tcp_timewait.c revision 161226
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 161226 2006-08-11 21:15:23Z mohans $ 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 180static int maxtcptw; 181SYSCTL_INT(_net_inet_tcp, OID_AUTO, maxtcptw, CTLFLAG_RDTUN, 182 &maxtcptw, 0, "Maximum number of compressed TCP TIME_WAIT entries"); 183 184/* 185 * TCP bandwidth limiting sysctls. Note that the default lower bound of 186 * 1024 exists only for debugging. A good production default would be 187 * something like 6100. 188 */ 189SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 190 "TCP inflight data limiting"); 191 192static int tcp_inflight_enable = 1; 193SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 194 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 195 196static int tcp_inflight_debug = 0; 197SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 198 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 199 200static int tcp_inflight_rttthresh; 201SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 202 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 203 "RTT threshold below which inflight will deactivate itself"); 204 205static int tcp_inflight_min = 6144; 206SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 207 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 208 209static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 210SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 211 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 212 213static int tcp_inflight_stab = 20; 214SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 215 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 216 217uma_zone_t sack_hole_zone; 218 219static struct inpcb *tcp_notify(struct inpcb *, int); 220static void tcp_isn_tick(void *); 221 222/* 223 * Target size of TCP PCB hash tables. Must be a power of two. 224 * 225 * Note that this can be overridden by the kernel environment 226 * variable net.inet.tcp.tcbhashsize 227 */ 228#ifndef TCBHASHSIZE 229#define TCBHASHSIZE 512 230#endif 231 232/* 233 * XXX 234 * Callouts should be moved into struct tcp directly. They are currently 235 * separate because the tcpcb structure is exported to userland for sysctl 236 * parsing purposes, which do not know about callouts. 237 */ 238struct tcpcb_mem { 239 struct tcpcb tcb; 240 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 241 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 242}; 243 244static uma_zone_t tcpcb_zone; 245static uma_zone_t tcptw_zone; 246struct callout isn_callout; 247static struct mtx isn_mtx; 248 249#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 250#define ISN_LOCK() mtx_lock(&isn_mtx) 251#define ISN_UNLOCK() mtx_unlock(&isn_mtx) 252 253/* 254 * TCP initialization. 255 */ 256static void 257tcp_zone_change(void *tag) 258{ 259 260 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 261 uma_zone_set_max(tcpcb_zone, maxsockets); 262 uma_zone_set_max(tcptw_zone, maxsockets / 5); 263} 264 265static int 266tcp_inpcb_init(void *mem, int size, int flags) 267{ 268 struct inpcb *inp = (struct inpcb *) mem; 269 INP_LOCK_INIT(inp, "inp", "tcpinp"); 270 return (0); 271} 272 273void 274tcp_init(void) 275{ 276 int hashsize = TCBHASHSIZE; 277 278 tcp_delacktime = TCPTV_DELACK; 279 tcp_keepinit = TCPTV_KEEP_INIT; 280 tcp_keepidle = TCPTV_KEEP_IDLE; 281 tcp_keepintvl = TCPTV_KEEPINTVL; 282 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 283 tcp_msl = TCPTV_MSL; 284 tcp_rexmit_min = TCPTV_MIN; 285 tcp_rexmit_slop = TCPTV_CPU_VAR; 286 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 287 288 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 289 LIST_INIT(&tcb); 290 tcbinfo.listhead = &tcb; 291 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 292 if (!powerof2(hashsize)) { 293 printf("WARNING: TCB hash size not a power of 2\n"); 294 hashsize = 512; /* safe default */ 295 } 296 tcp_tcbhashsize = hashsize; 297 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 298 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 299 &tcbinfo.porthashmask); 300 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 301 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 302 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 303#ifdef INET6 304#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 305#else /* INET6 */ 306#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 307#endif /* INET6 */ 308 if (max_protohdr < TCP_MINPROTOHDR) 309 max_protohdr = TCP_MINPROTOHDR; 310 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 311 panic("tcp_init"); 312#undef TCP_MINPROTOHDR 313 /* 314 * These have to be type stable for the benefit of the timers. 315 */ 316 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 317 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 318 uma_zone_set_max(tcpcb_zone, maxsockets); 319 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 320 if (maxtcptw == 0) 321 maxtcptw = maxsockets / 5; 322 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 323 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 324 uma_zone_set_max(tcptw_zone, maxtcptw); 325 tcp_timer_init(); 326 syncache_init(); 327 tcp_hc_init(); 328 tcp_reass_init(); 329 ISN_LOCK_INIT(); 330 callout_init(&isn_callout, CALLOUT_MPSAFE); 331 tcp_isn_tick(NULL); 332 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 333 SHUTDOWN_PRI_DEFAULT); 334 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 335 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 336 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 337 EVENTHANDLER_PRI_ANY); 338} 339 340void 341tcp_fini(void *xtp) 342{ 343 344 callout_stop(&isn_callout); 345} 346 347/* 348 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 349 * tcp_template used to store this data in mbufs, but we now recopy it out 350 * of the tcpcb each time to conserve mbufs. 351 */ 352void 353tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 354{ 355 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 356 357 INP_LOCK_ASSERT(inp); 358 359#ifdef INET6 360 if ((inp->inp_vflag & INP_IPV6) != 0) { 361 struct ip6_hdr *ip6; 362 363 ip6 = (struct ip6_hdr *)ip_ptr; 364 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 365 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 366 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 367 (IPV6_VERSION & IPV6_VERSION_MASK); 368 ip6->ip6_nxt = IPPROTO_TCP; 369 ip6->ip6_plen = sizeof(struct tcphdr); 370 ip6->ip6_src = inp->in6p_laddr; 371 ip6->ip6_dst = inp->in6p_faddr; 372 } else 373#endif 374 { 375 struct ip *ip; 376 377 ip = (struct ip *)ip_ptr; 378 ip->ip_v = IPVERSION; 379 ip->ip_hl = 5; 380 ip->ip_tos = inp->inp_ip_tos; 381 ip->ip_len = 0; 382 ip->ip_id = 0; 383 ip->ip_off = 0; 384 ip->ip_ttl = inp->inp_ip_ttl; 385 ip->ip_sum = 0; 386 ip->ip_p = IPPROTO_TCP; 387 ip->ip_src = inp->inp_laddr; 388 ip->ip_dst = inp->inp_faddr; 389 } 390 th->th_sport = inp->inp_lport; 391 th->th_dport = inp->inp_fport; 392 th->th_seq = 0; 393 th->th_ack = 0; 394 th->th_x2 = 0; 395 th->th_off = 5; 396 th->th_flags = 0; 397 th->th_win = 0; 398 th->th_urp = 0; 399 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 400} 401 402/* 403 * Create template to be used to send tcp packets on a connection. 404 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 405 * use for this function is in keepalives, which use tcp_respond. 406 */ 407struct tcptemp * 408tcpip_maketemplate(struct inpcb *inp) 409{ 410 struct mbuf *m; 411 struct tcptemp *n; 412 413 m = m_get(M_DONTWAIT, MT_DATA); 414 if (m == NULL) 415 return (0); 416 m->m_len = sizeof(struct tcptemp); 417 n = mtod(m, struct tcptemp *); 418 419 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 420 return (n); 421} 422 423/* 424 * Send a single message to the TCP at address specified by 425 * the given TCP/IP header. If m == NULL, then we make a copy 426 * of the tcpiphdr at ti and send directly to the addressed host. 427 * This is used to force keep alive messages out using the TCP 428 * template for a connection. If flags are given then we send 429 * a message back to the TCP which originated the * segment ti, 430 * and discard the mbuf containing it and any other attached mbufs. 431 * 432 * In any case the ack and sequence number of the transmitted 433 * segment are as specified by the parameters. 