1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 34 * $FreeBSD$ 35 */ 36 37 #include "opt_compat.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_tcpdebug.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/callout.h> 45 #include <sys/kernel.h> 46 #include <sys/sysctl.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #ifdef INET6 50 #include <sys/domain.h> 51 #endif 52 #include <sys/proc.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/protosw.h> 56 #include <sys/random.h> 57 58 #include <vm/vm_zone.h> 59 60 #include <net/route.h> 61 #include <net/if.h> 62 63 #define _IP_VHL 64 #include <netinet/in.h> 65 #include <netinet/in_systm.h> 66 #include <netinet/ip.h> 67 #ifdef INET6 68 #include <netinet/ip6.h> 69 #endif 70 #include <netinet/in_pcb.h> 71 #ifdef INET6 72 #include <netinet6/in6_pcb.h> 73 #endif 74 #include <netinet/in_var.h> 75 #include <netinet/ip_var.h> 76 #ifdef INET6 77 #include <netinet6/ip6_var.h> 78 #endif 79 #include <netinet/tcp.h> 80 #include <netinet/tcp_fsm.h> 81 #include <netinet/tcp_seq.h> 82 #include <netinet/tcp_timer.h> 83 #include <netinet/tcp_var.h> 84 #ifdef INET6 85 #include <netinet6/tcp6_var.h> 86 #endif 87 #include <netinet/tcpip.h> 88 #ifdef TCPDEBUG 89 #include <netinet/tcp_debug.h> 90 #endif 91 #include <netinet6/ip6protosw.h> 92 93 #ifdef IPSEC 94 #include <netinet6/ipsec.h> 95 #ifdef INET6 96 #include <netinet6/ipsec6.h> 97 #endif 98 #endif /*IPSEC*/ 99 100 #include <machine/in_cksum.h> 101 #include <sys/md5.h> 102 103 int tcp_mssdflt = TCP_MSS; 104 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 105 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 106 107 #ifdef INET6 108 int tcp_v6mssdflt = TCP6_MSS; 109 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 110 CTLFLAG_RW, &tcp_v6mssdflt , 0, 111 "Default TCP Maximum Segment Size for IPv6"); 112 #endif 113 114 #if 0 115 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 116 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 117 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 118 #endif 119 120 static int tcp_do_rfc1323 = 1; 121 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 122 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 123 124 static int tcp_do_rfc1644 = 0; 125 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW, 126 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions"); 127 128 static int tcp_tcbhashsize = 0; 129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, 130 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 131 132 static int do_tcpdrain = 1; 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 134 "Enable tcp_drain routine for extra help when low on mbufs"); 135 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 137 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 138 139 static int icmp_may_rst = 1; 140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 141 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 142 143 static int tcp_strict_rfc1948 = 0; 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, strict_rfc1948, CTLFLAG_RW, 145 &tcp_strict_rfc1948, 0, "Determines if RFC1948 is followed exactly"); 146 147 static int tcp_isn_reseed_interval = 0; 148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 149 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 150 151 static void tcp_cleartaocache __P((void)); 152 static void tcp_notify __P((struct inpcb *, int)); 153 154 /* 155 * Target size of TCP PCB hash tables. Must be a power of two. 156 * 157 * Note that this can be overridden by the kernel environment 158 * variable net.inet.tcp.tcbhashsize 159 */ 160 #ifndef TCBHASHSIZE 161 #define TCBHASHSIZE 512 162 #endif 163 164 /* 165 * This is the actual shape of what we allocate using the zone 166 * allocator. Doing it this way allows us to protect both structures 167 * using the same generation count, and also eliminates the overhead 168 * of allocating tcpcbs separately. By hiding the structure here, 169 * we avoid changing most of the rest of the code (although it needs 170 * to be changed, eventually, for greater efficiency). 171 */ 172 #define ALIGNMENT 32 173 #define ALIGNM1 (ALIGNMENT - 1) 174 struct inp_tp { 175 union { 176 struct inpcb inp; 177 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 178 } inp_tp_u; 179 struct tcpcb tcb; 180 struct callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl; 181 struct callout inp_tp_delack; 182 }; 183 #undef ALIGNMENT 184 #undef ALIGNM1 185 186 /* 187 * Tcp initialization 188 */ 189 void 190 tcp_init() 191 { 192 int hashsize = TCBHASHSIZE; 193 194 tcp_ccgen = 1; 195 tcp_cleartaocache(); 196 197 tcp_delacktime = TCPTV_DELACK; 198 tcp_keepinit = TCPTV_KEEP_INIT; 199 tcp_keepidle = TCPTV_KEEP_IDLE; 200 tcp_keepintvl = TCPTV_KEEPINTVL; 201 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 202 tcp_msl = TCPTV_MSL; 203 204 LIST_INIT(&tcb); 205 tcbinfo.listhead = &tcb; 206 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 207 if (!powerof2(hashsize)) { 208 printf("WARNING: TCB hash size not a power of 2\n"); 209 hashsize = 512; /* safe default */ 210 } 211 tcp_tcbhashsize = hashsize; 212 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 213 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 214 &tcbinfo.porthashmask); 215 tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets, 216 ZONE_INTERRUPT, 0); 217 #ifdef INET6 218 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 219 #else /* INET6 */ 220 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 221 #endif /* INET6 */ 222 if (max_protohdr < TCP_MINPROTOHDR) 223 max_protohdr = TCP_MINPROTOHDR; 224 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 225 panic("tcp_init"); 226 #undef TCP_MINPROTOHDR 227 } 228 229 /* 230 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 231 * tcp_template used to store this data in mbufs, but we now recopy it out 232 * of the tcpcb each time to conserve mbufs. 