tcp_timewait.c revision 167721
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 167721 2007-03-19 19:00:51Z andre $ 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/malloc.h> 47#include <sys/mbuf.h> 48#ifdef INET6 49#include <sys/domain.h> 50#endif 51#include <sys/priv.h> 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> 115 116#include <security/mac/mac_framework.h> 117 118int tcp_mssdflt = TCP_MSS; 119SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 120 &tcp_mssdflt, 0, "Default TCP Maximum Segment Size"); 121 122#ifdef INET6 123int tcp_v6mssdflt = TCP6_MSS; 124SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 125 CTLFLAG_RW, &tcp_v6mssdflt , 0, 126 "Default TCP Maximum Segment Size for IPv6"); 127#endif 128 129/* 130 * Minimum MSS we accept and use. This prevents DoS attacks where 131 * we are forced to a ridiculous low MSS like 20 and send hundreds 132 * of packets instead of one. The effect scales with the available 133 * bandwidth and quickly saturates the CPU and network interface 134 * with packet generation and sending. Set to zero to disable MINMSS 135 * checking. This setting prevents us from sending too small packets. 136 */ 137int tcp_minmss = TCP_MINMSS; 138SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 139 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 140/* 141 * Number of TCP segments per second we accept from remote host 142 * before we start to calculate average segment size. If average 143 * segment size drops below the minimum TCP MSS we assume a DoS 144 * attack and reset+drop the connection. Care has to be taken not to 145 * set this value too small to not kill interactive type connections 146 * (telnet, SSH) which send many small packets. 147 */ 148int tcp_minmssoverload = TCP_MINMSSOVERLOAD; 149SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, 150 &tcp_minmssoverload , 0, 151 "Number of TCP Segments per Second allowed to be under the MINMSS Size"); 152 153int tcp_do_rfc1323 = 1; 154SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 155 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions"); 156 157static int tcp_tcbhashsize = 0; 158SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 159 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 160 161static int do_tcpdrain = 1; 162SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, 163 &do_tcpdrain, 0, 164 "Enable tcp_drain routine for extra help when low on mbufs"); 165 166SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 167 &tcbinfo.ipi_count, 0, "Number of active PCBs"); 168 169static int icmp_may_rst = 1; 170SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, 171 &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 178static uma_zone_t tcptw_zone; 179static int maxtcptw; 180 181static int 182tcptw_auto_size(void) 183{ 184 int halfrange; 185 186 /* 187 * Max out at half the ephemeral port range so that TIME_WAIT 188 * sockets don't tie up too many ephemeral ports. 189 */ 190 if (ipport_lastauto > ipport_firstauto) 191 halfrange = (ipport_lastauto - ipport_firstauto) / 2; 192 else 193 halfrange = (ipport_firstauto - ipport_lastauto) / 2; 194 /* Protect against goofy port ranges smaller than 32. */ 195 return (imin(imax(halfrange, 32), maxsockets / 5)); 196} 197 198static int 199sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 200{ 201 int error, new; 202 203 if (maxtcptw == 0) 204 new = tcptw_auto_size(); 205 else 206 new = maxtcptw; 207 error = sysctl_handle_int(oidp, &new, sizeof(int), req); 208 if (error == 0 && req->newptr) 209 if (new >= 32) { 210 maxtcptw = new; 211 uma_zone_set_max(tcptw_zone, maxtcptw); 212 } 213 return (error); 214} 215SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 216 &maxtcptw, 0, sysctl_maxtcptw, "IU", 217 "Maximum number of compressed TCP TIME_WAIT entries"); 218 219static int nolocaltimewait = 0; 220SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW, 221 &nolocaltimewait, 0, 222 "Do not create compressed TCP TIME_WAIT entries for local connections"); 223 224/* 225 * TCP bandwidth limiting sysctls. Note that the default lower bound of 226 * 1024 exists only for debugging. A good production default would be 227 * something like 6100. 228 */ 229SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 230 "TCP inflight data limiting"); 231 232static int tcp_inflight_enable = 1; 233SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 234 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 235 236static int tcp_inflight_debug = 0; 237SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 238 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 239 240static int tcp_inflight_rttthresh; 241SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 242 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 243 "RTT threshold below which inflight will deactivate itself"); 244 245static int tcp_inflight_min = 6144; 246SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 247 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 248 249static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 250SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 251 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 252 253static int tcp_inflight_stab = 20; 254SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 255 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 256 257uma_zone_t sack_hole_zone; 258 259static struct inpcb *tcp_notify(struct inpcb *, int); 260static void tcp_isn_tick(void *); 261 262/* 263 * Target size of TCP PCB hash tables. Must be a power of two. 264 * 265 * Note that this can be overridden by the kernel environment 266 * variable net.inet.tcp.tcbhashsize 267 */ 268#ifndef TCBHASHSIZE 269#define TCBHASHSIZE 512 270#endif 271 272/* 273 * XXX 274 * Callouts should be moved into struct tcp directly. They are currently 275 * separate because the tcpcb structure is exported to userland for sysctl 276 * parsing purposes, which do not know about callouts. 277 */ 278struct tcpcb_mem { 279 struct tcpcb tcb; 280 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 281 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 282}; 283 284static uma_zone_t tcpcb_zone; 285struct callout isn_callout; 286static struct mtx isn_mtx; 287 288#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 289#define ISN_LOCK() mtx_lock(&isn_mtx) 290#define ISN_UNLOCK() mtx_unlock(&isn_mtx) 291 292/* 293 * TCP initialization. 294 */ 295static void 296tcp_zone_change(void *tag) 297{ 298 299 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 300 uma_zone_set_max(tcpcb_zone, maxsockets); 301 if (maxtcptw == 0) 302 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 303} 304 305static int 306tcp_inpcb_init(void *mem, int size, int flags) 307{ 308 struct inpcb *inp = mem; 309 310 INP_LOCK_INIT(inp, "inp", "tcpinp"); 311 return (0); 312} 313 314void 315tcp_init(void) 316{ 317 int hashsize = TCBHASHSIZE; 318 319 tcp_delacktime = TCPTV_DELACK; 320 tcp_keepinit = TCPTV_KEEP_INIT; 321 tcp_keepidle = TCPTV_KEEP_IDLE; 322 tcp_keepintvl = TCPTV_KEEPINTVL; 323 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 324 tcp_msl = TCPTV_MSL; 325 tcp_rexmit_min = TCPTV_MIN; 326 tcp_rexmit_slop = TCPTV_CPU_VAR; 327 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 328 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 329 330 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 331 LIST_INIT(&tcb); 332 tcbinfo.listhead = &tcb; 333 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 334 if (!powerof2(hashsize)) { 335 printf("WARNING: TCB hash size not a power of 2\n"); 336 hashsize = 512; /* safe default */ 337 } 338 tcp_tcbhashsize = hashsize; 339 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 340 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 341 &tcbinfo.porthashmask); 342 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 343 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 344 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 345#ifdef INET6 346#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 347#else /* INET6 */ 348#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 349#endif /* INET6 */ 350 if (max_protohdr < TCP_MINPROTOHDR) 351 max_protohdr = TCP_MINPROTOHDR; 352 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 353 panic("tcp_init"); 354#undef TCP_MINPROTOHDR 355 /* 356 * These have to be type stable for the benefit of the timers. 