tcp_timewait.c revision 104825
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: head/sys/netinet/tcp_timewait.c 104825 2002-10-10 21:41:30Z dillon $ 35 */ 36 37#include "opt_compat.h" 38#include "opt_inet6.h" 39#include "opt_ipsec.h" 40#include "opt_mac.h" 41#include "opt_tcpdebug.h" 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/callout.h> 46#include <sys/kernel.h> 47#include <sys/sysctl.h> 48#include <sys/mac.h> 49#include <sys/malloc.h> 50#include <sys/mbuf.h> 51#ifdef INET6 52#include <sys/domain.h> 53#endif 54#include <sys/proc.h> 55#include <sys/socket.h> 56#include <sys/socketvar.h> 57#include <sys/protosw.h> 58#include <sys/random.h> 59 60#include <vm/uma.h> 61 62#include <net/route.h> 63#include <net/if.h> 64 65#define _IP_VHL 66#include <netinet/in.h> 67#include <netinet/in_systm.h> 68#include <netinet/ip.h> 69#ifdef INET6 70#include <netinet/ip6.h> 71#endif 72#include <netinet/in_pcb.h> 73#ifdef INET6 74#include <netinet6/in6_pcb.h> 75#endif 76#include <netinet/in_var.h> 77#include <netinet/ip_var.h> 78#ifdef INET6 79#include <netinet6/ip6_var.h> 80#endif 81#include <netinet/tcp.h> 82#include <netinet/tcp_fsm.h> 83#include <netinet/tcp_seq.h> 84#include <netinet/tcp_timer.h> 85#include <netinet/tcp_var.h> 86#ifdef INET6 87#include <netinet6/tcp6_var.h> 88#endif 89#include <netinet/tcpip.h> 90#ifdef TCPDEBUG 91#include <netinet/tcp_debug.h> 92#endif 93#include <netinet6/ip6protosw.h> 94 95#ifdef IPSEC 96#include <netinet6/ipsec.h> 97#ifdef INET6 98#include <netinet6/ipsec6.h> 99#endif 100#endif /*IPSEC*/ 101 102#include <machine/in_cksum.h> 103#include <sys/md5.h> 104 105int tcp_mssdflt = TCP_MSS; 106SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 107 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 108 109#ifdef INET6 110int tcp_v6mssdflt = TCP6_MSS; 111SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 112 CTLFLAG_RW, &tcp_v6mssdflt , 0, 113 "Default TCP Maximum Segment Size for IPv6"); 114#endif 115 116#if 0 117static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 118SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 119 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 120#endif 121 122int tcp_do_rfc1323 = 1; 123SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 124 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 125 126int tcp_do_rfc1644 = 0; 127SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW, 128 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions"); 129 130static int tcp_tcbhashsize = 0; 131SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD, 132 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 133 134static int do_tcpdrain = 1; 135SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 136 "Enable tcp_drain routine for extra help when low on mbufs"); 137 138SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 139 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 140 141static int icmp_may_rst = 1; 142SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 143 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 144 145static int tcp_isn_reseed_interval = 0; 146SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 147 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 148 149/* 150 * TCP bandwidth limiting sysctls. Note that the default lower bound of 151 * 1024 exists only for debugging. A good production default would be 152 * something like 6100. 153 */ 154static int tcp_inflight_enable = 0; 155SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW, 156 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 157 158static int tcp_inflight_debug = 0; 159SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW, 160 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 161 162static int tcp_inflight_min = 1024; 163SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW, 164 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 165 166static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 167SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW, 168 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 169 170static void tcp_cleartaocache(void); 171static struct inpcb *tcp_notify(struct inpcb *, int); 172 173/* 174 * Target size of TCP PCB hash tables. Must be a power of two. 175 * 176 * Note that this can be overridden by the kernel environment 177 * variable net.inet.tcp.tcbhashsize 178 */ 179#ifndef TCBHASHSIZE 180#define TCBHASHSIZE 512 181#endif 182 183/* 184 * This is the actual shape of what we allocate using the zone 185 * allocator. Doing it this way allows us to protect both structures 186 * using the same generation count, and also eliminates the overhead 187 * of allocating tcpcbs separately. By hiding the structure here, 188 * we avoid changing most of the rest of the code (although it needs 189 * to be changed, eventually, for greater efficiency). 190 */ 191#define ALIGNMENT 32 192#define ALIGNM1 (ALIGNMENT - 1) 193struct inp_tp { 194 union { 195 struct inpcb inp; 196 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1]; 197 } inp_tp_u; 198 struct tcpcb tcb; 199 struct callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl; 200 struct callout inp_tp_delack; 201}; 202#undef ALIGNMENT 203#undef ALIGNM1 204 205/* 206 * Tcp initialization 207 */ 208void 209tcp_init() 210{ 211 int hashsize = TCBHASHSIZE; 212 213 tcp_ccgen = 1; 214 tcp_cleartaocache(); 215 216 tcp_delacktime = TCPTV_DELACK; 217 tcp_keepinit = TCPTV_KEEP_INIT; 218 tcp_keepidle = TCPTV_KEEP_IDLE; 219 tcp_keepintvl = TCPTV_KEEPINTVL; 220 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 221 tcp_msl = TCPTV_MSL; 222 tcp_rexmit_min = TCPTV_MIN; 223 tcp_rexmit_slop = TCPTV_CPU_VAR; 224 225 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 226 LIST_INIT(&tcb); 227 tcbinfo.listhead = &tcb; 228 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 229 if (!powerof2(hashsize)) { 230 printf("WARNING: TCB hash size not a power of 2\n"); 231 hashsize = 512; /* safe default */ 232 } 233 tcp_tcbhashsize = hashsize; 234 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 235 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 236 &tcbinfo.porthashmask); 237 tcbinfo.ipi_zone = uma_zcreate("tcpcb", sizeof(struct inp_tp), 238 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 239 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 240#ifdef INET6 241#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 242#else /* INET6 */ 243#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 244#endif /* INET6 */ 245 if (max_protohdr < TCP_MINPROTOHDR) 246 max_protohdr = TCP_MINPROTOHDR; 247 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 248 panic("tcp_init"); 249#undef TCP_MINPROTOHDR 250 251 syncache_init(); 252} 253 254/* 255 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 256 * tcp_template used to store this data in mbufs, but we now recopy it out 257 * of the tcpcb each time to conserve mbufs. 