tcp_timewait.c revision 138020
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 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 * $FreeBSD: head/sys/netinet/tcp_timewait.c 138020 2004-11-23 16:23:13Z rwatson $ 31 */ 32 33#include "opt_compat.h" 34#include "opt_inet.h" 35#include "opt_inet6.h" 36#include "opt_ipsec.h" 37#include "opt_mac.h" 38#include "opt_tcpdebug.h" 39#include "opt_tcp_sack.h" 40 41#include <sys/param.h> 42#include <sys/systm.h> 43#include <sys/callout.h> 44#include <sys/kernel.h> 45#include <sys/sysctl.h> 46#include <sys/mac.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/uma.h> 59 60#include <net/route.h> 61#include <net/if.h> 62 63#include <netinet/in.h> 64#include <netinet/in_systm.h> 65#include <netinet/ip.h> 66#ifdef INET6 67#include <netinet/ip6.h> 68#endif 69#include <netinet/in_pcb.h> 70#ifdef INET6 71#include <netinet6/in6_pcb.h> 72#endif 73#include <netinet/in_var.h> 74#include <netinet/ip_var.h> 75#ifdef INET6 76#include <netinet6/ip6_var.h> 77#include <netinet6/nd6.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#include <netkey/key.h> 99#endif /*IPSEC*/ 100 101#ifdef FAST_IPSEC 102#include <netipsec/ipsec.h> 103#include <netipsec/xform.h> 104#ifdef INET6 105#include <netipsec/ipsec6.h> 106#endif 107#include <netipsec/key.h> 108#define IPSEC 109#endif /*FAST_IPSEC*/ 110 111#include <machine/in_cksum.h> 112#include <sys/md5.h> 113 114int tcp_mssdflt = TCP_MSS; 115SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 116 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 117 118#ifdef INET6 119int tcp_v6mssdflt = TCP6_MSS; 120SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 121 CTLFLAG_RW, &tcp_v6mssdflt , 0, 122 "Default TCP Maximum Segment Size for IPv6"); 123#endif 124 125/* 126 * Minimum MSS we accept and use. This prevents DoS attacks where 127 * we are forced to a ridiculous low MSS like 20 and send hundreds 128 * of packets instead of one. The effect scales with the available 129 * bandwidth and quickly saturates the CPU and network interface 130 * with packet generation and sending. Set to zero to disable MINMSS 131 * checking. This setting prevents us from sending too small packets. 132 */ 133int tcp_minmss = TCP_MINMSS; 134SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 135 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 136/* 137 * Number of TCP segments per second we accept from remote host 138 * before we start to calculate average segment size. If average 139 * segment size drops below the minimum TCP MSS we assume a DoS 140 * attack and reset+drop the connection. Care has to be taken not to 141 * set this value too small to not kill interactive type connections 142 * (telnet, SSH) which send many small packets. 143 */ 144int tcp_minmssoverload = TCP_MINMSSOVERLOAD; 145SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, 146 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" 147 "be under the MINMSS Size"); 148 149#if 0 150static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 151SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 152 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 153#endif 154 155int tcp_do_rfc1323 = 1; 156SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 157 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 158 159static int tcp_tcbhashsize = 0; 160SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 161 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 162 163static int do_tcpdrain = 1; 164SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 165 "Enable tcp_drain routine for extra help when low on mbufs"); 166 167SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 168 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 169 170static int icmp_may_rst = 1; 171SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 172 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 173 174static int tcp_isn_reseed_interval = 0; 175SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 176 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 177 178/* 179 * TCP bandwidth limiting sysctls. Note that the default lower bound of 180 * 1024 exists only for debugging. A good production default would be 181 * something like 6100. 182 */ 183SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 184 "TCP inflight data limiting"); 185 186static int tcp_inflight_enable = 1; 187SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 188 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 189 190static int tcp_inflight_debug = 0; 191SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 192 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 193 194static int tcp_inflight_min = 6144; 195SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 196 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 197 198static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 199SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 200 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 201 202static int tcp_inflight_stab = 20; 203SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 204 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 205 206uma_zone_t sack_hole_zone; 207 208static struct inpcb *tcp_notify(struct inpcb *, int); 209static void tcp_discardcb(struct tcpcb *); 210static void tcp_isn_tick(void *); 211 212/* 213 * Target size of TCP PCB hash tables. Must be a power of two. 214 * 215 * Note that this can be overridden by the kernel environment 216 * variable net.inet.tcp.tcbhashsize 217 */ 218#ifndef TCBHASHSIZE 219#define TCBHASHSIZE 512 220#endif 221 222/* 223 * XXX 224 * Callouts should be moved into struct tcp directly. They are currently 225 * separate because the tcpcb structure is exported to userland for sysctl 226 * parsing purposes, which do not know about callouts. 227 */ 228struct tcpcb_mem { 229 struct tcpcb tcb; 230 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 231 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 232}; 233 234static uma_zone_t tcpcb_zone; 235static uma_zone_t tcptw_zone; 236struct callout isn_callout; 237 238/* 239 * Tcp initialization 240 */ 241void 242tcp_init() 243{ 244 int hashsize = TCBHASHSIZE; 245 246 tcp_delacktime = TCPTV_DELACK; 247 tcp_keepinit = TCPTV_KEEP_INIT; 248 tcp_keepidle = TCPTV_KEEP_IDLE; 249 tcp_keepintvl = TCPTV_KEEPINTVL; 250 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 251 tcp_msl = TCPTV_MSL; 252 tcp_rexmit_min = TCPTV_MIN; 253 tcp_rexmit_slop = TCPTV_CPU_VAR; 254 255 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 256 LIST_INIT(&tcb); 257 tcbinfo.listhead = &tcb; 258 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 259 if (!powerof2(hashsize)) { 260 printf("WARNING: TCB hash size not a power of 2\n"); 261 hashsize = 512; /* safe default */ 262 } 263 tcp_tcbhashsize = hashsize; 264 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 265 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 266 &tcbinfo.porthashmask); 267 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 268 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 269 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 270#ifdef INET6 271#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 272#else /* INET6 */ 273#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 274#endif /* INET6 */ 275 if (max_protohdr < TCP_MINPROTOHDR) 276 max_protohdr = TCP_MINPROTOHDR; 277 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 278 panic("tcp_init"); 279#undef TCP_MINPROTOHDR 280 /* 281 * These have to be type stable for the benefit of the timers. 