tcp_timewait.c revision 148385
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 148385 2005-07-25 12:31:43Z ume $ 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/scope6_var.h> 78#include <netinet6/nd6.h> 79#endif 80#include <netinet/ip_icmp.h> 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#include <netkey/key.h> 101#endif /*IPSEC*/ 102 103#ifdef FAST_IPSEC 104#include <netipsec/ipsec.h> 105#include <netipsec/xform.h> 106#ifdef INET6 107#include <netipsec/ipsec6.h> 108#endif 109#include <netipsec/key.h> 110#define IPSEC 111#endif /*FAST_IPSEC*/ 112 113#include <machine/in_cksum.h> 114#include <sys/md5.h> 115 116int tcp_mssdflt = TCP_MSS; 117SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 118 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 119 120#ifdef INET6 121int tcp_v6mssdflt = TCP6_MSS; 122SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 123 CTLFLAG_RW, &tcp_v6mssdflt , 0, 124 "Default TCP Maximum Segment Size for IPv6"); 125#endif 126 127/* 128 * Minimum MSS we accept and use. This prevents DoS attacks where 129 * we are forced to a ridiculous low MSS like 20 and send hundreds 130 * of packets instead of one. The effect scales with the available 131 * bandwidth and quickly saturates the CPU and network interface 132 * with packet generation and sending. Set to zero to disable MINMSS 133 * checking. This setting prevents us from sending too small packets. 134 */ 135int tcp_minmss = TCP_MINMSS; 136SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 137 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 138/* 139 * Number of TCP segments per second we accept from remote host 140 * before we start to calculate average segment size. If average 141 * segment size drops below the minimum TCP MSS we assume a DoS 142 * attack and reset+drop the connection. Care has to be taken not to 143 * set this value too small to not kill interactive type connections 144 * (telnet, SSH) which send many small packets. 145 */ 146int tcp_minmssoverload = TCP_MINMSSOVERLOAD; 147SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, 148 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" 149 "be under the MINMSS Size"); 150 151#if 0 152static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 153SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 154 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 155#endif 156 157int tcp_do_rfc1323 = 1; 158SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 159 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 160 161static int tcp_tcbhashsize = 0; 162SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 163 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 164 165static int do_tcpdrain = 1; 166SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 167 "Enable tcp_drain routine for extra help when low on mbufs"); 168 169SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 170 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 171 172static int icmp_may_rst = 1; 173SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 174 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 175 176static int tcp_isn_reseed_interval = 0; 177SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 178 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 179 180/* 181 * TCP bandwidth limiting sysctls. Note that the default lower bound of 182 * 1024 exists only for debugging. A good production default would be 183 * something like 6100. 184 */ 185SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 186 "TCP inflight data limiting"); 187 188static int tcp_inflight_enable = 1; 189SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 190 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 191 192static int tcp_inflight_debug = 0; 193SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 194 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 195 196static int tcp_inflight_min = 6144; 197SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 198 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 199 200static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 201SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 202 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 203 204static int tcp_inflight_stab = 20; 205SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 206 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 207 208uma_zone_t sack_hole_zone; 209 210static struct inpcb *tcp_notify(struct inpcb *, int); 211static void tcp_discardcb(struct tcpcb *); 212static void tcp_isn_tick(void *); 213 214/* 215 * Target size of TCP PCB hash tables. Must be a power of two. 216 * 217 * Note that this can be overridden by the kernel environment 218 * variable net.inet.tcp.tcbhashsize 219 */ 220#ifndef TCBHASHSIZE 221#define TCBHASHSIZE 512 222#endif 223 224/* 225 * XXX 226 * Callouts should be moved into struct tcp directly. They are currently 227 * separate because the tcpcb structure is exported to userland for sysctl 228 * parsing purposes, which do not know about callouts. 229 */ 230struct tcpcb_mem { 231 struct tcpcb tcb; 232 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 233 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 234}; 235 236static uma_zone_t tcpcb_zone; 237static uma_zone_t tcptw_zone; 238struct callout isn_callout; 239 240/* 241 * Tcp initialization 242 */ 243void 244tcp_init() 245{ 246 int hashsize = TCBHASHSIZE; 247 248 tcp_delacktime = TCPTV_DELACK; 249 tcp_keepinit = TCPTV_KEEP_INIT; 250 tcp_keepidle = TCPTV_KEEP_IDLE; 251 tcp_keepintvl = TCPTV_KEEPINTVL; 252 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 253 tcp_msl = TCPTV_MSL; 254 tcp_rexmit_min = TCPTV_MIN; 255 tcp_rexmit_slop = TCPTV_CPU_VAR; 256 257 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 258 LIST_INIT(&tcb); 259 tcbinfo.listhead = &tcb; 260 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 261 if (!powerof2(hashsize)) { 262 printf("WARNING: TCB hash size not a power of 2\n"); 263 hashsize = 512; /* safe default */ 264 } 265 tcp_tcbhashsize = hashsize; 266 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 267 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 268 &tcbinfo.porthashmask); 269 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 270 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 271 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 272#ifdef INET6 273#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 274#else /* INET6 */ 275#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 276#endif /* INET6 */ 277 if (max_protohdr < TCP_MINPROTOHDR) 278 max_protohdr = TCP_MINPROTOHDR; 279 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 280 panic("tcp_init"); 281#undef TCP_MINPROTOHDR 282 /* 283 * These have to be type stable for the benefit of the timers. 284 */ 285 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 286 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 287 uma_zone_set_max(tcpcb_zone, maxsockets); 288 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 289 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 290 uma_zone_set_max(tcptw_zone, maxsockets / 5); 291 tcp_timer_init(); 292 syncache_init(); 293 tcp_hc_init(); 294 tcp_reass_init(); 295 callout_init(&isn_callout, CALLOUT_MPSAFE); 296 tcp_isn_tick(NULL); 297 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 298 SHUTDOWN_PRI_DEFAULT); 299 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 300 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 301} 302 303void 304tcp_fini(xtp) 305 void *xtp; 306{ 307 callout_stop(&isn_callout); 308 309} 310 311/* 312 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 313 * tcp_template used to store this data in mbufs, but we now recopy it out 314 * of the tcpcb each time to conserve mbufs. 