258 */ 259void 260tcp_fillheaders(tp, ip_ptr, tcp_ptr) 261 struct tcpcb *tp; 262 void *ip_ptr; 263 void *tcp_ptr; 264{ 265 struct inpcb *inp = tp->t_inpcb; 266 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr; 267 268#ifdef INET6 269 if ((inp->inp_vflag & INP_IPV6) != 0) { 270 struct ip6_hdr *ip6; 271 272 ip6 = (struct ip6_hdr *)ip_ptr; 273 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 274 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 275 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 276 (IPV6_VERSION & IPV6_VERSION_MASK); 277 ip6->ip6_nxt = IPPROTO_TCP; 278 ip6->ip6_plen = sizeof(struct tcphdr); 279 ip6->ip6_src = inp->in6p_laddr; 280 ip6->ip6_dst = inp->in6p_faddr; 281 tcp_hdr->th_sum = 0; 282 } else 283#endif 284 { 285 struct ip *ip = (struct ip *) ip_ptr; 286 287 ip->ip_vhl = IP_VHL_BORING; 288 ip->ip_tos = 0; 289 ip->ip_len = 0; 290 ip->ip_id = 0; 291 ip->ip_off = 0; 292 ip->ip_ttl = 0; 293 ip->ip_sum = 0; 294 ip->ip_p = IPPROTO_TCP; 295 ip->ip_src = inp->inp_laddr; 296 ip->ip_dst = inp->inp_faddr; 297 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 298 htons(sizeof(struct tcphdr) + IPPROTO_TCP)); 299 } 300 301 tcp_hdr->th_sport = inp->inp_lport; 302 tcp_hdr->th_dport = inp->inp_fport; 303 tcp_hdr->th_seq = 0; 304 tcp_hdr->th_ack = 0; 305 tcp_hdr->th_x2 = 0; 306 tcp_hdr->th_off = 5; 307 tcp_hdr->th_flags = 0; 308 tcp_hdr->th_win = 0; 309 tcp_hdr->th_urp = 0; 310} 311 312/* 313 * Create template to be used to send tcp packets on a connection. 314 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 315 * use for this function is in keepalives, which use tcp_respond. 316 */ 317struct tcptemp * 318tcp_maketemplate(tp) 319 struct tcpcb *tp; 320{ 321 struct mbuf *m; 322 struct tcptemp *n; 323 324 m = m_get(M_DONTWAIT, MT_HEADER); 325 if (m == NULL) 326 return (0); 327 m->m_len = sizeof(struct tcptemp); 328 n = mtod(m, struct tcptemp *); 329 330 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 331 return (n); 332} 333 334/* 335 * Send a single message to the TCP at address specified by 336 * the given TCP/IP header. If m == 0, then we make a copy 337 * of the tcpiphdr at ti and send directly to the addressed host. 338 * This is used to force keep alive messages out using the TCP 339 * template for a connection. If flags are given then we send 340 * a message back to the TCP which originated the * segment ti, 341 * and discard the mbuf containing it and any other attached mbufs. 342 * 343 * In any case the ack and sequence number of the transmitted 344 * segment are as specified by the parameters. 345 * 346 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 347 */ 348void 349tcp_respond(tp, ipgen, th, m, ack, seq, flags) 350 struct tcpcb *tp; 351 void *ipgen; 352 register struct tcphdr *th; 353 register struct mbuf *m; 354 tcp_seq ack, seq; 355 int flags; 356{ 357 register int tlen; 358 int win = 0; 359 struct route *ro = 0; 360 struct route sro; 361 struct ip *ip; 362 struct tcphdr *nth; 363#ifdef INET6 364 struct route_in6 *ro6 = 0; 365 struct route_in6 sro6; 366 struct ip6_hdr *ip6; 367 int isipv6; 368#endif /* INET6 */ 369 int ipflags = 0; 370 371 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 372 373#ifdef INET6 374 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6; 375 ip6 = ipgen; 376#endif /* INET6 */ 377 ip = ipgen; 378 379 if (tp) { 380 if (!(flags & TH_RST)) { 381 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv); 382 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 383 win = (long)TCP_MAXWIN << tp->rcv_scale; 384 } 385#ifdef INET6 386 if (isipv6) 387 ro6 = &tp->t_inpcb->in6p_route; 388 else 389#endif /* INET6 */ 390 ro = &tp->t_inpcb->inp_route; 391 } else { 392#ifdef INET6 393 if (isipv6) { 394 ro6 = &sro6; 395 bzero(ro6, sizeof *ro6); 396 } else 397#endif /* INET6 */ 398 { 399 ro = &sro; 400 bzero(ro, sizeof *ro); 401 } 402 } 403 if (m == 0) { 404 m = m_gethdr(M_DONTWAIT, MT_HEADER); 405 if (m == NULL) 406 return; 407 tlen = 0; 408 m->m_data += max_linkhdr; 409#ifdef INET6 410 if (isipv6) { 411 bcopy((caddr_t)ip6, mtod(m, caddr_t), 412 sizeof(struct ip6_hdr)); 413 ip6 = mtod(m, struct ip6_hdr *); 414 nth = (struct tcphdr *)(ip6 + 1); 415 } else 416#endif /* INET6 */ 417 { 418 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 419 ip = mtod(m, struct ip *); 420 nth = (struct tcphdr *)(ip + 1); 421 } 422 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 423 flags = TH_ACK; 424 } else { 425 m_freem(m->m_next); 426 m->m_next = 0; 427 m->m_data = (caddr_t)ipgen; 428 /* m_len is set later */ 429 tlen = 0; 430#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 431#ifdef INET6 432 if (isipv6) { 433 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 434 nth = (struct tcphdr *)(ip6 + 1); 435 } else 436#endif /* INET6 */ 437 { 438 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 439 nth = (struct tcphdr *)(ip + 1); 440 } 441 if (th != nth) { 442 /* 443 * this is usually a case when an extension header 444 * exists between the IPv6 header and the 445 * TCP header. 