357 */ 358 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 359 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 360 uma_zone_set_max(tcpcb_zone, maxsockets); 361 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 362 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 363 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 364 if (maxtcptw == 0) 365 uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 366 else 367 uma_zone_set_max(tcptw_zone, maxtcptw); 368 tcp_timer_init(); 369 syncache_init(); 370 tcp_hc_init(); 371 tcp_reass_init(); 372 ISN_LOCK_INIT(); 373 callout_init(&isn_callout, CALLOUT_MPSAFE); 374 tcp_isn_tick(NULL); 375 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 376 SHUTDOWN_PRI_DEFAULT); 377 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 378 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 379 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 380 EVENTHANDLER_PRI_ANY); 381} 382 383void 384tcp_fini(void *xtp) 385{ 386 387 callout_stop(&isn_callout); 388} 389 390/* 391 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 392 * tcp_template used to store this data in mbufs, but we now recopy it out 393 * of the tcpcb each time to conserve mbufs. 394 */ 395void 396tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 397{ 398 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 399 400 INP_LOCK_ASSERT(inp); 401 402#ifdef INET6 403 if ((inp->inp_vflag & INP_IPV6) != 0) { 404 struct ip6_hdr *ip6; 405 406 ip6 = (struct ip6_hdr *)ip_ptr; 407 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 408 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 409 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 410 (IPV6_VERSION & IPV6_VERSION_MASK); 411 ip6->ip6_nxt = IPPROTO_TCP; 412 ip6->ip6_plen = sizeof(struct tcphdr); 413 ip6->ip6_src = inp->in6p_laddr; 414 ip6->ip6_dst = inp->in6p_faddr; 415 } else 416#endif 417 { 418 struct ip *ip; 419 420 ip = (struct ip *)ip_ptr; 421 ip->ip_v = IPVERSION; 422 ip->ip_hl = 5; 423 ip->ip_tos = inp->inp_ip_tos; 424 ip->ip_len = 0; 425 ip->ip_id = 0; 426 ip->ip_off = 0; 427 ip->ip_ttl = inp->inp_ip_ttl; 428 ip->ip_sum = 0; 429 ip->ip_p = IPPROTO_TCP; 430 ip->ip_src = inp->inp_laddr; 431 ip->ip_dst = inp->inp_faddr; 432 } 433 th->th_sport = inp->inp_lport; 434 th->th_dport = inp->inp_fport; 435 th->th_seq = 0; 436 th->th_ack = 0; 437 th->th_x2 = 0; 438 th->th_off = 5; 439 th->th_flags = 0; 440 th->th_win = 0; 441 th->th_urp = 0; 442 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 443} 444 445/* 446 * Create template to be used to send tcp packets on a connection. 447 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 448 * use for this function is in keepalives, which use tcp_respond. 449 */ 450struct tcptemp * 451tcpip_maketemplate(struct inpcb *inp) 452{ 453 struct mbuf *m; 454 struct tcptemp *n; 455 456 m = m_get(M_DONTWAIT, MT_DATA); 457 if (m == NULL) 458 return (0); 459 m->m_len = sizeof(struct tcptemp); 460 n = mtod(m, struct tcptemp *); 461 462 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 463 return (n); 464} 465 466/* 467 * Send a single message to the TCP at address specified by 468 * the given TCP/IP header. If m == NULL, then we make a copy 469 * of the tcpiphdr at ti and send directly to the addressed host. 470 * This is used to force keep alive messages out using the TCP 471 * template for a connection. If flags are given then we send 472 * a message back to the TCP which originated the * segment ti, 473 * and discard the mbuf containing it and any other attached mbufs. 474 * 475 * In any case the ack and sequence number of the transmitted 476 * segment are as specified by the parameters. 477 * 478 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 479 */ 480void 481tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th, 482 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 483{ 484 register int tlen; 485 int win = 0; 486 struct ip *ip; 487 struct tcphdr *nth; 488#ifdef INET6 489 struct ip6_hdr *ip6; 490 int isipv6; 491#endif /* INET6 */ 492 int ipflags = 0; 493 struct inpcb *inp; 494 495 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 496 497#ifdef INET6 498 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 499 ip6 = ipgen; 500#endif /* INET6 */ 501 ip = ipgen; 502 503 if (tp != NULL) { 504 inp = tp->t_inpcb; 505 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 506 INP_INFO_WLOCK_ASSERT(&tcbinfo); 507 INP_LOCK_ASSERT(inp); 508 } else 509 inp = NULL; 510 511 if (tp != NULL) { 512 if (!(flags & TH_RST)) { 513 win = sbspace(&inp->inp_socket->so_rcv); 514 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 515 win = (long)TCP_MAXWIN << tp->rcv_scale; 516 } 517 } 518 if (m == NULL) { 519 m = m_gethdr(M_DONTWAIT, MT_DATA); 520 if (m == NULL) 521 return; 522 tlen = 0; 523 m->m_data += max_linkhdr; 524#ifdef INET6 525 if (isipv6) { 526 bcopy((caddr_t)ip6, mtod(m, caddr_t), 527 sizeof(struct ip6_hdr)); 528 ip6 = mtod(m, struct ip6_hdr *); 529 nth = (struct tcphdr *)(ip6 + 1); 530 } else 531#endif /* INET6 */ 532 { 533 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 534 ip = mtod(m, struct ip *); 535 nth = (struct tcphdr *)(ip + 1); 536 } 537 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 538 flags = TH_ACK; 539 } else { 540 m_freem(m->m_next); 541 m->m_next = NULL; 542 m->m_data = (caddr_t)ipgen; 543 /* m_len is set later */ 544 tlen = 0; 545#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 546#ifdef INET6 547 if (isipv6) { 548 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 549 nth = (struct tcphdr *)(ip6 + 1); 550 } else 551#endif /* INET6 */ 552 { 553 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 554 nth = (struct tcphdr *)(ip + 1); 555 } 556 if (th != nth) { 557 /* 558 * this is usually a case when an extension header 559 * exists between the IPv6 header and the 560 * TCP header. 561 */ 562 nth->th_sport = th->th_sport; 563 nth->th_dport = th->th_dport; 564 } 565 xchg(nth->th_dport, nth->th_sport, n_short); 566#undef xchg 567 } 568#ifdef INET6 569 if (isipv6) { 570 ip6->ip6_flow = 0; 571 ip6->ip6_vfc = IPV6_VERSION; 572 ip6->ip6_nxt = IPPROTO_TCP; 573 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 574 tlen)); 575 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 576 } else 577#endif 578 { 579 tlen += sizeof (struct tcpiphdr); 580 ip->ip_len = tlen; 581 ip->ip_ttl = ip_defttl; 582 if (path_mtu_discovery) 583 ip->ip_off |= IP_DF; 584 } 585 m->m_len = tlen; 586 m->m_pkthdr.len = tlen; 587 m->m_pkthdr.rcvif = NULL; 588#ifdef MAC 589 if (inp != NULL) { 590 /* 591 * Packet is associated with a socket, so allow the 592 * label of the response to reflect the socket label. 593 */ 594 INP_LOCK_ASSERT(inp); 595 mac_create_mbuf_from_inpcb(inp, m); 596 } else { 597 /* 598 * Packet is not associated with a socket, so possibly 599 * update the label in place. 600 */ 601 mac_reflect_mbuf_tcp(m); 602 } 603#endif 604 nth->th_seq = htonl(seq); 605 nth->th_ack = htonl(ack); 606 nth->th_x2 = 0; 607 nth->th_off = sizeof (struct tcphdr) >> 2; 608 nth->th_flags = flags; 609 if (tp != NULL) 610 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 611 else 612 nth->th_win = htons((u_short)win); 613 nth->th_urp = 0; 614#ifdef INET6 615 if (isipv6) { 616 nth->th_sum = 0; 617 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 618 sizeof(struct ip6_hdr), 619 tlen - sizeof(struct ip6_hdr)); 620 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 621 NULL, NULL); 622 } else 623#endif /* INET6 */ 624 { 625 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 626 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 627 m->m_pkthdr.csum_flags = CSUM_TCP; 628 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 629 } 630#ifdef TCPDEBUG 631 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 632 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 633#endif 634#ifdef INET6 635 if (isipv6) 636 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 637 else 638#endif /* INET6 */ 639 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 640} 641 642/* 643 * Create a new TCP control block, making an 644 * empty reassembly queue and hooking it to the argument 645 * protocol control block. The `inp' parameter must have 646 * come from the zone allocator set up in tcp_init(). 647 */ 648struct tcpcb * 649tcp_newtcpcb(struct inpcb *inp) 650{ 651 struct tcpcb_mem *tm; 652 struct tcpcb *tp; 653#ifdef INET6 654 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 655#endif /* INET6 */ 656 657 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 658 if (tm == NULL) 659 return (NULL); 660 tp = &tm->tcb; 661 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 662 tp->t_maxseg = tp->t_maxopd = 663#ifdef INET6 664 isipv6 ? tcp_v6mssdflt : 665#endif /* INET6 */ 666 tcp_mssdflt; 667 668 /* Set up our timeouts. */ 669 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 670 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 671 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 672 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 673 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 674 675 if (tcp_do_rfc1323) 676 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 677 tp->sack_enable = tcp_do_sack; 678 TAILQ_INIT(&tp->snd_holes); 679 tp->t_inpcb = inp; /* XXX */ 680 /* 681 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 682 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 683 * reasonable initial retransmit time. 684 */ 685 tp->t_srtt = TCPTV_SRTTBASE; 686 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 687 tp->t_rttmin = tcp_rexmit_min; 688 tp->t_rxtcur = TCPTV_RTOBASE; 689 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 690 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 691 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 692 tp->t_rcvtime = ticks; 693 tp->t_bw_rtttime = ticks; 694 /* 695 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 696 * because the socket may be bound to an IPv6 wildcard address, 697 * which may match an IPv4-mapped IPv6 address. 