434 * 435 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 436 */ 437void 438tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th, 439 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 440{ 441 register int tlen; 442 int win = 0; 443 struct ip *ip; 444 struct tcphdr *nth; 445#ifdef INET6 446 struct ip6_hdr *ip6; 447 int isipv6; 448#endif /* INET6 */ 449 int ipflags = 0; 450 struct inpcb *inp; 451 452 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 453 454#ifdef INET6 455 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 456 ip6 = ipgen; 457#endif /* INET6 */ 458 ip = ipgen; 459 460 if (tp != NULL) { 461 inp = tp->t_inpcb; 462 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 463 INP_INFO_WLOCK_ASSERT(&tcbinfo); 464 INP_LOCK_ASSERT(inp); 465 } else 466 inp = NULL; 467 468 if (tp != NULL) { 469 if (!(flags & TH_RST)) { 470 win = sbspace(&inp->inp_socket->so_rcv); 471 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 472 win = (long)TCP_MAXWIN << tp->rcv_scale; 473 } 474 } 475 if (m == NULL) { 476 m = m_gethdr(M_DONTWAIT, MT_DATA); 477 if (m == NULL) 478 return; 479 tlen = 0; 480 m->m_data += max_linkhdr; 481#ifdef INET6 482 if (isipv6) { 483 bcopy((caddr_t)ip6, mtod(m, caddr_t), 484 sizeof(struct ip6_hdr)); 485 ip6 = mtod(m, struct ip6_hdr *); 486 nth = (struct tcphdr *)(ip6 + 1); 487 } else 488#endif /* INET6 */ 489 { 490 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 491 ip = mtod(m, struct ip *); 492 nth = (struct tcphdr *)(ip + 1); 493 } 494 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 495 flags = TH_ACK; 496 } else { 497 m_freem(m->m_next); 498 m->m_next = NULL; 499 m->m_data = (caddr_t)ipgen; 500 /* m_len is set later */ 501 tlen = 0; 502#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 503#ifdef INET6 504 if (isipv6) { 505 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 506 nth = (struct tcphdr *)(ip6 + 1); 507 } else 508#endif /* INET6 */ 509 { 510 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 511 nth = (struct tcphdr *)(ip + 1); 512 } 513 if (th != nth) { 514 /* 515 * this is usually a case when an extension header 516 * exists between the IPv6 header and the 517 * TCP header. 518 */ 519 nth->th_sport = th->th_sport; 520 nth->th_dport = th->th_dport; 521 } 522 xchg(nth->th_dport, nth->th_sport, n_short); 523#undef xchg 524 } 525#ifdef INET6 526 if (isipv6) { 527 ip6->ip6_flow = 0; 528 ip6->ip6_vfc = IPV6_VERSION; 529 ip6->ip6_nxt = IPPROTO_TCP; 530 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 531 tlen)); 532 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 533 } else 534#endif 535 { 536 tlen += sizeof (struct tcpiphdr); 537 ip->ip_len = tlen; 538 ip->ip_ttl = ip_defttl; 539 if (path_mtu_discovery) 540 ip->ip_off |= IP_DF; 541 } 542 m->m_len = tlen; 543 m->m_pkthdr.len = tlen; 544 m->m_pkthdr.rcvif = NULL; 545#ifdef MAC 546 if (inp != NULL) { 547 /* 548 * Packet is associated with a socket, so allow the 549 * label of the response to reflect the socket label. 550 */ 551 INP_LOCK_ASSERT(inp); 552 mac_create_mbuf_from_inpcb(inp, m); 553 } else { 554 /* 555 * Packet is not associated with a socket, so possibly 556 * update the label in place. 557 */ 558 mac_reflect_mbuf_tcp(m); 559 } 560#endif 561 nth->th_seq = htonl(seq); 562 nth->th_ack = htonl(ack); 563 nth->th_x2 = 0; 564 nth->th_off = sizeof (struct tcphdr) >> 2; 565 nth->th_flags = flags; 566 if (tp != NULL) 567 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 568 else 569 nth->th_win = htons((u_short)win); 570 nth->th_urp = 0; 571#ifdef INET6 572 if (isipv6) { 573 nth->th_sum = 0; 574 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 575 sizeof(struct ip6_hdr), 576 tlen - sizeof(struct ip6_hdr)); 577 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 578 NULL, NULL); 579 } else 580#endif /* INET6 */ 581 { 582 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 583 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 584 m->m_pkthdr.csum_flags = CSUM_TCP; 585 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 586 } 587#ifdef TCPDEBUG 588 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 589 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 590#endif 591#ifdef INET6 592 if (isipv6) 593 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 594 else 595#endif /* INET6 */ 596 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 597} 598 599/* 600 * Create a new TCP control block, making an 601 * empty reassembly queue and hooking it to the argument 602 * protocol control block. The `inp' parameter must have 603 * come from the zone allocator set up in tcp_init(). 604 */ 605struct tcpcb * 606tcp_newtcpcb(struct inpcb *inp) 607{ 608 struct tcpcb_mem *tm; 609 struct tcpcb *tp; 610#ifdef INET6 611 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 612#endif /* INET6 */ 613 614 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 615 if (tm == NULL) 616 return (NULL); 617 tp = &tm->tcb; 618 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 619 tp->t_maxseg = tp->t_maxopd = 620#ifdef INET6 621 isipv6 ? tcp_v6mssdflt : 622#endif /* INET6 */ 623 tcp_mssdflt; 624 625 /* Set up our timeouts. */ 626 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 627 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 628 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 629 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 630 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 631 632 if (tcp_do_rfc1323) 633 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 634 tp->sack_enable = tcp_do_sack; 635 TAILQ_INIT(&tp->snd_holes); 636 tp->t_inpcb = inp; /* XXX */ 637 /* 638 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 639 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 640 * reasonable initial retransmit time. 641 */ 642 tp->t_srtt = TCPTV_SRTTBASE; 643 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 644 tp->t_rttmin = tcp_rexmit_min; 645 tp->t_rxtcur = TCPTV_RTOBASE; 646 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 647 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 648 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 649 tp->t_rcvtime = ticks; 650 tp->t_bw_rtttime = ticks; 651 /* 652 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 653 * because the socket may be bound to an IPv6 wildcard address, 654 * which may match an IPv4-mapped IPv6 address. 655 */ 656 inp->inp_ip_ttl = ip_defttl; 657 inp->inp_ppcb = tp; 658 return (tp); /* XXX */ 659} 660 661/* 662 * Drop a TCP connection, reporting 663 * the specified error. If connection is synchronized, 664 * then send a RST to peer. 665 */ 666struct tcpcb * 667tcp_drop(struct tcpcb *tp, int errno) 668{ 669 struct socket *so = tp->t_inpcb->inp_socket; 670 671 INP_INFO_WLOCK_ASSERT(&tcbinfo); 672 INP_LOCK_ASSERT(tp->t_inpcb); 673 674 if (TCPS_HAVERCVDSYN(tp->t_state)) { 675 tp->t_state = TCPS_CLOSED; 676 (void) tcp_output(tp); 677 tcpstat.tcps_drops++; 678 } else 679 tcpstat.tcps_conndrops++; 680 if (errno == ETIMEDOUT && tp->t_softerror) 681 errno = tp->t_softerror; 682 so->so_error = errno; 683 return (tcp_close(tp)); 684} 685 686void 687tcp_discardcb(struct tcpcb *tp) 688{ 689 struct tseg_qent *q; 690 struct inpcb *inp = tp->t_inpcb; 691 struct socket *so = inp->inp_socket; 692#ifdef INET6 693 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 694#endif /* INET6 */ 695 696 INP_LOCK_ASSERT(inp); 697 698 /* 699 * Make sure that all of our timers are stopped before we 700 * delete the PCB. 701 */ 702 callout_stop(tp->tt_rexmt); 703 callout_stop(tp->tt_persist); 704 callout_stop(tp->tt_keep); 705 callout_stop(tp->tt_2msl); 706 callout_stop(tp->tt_delack); 707 708 /* 709 * If we got enough samples through the srtt filter, 710 * save the rtt and rttvar in the routing entry. 711 * 'Enough' is arbitrarily defined as 4 rtt samples. 712 * 4 samples is enough for the srtt filter to converge 713 * to within enough % of the correct value; fewer samples 714 * and we could save a bogus rtt. The danger is not high 715 * as tcp quickly recovers from everything. 716 * XXX: Works very well but needs some more statistics! 717 */ 718 if (tp->t_rttupdated >= 4) { 719 struct hc_metrics_lite metrics; 720 u_long ssthresh; 721 722 bzero(&metrics, sizeof(metrics)); 723 /* 724 * Update the ssthresh always when the conditions below 725 * are satisfied. This gives us better new start value 726 * for the congestion avoidance for new connections. 