282 */ 283 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 284 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 285 uma_zone_set_max(tcpcb_zone, maxsockets); 286 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 287 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 288 uma_zone_set_max(tcptw_zone, maxsockets / 5); 289 tcp_timer_init(); 290 syncache_init(); 291 tcp_hc_init(); 292 tcp_reass_init(); 293 callout_init(&isn_callout, CALLOUT_MPSAFE); 294 tcp_isn_tick(NULL); 295 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 296 SHUTDOWN_PRI_DEFAULT); 297 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 298 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 299} 300 301void 302tcp_fini(xtp) 303 void *xtp; 304{ 305 callout_stop(&isn_callout); 306 307} 308 309/* 310 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 311 * tcp_template used to store this data in mbufs, but we now recopy it out 312 * of the tcpcb each time to conserve mbufs. 313 */ 314void 315tcpip_fillheaders(inp, ip_ptr, tcp_ptr) 316 struct inpcb *inp; 317 void *ip_ptr; 318 void *tcp_ptr; 319{ 320 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 321 322#ifdef INET6 323 if ((inp->inp_vflag & INP_IPV6) != 0) { 324 struct ip6_hdr *ip6; 325 326 ip6 = (struct ip6_hdr *)ip_ptr; 327 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 328 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 329 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 330 (IPV6_VERSION & IPV6_VERSION_MASK); 331 ip6->ip6_nxt = IPPROTO_TCP; 332 ip6->ip6_plen = sizeof(struct tcphdr); 333 ip6->ip6_src = inp->in6p_laddr; 334 ip6->ip6_dst = inp->in6p_faddr; 335 } else 336#endif 337 { 338 struct ip *ip; 339 340 ip = (struct ip *)ip_ptr; 341 ip->ip_v = IPVERSION; 342 ip->ip_hl = 5; 343 ip->ip_tos = inp->inp_ip_tos; 344 ip->ip_len = 0; 345 ip->ip_id = 0; 346 ip->ip_off = 0; 347 ip->ip_ttl = inp->inp_ip_ttl; 348 ip->ip_sum = 0; 349 ip->ip_p = IPPROTO_TCP; 350 ip->ip_src = inp->inp_laddr; 351 ip->ip_dst = inp->inp_faddr; 352 } 353 th->th_sport = inp->inp_lport; 354 th->th_dport = inp->inp_fport; 355 th->th_seq = 0; 356 th->th_ack = 0; 357 th->th_x2 = 0; 358 th->th_off = 5; 359 th->th_flags = 0; 360 th->th_win = 0; 361 th->th_urp = 0; 362 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 363} 364 365/* 366 * Create template to be used to send tcp packets on a connection. 367 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 368 * use for this function is in keepalives, which use tcp_respond. 369 */ 370struct tcptemp * 371tcpip_maketemplate(inp) 372 struct inpcb *inp; 373{ 374 struct mbuf *m; 375 struct tcptemp *n; 376 377 m = m_get(M_DONTWAIT, MT_HEADER); 378 if (m == NULL) 379 return (0); 380 m->m_len = sizeof(struct tcptemp); 381 n = mtod(m, struct tcptemp *); 382 383 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 384 return (n); 385} 386 387/* 388 * Send a single message to the TCP at address specified by 389 * the given TCP/IP header. If m == NULL, then we make a copy 390 * of the tcpiphdr at ti and send directly to the addressed host. 391 * This is used to force keep alive messages out using the TCP 392 * template for a connection. If flags are given then we send 393 * a message back to the TCP which originated the * segment ti, 394 * and discard the mbuf containing it and any other attached mbufs. 395 * 396 * In any case the ack and sequence number of the transmitted 397 * segment are as specified by the parameters. 398 * 399 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 400 */ 401void 402tcp_respond(tp, ipgen, th, m, ack, seq, flags) 403 struct tcpcb *tp; 404 void *ipgen; 405 register struct tcphdr *th; 406 register struct mbuf *m; 407 tcp_seq ack, seq; 408 int flags; 409{ 410 register int tlen; 411 int win = 0; 412 struct ip *ip; 413 struct tcphdr *nth; 414#ifdef INET6 415 struct ip6_hdr *ip6; 416 int isipv6; 417#endif /* INET6 */ 418 int ipflags = 0; 419 struct inpcb *inp; 420 421 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 422 423#ifdef INET6 424 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 425 ip6 = ipgen; 426#endif /* INET6 */ 427 ip = ipgen; 428 429 if (tp != NULL) { 430 inp = tp->t_inpcb; 431 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 432 INP_INFO_WLOCK_ASSERT(&tcbinfo); 433 INP_LOCK_ASSERT(inp); 434 } else 435 inp = NULL; 436 437 if (tp != NULL) { 438 if (!(flags & TH_RST)) { 439 win = sbspace(&inp->inp_socket->so_rcv); 440 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 441 win = (long)TCP_MAXWIN << tp->rcv_scale; 442 } 443 } 444 if (m == NULL) { 445 m = m_gethdr(M_DONTWAIT, MT_HEADER); 446 if (m == NULL) 447 return; 448 tlen = 0; 449 m->m_data += max_linkhdr; 450#ifdef INET6 451 if (isipv6) { 452 bcopy((caddr_t)ip6, mtod(m, caddr_t), 453 sizeof(struct ip6_hdr)); 454 ip6 = mtod(m, struct ip6_hdr *); 455 nth = (struct tcphdr *)(ip6 + 1); 456 } else 457#endif /* INET6 */ 458 { 459 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 460 ip = mtod(m, struct ip *); 461 nth = (struct tcphdr *)(ip + 1); 462 } 463 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 464 flags = TH_ACK; 465 } else { 466 m_freem(m->m_next); 467 m->m_next = NULL; 468 m->m_data = (caddr_t)ipgen; 469 /* m_len is set later */ 470 tlen = 0; 471#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 472#ifdef INET6 473 if (isipv6) { 474 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 475 nth = (struct tcphdr *)(ip6 + 1); 476 } else 477#endif /* INET6 */ 478 { 479 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 480 nth = (struct tcphdr *)(ip + 1); 481 } 482 if (th != nth) { 483 /* 484 * this is usually a case when an extension header 485 * exists between the IPv6 header and the 486 * TCP header. 487 */ 488 nth->th_sport = th->th_sport; 489 nth->th_dport = th->th_dport; 490 } 491 xchg(nth->th_dport, nth->th_sport, n_short); 492#undef xchg 493 } 494#ifdef INET6 495 if (isipv6) { 496 ip6->ip6_flow = 0; 497 ip6->ip6_vfc = IPV6_VERSION; 498 ip6->ip6_nxt = IPPROTO_TCP; 499 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 500 tlen)); 501 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 502 } else 503#endif 504 { 505 tlen += sizeof (struct tcpiphdr); 506 ip->ip_len = tlen; 507 ip->ip_ttl = ip_defttl; 508 if (path_mtu_discovery) 509 ip->ip_off |= IP_DF; 510 } 511 m->m_len = tlen; 512 m->m_pkthdr.len = tlen; 513 m->m_pkthdr.rcvif = NULL; 514#ifdef MAC 515 if (inp != NULL) { 516 /* 517 * Packet is associated with a socket, so allow the 518 * label of the response to reflect the socket label. 519 */ 520 INP_LOCK_ASSERT(inp); 521 mac_create_mbuf_from_inpcb(inp, m); 522 } else { 523 /* 524 * Packet is not associated with a socket, so possibly 525 * update the label in place. 526 */ 527 mac_reflect_mbuf_tcp(m); 528 } 529#endif 530 nth->th_seq = htonl(seq); 531 nth->th_ack = htonl(ack); 532 nth->th_x2 = 0; 533 nth->th_off = sizeof (struct tcphdr) >> 2; 534 nth->th_flags = flags; 535 if (tp != NULL) 536 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 537 else 538 nth->th_win = htons((u_short)win); 539 nth->th_urp = 0; 540#ifdef INET6 541 if (isipv6) { 542 nth->th_sum = 0; 543 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 544 sizeof(struct ip6_hdr), 545 tlen - sizeof(struct ip6_hdr)); 546 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 547 NULL, NULL); 548 } else 549#endif /* INET6 */ 550 { 551 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 552 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 553 m->m_pkthdr.csum_flags = CSUM_TCP; 554 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 555 } 556#ifdef TCPDEBUG 557 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 558 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 559#endif 560#ifdef INET6 561 if (isipv6) 562 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 563 else 564#endif /* INET6 */ 565 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 566} 567 568/* 569 * Create a new TCP control block, making an 570 * empty reassembly queue and hooking it to the argument 571 * protocol control block. The `inp' parameter must have 572 * come from the zone allocator set up in tcp_init(). 