315 */ 316void 317tcpip_fillheaders(inp, ip_ptr, tcp_ptr) 318 struct inpcb *inp; 319 void *ip_ptr; 320 void *tcp_ptr; 321{ 322 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 323 324 INP_LOCK_ASSERT(inp); 325 326#ifdef INET6 327 if ((inp->inp_vflag & INP_IPV6) != 0) { 328 struct ip6_hdr *ip6; 329 330 ip6 = (struct ip6_hdr *)ip_ptr; 331 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 332 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 333 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 334 (IPV6_VERSION & IPV6_VERSION_MASK); 335 ip6->ip6_nxt = IPPROTO_TCP; 336 ip6->ip6_plen = sizeof(struct tcphdr); 337 ip6->ip6_src = inp->in6p_laddr; 338 ip6->ip6_dst = inp->in6p_faddr; 339 } else 340#endif 341 { 342 struct ip *ip; 343 344 ip = (struct ip *)ip_ptr; 345 ip->ip_v = IPVERSION; 346 ip->ip_hl = 5; 347 ip->ip_tos = inp->inp_ip_tos; 348 ip->ip_len = 0; 349 ip->ip_id = 0; 350 ip->ip_off = 0; 351 ip->ip_ttl = inp->inp_ip_ttl; 352 ip->ip_sum = 0; 353 ip->ip_p = IPPROTO_TCP; 354 ip->ip_src = inp->inp_laddr; 355 ip->ip_dst = inp->inp_faddr; 356 } 357 th->th_sport = inp->inp_lport; 358 th->th_dport = inp->inp_fport; 359 th->th_seq = 0; 360 th->th_ack = 0; 361 th->th_x2 = 0; 362 th->th_off = 5; 363 th->th_flags = 0; 364 th->th_win = 0; 365 th->th_urp = 0; 366 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 367} 368 369/* 370 * Create template to be used to send tcp packets on a connection. 371 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 372 * use for this function is in keepalives, which use tcp_respond. 373 */ 374struct tcptemp * 375tcpip_maketemplate(inp) 376 struct inpcb *inp; 377{ 378 struct mbuf *m; 379 struct tcptemp *n; 380 381 m = m_get(M_DONTWAIT, MT_HEADER); 382 if (m == NULL) 383 return (0); 384 m->m_len = sizeof(struct tcptemp); 385 n = mtod(m, struct tcptemp *); 386 387 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 388 return (n); 389} 390 391/* 392 * Send a single message to the TCP at address specified by 393 * the given TCP/IP header. If m == NULL, then we make a copy 394 * of the tcpiphdr at ti and send directly to the addressed host. 395 * This is used to force keep alive messages out using the TCP 396 * template for a connection. If flags are given then we send 397 * a message back to the TCP which originated the * segment ti, 398 * and discard the mbuf containing it and any other attached mbufs. 399 * 400 * In any case the ack and sequence number of the transmitted 401 * segment are as specified by the parameters. 402 * 403 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 404 */ 405void 406tcp_respond(tp, ipgen, th, m, ack, seq, flags) 407 struct tcpcb *tp; 408 void *ipgen; 409 register struct tcphdr *th; 410 register struct mbuf *m; 411 tcp_seq ack, seq; 412 int flags; 413{ 414 register int tlen; 415 int win = 0; 416 struct ip *ip; 417 struct tcphdr *nth; 418#ifdef INET6 419 struct ip6_hdr *ip6; 420 int isipv6; 421#endif /* INET6 */ 422 int ipflags = 0; 423 struct inpcb *inp; 424 425 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 426 427#ifdef INET6 428 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 429 ip6 = ipgen; 430#endif /* INET6 */ 431 ip = ipgen; 432 433 if (tp != NULL) { 434 inp = tp->t_inpcb; 435 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 436 INP_INFO_WLOCK_ASSERT(&tcbinfo); 437 INP_LOCK_ASSERT(inp); 438 } else 439 inp = NULL; 440 441 if (tp != NULL) { 442 if (!(flags & TH_RST)) { 443 win = sbspace(&inp->inp_socket->so_rcv); 444 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 445 win = (long)TCP_MAXWIN << tp->rcv_scale; 446 } 447 } 448 if (m == NULL) { 449 m = m_gethdr(M_DONTWAIT, MT_HEADER); 450 if (m == NULL) 451 return; 452 tlen = 0; 453 m->m_data += max_linkhdr; 454#ifdef INET6 455 if (isipv6) { 456 bcopy((caddr_t)ip6, mtod(m, caddr_t), 457 sizeof(struct ip6_hdr)); 458 ip6 = mtod(m, struct ip6_hdr *); 459 nth = (struct tcphdr *)(ip6 + 1); 460 } else 461#endif /* INET6 */ 462 { 463 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 464 ip = mtod(m, struct ip *); 465 nth = (struct tcphdr *)(ip + 1); 466 } 467 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 468 flags = TH_ACK; 469 } else { 470 m_freem(m->m_next); 471 m->m_next = NULL; 472 m->m_data = (caddr_t)ipgen; 473 /* m_len is set later */ 474 tlen = 0; 475#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 476#ifdef INET6 477 if (isipv6) { 478 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 479 nth = (struct tcphdr *)(ip6 + 1); 480 } else 481#endif /* INET6 */ 482 { 483 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 484 nth = (struct tcphdr *)(ip + 1); 485 } 486 if (th != nth) { 487 /* 488 * this is usually a case when an extension header 489 * exists between the IPv6 header and the 490 * TCP header. 491 */ 492 nth->th_sport = th->th_sport; 493 nth->th_dport = th->th_dport; 494 } 495 xchg(nth->th_dport, nth->th_sport, n_short); 496#undef xchg 497 } 498#ifdef INET6 499 if (isipv6) { 500 ip6->ip6_flow = 0; 501 ip6->ip6_vfc = IPV6_VERSION; 502 ip6->ip6_nxt = IPPROTO_TCP; 503 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 504 tlen)); 505 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 506 } else 507#endif 508 { 509 tlen += sizeof (struct tcpiphdr); 510 ip->ip_len = tlen; 511 ip->ip_ttl = ip_defttl; 512 if (path_mtu_discovery) 513 ip->ip_off |= IP_DF; 514 } 515 m->m_len = tlen; 516 m->m_pkthdr.len = tlen; 517 m->m_pkthdr.rcvif = NULL; 518#ifdef MAC 519 if (inp != NULL) { 520 /* 521 * Packet is associated with a socket, so allow the 522 * label of the response to reflect the socket label. 523 */ 524 INP_LOCK_ASSERT(inp); 525 mac_create_mbuf_from_inpcb(inp, m); 526 } else { 527 /* 528 * Packet is not associated with a socket, so possibly 529 * update the label in place. 530 */ 531 mac_reflect_mbuf_tcp(m); 532 } 533#endif 534 nth->th_seq = htonl(seq); 535 nth->th_ack = htonl(ack); 536 nth->th_x2 = 0; 537 nth->th_off = sizeof (struct tcphdr) >> 2; 538 nth->th_flags = flags; 539 if (tp != NULL) 540 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 541 else 542 nth->th_win = htons((u_short)win); 543 nth->th_urp = 0; 544#ifdef INET6 545 if (isipv6) { 546 nth->th_sum = 0; 547 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 548 sizeof(struct ip6_hdr), 549 tlen - sizeof(struct ip6_hdr)); 550 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 551 NULL, NULL); 552 } else 553#endif /* INET6 */ 554 { 555 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 556 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 557 m->m_pkthdr.csum_flags = CSUM_TCP; 558 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 559 } 560#ifdef TCPDEBUG 561 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 562 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 563#endif 564#ifdef INET6 565 if (isipv6) 566 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 567 else 568#endif /* INET6 */ 569 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 570} 571 572/* 573 * Create a new TCP control block, making an 574 * empty reassembly queue and hooking it to the argument 575 * protocol control block. The `inp' parameter must have 576 * come from the zone allocator set up in tcp_init(). 577 */ 578struct tcpcb * 579tcp_newtcpcb(inp) 580 struct inpcb *inp; 581{ 582 struct tcpcb_mem *tm; 583 struct tcpcb *tp; 584#ifdef INET6 585 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 586#endif /* INET6 */ 587 588 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 589 if (tm == NULL) 590 return (NULL); 591 tp = &tm->tcb; 592 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 593 tp->t_maxseg = tp->t_maxopd = 594#ifdef INET6 595 isipv6 ? tcp_v6mssdflt : 596#endif /* INET6 */ 597 tcp_mssdflt; 598 599 /* Set up our timeouts. */ 600 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 601 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 602 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 603 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 604 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 605 606 if (tcp_do_rfc1323) 607 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 608 tp->sack_enable = tcp_do_sack; 609 TAILQ_INIT(&tp->snd_holes); 610 tp->t_inpcb = inp; /* XXX */ 611 /* 612 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 613 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 614 * reasonable initial retransmit time. 615 */ 616 tp->t_srtt = TCPTV_SRTTBASE; 617 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 618 tp->t_rttmin = tcp_rexmit_min; 619 tp->t_rxtcur = TCPTV_RTOBASE; 620 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 621 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 622 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 623 tp->t_rcvtime = ticks; 624 tp->t_bw_rtttime = ticks; 625 /* 626 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 627 * because the socket may be bound to an IPv6 wildcard address, 628 * which may match an IPv4-mapped IPv6 address. 