727 * ssthresh is only set if packet loss occured on a session. 728 * 729 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 730 * being torn down. Ideally this code would not use 'so'. 731 */ 732 ssthresh = tp->snd_ssthresh; 733 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 734 /* 735 * convert the limit from user data bytes to 736 * packets then to packet data bytes. 737 */ 738 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 739 if (ssthresh < 2) 740 ssthresh = 2; 741 ssthresh *= (u_long)(tp->t_maxseg + 742#ifdef INET6 743 (isipv6 ? sizeof (struct ip6_hdr) + 744 sizeof (struct tcphdr) : 745#endif 746 sizeof (struct tcpiphdr) 747#ifdef INET6 748 ) 749#endif 750 ); 751 } else 752 ssthresh = 0; 753 metrics.rmx_ssthresh = ssthresh; 754 755 metrics.rmx_rtt = tp->t_srtt; 756 metrics.rmx_rttvar = tp->t_rttvar; 757 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 758 metrics.rmx_bandwidth = tp->snd_bandwidth; 759 metrics.rmx_cwnd = tp->snd_cwnd; 760 metrics.rmx_sendpipe = 0; 761 metrics.rmx_recvpipe = 0; 762 763 tcp_hc_update(&inp->inp_inc, &metrics); 764 } 765 766 /* free the reassembly queue, if any */ 767 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 768 LIST_REMOVE(q, tqe_q); 769 m_freem(q->tqe_m); 770 uma_zfree(tcp_reass_zone, q); 771 tp->t_segqlen--; 772 tcp_reass_qsize--; 773 } 774 tcp_free_sackholes(tp); 775 inp->inp_ppcb = NULL; 776 tp->t_inpcb = NULL; 777 uma_zfree(tcpcb_zone, tp); 778} 779 780/* 781 * Attempt to close a TCP control block, marking it as dropped, and freeing 782 * the socket if we hold the only reference. 783 */ 784struct tcpcb * 785tcp_close(struct tcpcb *tp) 786{ 787 struct inpcb *inp = tp->t_inpcb; 788 struct socket *so; 789 790 INP_INFO_WLOCK_ASSERT(&tcbinfo); 791 INP_LOCK_ASSERT(inp); 792 793 in_pcbdrop(inp); 794 tcpstat.tcps_closed++; 795 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 796 so = inp->inp_socket; 797 soisdisconnected(so); 798 if (inp->inp_vflag & INP_SOCKREF) { 799 KASSERT(so->so_state & SS_PROTOREF, 800 ("tcp_close: !SS_PROTOREF")); 801 inp->inp_vflag &= ~INP_SOCKREF; 802 INP_UNLOCK(inp); 803 ACCEPT_LOCK(); 804 SOCK_LOCK(so); 805 so->so_state &= ~SS_PROTOREF; 806 sofree(so); 807 return (NULL); 808 } 809 return (tp); 810} 811 812void 813tcp_drain(void) 814{ 815 816 if (do_tcpdrain) { 817 struct inpcb *inpb; 818 struct tcpcb *tcpb; 819 struct tseg_qent *te; 820 821 /* 822 * Walk the tcpbs, if existing, and flush the reassembly queue, 823 * if there is one... 824 * XXX: The "Net/3" implementation doesn't imply that the TCP 825 * reassembly queue should be flushed, but in a situation 826 * where we're really low on mbufs, this is potentially 827 * usefull. 828 */ 829 INP_INFO_RLOCK(&tcbinfo); 830 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 831 if (inpb->inp_vflag & INP_TIMEWAIT) 832 continue; 833 INP_LOCK(inpb); 834 if ((tcpb = intotcpcb(inpb)) != NULL) { 835 while ((te = LIST_FIRST(&tcpb->t_segq)) 836 != NULL) { 837 LIST_REMOVE(te, tqe_q); 838 m_freem(te->tqe_m); 839 uma_zfree(tcp_reass_zone, te); 840 tcpb->t_segqlen--; 841 tcp_reass_qsize--; 842 } 843 tcp_clean_sackreport(tcpb); 844 } 845 INP_UNLOCK(inpb); 846 } 847 INP_INFO_RUNLOCK(&tcbinfo); 848 } 849} 850 851/* 852 * Notify a tcp user of an asynchronous error; 853 * store error as soft error, but wake up user 854 * (for now, won't do anything until can select for soft error). 855 * 856 * Do not wake up user since there currently is no mechanism for 857 * reporting soft errors (yet - a kqueue filter may be added). 858 */ 859static struct inpcb * 860tcp_notify(struct inpcb *inp, int error) 861{ 862 struct tcpcb *tp; 863 864 INP_INFO_WLOCK_ASSERT(&tcbinfo); 865 INP_LOCK_ASSERT(inp); 866 867 if ((inp->inp_vflag & INP_TIMEWAIT) || 868 (inp->inp_vflag & INP_DROPPED)) 869 return (inp); 870 871 tp = intotcpcb(inp); 872 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 873 874 /* 875 * Ignore some errors if we are hooked up. 876 * If connection hasn't completed, has retransmitted several times, 877 * and receives a second error, give up now. This is better 878 * than waiting a long time to establish a connection that 879 * can never complete. 880 */ 881 if (tp->t_state == TCPS_ESTABLISHED && 882 (error == EHOSTUNREACH || error == ENETUNREACH || 883 error == EHOSTDOWN)) { 884 return (inp); 885 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 886 tp->t_softerror) { 887 tp = tcp_drop(tp, error); 888 if (tp != NULL) 889 return (inp); 890 else 891 return (NULL); 892 } else { 893 tp->t_softerror = error; 894 return (inp); 895 } 896#if 0 897 wakeup( &so->so_timeo); 898 sorwakeup(so); 899 sowwakeup(so); 900#endif 901} 902 903static int 904tcp_pcblist(SYSCTL_HANDLER_ARGS) 905{ 906 int error, i, n; 907 struct inpcb *inp, **inp_list; 908 inp_gen_t gencnt; 909 struct xinpgen xig; 910 911 /* 912 * The process of preparing the TCB list is too time-consuming and 913 * resource-intensive to repeat twice on every request. 914 */ 915 if (req->oldptr == NULL) { 916 n = tcbinfo.ipi_count; 917 req->oldidx = 2 * (sizeof xig) 918 + (n + n/8) * sizeof(struct xtcpcb); 919 return (0); 920 } 921 922 if (req->newptr != NULL) 923 return (EPERM); 924 925 /* 926 * OK, now we're committed to doing something. 927 */ 928 INP_INFO_RLOCK(&tcbinfo); 929 gencnt = tcbinfo.ipi_gencnt; 930 n = tcbinfo.ipi_count; 931 INP_INFO_RUNLOCK(&tcbinfo); 932 933 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 934 + n * sizeof(struct xtcpcb)); 935 if (error != 0) 936 return (error); 937 938 xig.xig_len = sizeof xig; 939 xig.xig_count = n; 940 xig.xig_gen = gencnt; 941 xig.xig_sogen = so_gencnt; 942 error = SYSCTL_OUT(req, &xig, sizeof xig); 943 if (error) 944 return (error); 945 946 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 947 if (inp_list == NULL) 948 return (ENOMEM); 949 950 INP_INFO_RLOCK(&tcbinfo); 951 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 952 inp = LIST_NEXT(inp, inp_list)) { 953 INP_LOCK(inp); 954 if (inp->inp_gencnt <= gencnt) { 955 /* 956 * XXX: This use of cr_cansee(), introduced with 957 * TCP state changes, is not quite right, but for 958 * now, better than nothing. 959 */ 960 if (inp->inp_vflag & INP_TIMEWAIT) { 961 if (intotw(inp) != NULL) 962 error = cr_cansee(req->td->td_ucred, 963 intotw(inp)->tw_cred); 964 else 965 error = EINVAL; /* Skip this inp. */ 966 } else 967 error = cr_canseesocket(req->td->td_ucred, 968 inp->inp_socket); 969 if (error == 0) 970 inp_list[i++] = inp; 971 } 972 INP_UNLOCK(inp); 973 } 974 INP_INFO_RUNLOCK(&tcbinfo); 975 n = i; 976 977 error = 0; 978 for (i = 0; i < n; i++) { 979 inp = inp_list[i]; 980 INP_LOCK(inp); 981 if (inp->inp_gencnt <= gencnt) { 982 struct xtcpcb xt; 983 void *inp_ppcb; 984 985 bzero(&xt, sizeof(xt)); 986 xt.xt_len = sizeof xt; 987 /* XXX should avoid extra copy */ 988 bcopy(inp, &xt.xt_inp, sizeof *inp); 989 inp_ppcb = inp->inp_ppcb; 990 if (inp_ppcb == NULL) 991 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 992 else if (inp->inp_vflag & INP_TIMEWAIT) { 993 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 994 xt.xt_tp.t_state = TCPS_TIME_WAIT; 995 } else 996 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 997 if (inp->inp_socket != NULL) 998 sotoxsocket(inp->inp_socket, &xt.xt_socket); 999 else { 1000 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1001 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1002 } 1003 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1004 INP_UNLOCK(inp); 1005 error = SYSCTL_OUT(req, &xt, sizeof xt); 1006 } else 1007 INP_UNLOCK(inp); 1008 1009 } 1010 if (!error) { 1011 /* 1012 * Give the user an updated idea of our state. 1013 * If the generation differs from what we told 1014 * her before, she knows that something happened 1015 * while we were processing this request, and it 1016 * might be necessary to retry. 