446 */ 447 nth->th_sport = th->th_sport; 448 nth->th_dport = th->th_dport; 449 } 450 xchg(nth->th_dport, nth->th_sport, n_short); 451#undef xchg 452 } 453#ifdef INET6 454 if (isipv6) { 455 ip6->ip6_flow = 0; 456 ip6->ip6_vfc = IPV6_VERSION; 457 ip6->ip6_nxt = IPPROTO_TCP; 458 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 459 tlen)); 460 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 461 } else 462#endif 463 { 464 tlen += sizeof (struct tcpiphdr); 465 ip->ip_len = tlen; 466 ip->ip_ttl = ip_defttl; 467 } 468 m->m_len = tlen; 469 m->m_pkthdr.len = tlen; 470 m->m_pkthdr.rcvif = (struct ifnet *) 0; 471#ifdef MAC 472 if (tp != NULL) { 473 /* 474 * Packet is associated with a socket, so allow the 475 * label of the response to reflect the socket label. 476 */ 477 mac_create_mbuf_from_socket(tp->t_inpcb->inp_socket, m); 478 } else { 479 /* 480 * XXXMAC: This will need to call a mac function that 481 * modifies the mbuf label in place for TCP datagrams 482 * not associated with a PCB. 483 */ 484 } 485#endif 486 nth->th_seq = htonl(seq); 487 nth->th_ack = htonl(ack); 488 nth->th_x2 = 0; 489 nth->th_off = sizeof (struct tcphdr) >> 2; 490 nth->th_flags = flags; 491 if (tp) 492 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 493 else 494 nth->th_win = htons((u_short)win); 495 nth->th_urp = 0; 496#ifdef INET6 497 if (isipv6) { 498 nth->th_sum = 0; 499 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 500 sizeof(struct ip6_hdr), 501 tlen - sizeof(struct ip6_hdr)); 502 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, 503 ro6 && ro6->ro_rt ? 504 ro6->ro_rt->rt_ifp : 505 NULL); 506 } else 507#endif /* INET6 */ 508 { 509 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 510 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 511 m->m_pkthdr.csum_flags = CSUM_TCP; 512 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 513 } 514#ifdef TCPDEBUG 515 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 516 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 517#endif 518#ifdef IPSEC 519 if (ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) { 520 m_freem(m); 521 return; 522 } 523#endif 524#ifdef INET6 525 if (isipv6) { 526 (void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL); 527 if (ro6 == &sro6 && ro6->ro_rt) { 528 RTFREE(ro6->ro_rt); 529 ro6->ro_rt = NULL; 530 } 531 } else 532#endif /* INET6 */ 533 { 534 (void) ip_output(m, NULL, ro, ipflags, NULL); 535 if (ro == &sro && ro->ro_rt) { 536 RTFREE(ro->ro_rt); 537 ro->ro_rt = NULL; 538 } 539 } 540} 541 542/* 543 * Create a new TCP control block, making an 544 * empty reassembly queue and hooking it to the argument 545 * protocol control block. The `inp' parameter must have 546 * come from the zone allocator set up in tcp_init(). 547 */ 548struct tcpcb * 549tcp_newtcpcb(inp) 550 struct inpcb *inp; 551{ 552 struct inp_tp *it; 553 register struct tcpcb *tp; 554#ifdef INET6 555 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 556#endif /* INET6 */ 557 558 it = (struct inp_tp *)inp; 559 tp = &it->tcb; 560 bzero((char *) tp, sizeof(struct tcpcb)); 561 LIST_INIT(&tp->t_segq); 562 tp->t_maxseg = tp->t_maxopd = 563#ifdef INET6 564 isipv6 ? tcp_v6mssdflt : 565#endif /* INET6 */ 566 tcp_mssdflt; 567 568 /* Set up our timeouts. */ 569 callout_init(tp->tt_rexmt = &it->inp_tp_rexmt, 0); 570 callout_init(tp->tt_persist = &it->inp_tp_persist, 0); 571 callout_init(tp->tt_keep = &it->inp_tp_keep, 0); 572 callout_init(tp->tt_2msl = &it->inp_tp_2msl, 0); 573 callout_init(tp->tt_delack = &it->inp_tp_delack, 0); 574 575 if (tcp_do_rfc1323) 576 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 577 if (tcp_do_rfc1644) 578 tp->t_flags |= TF_REQ_CC; 579 tp->t_inpcb = inp; /* XXX */ 580 /* 581 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 582 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 583 * reasonable initial retransmit time. 584 */ 585 tp->t_srtt = TCPTV_SRTTBASE; 586 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 587 tp->t_rttmin = tcp_rexmit_min; 588 tp->t_rxtcur = TCPTV_RTOBASE; 589 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 590 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 591 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 592 tp->t_rcvtime = ticks; 593 tp->t_bw_rtttime = ticks; 594 /* 595 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 596 * because the socket may be bound to an IPv6 wildcard address, 597 * which may match an IPv4-mapped IPv6 address. 598 */ 599 inp->inp_ip_ttl = ip_defttl; 600 inp->inp_ppcb = (caddr_t)tp; 601 return (tp); /* XXX */ 602} 603 604/* 605 * Drop a TCP connection, reporting 606 * the specified error. If connection is synchronized, 607 * then send a RST to peer. 608 */ 609struct tcpcb * 610tcp_drop(tp, errno) 611 register struct tcpcb *tp; 612 int errno; 613{ 614 struct socket *so = tp->t_inpcb->inp_socket; 615 616 if (TCPS_HAVERCVDSYN(tp->t_state)) { 617 tp->t_state = TCPS_CLOSED; 618 (void) tcp_output(tp); 619 tcpstat.tcps_drops++; 620 } else 621 tcpstat.tcps_conndrops++; 622 if (errno == ETIMEDOUT && tp->t_softerror) 623 errno = tp->t_softerror; 624 so->so_error = errno; 625 return (tcp_close(tp)); 626} 627 628/* 629 * Close a TCP control block: 630 * discard all space held by the tcp 631 * discard internet protocol block 632 * wake up any sleepers 633 */ 634struct tcpcb * 635tcp_close(tp) 636 register struct tcpcb *tp; 637{ 638 register struct tseg_qent *q; 639 struct inpcb *inp = tp->t_inpcb; 640 struct socket *so = inp->inp_socket; 641#ifdef INET6 642 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 643#endif /* INET6 */ 644 register struct rtentry *rt; 645 int dosavessthresh; 646 647 /* 648 * Make sure that all of our timers are stopped before we 649 * delete the PCB. 650 */ 651 callout_stop(tp->tt_rexmt); 652 callout_stop(tp->tt_persist); 653 callout_stop(tp->tt_keep); 654 callout_stop(tp->tt_2msl); 655 callout_stop(tp->tt_delack); 656 657 /* 658 * If we got enough samples through the srtt filter, 659 * save the rtt and rttvar in the routing entry. 660 * 'Enough' is arbitrarily defined as the 16 samples. 661 * 16 samples is enough for the srtt filter to converge 662 * to within 5% of the correct value; fewer samples and 663 * we could save a very bogus rtt. 664 * 665 * Don't update the default route's characteristics and don't 666 * update anything that the user "locked". 