698 */ 699 inp->inp_ip_ttl = ip_defttl; 700 inp->inp_ppcb = tp; 701 return (tp); /* XXX */ 702} 703 704/* 705 * Drop a TCP connection, reporting 706 * the specified error. If connection is synchronized, 707 * then send a RST to peer. 708 */ 709struct tcpcb * 710tcp_drop(struct tcpcb *tp, int errno) 711{ 712 struct socket *so = tp->t_inpcb->inp_socket; 713 714 INP_INFO_WLOCK_ASSERT(&tcbinfo); 715 INP_LOCK_ASSERT(tp->t_inpcb); 716 717 if (TCPS_HAVERCVDSYN(tp->t_state)) { 718 tp->t_state = TCPS_CLOSED; 719 (void) tcp_output(tp); 720 tcpstat.tcps_drops++; 721 } else 722 tcpstat.tcps_conndrops++; 723 if (errno == ETIMEDOUT && tp->t_softerror) 724 errno = tp->t_softerror; 725 so->so_error = errno; 726 return (tcp_close(tp)); 727} 728 729void 730tcp_discardcb(struct tcpcb *tp) 731{ 732 struct tseg_qent *q; 733 struct inpcb *inp = tp->t_inpcb; 734 struct socket *so = inp->inp_socket; 735#ifdef INET6 736 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 737#endif /* INET6 */ 738 739 INP_LOCK_ASSERT(inp); 740 741 /* 742 * Make sure that all of our timers are stopped before we 743 * delete the PCB. 744 */ 745 callout_stop(tp->tt_rexmt); 746 callout_stop(tp->tt_persist); 747 callout_stop(tp->tt_keep); 748 callout_stop(tp->tt_2msl); 749 callout_stop(tp->tt_delack); 750 751 /* 752 * If we got enough samples through the srtt filter, 753 * save the rtt and rttvar in the routing entry. 754 * 'Enough' is arbitrarily defined as 4 rtt samples. 755 * 4 samples is enough for the srtt filter to converge 756 * to within enough % of the correct value; fewer samples 757 * and we could save a bogus rtt. The danger is not high 758 * as tcp quickly recovers from everything. 759 * XXX: Works very well but needs some more statistics! 760 */ 761 if (tp->t_rttupdated >= 4) { 762 struct hc_metrics_lite metrics; 763 u_long ssthresh; 764 765 bzero(&metrics, sizeof(metrics)); 766 /* 767 * Update the ssthresh always when the conditions below 768 * are satisfied. This gives us better new start value 769 * for the congestion avoidance for new connections. 770 * ssthresh is only set if packet loss occured on a session. 771 * 772 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 773 * being torn down. Ideally this code would not use 'so'. 774 */ 775 ssthresh = tp->snd_ssthresh; 776 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 777 /* 778 * convert the limit from user data bytes to 779 * packets then to packet data bytes. 780 */ 781 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 782 if (ssthresh < 2) 783 ssthresh = 2; 784 ssthresh *= (u_long)(tp->t_maxseg + 785#ifdef INET6 786 (isipv6 ? sizeof (struct ip6_hdr) + 787 sizeof (struct tcphdr) : 788#endif 789 sizeof (struct tcpiphdr) 790#ifdef INET6 791 ) 792#endif 793 ); 794 } else 795 ssthresh = 0; 796 metrics.rmx_ssthresh = ssthresh; 797 798 metrics.rmx_rtt = tp->t_srtt; 799 metrics.rmx_rttvar = tp->t_rttvar; 800 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 801 metrics.rmx_bandwidth = tp->snd_bandwidth; 802 metrics.rmx_cwnd = tp->snd_cwnd; 803 metrics.rmx_sendpipe = 0; 804 metrics.rmx_recvpipe = 0; 805 806 tcp_hc_update(&inp->inp_inc, &metrics); 807 } 808 809 /* free the reassembly queue, if any */ 810 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 811 LIST_REMOVE(q, tqe_q); 812 m_freem(q->tqe_m); 813 uma_zfree(tcp_reass_zone, q); 814 tp->t_segqlen--; 815 tcp_reass_qsize--; 816 } 817 tcp_free_sackholes(tp); 818 inp->inp_ppcb = NULL; 819 tp->t_inpcb = NULL; 820 uma_zfree(tcpcb_zone, tp); 821} 822 823/* 824 * Attempt to close a TCP control block, marking it as dropped, and freeing 825 * the socket if we hold the only reference. 826 */ 827struct tcpcb * 828tcp_close(struct tcpcb *tp) 829{ 830 struct inpcb *inp = tp->t_inpcb; 831 struct socket *so; 832 833 INP_INFO_WLOCK_ASSERT(&tcbinfo); 834 INP_LOCK_ASSERT(inp); 835 836 in_pcbdrop(inp); 837 tcpstat.tcps_closed++; 838 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 839 so = inp->inp_socket; 840 soisdisconnected(so); 841 if (inp->inp_vflag & INP_SOCKREF) { 842 KASSERT(so->so_state & SS_PROTOREF, 843 ("tcp_close: !SS_PROTOREF")); 844 inp->inp_vflag &= ~INP_SOCKREF; 845 INP_UNLOCK(inp); 846 ACCEPT_LOCK(); 847 SOCK_LOCK(so); 848 so->so_state &= ~SS_PROTOREF; 849 sofree(so); 850 return (NULL); 851 } 852 return (tp); 853} 854 855void 856tcp_drain(void) 857{ 858 859 if (do_tcpdrain) { 860 struct inpcb *inpb; 861 struct tcpcb *tcpb; 862 struct tseg_qent *te; 863 864 /* 865 * Walk the tcpbs, if existing, and flush the reassembly queue, 866 * if there is one... 867 * XXX: The "Net/3" implementation doesn't imply that the TCP 868 * reassembly queue should be flushed, but in a situation 869 * where we're really low on mbufs, this is potentially 870 * usefull. 871 */ 872 INP_INFO_RLOCK(&tcbinfo); 873 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 874 if (inpb->inp_vflag & INP_TIMEWAIT) 875 continue; 876 INP_LOCK(inpb); 877 if ((tcpb = intotcpcb(inpb)) != NULL) { 878 while ((te = LIST_FIRST(&tcpb->t_segq)) 879 != NULL) { 880 LIST_REMOVE(te, tqe_q); 881 m_freem(te->tqe_m); 882 uma_zfree(tcp_reass_zone, te); 883 tcpb->t_segqlen--; 884 tcp_reass_qsize--; 885 } 886 tcp_clean_sackreport(tcpb); 887 } 888 INP_UNLOCK(inpb); 889 } 890 INP_INFO_RUNLOCK(&tcbinfo); 891 } 892} 893 894/* 895 * Notify a tcp user of an asynchronous error; 896 * store error as soft error, but wake up user 897 * (for now, won't do anything until can select for soft error). 898 * 899 * Do not wake up user since there currently is no mechanism for 900 * reporting soft errors (yet - a kqueue filter may be added). 901 */ 902static struct inpcb * 903tcp_notify(struct inpcb *inp, int error) 904{ 905 struct tcpcb *tp; 906 907 INP_INFO_WLOCK_ASSERT(&tcbinfo); 908 INP_LOCK_ASSERT(inp); 909 910 if ((inp->inp_vflag & INP_TIMEWAIT) || 911 (inp->inp_vflag & INP_DROPPED)) 912 return (inp); 913 914 tp = intotcpcb(inp); 915 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 916 917 /* 918 * Ignore some errors if we are hooked up. 919 * If connection hasn't completed, has retransmitted several times, 920 * and receives a second error, give up now. This is better 921 * than waiting a long time to establish a connection that 922 * can never complete. 923 */ 924 if (tp->t_state == TCPS_ESTABLISHED && 925 (error == EHOSTUNREACH || error == ENETUNREACH || 926 error == EHOSTDOWN)) { 927 return (inp); 928 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 929 tp->t_softerror) { 930 tp = tcp_drop(tp, error); 931 if (tp != NULL) 932 return (inp); 933 else 934 return (NULL); 935 } else { 936 tp->t_softerror = error; 937 return (inp); 938 } 939#if 0 940 wakeup( &so->so_timeo); 941 sorwakeup(so); 942 sowwakeup(so); 943#endif 944} 945 946static int 947tcp_pcblist(SYSCTL_HANDLER_ARGS) 948{ 949 int error, i, n; 950 struct inpcb *inp, **inp_list; 951 inp_gen_t gencnt; 952 struct xinpgen xig; 953 954 /* 955 * The process of preparing the TCB list is too time-consuming and 956 * resource-intensive to repeat twice on every request. 957 */ 958 if (req->oldptr == NULL) { 959 n = tcbinfo.ipi_count; 960 req->oldidx = 2 * (sizeof xig) 961 + (n + n/8) * sizeof(struct xtcpcb); 962 return (0); 963 } 964 965 if (req->newptr != NULL) 966 return (EPERM); 967 968 /* 969 * OK, now we're committed to doing something. 970 */ 971 INP_INFO_RLOCK(&tcbinfo); 972 gencnt = tcbinfo.ipi_gencnt; 973 n = tcbinfo.ipi_count; 974 INP_INFO_RUNLOCK(&tcbinfo); 975 976 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 977 + n * sizeof(struct xtcpcb)); 978 if (error != 0) 979 return (error); 980 981 xig.xig_len = sizeof xig; 982 xig.xig_count = n; 983 xig.xig_gen = gencnt; 984 xig.xig_sogen = so_gencnt; 985 error = SYSCTL_OUT(req, &xig, sizeof xig); 986 if (error) 987 return (error); 988 989 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 990 if (inp_list == NULL) 991 return (ENOMEM); 992 993 INP_INFO_RLOCK(&tcbinfo); 994 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 995 inp = LIST_NEXT(inp, inp_list)) { 996 INP_LOCK(inp); 997 if (inp->inp_gencnt <= gencnt) { 998 /* 999 * XXX: This use of cr_cansee(), introduced with 1000 * TCP state changes, is not quite right, but for 1001 * now, better than nothing. 