233 */ 234 void 235 tcp_fillheaders(tp, ip_ptr, tcp_ptr) 236 struct tcpcb *tp; 237 void *ip_ptr; 238 void *tcp_ptr; 239 { 240 struct inpcb *inp = tp->t_inpcb; 241 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr; 242 243 #ifdef INET6 244 if ((inp->inp_vflag & INP_IPV6) != 0) { 245 struct ip6_hdr *ip6; 246 247 ip6 = (struct ip6_hdr *)ip_ptr; 248 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 249 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 250 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 251 (IPV6_VERSION & IPV6_VERSION_MASK); 252 ip6->ip6_nxt = IPPROTO_TCP; 253 ip6->ip6_plen = sizeof(struct tcphdr); 254 ip6->ip6_src = inp->in6p_laddr; 255 ip6->ip6_dst = inp->in6p_faddr; 256 tcp_hdr->th_sum = 0; 257 } else 258 #endif 259 { 260 struct ip *ip = (struct ip *) ip_ptr; 261 262 ip->ip_vhl = IP_VHL_BORING; 263 ip->ip_tos = 0; 264 ip->ip_len = 0; 265 ip->ip_id = 0; 266 ip->ip_off = 0; 267 ip->ip_ttl = 0; 268 ip->ip_sum = 0; 269 ip->ip_p = IPPROTO_TCP; 270 ip->ip_src = inp->inp_laddr; 271 ip->ip_dst = inp->inp_faddr; 272 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 273 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 274 } 275 276 tcp_hdr->th_sport = inp->inp_lport; 277 tcp_hdr->th_dport = inp->inp_fport; 278 tcp_hdr->th_seq = 0; 279 tcp_hdr->th_ack = 0; 280 tcp_hdr->th_x2 = 0; 281 tcp_hdr->th_off = 5; 282 tcp_hdr->th_flags = 0; 283 tcp_hdr->th_win = 0; 284 tcp_hdr->th_urp = 0; 285 } 286 287 /* 288 * Create template to be used to send tcp packets on a connection. 289 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 290 * use for this function is in keepalives, which use tcp_respond. 291 */ 292 struct tcptemp * 293 tcp_maketemplate(tp) 294 struct tcpcb *tp; 295 { 296 struct mbuf *m; 297 struct tcptemp *n; 298 299 m = m_get(M_DONTWAIT, MT_HEADER); 300 if (m == NULL) 301 return (0); 302 m->m_len = sizeof(struct tcptemp); 303 n = mtod(m, struct tcptemp *); 304 305 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 306 return (n); 307 } 308 309 /* 310 * Send a single message to the TCP at address specified by 311 * the given TCP/IP header. If m == 0, then we make a copy 312 * of the tcpiphdr at ti and send directly to the addressed host. 313 * This is used to force keep alive messages out using the TCP 314 * template for a connection. If flags are given then we send 315 * a message back to the TCP which originated the * segment ti, 316 * and discard the mbuf containing it and any other attached mbufs. 317 * 318 * In any case the ack and sequence number of the transmitted 319 * segment are as specified by the parameters. 320 * 321 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 322 */ 323 void 324 tcp_respond(tp, ipgen, th, m, ack, seq, flags) 325 struct tcpcb *tp; 326 void *ipgen; 327 register struct tcphdr *th; 328 register struct mbuf *m; 329 tcp_seq ack, seq; 330 int flags; 331 { 332 register int tlen; 333 int win = 0; 334 struct route *ro = 0; 335 struct route sro; 336 struct ip *ip; 337 struct tcphdr *nth; 338 #ifdef INET6 339 struct route_in6 *ro6 = 0; 340 struct route_in6 sro6; 341 struct ip6_hdr *ip6; 342 int isipv6; 343 #endif /* INET6 */ 344 int ipflags = 0; 345 346 #ifdef INET6 347 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6; 348 ip6 = ipgen; 349 #endif /* INET6 */ 350 ip = ipgen; 351 352 if (tp) { 353 if (!(flags & TH_RST)) { 354 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 355 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 356 win = (long)TCP_MAXWIN << tp->rcv_scale; 357 } 358 #ifdef INET6 359 if (isipv6) 360 ro6 = &tp->t_inpcb->in6p_route; 361 else 362 #endif /* INET6 */ 363 ro = &tp->t_inpcb->inp_route; 364 } else { 365 #ifdef INET6 366 if (isipv6) { 367 ro6 = &sro6; 368 bzero(ro6, sizeof *ro6); 369 } else 370 #endif /* INET6 */ 371 { 372 ro = &sro; 373 bzero(ro, sizeof *ro); 374 } 375 } 376 if (m == 0) { 377 m = m_gethdr(M_DONTWAIT, MT_HEADER); 378 if (m == NULL) 379 return; 380 tlen = 0; 381 m->m_data += max_linkhdr; 382 #ifdef INET6 383 if (isipv6) { 384 bcopy((caddr_t)ip6, mtod(m, caddr_t), 385 sizeof(struct ip6_hdr)); 386 ip6 = mtod(m, struct ip6_hdr *); 387 nth = (struct tcphdr *)(ip6 + 1); 388 } else 389 #endif /* INET6 */ 390 { 391 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 392 ip = mtod(m, struct ip *); 393 nth = (struct tcphdr *)(ip + 1); 394 } 395 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 396 flags = TH_ACK; 397 } else { 398 m_freem(m->m_next); 399 m->m_next = 0; 400 m->m_data = (caddr_t)ipgen; 401 /* m_len is set later */ 402 tlen = 0; 403 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 404 #ifdef INET6 405 if (isipv6) { 406 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 407 nth = (struct tcphdr *)(ip6 + 1); 408 } else 409 #endif /* INET6 */ 410 { 411 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 412 nth = (struct tcphdr *)(ip + 1); 413 } 414 if (th != nth) { 415 /* 416 * this is usually a case when an extension header 417 * exists between the IPv6 header and the 418 * TCP header. 