573 */ 574struct tcpcb * 575tcp_newtcpcb(inp) 576 struct inpcb *inp; 577{ 578 struct tcpcb_mem *tm; 579 struct tcpcb *tp; 580#ifdef INET6 581 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 582#endif /* INET6 */ 583 int callout_flag; 584 585 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 586 if (tm == NULL) 587 return (NULL); 588 tp = &tm->tcb; 589 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 590 tp->t_maxseg = tp->t_maxopd = 591#ifdef INET6 592 isipv6 ? tcp_v6mssdflt : 593#endif /* INET6 */ 594 tcp_mssdflt; 595 596 /* Set up our timeouts. */ 597 /* 598 * XXXRW: Are these actually MPSAFE? I think so, but need to 599 * review the timed wait code, as it has some list variables, 600 * etc, that are global. 601 */ 602 callout_flag = debug_mpsafenet ? CALLOUT_MPSAFE : 0; 603 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, callout_flag); 604 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, callout_flag); 605 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, callout_flag); 606 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, callout_flag); 607 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, callout_flag); 608 609 if (tcp_do_rfc1323) 610 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 611 tp->sack_enable = tcp_do_sack; 612 tp->t_inpcb = inp; /* XXX */ 613 /* 614 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 615 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 616 * reasonable initial retransmit time. 617 */ 618 tp->t_srtt = TCPTV_SRTTBASE; 619 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 620 tp->t_rttmin = tcp_rexmit_min; 621 tp->t_rxtcur = TCPTV_RTOBASE; 622 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 623 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 624 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 625 tp->t_rcvtime = ticks; 626 tp->t_bw_rtttime = ticks; 627 /* 628 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 629 * because the socket may be bound to an IPv6 wildcard address, 630 * which may match an IPv4-mapped IPv6 address. 631 */ 632 inp->inp_ip_ttl = ip_defttl; 633 inp->inp_ppcb = (caddr_t)tp; 634 return (tp); /* XXX */ 635} 636 637/* 638 * Drop a TCP connection, reporting 639 * the specified error. If connection is synchronized, 640 * then send a RST to peer. 641 */ 642struct tcpcb * 643tcp_drop(tp, errno) 644 register struct tcpcb *tp; 645 int errno; 646{ 647 struct socket *so = tp->t_inpcb->inp_socket; 648 649 INP_LOCK_ASSERT(tp->t_inpcb); 650 if (TCPS_HAVERCVDSYN(tp->t_state)) { 651 tp->t_state = TCPS_CLOSED; 652 (void) tcp_output(tp); 653 tcpstat.tcps_drops++; 654 } else 655 tcpstat.tcps_conndrops++; 656 if (errno == ETIMEDOUT && tp->t_softerror) 657 errno = tp->t_softerror; 658 so->so_error = errno; 659 return (tcp_close(tp)); 660} 661 662static void 663tcp_discardcb(tp) 664 struct tcpcb *tp; 665{ 666 struct tseg_qent *q; 667 struct inpcb *inp = tp->t_inpcb; 668 struct socket *so = inp->inp_socket; 669#ifdef INET6 670 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 671#endif /* INET6 */ 672 673 /* 674 * Make sure that all of our timers are stopped before we 675 * delete the PCB. 676 */ 677 callout_stop(tp->tt_rexmt); 678 callout_stop(tp->tt_persist); 679 callout_stop(tp->tt_keep); 680 callout_stop(tp->tt_2msl); 681 callout_stop(tp->tt_delack); 682 683 /* 684 * If we got enough samples through the srtt filter, 685 * save the rtt and rttvar in the routing entry. 686 * 'Enough' is arbitrarily defined as 4 rtt samples. 687 * 4 samples is enough for the srtt filter to converge 688 * to within enough % of the correct value; fewer samples 689 * and we could save a bogus rtt. The danger is not high 690 * as tcp quickly recovers from everything. 691 * XXX: Works very well but needs some more statistics! 692 */ 693 if (tp->t_rttupdated >= 4) { 694 struct hc_metrics_lite metrics; 695 u_long ssthresh; 696 697 bzero(&metrics, sizeof(metrics)); 698 /* 699 * Update the ssthresh always when the conditions below 700 * are satisfied. This gives us better new start value 701 * for the congestion avoidance for new connections. 702 * ssthresh is only set if packet loss occured on a session. 703 */ 704 ssthresh = tp->snd_ssthresh; 705 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 706 /* 707 * convert the limit from user data bytes to 708 * packets then to packet data bytes. 709 */ 710 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 711 if (ssthresh < 2) 712 ssthresh = 2; 713 ssthresh *= (u_long)(tp->t_maxseg + 714#ifdef INET6 715 (isipv6 ? sizeof (struct ip6_hdr) + 716 sizeof (struct tcphdr) : 717#endif 718 sizeof (struct tcpiphdr) 719#ifdef INET6 720 ) 721#endif 722 ); 723 } else 724 ssthresh = 0; 725 metrics.rmx_ssthresh = ssthresh; 726 727 metrics.rmx_rtt = tp->t_srtt; 728 metrics.rmx_rttvar = tp->t_rttvar; 729 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 730 metrics.rmx_bandwidth = tp->snd_bandwidth; 731 metrics.rmx_cwnd = tp->snd_cwnd; 732 metrics.rmx_sendpipe = 0; 733 metrics.rmx_recvpipe = 0; 734 735 tcp_hc_update(&inp->inp_inc, &metrics); 736 } 737 738 /* free the reassembly queue, if any */ 739 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 740 LIST_REMOVE(q, tqe_q); 741 m_freem(q->tqe_m); 742 uma_zfree(tcp_reass_zone, q); 743 tp->t_segqlen--; 744 tcp_reass_qsize--; 745 } 746 tcp_free_sackholes(tp); 747 inp->inp_ppcb = NULL; 748 tp->t_inpcb = NULL; 749 uma_zfree(tcpcb_zone, tp); 750 soisdisconnected(so); 751} 752 753/* 754 * Close a TCP control block: 755 * discard all space held by the tcp 756 * discard internet protocol block 757 * wake up any sleepers 758 */ 759struct tcpcb * 760tcp_close(tp) 761 struct tcpcb *tp; 762{ 763 struct inpcb *inp = tp->t_inpcb; 764#ifdef INET6 765 struct socket *so = inp->inp_socket; 766#endif 767 768 tcp_discardcb(tp); 769#ifdef INET6 770 if (INP_CHECK_SOCKAF(so, AF_INET6)) 771 in6_pcbdetach(inp); 772 else 773#endif 774 in_pcbdetach(inp); 775 tcpstat.tcps_closed++; 776 return (NULL); 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 if (inpb->inp_vflag & INP_TIMEWAIT) 799 continue; 800 INP_LOCK(inpb); 801 if ((tcpb = intotcpcb(inpb)) != NULL) { 802 while ((te = LIST_FIRST(&tcpb->t_segq)) 803 != NULL) { 804 LIST_REMOVE(te, tqe_q); 805 m_freem(te->tqe_m); 806 uma_zfree(tcp_reass_zone, te); 807 tcpb->t_segqlen--; 808 tcp_reass_qsize--; 809 } 810 } 811 INP_UNLOCK(inpb); 812 } 813 INP_INFO_RUNLOCK(&tcbinfo); 814 } 815} 816 817/* 818 * Notify a tcp user of an asynchronous error; 819 * store error as soft error, but wake up user 820 * (for now, won't do anything until can select for soft error). 821 * 822 * Do not wake up user since there currently is no mechanism for 823 * reporting soft errors (yet - a kqueue filter may be added). 824 */ 825static struct inpcb * 826tcp_notify(inp, error) 827 struct inpcb *inp; 828 int error; 829{ 830 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 831 832 /* 833 * Ignore some errors if we are hooked up. 834 * If connection hasn't completed, has retransmitted several times, 835 * and receives a second error, give up now. This is better 836 * than waiting a long time to establish a connection that 837 * can never complete. 