1002 */ 1003 if (inp->inp_vflag & INP_TIMEWAIT) { 1004 if (intotw(inp) != NULL) 1005 error = cr_cansee(req->td->td_ucred, 1006 intotw(inp)->tw_cred); 1007 else 1008 error = EINVAL; /* Skip this inp. */ 1009 } else 1010 error = cr_canseesocket(req->td->td_ucred, 1011 inp->inp_socket); 1012 if (error == 0) 1013 inp_list[i++] = inp; 1014 } 1015 INP_UNLOCK(inp); 1016 } 1017 INP_INFO_RUNLOCK(&tcbinfo); 1018 n = i; 1019 1020 error = 0; 1021 for (i = 0; i < n; i++) { 1022 inp = inp_list[i]; 1023 INP_LOCK(inp); 1024 if (inp->inp_gencnt <= gencnt) { 1025 struct xtcpcb xt; 1026 void *inp_ppcb; 1027 1028 bzero(&xt, sizeof(xt)); 1029 xt.xt_len = sizeof xt; 1030 /* XXX should avoid extra copy */ 1031 bcopy(inp, &xt.xt_inp, sizeof *inp); 1032 inp_ppcb = inp->inp_ppcb; 1033 if (inp_ppcb == NULL) 1034 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1035 else if (inp->inp_vflag & INP_TIMEWAIT) { 1036 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1037 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1038 } else 1039 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1040 if (inp->inp_socket != NULL) 1041 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1042 else { 1043 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1044 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1045 } 1046 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1047 INP_UNLOCK(inp); 1048 error = SYSCTL_OUT(req, &xt, sizeof xt); 1049 } else 1050 INP_UNLOCK(inp); 1051 1052 } 1053 if (!error) { 1054 /* 1055 * Give the user an updated idea of our state. 1056 * If the generation differs from what we told 1057 * her before, she knows that something happened 1058 * while we were processing this request, and it 1059 * might be necessary to retry. 1060 */ 1061 INP_INFO_RLOCK(&tcbinfo); 1062 xig.xig_gen = tcbinfo.ipi_gencnt; 1063 xig.xig_sogen = so_gencnt; 1064 xig.xig_count = tcbinfo.ipi_count; 1065 INP_INFO_RUNLOCK(&tcbinfo); 1066 error = SYSCTL_OUT(req, &xig, sizeof xig); 1067 } 1068 free(inp_list, M_TEMP); 1069 return (error); 1070} 1071 1072SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1073 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1074 1075static int 1076tcp_getcred(SYSCTL_HANDLER_ARGS) 1077{ 1078 struct xucred xuc; 1079 struct sockaddr_in addrs[2]; 1080 struct inpcb *inp; 1081 int error; 1082 1083 error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED, 1084 SUSER_ALLOWJAIL); 1085 if (error) 1086 return (error); 1087 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1088 if (error) 1089 return (error); 1090 INP_INFO_RLOCK(&tcbinfo); 1091 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1092 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1093 if (inp == NULL) { 1094 error = ENOENT; 1095 goto outunlocked; 1096 } 1097 INP_LOCK(inp); 1098 if (inp->inp_socket == NULL) { 1099 error = ENOENT; 1100 goto out; 1101 } 1102 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1103 if (error) 1104 goto out; 1105 cru2x(inp->inp_socket->so_cred, &xuc); 1106out: 1107 INP_UNLOCK(inp); 1108outunlocked: 1109 INP_INFO_RUNLOCK(&tcbinfo); 1110 if (error == 0) 1111 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1112 return (error); 1113} 1114 1115SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1116 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1117 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1118 1119#ifdef INET6 1120static int 1121tcp6_getcred(SYSCTL_HANDLER_ARGS) 1122{ 1123 struct xucred xuc; 1124 struct sockaddr_in6 addrs[2]; 1125 struct inpcb *inp; 1126 int error, mapped = 0; 1127 1128 error = priv_check_cred(req->td->td_ucred, PRIV_NETINET_GETCRED, 1129 SUSER_ALLOWJAIL); 1130 if (error) 1131 return (error); 1132 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1133 if (error) 1134 return (error); 1135 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1136 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1137 return (error); 1138 } 1139 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1140 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1141 mapped = 1; 1142 else 1143 return (EINVAL); 1144 } 1145 1146 INP_INFO_RLOCK(&tcbinfo); 1147 if (mapped == 1) 1148 inp = in_pcblookup_hash(&tcbinfo, 1149 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1150 addrs[1].sin6_port, 1151 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1152 addrs[0].sin6_port, 1153 0, NULL); 1154 else 1155 inp = in6_pcblookup_hash(&tcbinfo, 1156 &addrs[1].sin6_addr, addrs[1].sin6_port, 1157 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1158 if (inp == NULL) { 1159 error = ENOENT; 1160 goto outunlocked; 1161 } 1162 INP_LOCK(inp); 1163 if (inp->inp_socket == NULL) { 1164 error = ENOENT; 1165 goto out; 1166 } 1167 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1168 if (error) 1169 goto out; 1170 cru2x(inp->inp_socket->so_cred, &xuc); 1171out: 1172 INP_UNLOCK(inp); 1173outunlocked: 1174 INP_INFO_RUNLOCK(&tcbinfo); 1175 if (error == 0) 1176 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1177 return (error); 1178} 1179 1180SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1181 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1182 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1183#endif 1184 1185 1186void 1187tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1188{ 1189 struct ip *ip = vip; 1190 struct tcphdr *th; 1191 struct in_addr faddr; 1192 struct inpcb *inp; 1193 struct tcpcb *tp; 1194 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1195 struct icmp *icp; 1196 struct in_conninfo inc; 1197 tcp_seq icmp_tcp_seq; 1198 int mtu; 1199 1200 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1201 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1202 return; 1203 1204 if (cmd == PRC_MSGSIZE) 1205 notify = tcp_mtudisc; 1206 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1207 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1208 notify = tcp_drop_syn_sent; 1209 /* 1210 * Redirects don't need to be handled up here. 1211 */ 1212 else if (PRC_IS_REDIRECT(cmd)) 1213 return; 1214 /* 1215 * Source quench is depreciated. 1216 */ 1217 else if (cmd == PRC_QUENCH) 1218 return; 1219 /* 1220 * Hostdead is ugly because it goes linearly through all PCBs. 1221 * XXX: We never get this from ICMP, otherwise it makes an 1222 * excellent DoS attack on machines with many connections. 1223 */ 1224 else if (cmd == PRC_HOSTDEAD) 1225 ip = NULL; 1226 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1227 return; 1228 if (ip != NULL) { 1229 icp = (struct icmp *)((caddr_t)ip 1230 - offsetof(struct icmp, icmp_ip)); 1231 th = (struct tcphdr *)((caddr_t)ip 1232 + (ip->ip_hl << 2)); 1233 INP_INFO_WLOCK(&tcbinfo); 1234 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1235 ip->ip_src, th->th_sport, 0, NULL); 1236 if (inp != NULL) { 1237 INP_LOCK(inp); 1238 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1239 !(inp->inp_vflag & INP_DROPPED) && 1240 !(inp->inp_socket == NULL)) { 1241 icmp_tcp_seq = htonl(th->th_seq); 1242 tp = intotcpcb(inp); 1243 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1244 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1245 if (cmd == PRC_MSGSIZE) { 1246 /* 1247 * MTU discovery: 1248 * If we got a needfrag set the MTU 1249 * in the route to the suggested new 1250 * value (if given) and then notify. 1251 */ 1252 bzero(&inc, sizeof(inc)); 1253 inc.inc_flags = 0; /* IPv4 */ 1254 inc.inc_faddr = faddr; 1255 1256 mtu = ntohs(icp->icmp_nextmtu); 1257 /* 1258 * If no alternative MTU was 1259 * proposed, try the next smaller 1260 * one. ip->ip_len has already 1261 * been swapped in icmp_input(). 1262 */ 1263 if (!mtu) 1264 mtu = ip_next_mtu(ip->ip_len, 1265 1); 1266 if (mtu < max(296, (tcp_minmss) 1267 + sizeof(struct tcpiphdr))) 1268 mtu = 0; 1269 if (!mtu) 1270 mtu = tcp_mssdflt 1271 + sizeof(struct tcpiphdr); 1272 /* 1273 * Only cache the the MTU if it 1274 * is smaller than the interface 1275 * or route MTU. tcp_mtudisc() 1276 * will do right thing by itself. 