419 */ 420 nth->th_sport = th->th_sport; 421 nth->th_dport = th->th_dport; 422 } 423 xchg(nth->th_dport, nth->th_sport, n_short); 424 #undef xchg 425 } 426 #ifdef INET6 427 if (isipv6) { 428 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 429 tlen)); 430 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 431 } else 432 #endif 433 { 434 tlen += sizeof (struct tcpiphdr); 435 ip->ip_len = tlen; 436 ip->ip_ttl = ip_defttl; 437 } 438 m->m_len = tlen; 439 m->m_pkthdr.len = tlen; 440 m->m_pkthdr.rcvif = (struct ifnet *) 0; 441 nth->th_seq = htonl(seq); 442 nth->th_ack = htonl(ack); 443 nth->th_x2 = 0; 444 nth->th_off = sizeof (struct tcphdr) >> 2; 445 nth->th_flags = flags; 446 if (tp) 447 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 448 else 449 nth->th_win = htons((u_short)win); 450 nth->th_urp = 0; 451 #ifdef INET6 452 if (isipv6) { 453 nth->th_sum = 0; 454 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 455 sizeof(struct ip6_hdr), 456 tlen - sizeof(struct ip6_hdr)); 457 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, 458 ro6 && ro6->ro_rt ? 459 ro6->ro_rt->rt_ifp : 460 NULL); 461 } else 462 #endif /* INET6 */ 463 { 464 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 465 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 466 m->m_pkthdr.csum_flags = CSUM_TCP; 467 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 468 } 469 #ifdef TCPDEBUG 470 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 471 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 472 #endif 473 #ifdef IPSEC 474 if (ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) { 475 m_freem(m); 476 return; 477 } 478 #endif 479 #ifdef INET6 480 if (isipv6) { 481 (void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL); 482 if (ro6 == &sro6 && ro6->ro_rt) { 483 RTFREE(ro6->ro_rt); 484 ro6->ro_rt = NULL; 485 } 486 } else 487 #endif /* INET6 */ 488 { 489 (void) ip_output(m, NULL, ro, ipflags, NULL); 490 if (ro == &sro && ro->ro_rt) { 491 RTFREE(ro->ro_rt); 492 ro->ro_rt = NULL; 493 } 494 } 495 } 496 497 /* 498 * Create a new TCP control block, making an 499 * empty reassembly queue and hooking it to the argument 500 * protocol control block. The `inp' parameter must have 501 * come from the zone allocator set up in tcp_init(). 502 */ 503 struct tcpcb * 504 tcp_newtcpcb(inp) 505 struct inpcb *inp; 506 { 507 struct inp_tp *it; 508 register struct tcpcb *tp; 509 #ifdef INET6 510 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 511 #endif /* INET6 */ 512 513 it = (struct inp_tp *)inp; 514 tp = &it->tcb; 515 bzero((char *) tp, sizeof(struct tcpcb)); 516 LIST_INIT(&tp->t_segq); 517 tp->t_maxseg = tp->t_maxopd = 518 #ifdef INET6 519 isipv6 ? tcp_v6mssdflt : 520 #endif /* INET6 */ 521 tcp_mssdflt; 522 523 /* Set up our timeouts. */ 524 callout_init(tp->tt_rexmt = &it->inp_tp_rexmt, 0); 525 callout_init(tp->tt_persist = &it->inp_tp_persist, 0); 526 callout_init(tp->tt_keep = &it->inp_tp_keep, 0); 527 callout_init(tp->tt_2msl = &it->inp_tp_2msl, 0); 528 callout_init(tp->tt_delack = &it->inp_tp_delack, 0); 529 530 if (tcp_do_rfc1323) 531 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 532 if (tcp_do_rfc1644) 533 tp->t_flags |= TF_REQ_CC; 534 tp->t_inpcb = inp; /* XXX */ 535 /* 536 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 537 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 538 * reasonable initial retransmit time. 539 */ 540 tp->t_srtt = TCPTV_SRTTBASE; 541 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 542 tp->t_rttmin = TCPTV_MIN; 543 tp->t_rxtcur = TCPTV_RTOBASE; 544 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 545 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 546 tp->t_rcvtime = ticks; 547 /* 548 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 549 * because the socket may be bound to an IPv6 wildcard address, 550 * which may match an IPv4-mapped IPv6 address. 