667 */ 668 if (tp->t_rttupdated >= 16) { 669 register u_long i = 0; 670#ifdef INET6 671 if (isipv6) { 672 struct sockaddr_in6 *sin6; 673 674 if ((rt = inp->in6p_route.ro_rt) == NULL) 675 goto no_valid_rt; 676 sin6 = (struct sockaddr_in6 *)rt_key(rt); 677 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 678 goto no_valid_rt; 679 } 680 else 681#endif /* INET6 */ 682 if ((rt = inp->inp_route.ro_rt) == NULL || 683 ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr 684 == INADDR_ANY) 685 goto no_valid_rt; 686 687 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) { 688 i = tp->t_srtt * 689 (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 690 if (rt->rt_rmx.rmx_rtt && i) 691 /* 692 * filter this update to half the old & half 693 * the new values, converting scale. 694 * See route.h and tcp_var.h for a 695 * description of the scaling constants. 696 */ 697 rt->rt_rmx.rmx_rtt = 698 (rt->rt_rmx.rmx_rtt + i) / 2; 699 else 700 rt->rt_rmx.rmx_rtt = i; 701 tcpstat.tcps_cachedrtt++; 702 } 703 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) { 704 i = tp->t_rttvar * 705 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 706 if (rt->rt_rmx.rmx_rttvar && i) 707 rt->rt_rmx.rmx_rttvar = 708 (rt->rt_rmx.rmx_rttvar + i) / 2; 709 else 710 rt->rt_rmx.rmx_rttvar = i; 711 tcpstat.tcps_cachedrttvar++; 712 } 713 /* 714 * The old comment here said: 715 * update the pipelimit (ssthresh) if it has been updated 716 * already or if a pipesize was specified & the threshhold 717 * got below half the pipesize. I.e., wait for bad news 718 * before we start updating, then update on both good 719 * and bad news. 720 * 721 * But we want to save the ssthresh even if no pipesize is 722 * specified explicitly in the route, because such 723 * connections still have an implicit pipesize specified 724 * by the global tcp_sendspace. In the absence of a reliable 725 * way to calculate the pipesize, it will have to do. 726 */ 727 i = tp->snd_ssthresh; 728 if (rt->rt_rmx.rmx_sendpipe != 0) 729 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2); 730 else 731 dosavessthresh = (i < so->so_snd.sb_hiwat / 2); 732 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 && 733 i != 0 && rt->rt_rmx.rmx_ssthresh != 0) 734 || dosavessthresh) { 735 /* 736 * convert the limit from user data bytes to 737 * packets then to packet data bytes. 738 */ 739 i = (i + tp->t_maxseg / 2) / tp->t_maxseg; 740 if (i < 2) 741 i = 2; 742 i *= (u_long)(tp->t_maxseg + 743#ifdef INET6 744 (isipv6 ? sizeof (struct ip6_hdr) + 745 sizeof (struct tcphdr) : 746#endif 747 sizeof (struct tcpiphdr) 748#ifdef INET6 749 ) 750#endif 751 ); 752 if (rt->rt_rmx.rmx_ssthresh) 753 rt->rt_rmx.rmx_ssthresh = 754 (rt->rt_rmx.rmx_ssthresh + i) / 2; 755 else 756 rt->rt_rmx.rmx_ssthresh = i; 757 tcpstat.tcps_cachedssthresh++; 758 } 759 } 760 no_valid_rt: 761 /* free the reassembly queue, if any */ 762 while((q = LIST_FIRST(&tp->t_segq)) != NULL) { 763 LIST_REMOVE(q, tqe_q); 764 m_freem(q->tqe_m); 765 FREE(q, M_TSEGQ); 766 } 767 inp->inp_ppcb = NULL; 768 soisdisconnected(so); 769#ifdef INET6 770 if (INP_CHECK_SOCKAF(so, AF_INET6)) 771 in6_pcbdetach(inp); 772 else 773#endif /* INET6 */ 774 in_pcbdetach(inp); 775 tcpstat.tcps_closed++; 776 return ((struct tcpcb *)0); 777} 778 779void 780tcp_drain() 781{ 782 if (do_tcpdrain) 783 { 784 struct inpcb *inpb; 785 struct tcpcb *tcpb; 786 struct tseg_qent *te; 787 788 /* 789 * Walk the tcpbs, if existing, and flush the reassembly queue, 790 * if there is one... 791 * XXX: The "Net/3" implementation doesn't imply that the TCP 792 * reassembly queue should be flushed, but in a situation 793 * where we're really low on mbufs, this is potentially 794 * usefull. 795 */ 796 INP_INFO_RLOCK(&tcbinfo); 797 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 798 INP_LOCK(inpb); 799 if ((tcpb = intotcpcb(inpb))) { 800 while ((te = LIST_FIRST(&tcpb->t_segq)) 801 != NULL) { 802 LIST_REMOVE(te, tqe_q); 803 m_freem(te->tqe_m); 804 FREE(te, M_TSEGQ); 805 } 806 } 807 INP_UNLOCK(inpb); 808 } 809 INP_INFO_RUNLOCK(&tcbinfo); 810 } 811} 812 813/* 814 * Notify a tcp user of an asynchronous error; 815 * store error as soft error, but wake up user 816 * (for now, won't do anything until can select for soft error). 817 * 818 * Do not wake up user since there currently is no mechanism for 819 * reporting soft errors (yet - a kqueue filter may be added). 820 */ 821static struct inpcb * 822tcp_notify(inp, error) 823 struct inpcb *inp; 824 int error; 825{ 826 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 827 828 /* 829 * Ignore some errors if we are hooked up. 830 * If connection hasn't completed, has retransmitted several times, 831 * and receives a second error, give up now. This is better 832 * than waiting a long time to establish a connection that 833 * can never complete. 834 */ 835 if (tp->t_state == TCPS_ESTABLISHED && 836 (error == EHOSTUNREACH || error == ENETUNREACH || 837 error == EHOSTDOWN)) { 838 return inp; 839 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 840 tp->t_softerror) { 841 tcp_drop(tp, error); 842 return (struct inpcb *)0; 843 } else { 844 tp->t_softerror = error; 845 return inp; 846 } 847#if 0 848 wakeup((caddr_t) &so->so_timeo); 849 sorwakeup(so); 850 sowwakeup(so); 851#endif 852} 853 854static int 855tcp_pcblist(SYSCTL_HANDLER_ARGS) 856{ 857 int error, i, n, s; 858 struct inpcb *inp, **inp_list; 859 inp_gen_t gencnt; 860 struct xinpgen xig; 861 862 /* 863 * The process of preparing the TCB list is too time-consuming and 864 * resource-intensive to repeat twice on every request. 865 */ 866 if (req->oldptr == 0) { 867 n = tcbinfo.ipi_count; 868 req->oldidx = 2 * (sizeof xig) 869 + (n + n/8) * sizeof(struct xtcpcb); 870 return 0; 871 } 872 873 if (req->newptr != 0) 874 return EPERM; 875 876 /* 877 * OK, now we're committed to doing something. 