1277 */ 1278 if (mtu <= tcp_maxmtu(&inc, NULL)) 1279 tcp_hc_updatemtu(&inc, mtu); 1280 } 1281 1282 inp = (*notify)(inp, inetctlerrmap[cmd]); 1283 } 1284 } 1285 if (inp != NULL) 1286 INP_UNLOCK(inp); 1287 } else { 1288 inc.inc_fport = th->th_dport; 1289 inc.inc_lport = th->th_sport; 1290 inc.inc_faddr = faddr; 1291 inc.inc_laddr = ip->ip_src; 1292#ifdef INET6 1293 inc.inc_isipv6 = 0; 1294#endif 1295 syncache_unreach(&inc, th); 1296 } 1297 INP_INFO_WUNLOCK(&tcbinfo); 1298 } else 1299 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1300} 1301 1302#ifdef INET6 1303void 1304tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1305{ 1306 struct tcphdr th; 1307 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1308 struct ip6_hdr *ip6; 1309 struct mbuf *m; 1310 struct ip6ctlparam *ip6cp = NULL; 1311 const struct sockaddr_in6 *sa6_src = NULL; 1312 int off; 1313 struct tcp_portonly { 1314 u_int16_t th_sport; 1315 u_int16_t th_dport; 1316 } *thp; 1317 1318 if (sa->sa_family != AF_INET6 || 1319 sa->sa_len != sizeof(struct sockaddr_in6)) 1320 return; 1321 1322 if (cmd == PRC_MSGSIZE) 1323 notify = tcp_mtudisc; 1324 else if (!PRC_IS_REDIRECT(cmd) && 1325 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1326 return; 1327 /* Source quench is depreciated. */ 1328 else if (cmd == PRC_QUENCH) 1329 return; 1330 1331 /* if the parameter is from icmp6, decode it. */ 1332 if (d != NULL) { 1333 ip6cp = (struct ip6ctlparam *)d; 1334 m = ip6cp->ip6c_m; 1335 ip6 = ip6cp->ip6c_ip6; 1336 off = ip6cp->ip6c_off; 1337 sa6_src = ip6cp->ip6c_src; 1338 } else { 1339 m = NULL; 1340 ip6 = NULL; 1341 off = 0; /* fool gcc */ 1342 sa6_src = &sa6_any; 1343 } 1344 1345 if (ip6 != NULL) { 1346 struct in_conninfo inc; 1347 /* 1348 * XXX: We assume that when IPV6 is non NULL, 1349 * M and OFF are valid. 1350 */ 1351 1352 /* check if we can safely examine src and dst ports */ 1353 if (m->m_pkthdr.len < off + sizeof(*thp)) 1354 return; 1355 1356 bzero(&th, sizeof(th)); 1357 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1358 1359 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1360 (struct sockaddr *)ip6cp->ip6c_src, 1361 th.th_sport, cmd, NULL, notify); 1362 1363 inc.inc_fport = th.th_dport; 1364 inc.inc_lport = th.th_sport; 1365 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1366 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1367 inc.inc_isipv6 = 1; 1368 INP_INFO_WLOCK(&tcbinfo); 1369 syncache_unreach(&inc, &th); 1370 INP_INFO_WUNLOCK(&tcbinfo); 1371 } else 1372 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1373 0, cmd, NULL, notify); 1374} 1375#endif /* INET6 */ 1376 1377 1378/* 1379 * Following is where TCP initial sequence number generation occurs. 1380 * 1381 * There are two places where we must use initial sequence numbers: 1382 * 1. In SYN-ACK packets. 1383 * 2. In SYN packets. 1384 * 1385 * All ISNs for SYN-ACK packets are generated by the syncache. See 1386 * tcp_syncache.c for details. 1387 * 1388 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1389 * depends on this property. In addition, these ISNs should be 1390 * unguessable so as to prevent connection hijacking. To satisfy 1391 * the requirements of this situation, the algorithm outlined in 1392 * RFC 1948 is used, with only small modifications. 1393 * 1394 * Implementation details: 1395 * 1396 * Time is based off the system timer, and is corrected so that it 1397 * increases by one megabyte per second. This allows for proper 1398 * recycling on high speed LANs while still leaving over an hour 1399 * before rollover. 1400 * 1401 * As reading the *exact* system time is too expensive to be done 1402 * whenever setting up a TCP connection, we increment the time 1403 * offset in two ways. First, a small random positive increment 1404 * is added to isn_offset for each connection that is set up. 1405 * Second, the function tcp_isn_tick fires once per clock tick 1406 * and increments isn_offset as necessary so that sequence numbers 1407 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1408 * random positive increments serve only to ensure that the same 1409 * exact sequence number is never sent out twice (as could otherwise 1410 * happen when a port is recycled in less than the system tick 1411 * interval.) 1412 * 1413 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1414 * between seeding of isn_secret. This is normally set to zero, 1415 * as reseeding should not be necessary. 1416 * 1417 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1418 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1419 * general, this means holding an exclusive (write) lock. 1420 */ 1421 1422#define ISN_BYTES_PER_SECOND 1048576 1423#define ISN_STATIC_INCREMENT 4096 1424#define ISN_RANDOM_INCREMENT (4096 - 1) 1425 1426static u_char isn_secret[32]; 1427static int isn_last_reseed; 1428static u_int32_t isn_offset, isn_offset_old; 1429static MD5_CTX isn_ctx; 1430 1431tcp_seq 1432tcp_new_isn(struct tcpcb *tp) 1433{ 1434 u_int32_t md5_buffer[4]; 1435 tcp_seq new_isn; 1436 1437 INP_LOCK_ASSERT(tp->t_inpcb); 1438 1439 ISN_LOCK(); 1440 /* Seed if this is the first use, reseed if requested. */ 1441 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1442 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1443 < (u_int)ticks))) { 1444 read_random(&isn_secret, sizeof(isn_secret)); 1445 isn_last_reseed = ticks; 1446 } 1447 1448 /* Compute the md5 hash and return the ISN. */ 1449 MD5Init(&isn_ctx); 1450 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1451 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1452#ifdef INET6 1453 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1454 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1455 sizeof(struct in6_addr)); 1456 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1457 sizeof(struct in6_addr)); 1458 } else 1459#endif 1460 { 1461 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1462 sizeof(struct in_addr)); 1463 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1464 sizeof(struct in_addr)); 1465 } 1466 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1467 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1468 new_isn = (tcp_seq) md5_buffer[0]; 1469 isn_offset += ISN_STATIC_INCREMENT + 1470 (arc4random() & ISN_RANDOM_INCREMENT); 1471 new_isn += isn_offset; 1472 ISN_UNLOCK(); 1473 return (new_isn); 1474} 1475 1476/* 1477 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1478 * to keep time flowing at a relatively constant rate. If the random 1479 * increments have already pushed us past the projected offset, do nothing. 1480 */ 1481static void 1482tcp_isn_tick(void *xtp) 1483{ 1484 u_int32_t projected_offset; 1485 1486 ISN_LOCK(); 1487 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1488 1489 if (projected_offset > isn_offset) 1490 isn_offset = projected_offset; 1491 1492 isn_offset_old = isn_offset; 1493 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1494 ISN_UNLOCK(); 1495} 1496 1497/* 1498 * When a specific ICMP unreachable message is received and the 1499 * connection state is SYN-SENT, drop the connection. This behavior 1500 * is controlled by the icmp_may_rst sysctl. 1501 */ 1502struct inpcb * 1503tcp_drop_syn_sent(struct inpcb *inp, int errno) 1504{ 1505 struct tcpcb *tp; 1506 1507 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1508 INP_LOCK_ASSERT(inp); 1509 1510 if ((inp->inp_vflag & INP_TIMEWAIT) || 1511 (inp->inp_vflag & INP_DROPPED)) 1512 return (inp); 1513 1514 tp = intotcpcb(inp); 1515 if (tp->t_state != TCPS_SYN_SENT) 1516 return (inp); 1517 1518 tp = tcp_drop(tp, errno); 1519 if (tp != NULL) 1520 return (inp); 1521 else 1522 return (NULL); 1523} 1524 1525/* 1526 * When `need fragmentation' ICMP is received, update our idea of the MSS 1527 * based on the new value in the route. Also nudge TCP to send something, 1528 * since we know the packet we just sent was dropped. 1529 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1530 */ 1531struct inpcb * 1532tcp_mtudisc(struct inpcb *inp, int errno) 1533{ 1534 struct tcpcb *tp; 1535 struct socket *so = inp->inp_socket; 1536 u_int maxmtu; 1537 u_int romtu; 1538 int mss; 1539#ifdef INET6 1540 int isipv6; 1541#endif /* INET6 */ 1542 1543 INP_LOCK_ASSERT(inp); 1544 if ((inp->inp_vflag & INP_TIMEWAIT) || 1545 (inp->inp_vflag & INP_DROPPED)) 1546 return (inp); 1547 1548 tp = intotcpcb(inp); 1549 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1550 1551#ifdef INET6 1552 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1553#endif 1554 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1555 romtu = 1556#ifdef INET6 1557 isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) : 1558#endif /* INET6 */ 1559 tcp_maxmtu(&inp->inp_inc, NULL); 1560 if (!maxmtu) 1561 maxmtu = romtu; 1562 else 1563 maxmtu = min(maxmtu, romtu); 1564 if (!maxmtu) { 1565 tp->t_maxopd = tp->t_maxseg = 1566#ifdef INET6 1567 isipv6 ? tcp_v6mssdflt : 1568#endif /* INET6 */ 1569 tcp_mssdflt; 1570 return (inp); 1571 } 1572 mss = maxmtu - 1573#ifdef INET6 1574 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1575#endif /* INET6 */ 1576 sizeof(struct tcpiphdr) 1577#ifdef INET6 1578 ) 1579#endif /* INET6 */ 1580 ; 1581 1582 /* 1583 * XXX - The above conditional probably violates the TCP 1584 * spec. The problem is that, since we don't know the 1585 * other end's MSS, we are supposed to use a conservative 1586 * default. But, if we do that, then MTU discovery will 1587 * never actually take place, because the conservative 1588 * default is much less than the MTUs typically seen 1589 * on the Internet today. For the moment, we'll sweep 1590 * this under the carpet. 