2194 2195 int 2196 syn_cache_add(so, m, optp, optlen, oi) 2197 struct socket *so; 2198 struct mbuf *m; 2199 u_char *optp; 2200 int optlen; 2201 struct tcp_opt_info *oi; 2202 { 2203 register struct tcpiphdr *ti; 2204 struct tcpcb tb, *tp; 2205 long win; 2206 struct syn_cache *sc; 2207 struct syn_cache_head *scp; 2208 struct mbuf *ipopts; 2209 2210 tp = sototcpcb(so); 2211 ti = mtod(m, struct tcpiphdr *); 2212 2213 /* 2214 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 2215 * in_broadcast() should never return true on a received 2216 * packet with M_BCAST not set. 2217 */ 2218 if (m->m_flags & (M_BCAST|M_MCAST) || 2219 IN_MULTICAST(ti->ti_src.s_addr) || 2220 IN_MULTICAST(ti->ti_dst.s_addr)) 2221 return (0); 2222 2223 /* 2224 * Initialize some local state. 2225 */ 2226 win = sbspace(&so->so_rcv); 2227 if (win > TCP_MAXWIN) 2228 win = TCP_MAXWIN; 2229 2230 /* 2231 * Remember the IP options, if any. 2232 */ 2233 ipopts = ip_srcroute(); 2234 2235 if (optp) { 2236 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 2237 tcp_dooptions(&tb, optp, optlen, ti, oi); 2238 } else 2239 tb.t_flags = 0; 2240 2241 /* 2242 * See if we already have an entry for this connection. 2243 * If we do, resend the SYN,ACK, and remember since the 2244 * initial congestion window must be initialized to 1 2245 * segment when the connection completes. 2246 */ 2247 if ((sc = syn_cache_lookup(ti, &scp)) != NULL) { 2248 tcpstat.tcps_sc_dupesyn++; 2249 sc->sc_flags |= SCF_SYNACK_REXMT; 2250 2251 if (ipopts) { 2252 /* 2253 * If we were remembering a previous source route, 2254 * forget it and use the new one we've been given. 2255 */ 2256 if (sc->sc_ipopts) 2257 (void) m_free(sc->sc_ipopts); 2258 sc->sc_ipopts = ipopts; 2259 } 2260 2261 if (syn_cache_respond(sc, m, ti, win, tb.ts_recent) == 0) { 2262 tcpstat.tcps_sndacks++; 2263 tcpstat.tcps_sndtotal++; 2264 } 2265 return (1); 2266 } 2267 2268 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 2269 if (sc == NULL) { 2270 if (ipopts) 2271 (void) m_free(ipopts); 2272 return (0); 2273 } 2274 2275 /* 2276 * Fill in the cache, and put the necessary IP and TCP 2277 * options into the reply. 2278 */ 2279 sc->sc_src.s_addr = ti->ti_src.s_addr; 2280 sc->sc_dst.s_addr = ti->ti_dst.s_addr; 2281 sc->sc_sport = ti->ti_sport; 2282 sc->sc_dport = ti->ti_dport; 2283 sc->sc_flags = 0; 2284 sc->sc_ipopts = ipopts; 2285 sc->sc_irs = ti->ti_seq; 2286 sc->sc_iss = tcp_new_iss(sc, sizeof(struct syn_cache), 0); 2287 sc->sc_peermaxseg = oi->maxseg; 2288 sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ? 2289 m->m_pkthdr.rcvif : NULL); 2290 if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) 2291 sc->sc_flags |= SCF_TIMESTAMP; 2292 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2293 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2294 sc->sc_requested_s_scale = tb.requested_s_scale; 2295 sc->sc_request_r_scale = 0; 2296 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 2297 TCP_MAXWIN << sc->sc_request_r_scale < 2298 so->so_rcv.sb_hiwat) 2299 sc->sc_request_r_scale++; 2300 } else { 2301 sc->sc_requested_s_scale = 15; 2302 sc->sc_request_r_scale = 15; 2303 } 2304 if (syn_cache_respond(sc, m, ti, win, tb.ts_recent) == 0) { 2305 syn_cache_insert(sc); 2306 tcpstat.tcps_sndacks++; 2307 tcpstat.tcps_sndtotal++; 2308 } else { 2309 if (sc->sc_ipopts) 2310 (void) m_free(sc->sc_ipopts); 2311 pool_put(&syn_cache_pool, sc); 2312 tcpstat.tcps_sc_dropped++; 2313 } 2314 return (1); 2315 } 2316 2317 int 2318 syn_cache_respond(sc, m, ti, win, ts) 2319 struct syn_cache *sc; 2320 struct mbuf *m; 2321 register struct tcpiphdr *ti; 2322 long win; 2323 u_long ts; 2324 { 2325 u_int8_t *optp; 2326 int optlen; 2327 2328 /* 2329 * Tack on the TCP options. If there isn't enough trailing 2330 * space for them, move up the fixed header to make space. 2331 */ 2332 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 2333 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 2334 if (optlen > M_TRAILINGSPACE(m)) { 2335 if (M_LEADINGSPACE(m) >= optlen) { 2336 m->m_data -= optlen; 2337 m->m_len += optlen; 2338 } else { 2339 struct mbuf *m0 = m; 2340 if ((m = m_gethdr(M_DONTWAIT, MT_HEADER)) == NULL) { 2341 m_freem(m0); 2342 return (ENOBUFS); 2343 } 2344 MH_ALIGN(m, sizeof(*ti) + optlen); 2345 m->m_next = m0; /* this gets freed below */ 2346 } 2347 bcopy((caddr_t)ti, mtod(m, caddr_t), sizeof(*ti)); 2348 ti = mtod(m, struct tcpiphdr *); 2349 } 2350 2351 optp = (u_int8_t *)(ti + 1); 2352 optp[0] = TCPOPT_MAXSEG; 2353 optp[1] = 4; 2354 optp[2] = (sc->sc_ourmaxseg >> 8) & 0xff; 2355 optp[3] = sc->sc_ourmaxseg & 0xff; 2356 optlen = 4; 2357 2358 if (sc->sc_request_r_scale != 15) { 2359 *((u_int32_t *)(optp + optlen)) = htonl(TCPOPT_NOP << 24 | 2360 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 2361 sc->sc_request_r_scale); 2362 optlen += 4; 2363 } 2364 2365 if (sc->sc_flags & SCF_TIMESTAMP) { 2366 u_int32_t *lp = (u_int32_t *)(optp + optlen); 2367 /* Form timestamp option as shown in appendix A of RFC 1323. */ 2368 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 2369 *lp++ = htonl(tcp_now); 2370 *lp = htonl(ts); 2371 optlen += TCPOLEN_TSTAMP_APPA; 2372 } 2373 2374 /* 2375 * Toss any trailing mbufs. No need to worry about 2376 * m_len and m_pkthdr.len, since tcp_respond() will 2377 * unconditionally set them. 2378 */ 2379 if (m->m_next) { 2380 m_freem(m->m_next); 2381 m->m_next = NULL; 2382 } 2383 2384 /* 2385 * Fill in the fields that tcp_respond() will not touch, and 2386 * then send the response. 2387 */ 2388 ti->ti_off = (sizeof(struct tcphdr) + optlen) >> 2; 2389 ti->ti_win = htons(win); 2390 return (tcp_respond(NULL, ti, m, sc->sc_irs + 1, sc->sc_iss, 2391 TH_SYN|TH_ACK)); 2392 } 2393
Indexes created Mon Sep 22 10:09:38 GMT 2025