1017 */ 1018 INP_INFO_RLOCK(&tcbinfo); 1019 xig.xig_gen = tcbinfo.ipi_gencnt; 1020 xig.xig_sogen = so_gencnt; 1021 xig.xig_count = tcbinfo.ipi_count; 1022 INP_INFO_RUNLOCK(&tcbinfo); 1023 error = SYSCTL_OUT(req, &xig, sizeof xig); 1024 } 1025 free(inp_list, M_TEMP); 1026 return (error); 1027} 1028 1029SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1030 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1031 1032static int 1033tcp_getcred(SYSCTL_HANDLER_ARGS) 1034{ 1035 struct xucred xuc; 1036 struct sockaddr_in addrs[2]; 1037 struct inpcb *inp; 1038 int error; 1039 1040 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1041 if (error) 1042 return (error); 1043 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1044 if (error) 1045 return (error); 1046 INP_INFO_RLOCK(&tcbinfo); 1047 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1048 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1049 if (inp == NULL) { 1050 error = ENOENT; 1051 goto outunlocked; 1052 } 1053 INP_LOCK(inp); 1054 if (inp->inp_socket == NULL) { 1055 error = ENOENT; 1056 goto out; 1057 } 1058 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1059 if (error) 1060 goto out; 1061 cru2x(inp->inp_socket->so_cred, &xuc); 1062out: 1063 INP_UNLOCK(inp); 1064outunlocked: 1065 INP_INFO_RUNLOCK(&tcbinfo); 1066 if (error == 0) 1067 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1068 return (error); 1069} 1070 1071SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1072 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1073 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1074 1075#ifdef INET6 1076static int 1077tcp6_getcred(SYSCTL_HANDLER_ARGS) 1078{ 1079 struct xucred xuc; 1080 struct sockaddr_in6 addrs[2]; 1081 struct inpcb *inp; 1082 int error, mapped = 0; 1083 1084 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1085 if (error) 1086 return (error); 1087 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1088 if (error) 1089 return (error); 1090 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1091 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1092 return (error); 1093 } 1094 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1095 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1096 mapped = 1; 1097 else 1098 return (EINVAL); 1099 } 1100 1101 INP_INFO_RLOCK(&tcbinfo); 1102 if (mapped == 1) 1103 inp = in_pcblookup_hash(&tcbinfo, 1104 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1105 addrs[1].sin6_port, 1106 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1107 addrs[0].sin6_port, 1108 0, NULL); 1109 else 1110 inp = in6_pcblookup_hash(&tcbinfo, 1111 &addrs[1].sin6_addr, addrs[1].sin6_port, 1112 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1113 if (inp == NULL) { 1114 error = ENOENT; 1115 goto outunlocked; 1116 } 1117 INP_LOCK(inp); 1118 if (inp->inp_socket == NULL) { 1119 error = ENOENT; 1120 goto out; 1121 } 1122 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1123 if (error) 1124 goto out; 1125 cru2x(inp->inp_socket->so_cred, &xuc); 1126out: 1127 INP_UNLOCK(inp); 1128outunlocked: 1129 INP_INFO_RUNLOCK(&tcbinfo); 1130 if (error == 0) 1131 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1132 return (error); 1133} 1134 1135SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1136 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1137 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1138#endif 1139 1140 1141void 1142tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1143{ 1144 struct ip *ip = vip; 1145 struct tcphdr *th; 1146 struct in_addr faddr; 1147 struct inpcb *inp; 1148 struct tcpcb *tp; 1149 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1150 struct icmp *icp; 1151 struct in_conninfo inc; 1152 tcp_seq icmp_tcp_seq; 1153 int mtu; 1154 1155 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1156 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1157 return; 1158 1159 if (cmd == PRC_MSGSIZE) 1160 notify = tcp_mtudisc; 1161 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1162 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1163 notify = tcp_drop_syn_sent; 1164 /* 1165 * Redirects don't need to be handled up here. 1166 */ 1167 else if (PRC_IS_REDIRECT(cmd)) 1168 return; 1169 /* 1170 * Source quench is depreciated. 1171 */ 1172 else if (cmd == PRC_QUENCH) 1173 return; 1174 /* 1175 * Hostdead is ugly because it goes linearly through all PCBs. 1176 * XXX: We never get this from ICMP, otherwise it makes an 1177 * excellent DoS attack on machines with many connections. 1178 */ 1179 else if (cmd == PRC_HOSTDEAD) 1180 ip = NULL; 1181 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1182 return; 1183 if (ip != NULL) { 1184 icp = (struct icmp *)((caddr_t)ip 1185 - offsetof(struct icmp, icmp_ip)); 1186 th = (struct tcphdr *)((caddr_t)ip 1187 + (ip->ip_hl << 2)); 1188 INP_INFO_WLOCK(&tcbinfo); 1189 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1190 ip->ip_src, th->th_sport, 0, NULL); 1191 if (inp != NULL) { 1192 INP_LOCK(inp); 1193 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1194 !(inp->inp_vflag & INP_DROPPED) && 1195 !(inp->inp_socket == NULL)) { 1196 icmp_tcp_seq = htonl(th->th_seq); 1197 tp = intotcpcb(inp); 1198 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1199 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1200 if (cmd == PRC_MSGSIZE) { 1201 /* 1202 * MTU discovery: 1203 * If we got a needfrag set the MTU 1204 * in the route to the suggested new 1205 * value (if given) and then notify. 1206 */ 1207 bzero(&inc, sizeof(inc)); 1208 inc.inc_flags = 0; /* IPv4 */ 1209 inc.inc_faddr = faddr; 1210 1211 mtu = ntohs(icp->icmp_nextmtu); 1212 /* 1213 * If no alternative MTU was 1214 * proposed, try the next smaller 1215 * one. ip->ip_len has already 1216 * been swapped in icmp_input(). 1217 */ 1218 if (!mtu) 1219 mtu = ip_next_mtu(ip->ip_len, 1220 1); 1221 if (mtu < max(296, (tcp_minmss) 1222 + sizeof(struct tcpiphdr))) 1223 mtu = 0; 1224 if (!mtu) 1225 mtu = tcp_mssdflt 1226 + sizeof(struct tcpiphdr); 1227 /* 1228 * Only cache the the MTU if it 1229 * is smaller than the interface 1230 * or route MTU. tcp_mtudisc() 1231 * will do right thing by itself. 1232 */ 1233 if (mtu <= tcp_maxmtu(&inc)) 1234 tcp_hc_updatemtu(&inc, mtu); 1235 } 1236 1237 inp = (*notify)(inp, inetctlerrmap[cmd]); 1238 } 1239 } 1240 if (inp != NULL) 1241 INP_UNLOCK(inp); 1242 } else { 1243 inc.inc_fport = th->th_dport; 1244 inc.inc_lport = th->th_sport; 1245 inc.inc_faddr = faddr; 1246 inc.inc_laddr = ip->ip_src; 1247#ifdef INET6 1248 inc.inc_isipv6 = 0; 1249#endif 1250 syncache_unreach(&inc, th); 1251 } 1252 INP_INFO_WUNLOCK(&tcbinfo); 1253 } else 1254 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1255} 1256 1257#ifdef INET6 1258void 1259tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1260{ 1261 struct tcphdr th; 1262 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1263 struct ip6_hdr *ip6; 1264 struct mbuf *m; 1265 struct ip6ctlparam *ip6cp = NULL; 1266 const struct sockaddr_in6 *sa6_src = NULL; 1267 int off; 1268 struct tcp_portonly { 1269 u_int16_t th_sport; 1270 u_int16_t th_dport; 1271 } *thp; 1272 1273 if (sa->sa_family != AF_INET6 || 1274 sa->sa_len != sizeof(struct sockaddr_in6)) 1275 return; 1276 1277 if (cmd == PRC_MSGSIZE) 1278 notify = tcp_mtudisc; 1279 else if (!PRC_IS_REDIRECT(cmd) && 1280 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1281 return; 1282 /* Source quench is depreciated. */ 1283 else if (cmd == PRC_QUENCH) 1284 return; 1285 1286 /* if the parameter is from icmp6, decode it. */ 1287 if (d != NULL) { 1288 ip6cp = (struct ip6ctlparam *)d; 1289 m = ip6cp->ip6c_m; 1290 ip6 = ip6cp->ip6c_ip6; 1291 off = ip6cp->ip6c_off; 1292 sa6_src = ip6cp->ip6c_src; 1293 } else { 1294 m = NULL; 1295 ip6 = NULL; 1296 off = 0; /* fool gcc */ 1297 sa6_src = &sa6_any; 1298 } 1299 1300 if (ip6 != NULL) { 1301 struct in_conninfo inc; 1302 /* 1303 * XXX: We assume that when IPV6 is non NULL, 1304 * M and OFF are valid. 1305 */ 1306 1307 /* check if we can safely examine src and dst ports */ 1308 if (m->m_pkthdr.len < off + sizeof(*thp)) 1309 return; 1310 1311 bzero(&th, sizeof(th)); 1312 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1313 1314 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1315 (struct sockaddr *)ip6cp->ip6c_src, 1316 th.th_sport, cmd, NULL, notify); 1317 1318 inc.inc_fport = th.th_dport; 1319 inc.inc_lport = th.th_sport; 1320 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1321 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1322 inc.inc_isipv6 = 1; 1323 INP_INFO_WLOCK(&tcbinfo); 1324 syncache_unreach(&inc, &th); 1325 INP_INFO_WUNLOCK(&tcbinfo); 1326 } else 1327 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1328 0, cmd, NULL, notify); 1329} 1330#endif /* INET6 */ 1331 1332 1333/* 1334 * Following is where TCP initial sequence number generation occurs. 