1591 * 1592 * The conservative default might not actually be a problem 1593 * if the only case this occurs is when sending an initial 1594 * SYN with options and data to a host we've never talked 1595 * to before. Then, they will reply with an MSS value which 1596 * will get recorded and the new parameters should get 1597 * recomputed. For Further Study. 1598 */ 1599 if (tp->t_maxopd <= mss) 1600 return (inp); 1601 tp->t_maxopd = mss; 1602 1603 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1604 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1605 mss -= TCPOLEN_TSTAMP_APPA; 1606#if (MCLBYTES & (MCLBYTES - 1)) == 0 1607 if (mss > MCLBYTES) 1608 mss &= ~(MCLBYTES-1); 1609#else 1610 if (mss > MCLBYTES) 1611 mss = mss / MCLBYTES * MCLBYTES; 1612#endif 1613 if (so->so_snd.sb_hiwat < mss) 1614 mss = so->so_snd.sb_hiwat; 1615 1616 tp->t_maxseg = mss; 1617 1618 tcpstat.tcps_mturesent++; 1619 tp->t_rtttime = 0; 1620 tp->snd_nxt = tp->snd_una; 1621 tcp_free_sackholes(tp); 1622 tp->snd_recover = tp->snd_max; 1623 if (tp->sack_enable) 1624 EXIT_FASTRECOVERY(tp); 1625 tcp_output(tp); 1626 return (inp); 1627} 1628 1629/* 1630 * Look-up the routing entry to the peer of this inpcb. If no route 1631 * is found and it cannot be allocated, then return NULL. This routine 1632 * is called by TCP routines that access the rmx structure and by tcp_mss 1633 * to get the interface MTU. 1634 */ 1635u_long 1636tcp_maxmtu(struct in_conninfo *inc, int *flags) 1637{ 1638 struct route sro; 1639 struct sockaddr_in *dst; 1640 struct ifnet *ifp; 1641 u_long maxmtu = 0; 1642 1643 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1644 1645 bzero(&sro, sizeof(sro)); 1646 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1647 dst = (struct sockaddr_in *)&sro.ro_dst; 1648 dst->sin_family = AF_INET; 1649 dst->sin_len = sizeof(*dst); 1650 dst->sin_addr = inc->inc_faddr; 1651 rtalloc_ign(&sro, RTF_CLONING); 1652 } 1653 if (sro.ro_rt != NULL) { 1654 ifp = sro.ro_rt->rt_ifp; 1655 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1656 maxmtu = ifp->if_mtu; 1657 else 1658 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1659 1660 /* Report additional interface capabilities. */ 1661 if (flags != NULL) { 1662 if (ifp->if_capenable & IFCAP_TSO4 && 1663 ifp->if_hwassist & CSUM_TSO) 1664 *flags |= CSUM_TSO; 1665 } 1666 RTFREE(sro.ro_rt); 1667 } 1668 return (maxmtu); 1669} 1670 1671#ifdef INET6 1672u_long 1673tcp_maxmtu6(struct in_conninfo *inc, int *flags) 1674{ 1675 struct route_in6 sro6; 1676 struct ifnet *ifp; 1677 u_long maxmtu = 0; 1678 1679 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1680 1681 bzero(&sro6, sizeof(sro6)); 1682 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1683 sro6.ro_dst.sin6_family = AF_INET6; 1684 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1685 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1686 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1687 } 1688 if (sro6.ro_rt != NULL) { 1689 ifp = sro6.ro_rt->rt_ifp; 1690 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1691 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1692 else 1693 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1694 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1695 1696 /* Report additional interface capabilities. */ 1697 if (flags != NULL) { 1698 if (ifp->if_capenable & IFCAP_TSO6 && 1699 ifp->if_hwassist & CSUM_TSO) 1700 *flags |= CSUM_TSO; 1701 } 1702 RTFREE(sro6.ro_rt); 1703 } 1704 1705 return (maxmtu); 1706} 1707#endif /* INET6 */ 1708 1709#ifdef IPSEC 1710/* compute ESP/AH header size for TCP, including outer IP header. */ 1711size_t 1712ipsec_hdrsiz_tcp(struct tcpcb *tp) 1713{ 1714 struct inpcb *inp; 1715 struct mbuf *m; 1716 size_t hdrsiz; 1717 struct ip *ip; 1718#ifdef INET6 1719 struct ip6_hdr *ip6; 1720#endif 1721 struct tcphdr *th; 1722 1723 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1724 return (0); 1725 MGETHDR(m, M_DONTWAIT, MT_DATA); 1726 if (!m) 1727 return (0); 1728 1729#ifdef INET6 1730 if ((inp->inp_vflag & INP_IPV6) != 0) { 1731 ip6 = mtod(m, struct ip6_hdr *); 1732 th = (struct tcphdr *)(ip6 + 1); 1733 m->m_pkthdr.len = m->m_len = 1734 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1735 tcpip_fillheaders(inp, ip6, th); 1736 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1737 } else 1738#endif /* INET6 */ 1739 { 1740 ip = mtod(m, struct ip *); 1741 th = (struct tcphdr *)(ip + 1); 1742 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1743 tcpip_fillheaders(inp, ip, th); 1744 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1745 } 1746 1747 m_free(m); 1748 return (hdrsiz); 1749} 1750#endif /*IPSEC*/ 1751 1752/* 1753 * Move a TCP connection into TIME_WAIT state. 1754 * tcbinfo is locked. 1755 * inp is locked, and is unlocked before returning. 1756 */ 1757void 1758tcp_twstart(struct tcpcb *tp) 1759{ 1760 struct tcptw *tw; 1761 struct inpcb *inp = tp->t_inpcb; 1762 int acknow; 1763 struct socket *so; 1764 1765 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1766 INP_LOCK_ASSERT(inp); 1767 1768 if (nolocaltimewait && in_localip(inp->inp_faddr)) { 1769 tp = tcp_close(tp); 1770 if (tp != NULL) 1771 INP_UNLOCK(inp); 1772 return; 1773 } 1774 1775 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1776 if (tw == NULL) { 1777 tw = tcp_timer_2msl_tw(1); 1778 if (tw == NULL) { 1779 tp = tcp_close(tp); 1780 if (tp != NULL) 1781 INP_UNLOCK(inp); 1782 return; 1783 } 1784 } 1785 tw->tw_inpcb = inp; 1786 1787 /* 1788 * Recover last window size sent. 1789 */ 1790 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1791 1792 /* 1793 * Set t_recent if timestamps are used on the connection. 1794 */ 1795 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1796 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1797 tw->t_recent = tp->ts_recent; 1798 else 1799 tw->t_recent = 0; 1800 1801 tw->snd_nxt = tp->snd_nxt; 1802 tw->rcv_nxt = tp->rcv_nxt; 1803 tw->iss = tp->iss; 1804 tw->irs = tp->irs; 1805 tw->t_starttime = tp->t_starttime; 1806 tw->tw_time = 0; 1807 1808/* XXX 1809 * If this code will 1810 * be used for fin-wait-2 state also, then we may need 1811 * a ts_recent from the last segment. 1812 */ 1813 acknow = tp->t_flags & TF_ACKNOW; 1814 1815 /* 1816 * First, discard tcpcb state, which includes stopping its timers and 1817 * freeing it. tcp_discardcb() used to also release the inpcb, but 1818 * that work is now done in the caller. 1819 * 1820 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 1821 * and might not be needed here any longer. 1822 */ 1823 tcp_discardcb(tp); 1824 so = inp->inp_socket; 1825 soisdisconnected(so); 1826 SOCK_LOCK(so); 1827 tw->tw_cred = crhold(so->so_cred); 1828 tw->tw_so_options = so->so_options; 1829 SOCK_UNLOCK(so); 1830 if (acknow) 1831 tcp_twrespond(tw, TH_ACK); 1832 inp->inp_ppcb = tw; 1833 inp->inp_vflag |= INP_TIMEWAIT; 1834 tcp_timer_2msl_reset(tw, 0); 1835 1836 /* 1837 * If the inpcb owns the sole reference to the socket, then we can 1838 * detach and free the socket as it is not needed in time wait. 1839 */ 1840 if (inp->inp_vflag & INP_SOCKREF) { 1841 KASSERT(so->so_state & SS_PROTOREF, 1842 ("tcp_twstart: !SS_PROTOREF")); 1843 inp->inp_vflag &= ~INP_SOCKREF; 1844 INP_UNLOCK(inp); 1845 ACCEPT_LOCK(); 1846 SOCK_LOCK(so); 1847 so->so_state &= ~SS_PROTOREF; 1848 sofree(so); 1849 } else 1850 INP_UNLOCK(inp); 1851} 1852 1853#if 0 1854/* 1855 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1856 * the actual rate is slightly higher due to the addition of 1857 * random positive increments. 1858 * 1859 * Most other new OSes use semi-randomized ISN values, so we 1860 * do not need to worry about them. 1861 */ 1862#define MS_ISN_BYTES_PER_SECOND 250000 1863 1864/* 1865 * Determine if the ISN we will generate has advanced beyond the last 1866 * sequence number used by the previous connection. If so, indicate 1867 * that it is safe to recycle this tw socket by returning 1. 1868 */ 1869int 1870tcp_twrecycleable(struct tcptw *tw) 1871{ 1872 tcp_seq new_iss = tw->iss; 1873 tcp_seq new_irs = tw->irs; 1874 1875 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1876 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1877 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1878 1879 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1880 return (1); 1881 else 1882 return (0); 1883} 1884#endif 1885 1886void 1887tcp_twclose(struct tcptw *tw, int reuse) 1888{ 1889 struct socket *so; 1890 struct inpcb *inp; 1891 1892 /* 1893 * At this point, we are in one of two situations: 1894 * 1895 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1896 * all state. 1897 * 1898 * (2) We have a socket -- if we own a reference, release it and 1899 * notify the socket layer. 