878 */ 879 s = splnet(); 880 INP_INFO_RLOCK(&tcbinfo); 881 gencnt = tcbinfo.ipi_gencnt; 882 n = tcbinfo.ipi_count; 883 INP_INFO_RUNLOCK(&tcbinfo); 884 splx(s); 885 886 sysctl_wire_old_buffer(req, 2 * (sizeof xig) 887 + n * sizeof(struct xtcpcb)); 888 889 xig.xig_len = sizeof xig; 890 xig.xig_count = n; 891 xig.xig_gen = gencnt; 892 xig.xig_sogen = so_gencnt; 893 error = SYSCTL_OUT(req, &xig, sizeof xig); 894 if (error) 895 return error; 896 897 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 898 if (inp_list == 0) 899 return ENOMEM; 900 901 s = splnet(); 902 INP_INFO_RLOCK(&tcbinfo); 903 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n; 904 inp = LIST_NEXT(inp, inp_list)) { 905 INP_LOCK(inp); 906 if (inp->inp_gencnt <= gencnt && 907 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0) 908 inp_list[i++] = inp; 909 INP_UNLOCK(inp); 910 } 911 INP_INFO_RUNLOCK(&tcbinfo); 912 splx(s); 913 n = i; 914 915 error = 0; 916 for (i = 0; i < n; i++) { 917 inp = inp_list[i]; 918 INP_LOCK(inp); 919 if (inp->inp_gencnt <= gencnt) { 920 struct xtcpcb xt; 921 caddr_t inp_ppcb; 922 xt.xt_len = sizeof xt; 923 /* XXX should avoid extra copy */ 924 bcopy(inp, &xt.xt_inp, sizeof *inp); 925 inp_ppcb = inp->inp_ppcb; 926 if (inp_ppcb != NULL) 927 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 928 else 929 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 930 if (inp->inp_socket) 931 sotoxsocket(inp->inp_socket, &xt.xt_socket); 932 error = SYSCTL_OUT(req, &xt, sizeof xt); 933 } 934 INP_UNLOCK(inp); 935 } 936 if (!error) { 937 /* 938 * Give the user an updated idea of our state. 939 * If the generation differs from what we told 940 * her before, she knows that something happened 941 * while we were processing this request, and it 942 * might be necessary to retry. 943 */ 944 s = splnet(); 945 INP_INFO_RLOCK(&tcbinfo); 946 xig.xig_gen = tcbinfo.ipi_gencnt; 947 xig.xig_sogen = so_gencnt; 948 xig.xig_count = tcbinfo.ipi_count; 949 INP_INFO_RUNLOCK(&tcbinfo); 950 splx(s); 951 error = SYSCTL_OUT(req, &xig, sizeof xig); 952 } 953 free(inp_list, M_TEMP); 954 return error; 955} 956 957SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 958 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 959 960static int 961tcp_getcred(SYSCTL_HANDLER_ARGS) 962{ 963 struct xucred xuc; 964 struct sockaddr_in addrs[2]; 965 struct inpcb *inp; 966 int error, s; 967 968 error = suser_cred(req->td->td_ucred, PRISON_ROOT); 969 if (error) 970 return (error); 971 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 972 if (error) 973 return (error); 974 s = splnet(); 975 INP_INFO_RLOCK(&tcbinfo); 976 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 977 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 978 if (inp == NULL) { 979 error = ENOENT; 980 goto outunlocked; 981 } 982 INP_LOCK(inp); 983 if (inp->inp_socket == NULL) { 984 error = ENOENT; 985 goto out; 986 } 987 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 988 if (error) 989 goto out; 990 cru2x(inp->inp_socket->so_cred, &xuc); 991out: 992 INP_UNLOCK(inp); 993outunlocked: 994 INP_INFO_RUNLOCK(&tcbinfo); 995 splx(s); 996 if (error == 0) 997 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 998 return (error); 999} 1000 1001SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1002 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1003 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1004 1005#ifdef INET6 1006static int 1007tcp6_getcred(SYSCTL_HANDLER_ARGS) 1008{ 1009 struct xucred xuc; 1010 struct sockaddr_in6 addrs[2]; 1011 struct inpcb *inp; 1012 int error, s, mapped = 0; 1013 1014 error = suser_cred(req->td->td_ucred, PRISON_ROOT); 1015 if (error) 1016 return (error); 1017 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1018 if (error) 1019 return (error); 1020 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1021 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1022 mapped = 1; 1023 else 1024 return (EINVAL); 1025 } 1026 s = splnet(); 1027 INP_INFO_RLOCK(&tcbinfo); 1028 if (mapped == 1) 1029 inp = in_pcblookup_hash(&tcbinfo, 1030 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1031 addrs[1].sin6_port, 1032 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1033 addrs[0].sin6_port, 1034 0, NULL); 1035 else 1036 inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr, 1037 addrs[1].sin6_port, 1038 &addrs[0].sin6_addr, addrs[0].sin6_port, 1039 0, NULL); 1040 if (inp == NULL) { 1041 error = ENOENT; 1042 goto outunlocked; 1043 } 1044 INP_LOCK(inp); 1045 if (inp->inp_socket == NULL) { 1046 error = ENOENT; 1047 goto out; 1048 } 1049 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1050 if (error) 1051 goto out; 1052 cru2x(inp->inp_socket->so_cred, &xuc); 1053out: 1054 INP_UNLOCK(inp); 1055outunlocked: 1056 INP_INFO_RUNLOCK(&tcbinfo); 1057 splx(s); 1058 if (error == 0) 1059 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1060 return (error); 1061} 1062 1063SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1064 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1065 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1066#endif 1067 1068 1069void 1070tcp_ctlinput(cmd, sa, vip) 1071 int cmd; 1072 struct sockaddr *sa; 1073 void *vip; 1074{ 1075 struct ip *ip = vip; 1076 struct tcphdr *th; 1077 struct in_addr faddr; 1078 struct inpcb *inp; 1079 struct tcpcb *tp; 1080 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1081 tcp_seq icmp_seq; 1082 int s; 1083 1084 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1085 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1086 return; 1087 1088 if (cmd == PRC_QUENCH) 1089 notify = tcp_quench; 1090 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1091 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1092 notify = tcp_drop_syn_sent; 1093 else if (cmd == PRC_MSGSIZE) 1094 notify = tcp_mtudisc; 1095 else if (PRC_IS_REDIRECT(cmd)) { 1096 ip = 0; 1097 notify = in_rtchange; 1098 } else if (cmd == PRC_HOSTDEAD) 1099 ip = 0; 1100 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0) 1101 return; 1102 if (ip) { 1103 s = splnet(); 1104 th = (struct tcphdr *)((caddr_t)ip 1105 + (IP_VHL_HL(ip->ip_vhl) << 2)); 1106 INP_INFO_WLOCK(&tcbinfo); 1107 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1108 ip->ip_src, th->th_sport, 0, NULL); 1109 if (inp != NULL) { 1110 INP_LOCK(inp); 1111 if (inp->inp_socket != NULL) { 1112 icmp_seq = htonl(th->th_seq); 1113 tp = intotcpcb(inp); 1114 if (SEQ_GEQ(icmp_seq, tp->snd_una) && 1115 SEQ_LT(icmp_seq, tp->snd_max)) 1116 inp = (*notify)(inp, inetctlerrmap[cmd]); 1117 } 1118 if (inp) 1119 INP_UNLOCK(inp); 