1335 * 1336 * There are two places where we must use initial sequence numbers: 1337 * 1. In SYN-ACK packets. 1338 * 2. In SYN packets. 1339 * 1340 * All ISNs for SYN-ACK packets are generated by the syncache. See 1341 * tcp_syncache.c for details. 1342 * 1343 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1344 * depends on this property. In addition, these ISNs should be 1345 * unguessable so as to prevent connection hijacking. To satisfy 1346 * the requirements of this situation, the algorithm outlined in 1347 * RFC 1948 is used, with only small modifications. 1348 * 1349 * Implementation details: 1350 * 1351 * Time is based off the system timer, and is corrected so that it 1352 * increases by one megabyte per second. This allows for proper 1353 * recycling on high speed LANs while still leaving over an hour 1354 * before rollover. 1355 * 1356 * As reading the *exact* system time is too expensive to be done 1357 * whenever setting up a TCP connection, we increment the time 1358 * offset in two ways. First, a small random positive increment 1359 * is added to isn_offset for each connection that is set up. 1360 * Second, the function tcp_isn_tick fires once per clock tick 1361 * and increments isn_offset as necessary so that sequence numbers 1362 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1363 * random positive increments serve only to ensure that the same 1364 * exact sequence number is never sent out twice (as could otherwise 1365 * happen when a port is recycled in less than the system tick 1366 * interval.) 1367 * 1368 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1369 * between seeding of isn_secret. This is normally set to zero, 1370 * as reseeding should not be necessary. 1371 * 1372 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1373 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1374 * general, this means holding an exclusive (write) lock. 1375 */ 1376 1377#define ISN_BYTES_PER_SECOND 1048576 1378#define ISN_STATIC_INCREMENT 4096 1379#define ISN_RANDOM_INCREMENT (4096 - 1) 1380 1381static u_char isn_secret[32]; 1382static int isn_last_reseed; 1383static u_int32_t isn_offset, isn_offset_old; 1384static MD5_CTX isn_ctx; 1385 1386tcp_seq 1387tcp_new_isn(struct tcpcb *tp) 1388{ 1389 u_int32_t md5_buffer[4]; 1390 tcp_seq new_isn; 1391 1392 INP_LOCK_ASSERT(tp->t_inpcb); 1393 1394 ISN_LOCK(); 1395 /* Seed if this is the first use, reseed if requested. */ 1396 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1397 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1398 < (u_int)ticks))) { 1399 read_random(&isn_secret, sizeof(isn_secret)); 1400 isn_last_reseed = ticks; 1401 } 1402 1403 /* Compute the md5 hash and return the ISN. */ 1404 MD5Init(&isn_ctx); 1405 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1406 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1407#ifdef INET6 1408 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1409 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1410 sizeof(struct in6_addr)); 1411 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1412 sizeof(struct in6_addr)); 1413 } else 1414#endif 1415 { 1416 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1417 sizeof(struct in_addr)); 1418 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1419 sizeof(struct in_addr)); 1420 } 1421 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1422 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1423 new_isn = (tcp_seq) md5_buffer[0]; 1424 isn_offset += ISN_STATIC_INCREMENT + 1425 (arc4random() & ISN_RANDOM_INCREMENT); 1426 new_isn += isn_offset; 1427 ISN_UNLOCK(); 1428 return (new_isn); 1429} 1430 1431/* 1432 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1433 * to keep time flowing at a relatively constant rate. If the random 1434 * increments have already pushed us past the projected offset, do nothing. 1435 */ 1436static void 1437tcp_isn_tick(void *xtp) 1438{ 1439 u_int32_t projected_offset; 1440 1441 ISN_LOCK(); 1442 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1443 1444 if (projected_offset > isn_offset) 1445 isn_offset = projected_offset; 1446 1447 isn_offset_old = isn_offset; 1448 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1449 ISN_UNLOCK(); 1450} 1451 1452/* 1453 * When a specific ICMP unreachable message is received and the 1454 * connection state is SYN-SENT, drop the connection. This behavior 1455 * is controlled by the icmp_may_rst sysctl. 1456 */ 1457struct inpcb * 1458tcp_drop_syn_sent(struct inpcb *inp, int errno) 1459{ 1460 struct tcpcb *tp; 1461 1462 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1463 INP_LOCK_ASSERT(inp); 1464 1465 if ((inp->inp_vflag & INP_TIMEWAIT) || 1466 (inp->inp_vflag & INP_DROPPED)) 1467 return (inp); 1468 1469 tp = intotcpcb(inp); 1470 if (tp->t_state != TCPS_SYN_SENT) 1471 return (inp); 1472 1473 tp = tcp_drop(tp, errno); 1474 if (tp != NULL) 1475 return (inp); 1476 else 1477 return (NULL); 1478} 1479 1480/* 1481 * When `need fragmentation' ICMP is received, update our idea of the MSS 1482 * based on the new value in the route. Also nudge TCP to send something, 1483 * since we know the packet we just sent was dropped. 1484 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1485 */ 1486struct inpcb * 1487tcp_mtudisc(struct inpcb *inp, int errno) 1488{ 1489 struct tcpcb *tp; 1490 struct socket *so = inp->inp_socket; 1491 u_int maxmtu; 1492 u_int romtu; 1493 int mss; 1494#ifdef INET6 1495 int isipv6; 1496#endif /* INET6 */ 1497 1498 INP_LOCK_ASSERT(inp); 1499 if ((inp->inp_vflag & INP_TIMEWAIT) || 1500 (inp->inp_vflag & INP_DROPPED)) 1501 return (inp); 1502 1503 tp = intotcpcb(inp); 1504 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1505 1506#ifdef INET6 1507 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1508#endif 1509 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1510 romtu = 1511#ifdef INET6 1512 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1513#endif /* INET6 */ 1514 tcp_maxmtu(&inp->inp_inc); 1515 if (!maxmtu) 1516 maxmtu = romtu; 1517 else 1518 maxmtu = min(maxmtu, romtu); 1519 if (!maxmtu) { 1520 tp->t_maxopd = tp->t_maxseg = 1521#ifdef INET6 1522 isipv6 ? tcp_v6mssdflt : 1523#endif /* INET6 */ 1524 tcp_mssdflt; 1525 return (inp); 1526 } 1527 mss = maxmtu - 1528#ifdef INET6 1529 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1530#endif /* INET6 */ 1531 sizeof(struct tcpiphdr) 1532#ifdef INET6 1533 ) 1534#endif /* INET6 */ 1535 ; 1536 1537 /* 1538 * XXX - The above conditional probably violates the TCP 1539 * spec. The problem is that, since we don't know the 1540 * other end's MSS, we are supposed to use a conservative 1541 * default. But, if we do that, then MTU discovery will 1542 * never actually take place, because the conservative 1543 * default is much less than the MTUs typically seen 1544 * on the Internet today. For the moment, we'll sweep 1545 * this under the carpet. 1546 * 1547 * The conservative default might not actually be a problem 1548 * if the only case this occurs is when sending an initial 1549 * SYN with options and data to a host we've never talked 1550 * to before. Then, they will reply with an MSS value which 1551 * will get recorded and the new parameters should get 1552 * recomputed. For Further Study. 