629 */ 630 inp->inp_ip_ttl = ip_defttl; 631 inp->inp_ppcb = (caddr_t)tp; 632 return (tp); /* XXX */ 633} 634 635/* 636 * Drop a TCP connection, reporting 637 * the specified error. If connection is synchronized, 638 * then send a RST to peer. 639 */ 640struct tcpcb * 641tcp_drop(tp, errno) 642 register struct tcpcb *tp; 643 int errno; 644{ 645 struct socket *so = tp->t_inpcb->inp_socket; 646 647 INP_INFO_WLOCK_ASSERT(&tcbinfo); 648 INP_LOCK_ASSERT(tp->t_inpcb); 649 650 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 INP_LOCK_ASSERT(inp); 674 675 /* 676 * Make sure that all of our timers are stopped before we 677 * delete the PCB. 678 */ 679 callout_stop(tp->tt_rexmt); 680 callout_stop(tp->tt_persist); 681 callout_stop(tp->tt_keep); 682 callout_stop(tp->tt_2msl); 683 callout_stop(tp->tt_delack); 684 685 /* 686 * If we got enough samples through the srtt filter, 687 * save the rtt and rttvar in the routing entry. 688 * 'Enough' is arbitrarily defined as 4 rtt samples. 689 * 4 samples is enough for the srtt filter to converge 690 * to within enough % of the correct value; fewer samples 691 * and we could save a bogus rtt. The danger is not high 692 * as tcp quickly recovers from everything. 693 * XXX: Works very well but needs some more statistics! 694 */ 695 if (tp->t_rttupdated >= 4) { 696 struct hc_metrics_lite metrics; 697 u_long ssthresh; 698 699 bzero(&metrics, sizeof(metrics)); 700 /* 701 * Update the ssthresh always when the conditions below 702 * are satisfied. This gives us better new start value 703 * for the congestion avoidance for new connections. 704 * ssthresh is only set if packet loss occured on a session. 705 */ 706 ssthresh = tp->snd_ssthresh; 707 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 708 /* 709 * convert the limit from user data bytes to 710 * packets then to packet data bytes. 711 */ 712 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 713 if (ssthresh < 2) 714 ssthresh = 2; 715 ssthresh *= (u_long)(tp->t_maxseg + 716#ifdef INET6 717 (isipv6 ? sizeof (struct ip6_hdr) + 718 sizeof (struct tcphdr) : 719#endif 720 sizeof (struct tcpiphdr) 721#ifdef INET6 722 ) 723#endif 724 ); 725 } else 726 ssthresh = 0; 727 metrics.rmx_ssthresh = ssthresh; 728 729 metrics.rmx_rtt = tp->t_srtt; 730 metrics.rmx_rttvar = tp->t_rttvar; 731 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 732 metrics.rmx_bandwidth = tp->snd_bandwidth; 733 metrics.rmx_cwnd = tp->snd_cwnd; 734 metrics.rmx_sendpipe = 0; 735 metrics.rmx_recvpipe = 0; 736 737 tcp_hc_update(&inp->inp_inc, &metrics); 738 } 739 740 /* free the reassembly queue, if any */ 741 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 742 LIST_REMOVE(q, tqe_q); 743 m_freem(q->tqe_m); 744 uma_zfree(tcp_reass_zone, q); 745 tp->t_segqlen--; 746 tcp_reass_qsize--; 747 } 748 tcp_free_sackholes(tp); 749 inp->inp_ppcb = NULL; 750 tp->t_inpcb = NULL; 751 uma_zfree(tcpcb_zone, tp); 752 soisdisconnected(so); 753} 754 755/* 756 * Close a TCP control block: 757 * discard all space held by the tcp 758 * discard internet protocol block 759 * wake up any sleepers 760 */ 761struct tcpcb * 762tcp_close(tp) 763 struct tcpcb *tp; 764{ 765 struct inpcb *inp = tp->t_inpcb; 766#ifdef INET6 767 struct socket *so = inp->inp_socket; 768#endif 769 770 INP_INFO_WLOCK_ASSERT(&tcbinfo); 771 INP_LOCK_ASSERT(inp); 772 773 tcp_discardcb(tp); 774#ifdef INET6 775 if (INP_CHECK_SOCKAF(so, AF_INET6)) 776 in6_pcbdetach(inp); 777 else 778#endif 779 in_pcbdetach(inp); 780 tcpstat.tcps_closed++; 781 return (NULL); 782} 783 784void 785tcp_drain() 786{ 787 if (do_tcpdrain) 788 { 789 struct inpcb *inpb; 790 struct tcpcb *tcpb; 791 struct tseg_qent *te; 792 793 /* 794 * Walk the tcpbs, if existing, and flush the reassembly queue, 795 * if there is one... 796 * XXX: The "Net/3" implementation doesn't imply that the TCP 797 * reassembly queue should be flushed, but in a situation 798 * where we're really low on mbufs, this is potentially 799 * usefull. 800 */ 801 INP_INFO_RLOCK(&tcbinfo); 802 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 803 if (inpb->inp_vflag & INP_TIMEWAIT) 804 continue; 805 INP_LOCK(inpb); 806 if ((tcpb = intotcpcb(inpb)) != NULL) { 807 while ((te = LIST_FIRST(&tcpb->t_segq)) 808 != NULL) { 809 LIST_REMOVE(te, tqe_q); 810 m_freem(te->tqe_m); 811 uma_zfree(tcp_reass_zone, te); 812 tcpb->t_segqlen--; 813 tcp_reass_qsize--; 814 } 815 tcp_clean_sackreport(tcpb); 816 } 817 INP_UNLOCK(inpb); 818 } 819 INP_INFO_RUNLOCK(&tcbinfo); 820 } 821} 822 823/* 824 * Notify a tcp user of an asynchronous error; 825 * store error as soft error, but wake up user 826 * (for now, won't do anything until can select for soft error). 827 * 828 * Do not wake up user since there currently is no mechanism for 829 * reporting soft errors (yet - a kqueue filter may be added). 830 */ 831static struct inpcb * 832tcp_notify(inp, error) 833 struct inpcb *inp; 834 int error; 835{ 836 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 837 838 INP_INFO_WLOCK_ASSERT(&tcbinfo); 839 INP_LOCK_ASSERT(inp); 840 841 /* 842 * Ignore some errors if we are hooked up. 843 * If connection hasn't completed, has retransmitted several times, 844 * and receives a second error, give up now. This is better 845 * than waiting a long time to establish a connection that 846 * can never complete. 847 */ 848 if (tp->t_state == TCPS_ESTABLISHED && 849 (error == EHOSTUNREACH || error == ENETUNREACH || 850 error == EHOSTDOWN)) { 851 return (inp); 852 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 853 tp->t_softerror) { 854 tcp_drop(tp, error); 855 return (struct inpcb *)0; 856 } else { 857 tp->t_softerror = error; 858 return (inp); 859 } 860#if 0 861 wakeup( &so->so_timeo); 862 sorwakeup(so); 863 sowwakeup(so); 864#endif 865} 866 867static int 868tcp_pcblist(SYSCTL_HANDLER_ARGS) 869{ 870 int error, i, n; 871 struct inpcb *inp, **inp_list; 872 inp_gen_t gencnt; 873 struct xinpgen xig; 874 875 /* 876 * The process of preparing the TCB list is too time-consuming and 877 * resource-intensive to repeat twice on every request. 878 */ 879 if (req->oldptr == NULL) { 880 n = tcbinfo.ipi_count; 881 req->oldidx = 2 * (sizeof xig) 882 + (n + n/8) * sizeof(struct xtcpcb); 883 return (0); 884 } 885 886 if (req->newptr != NULL) 887 return (EPERM); 888 889 /* 890 * OK, now we're committed to doing something. 891 */ 892 INP_INFO_RLOCK(&tcbinfo); 893 gencnt = tcbinfo.ipi_gencnt; 894 n = tcbinfo.ipi_count; 895 INP_INFO_RUNLOCK(&tcbinfo); 896 897 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 898 + n * sizeof(struct xtcpcb)); 899 if (error != 0) 900 return (error); 901 902 xig.xig_len = sizeof xig; 903 xig.xig_count = n; 904 xig.xig_gen = gencnt; 905 xig.xig_sogen = so_gencnt; 906 error = SYSCTL_OUT(req, &xig, sizeof xig); 907 if (error) 908 return (error); 909 910 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 911 if (inp_list == NULL) 912 return (ENOMEM); 913 914 INP_INFO_RLOCK(&tcbinfo); 915 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 916 inp = LIST_NEXT(inp, inp_list)) { 917 INP_LOCK(inp); 918 if (inp->inp_gencnt <= gencnt) { 919 /* 920 * XXX: This use of cr_cansee(), introduced with 921 * TCP state changes, is not quite right, but for 922 * now, better than nothing. 