1120 } else { 1121 struct in_conninfo inc; 1122 1123 inc.inc_fport = th->th_dport; 1124 inc.inc_lport = th->th_sport; 1125 inc.inc_faddr = faddr; 1126 inc.inc_laddr = ip->ip_src; 1127#ifdef INET6 1128 inc.inc_isipv6 = 0; 1129#endif 1130 syncache_unreach(&inc, th); 1131 } 1132 INP_INFO_WUNLOCK(&tcbinfo); 1133 splx(s); 1134 } else 1135 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1136} 1137 1138#ifdef INET6 1139void 1140tcp6_ctlinput(cmd, sa, d) 1141 int cmd; 1142 struct sockaddr *sa; 1143 void *d; 1144{ 1145 struct tcphdr th; 1146 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1147 struct ip6_hdr *ip6; 1148 struct mbuf *m; 1149 struct ip6ctlparam *ip6cp = NULL; 1150 const struct sockaddr_in6 *sa6_src = NULL; 1151 int off; 1152 struct tcp_portonly { 1153 u_int16_t th_sport; 1154 u_int16_t th_dport; 1155 } *thp; 1156 1157 if (sa->sa_family != AF_INET6 || 1158 sa->sa_len != sizeof(struct sockaddr_in6)) 1159 return; 1160 1161 if (cmd == PRC_QUENCH) 1162 notify = tcp_quench; 1163 else if (cmd == PRC_MSGSIZE) 1164 notify = tcp_mtudisc; 1165 else if (!PRC_IS_REDIRECT(cmd) && 1166 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1167 return; 1168 1169 /* if the parameter is from icmp6, decode it. */ 1170 if (d != NULL) { 1171 ip6cp = (struct ip6ctlparam *)d; 1172 m = ip6cp->ip6c_m; 1173 ip6 = ip6cp->ip6c_ip6; 1174 off = ip6cp->ip6c_off; 1175 sa6_src = ip6cp->ip6c_src; 1176 } else { 1177 m = NULL; 1178 ip6 = NULL; 1179 off = 0; /* fool gcc */ 1180 sa6_src = &sa6_any; 1181 } 1182 1183 if (ip6) { 1184 struct in_conninfo inc; 1185 /* 1186 * XXX: We assume that when IPV6 is non NULL, 1187 * M and OFF are valid. 1188 */ 1189 1190 /* check if we can safely examine src and dst ports */ 1191 if (m->m_pkthdr.len < off + sizeof(*thp)) 1192 return; 1193 1194 bzero(&th, sizeof(th)); 1195 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1196 1197 in6_pcbnotify(&tcb, sa, th.th_dport, 1198 (struct sockaddr *)ip6cp->ip6c_src, 1199 th.th_sport, cmd, notify); 1200 1201 inc.inc_fport = th.th_dport; 1202 inc.inc_lport = th.th_sport; 1203 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1204 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1205 inc.inc_isipv6 = 1; 1206 syncache_unreach(&inc, &th); 1207 } else 1208 in6_pcbnotify(&tcb, sa, 0, (const struct sockaddr *)sa6_src, 1209 0, cmd, notify); 1210} 1211#endif /* INET6 */ 1212 1213 1214/* 1215 * Following is where TCP initial sequence number generation occurs. 1216 * 1217 * There are two places where we must use initial sequence numbers: 1218 * 1. In SYN-ACK packets. 1219 * 2. In SYN packets. 1220 * 1221 * All ISNs for SYN-ACK packets are generated by the syncache. See 1222 * tcp_syncache.c for details. 1223 * 1224 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1225 * depends on this property. In addition, these ISNs should be 1226 * unguessable so as to prevent connection hijacking. To satisfy 1227 * the requirements of this situation, the algorithm outlined in 1228 * RFC 1948 is used to generate sequence numbers. 1229 * 1230 * Implementation details: 1231 * 1232 * Time is based off the system timer, and is corrected so that it 1233 * increases by one megabyte per second. This allows for proper 1234 * recycling on high speed LANs while still leaving over an hour 1235 * before rollover. 1236 * 1237 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1238 * between seeding of isn_secret. This is normally set to zero, 1239 * as reseeding should not be necessary. 1240 * 1241 */ 1242 1243#define ISN_BYTES_PER_SECOND 1048576 1244 1245u_char isn_secret[32]; 1246int isn_last_reseed; 1247MD5_CTX isn_ctx; 1248 1249tcp_seq 1250tcp_new_isn(tp) 1251 struct tcpcb *tp; 1252{ 1253 u_int32_t md5_buffer[4]; 1254 tcp_seq new_isn; 1255 1256 /* Seed if this is the first use, reseed if requested. */ 1257 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1258 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1259 < (u_int)ticks))) { 1260 read_random(&isn_secret, sizeof(isn_secret)); 1261 isn_last_reseed = ticks; 1262 } 1263 1264 /* Compute the md5 hash and return the ISN. */ 1265 MD5Init(&isn_ctx); 1266 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1267 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1268#ifdef INET6 1269 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1270 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1271 sizeof(struct in6_addr)); 1272 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1273 sizeof(struct in6_addr)); 1274 } else 1275#endif 1276 { 1277 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1278 sizeof(struct in_addr)); 1279 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1280 sizeof(struct in_addr)); 1281 } 1282 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1283 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1284 new_isn = (tcp_seq) md5_buffer[0]; 1285 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz); 1286 return new_isn; 1287} 1288 1289/* 1290 * When a source quench is received, close congestion window 1291 * to one segment. We will gradually open it again as we proceed. 1292 */ 1293struct inpcb * 1294tcp_quench(inp, errno) 1295 struct inpcb *inp; 1296 int errno; 1297{ 1298 struct tcpcb *tp = intotcpcb(inp); 1299 1300 if (tp) 1301 tp->snd_cwnd = tp->t_maxseg; 1302 return (inp); 1303} 1304 1305/* 1306 * When a specific ICMP unreachable message is received and the 1307 * connection state is SYN-SENT, drop the connection. This behavior 1308 * is controlled by the icmp_may_rst sysctl. 1309 */ 1310struct inpcb * 1311tcp_drop_syn_sent(inp, errno) 1312 struct inpcb *inp; 1313 int errno; 1314{ 1315 struct tcpcb *tp = intotcpcb(inp); 1316 1317 if (tp && tp->t_state == TCPS_SYN_SENT) { 1318 tcp_drop(tp, errno); 1319 return (struct inpcb *)0; 1320 } 1321 return inp; 1322} 1323 1324/* 1325 * When `need fragmentation' ICMP is received, update our idea of the MSS 1326 * based on the new value in the route. Also nudge TCP to send something, 1327 * since we know the packet we just sent was dropped. 