551 */ 552 inp->inp_ip_ttl = ip_defttl; 553 inp->inp_ppcb = (caddr_t)tp; 554 return (tp); /* XXX */ 555 } 556 557 /* 558 * Drop a TCP connection, reporting 559 * the specified error. If connection is synchronized, 560 * then send a RST to peer. 561 */ 562 struct tcpcb * 563 tcp_drop(tp, errno) 564 register struct tcpcb *tp; 565 int errno; 566 { 567 struct socket *so = tp->t_inpcb->inp_socket; 568 569 if (TCPS_HAVERCVDSYN(tp->t_state)) { 570 tp->t_state = TCPS_CLOSED; 571 (void) tcp_output(tp); 572 tcpstat.tcps_drops++; 573 } else 574 tcpstat.tcps_conndrops++; 575 if (errno == ETIMEDOUT && tp->t_softerror) 576 errno = tp->t_softerror; 577 so->so_error = errno; 578 return (tcp_close(tp)); 579 } 580 581 /* 582 * Close a TCP control block: 583 * discard all space held by the tcp 584 * discard internet protocol block 585 * wake up any sleepers 586 */ 587 struct tcpcb * 588 tcp_close(tp) 589 register struct tcpcb *tp; 590 { 591 register struct tseg_qent *q; 592 struct inpcb *inp = tp->t_inpcb; 593 struct socket *so = inp->inp_socket; 594 #ifdef INET6 595 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 596 #endif /* INET6 */ 597 register struct rtentry *rt; 598 int dosavessthresh; 599 600 /* 601 * Make sure that all of our timers are stopped before we 602 * delete the PCB. 603 */ 604 callout_stop(tp->tt_rexmt); 605 callout_stop(tp->tt_persist); 606 callout_stop(tp->tt_keep); 607 callout_stop(tp->tt_2msl); 608 callout_stop(tp->tt_delack); 609 610 /* 611 * If we got enough samples through the srtt filter, 612 * save the rtt and rttvar in the routing entry. 613 * 'Enough' is arbitrarily defined as the 16 samples. 614 * 16 samples is enough for the srtt filter to converge 615 * to within 5% of the correct value; fewer samples and 616 * we could save a very bogus rtt. 617 * 618 * Don't update the default route's characteristics and don't 619 * update anything that the user "locked". 620 */ 621 if (tp->t_rttupdated >= 16) { 622 register u_long i = 0; 623 #ifdef INET6 624 if (isipv6) { 625 struct sockaddr_in6 *sin6; 626 627 if ((rt = inp->in6p_route.ro_rt) == NULL) 628 goto no_valid_rt; 629 sin6 = (struct sockaddr_in6 *)rt_key(rt); 630 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 631 goto no_valid_rt; 632 } 633 else 634 #endif /* INET6 */ 635 if ((rt = inp->inp_route.ro_rt) == NULL || 636 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr 637 == INADDR_ANY) 638 goto no_valid_rt; 639 640 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 641 i = tp->t_srtt * 642 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 643 if (rt->rt_rmx.rmx_rtt && i) 644 /* 645 * filter this update to half the old & half 646 * the new values, converting scale. 647 * See route.h and tcp_var.h for a 648 * description of the scaling constants. 649 */ 650 rt->rt_rmx.rmx_rtt = 651 (rt->rt_rmx.rmx_rtt + i) / 2; 652 else 653 rt->rt_rmx.rmx_rtt = i; 654 tcpstat.tcps_cachedrtt++; 655 } 656 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 657 i = tp->t_rttvar * 658 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 659 if (rt->rt_rmx.rmx_rttvar && i) 660 rt->rt_rmx.rmx_rttvar = 661 (rt->rt_rmx.rmx_rttvar + i) / 2; 662 else 663 rt->rt_rmx.rmx_rttvar = i; 664 tcpstat.tcps_cachedrttvar++; 665 } 666 /* 667 * The old comment here said: 668 * update the pipelimit (ssthresh) if it has been updated 669 * already or if a pipesize was specified & the threshhold 670 * got below half the pipesize. I.e., wait for bad news 671 * before we start updating, then update on both good 672 * and bad news. 673 * 674 * But we want to save the ssthresh even if no pipesize is 675 * specified explicitly in the route, because such 676 * connections still have an implicit pipesize specified 677 * by the global tcp_sendspace. In the absence of a reliable 678 * way to calculate the pipesize, it will have to do. 679 */ 680 i = tp->snd_ssthresh; 681 if (rt->rt_rmx.rmx_sendpipe != 0) 682 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2); 683 else 684 dosavessthresh = (i < so->so_snd.sb_hiwat / 2); 685 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 686 i != 0 && rt->rt_rmx.rmx_ssthresh != 0) 687 || dosavessthresh) { 688 /* 689 * convert the limit from user data bytes to 690 * packets then to packet data bytes. 691 */ 692 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 693 if (i < 2) 694 i = 2; 695 i *= (u_long)(tp->t_maxseg + 696 #ifdef INET6 697 (isipv6 ? sizeof (struct ip6_hdr) + 698 sizeof (struct tcphdr) : 699 #endif 700 sizeof (struct tcpiphdr) 701 #ifdef INET6 702 ) 703 #endif 704 ); 705 if (rt->rt_rmx.rmx_ssthresh) 706 rt->rt_rmx.rmx_ssthresh = 707 (rt->rt_rmx.rmx_ssthresh + i) / 2; 708 else 709 rt->rt_rmx.rmx_ssthresh = i; 710 tcpstat.tcps_cachedssthresh++; 711 } 712 } 713 rt = inp->inp_route.ro_rt; 714 if (rt) { 715 /* 716 * mark route for deletion if no information is 717 * cached. 718 */ 719 if ((tp->t_flags & TF_LQ_OVERFLOW) && 720 ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0)){ 721 if (rt->rt_rmx.rmx_rtt == 0) 722 rt->rt_flags |= RTF_DELCLONE; 723 } 724 } 725 no_valid_rt: 726 /* free the reassembly queue, if any */ 727 while((q = LIST_FIRST(&tp->t_segq)) != NULL) { 728 LIST_REMOVE(q, tqe_q); 729 m_freem(q->tqe_m); 730 FREE(q, M_TSEGQ); 731 } 732 inp->inp_ppcb = NULL; 733 soisdisconnected(so); 734 #ifdef INET6 735 if (INP_CHECK_SOCKAF(so, AF_INET6)) 736 in6_pcbdetach(inp); 737 else 738 #endif /* INET6 */ 739 in_pcbdetach(inp); 740 tcpstat.tcps_closed++; 741 return ((struct tcpcb *)0); 742 } 743 744 void 745 tcp_drain() 746 { 747 if (do_tcpdrain) 748 { 749 struct inpcb *inpb; 750 struct tcpcb *tcpb; 751 struct tseg_qent *te; 752 753 /* 754 * Walk the tcpbs, if existing, and flush the reassembly queue, 755 * if there is one... 756 * XXX: The "Net/3" implementation doesn't imply that the TCP 757 * reassembly queue should be flushed, but in a situation 758 * where we're really low on mbufs, this is potentially 759 * usefull. 