838 */ 839 if (tp->t_state == TCPS_ESTABLISHED && 840 (error == EHOSTUNREACH || error == ENETUNREACH || 841 error == EHOSTDOWN)) { 842 return inp; 843 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 844 tp->t_softerror) { 845 tcp_drop(tp, error); 846 return (struct inpcb *)0; 847 } else { 848 tp->t_softerror = error; 849 return inp; 850 } 851#if 0 852 wakeup( &so->so_timeo); 853 sorwakeup(so); 854 sowwakeup(so); 855#endif 856} 857 858static int 859tcp_pcblist(SYSCTL_HANDLER_ARGS) 860{ 861 int error, i, n, s; 862 struct inpcb *inp, **inp_list; 863 inp_gen_t gencnt; 864 struct xinpgen xig; 865 866 /* 867 * The process of preparing the TCB list is too time-consuming and 868 * resource-intensive to repeat twice on every request. 869 */ 870 if (req->oldptr == NULL) { 871 n = tcbinfo.ipi_count; 872 req->oldidx = 2 * (sizeof xig) 873 + (n + n/8) * sizeof(struct xtcpcb); 874 return 0; 875 } 876 877 if (req->newptr != NULL) 878 return EPERM; 879 880 /* 881 * OK, now we're committed to doing something. 882 */ 883 s = splnet(); 884 INP_INFO_RLOCK(&tcbinfo); 885 gencnt = tcbinfo.ipi_gencnt; 886 n = tcbinfo.ipi_count; 887 INP_INFO_RUNLOCK(&tcbinfo); 888 splx(s); 889 890 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 891 + n * sizeof(struct xtcpcb)); 892 if (error != 0) 893 return (error); 894 895 xig.xig_len = sizeof xig; 896 xig.xig_count = n; 897 xig.xig_gen = gencnt; 898 xig.xig_sogen = so_gencnt; 899 error = SYSCTL_OUT(req, &xig, sizeof xig); 900 if (error) 901 return error; 902 903 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 904 if (inp_list == NULL) 905 return ENOMEM; 906 907 s = splnet(); 908 INP_INFO_RLOCK(&tcbinfo); 909 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 910 inp = LIST_NEXT(inp, inp_list)) { 911 INP_LOCK(inp); 912 if (inp->inp_gencnt <= gencnt) { 913 /* 914 * XXX: This use of cr_cansee(), introduced with 915 * TCP state changes, is not quite right, but for 916 * now, better than nothing. 917 */ 918 if (inp->inp_vflag & INP_TIMEWAIT) 919 error = cr_cansee(req->td->td_ucred, 920 intotw(inp)->tw_cred); 921 else 922 error = cr_canseesocket(req->td->td_ucred, 923 inp->inp_socket); 924 if (error == 0) 925 inp_list[i++] = inp; 926 } 927 INP_UNLOCK(inp); 928 } 929 INP_INFO_RUNLOCK(&tcbinfo); 930 splx(s); 931 n = i; 932 933 error = 0; 934 for (i = 0; i < n; i++) { 935 inp = inp_list[i]; 936 if (inp->inp_gencnt <= gencnt) { 937 struct xtcpcb xt; 938 caddr_t inp_ppcb; 939 xt.xt_len = sizeof xt; 940 /* XXX should avoid extra copy */ 941 bcopy(inp, &xt.xt_inp, sizeof *inp); 942 inp_ppcb = inp->inp_ppcb; 943 if (inp_ppcb == NULL) 944 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 945 else if (inp->inp_vflag & INP_TIMEWAIT) { 946 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 947 xt.xt_tp.t_state = TCPS_TIME_WAIT; 948 } else 949 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 950 if (inp->inp_socket != NULL) 951 sotoxsocket(inp->inp_socket, &xt.xt_socket); 952 else { 953 bzero(&xt.xt_socket, sizeof xt.xt_socket); 954 xt.xt_socket.xso_protocol = IPPROTO_TCP; 955 } 956 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 957 error = SYSCTL_OUT(req, &xt, sizeof xt); 958 } 959 } 960 if (!error) { 961 /* 962 * Give the user an updated idea of our state. 963 * If the generation differs from what we told 964 * her before, she knows that something happened 965 * while we were processing this request, and it 966 * might be necessary to retry. 967 */ 968 s = splnet(); 969 INP_INFO_RLOCK(&tcbinfo); 970 xig.xig_gen = tcbinfo.ipi_gencnt; 971 xig.xig_sogen = so_gencnt; 972 xig.xig_count = tcbinfo.ipi_count; 973 INP_INFO_RUNLOCK(&tcbinfo); 974 splx(s); 975 error = SYSCTL_OUT(req, &xig, sizeof xig); 976 } 977 free(inp_list, M_TEMP); 978 return error; 979} 980 981SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 982 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 983 984static int 985tcp_getcred(SYSCTL_HANDLER_ARGS) 986{ 987 struct xucred xuc; 988 struct sockaddr_in addrs[2]; 989 struct inpcb *inp; 990 int error, s; 991 992 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 993 if (error) 994 return (error); 995 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 996 if (error) 997 return (error); 998 s = splnet(); 999 INP_INFO_RLOCK(&tcbinfo); 1000 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1001 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1002 if (inp == NULL) { 1003 error = ENOENT; 1004 goto outunlocked; 1005 } 1006 INP_LOCK(inp); 1007 if (inp->inp_socket == NULL) { 1008 error = ENOENT; 1009 goto out; 1010 } 1011 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1012 if (error) 1013 goto out; 1014 cru2x(inp->inp_socket->so_cred, &xuc); 1015out: 1016 INP_UNLOCK(inp); 1017outunlocked: 1018 INP_INFO_RUNLOCK(&tcbinfo); 1019 splx(s); 1020 if (error == 0) 1021 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1022 return (error); 1023} 1024 1025SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1026 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1027 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1028 1029#ifdef INET6 1030static int 1031tcp6_getcred(SYSCTL_HANDLER_ARGS) 1032{ 1033 struct xucred xuc; 1034 struct sockaddr_in6 addrs[2]; 1035 struct inpcb *inp; 1036 int error, s, mapped = 0; 1037 1038 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1039 if (error) 1040 return (error); 1041 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1042 if (error) 1043 return (error); 1044 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1045 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1046 mapped = 1; 1047 else 1048 return (EINVAL); 1049 } 1050 s = splnet(); 1051 INP_INFO_RLOCK(&tcbinfo); 1052 if (mapped == 1) 1053 inp = in_pcblookup_hash(&tcbinfo, 1054 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1055 addrs[1].sin6_port, 1056 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1057 addrs[0].sin6_port, 1058 0, NULL); 1059 else 1060 inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr, 1061 addrs[1].sin6_port, 1062 &addrs[0].sin6_addr, addrs[0].sin6_port, 1063 0, NULL); 1064 if (inp == NULL) { 1065 error = ENOENT; 1066 goto outunlocked; 1067 } 1068 INP_LOCK(inp); 1069 if (inp->inp_socket == NULL) { 1070 error = ENOENT; 1071 goto out; 1072 } 1073 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1074 if (error) 1075 goto out; 1076 cru2x(inp->inp_socket->so_cred, &xuc); 1077out: 1078 INP_UNLOCK(inp); 1079outunlocked: 1080 INP_INFO_RUNLOCK(&tcbinfo); 1081 splx(s); 1082 if (error == 0) 1083 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1084 return (error); 1085} 1086 1087SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1088 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1089 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1090#endif 1091 1092 1093void 1094tcp_ctlinput(cmd, sa, vip) 1095 int cmd; 1096 struct sockaddr *sa; 1097 void *vip; 1098{ 1099 struct ip *ip = vip; 1100 struct tcphdr *th; 1101 struct in_addr faddr; 1102 struct inpcb *inp; 1103 struct tcpcb *tp; 1104 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1105 tcp_seq icmp_seq; 1106 int s; 1107 1108 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1109 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1110 return; 1111 1112 if (cmd == PRC_QUENCH) 1113 notify = tcp_quench; 1114 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1115 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1116 notify = tcp_drop_syn_sent; 1117 else if (cmd == PRC_MSGSIZE) 1118 notify = tcp_mtudisc; 1119 /* 1120 * Redirects don't need to be handled up here. 1121 */ 1122 else if (PRC_IS_REDIRECT(cmd)) 1123 return; 1124 /* 1125 * Hostdead is ugly because it goes linearly through all PCBs. 1126 * XXX: We never get this from ICMP, otherwise it makes an 1127 * excellent DoS attack on machines with many connections. 