1900 */ 1901 inp = tw->tw_inpcb; 1902 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1903 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1904 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1905 INP_LOCK_ASSERT(inp); 1906 1907 tw->tw_inpcb = NULL; 1908 tcp_timer_2msl_stop(tw); 1909 inp->inp_ppcb = NULL; 1910 in_pcbdrop(inp); 1911 1912 so = inp->inp_socket; 1913 if (so != NULL) { 1914 /* 1915 * If there's a socket, handle two cases: first, we own a 1916 * strong reference, which we will now release, or we don't 1917 * in which case another reference exists (XXXRW: think 1918 * about this more), and we don't need to take action. 1919 */ 1920 if (inp->inp_vflag & INP_SOCKREF) { 1921 inp->inp_vflag &= ~INP_SOCKREF; 1922 INP_UNLOCK(inp); 1923 ACCEPT_LOCK(); 1924 SOCK_LOCK(so); 1925 KASSERT(so->so_state & SS_PROTOREF, 1926 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1927 so->so_state &= ~SS_PROTOREF; 1928 sofree(so); 1929 } else { 1930 /* 1931 * If we don't own the only reference, the socket and 1932 * inpcb need to be left around to be handled by 1933 * tcp_usr_detach() later. 1934 */ 1935 INP_UNLOCK(inp); 1936 } 1937 } else { 1938#ifdef INET6 1939 if (inp->inp_vflag & INP_IPV6PROTO) 1940 in6_pcbfree(inp); 1941 else 1942#endif 1943 in_pcbfree(inp); 1944 } 1945 tcpstat.tcps_closed++; 1946 crfree(tw->tw_cred); 1947 tw->tw_cred = NULL; 1948 if (reuse) 1949 return; 1950 uma_zfree(tcptw_zone, tw); 1951} 1952 1953int 1954tcp_twrespond(struct tcptw *tw, int flags) 1955{ 1956 struct inpcb *inp = tw->tw_inpcb; 1957 struct tcphdr *th; 1958 struct mbuf *m; 1959 struct ip *ip = NULL; 1960 u_int8_t *optp; 1961 u_int hdrlen, optlen; 1962 int error; 1963#ifdef INET6 1964 struct ip6_hdr *ip6 = NULL; 1965 int isipv6 = inp->inp_inc.inc_isipv6; 1966#endif 1967 1968 INP_LOCK_ASSERT(inp); 1969 1970 m = m_gethdr(M_DONTWAIT, MT_DATA); 1971 if (m == NULL) 1972 return (ENOBUFS); 1973 m->m_data += max_linkhdr; 1974 1975#ifdef MAC 1976 mac_create_mbuf_from_inpcb(inp, m); 1977#endif 1978 1979#ifdef INET6 1980 if (isipv6) { 1981 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1982 ip6 = mtod(m, struct ip6_hdr *); 1983 th = (struct tcphdr *)(ip6 + 1); 1984 tcpip_fillheaders(inp, ip6, th); 1985 } else 1986#endif 1987 { 1988 hdrlen = sizeof(struct tcpiphdr); 1989 ip = mtod(m, struct ip *); 1990 th = (struct tcphdr *)(ip + 1); 1991 tcpip_fillheaders(inp, ip, th); 1992 } 1993 optp = (u_int8_t *)(th + 1); 1994 1995 /* 1996 * Send a timestamp and echo-reply if both our side and our peer 1997 * have sent timestamps in our SYN's and this is not a RST. 1998 */ 1999 if (tw->t_recent && flags == TH_ACK) { 2000 u_int32_t *lp = (u_int32_t *)optp; 2001 2002 /* Form timestamp option as shown in appendix A of RFC 1323. */ 2003 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 2004 *lp++ = htonl(ticks); 2005 *lp = htonl(tw->t_recent); 2006 optp += TCPOLEN_TSTAMP_APPA; 2007 } 2008 2009 optlen = optp - (u_int8_t *)(th + 1); 2010 2011 m->m_len = hdrlen + optlen; 2012 m->m_pkthdr.len = m->m_len; 2013 2014 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 2015 2016 th->th_seq = htonl(tw->snd_nxt); 2017 th->th_ack = htonl(tw->rcv_nxt); 2018 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 2019 th->th_flags = flags; 2020 th->th_win = htons(tw->last_win); 2021 2022#ifdef INET6 2023 if (isipv6) { 2024 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 2025 sizeof(struct tcphdr) + optlen); 2026 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2027 error = ip6_output(m, inp->in6p_outputopts, NULL, 2028 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 2029 } else 2030#endif 2031 { 2032 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2033 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 2034 m->m_pkthdr.csum_flags = CSUM_TCP; 2035 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2036 ip->ip_len = m->m_pkthdr.len; 2037 if (path_mtu_discovery) 2038 ip->ip_off |= IP_DF; 2039 error = ip_output(m, inp->inp_options, NULL, 2040 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 2041 NULL, inp); 2042 } 2043 if (flags & TH_ACK) 2044 tcpstat.tcps_sndacks++; 2045 else 2046 tcpstat.tcps_sndctrl++; 2047 tcpstat.tcps_sndtotal++; 2048 return (error); 2049} 2050 2051/* 2052 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 2053 * 2054 * This code attempts to calculate the bandwidth-delay product as a 2055 * means of determining the optimal window size to maximize bandwidth, 2056 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 2057 * routers. This code also does a fairly good job keeping RTTs in check 2058 * across slow links like modems. We implement an algorithm which is very 2059 * similar (but not meant to be) TCP/Vegas. The code operates on the 2060 * transmitter side of a TCP connection and so only effects the transmit 2061 * side of the connection. 2062 * 2063 * BACKGROUND: TCP makes no provision for the management of buffer space 2064 * at the end points or at the intermediate routers and switches. A TCP 2065 * stream, whether using NewReno or not, will eventually buffer as 2066 * many packets as it is able and the only reason this typically works is 2067 * due to the fairly small default buffers made available for a connection 2068 * (typicaly 16K or 32K). As machines use larger windows and/or window 2069 * scaling it is now fairly easy for even a single TCP connection to blow-out 2070 * all available buffer space not only on the local interface, but on 2071 * intermediate routers and switches as well. NewReno makes a misguided 2072 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2073 * then backing off, then steadily increasing the window again until another 2074 * failure occurs, ad-infinitum. This results in terrible oscillation that 2075 * is only made worse as network loads increase and the idea of intentionally 2076 * blowing out network buffers is, frankly, a terrible way to manage network 2077 * resources. 2078 * 2079 * It is far better to limit the transmit window prior to the failure 2080 * condition being achieved. There are two general ways to do this: First 2081 * you can 'scan' through different transmit window sizes and locate the 2082 * point where the RTT stops increasing, indicating that you have filled the 2083 * pipe, then scan backwards until you note that RTT stops decreasing, then 2084 * repeat ad-infinitum. This method works in principle but has severe 2085 * implementation issues due to RTT variances, timer granularity, and 2086 * instability in the algorithm which can lead to many false positives and 2087 * create oscillations as well as interact badly with other TCP streams 2088 * implementing the same algorithm. 2089 * 2090 * The second method is to limit the window to the bandwidth delay product 2091 * of the link. This is the method we implement. RTT variances and our 2092 * own manipulation of the congestion window, bwnd, can potentially 2093 * destabilize the algorithm. For this reason we have to stabilize the 2094 * elements used to calculate the window. We do this by using the minimum 2095 * observed RTT, the long term average of the observed bandwidth, and 2096 * by adding two segments worth of slop. It isn't perfect but it is able 2097 * to react to changing conditions and gives us a very stable basis on 2098 * which to extend the algorithm. 2099 */ 2100void 2101tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2102{ 2103 u_long bw; 2104 u_long bwnd; 2105 int save_ticks; 2106 2107 INP_LOCK_ASSERT(tp->t_inpcb); 2108 2109 /* 2110 * If inflight_enable is disabled in the middle of a tcp connection, 2111 * make sure snd_bwnd is effectively disabled. 2112 */ 2113 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2114 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2115 tp->snd_bandwidth = 0; 2116 return; 2117 } 2118 2119 /* 2120 * Figure out the bandwidth. Due to the tick granularity this 2121 * is a very rough number and it MUST be averaged over a fairly 2122 * long period of time. XXX we need to take into account a link 2123 * that is not using all available bandwidth, but for now our 2124 * slop will ramp us up if this case occurs and the bandwidth later 2125 * increases. 2126 * 2127 * Note: if ticks rollover 'bw' may wind up negative. We must 2128 * effectively reset t_bw_rtttime for this case. 2129 */ 2130 save_ticks = ticks; 2131 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2132 return; 2133 2134 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2135 (save_ticks - tp->t_bw_rtttime); 2136 tp->t_bw_rtttime = save_ticks; 2137 tp->t_bw_rtseq = ack_seq; 2138 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2139 return; 2140 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2141 2142 tp->snd_bandwidth = bw; 2143 2144 /* 2145 * Calculate the semi-static bandwidth delay product, plus two maximal 2146 * segments. The additional slop puts us squarely in the sweet 2147 * spot and also handles the bandwidth run-up case and stabilization. 