923 */ 924 if (inp->inp_vflag & INP_TIMEWAIT) 925 error = cr_cansee(req->td->td_ucred, 926 intotw(inp)->tw_cred); 927 else 928 error = cr_canseesocket(req->td->td_ucred, 929 inp->inp_socket); 930 if (error == 0) 931 inp_list[i++] = inp; 932 } 933 INP_UNLOCK(inp); 934 } 935 INP_INFO_RUNLOCK(&tcbinfo); 936 n = i; 937 938 error = 0; 939 for (i = 0; i < n; i++) { 940 inp = inp_list[i]; 941 if (inp->inp_gencnt <= gencnt) { 942 struct xtcpcb xt; 943 caddr_t inp_ppcb; 944 945 bzero(&xt, sizeof(xt)); 946 xt.xt_len = sizeof xt; 947 /* XXX should avoid extra copy */ 948 bcopy(inp, &xt.xt_inp, sizeof *inp); 949 inp_ppcb = inp->inp_ppcb; 950 if (inp_ppcb == NULL) 951 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 952 else if (inp->inp_vflag & INP_TIMEWAIT) { 953 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 954 xt.xt_tp.t_state = TCPS_TIME_WAIT; 955 } else 956 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 957 if (inp->inp_socket != NULL) 958 sotoxsocket(inp->inp_socket, &xt.xt_socket); 959 else { 960 bzero(&xt.xt_socket, sizeof xt.xt_socket); 961 xt.xt_socket.xso_protocol = IPPROTO_TCP; 962 } 963 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 964 error = SYSCTL_OUT(req, &xt, sizeof xt); 965 } 966 } 967 if (!error) { 968 /* 969 * Give the user an updated idea of our state. 970 * If the generation differs from what we told 971 * her before, she knows that something happened 972 * while we were processing this request, and it 973 * might be necessary to retry. 974 */ 975 INP_INFO_RLOCK(&tcbinfo); 976 xig.xig_gen = tcbinfo.ipi_gencnt; 977 xig.xig_sogen = so_gencnt; 978 xig.xig_count = tcbinfo.ipi_count; 979 INP_INFO_RUNLOCK(&tcbinfo); 980 error = SYSCTL_OUT(req, &xig, sizeof xig); 981 } 982 free(inp_list, M_TEMP); 983 return (error); 984} 985 986SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 987 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 988 989static int 990tcp_getcred(SYSCTL_HANDLER_ARGS) 991{ 992 struct xucred xuc; 993 struct sockaddr_in addrs[2]; 994 struct inpcb *inp; 995 int error; 996 997 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 998 if (error) 999 return (error); 1000 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1001 if (error) 1002 return (error); 1003 INP_INFO_RLOCK(&tcbinfo); 1004 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1005 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1006 if (inp == NULL) { 1007 error = ENOENT; 1008 goto outunlocked; 1009 } 1010 INP_LOCK(inp); 1011 if (inp->inp_socket == NULL) { 1012 error = ENOENT; 1013 goto out; 1014 } 1015 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1016 if (error) 1017 goto out; 1018 cru2x(inp->inp_socket->so_cred, &xuc); 1019out: 1020 INP_UNLOCK(inp); 1021outunlocked: 1022 INP_INFO_RUNLOCK(&tcbinfo); 1023 if (error == 0) 1024 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1025 return (error); 1026} 1027 1028SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1029 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1030 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1031 1032#ifdef INET6 1033static int 1034tcp6_getcred(SYSCTL_HANDLER_ARGS) 1035{ 1036 struct xucred xuc; 1037 struct sockaddr_in6 addrs[2]; 1038 struct in6_addr a6[2]; 1039 struct inpcb *inp; 1040 int error, mapped = 0; 1041 1042 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1043 if (error) 1044 return (error); 1045 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1046 if (error) 1047 return (error); 1048 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1049 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1050 return (error); 1051 } 1052 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1053 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1054 mapped = 1; 1055 else 1056 return (EINVAL); 1057 } 1058 1059 INP_INFO_RLOCK(&tcbinfo); 1060 if (mapped == 1) 1061 inp = in_pcblookup_hash(&tcbinfo, 1062 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1063 addrs[1].sin6_port, 1064 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1065 addrs[0].sin6_port, 1066 0, NULL); 1067 else 1068 inp = in6_pcblookup_hash(&tcbinfo, &a6[1], addrs[1].sin6_port, 1069 &a6[0], addrs[0].sin6_port, 0, NULL); 1070 if (inp == NULL) { 1071 error = ENOENT; 1072 goto outunlocked; 1073 } 1074 INP_LOCK(inp); 1075 if (inp->inp_socket == NULL) { 1076 error = ENOENT; 1077 goto out; 1078 } 1079 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1080 if (error) 1081 goto out; 1082 cru2x(inp->inp_socket->so_cred, &xuc); 1083out: 1084 INP_UNLOCK(inp); 1085outunlocked: 1086 INP_INFO_RUNLOCK(&tcbinfo); 1087 if (error == 0) 1088 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1089 return (error); 1090} 1091 1092SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1093 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1094 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1095#endif 1096 1097 1098void 1099tcp_ctlinput(cmd, sa, vip) 1100 int cmd; 1101 struct sockaddr *sa; 1102 void *vip; 1103{ 1104 struct ip *ip = vip; 1105 struct tcphdr *th; 1106 struct in_addr faddr; 1107 struct inpcb *inp; 1108 struct tcpcb *tp; 1109 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1110 struct icmp *icp; 1111 struct in_conninfo inc; 1112 tcp_seq icmp_tcp_seq; 1113 int mtu; 1114 1115 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1116 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1117 return; 1118 1119 if (cmd == PRC_MSGSIZE) 1120 notify = tcp_mtudisc; 1121 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1122 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1123 notify = tcp_drop_syn_sent; 1124 /* 1125 * Redirects don't need to be handled up here. 1126 */ 1127 else if (PRC_IS_REDIRECT(cmd)) 1128 return; 1129 /* 1130 * Source quench is depreciated. 1131 */ 1132 else if (cmd == PRC_QUENCH) 1133 return; 1134 /* 1135 * Hostdead is ugly because it goes linearly through all PCBs. 1136 * XXX: We never get this from ICMP, otherwise it makes an 1137 * excellent DoS attack on machines with many connections. 1138 */ 1139 else if (cmd == PRC_HOSTDEAD) 1140 ip = NULL; 1141 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1142 return; 1143 if (ip != NULL) { 1144 icp = (struct icmp *)((caddr_t)ip 1145 - offsetof(struct icmp, icmp_ip)); 1146 th = (struct tcphdr *)((caddr_t)ip 1147 + (ip->ip_hl << 2)); 1148 INP_INFO_WLOCK(&tcbinfo); 1149 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1150 ip->ip_src, th->th_sport, 0, NULL); 1151 if (inp != NULL) { 1152 INP_LOCK(inp); 1153 if (inp->inp_socket != NULL) { 1154 icmp_tcp_seq = htonl(th->th_seq); 1155 tp = intotcpcb(inp); 1156 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1157 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1158 if (cmd == PRC_MSGSIZE) { 1159 /* 1160 * MTU discovery: 1161 * If we got a needfrag set the MTU 1162 * in the route to the suggested new 1163 * value (if given) and then notify. 1164 */ 1165 bzero(&inc, sizeof(inc)); 1166 inc.inc_flags = 0; /* IPv4 */ 1167 inc.inc_faddr = faddr; 1168 1169 mtu = ntohs(icp->icmp_nextmtu); 1170 /* 1171 * If no alternative MTU was 1172 * proposed, try the next smaller 1173 * one. 1174 */ 1175 if (!mtu) 1176 mtu = ip_next_mtu(ntohs(ip->ip_len), 1177 1); 1178 if (mtu < max(296, (tcp_minmss) 1179 + sizeof(struct tcpiphdr))) 1180 mtu = 0; 1181 if (!mtu) 1182 mtu = tcp_mssdflt 1183 + sizeof(struct tcpiphdr); 1184 /* 1185 * Only cache the the MTU if it 1186 * is smaller than the interface 1187 * or route MTU. tcp_mtudisc() 1188 * will do right thing by itself. 