1128 */ 1129 else if (cmd == PRC_HOSTDEAD) 1130 ip = NULL; 1131 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1132 return; 1133 if (ip != NULL) { 1134 s = splnet(); 1135 th = (struct tcphdr *)((caddr_t)ip 1136 + (ip->ip_hl << 2)); 1137 INP_INFO_WLOCK(&tcbinfo); 1138 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1139 ip->ip_src, th->th_sport, 0, NULL); 1140 if (inp != NULL) { 1141 INP_LOCK(inp); 1142 if (inp->inp_socket != NULL) { 1143 icmp_seq = htonl(th->th_seq); 1144 tp = intotcpcb(inp); 1145 if (SEQ_GEQ(icmp_seq, tp->snd_una) && 1146 SEQ_LT(icmp_seq, tp->snd_max)) 1147 inp = (*notify)(inp, inetctlerrmap[cmd]); 1148 } 1149 if (inp != NULL) 1150 INP_UNLOCK(inp); 1151 } else { 1152 struct in_conninfo inc; 1153 1154 inc.inc_fport = th->th_dport; 1155 inc.inc_lport = th->th_sport; 1156 inc.inc_faddr = faddr; 1157 inc.inc_laddr = ip->ip_src; 1158#ifdef INET6 1159 inc.inc_isipv6 = 0; 1160#endif 1161 syncache_unreach(&inc, th); 1162 } 1163 INP_INFO_WUNLOCK(&tcbinfo); 1164 splx(s); 1165 } else 1166 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1167} 1168 1169#ifdef INET6 1170void 1171tcp6_ctlinput(cmd, sa, d) 1172 int cmd; 1173 struct sockaddr *sa; 1174 void *d; 1175{ 1176 struct tcphdr th; 1177 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1178 struct ip6_hdr *ip6; 1179 struct mbuf *m; 1180 struct ip6ctlparam *ip6cp = NULL; 1181 const struct sockaddr_in6 *sa6_src = NULL; 1182 int off; 1183 struct tcp_portonly { 1184 u_int16_t th_sport; 1185 u_int16_t th_dport; 1186 } *thp; 1187 1188 if (sa->sa_family != AF_INET6 || 1189 sa->sa_len != sizeof(struct sockaddr_in6)) 1190 return; 1191 1192 if (cmd == PRC_QUENCH) 1193 notify = tcp_quench; 1194 else if (cmd == PRC_MSGSIZE) 1195 notify = tcp_mtudisc; 1196 else if (!PRC_IS_REDIRECT(cmd) && 1197 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1198 return; 1199 1200 /* if the parameter is from icmp6, decode it. */ 1201 if (d != NULL) { 1202 ip6cp = (struct ip6ctlparam *)d; 1203 m = ip6cp->ip6c_m; 1204 ip6 = ip6cp->ip6c_ip6; 1205 off = ip6cp->ip6c_off; 1206 sa6_src = ip6cp->ip6c_src; 1207 } else { 1208 m = NULL; 1209 ip6 = NULL; 1210 off = 0; /* fool gcc */ 1211 sa6_src = &sa6_any; 1212 } 1213 1214 if (ip6 != NULL) { 1215 struct in_conninfo inc; 1216 /* 1217 * XXX: We assume that when IPV6 is non NULL, 1218 * M and OFF are valid. 1219 */ 1220 1221 /* check if we can safely examine src and dst ports */ 1222 if (m->m_pkthdr.len < off + sizeof(*thp)) 1223 return; 1224 1225 bzero(&th, sizeof(th)); 1226 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1227 1228 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1229 (struct sockaddr *)ip6cp->ip6c_src, 1230 th.th_sport, cmd, NULL, notify); 1231 1232 inc.inc_fport = th.th_dport; 1233 inc.inc_lport = th.th_sport; 1234 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1235 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1236 inc.inc_isipv6 = 1; 1237 INP_INFO_WLOCK(&tcbinfo); 1238 syncache_unreach(&inc, &th); 1239 INP_INFO_WUNLOCK(&tcbinfo); 1240 } else 1241 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1242 0, cmd, NULL, notify); 1243} 1244#endif /* INET6 */ 1245 1246 1247/* 1248 * Following is where TCP initial sequence number generation occurs. 1249 * 1250 * There are two places where we must use initial sequence numbers: 1251 * 1. In SYN-ACK packets. 1252 * 2. In SYN packets. 1253 * 1254 * All ISNs for SYN-ACK packets are generated by the syncache. See 1255 * tcp_syncache.c for details. 1256 * 1257 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1258 * depends on this property. In addition, these ISNs should be 1259 * unguessable so as to prevent connection hijacking. To satisfy 1260 * the requirements of this situation, the algorithm outlined in 1261 * RFC 1948 is used, with only small modifications. 1262 * 1263 * Implementation details: 1264 * 1265 * Time is based off the system timer, and is corrected so that it 1266 * increases by one megabyte per second. This allows for proper 1267 * recycling on high speed LANs while still leaving over an hour 1268 * before rollover. 1269 * 1270 * As reading the *exact* system time is too expensive to be done 1271 * whenever setting up a TCP connection, we increment the time 1272 * offset in two ways. First, a small random positive increment 1273 * is added to isn_offset for each connection that is set up. 1274 * Second, the function tcp_isn_tick fires once per clock tick 1275 * and increments isn_offset as necessary so that sequence numbers 1276 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1277 * random positive increments serve only to ensure that the same 1278 * exact sequence number is never sent out twice (as could otherwise 1279 * happen when a port is recycled in less than the system tick 1280 * interval.) 1281 * 1282 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1283 * between seeding of isn_secret. This is normally set to zero, 1284 * as reseeding should not be necessary. 1285 * 1286 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1287 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1288 * general, this means holding an exclusive (write) lock. 1289 */ 1290 1291#define ISN_BYTES_PER_SECOND 1048576 1292#define ISN_STATIC_INCREMENT 4096 1293#define ISN_RANDOM_INCREMENT (4096 - 1) 1294 1295static u_char isn_secret[32]; 1296static int isn_last_reseed; 1297static u_int32_t isn_offset, isn_offset_old; 1298static MD5_CTX isn_ctx; 1299 1300tcp_seq 1301tcp_new_isn(tp) 1302 struct tcpcb *tp; 1303{ 1304 u_int32_t md5_buffer[4]; 1305 tcp_seq new_isn; 1306 1307 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1308 1309 /* Seed if this is the first use, reseed if requested. */ 1310 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1311 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1312 < (u_int)ticks))) { 1313 read_random(&isn_secret, sizeof(isn_secret)); 1314 isn_last_reseed = ticks; 1315 } 1316 1317 /* Compute the md5 hash and return the ISN. */ 1318 MD5Init(&isn_ctx); 1319 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1320 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1321#ifdef INET6 1322 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1323 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1324 sizeof(struct in6_addr)); 1325 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1326 sizeof(struct in6_addr)); 1327 } else 1328#endif 1329 { 1330 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1331 sizeof(struct in_addr)); 1332 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1333 sizeof(struct in_addr)); 1334 } 1335 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1336 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1337 new_isn = (tcp_seq) md5_buffer[0]; 1338 isn_offset += ISN_STATIC_INCREMENT + 1339 (arc4random() & ISN_RANDOM_INCREMENT); 1340 new_isn += isn_offset; 1341 return new_isn; 1342} 1343 1344/* 1345 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1346 * to keep time flowing at a relatively constant rate. If the random 1347 * increments have already pushed us past the projected offset, do nothing. 1348 */ 1349static void 1350tcp_isn_tick(xtp) 1351 void *xtp; 1352{ 1353 u_int32_t projected_offset; 1354 1355 INP_INFO_WLOCK(&tcbinfo); 1356 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / hz; 1357 1358 if (projected_offset > isn_offset) 1359 isn_offset = projected_offset; 1360 1361 isn_offset_old = isn_offset; 1362 callout_reset(&isn_callout, 1, tcp_isn_tick, NULL); 1363 INP_INFO_WUNLOCK(&tcbinfo); 1364} 1365 1366/* 1367 * When a source quench is received, close congestion window 1368 * to one segment. We will gradually open it again as we proceed. 1369 */ 1370struct inpcb * 1371tcp_quench(inp, errno) 1372 struct inpcb *inp; 1373 int errno; 1374{ 1375 struct tcpcb *tp = intotcpcb(inp); 1376 1377 INP_LOCK_ASSERT(inp); 1378 if (tp != NULL) 1379 tp->snd_cwnd = tp->t_maxseg; 1380 return (inp); 1381} 1382 1383/* 1384 * When a specific ICMP unreachable message is received and the 1385 * connection state is SYN-SENT, drop the connection. This behavior 1386 * is controlled by the icmp_may_rst sysctl. 