1553 */ 1554 if (tp->t_maxopd <= mss) 1555 return (inp); 1556 tp->t_maxopd = mss; 1557 1558 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1559 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1560 mss -= TCPOLEN_TSTAMP_APPA; 1561#if (MCLBYTES & (MCLBYTES - 1)) == 0 1562 if (mss > MCLBYTES) 1563 mss &= ~(MCLBYTES-1); 1564#else 1565 if (mss > MCLBYTES) 1566 mss = mss / MCLBYTES * MCLBYTES; 1567#endif 1568 if (so->so_snd.sb_hiwat < mss) 1569 mss = so->so_snd.sb_hiwat; 1570 1571 tp->t_maxseg = mss; 1572 1573 tcpstat.tcps_mturesent++; 1574 tp->t_rtttime = 0; 1575 tp->snd_nxt = tp->snd_una; 1576 tcp_output(tp); 1577 return (inp); 1578} 1579 1580/* 1581 * Look-up the routing entry to the peer of this inpcb. If no route 1582 * is found and it cannot be allocated, then return NULL. This routine 1583 * is called by TCP routines that access the rmx structure and by tcp_mss 1584 * to get the interface MTU. 1585 */ 1586u_long 1587tcp_maxmtu(struct in_conninfo *inc) 1588{ 1589 struct route sro; 1590 struct sockaddr_in *dst; 1591 struct ifnet *ifp; 1592 u_long maxmtu = 0; 1593 1594 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1595 1596 bzero(&sro, sizeof(sro)); 1597 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1598 dst = (struct sockaddr_in *)&sro.ro_dst; 1599 dst->sin_family = AF_INET; 1600 dst->sin_len = sizeof(*dst); 1601 dst->sin_addr = inc->inc_faddr; 1602 rtalloc_ign(&sro, RTF_CLONING); 1603 } 1604 if (sro.ro_rt != NULL) { 1605 ifp = sro.ro_rt->rt_ifp; 1606 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1607 maxmtu = ifp->if_mtu; 1608 else 1609 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1610 RTFREE(sro.ro_rt); 1611 } 1612 return (maxmtu); 1613} 1614 1615#ifdef INET6 1616u_long 1617tcp_maxmtu6(struct in_conninfo *inc) 1618{ 1619 struct route_in6 sro6; 1620 struct ifnet *ifp; 1621 u_long maxmtu = 0; 1622 1623 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1624 1625 bzero(&sro6, sizeof(sro6)); 1626 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1627 sro6.ro_dst.sin6_family = AF_INET6; 1628 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1629 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1630 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1631 } 1632 if (sro6.ro_rt != NULL) { 1633 ifp = sro6.ro_rt->rt_ifp; 1634 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1635 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1636 else 1637 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1638 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1639 RTFREE(sro6.ro_rt); 1640 } 1641 1642 return (maxmtu); 1643} 1644#endif /* INET6 */ 1645 1646#ifdef IPSEC 1647/* compute ESP/AH header size for TCP, including outer IP header. */ 1648size_t 1649ipsec_hdrsiz_tcp(struct tcpcb *tp) 1650{ 1651 struct inpcb *inp; 1652 struct mbuf *m; 1653 size_t hdrsiz; 1654 struct ip *ip; 1655#ifdef INET6 1656 struct ip6_hdr *ip6; 1657#endif 1658 struct tcphdr *th; 1659 1660 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1661 return (0); 1662 MGETHDR(m, M_DONTWAIT, MT_DATA); 1663 if (!m) 1664 return (0); 1665 1666#ifdef INET6 1667 if ((inp->inp_vflag & INP_IPV6) != 0) { 1668 ip6 = mtod(m, struct ip6_hdr *); 1669 th = (struct tcphdr *)(ip6 + 1); 1670 m->m_pkthdr.len = m->m_len = 1671 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1672 tcpip_fillheaders(inp, ip6, th); 1673 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1674 } else 1675#endif /* INET6 */ 1676 { 1677 ip = mtod(m, struct ip *); 1678 th = (struct tcphdr *)(ip + 1); 1679 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1680 tcpip_fillheaders(inp, ip, th); 1681 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1682 } 1683 1684 m_free(m); 1685 return (hdrsiz); 1686} 1687#endif /*IPSEC*/ 1688 1689/* 1690 * Move a TCP connection into TIME_WAIT state. 1691 * tcbinfo is locked. 1692 * inp is locked, and is unlocked before returning. 1693 */ 1694void 1695tcp_twstart(struct tcpcb *tp) 1696{ 1697 struct tcptw *tw; 1698 struct inpcb *inp; 1699 int tw_time, acknow; 1700 struct socket *so; 1701 1702 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1703 INP_LOCK_ASSERT(tp->t_inpcb); 1704 1705 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1706 if (tw == NULL) { 1707 tw = tcp_timer_2msl_tw(1); 1708 if (tw == NULL) { 1709 tp = tcp_close(tp); 1710 if (tp != NULL) 1711 INP_UNLOCK(tp->t_inpcb); 1712 return; 1713 } 1714 } 1715 inp = tp->t_inpcb; 1716 tw->tw_inpcb = inp; 1717 1718 /* 1719 * Recover last window size sent. 1720 */ 1721 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1722 1723 /* 1724 * Set t_recent if timestamps are used on the connection. 1725 */ 1726 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1727 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1728 tw->t_recent = tp->ts_recent; 1729 else 1730 tw->t_recent = 0; 1731 1732 tw->snd_nxt = tp->snd_nxt; 1733 tw->rcv_nxt = tp->rcv_nxt; 1734 tw->iss = tp->iss; 1735 tw->irs = tp->irs; 1736 tw->t_starttime = tp->t_starttime; 1737 tw->tw_time = 0; 1738 1739/* XXX 1740 * If this code will 1741 * be used for fin-wait-2 state also, then we may need 1742 * a ts_recent from the last segment. 1743 */ 1744 tw_time = 2 * tcp_msl; 1745 acknow = tp->t_flags & TF_ACKNOW; 1746 1747 /* 1748 * First, discard tcpcb state, which includes stopping its timers and 1749 * freeing it. tcp_discardcb() used to also release the inpcb, but 1750 * that work is now done in the caller. 1751 * 1752 * Note: soisdisconnected() call used to be made in tcp_discardcb(), 1753 * and might not be needed here any longer. 1754 */ 1755 tcp_discardcb(tp); 1756 so = inp->inp_socket; 1757 soisdisconnected(so); 1758 SOCK_LOCK(so); 1759 tw->tw_cred = crhold(so->so_cred); 1760 tw->tw_so_options = so->so_options; 1761 SOCK_UNLOCK(so); 1762 if (acknow) 1763 tcp_twrespond(tw, TH_ACK); 1764 inp->inp_ppcb = tw; 1765 inp->inp_vflag |= INP_TIMEWAIT; 1766 tcp_timer_2msl_reset(tw, tw_time, 0); 1767 1768 /* 1769 * If the inpcb owns the sole reference to the socket, then we can 1770 * detach and free the socket as it is not needed in time wait. 1771 */ 1772 if (inp->inp_vflag & INP_SOCKREF) { 1773 KASSERT(so->so_state & SS_PROTOREF, 1774 ("tcp_twstart: !SS_PROTOREF")); 1775 inp->inp_vflag &= ~INP_SOCKREF; 1776 INP_UNLOCK(inp); 1777 ACCEPT_LOCK(); 1778 SOCK_LOCK(so); 1779 so->so_state &= ~SS_PROTOREF; 1780 sofree(so); 1781 } else 1782 INP_UNLOCK(inp); 1783} 1784 1785/* 1786 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1787 * the actual rate is slightly higher due to the addition of 1788 * random positive increments. 1789 * 1790 * Most other new OSes use semi-randomized ISN values, so we 1791 * do not need to worry about them. 1792 */ 1793#define MS_ISN_BYTES_PER_SECOND 250000 1794 1795/* 1796 * Determine if the ISN we will generate has advanced beyond the last 1797 * sequence number used by the previous connection. If so, indicate 1798 * that it is safe to recycle this tw socket by returning 1. 1799 * 1800 * XXXRW: This function should assert the inpcb lock as it does multiple 1801 * non-atomic reads from the tcptw, but is currently called without it from 1802 * in_pcb.c:in_pcblookup_local(). 1803 */ 1804int 1805tcp_twrecycleable(struct tcptw *tw) 1806{ 1807 tcp_seq new_iss = tw->iss; 1808 tcp_seq new_irs = tw->irs; 1809 1810 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1811 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1812 1813 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1814 return (1); 1815 else 1816 return (0); 1817} 1818 1819void 1820tcp_twclose(struct tcptw *tw, int reuse) 1821{ 1822 struct socket *so; 1823 struct inpcb *inp; 1824 1825 /* 1826 * At this point, we are in one of two situations: 1827 * 1828 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1829 * all state. 1830 * 1831 * (2) We have a socket -- if we own a reference, release it and 1832 * notify the socket layer. 1833 */ 1834 inp = tw->tw_inpcb; 1835 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1836 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1837 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1838 INP_LOCK_ASSERT(inp); 1839 1840 tw->tw_inpcb = NULL; 1841 tcp_timer_2msl_stop(tw); 1842 inp->inp_ppcb = NULL; 1843 in_pcbdrop(inp); 1844 1845 so = inp->inp_socket; 1846 if (so != NULL) { 1847 /* 1848 * If there's a socket, handle two cases: first, we own a 1849 * strong reference, which we will now release, or we don't 1850 * in which case another reference exists (XXXRW: think 1851 * about this more), and we don't need to take action. 1852 */ 1853 if (inp->inp_vflag & INP_SOCKREF) { 1854 inp->inp_vflag &= ~INP_SOCKREF; 1855 INP_UNLOCK(inp); 1856 ACCEPT_LOCK(); 1857 SOCK_LOCK(so); 1858 KASSERT(so->so_state & SS_PROTOREF, 1859 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1860 so->so_state &= ~SS_PROTOREF; 1861 sofree(so); 1862 } else { 1863 /* 1864 * If we don't own the only reference, the socket and 1865 * inpcb need to be left around to be handled by 1866 * tcp_usr_detach() later. 