1189 */ 1190 if (mtu <= tcp_maxmtu(&inc)) 1191 tcp_hc_updatemtu(&inc, mtu); 1192 } 1193 1194 inp = (*notify)(inp, inetctlerrmap[cmd]); 1195 } 1196 } 1197 if (inp != NULL) 1198 INP_UNLOCK(inp); 1199 } else { 1200 inc.inc_fport = th->th_dport; 1201 inc.inc_lport = th->th_sport; 1202 inc.inc_faddr = faddr; 1203 inc.inc_laddr = ip->ip_src; 1204#ifdef INET6 1205 inc.inc_isipv6 = 0; 1206#endif 1207 syncache_unreach(&inc, th); 1208 } 1209 INP_INFO_WUNLOCK(&tcbinfo); 1210 } else 1211 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1212} 1213 1214#ifdef INET6 1215void 1216tcp6_ctlinput(cmd, sa, d) 1217 int cmd; 1218 struct sockaddr *sa; 1219 void *d; 1220{ 1221 struct tcphdr th; 1222 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1223 struct ip6_hdr *ip6; 1224 struct mbuf *m; 1225 struct ip6ctlparam *ip6cp = NULL; 1226 const struct sockaddr_in6 *sa6_src = NULL; 1227 int off; 1228 struct tcp_portonly { 1229 u_int16_t th_sport; 1230 u_int16_t th_dport; 1231 } *thp; 1232 1233 if (sa->sa_family != AF_INET6 || 1234 sa->sa_len != sizeof(struct sockaddr_in6)) 1235 return; 1236 1237 if (cmd == PRC_MSGSIZE) 1238 notify = tcp_mtudisc; 1239 else if (!PRC_IS_REDIRECT(cmd) && 1240 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1241 return; 1242 /* Source quench is depreciated. */ 1243 else if (cmd == PRC_QUENCH) 1244 return; 1245 1246 /* if the parameter is from icmp6, decode it. */ 1247 if (d != NULL) { 1248 ip6cp = (struct ip6ctlparam *)d; 1249 m = ip6cp->ip6c_m; 1250 ip6 = ip6cp->ip6c_ip6; 1251 off = ip6cp->ip6c_off; 1252 sa6_src = ip6cp->ip6c_src; 1253 } else { 1254 m = NULL; 1255 ip6 = NULL; 1256 off = 0; /* fool gcc */ 1257 sa6_src = &sa6_any; 1258 } 1259 1260 if (ip6 != NULL) { 1261 struct in_conninfo inc; 1262 /* 1263 * XXX: We assume that when IPV6 is non NULL, 1264 * M and OFF are valid. 1265 */ 1266 1267 /* check if we can safely examine src and dst ports */ 1268 if (m->m_pkthdr.len < off + sizeof(*thp)) 1269 return; 1270 1271 bzero(&th, sizeof(th)); 1272 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1273 1274 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1275 (struct sockaddr *)ip6cp->ip6c_src, 1276 th.th_sport, cmd, NULL, notify); 1277 1278 inc.inc_fport = th.th_dport; 1279 inc.inc_lport = th.th_sport; 1280 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1281 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1282 inc.inc_isipv6 = 1; 1283 INP_INFO_WLOCK(&tcbinfo); 1284 syncache_unreach(&inc, &th); 1285 INP_INFO_WUNLOCK(&tcbinfo); 1286 } else 1287 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1288 0, cmd, NULL, notify); 1289} 1290#endif /* INET6 */ 1291 1292 1293/* 1294 * Following is where TCP initial sequence number generation occurs. 1295 * 1296 * There are two places where we must use initial sequence numbers: 1297 * 1. In SYN-ACK packets. 1298 * 2. In SYN packets. 1299 * 1300 * All ISNs for SYN-ACK packets are generated by the syncache. See 1301 * tcp_syncache.c for details. 1302 * 1303 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1304 * depends on this property. In addition, these ISNs should be 1305 * unguessable so as to prevent connection hijacking. To satisfy 1306 * the requirements of this situation, the algorithm outlined in 1307 * RFC 1948 is used, with only small modifications. 1308 * 1309 * Implementation details: 1310 * 1311 * Time is based off the system timer, and is corrected so that it 1312 * increases by one megabyte per second. This allows for proper 1313 * recycling on high speed LANs while still leaving over an hour 1314 * before rollover. 1315 * 1316 * As reading the *exact* system time is too expensive to be done 1317 * whenever setting up a TCP connection, we increment the time 1318 * offset in two ways. First, a small random positive increment 1319 * is added to isn_offset for each connection that is set up. 1320 * Second, the function tcp_isn_tick fires once per clock tick 1321 * and increments isn_offset as necessary so that sequence numbers 1322 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1323 * random positive increments serve only to ensure that the same 1324 * exact sequence number is never sent out twice (as could otherwise 1325 * happen when a port is recycled in less than the system tick 1326 * interval.) 1327 * 1328 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1329 * between seeding of isn_secret. This is normally set to zero, 1330 * as reseeding should not be necessary. 1331 * 1332 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1333 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1334 * general, this means holding an exclusive (write) lock. 1335 */ 1336 1337#define ISN_BYTES_PER_SECOND 1048576 1338#define ISN_STATIC_INCREMENT 4096 1339#define ISN_RANDOM_INCREMENT (4096 - 1) 1340 1341static u_char isn_secret[32]; 1342static int isn_last_reseed; 1343static u_int32_t isn_offset, isn_offset_old; 1344static MD5_CTX isn_ctx; 1345 1346tcp_seq 1347tcp_new_isn(tp) 1348 struct tcpcb *tp; 1349{ 1350 u_int32_t md5_buffer[4]; 1351 tcp_seq new_isn; 1352 1353 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1354 INP_LOCK_ASSERT(tp->t_inpcb); 1355 1356 /* Seed if this is the first use, reseed if requested. */ 1357 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1358 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1359 < (u_int)ticks))) { 1360 read_random(&isn_secret, sizeof(isn_secret)); 1361 isn_last_reseed = ticks; 1362 } 1363 1364 /* Compute the md5 hash and return the ISN. */ 1365 MD5Init(&isn_ctx); 1366 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1367 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1368#ifdef INET6 1369 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1370 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1371 sizeof(struct in6_addr)); 1372 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1373 sizeof(struct in6_addr)); 1374 } else 1375#endif 1376 { 1377 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1378 sizeof(struct in_addr)); 1379 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1380 sizeof(struct in_addr)); 1381 } 1382 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1383 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1384 new_isn = (tcp_seq) md5_buffer[0]; 1385 isn_offset += ISN_STATIC_INCREMENT + 1386 (arc4random() & ISN_RANDOM_INCREMENT); 1387 new_isn += isn_offset; 1388 return (new_isn); 1389} 1390 1391/* 1392 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1393 * to keep time flowing at a relatively constant rate. If the random 1394 * increments have already pushed us past the projected offset, do nothing. 1395 */ 1396static void 1397tcp_isn_tick(xtp) 1398 void *xtp; 1399{ 1400 u_int32_t projected_offset; 1401 1402 INP_INFO_WLOCK(&tcbinfo); 1403 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1404 1405 if (projected_offset > isn_offset) 1406 isn_offset = projected_offset; 1407 1408 isn_offset_old = isn_offset; 1409 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1410 INP_INFO_WUNLOCK(&tcbinfo); 1411} 1412 1413/* 1414 * When a specific ICMP unreachable message is received and the 1415 * connection state is SYN-SENT, drop the connection. This behavior 1416 * is controlled by the icmp_may_rst sysctl. 1417 */ 1418struct inpcb * 1419tcp_drop_syn_sent(inp, errno) 1420 struct inpcb *inp; 1421 int errno; 1422{ 1423 struct tcpcb *tp = intotcpcb(inp); 1424 1425 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1426 INP_LOCK_ASSERT(inp); 1427 1428 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) { 1429 tcp_drop(tp, errno); 1430 return (NULL); 1431 } 1432 return (inp); 1433} 1434 1435/* 1436 * When `need fragmentation' ICMP is received, update our idea of the MSS 1437 * based on the new value in the route. Also nudge TCP to send something, 1438 * since we know the packet we just sent was dropped. 1439 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1440 */ 1441struct inpcb * 1442tcp_mtudisc(inp, errno) 1443 struct inpcb *inp; 1444 int errno; 1445{ 1446 struct tcpcb *tp = intotcpcb(inp); 1447 struct socket *so = inp->inp_socket; 1448 u_int maxmtu; 1449 u_int romtu; 1450 int mss; 1451#ifdef INET6 1452 int isipv6; 1453#endif /* INET6 */ 1454 1455 INP_LOCK_ASSERT(inp); 1456 if (tp != NULL) { 1457#ifdef INET6 1458 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1459#endif 1460 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1461 romtu = 1462#ifdef INET6 1463 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1464#endif /* INET6 */ 1465 tcp_maxmtu(&inp->inp_inc); 1466 if (!maxmtu) 1467 maxmtu = romtu; 1468 else 1469 maxmtu = min(maxmtu, romtu); 1470 if (!maxmtu) { 1471 tp->t_maxopd = tp->t_maxseg = 1472#ifdef INET6 1473 isipv6 ? tcp_v6mssdflt : 1474#endif /* INET6 */ 1475 tcp_mssdflt; 1476 return (inp); 1477 } 1478 mss = maxmtu - 1479#ifdef INET6 1480 (isipv6 ? 