1328 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1329 */ 1330struct inpcb * 1331tcp_mtudisc(inp, errno) 1332 struct inpcb *inp; 1333 int errno; 1334{ 1335 struct tcpcb *tp = intotcpcb(inp); 1336 struct rtentry *rt; 1337 struct rmxp_tao *taop; 1338 struct socket *so = inp->inp_socket; 1339 int offered; 1340 int mss; 1341#ifdef INET6 1342 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1343#endif /* INET6 */ 1344 1345 if (tp) { 1346#ifdef INET6 1347 if (isipv6) 1348 rt = tcp_rtlookup6(&inp->inp_inc); 1349 else 1350#endif /* INET6 */ 1351 rt = tcp_rtlookup(&inp->inp_inc); 1352 if (!rt || !rt->rt_rmx.rmx_mtu) { 1353 tp->t_maxopd = tp->t_maxseg = 1354#ifdef INET6 1355 isipv6 ? tcp_v6mssdflt : 1356#endif /* INET6 */ 1357 tcp_mssdflt; 1358 return inp; 1359 } 1360 taop = rmx_taop(rt->rt_rmx); 1361 offered = taop->tao_mssopt; 1362 mss = rt->rt_rmx.rmx_mtu - 1363#ifdef INET6 1364 (isipv6 ? 1365 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1366#endif /* INET6 */ 1367 sizeof(struct tcpiphdr) 1368#ifdef INET6 1369 ) 1370#endif /* INET6 */ 1371 ; 1372 1373 if (offered) 1374 mss = min(mss, offered); 1375 /* 1376 * XXX - The above conditional probably violates the TCP 1377 * spec. The problem is that, since we don't know the 1378 * other end's MSS, we are supposed to use a conservative 1379 * default. But, if we do that, then MTU discovery will 1380 * never actually take place, because the conservative 1381 * default is much less than the MTUs typically seen 1382 * on the Internet today. For the moment, we'll sweep 1383 * this under the carpet. 1384 * 1385 * The conservative default might not actually be a problem 1386 * if the only case this occurs is when sending an initial 1387 * SYN with options and data to a host we've never talked 1388 * to before. Then, they will reply with an MSS value which 1389 * will get recorded and the new parameters should get 1390 * recomputed. For Further Study. 1391 */ 1392 if (tp->t_maxopd <= mss) 1393 return inp; 1394 tp->t_maxopd = mss; 1395 1396 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1397 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1398 mss -= TCPOLEN_TSTAMP_APPA; 1399 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 1400 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 1401 mss -= TCPOLEN_CC_APPA; 1402#if (MCLBYTES & (MCLBYTES - 1)) == 0 1403 if (mss > MCLBYTES) 1404 mss &= ~(MCLBYTES-1); 1405#else 1406 if (mss > MCLBYTES) 1407 mss = mss / MCLBYTES * MCLBYTES; 1408#endif 1409 if (so->so_snd.sb_hiwat < mss) 1410 mss = so->so_snd.sb_hiwat; 1411 1412 tp->t_maxseg = mss; 1413 1414 tcpstat.tcps_mturesent++; 1415 tp->t_rtttime = 0; 1416 tp->snd_nxt = tp->snd_una; 1417 tcp_output(tp); 1418 } 1419 return inp; 1420} 1421 1422/* 1423 * Look-up the routing entry to the peer of this inpcb. If no route 1424 * is found and it cannot be allocated the return NULL. This routine 1425 * is called by TCP routines that access the rmx structure and by tcp_mss 1426 * to get the interface MTU. 1427 */ 1428struct rtentry * 1429tcp_rtlookup(inc) 1430 struct in_conninfo *inc; 1431{ 1432 struct route *ro; 1433 struct rtentry *rt; 1434 1435 ro = &inc->inc_route; 1436 rt = ro->ro_rt; 1437 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1438 /* No route yet, so try to acquire one */ 1439 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1440 ro->ro_dst.sa_family = AF_INET; 1441 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1442 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 1443 inc->inc_faddr; 1444 rtalloc(ro); 1445 rt = ro->ro_rt; 1446 } 1447 } 1448 return rt; 1449} 1450 1451#ifdef INET6 1452struct rtentry * 1453tcp_rtlookup6(inc) 1454 struct in_conninfo *inc; 1455{ 1456 struct route_in6 *ro6; 1457 struct rtentry *rt; 1458 1459 ro6 = &inc->inc6_route; 1460 rt = ro6->ro_rt; 1461 if (rt == NULL || !(rt->rt_flags & RTF_UP)) { 1462 /* No route yet, so try to acquire one */ 1463 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1464 ro6->ro_dst.sin6_family = AF_INET6; 1465 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1466 ro6->ro_dst.sin6_addr = inc->inc6_faddr; 1467 rtalloc((struct route *)ro6); 1468 rt = ro6->ro_rt; 1469 } 1470 } 1471 return rt; 1472} 1473#endif /* INET6 */ 1474 1475#ifdef IPSEC 1476/* compute ESP/AH header size for TCP, including outer IP header. */ 1477size_t 1478ipsec_hdrsiz_tcp(tp) 1479 struct tcpcb *tp; 1480{ 1481 struct inpcb *inp; 1482 struct mbuf *m; 1483 size_t hdrsiz; 1484 struct ip *ip; 1485#ifdef INET6 1486 struct ip6_hdr *ip6; 1487#endif /* INET6 */ 1488 struct tcphdr *th; 1489 1490 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1491 return 0; 1492 MGETHDR(m, M_DONTWAIT, MT_DATA); 1493 if (!m) 1494 return 0; 1495 1496#ifdef INET6 1497 if ((inp->inp_vflag & INP_IPV6) != 0) { 1498 ip6 = mtod(m, struct ip6_hdr *); 1499 th = (struct tcphdr *)(ip6 + 1); 1500 m->m_pkthdr.len = m->m_len = 1501 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1502 tcp_fillheaders(tp, ip6, th); 1503 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1504 } else 1505#endif /* INET6 */ 1506 { 1507 ip = mtod(m, struct ip *); 1508 th = (struct tcphdr *)(ip + 1); 1509 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1510 tcp_fillheaders(tp, ip, th); 1511 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1512 } 1513 1514 m_free(m); 1515 return hdrsiz; 1516} 1517#endif /*IPSEC*/ 1518 1519/* 1520 * Return a pointer to the cached information about the remote host. 1521 * The cached information is stored in the protocol specific part of 1522 * the route metrics. 1523 */ 1524struct rmxp_tao * 1525tcp_gettaocache(inc) 1526 struct in_conninfo *inc; 1527{ 1528 struct rtentry *rt; 1529 1530#ifdef INET6 1531 if (inc->inc_isipv6) 1532 rt = tcp_rtlookup6(inc); 1533 else 1534#endif /* INET6 */ 1535 rt = tcp_rtlookup(inc); 1536 1537 /* Make sure this is a host route and is up. */ 1538 if (rt == NULL || 1539 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) 1540 return NULL; 1541 1542 return rmx_taop(rt->rt_rmx); 1543} 1544 1545/* 1546 * Clear all the TAO cache entries, called from tcp_init. 1547 * 1548 * XXX 1549 * This routine is just an empty one, because we assume that the routing 1550 * routing tables are initialized at the same time when TCP, so there is 1551 * nothing in the cache left over. 1552 */ 1553static void 1554tcp_cleartaocache() 1555{ 1556} 1557 1558/* 1559 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1560 * 1561 * This code attempts to calculate the bandwidth-delay product as a 1562 * means of determining the optimal window size to maximize bandwidth, 1563 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1564 * routers. This code also does a fairly good job keeping RTTs in check 1565 * across slow links like modems. We implement an algorithm which is very 1566 * similar (but not meant to be) TCP/Vegas. The code operates on the 1567 * transmitter side of a TCP connection and so only effects the transmit 1568 * side of the connection. 