760 */ 761 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 762 if ((tcpb = intotcpcb(inpb))) { 763 while ((te = LIST_FIRST(&tcpb->t_segq)) 764 != NULL) { 765 LIST_REMOVE(te, tqe_q); 766 m_freem(te->tqe_m); 767 FREE(te, M_TSEGQ); 768 } 769 } 770 } 771 } 772 } 773 774 /* 775 * Notify a tcp user of an asynchronous error; 776 * store error as soft error, but wake up user 777 * (for now, won't do anything until can select for soft error). 778 * 779 * Do not wake up user since there currently is no mechanism for 780 * reporting soft errors (yet - a kqueue filter may be added). 781 */ 782 static void 783 tcp_notify(inp, error) 784 struct inpcb *inp; 785 int error; 786 { 787 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 788 789 /* 790 * Ignore some errors if we are hooked up. 791 * If connection hasn't completed, has retransmitted several times, 792 * and receives a second error, give up now. This is better 793 * than waiting a long time to establish a connection that 794 * can never complete. 795 */ 796 if (tp->t_state == TCPS_ESTABLISHED && 797 (error == EHOSTUNREACH || error == ENETUNREACH || 798 error == EHOSTDOWN)) { 799 return; 800 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 801 tp->t_softerror) 802 tcp_drop(tp, error); 803 else 804 tp->t_softerror = error; 805 #if 0 806 wakeup((caddr_t) &so->so_timeo); 807 sorwakeup(so); 808 sowwakeup(so); 809 #endif 810 } 811 812 static int 813 tcp_pcblist(SYSCTL_HANDLER_ARGS) 814 { 815 int error, i, n, s; 816 struct inpcb *inp, **inp_list; 817 inp_gen_t gencnt; 818 struct xinpgen xig; 819 820 /* 821 * The process of preparing the TCB list is too time-consuming and 822 * resource-intensive to repeat twice on every request. 823 */ 824 if (req->oldptr == 0) { 825 n = tcbinfo.ipi_count; 826 req->oldidx = 2 * (sizeof xig) 827 + (n + n/8) * sizeof(struct xtcpcb); 828 return 0; 829 } 830 831 if (req->newptr != 0) 832 return EPERM; 833 834 /* 835 * OK, now we're committed to doing something. 836 */ 837 s = splnet(); 838 gencnt = tcbinfo.ipi_gencnt; 839 n = tcbinfo.ipi_count; 840 splx(s); 841 842 xig.xig_len = sizeof xig; 843 xig.xig_count = n; 844 xig.xig_gen = gencnt; 845 xig.xig_sogen = so_gencnt; 846 error = SYSCTL_OUT(req, &xig, sizeof xig); 847 if (error) 848 return error; 849 850 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 851 if (inp_list == 0) 852 return ENOMEM; 853 854 s = splnet(); 855 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n; 856 inp = LIST_NEXT(inp, inp_list)) { 857 if (inp->inp_gencnt <= gencnt) { 858 if (cr_cansee(req->p->p_ucred, 859 inp->inp_socket->so_cred)) 860 continue; 861 inp_list[i++] = inp; 862 } 863 } 864 splx(s); 865 n = i; 866 867 error = 0; 868 for (i = 0; i < n; i++) { 869 inp = inp_list[i]; 870 if (inp->inp_gencnt <= gencnt) { 871 struct xtcpcb xt; 872 caddr_t inp_ppcb; 873 xt.xt_len = sizeof xt; 874 /* XXX should avoid extra copy */ 875 bcopy(inp, &xt.xt_inp, sizeof *inp); 876 inp_ppcb = inp->inp_ppcb; 877 if (inp_ppcb != NULL) 878 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 879 else 880 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 881 if (inp->inp_socket) 882 sotoxsocket(inp->inp_socket, &xt.xt_socket); 883 error = SYSCTL_OUT(req, &xt, sizeof xt); 884 } 885 } 886 if (!error) { 887 /* 888 * Give the user an updated idea of our state. 889 * If the generation differs from what we told 890 * her before, she knows that something happened 891 * while we were processing this request, and it 892 * might be necessary to retry. 893 */ 894 s = splnet(); 895 xig.xig_gen = tcbinfo.ipi_gencnt; 896 xig.xig_sogen = so_gencnt; 897 xig.xig_count = tcbinfo.ipi_count; 898 splx(s); 899 error = SYSCTL_OUT(req, &xig, sizeof xig); 900 } 901 free(inp_list, M_TEMP); 902 return error; 903 } 904 905 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 906 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 907 908 static int 909 tcp_getcred(SYSCTL_HANDLER_ARGS) 910 { 911 struct xucred xuc; 912 struct sockaddr_in addrs[2]; 913 struct inpcb *inp; 914 int error, s; 915 916 error = suser_xxx(0, req->p, PRISON_ROOT); 917 if (error) 918 return (error); 919 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 920 if (error) 921 return (error); 922 s = splnet(); 923 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 924 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 925 if (inp == NULL || inp->inp_socket == NULL) { 926 error = ENOENT; 927 goto out; 928 } 929 error = cr_cansee(req->p->p_ucred, inp->inp_socket->so_cred); 930 if (error) 931 goto out; 932 bzero(&xuc, sizeof(xuc)); 933 xuc.cr_uid = inp->inp_socket->so_cred->cr_uid; 934 xuc.cr_ngroups = inp->inp_socket->so_cred->cr_ngroups; 935 bcopy(inp->inp_socket->so_cred->cr_groups, xuc.cr_groups, 936 sizeof(xuc.cr_groups)); 937 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 938 out: 939 splx(s); 940 return (error); 941 } 942 943 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 944 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 945 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 946 947 #ifdef INET6 948 static int 949 tcp6_getcred(SYSCTL_HANDLER_ARGS) 950 { 951 struct xucred xuc; 952 struct sockaddr_in6 addrs[2]; 953 struct inpcb *inp; 954 int error, s, mapped = 0; 955 956 