1481 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1482#endif /* INET6 */ 1483 sizeof(struct tcpiphdr) 1484#ifdef INET6 1485 ) 1486#endif /* INET6 */ 1487 ; 1488 1489 /* 1490 * XXX - The above conditional probably violates the TCP 1491 * spec. The problem is that, since we don't know the 1492 * other end's MSS, we are supposed to use a conservative 1493 * default. But, if we do that, then MTU discovery will 1494 * never actually take place, because the conservative 1495 * default is much less than the MTUs typically seen 1496 * on the Internet today. For the moment, we'll sweep 1497 * this under the carpet. 1498 * 1499 * The conservative default might not actually be a problem 1500 * if the only case this occurs is when sending an initial 1501 * SYN with options and data to a host we've never talked 1502 * to before. Then, they will reply with an MSS value which 1503 * will get recorded and the new parameters should get 1504 * recomputed. For Further Study. 1505 */ 1506 if (tp->t_maxopd <= mss) 1507 return (inp); 1508 tp->t_maxopd = mss; 1509 1510 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1511 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1512 mss -= TCPOLEN_TSTAMP_APPA; 1513#if (MCLBYTES & (MCLBYTES - 1)) == 0 1514 if (mss > MCLBYTES) 1515 mss &= ~(MCLBYTES-1); 1516#else 1517 if (mss > MCLBYTES) 1518 mss = mss / MCLBYTES * MCLBYTES; 1519#endif 1520 if (so->so_snd.sb_hiwat < mss) 1521 mss = so->so_snd.sb_hiwat; 1522 1523 tp->t_maxseg = mss; 1524 1525 tcpstat.tcps_mturesent++; 1526 tp->t_rtttime = 0; 1527 tp->snd_nxt = tp->snd_una; 1528 tcp_output(tp); 1529 } 1530 return (inp); 1531} 1532 1533/* 1534 * Look-up the routing entry to the peer of this inpcb. If no route 1535 * is found and it cannot be allocated, then return NULL. This routine 1536 * is called by TCP routines that access the rmx structure and by tcp_mss 1537 * to get the interface MTU. 1538 */ 1539u_long 1540tcp_maxmtu(inc) 1541 struct in_conninfo *inc; 1542{ 1543 struct route sro; 1544 struct sockaddr_in *dst; 1545 struct ifnet *ifp; 1546 u_long maxmtu = 0; 1547 1548 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1549 1550 bzero(&sro, sizeof(sro)); 1551 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1552 dst = (struct sockaddr_in *)&sro.ro_dst; 1553 dst->sin_family = AF_INET; 1554 dst->sin_len = sizeof(*dst); 1555 dst->sin_addr = inc->inc_faddr; 1556 rtalloc_ign(&sro, RTF_CLONING); 1557 } 1558 if (sro.ro_rt != NULL) { 1559 ifp = sro.ro_rt->rt_ifp; 1560 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1561 maxmtu = ifp->if_mtu; 1562 else 1563 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1564 RTFREE(sro.ro_rt); 1565 } 1566 return (maxmtu); 1567} 1568 1569#ifdef INET6 1570u_long 1571tcp_maxmtu6(inc) 1572 struct in_conninfo *inc; 1573{ 1574 struct route_in6 sro6; 1575 struct ifnet *ifp; 1576 u_long maxmtu = 0; 1577 1578 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1579 1580 bzero(&sro6, sizeof(sro6)); 1581 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1582 sro6.ro_dst.sin6_family = AF_INET6; 1583 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1584 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1585 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1586 } 1587 if (sro6.ro_rt != NULL) { 1588 ifp = sro6.ro_rt->rt_ifp; 1589 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1590 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1591 else 1592 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1593 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1594 RTFREE(sro6.ro_rt); 1595 } 1596 1597 return (maxmtu); 1598} 1599#endif /* INET6 */ 1600 1601#ifdef IPSEC 1602/* compute ESP/AH header size for TCP, including outer IP header. */ 1603size_t 1604ipsec_hdrsiz_tcp(tp) 1605 struct tcpcb *tp; 1606{ 1607 struct inpcb *inp; 1608 struct mbuf *m; 1609 size_t hdrsiz; 1610 struct ip *ip; 1611#ifdef INET6 1612 struct ip6_hdr *ip6; 1613#endif 1614 struct tcphdr *th; 1615 1616 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1617 return (0); 1618 MGETHDR(m, M_DONTWAIT, MT_DATA); 1619 if (!m) 1620 return (0); 1621 1622#ifdef INET6 1623 if ((inp->inp_vflag & INP_IPV6) != 0) { 1624 ip6 = mtod(m, struct ip6_hdr *); 1625 th = (struct tcphdr *)(ip6 + 1); 1626 m->m_pkthdr.len = m->m_len = 1627 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1628 tcpip_fillheaders(inp, ip6, th); 1629 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1630 } else 1631#endif /* INET6 */ 1632 { 1633 ip = mtod(m, struct ip *); 1634 th = (struct tcphdr *)(ip + 1); 1635 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1636 tcpip_fillheaders(inp, ip, th); 1637 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1638 } 1639 1640 m_free(m); 1641 return (hdrsiz); 1642} 1643#endif /*IPSEC*/ 1644 1645/* 1646 * Move a TCP connection into TIME_WAIT state. 1647 * tcbinfo is locked. 1648 * inp is locked, and is unlocked before returning. 1649 */ 1650void 1651tcp_twstart(tp) 1652 struct tcpcb *tp; 1653{ 1654 struct tcptw *tw; 1655 struct inpcb *inp; 1656 int tw_time, acknow; 1657 struct socket *so; 1658 1659 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1660 INP_LOCK_ASSERT(tp->t_inpcb); 1661 1662 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1663 if (tw == NULL) { 1664 tw = tcp_timer_2msl_tw(1); 1665 if (tw == NULL) { 1666 tcp_close(tp); 1667 return; 1668 } 1669 } 1670 inp = tp->t_inpcb; 1671 tw->tw_inpcb = inp; 1672 1673 /* 1674 * Recover last window size sent. 1675 */ 1676 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1677 1678 /* 1679 * Set t_recent if timestamps are used on the connection. 1680 */ 1681 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1682 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1683 tw->t_recent = tp->ts_recent; 1684 else 1685 tw->t_recent = 0; 1686 1687 tw->snd_nxt = tp->snd_nxt; 1688 tw->rcv_nxt = tp->rcv_nxt; 1689 tw->iss = tp->iss; 1690 tw->irs = tp->irs; 1691 tw->t_starttime = tp->t_starttime; 1692 tw->tw_time = 0; 1693 1694/* XXX 1695 * If this code will 1696 * be used for fin-wait-2 state also, then we may need 1697 * a ts_recent from the last segment. 1698 */ 1699 tw_time = 2 * tcp_msl; 1700 acknow = tp->t_flags & TF_ACKNOW; 1701 tcp_discardcb(tp); 1702 so = inp->inp_socket; 1703 ACCEPT_LOCK(); 1704 SOCK_LOCK(so); 1705 so->so_pcb = NULL; 1706 tw->tw_cred = crhold(so->so_cred); 1707 tw->tw_so_options = so->so_options; 1708 sotryfree(so); 1709 inp->inp_socket = NULL; 1710 if (acknow) 1711 tcp_twrespond(tw, TH_ACK); 1712 inp->inp_ppcb = (caddr_t)tw; 1713 inp->inp_vflag |= INP_TIMEWAIT; 1714 tcp_timer_2msl_reset(tw, tw_time); 1715 INP_UNLOCK(inp); 1716} 1717 1718/* 1719 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1720 * the actual rate is slightly higher due to the addition of 1721 * random positive increments. 1722 * 1723 * Most other new OSes use semi-randomized ISN values, so we 1724 * do not need to worry about them. 1725 */ 1726#define MS_ISN_BYTES_PER_SECOND 250000 1727 1728/* 1729 * Determine if the ISN we will generate has advanced beyond the last 1730 * sequence number used by the previous connection. If so, indicate 1731 * that it is safe to recycle this tw socket by returning 1. 1732 * 1733 * XXXRW: This function should assert the inpcb lock as it does multiple 1734 * non-atomic reads from the tcptw, but is currently called without it from 1735 * in_pcb.c:in_pcblookup_local(). 