1867 */ 1868 INP_UNLOCK(inp); 1869 } 1870 } else { 1871#ifdef INET6 1872 if (inp->inp_vflag & INP_IPV6PROTO) 1873 in6_pcbfree(inp); 1874 else 1875#endif 1876 in_pcbfree(inp); 1877 } 1878 tcpstat.tcps_closed++; 1879 crfree(tw->tw_cred); 1880 tw->tw_cred = NULL; 1881 if (reuse) 1882 return; 1883 uma_zfree(tcptw_zone, tw); 1884} 1885 1886int 1887tcp_twrespond(struct tcptw *tw, int flags) 1888{ 1889 struct inpcb *inp = tw->tw_inpcb; 1890 struct tcphdr *th; 1891 struct mbuf *m; 1892 struct ip *ip = NULL; 1893 u_int8_t *optp; 1894 u_int hdrlen, optlen; 1895 int error; 1896#ifdef INET6 1897 struct ip6_hdr *ip6 = NULL; 1898 int isipv6 = inp->inp_inc.inc_isipv6; 1899#endif 1900 1901 INP_LOCK_ASSERT(inp); 1902 1903 m = m_gethdr(M_DONTWAIT, MT_DATA); 1904 if (m == NULL) 1905 return (ENOBUFS); 1906 m->m_data += max_linkhdr; 1907 1908#ifdef MAC 1909 mac_create_mbuf_from_inpcb(inp, m); 1910#endif 1911 1912#ifdef INET6 1913 if (isipv6) { 1914 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1915 ip6 = mtod(m, struct ip6_hdr *); 1916 th = (struct tcphdr *)(ip6 + 1); 1917 tcpip_fillheaders(inp, ip6, th); 1918 } else 1919#endif 1920 { 1921 hdrlen = sizeof(struct tcpiphdr); 1922 ip = mtod(m, struct ip *); 1923 th = (struct tcphdr *)(ip + 1); 1924 tcpip_fillheaders(inp, ip, th); 1925 } 1926 optp = (u_int8_t *)(th + 1); 1927 1928 /* 1929 * Send a timestamp and echo-reply if both our side and our peer 1930 * have sent timestamps in our SYN's and this is not a RST. 1931 */ 1932 if (tw->t_recent && flags == TH_ACK) { 1933 u_int32_t *lp = (u_int32_t *)optp; 1934 1935 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1936 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1937 *lp++ = htonl(ticks); 1938 *lp = htonl(tw->t_recent); 1939 optp += TCPOLEN_TSTAMP_APPA; 1940 } 1941 1942 optlen = optp - (u_int8_t *)(th + 1); 1943 1944 m->m_len = hdrlen + optlen; 1945 m->m_pkthdr.len = m->m_len; 1946 1947 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1948 1949 th->th_seq = htonl(tw->snd_nxt); 1950 th->th_ack = htonl(tw->rcv_nxt); 1951 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1952 th->th_flags = flags; 1953 th->th_win = htons(tw->last_win); 1954 1955#ifdef INET6 1956 if (isipv6) { 1957 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1958 sizeof(struct tcphdr) + optlen); 1959 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1960 error = ip6_output(m, inp->in6p_outputopts, NULL, 1961 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1962 } else 1963#endif 1964 { 1965 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1966 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1967 m->m_pkthdr.csum_flags = CSUM_TCP; 1968 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1969 ip->ip_len = m->m_pkthdr.len; 1970 if (path_mtu_discovery) 1971 ip->ip_off |= IP_DF; 1972 error = ip_output(m, inp->inp_options, NULL, 1973 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1974 NULL, inp); 1975 } 1976 if (flags & TH_ACK) 1977 tcpstat.tcps_sndacks++; 1978 else 1979 tcpstat.tcps_sndctrl++; 1980 tcpstat.tcps_sndtotal++; 1981 return (error); 1982} 1983 1984/* 1985 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1986 * 1987 * This code attempts to calculate the bandwidth-delay product as a 1988 * means of determining the optimal window size to maximize bandwidth, 1989 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1990 * routers. This code also does a fairly good job keeping RTTs in check 1991 * across slow links like modems. We implement an algorithm which is very 1992 * similar (but not meant to be) TCP/Vegas. The code operates on the 1993 * transmitter side of a TCP connection and so only effects the transmit 1994 * side of the connection. 1995 * 1996 * BACKGROUND: TCP makes no provision for the management of buffer space 1997 * at the end points or at the intermediate routers and switches. A TCP 1998 * stream, whether using NewReno or not, will eventually buffer as 1999 * many packets as it is able and the only reason this typically works is 2000 * due to the fairly small default buffers made available for a connection 2001 * (typicaly 16K or 32K). As machines use larger windows and/or window 2002 * scaling it is now fairly easy for even a single TCP connection to blow-out 2003 * all available buffer space not only on the local interface, but on 2004 * intermediate routers and switches as well. NewReno makes a misguided 2005 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2006 * then backing off, then steadily increasing the window again until another 2007 * failure occurs, ad-infinitum. This results in terrible oscillation that 2008 * is only made worse as network loads increase and the idea of intentionally 2009 * blowing out network buffers is, frankly, a terrible way to manage network 2010 * resources. 2011 * 2012 * It is far better to limit the transmit window prior to the failure 2013 * condition being achieved. There are two general ways to do this: First 2014 * you can 'scan' through different transmit window sizes and locate the 2015 * point where the RTT stops increasing, indicating that you have filled the 2016 * pipe, then scan backwards until you note that RTT stops decreasing, then 2017 * repeat ad-infinitum. This method works in principle but has severe 2018 * implementation issues due to RTT variances, timer granularity, and 2019 * instability in the algorithm which can lead to many false positives and 2020 * create oscillations as well as interact badly with other TCP streams 2021 * implementing the same algorithm. 2022 * 2023 * The second method is to limit the window to the bandwidth delay product 2024 * of the link. This is the method we implement. RTT variances and our 2025 * own manipulation of the congestion window, bwnd, can potentially 2026 * destabilize the algorithm. For this reason we have to stabilize the 2027 * elements used to calculate the window. We do this by using the minimum 2028 * observed RTT, the long term average of the observed bandwidth, and 2029 * by adding two segments worth of slop. It isn't perfect but it is able 2030 * to react to changing conditions and gives us a very stable basis on 2031 * which to extend the algorithm. 2032 */ 2033void 2034tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2035{ 2036 u_long bw; 2037 u_long bwnd; 2038 int save_ticks; 2039 2040 INP_LOCK_ASSERT(tp->t_inpcb); 2041 2042 /* 2043 * If inflight_enable is disabled in the middle of a tcp connection, 2044 * make sure snd_bwnd is effectively disabled. 2045 */ 2046 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2047 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2048 tp->snd_bandwidth = 0; 2049 return; 2050 } 2051 2052 /* 2053 * Figure out the bandwidth. Due to the tick granularity this 2054 * is a very rough number and it MUST be averaged over a fairly 2055 * long period of time. XXX we need to take into account a link 2056 * that is not using all available bandwidth, but for now our 2057 * slop will ramp us up if this case occurs and the bandwidth later 2058 * increases. 2059 * 2060 * Note: if ticks rollover 'bw' may wind up negative. We must 2061 * effectively reset t_bw_rtttime for this case. 2062 */ 2063 save_ticks = ticks; 2064 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2065 return; 2066 2067 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2068 (save_ticks - tp->t_bw_rtttime); 2069 tp->t_bw_rtttime = save_ticks; 2070 tp->t_bw_rtseq = ack_seq; 2071 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2072 return; 2073 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2074 2075 tp->snd_bandwidth = bw; 2076 2077 /* 2078 * Calculate the semi-static bandwidth delay product, plus two maximal 2079 * segments. The additional slop puts us squarely in the sweet 2080 * spot and also handles the bandwidth run-up case and stabilization. 2081 * Without the slop we could be locking ourselves into a lower 2082 * bandwidth. 2083 * 2084 * Situations Handled: 2085 * (1) Prevents over-queueing of packets on LANs, especially on 2086 * high speed LANs, allowing larger TCP buffers to be 2087 * specified, and also does a good job preventing 2088 * over-queueing of packets over choke points like modems 2089 * (at least for the transmit side). 2090 * 2091 * (2) Is able to handle changing network loads (bandwidth 2092 * drops so bwnd drops, bandwidth increases so bwnd 2093 * increases). 