error = suser_xxx(0, req->p, PRISON_ROOT); 957 if (error) 958 return (error); 959 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 960 if (error) 961 return (error); 962 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 963 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 964 mapped = 1; 965 else 966 return (EINVAL); 967 } 968 s = splnet(); 969 if (mapped == 1) 970 inp = in_pcblookup_hash(&tcbinfo, 971 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 972 addrs[1].sin6_port, 973 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 974 addrs[0].sin6_port, 975 0, NULL); 976 else 977 inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr, 978 addrs[1].sin6_port, 979 &addrs[0].sin6_addr, addrs[0].sin6_port, 980 0, NULL); 981 if (inp == NULL || inp->inp_socket == NULL) { 982 error = ENOENT; 983 goto out; 984 } 985 error = cr_cansee(req->p->p_ucred, inp->inp_socket->so_cred); 986 if (error) 987 goto out; 988 bzero(&xuc, sizeof(xuc)); 989 xuc.cr_uid = inp->inp_socket->so_cred->cr_uid; 990 xuc.cr_ngroups = inp->inp_socket->so_cred->cr_ngroups; 991 bcopy(inp->inp_socket->so_cred->cr_groups, xuc.cr_groups, 992 sizeof(xuc.cr_groups)); 993 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 994 out: 995 splx(s); 996 return (error); 997 } 998 999 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1000 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1001 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1002 #endif 1003 1004 1005 void 1006 tcp_ctlinput(cmd, sa, vip) 1007 int cmd; 1008 struct sockaddr *sa; 1009 void *vip; 1010 { 1011 struct ip *ip = vip; 1012 struct tcphdr *th; 1013 struct in_addr faddr; 1014 struct inpcb *inp; 1015 struct tcpcb *tp; 1016 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1017 tcp_seq icmp_seq; 1018 int s; 1019 1020 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1021 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1022 return; 1023 1024 if (cmd == PRC_QUENCH) 1025 notify = tcp_quench; 1026 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1027 cmd == PRC_UNREACH_PORT) && ip) 1028 notify = tcp_drop_syn_sent; 1029 else if (cmd == PRC_MSGSIZE) 1030 notify = tcp_mtudisc; 1031 else if (PRC_IS_REDIRECT(cmd)) { 1032 ip = 0; 1033 notify = in_rtchange; 1034 } else if (cmd == PRC_HOSTDEAD) 1035 ip = 0; 1036 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0) 1037 return; 1038 if (ip) { 1039 s = splnet(); 1040 th = (struct tcphdr *)((caddr_t)ip 1041 + (IP_VHL_HL(ip->ip_vhl) << 2)); 1042 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1043 ip->ip_src, th->th_sport, 0, NULL); 1044 if (inp != NULL && inp->inp_socket != NULL) { 1045 icmp_seq = htonl(th->th_seq); 1046 tp = intotcpcb(inp); 1047 if (SEQ_GEQ(icmp_seq, tp->snd_una) && 1048 SEQ_LT(icmp_seq, tp->snd_max)) 1049 (*notify)(inp, inetctlerrmap[cmd]); 1050 } 1051 splx(s); 1052 } else 1053 in_pcbnotifyall(&tcb, faddr, inetctlerrmap[cmd], notify); 1054 } 1055 1056 #ifdef INET6 1057 void 1058 tcp6_ctlinput(cmd, sa, d) 1059 int cmd; 1060 struct sockaddr *sa; 1061 void *d; 1062 { 1063 struct tcphdr th; 1064 void (*notify) __P((struct inpcb *, int)) = tcp_notify; 1065 struct ip6_hdr *ip6; 1066 struct mbuf *m; 1067 struct ip6ctlparam *ip6cp = NULL; 1068 const struct sockaddr_in6 *sa6_src = NULL; 1069 int off; 1070 struct tcp_portonly { 1071 u_int16_t th_sport; 1072 u_int16_t th_dport; 1073 } *thp; 1074 1075 if (sa->sa_family != AF_INET6 || 1076 sa->sa_len != sizeof(struct sockaddr_in6)) 1077 return; 1078 1079 if (cmd == PRC_QUENCH) 1080 notify = tcp_quench; 1081 else if (cmd == PRC_MSGSIZE) 1082 notify = tcp_mtudisc; 1083 else if (!PRC_IS_REDIRECT(cmd) && 1084 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1085 return; 1086 1087 /* if the parameter is from icmp6, decode it. */ 1088 if (d != NULL) { 1089 ip6cp = (struct ip6ctlparam *)d; 1090 m = ip6cp->ip6c_m; 1091 ip6 = ip6cp->ip6c_ip6; 1092 off = ip6cp->ip6c_off; 1093 sa6_src = ip6cp->ip6c_src; 1094 } else { 1095 m = NULL; 1096 ip6 = NULL; 1097 off = 0; /* fool gcc */ 1098 sa6_src = &sa6_any; 1099 } 1100 1101 if (ip6) { 1102 /* 1103 * XXX: We assume that when IPV6 is non NULL, 1104 * M and OFF are valid. 1105 */ 1106 1107 /* check if we can safely examine src and dst ports */ 1108 if (m->m_pkthdr.len < off + sizeof(*thp)) 1109 return; 1110 1111 bzero(&th, sizeof(th)); 1112 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1113 1114 in6_pcbnotify(&tcb, sa, th.th_dport, 1115 (struct sockaddr *)ip6cp->ip6c_src, 1116 th.th_sport, cmd, notify); 1117 } else 1118 in6_pcbnotify(&tcb, sa, 0, (struct sockaddr *)sa6_src, 1119 0, cmd, notify); 1120 } 1121 #endif /* INET6 */ 1122 1123 1124 /* 1125 * Following is where TCP initial sequence number generation occurs. 1126 * 1127 * There are two places where we must use initial sequence numbers: 1128 * 1. In SYN-ACK packets. 1129 * 2. In SYN packets. 1130 * 1131 * The ISNs in SYN-ACK packets have no monotonicity requirement, 1132 * and should be as unpredictable as possible to avoid the possibility 1133 * of spoofing and/or connection hijacking. To satisfy this 1134 * requirement, SYN-ACK ISNs are generated via the arc4random() 1135 * function. If exact RFC 1948 compliance is requested via sysctl, 1136 * these ISNs will be generated just like those in SYN packets. 