1387 */ 1388struct inpcb * 1389tcp_drop_syn_sent(inp, errno) 1390 struct inpcb *inp; 1391 int errno; 1392{ 1393 struct tcpcb *tp = intotcpcb(inp); 1394 1395 INP_LOCK_ASSERT(inp); 1396 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) { 1397 tcp_drop(tp, errno); 1398 return (struct inpcb *)0; 1399 } 1400 return inp; 1401} 1402 1403/* 1404 * When `need fragmentation' ICMP is received, update our idea of the MSS 1405 * based on the new value in the route. Also nudge TCP to send something, 1406 * since we know the packet we just sent was dropped. 1407 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1408 */ 1409struct inpcb * 1410tcp_mtudisc(inp, errno) 1411 struct inpcb *inp; 1412 int errno; 1413{ 1414 struct tcpcb *tp = intotcpcb(inp); 1415 struct socket *so = inp->inp_socket; 1416 u_int maxmtu; 1417 u_int romtu; 1418 int mss; 1419#ifdef INET6 1420 int isipv6; 1421#endif /* INET6 */ 1422 1423 INP_LOCK_ASSERT(inp); 1424 if (tp != NULL) { 1425#ifdef INET6 1426 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1427#endif 1428 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1429 romtu = 1430#ifdef INET6 1431 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1432#endif /* INET6 */ 1433 tcp_maxmtu(&inp->inp_inc); 1434 if (!maxmtu) 1435 maxmtu = romtu; 1436 else 1437 maxmtu = min(maxmtu, romtu); 1438 if (!maxmtu) { 1439 tp->t_maxopd = tp->t_maxseg = 1440#ifdef INET6 1441 isipv6 ? tcp_v6mssdflt : 1442#endif /* INET6 */ 1443 tcp_mssdflt; 1444 return inp; 1445 } 1446 mss = maxmtu - 1447#ifdef INET6 1448 (isipv6 ? 1449 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1450#endif /* INET6 */ 1451 sizeof(struct tcpiphdr) 1452#ifdef INET6 1453 ) 1454#endif /* INET6 */ 1455 ; 1456 1457 /* 1458 * XXX - The above conditional probably violates the TCP 1459 * spec. The problem is that, since we don't know the 1460 * other end's MSS, we are supposed to use a conservative 1461 * default. But, if we do that, then MTU discovery will 1462 * never actually take place, because the conservative 1463 * default is much less than the MTUs typically seen 1464 * on the Internet today. For the moment, we'll sweep 1465 * this under the carpet. 1466 * 1467 * The conservative default might not actually be a problem 1468 * if the only case this occurs is when sending an initial 1469 * SYN with options and data to a host we've never talked 1470 * to before. Then, they will reply with an MSS value which 1471 * will get recorded and the new parameters should get 1472 * recomputed. For Further Study. 1473 */ 1474 if (tp->t_maxopd <= mss) 1475 return inp; 1476 tp->t_maxopd = mss; 1477 1478 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1479 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1480 mss -= TCPOLEN_TSTAMP_APPA; 1481#if (MCLBYTES & (MCLBYTES - 1)) == 0 1482 if (mss > MCLBYTES) 1483 mss &= ~(MCLBYTES-1); 1484#else 1485 if (mss > MCLBYTES) 1486 mss = mss / MCLBYTES * MCLBYTES; 1487#endif 1488 if (so->so_snd.sb_hiwat < mss) 1489 mss = so->so_snd.sb_hiwat; 1490 1491 tp->t_maxseg = mss; 1492 1493 tcpstat.tcps_mturesent++; 1494 tp->t_rtttime = 0; 1495 tp->snd_nxt = tp->snd_una; 1496 tcp_output(tp); 1497 } 1498 return inp; 1499} 1500 1501/* 1502 * Look-up the routing entry to the peer of this inpcb. If no route 1503 * is found and it cannot be allocated, then return NULL. This routine 1504 * is called by TCP routines that access the rmx structure and by tcp_mss 1505 * to get the interface MTU. 1506 */ 1507u_long 1508tcp_maxmtu(inc) 1509 struct in_conninfo *inc; 1510{ 1511 struct route sro; 1512 struct sockaddr_in *dst; 1513 struct ifnet *ifp; 1514 u_long maxmtu = 0; 1515 1516 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1517 1518 bzero(&sro, sizeof(sro)); 1519 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1520 dst = (struct sockaddr_in *)&sro.ro_dst; 1521 dst->sin_family = AF_INET; 1522 dst->sin_len = sizeof(*dst); 1523 dst->sin_addr = inc->inc_faddr; 1524 rtalloc_ign(&sro, RTF_CLONING); 1525 } 1526 if (sro.ro_rt != NULL) { 1527 ifp = sro.ro_rt->rt_ifp; 1528 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1529 maxmtu = ifp->if_mtu; 1530 else 1531 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1532 RTFREE(sro.ro_rt); 1533 } 1534 return (maxmtu); 1535} 1536 1537#ifdef INET6 1538u_long 1539tcp_maxmtu6(inc) 1540 struct in_conninfo *inc; 1541{ 1542 struct route_in6 sro6; 1543 struct ifnet *ifp; 1544 u_long maxmtu = 0; 1545 1546 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1547 1548 bzero(&sro6, sizeof(sro6)); 1549 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1550 sro6.ro_dst.sin6_family = AF_INET6; 1551 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1552 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1553 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1554 } 1555 if (sro6.ro_rt != NULL) { 1556 ifp = sro6.ro_rt->rt_ifp; 1557 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1558 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1559 else 1560 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1561 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1562 RTFREE(sro6.ro_rt); 1563 } 1564 1565 return (maxmtu); 1566} 1567#endif /* INET6 */ 1568 1569#ifdef IPSEC 1570/* compute ESP/AH header size for TCP, including outer IP header. */ 1571size_t 1572ipsec_hdrsiz_tcp(tp) 1573 struct tcpcb *tp; 1574{ 1575 struct inpcb *inp; 1576 struct mbuf *m; 1577 size_t hdrsiz; 1578 struct ip *ip; 1579#ifdef INET6 1580 struct ip6_hdr *ip6; 1581#endif 1582 struct tcphdr *th; 1583 1584 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1585 return 0; 1586 MGETHDR(m, M_DONTWAIT, MT_DATA); 1587 if (!m) 1588 return 0; 1589 1590#ifdef INET6 1591 if ((inp->inp_vflag & INP_IPV6) != 0) { 1592 ip6 = mtod(m, struct ip6_hdr *); 1593 th = (struct tcphdr *)(ip6 + 1); 1594 m->m_pkthdr.len = m->m_len = 1595 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1596 tcpip_fillheaders(inp, ip6, th); 1597 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1598 } else 1599#endif /* INET6 */ 1600 { 1601 ip = mtod(m, struct ip *); 1602 th = (struct tcphdr *)(ip + 1); 1603 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1604 tcpip_fillheaders(inp, ip, th); 1605 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1606 } 1607 1608 m_free(m); 1609 return hdrsiz; 1610} 1611#endif /*IPSEC*/ 1612 1613/* 1614 * Move a TCP connection into TIME_WAIT state. 1615 * tcbinfo is unlocked. 1616 * inp is locked, and is unlocked before returning. 1617 */ 1618void 1619tcp_twstart(tp) 1620 struct tcpcb *tp; 1621{ 1622 struct tcptw *tw; 1623 struct inpcb *inp; 1624 int tw_time, acknow; 1625 struct socket *so; 1626 1627 INP_LOCK_ASSERT(tp->t_inpcb); 1628 1629 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1630 if (tw == NULL) { 1631 tw = tcp_timer_2msl_tw(1); 1632 if (tw == NULL) { 1633 tcp_close(tp); 1634 return; 1635 } 1636 } 1637 inp = tp->t_inpcb; 1638 tw->tw_inpcb = inp; 1639 1640 /* 1641 * Recover last window size sent. 1642 */ 1643 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1644 1645 /* 1646 * Set t_recent if timestamps are used on the connection. 1647 */ 1648 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1649 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1650 tw->t_recent = tp->ts_recent; 1651 else 1652 tw->t_recent = 0; 1653 1654 tw->snd_nxt = tp->snd_nxt; 1655 tw->rcv_nxt = tp->rcv_nxt; 1656 tw->iss = tp->iss; 1657 tw->irs = tp->irs; 1658 tw->t_starttime = tp->t_starttime; 1659 tw->tw_time = 0; 1660 1661/* XXX 1662 * If this code will 1663 * be used for fin-wait-2 state also, then we may need 1664 * a ts_recent from the last segment. 