2094 * 2095 * (3) Theoretically should stabilize in the face of multiple 2096 * connections implementing the same algorithm (this may need 2097 * a little work). 2098 * 2099 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2100 * be adjusted with a sysctl but typically only needs to be 2101 * on very slow connections. A value no smaller then 5 2102 * should be used, but only reduce this default if you have 2103 * no other choice. 2104 */ 2105#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2106 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2107#undef USERTT 2108 2109 if (tcp_inflight_debug > 0) { 2110 static int ltime; 2111 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2112 ltime = ticks; 2113 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2114 tp, 2115 bw, 2116 tp->t_rttbest, 2117 tp->t_srtt, 2118 bwnd 2119 ); 2120 } 2121 } 2122 if ((long)bwnd < tcp_inflight_min) 2123 bwnd = tcp_inflight_min; 2124 if (bwnd > tcp_inflight_max) 2125 bwnd = tcp_inflight_max; 2126 if ((long)bwnd < tp->t_maxseg * 2) 2127 bwnd = tp->t_maxseg * 2; 2128 tp->snd_bwnd = bwnd; 2129} 2130 2131#ifdef TCP_SIGNATURE 2132/* 2133 * Callback function invoked by m_apply() to digest TCP segment data 2134 * contained within an mbuf chain. 2135 */ 2136static int 2137tcp_signature_apply(void *fstate, void *data, u_int len) 2138{ 2139 2140 MD5Update(fstate, (u_char *)data, len); 2141 return (0); 2142} 2143 2144/* 2145 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2146 * 2147 * Parameters: 2148 * m pointer to head of mbuf chain 2149 * off0 offset to TCP header within the mbuf chain 2150 * len length of TCP segment data, excluding options 2151 * optlen length of TCP segment options 2152 * buf pointer to storage for computed MD5 digest 2153 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2154 * 2155 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2156 * When called from tcp_input(), we can be sure that th_sum has been 2157 * zeroed out and verified already. 2158 * 2159 * This function is for IPv4 use only. Calling this function with an 2160 * IPv6 packet in the mbuf chain will yield undefined results. 2161 * 2162 * Return 0 if successful, otherwise return -1. 2163 * 2164 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2165 * search with the destination IP address, and a 'magic SPI' to be 2166 * determined by the application. This is hardcoded elsewhere to 1179 2167 * right now. Another branch of this code exists which uses the SPD to 2168 * specify per-application flows but it is unstable. 2169 */ 2170int 2171tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2172 u_char *buf, u_int direction) 2173{ 2174 union sockaddr_union dst; 2175 struct ippseudo ippseudo; 2176 MD5_CTX ctx; 2177 int doff; 2178 struct ip *ip; 2179 struct ipovly *ipovly; 2180 struct secasvar *sav; 2181 struct tcphdr *th; 2182 u_short savecsum; 2183 2184 KASSERT(m != NULL, ("NULL mbuf chain")); 2185 KASSERT(buf != NULL, ("NULL signature pointer")); 2186 2187 /* Extract the destination from the IP header in the mbuf. */ 2188 ip = mtod(m, struct ip *); 2189 bzero(&dst, sizeof(union sockaddr_union)); 2190 dst.sa.sa_len = sizeof(struct sockaddr_in); 2191 dst.sa.sa_family = AF_INET; 2192 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2193 ip->ip_src : ip->ip_dst; 2194 2195 /* Look up an SADB entry which matches the address of the peer. */ 2196 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2197 if (sav == NULL) { 2198 printf("%s: SADB lookup failed for %s\n", __func__, 2199 inet_ntoa(dst.sin.sin_addr)); 2200 return (EINVAL); 2201 } 2202 2203 MD5Init(&ctx); 2204 ipovly = (struct ipovly *)ip; 2205 th = (struct tcphdr *)((u_char *)ip + off0); 2206 doff = off0 + sizeof(struct tcphdr) + optlen; 2207 2208 /* 2209 * Step 1: Update MD5 hash with IP pseudo-header. 2210 * 2211 * XXX The ippseudo header MUST be digested in network byte order, 2212 * or else we'll fail the regression test. Assume all fields we've 2213 * been doing arithmetic on have been in host byte order. 2214 * XXX One cannot depend on ipovly->ih_len here. When called from 2215 * tcp_output(), the underlying ip_len member has not yet been set. 2216 */ 2217 ippseudo.ippseudo_src = ipovly->ih_src; 2218 ippseudo.ippseudo_dst = ipovly->ih_dst; 2219 ippseudo.ippseudo_pad = 0; 2220 ippseudo.ippseudo_p = IPPROTO_TCP; 2221 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2222 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2223 2224 /* 2225 * Step 2: Update MD5 hash with TCP header, excluding options. 2226 * The TCP checksum must be set to zero. 2227 */ 2228 savecsum = th->th_sum; 2229 th->th_sum = 0; 2230 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2231 th->th_sum = savecsum; 2232 2233 /* 2234 * Step 3: Update MD5 hash with TCP segment data. 2235 * Use m_apply() to avoid an early m_pullup(). 2236 */ 2237 if (len > 0) 2238 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2239 2240 /* 2241 * Step 4: Update MD5 hash with shared secret. 2242 */ 2243 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2244 MD5Final(buf, &ctx); 2245 2246 key_sa_recordxfer(sav, m); 2247 KEY_FREESAV(&sav); 2248 return (0); 2249} 2250#endif /* TCP_SIGNATURE */ 2251 2252static int 2253sysctl_drop(SYSCTL_HANDLER_ARGS) 2254{ 2255 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2256 struct sockaddr_storage addrs[2]; 2257 struct inpcb *inp; 2258 struct tcpcb *tp; 2259 struct tcptw *tw; 2260 struct sockaddr_in *fin, *lin; 2261#ifdef INET6 2262 struct sockaddr_in6 *fin6, *lin6; 2263 struct in6_addr f6, l6; 2264#endif 2265 int error; 2266 2267 inp = NULL; 2268 fin = lin = NULL; 2269#ifdef INET6 2270 fin6 = lin6 = NULL; 2271#endif 2272 error = 0; 2273 2274 if (req->oldptr != NULL || req->oldlen != 0) 2275 return (EINVAL); 2276 if (req->newptr == NULL) 2277 return (EPERM); 2278 if (req->newlen < sizeof(addrs)) 2279 return (ENOMEM); 2280 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2281 if (error) 2282 return (error); 2283 2284 switch (addrs[0].ss_family) { 2285#ifdef INET6 2286 case AF_INET6: 2287 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2288 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2289 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2290 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2291 return (EINVAL); 2292 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2293 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2294 return (EINVAL); 2295 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2296 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2297 fin = (struct sockaddr_in *)&addrs[0]; 2298 lin = (struct sockaddr_in *)&addrs[1]; 2299 break; 2300 } 2301 error = sa6_embedscope(fin6, ip6_use_defzone); 2302 if (error) 2303 return (error); 2304 error = sa6_embedscope(lin6, ip6_use_defzone); 2305 if (error) 2306 return (error); 2307 break; 2308#endif 2309 case AF_INET: 2310 fin = (struct sockaddr_in *)&addrs[0]; 2311 lin = (struct sockaddr_in *)&addrs[1]; 2312 if (fin->sin_len != sizeof(struct sockaddr_in) || 2313 lin->sin_len != sizeof(struct sockaddr_in)) 2314 return (EINVAL); 2315 break; 2316 default: 2317 return (EINVAL); 2318 } 2319 INP_INFO_WLOCK(&tcbinfo); 2320 switch (addrs[0].ss_family) { 2321#ifdef INET6 2322 case AF_INET6: 2323 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2324 &l6, lin6->sin6_port, 0, NULL); 2325 break; 2326#endif 2327 case AF_INET: 2328 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2329 lin->sin_addr, lin->sin_port, 0, NULL); 2330 break; 2331 } 2332 if (inp != NULL) { 2333 INP_LOCK(inp); 2334 if (inp->inp_vflag & INP_TIMEWAIT) { 2335 /* 2336 * XXXRW: There currently exists a state where an 2337 * inpcb is present, but its timewait state has been 2338 * discarded. For now, don't allow dropping of this 2339 * type of inpcb. 2340 */ 2341 tw = intotw(inp); 2342 if (tw != NULL) 2343 tcp_twclose(tw, 0); 2344 } else if (!(inp->inp_vflag & INP_DROPPED) && 2345 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2346 tp = intotcpcb(inp); 2347 tcp_drop(tp, ECONNABORTED); 2348 } 2349 INP_UNLOCK(inp); 2350 } else 2351 error = ESRCH; 2352 INP_INFO_WUNLOCK(&tcbinfo); 2353 return (error); 2354} 2355 2356SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2357 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2358 0, sysctl_drop, "", "Drop TCP connection"); 2359