1137 * 1138 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1139 * depends on this property. In addition, these ISNs should be 1140 * unguessable so as to prevent connection hijacking. To satisfy 1141 * the requirements of this situation, the algorithm outlined in 1142 * RFC 1948 is used to generate sequence numbers. 1143 * 1144 * For more information on the theory of operation, please see 1145 * RFC 1948. 1146 * 1147 * Implementation details: 1148 * 1149 * Time is based off the system timer, and is corrected so that it 1150 * increases by one megabyte per second. This allows for proper 1151 * recycling on high speed LANs while still leaving over an hour 1152 * before rollover. 1153 * 1154 * Two sysctls control the generation of ISNs: 1155 * 1156 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1157 * between seeding of isn_secret. This is normally set to zero, 1158 * as reseeding should not be necessary. 1159 * 1160 * net.inet.tcp.strict_rfc1948 controls whether RFC 1948 is followed 1161 * strictly. When strict compliance is requested, reseeding is 1162 * disabled and SYN-ACKs will be generated in the same manner as 1163 * SYNs. Strict mode is disabled by default. 1164 * 1165 */ 1166 1167 #define ISN_BYTES_PER_SECOND 1048576 1168 1169 u_char isn_secret[32]; 1170 int isn_last_reseed; 1171 MD5_CTX isn_ctx; 1172 1173 tcp_seq 1174 tcp_new_isn(tp) 1175 struct tcpcb *tp; 1176 { 1177 u_int32_t md5_buffer[4]; 1178 tcp_seq new_isn; 1179 1180 /* Use arc4random for SYN-ACKs when not in exact RFC1948 mode. */ 1181 if (((tp->t_state == TCPS_LISTEN) || (tp->t_state == TCPS_TIME_WAIT)) 1182 && tcp_strict_rfc1948 == 0) 1183 return arc4random(); 1184 1185 /* Seed if this is the first use, reseed if requested. */ 1186 if ((isn_last_reseed == 0) || 1187 ((tcp_strict_rfc1948 == 0) && (tcp_isn_reseed_interval > 0) && 1188 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1189 < (u_int)ticks))) { 1190 read_random(&isn_secret, sizeof(isn_secret)); 1191 isn_last_reseed = ticks; 1192 } 1193 1194 /* Compute the md5 hash and return the ISN. */ 1195 MD5Init(&isn_ctx); 1196 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1197 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1198 #ifdef INET6 1199 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1200 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1201 sizeof(struct in6_addr)); 1202 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1203 sizeof(struct in6_addr)); 1204 } else 1205 #endif 1206 { 1207 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1208 sizeof(struct in_addr)); 1209 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1210 sizeof(struct in_addr)); 1211 } 1212 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1213 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1214 new_isn = (tcp_seq) md5_buffer[0]; 1215 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz); 1216 return new_isn; 1217 } 1218 1219 /* 1220 * When a source quench is received, close congestion window 1221 * to one segment. We will gradually open it again as we proceed. 1222 */ 1223 void 1224 tcp_quench(inp, errno) 1225 struct inpcb *inp; 1226 int errno; 1227 { 1228 struct tcpcb *tp = intotcpcb(inp); 1229 1230 if (tp) 1231 tp->snd_cwnd = tp->t_maxseg; 1232 } 1233 1234 /* 1235 * When a specific ICMP unreachable message is received and the 1236 * connection state is SYN-SENT, drop the connection. This behavior 1237 * is controlled by the icmp_may_rst sysctl. 1238 */ 1239 void 1240 tcp_drop_syn_sent(inp, errno) 1241 struct inpcb *inp; 1242 int errno; 1243 { 1244 struct tcpcb *tp = intotcpcb(inp); 1245 1246 if (tp && tp->t_state == TCPS_SYN_SENT) 1247 tcp_drop(tp, errno); 1248 } 1249 1250 /* 1251 * When `need fragmentation' ICMP is received, update our idea of the MSS 1252 * based on the new value in the route. Also nudge TCP to send something, 1253 * since we know the packet we just sent was dropped. 1254 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1255 */ 1256 void 1257 tcp_mtudisc(inp, errno) 1258 struct inpcb *inp; 1259 int errno; 1260 { 1261 struct tcpcb *tp = intotcpcb(inp); 1262 struct rtentry *rt; 1263 struct rmxp_tao *taop; 1264 struct socket *so = inp->inp_socket; 1265 int offered; 1266 int mss; 1267 #ifdef INET6 1268 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1269 #endif /* INET6 */ 1270 1271 if (tp) { 1272 #ifdef INET6 1273 if (isipv6) 1274 rt = tcp_rtlookup6(inp); 1275 else 1276 #endif /* INET6 */ 1277 rt = tcp_rtlookup(inp); 1278 if (!rt || !rt->rt_rmx.rmx_mtu) { 1279 tp->t_maxopd = tp->t_maxseg = 1280 #ifdef INET6 1281 isipv6 ? tcp_v6mssdflt : 1282 #endif /* INET6 */ 1283 tcp_mssdflt; 1284 return; 1285 } 1286 taop = rmx_taop(rt->rt_rmx); 1287 offered = taop->tao_mssopt; 1288 mss = rt->rt_rmx.rmx_mtu - 1289 #ifdef INET6 1290 (isipv6 ? 1291 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1292 #endif /* INET6 */ 1293 sizeof(struct tcpiphdr) 1294 #ifdef INET6 1295 ) 1296 #endif /* INET6 */ 1297 ; 1298 1299 if (offered) 1300 mss = min(mss, offered); 1301 /* 1302 * XXX - The above conditional probably violates the TCP 1303 * spec. The problem is that, since we don't know the 1304 * other end's MSS, we are supposed to use a conservative 1305 * default. But, if we do that, then MTU discovery will 1306 * never actually take place, because the conservative 1307 * default is much less than the MTUs typically seen 1308 * on the Internet today. For the moment, we'll sweep 1309 * this under the carpet. 