2148 * Without the slop we could be locking ourselves into a lower 2149 * bandwidth. 2150 * 2151 * Situations Handled: 2152 * (1) Prevents over-queueing of packets on LANs, especially on 2153 * high speed LANs, allowing larger TCP buffers to be 2154 * specified, and also does a good job preventing 2155 * over-queueing of packets over choke points like modems 2156 * (at least for the transmit side). 2157 * 2158 * (2) Is able to handle changing network loads (bandwidth 2159 * drops so bwnd drops, bandwidth increases so bwnd 2160 * increases). 2161 * 2162 * (3) Theoretically should stabilize in the face of multiple 2163 * connections implementing the same algorithm (this may need 2164 * a little work). 2165 * 2166 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2167 * be adjusted with a sysctl but typically only needs to be 2168 * on very slow connections. A value no smaller then 5 2169 * should be used, but only reduce this default if you have 2170 * no other choice. 2171 */ 2172#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2173 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2174#undef USERTT 2175 2176 if (tcp_inflight_debug > 0) { 2177 static int ltime; 2178 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2179 ltime = ticks; 2180 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2181 tp, 2182 bw, 2183 tp->t_rttbest, 2184 tp->t_srtt, 2185 bwnd 2186 ); 2187 } 2188 } 2189 if ((long)bwnd < tcp_inflight_min) 2190 bwnd = tcp_inflight_min; 2191 if (bwnd > tcp_inflight_max) 2192 bwnd = tcp_inflight_max; 2193 if ((long)bwnd < tp->t_maxseg * 2) 2194 bwnd = tp->t_maxseg * 2; 2195 tp->snd_bwnd = bwnd; 2196} 2197 2198#ifdef TCP_SIGNATURE 2199/* 2200 * Callback function invoked by m_apply() to digest TCP segment data 2201 * contained within an mbuf chain. 2202 */ 2203static int 2204tcp_signature_apply(void *fstate, void *data, u_int len) 2205{ 2206 2207 MD5Update(fstate, (u_char *)data, len); 2208 return (0); 2209} 2210 2211/* 2212 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2213 * 2214 * Parameters: 2215 * m pointer to head of mbuf chain 2216 * off0 offset to TCP header within the mbuf chain 2217 * len length of TCP segment data, excluding options 2218 * optlen length of TCP segment options 2219 * buf pointer to storage for computed MD5 digest 2220 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2221 * 2222 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2223 * When called from tcp_input(), we can be sure that th_sum has been 2224 * zeroed out and verified already. 2225 * 2226 * This function is for IPv4 use only. Calling this function with an 2227 * IPv6 packet in the mbuf chain will yield undefined results. 2228 * 2229 * Return 0 if successful, otherwise return -1. 2230 * 2231 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2232 * search with the destination IP address, and a 'magic SPI' to be 2233 * determined by the application. This is hardcoded elsewhere to 1179 2234 * right now. Another branch of this code exists which uses the SPD to 2235 * specify per-application flows but it is unstable. 2236 */ 2237int 2238tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2239 u_char *buf, u_int direction) 2240{ 2241 union sockaddr_union dst; 2242 struct ippseudo ippseudo; 2243 MD5_CTX ctx; 2244 int doff; 2245 struct ip *ip; 2246 struct ipovly *ipovly; 2247 struct secasvar *sav; 2248 struct tcphdr *th; 2249 u_short savecsum; 2250 2251 KASSERT(m != NULL, ("NULL mbuf chain")); 2252 KASSERT(buf != NULL, ("NULL signature pointer")); 2253 2254 /* Extract the destination from the IP header in the mbuf. */ 2255 ip = mtod(m, struct ip *); 2256 bzero(&dst, sizeof(union sockaddr_union)); 2257 dst.sa.sa_len = sizeof(struct sockaddr_in); 2258 dst.sa.sa_family = AF_INET; 2259 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2260 ip->ip_src : ip->ip_dst; 2261 2262 /* Look up an SADB entry which matches the address of the peer. */ 2263 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2264 if (sav == NULL) { 2265 printf("%s: SADB lookup failed for %s\n", __func__, 2266 inet_ntoa(dst.sin.sin_addr)); 2267 return (EINVAL); 2268 } 2269 2270 MD5Init(&ctx); 2271 ipovly = (struct ipovly *)ip; 2272 th = (struct tcphdr *)((u_char *)ip + off0); 2273 doff = off0 + sizeof(struct tcphdr) + optlen; 2274 2275 /* 2276 * Step 1: Update MD5 hash with IP pseudo-header. 2277 * 2278 * XXX The ippseudo header MUST be digested in network byte order, 2279 * or else we'll fail the regression test. Assume all fields we've 2280 * been doing arithmetic on have been in host byte order. 2281 * XXX One cannot depend on ipovly->ih_len here. When called from 2282 * tcp_output(), the underlying ip_len member has not yet been set. 2283 */ 2284 ippseudo.ippseudo_src = ipovly->ih_src; 2285 ippseudo.ippseudo_dst = ipovly->ih_dst; 2286 ippseudo.ippseudo_pad = 0; 2287 ippseudo.ippseudo_p = IPPROTO_TCP; 2288 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2289 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2290 2291 /* 2292 * Step 2: Update MD5 hash with TCP header, excluding options. 2293 * The TCP checksum must be set to zero. 2294 */ 2295 savecsum = th->th_sum; 2296 th->th_sum = 0; 2297 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2298 th->th_sum = savecsum; 2299 2300 /* 2301 * Step 3: Update MD5 hash with TCP segment data. 2302 * Use m_apply() to avoid an early m_pullup(). 2303 */ 2304 if (len > 0) 2305 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2306 2307 /* 2308 * Step 4: Update MD5 hash with shared secret. 2309 */ 2310 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2311 MD5Final(buf, &ctx); 2312 2313 key_sa_recordxfer(sav, m); 2314 KEY_FREESAV(&sav); 2315 return (0); 2316} 2317#endif /* TCP_SIGNATURE */ 2318 2319static int 2320sysctl_drop(SYSCTL_HANDLER_ARGS) 2321{ 2322 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2323 struct sockaddr_storage addrs[2]; 2324 struct inpcb *inp; 2325 struct tcpcb *tp; 2326 struct tcptw *tw; 2327 struct sockaddr_in *fin, *lin; 2328#ifdef INET6 2329 struct sockaddr_in6 *fin6, *lin6; 2330 struct in6_addr f6, l6; 2331#endif 2332 int error; 2333 2334 inp = NULL; 2335 fin = lin = NULL; 2336#ifdef INET6 2337 fin6 = lin6 = NULL; 2338#endif 2339 error = 0; 2340 2341 if (req->oldptr != NULL || req->oldlen != 0) 2342 return (EINVAL); 2343 if (req->newptr == NULL) 2344 return (EPERM); 2345 if (req->newlen < sizeof(addrs)) 2346 return (ENOMEM); 2347 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2348 if (error) 2349 return (error); 2350 2351 switch (addrs[0].ss_family) { 2352#ifdef INET6 2353 case AF_INET6: 2354 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2355 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2356 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2357 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2358 return (EINVAL); 2359 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2360 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2361 return (EINVAL); 2362 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2363 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2364 fin = (struct sockaddr_in *)&addrs[0]; 2365 lin = (struct sockaddr_in *)&addrs[1]; 2366 break; 2367 } 2368 error = sa6_embedscope(fin6, ip6_use_defzone); 2369 if (error) 2370 return (error); 2371 error = sa6_embedscope(lin6, ip6_use_defzone); 2372 if (error) 2373 return (error); 2374 break; 2375#endif 2376 case AF_INET: 2377 fin = (struct sockaddr_in *)&addrs[0]; 2378 lin = (struct sockaddr_in *)&addrs[1]; 2379 if (fin->sin_len != sizeof(struct sockaddr_in) || 2380 lin->sin_len != sizeof(struct sockaddr_in)) 2381 return (EINVAL); 2382 break; 2383 default: 2384 return (EINVAL); 2385 } 2386 INP_INFO_WLOCK(&tcbinfo); 2387 switch (addrs[0].ss_family) { 2388#ifdef INET6 2389 case AF_INET6: 2390 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2391 &l6, lin6->sin6_port, 0, NULL); 2392 break; 2393#endif 2394 case AF_INET: 2395 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2396 lin->sin_addr, lin->sin_port, 0, NULL); 2397 break; 2398 } 2399 if (inp != NULL) { 2400 INP_LOCK(inp); 2401 if (inp->inp_vflag & INP_TIMEWAIT) { 2402 /* 2403 * XXXRW: There currently exists a state where an 2404 * inpcb is present, but its timewait state has been 2405 * discarded. For now, don't allow dropping of this 2406 * type of inpcb. 2407 */ 2408 tw = intotw(inp); 2409 if (tw != NULL) 2410 tcp_twclose(tw, 0); 2411 } else if (!(inp->inp_vflag & INP_DROPPED) && 2412 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2413 tp = intotcpcb(inp); 2414 tcp_drop(tp, ECONNABORTED); 2415 } 2416 INP_UNLOCK(inp); 2417 } else 2418 error = ESRCH; 2419 INP_INFO_WUNLOCK(&tcbinfo); 2420 return (error); 2421} 2422 2423SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2424 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2425 0, sysctl_drop, "", "Drop TCP connection"); 2426