1736 */ 1737int 1738tcp_twrecycleable(struct tcptw *tw) 1739{ 1740 tcp_seq new_iss = tw->iss; 1741 tcp_seq new_irs = tw->irs; 1742 1743 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1744 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1745 1746 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1747 return (1); 1748 else 1749 return (0); 1750} 1751 1752struct tcptw * 1753tcp_twclose(struct tcptw *tw, int reuse) 1754{ 1755 struct inpcb *inp; 1756 1757 inp = tw->tw_inpcb; 1758 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1759 INP_LOCK_ASSERT(inp); 1760 1761 tw->tw_inpcb = NULL; 1762 tcp_timer_2msl_stop(tw); 1763 inp->inp_ppcb = NULL; 1764#ifdef INET6 1765 if (inp->inp_vflag & INP_IPV6PROTO) 1766 in6_pcbdetach(inp); 1767 else 1768#endif 1769 in_pcbdetach(inp); 1770 tcpstat.tcps_closed++; 1771 crfree(tw->tw_cred); 1772 tw->tw_cred = NULL; 1773 if (reuse) 1774 return (tw); 1775 uma_zfree(tcptw_zone, tw); 1776 return (NULL); 1777} 1778 1779int 1780tcp_twrespond(struct tcptw *tw, int flags) 1781{ 1782 struct inpcb *inp = tw->tw_inpcb; 1783 struct tcphdr *th; 1784 struct mbuf *m; 1785 struct ip *ip = NULL; 1786 u_int8_t *optp; 1787 u_int hdrlen, optlen; 1788 int error; 1789#ifdef INET6 1790 struct ip6_hdr *ip6 = NULL; 1791 int isipv6 = inp->inp_inc.inc_isipv6; 1792#endif 1793 1794 INP_LOCK_ASSERT(inp); 1795 1796 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1797 if (m == NULL) 1798 return (ENOBUFS); 1799 m->m_data += max_linkhdr; 1800 1801#ifdef MAC 1802 mac_create_mbuf_from_inpcb(inp, m); 1803#endif 1804 1805#ifdef INET6 1806 if (isipv6) { 1807 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1808 ip6 = mtod(m, struct ip6_hdr *); 1809 th = (struct tcphdr *)(ip6 + 1); 1810 tcpip_fillheaders(inp, ip6, th); 1811 } else 1812#endif 1813 { 1814 hdrlen = sizeof(struct tcpiphdr); 1815 ip = mtod(m, struct ip *); 1816 th = (struct tcphdr *)(ip + 1); 1817 tcpip_fillheaders(inp, ip, th); 1818 } 1819 optp = (u_int8_t *)(th + 1); 1820 1821 /* 1822 * Send a timestamp and echo-reply if both our side and our peer 1823 * have sent timestamps in our SYN's and this is not a RST. 1824 */ 1825 if (tw->t_recent && flags == TH_ACK) { 1826 u_int32_t *lp = (u_int32_t *)optp; 1827 1828 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1829 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1830 *lp++ = htonl(ticks); 1831 *lp = htonl(tw->t_recent); 1832 optp += TCPOLEN_TSTAMP_APPA; 1833 } 1834 1835 optlen = optp - (u_int8_t *)(th + 1); 1836 1837 m->m_len = hdrlen + optlen; 1838 m->m_pkthdr.len = m->m_len; 1839 1840 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1841 1842 th->th_seq = htonl(tw->snd_nxt); 1843 th->th_ack = htonl(tw->rcv_nxt); 1844 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1845 th->th_flags = flags; 1846 th->th_win = htons(tw->last_win); 1847 1848#ifdef INET6 1849 if (isipv6) { 1850 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1851 sizeof(struct tcphdr) + optlen); 1852 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1853 error = ip6_output(m, inp->in6p_outputopts, NULL, 1854 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1855 } else 1856#endif 1857 { 1858 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1859 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1860 m->m_pkthdr.csum_flags = CSUM_TCP; 1861 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1862 ip->ip_len = m->m_pkthdr.len; 1863 if (path_mtu_discovery) 1864 ip->ip_off |= IP_DF; 1865 error = ip_output(m, inp->inp_options, NULL, 1866 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1867 NULL, inp); 1868 } 1869 if (flags & TH_ACK) 1870 tcpstat.tcps_sndacks++; 1871 else 1872 tcpstat.tcps_sndctrl++; 1873 tcpstat.tcps_sndtotal++; 1874 return (error); 1875} 1876 1877/* 1878 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1879 * 1880 * This code attempts to calculate the bandwidth-delay product as a 1881 * means of determining the optimal window size to maximize bandwidth, 1882 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1883 * routers. This code also does a fairly good job keeping RTTs in check 1884 * across slow links like modems. We implement an algorithm which is very 1885 * similar (but not meant to be) TCP/Vegas. The code operates on the 1886 * transmitter side of a TCP connection and so only effects the transmit 1887 * side of the connection. 1888 * 1889 * BACKGROUND: TCP makes no provision for the management of buffer space 1890 * at the end points or at the intermediate routers and switches. A TCP 1891 * stream, whether using NewReno or not, will eventually buffer as 1892 * many packets as it is able and the only reason this typically works is 1893 * due to the fairly small default buffers made available for a connection 1894 * (typicaly 16K or 32K). As machines use larger windows and/or window 1895 * scaling it is now fairly easy for even a single TCP connection to blow-out 1896 * all available buffer space not only on the local interface, but on 1897 * intermediate routers and switches as well. NewReno makes a misguided 1898 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1899 * then backing off, then steadily increasing the window again until another 1900 * failure occurs, ad-infinitum. This results in terrible oscillation that 1901 * is only made worse as network loads increase and the idea of intentionally 1902 * blowing out network buffers is, frankly, a terrible way to manage network 1903 * resources. 1904 * 1905 * It is far better to limit the transmit window prior to the failure 1906 * condition being achieved. There are two general ways to do this: First 1907 * you can 'scan' through different transmit window sizes and locate the 1908 * point where the RTT stops increasing, indicating that you have filled the 1909 * pipe, then scan backwards until you note that RTT stops decreasing, then 1910 * repeat ad-infinitum. This method works in principle but has severe 1911 * implementation issues due to RTT variances, timer granularity, and 1912 * instability in the algorithm which can lead to many false positives and 1913 * create oscillations as well as interact badly with other TCP streams 1914 * implementing the same algorithm. 1915 * 1916 * The second method is to limit the window to the bandwidth delay product 1917 * of the link. This is the method we implement. RTT variances and our 1918 * own manipulation of the congestion window, bwnd, can potentially 1919 * destabilize the algorithm. For this reason we have to stabilize the 1920 * elements used to calculate the window. We do this by using the minimum 1921 * observed RTT, the long term average of the observed bandwidth, and 1922 * by adding two segments worth of slop. It isn't perfect but it is able 1923 * to react to changing conditions and gives us a very stable basis on 1924 * which to extend the algorithm. 1925 */ 1926void 1927tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1928{ 1929 u_long bw; 1930 u_long bwnd; 1931 int save_ticks; 1932 1933 INP_LOCK_ASSERT(tp->t_inpcb); 1934 1935 /* 1936 * If inflight_enable is disabled in the middle of a tcp connection, 1937 * make sure snd_bwnd is effectively disabled. 1938 */ 1939 if (tcp_inflight_enable == 0) { 1940 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1941 tp->snd_bandwidth = 0; 1942 return; 1943 } 1944 1945 /* 1946 * Figure out the bandwidth. Due to the tick granularity this 1947 * is a very rough number and it MUST be averaged over a fairly 1948 * long period of time. XXX we need to take into account a link 1949 * that is not using all available bandwidth, but for now our 1950 * slop will ramp us up if this case occurs and the bandwidth later 1951 * increases. 1952 * 1953 * Note: if ticks rollover 'bw' may wind up negative. We must 1954 * effectively reset t_bw_rtttime for this case. 1955 */ 1956 save_ticks = ticks; 1957 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 1958 return; 1959 1960 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 1961 (save_ticks - tp->t_bw_rtttime); 1962 tp->t_bw_rtttime = save_ticks; 1963 tp->t_bw_rtseq = ack_seq; 1964 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 1965 return; 1966 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1967 1968 tp->snd_bandwidth = bw; 1969 1970 /* 1971 * Calculate the semi-static bandwidth delay product, plus two maximal 1972 * segments. The additional slop puts us squarely in the sweet 1973 * spot and also handles the bandwidth run-up case and stabilization. 1974 * Without the slop we could be locking ourselves into a lower 1975 * bandwidth. 1976 * 1977 * Situations Handled: 1978 * (1) Prevents over-queueing of packets on LANs, especially on 1979 * high speed LANs, allowing larger TCP buffers to be 1980 * specified, and also does a good job preventing 1981 * over-queueing of packets over choke points like modems 1982 * (at least for the transmit side). 