1310 * 1311 * The conservative default might not actually be a problem 1312 * if the only case this occurs is when sending an initial 1313 * SYN with options and data to a host we've never talked 1314 * to before. Then, they will reply with an MSS value which 1315 * will get recorded and the new parameters should get 1316 * recomputed. For Further Study. 1317 */ 1318 if (tp->t_maxopd <= mss) 1319 return; 1320 tp->t_maxopd = mss; 1321 1322 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1323 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1324 mss -= TCPOLEN_TSTAMP_APPA; 1325 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 1326 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 1327 mss -= TCPOLEN_CC_APPA; 1328 #if (MCLBYTES & (MCLBYTES - 1)) == 0 1329 if (mss > MCLBYTES) 1330 mss &= ~(MCLBYTES-1); 1331 #else 1332 if (mss > MCLBYTES) 1333 mss = mss / MCLBYTES * MCLBYTES; 1334 #endif 1335 if (so->so_snd.sb_hiwat < mss) 1336 mss = so->so_snd.sb_hiwat; 1337 1338 tp->t_maxseg = mss; 1339 1340 tcpstat.tcps_mturesent++; 1341 tp->t_rtttime = 0; 1342 tp->snd_nxt = tp->snd_una; 1343 tcp_output(tp); 1344 } 1345 } 1346 1347 /* 1348 * Look-up the routing entry to the peer of this inpcb. If no route 1349 * is found and it cannot be allocated the return NULL. This routine 1350 * is called by TCP routines that access the rmx structure and by tcp_mss 1351 * to get the interface MTU. 1352 */ 1353 struct rtentry * 1354 tcp_rtlookup(inp) 1355 struct inpcb *inp; 1356 { 1357 struct route *ro; 1358 struct rtentry *rt; 1359 1360 ro = &inp->inp_route; 1361 rt = ro->ro_rt; 1362 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1363 /* No route yet, so try to acquire one */ 1364 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1365 ro->ro_dst.sa_family = AF_INET; 1366 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1367 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1368 inp->inp_faddr; 1369 rtalloc(ro); 1370 rt = ro->ro_rt; 1371 } 1372 } 1373 return rt; 1374 } 1375 1376 #ifdef INET6 1377 struct rtentry * 1378 tcp_rtlookup6(inp) 1379 struct inpcb *inp; 1380 { 1381 struct route_in6 *ro6; 1382 struct rtentry *rt; 1383 1384 ro6 = &inp->in6p_route; 1385 rt = ro6->ro_rt; 1386 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1387 /* No route yet, so try to acquire one */ 1388 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) { 1389 struct sockaddr_in6 *dst6; 1390 1391 dst6 = (struct sockaddr_in6 *)&ro6->ro_dst; 1392 dst6->sin6_family = AF_INET6; 1393 dst6->sin6_len = sizeof(*dst6); 1394 dst6->sin6_addr = inp->in6p_faddr; 1395 rtalloc((struct route *)ro6); 1396 rt = ro6->ro_rt; 1397 } 1398 } 1399 return rt; 1400 } 1401 #endif /* INET6 */ 1402 1403 #ifdef IPSEC 1404 /* compute ESP/AH header size for TCP, including outer IP header. */ 1405 size_t 1406 ipsec_hdrsiz_tcp(tp) 1407 struct tcpcb *tp; 1408 { 1409 struct inpcb *inp; 1410 struct mbuf *m; 1411 size_t hdrsiz; 1412 struct ip *ip; 1413 #ifdef INET6 1414 struct ip6_hdr *ip6; 1415 #endif /* INET6 */ 1416 struct tcphdr *th; 1417 1418 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1419 return 0; 1420 MGETHDR(m, M_DONTWAIT, MT_DATA); 1421 if (!m) 1422 return 0; 1423 1424 #ifdef INET6 1425 if ((inp->inp_vflag & INP_IPV6) != 0) { 1426 ip6 = mtod(m, struct ip6_hdr *); 1427 th = (struct tcphdr *)(ip6 + 1); 1428 m->m_pkthdr.len = m->m_len = 1429 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1430 tcp_fillheaders(tp, ip6, th); 1431 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1432 } else 1433 #endif /* INET6 */ 1434 { 1435 ip = mtod(m, struct ip *); 1436 th = (struct tcphdr *)(ip + 1); 1437 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1438 tcp_fillheaders(tp, ip, th); 1439 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1440 } 1441 1442 m_free(m); 1443 return hdrsiz; 1444 } 1445 #endif /*IPSEC*/ 1446 1447 /* 1448 * Return a pointer to the cached information about the remote host. 1449 * The cached information is stored in the protocol specific part of 1450 * the route metrics. 1451 */ 1452 struct rmxp_tao * 1453 tcp_gettaocache(inp) 1454 struct inpcb *inp; 1455 { 1456 struct rtentry *rt; 1457 1458 #ifdef INET6 1459 if ((inp->inp_vflag & INP_IPV6) != 0) 1460 rt = tcp_rtlookup6(inp); 1461 else 1462 #endif /* INET6 */ 1463 rt = tcp_rtlookup(inp); 1464 1465 /* Make sure this is a host route and is up. */ 1466 if (rt == NULL || 1467 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) 1468 return NULL; 1469 1470 return rmx_taop(rt->rt_rmx); 1471 } 1472 1473 /* 1474 * Clear all the TAO cache entries, called from tcp_init. 1475 * 1476 * XXX 1477 * This routine is just an empty one, because we assume that the routing 1478 * routing tables are initialized at the same time when TCP, so there is 1479 * nothing in the cache left over. 1480 */ 1481 static void 1482 tcp_cleartaocache() 1483 { 1484 } 1485