1569 * 1570 * BACKGROUND: TCP makes no provision for the management of buffer space 1571 * at the end points or at the intermediate routers and switches. A TCP 1572 * stream, whether using NewReno or not, will eventually buffer as 1573 * many packets as it is able and the only reason this typically works is 1574 * due to the fairly small default buffers made available for a connection 1575 * (typicaly 16K or 32K). As machines use larger windows and/or window 1576 * scaling it is now fairly easy for even a single TCP connection to blow-out 1577 * all available buffer space not only on the local interface, but on 1578 * intermediate routers and switches as well. NewReno makes a misguided 1579 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1580 * then backing off, then steadily increasing the window again until another 1581 * failure occurs, ad-infinitum. This results in terrible oscillation that 1582 * is only made worse as network loads increase and the idea of intentionally 1583 * blowing out network buffers is, frankly, a terrible way to manage network 1584 * resources. 1585 * 1586 * It is far better to limit the transmit window prior to the failure 1587 * condition being achieved. There are two general ways to do this: First 1588 * you can 'scan' through different transmit window sizes and locate the 1589 * point where the RTT stops increasing, indicating that you have filled the 1590 * pipe, then scan backwards until you note that RTT stops decreasing, then 1591 * repeat ad-infinitum. This method works in principle but has severe 1592 * implementation issues due to RTT variances, timer granularity, and 1593 * instability in the algorithm which can lead to many false positives and 1594 * create oscillations as well as interact badly with other TCP streams 1595 * implementing the same algorithm. 1596 * 1597 * The second method is to limit the window to the bandwidth delay product 1598 * of the link. This is the method we implement. RTT variances and our 1599 * own manipulation of the congestion window, bwnd, can potentially 1600 * destabilize the algorithm. For this reason we have to stabilize the 1601 * elements used to calculate the window. We do this by using the minimum 1602 * observed RTT, the long term average of the observed bandwidth, and 1603 * by adding two segments worth of slop. It isn't perfect but it is able 1604 * to react to changing conditions and gives us a very stable basis on 1605 * which to extend the algorithm. 1606 */ 1607void 1608tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1609{ 1610 u_long bw; 1611 u_long bwnd; 1612 int save_ticks; 1613 1614 /* 1615 * If inflight_enable is disabled in the middle of a tcp connection, 1616 * make sure snd_bwnd is effectively disabled. 1617 */ 1618 if (tcp_inflight_enable == 0) { 1619 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1620 tp->snd_bandwidth = 0; 1621 return; 1622 } 1623 1624 /* 1625 * Figure out the bandwidth. Due to the tick granularity this 1626 * is a very rough number and it MUST be averaged over a fairly 1627 * long period of time. XXX we need to take into account a link 1628 * that is not using all available bandwidth, but for now our 1629 * slop will ramp us up if this case occurs and the bandwidth later 1630 * increases. 1631 * 1632 * Note: if ticks rollover 'bw' may wind up negative. We must 1633 * effectively reset t_bw_rtttime for this case. 1634 */ 1635 save_ticks = ticks; 1636 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 1637 return; 1638 1639 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 1640 (save_ticks - tp->t_bw_rtttime); 1641 tp->t_bw_rtttime = save_ticks; 1642 tp->t_bw_rtseq = ack_seq; 1643 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 1644 return; 1645 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1646 1647 tp->snd_bandwidth = bw; 1648 1649 /* 1650 * Calculate the semi-static bandwidth delay product, plus two maximal 1651 * segments. The additional slop puts us squarely in the sweet 1652 * spot and also handles the bandwidth run-up case. Without the 1653 * slop we could be locking ourselves into a lower bandwidth. 1654 * 1655 * Situations Handled: 1656 * (1) Prevents over-queueing of packets on LANs, especially on 1657 * high speed LANs, allowing larger TCP buffers to be 1658 * specified, and also does a good job preventing 1659 * over-queueing of packets over choke points like modems 1660 * (at least for the transmit side). 1661 * 1662 * (2) Is able to handle changing network loads (bandwidth 1663 * drops so bwnd drops, bandwidth increases so bwnd 1664 * increases). 1665 * 1666 * (3) Theoretically should stabilize in the face of multiple 1667 * connections implementing the same algorithm (this may need 1668 * a little work). 1669 */ 1670#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1671 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + 2 * tp->t_maxseg; 1672#undef USERTT 1673 1674 if (tcp_inflight_debug > 0) { 1675 static int ltime; 1676 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 1677 ltime = ticks; 1678 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 1679 tp, 1680 bw, 1681 tp->t_rttbest, 1682 tp->t_srtt, 1683 bwnd 1684 ); 1685 } 1686 } 1687 if ((long)bwnd < tcp_inflight_min) 1688 bwnd = tcp_inflight_min; 1689 if (bwnd > tcp_inflight_max) 1690 bwnd = tcp_inflight_max; 1691 if ((long)bwnd < tp->t_maxseg * 2) 1692 bwnd = tp->t_maxseg * 2; 1693 tp->snd_bwnd = bwnd; 1694} 1695 1696