1665 */ 1666 tw_time = 2 * tcp_msl; 1667 acknow = tp->t_flags & TF_ACKNOW; 1668 tcp_discardcb(tp); 1669 so = inp->inp_socket; 1670 ACCEPT_LOCK(); 1671 SOCK_LOCK(so); 1672 so->so_pcb = NULL; 1673 tw->tw_cred = crhold(so->so_cred); 1674 tw->tw_so_options = so->so_options; 1675 sotryfree(so); 1676 inp->inp_socket = NULL; 1677 if (acknow) 1678 tcp_twrespond(tw, TH_ACK); 1679 inp->inp_ppcb = (caddr_t)tw; 1680 inp->inp_vflag |= INP_TIMEWAIT; 1681 tcp_timer_2msl_reset(tw, tw_time); 1682 INP_UNLOCK(inp); 1683} 1684 1685/* 1686 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1687 * the actual rate is slightly higher due to the addition of 1688 * random positive increments. 1689 * 1690 * Most other new OSes use semi-randomized ISN values, so we 1691 * do not need to worry about them. 1692 */ 1693#define MS_ISN_BYTES_PER_SECOND 250000 1694 1695/* 1696 * Determine if the ISN we will generate has advanced beyond the last 1697 * sequence number used by the previous connection. If so, indicate 1698 * that it is safe to recycle this tw socket by returning 1. 1699 * 1700 * XXXRW: This function should assert the inpcb lock as it does multiple 1701 * non-atomic reads from the tcptw, but is currently * called without it from 1702 * in_pcb.c:in_pcblookup_local(). 1703 */ 1704int 1705tcp_twrecycleable(struct tcptw *tw) 1706{ 1707 tcp_seq new_iss = tw->iss; 1708 tcp_seq new_irs = tw->irs; 1709 1710 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1711 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1712 1713 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1714 return 1; 1715 else 1716 return 0; 1717} 1718 1719struct tcptw * 1720tcp_twclose(struct tcptw *tw, int reuse) 1721{ 1722 struct inpcb *inp; 1723 1724 inp = tw->tw_inpcb; 1725 INP_LOCK_ASSERT(inp); 1726 1727 tw->tw_inpcb = NULL; 1728 tcp_timer_2msl_stop(tw); 1729 inp->inp_ppcb = NULL; 1730#ifdef INET6 1731 if (inp->inp_vflag & INP_IPV6PROTO) 1732 in6_pcbdetach(inp); 1733 else 1734#endif 1735 in_pcbdetach(inp); 1736 tcpstat.tcps_closed++; 1737 crfree(tw->tw_cred); 1738 tw->tw_cred = NULL; 1739 if (reuse) 1740 return (tw); 1741 uma_zfree(tcptw_zone, tw); 1742 return (NULL); 1743} 1744 1745int 1746tcp_twrespond(struct tcptw *tw, int flags) 1747{ 1748 struct inpcb *inp = tw->tw_inpcb; 1749 struct tcphdr *th; 1750 struct mbuf *m; 1751 struct ip *ip = NULL; 1752 u_int8_t *optp; 1753 u_int hdrlen, optlen; 1754 int error; 1755#ifdef INET6 1756 struct ip6_hdr *ip6 = NULL; 1757 int isipv6 = inp->inp_inc.inc_isipv6; 1758#endif 1759 1760 INP_LOCK_ASSERT(inp); 1761 1762 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1763 if (m == NULL) 1764 return (ENOBUFS); 1765 m->m_data += max_linkhdr; 1766 1767#ifdef MAC 1768 mac_create_mbuf_from_inpcb(inp, m); 1769#endif 1770 1771#ifdef INET6 1772 if (isipv6) { 1773 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1774 ip6 = mtod(m, struct ip6_hdr *); 1775 th = (struct tcphdr *)(ip6 + 1); 1776 tcpip_fillheaders(inp, ip6, th); 1777 } else 1778#endif 1779 { 1780 hdrlen = sizeof(struct tcpiphdr); 1781 ip = mtod(m, struct ip *); 1782 th = (struct tcphdr *)(ip + 1); 1783 tcpip_fillheaders(inp, ip, th); 1784 } 1785 optp = (u_int8_t *)(th + 1); 1786 1787 /* 1788 * Send a timestamp and echo-reply if both our side and our peer 1789 * have sent timestamps in our SYN's and this is not a RST. 1790 */ 1791 if (tw->t_recent && flags == TH_ACK) { 1792 u_int32_t *lp = (u_int32_t *)optp; 1793 1794 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1795 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1796 *lp++ = htonl(ticks); 1797 *lp = htonl(tw->t_recent); 1798 optp += TCPOLEN_TSTAMP_APPA; 1799 } 1800 1801 optlen = optp - (u_int8_t *)(th + 1); 1802 1803 m->m_len = hdrlen + optlen; 1804 m->m_pkthdr.len = m->m_len; 1805 1806 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1807 1808 th->th_seq = htonl(tw->snd_nxt); 1809 th->th_ack = htonl(tw->rcv_nxt); 1810 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1811 th->th_flags = flags; 1812 th->th_win = htons(tw->last_win); 1813 1814#ifdef INET6 1815 if (isipv6) { 1816 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1817 sizeof(struct tcphdr) + optlen); 1818 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1819 error = ip6_output(m, inp->in6p_outputopts, NULL, 1820 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1821 } else 1822#endif 1823 { 1824 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1825 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1826 m->m_pkthdr.csum_flags = CSUM_TCP; 1827 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1828 ip->ip_len = m->m_pkthdr.len; 1829 if (path_mtu_discovery) 1830 ip->ip_off |= IP_DF; 1831 error = ip_output(m, inp->inp_options, NULL, 1832 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1833 NULL, inp); 1834 } 1835 if (flags & TH_ACK) 1836 tcpstat.tcps_sndacks++; 1837 else 1838 tcpstat.tcps_sndctrl++; 1839 tcpstat.tcps_sndtotal++; 1840 return (error); 1841} 1842 1843/* 1844 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1845 * 1846 * This code attempts to calculate the bandwidth-delay product as a 1847 * means of determining the optimal window size to maximize bandwidth, 1848 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1849 * routers. This code also does a fairly good job keeping RTTs in check 1850 * across slow links like modems. We implement an algorithm which is very 1851 * similar (but not meant to be) TCP/Vegas. The code operates on the 1852 * transmitter side of a TCP connection and so only effects the transmit 1853 * side of the connection. 1854 * 1855 * BACKGROUND: TCP makes no provision for the management of buffer space 1856 * at the end points or at the intermediate routers and switches. A TCP 1857 * stream, whether using NewReno or not, will eventually buffer as 1858 * many packets as it is able and the only reason this typically works is 1859 * due to the fairly small default buffers made available for a connection 1860 * (typicaly 16K or 32K). As machines use larger windows and/or window 1861 * scaling it is now fairly easy for even a single TCP connection to blow-out 1862 * all available buffer space not only on the local interface, but on 1863 * intermediate routers and switches as well. NewReno makes a misguided 1864 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1865 * then backing off, then steadily increasing the window again until another 1866 * failure occurs, ad-infinitum. This results in terrible oscillation that 1867 * is only made worse as network loads increase and the idea of intentionally 1868 * blowing out network buffers is, frankly, a terrible way to manage network 1869 * resources. 1870 * 1871 * It is far better to limit the transmit window prior to the failure 1872 * condition being achieved. There are two general ways to do this: First 1873 * you can 'scan' through different transmit window sizes and locate the 1874 * point where the RTT stops increasing, indicating that you have filled the 1875 * pipe, then scan backwards until you note that RTT stops decreasing, then 1876 * repeat ad-infinitum. This method works in principle but has severe 1877 * implementation issues due to RTT variances, timer granularity, and 1878 * instability in the algorithm which can lead to many false positives and 1879 * create oscillations as well as interact badly with other TCP streams 1880 * implementing the same algorithm. 