1983 * 1984 * (2) Is able to handle changing network loads (bandwidth 1985 * drops so bwnd drops, bandwidth increases so bwnd 1986 * increases). 1987 * 1988 * (3) Theoretically should stabilize in the face of multiple 1989 * connections implementing the same algorithm (this may need 1990 * a little work). 1991 * 1992 * (4) Stability value (defaults to 20 = 2 maximal packets) can 1993 * be adjusted with a sysctl but typically only needs to be 1994 * on very slow connections. A value no smaller then 5 1995 * should be used, but only reduce this default if you have 1996 * no other choice. 1997 */ 1998#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1999 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2000#undef USERTT 2001 2002 if (tcp_inflight_debug > 0) { 2003 static int ltime; 2004 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2005 ltime = ticks; 2006 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2007 tp, 2008 bw, 2009 tp->t_rttbest, 2010 tp->t_srtt, 2011 bwnd 2012 ); 2013 } 2014 } 2015 if ((long)bwnd < tcp_inflight_min) 2016 bwnd = tcp_inflight_min; 2017 if (bwnd > tcp_inflight_max) 2018 bwnd = tcp_inflight_max; 2019 if ((long)bwnd < tp->t_maxseg * 2) 2020 bwnd = tp->t_maxseg * 2; 2021 tp->snd_bwnd = bwnd; 2022} 2023 2024#ifdef TCP_SIGNATURE 2025/* 2026 * Callback function invoked by m_apply() to digest TCP segment data 2027 * contained within an mbuf chain. 2028 */ 2029static int 2030tcp_signature_apply(void *fstate, void *data, u_int len) 2031{ 2032 2033 MD5Update(fstate, (u_char *)data, len); 2034 return (0); 2035} 2036 2037/* 2038 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2039 * 2040 * Parameters: 2041 * m pointer to head of mbuf chain 2042 * off0 offset to TCP header within the mbuf chain 2043 * len length of TCP segment data, excluding options 2044 * optlen length of TCP segment options 2045 * buf pointer to storage for computed MD5 digest 2046 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2047 * 2048 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2049 * When called from tcp_input(), we can be sure that th_sum has been 2050 * zeroed out and verified already. 2051 * 2052 * This function is for IPv4 use only. Calling this function with an 2053 * IPv6 packet in the mbuf chain will yield undefined results. 2054 * 2055 * Return 0 if successful, otherwise return -1. 2056 * 2057 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2058 * search with the destination IP address, and a 'magic SPI' to be 2059 * determined by the application. This is hardcoded elsewhere to 1179 2060 * right now. Another branch of this code exists which uses the SPD to 2061 * specify per-application flows but it is unstable. 2062 */ 2063int 2064tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2065 u_char *buf, u_int direction) 2066{ 2067 union sockaddr_union dst; 2068 struct ippseudo ippseudo; 2069 MD5_CTX ctx; 2070 int doff; 2071 struct ip *ip; 2072 struct ipovly *ipovly; 2073 struct secasvar *sav; 2074 struct tcphdr *th; 2075 u_short savecsum; 2076 2077 KASSERT(m != NULL, ("NULL mbuf chain")); 2078 KASSERT(buf != NULL, ("NULL signature pointer")); 2079 2080 /* Extract the destination from the IP header in the mbuf. */ 2081 ip = mtod(m, struct ip *); 2082 bzero(&dst, sizeof(union sockaddr_union)); 2083 dst.sa.sa_len = sizeof(struct sockaddr_in); 2084 dst.sa.sa_family = AF_INET; 2085 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2086 ip->ip_src : ip->ip_dst; 2087 2088 /* Look up an SADB entry which matches the address of the peer. */ 2089 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2090 if (sav == NULL) { 2091 printf("%s: SADB lookup failed for %s\n", __func__, 2092 inet_ntoa(dst.sin.sin_addr)); 2093 return (EINVAL); 2094 } 2095 2096 MD5Init(&ctx); 2097 ipovly = (struct ipovly *)ip; 2098 th = (struct tcphdr *)((u_char *)ip + off0); 2099 doff = off0 + sizeof(struct tcphdr) + optlen; 2100 2101 /* 2102 * Step 1: Update MD5 hash with IP pseudo-header. 2103 * 2104 * XXX The ippseudo header MUST be digested in network byte order, 2105 * or else we'll fail the regression test. Assume all fields we've 2106 * been doing arithmetic on have been in host byte order. 2107 * XXX One cannot depend on ipovly->ih_len here. When called from 2108 * tcp_output(), the underlying ip_len member has not yet been set. 2109 */ 2110 ippseudo.ippseudo_src = ipovly->ih_src; 2111 ippseudo.ippseudo_dst = ipovly->ih_dst; 2112 ippseudo.ippseudo_pad = 0; 2113 ippseudo.ippseudo_p = IPPROTO_TCP; 2114 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2115 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2116 2117 /* 2118 * Step 2: Update MD5 hash with TCP header, excluding options. 2119 * The TCP checksum must be set to zero. 2120 */ 2121 savecsum = th->th_sum; 2122 th->th_sum = 0; 2123 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2124 th->th_sum = savecsum; 2125 2126 /* 2127 * Step 3: Update MD5 hash with TCP segment data. 2128 * Use m_apply() to avoid an early m_pullup(). 2129 */ 2130 if (len > 0) 2131 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2132 2133 /* 2134 * Step 4: Update MD5 hash with shared secret. 2135 */ 2136 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2137 MD5Final(buf, &ctx); 2138 2139 key_sa_recordxfer(sav, m); 2140 KEY_FREESAV(&sav); 2141 return (0); 2142} 2143#endif /* TCP_SIGNATURE */ 2144 2145static int 2146sysctl_drop(SYSCTL_HANDLER_ARGS) 2147{ 2148 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2149 struct sockaddr_storage addrs[2]; 2150 struct inpcb *inp; 2151 struct tcpcb *tp; 2152 struct sockaddr_in *fin, *lin; 2153#ifdef INET6 2154 struct sockaddr_in6 *fin6, *lin6; 2155 struct in6_addr f6, l6; 2156#endif 2157 int error; 2158 2159 inp = NULL; 2160 fin = lin = NULL; 2161#ifdef INET6 2162 fin6 = lin6 = NULL; 2163#endif 2164 error = 0; 2165 2166 if (req->oldptr != NULL || req->oldlen != 0) 2167 return (EINVAL); 2168 if (req->newptr == NULL) 2169 return (EPERM); 2170 if (req->newlen < sizeof(addrs)) 2171 return (ENOMEM); 2172 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2173 if (error) 2174 return (error); 2175 2176 switch (addrs[0].ss_family) { 2177#ifdef INET6 2178 case AF_INET6: 2179 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2180 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2181 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2182 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2183 return (EINVAL); 2184 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2185 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2186 return (EINVAL); 2187 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2188 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2189 fin = (struct sockaddr_in *)&addrs[0]; 2190 lin = (struct sockaddr_in *)&addrs[1]; 2191 break; 2192 } 2193 error = sa6_embedscope(fin6, ip6_use_defzone); 2194 if (error) 2195 return (error); 2196 error = sa6_embedscope(lin6, ip6_use_defzone); 2197 return (error); 2198 break; 2199#endif 2200 case AF_INET: 2201 fin = (struct sockaddr_in *)&addrs[0]; 2202 lin = (struct sockaddr_in *)&addrs[1]; 2203 if (fin->sin_len != sizeof(struct sockaddr_in) || 2204 lin->sin_len != sizeof(struct sockaddr_in)) 2205 return (EINVAL); 2206 break; 2207 default: 2208 return (EINVAL); 2209 } 2210 INP_INFO_WLOCK(&tcbinfo); 2211 switch (addrs[0].ss_family) { 2212#ifdef INET6 2213 case AF_INET6: 2214 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2215 &l6, lin6->sin6_port, 0, NULL); 2216 break; 2217#endif 2218 case AF_INET: 2219 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2220 lin->sin_addr, lin->sin_port, 0, NULL); 2221 break; 2222 } 2223 if (inp != NULL) { 2224 INP_LOCK(inp); 2225 if ((tp = intotcpcb(inp)) && 2226 ((inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 2227 tp = tcp_drop(tp, ECONNABORTED); 2228 if (tp != NULL) 2229 INP_UNLOCK(inp); 2230 } else 2231 INP_UNLOCK(inp); 2232 } else 2233 error = ESRCH; 2234 INP_INFO_WUNLOCK(&tcbinfo); 2235 return (error); 2236} 2237 2238SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2239 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2240 0, sysctl_drop, "", "Drop TCP connection"); 2241