1881 * 1882 * The second method is to limit the window to the bandwidth delay product 1883 * of the link. This is the method we implement. RTT variances and our 1884 * own manipulation of the congestion window, bwnd, can potentially 1885 * destabilize the algorithm. For this reason we have to stabilize the 1886 * elements used to calculate the window. We do this by using the minimum 1887 * observed RTT, the long term average of the observed bandwidth, and 1888 * by adding two segments worth of slop. It isn't perfect but it is able 1889 * to react to changing conditions and gives us a very stable basis on 1890 * which to extend the algorithm. 1891 */ 1892void 1893tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1894{ 1895 u_long bw; 1896 u_long bwnd; 1897 int save_ticks; 1898 1899 /* 1900 * If inflight_enable is disabled in the middle of a tcp connection, 1901 * make sure snd_bwnd is effectively disabled. 1902 */ 1903 if (tcp_inflight_enable == 0) { 1904 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1905 tp->snd_bandwidth = 0; 1906 return; 1907 } 1908 1909 /* 1910 * Figure out the bandwidth. Due to the tick granularity this 1911 * is a very rough number and it MUST be averaged over a fairly 1912 * long period of time. XXX we need to take into account a link 1913 * that is not using all available bandwidth, but for now our 1914 * slop will ramp us up if this case occurs and the bandwidth later 1915 * increases. 1916 * 1917 * Note: if ticks rollover 'bw' may wind up negative. We must 1918 * effectively reset t_bw_rtttime for this case. 1919 */ 1920 save_ticks = ticks; 1921 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 1922 return; 1923 1924 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 1925 (save_ticks - tp->t_bw_rtttime); 1926 tp->t_bw_rtttime = save_ticks; 1927 tp->t_bw_rtseq = ack_seq; 1928 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 1929 return; 1930 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1931 1932 tp->snd_bandwidth = bw; 1933 1934 /* 1935 * Calculate the semi-static bandwidth delay product, plus two maximal 1936 * segments. The additional slop puts us squarely in the sweet 1937 * spot and also handles the bandwidth run-up case and stabilization. 1938 * Without the slop we could be locking ourselves into a lower 1939 * bandwidth. 1940 * 1941 * Situations Handled: 1942 * (1) Prevents over-queueing of packets on LANs, especially on 1943 * high speed LANs, allowing larger TCP buffers to be 1944 * specified, and also does a good job preventing 1945 * over-queueing of packets over choke points like modems 1946 * (at least for the transmit side). 1947 * 1948 * (2) Is able to handle changing network loads (bandwidth 1949 * drops so bwnd drops, bandwidth increases so bwnd 1950 * increases). 1951 * 1952 * (3) Theoretically should stabilize in the face of multiple 1953 * connections implementing the same algorithm (this may need 1954 * a little work). 1955 * 1956 * (4) Stability value (defaults to 20 = 2 maximal packets) can 1957 * be adjusted with a sysctl but typically only needs to be 1958 * on very slow connections. A value no smaller then 5 1959 * should be used, but only reduce this default if you have 1960 * no other choice. 1961 */ 1962#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1963 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 1964#undef USERTT 1965 1966 if (tcp_inflight_debug > 0) { 1967 static int ltime; 1968 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 1969 ltime = ticks; 1970 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 1971 tp, 1972 bw, 1973 tp->t_rttbest, 1974 tp->t_srtt, 1975 bwnd 1976 ); 1977 } 1978 } 1979 if ((long)bwnd < tcp_inflight_min) 1980 bwnd = tcp_inflight_min; 1981 if (bwnd > tcp_inflight_max) 1982 bwnd = tcp_inflight_max; 1983 if ((long)bwnd < tp->t_maxseg * 2) 1984 bwnd = tp->t_maxseg * 2; 1985 tp->snd_bwnd = bwnd; 1986} 1987 1988#ifdef TCP_SIGNATURE 1989/* 1990 * Callback function invoked by m_apply() to digest TCP segment data 1991 * contained within an mbuf chain. 1992 */ 1993static int 1994tcp_signature_apply(void *fstate, void *data, u_int len) 1995{ 1996 1997 MD5Update(fstate, (u_char *)data, len); 1998 return (0); 1999} 2000 2001/* 2002 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2003 * 2004 * Parameters: 2005 * m pointer to head of mbuf chain 2006 * off0 offset to TCP header within the mbuf chain 2007 * len length of TCP segment data, excluding options 2008 * optlen length of TCP segment options 2009 * buf pointer to storage for computed MD5 digest 2010 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2011 * 2012 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2013 * When called from tcp_input(), we can be sure that th_sum has been 2014 * zeroed out and verified already. 2015 * 2016 * This function is for IPv4 use only. Calling this function with an 2017 * IPv6 packet in the mbuf chain will yield undefined results. 2018 * 2019 * Return 0 if successful, otherwise return -1. 2020 * 2021 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2022 * search with the destination IP address, and a 'magic SPI' to be 2023 * determined by the application. This is hardcoded elsewhere to 1179 2024 * right now. Another branch of this code exists which uses the SPD to 2025 * specify per-application flows but it is unstable. 2026 */ 2027int 2028tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2029 u_char *buf, u_int direction) 2030{ 2031 union sockaddr_union dst; 2032 struct ippseudo ippseudo; 2033 MD5_CTX ctx; 2034 int doff; 2035 struct ip *ip; 2036 struct ipovly *ipovly; 2037 struct secasvar *sav; 2038 struct tcphdr *th; 2039 u_short savecsum; 2040 2041 KASSERT(m != NULL, ("NULL mbuf chain")); 2042 KASSERT(buf != NULL, ("NULL signature pointer")); 2043 2044 /* Extract the destination from the IP header in the mbuf. */ 2045 ip = mtod(m, struct ip *); 2046 bzero(&dst, sizeof(union sockaddr_union)); 2047 dst.sa.sa_len = sizeof(struct sockaddr_in); 2048 dst.sa.sa_family = AF_INET; 2049 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2050 ip->ip_src : ip->ip_dst; 2051 2052 /* Look up an SADB entry which matches the address of the peer. */ 2053 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2054 if (sav == NULL) { 2055 printf("%s: SADB lookup failed for %s\n", __func__, 2056 inet_ntoa(dst.sin.sin_addr)); 2057 return (EINVAL); 2058 } 2059 2060 MD5Init(&ctx); 2061 ipovly = (struct ipovly *)ip; 2062 th = (struct tcphdr *)((u_char *)ip + off0); 2063 doff = off0 + sizeof(struct tcphdr) + optlen; 2064 2065 /* 2066 * Step 1: Update MD5 hash with IP pseudo-header. 2067 * 2068 * XXX The ippseudo header MUST be digested in network byte order, 2069 * or else we'll fail the regression test. Assume all fields we've 2070 * been doing arithmetic on have been in host byte order. 2071 * XXX One cannot depend on ipovly->ih_len here. When called from 2072 * tcp_output(), the underlying ip_len member has not yet been set. 2073 */ 2074 ippseudo.ippseudo_src = ipovly->ih_src; 2075 ippseudo.ippseudo_dst = ipovly->ih_dst; 2076 ippseudo.ippseudo_pad = 0; 2077 ippseudo.ippseudo_p = IPPROTO_TCP; 2078 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2079 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2080 2081 /* 2082 * Step 2: Update MD5 hash with TCP header, excluding options. 2083 * The TCP checksum must be set to zero. 2084 */ 2085 savecsum = th->th_sum; 2086 th->th_sum = 0; 2087 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2088 th->th_sum = savecsum; 2089 2090 /* 2091 * Step 3: Update MD5 hash with TCP segment data. 2092 * Use m_apply() to avoid an early m_pullup(). 2093 */ 2094 if (len > 0) 2095 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2096 2097 /* 2098 * Step 4: Update MD5 hash with shared secret. 2099 */ 2100 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2101 MD5Final(buf, &ctx); 2102 2103 key_sa_recordxfer(sav, m); 2104 KEY_FREESAV(&sav); 2105 return (0); 2106} 2107#endif /* TCP_SIGNATURE */ 2108