tcp_timewait.c revision 160549
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 160549 2006-07-21 17:11:15Z rwatson $ 31 */ 32 33#include "opt_compat.h" 34#include "opt_inet.h" 35#include "opt_inet6.h" 36#include "opt_ipsec.h" 37#include "opt_mac.h" 38#include "opt_tcpdebug.h" 39#include "opt_tcp_sack.h" 40 41#include <sys/param.h> 42#include <sys/systm.h> 43#include <sys/callout.h> 44#include <sys/kernel.h> 45#include <sys/sysctl.h> 46#include <sys/mac.h> 47#include <sys/malloc.h> 48#include <sys/mbuf.h> 49#ifdef INET6 50#include <sys/domain.h> 51#endif 52#include <sys/proc.h> 53#include <sys/socket.h> 54#include <sys/socketvar.h> 55#include <sys/protosw.h> 56#include <sys/random.h> 57 58#include <vm/uma.h> 59 60#include <net/route.h> 61#include <net/if.h> 62 63#include <netinet/in.h> 64#include <netinet/in_systm.h> 65#include <netinet/ip.h> 66#ifdef INET6 67#include <netinet/ip6.h> 68#endif 69#include <netinet/in_pcb.h> 70#ifdef INET6 71#include <netinet6/in6_pcb.h> 72#endif 73#include <netinet/in_var.h> 74#include <netinet/ip_var.h> 75#ifdef INET6 76#include <netinet6/ip6_var.h> 77#include <netinet6/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 /* 697 * XXXRW: This is all very well and good, but actually, we might be 698 * discarding the tcpcb after the socket is gone, so we can't do 699 * this: 700 KASSERT(so != NULL, ("tcp_discardcb: so == NULL")); 701 */ 702 INP_LOCK_ASSERT(inp); 703 704 /* 705 * Make sure that all of our timers are stopped before we 706 * delete the PCB. 707 */ 708 callout_stop(tp->tt_rexmt); 709 callout_stop(tp->tt_persist); 710 callout_stop(tp->tt_keep); 711 callout_stop(tp->tt_2msl); 712 callout_stop(tp->tt_delack); 713 714 /* 715 * If we got enough samples through the srtt filter, 716 * save the rtt and rttvar in the routing entry. 717 * 'Enough' is arbitrarily defined as 4 rtt samples. 718 * 4 samples is enough for the srtt filter to converge 719 * to within enough % of the correct value; fewer samples 720 * and we could save a bogus rtt. The danger is not high 721 * as tcp quickly recovers from everything. 722 * XXX: Works very well but needs some more statistics! 723 */ 724 if (tp->t_rttupdated >= 4) { 725 struct hc_metrics_lite metrics; 726 u_long ssthresh; 727 728 bzero(&metrics, sizeof(metrics)); 729 /* 730 * Update the ssthresh always when the conditions below 731 * are satisfied. This gives us better new start value 732 * for the congestion avoidance for new connections. 733 * ssthresh is only set if packet loss occured on a session. 734 */ 735 ssthresh = tp->snd_ssthresh; 736 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 737 /* 738 * convert the limit from user data bytes to 739 * packets then to packet data bytes. 740 */ 741 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 742 if (ssthresh < 2) 743 ssthresh = 2; 744 ssthresh *= (u_long)(tp->t_maxseg + 745#ifdef INET6 746 (isipv6 ? sizeof (struct ip6_hdr) + 747 sizeof (struct tcphdr) : 748#endif 749 sizeof (struct tcpiphdr) 750#ifdef INET6 751 ) 752#endif 753 ); 754 } else 755 ssthresh = 0; 756 metrics.rmx_ssthresh = ssthresh; 757 758 metrics.rmx_rtt = tp->t_srtt; 759 metrics.rmx_rttvar = tp->t_rttvar; 760 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 761 metrics.rmx_bandwidth = tp->snd_bandwidth; 762 metrics.rmx_cwnd = tp->snd_cwnd; 763 metrics.rmx_sendpipe = 0; 764 metrics.rmx_recvpipe = 0; 765 766 tcp_hc_update(&inp->inp_inc, &metrics); 767 } 768 769 /* free the reassembly queue, if any */ 770 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 771 LIST_REMOVE(q, tqe_q); 772 m_freem(q->tqe_m); 773 uma_zfree(tcp_reass_zone, q); 774 tp->t_segqlen--; 775 tcp_reass_qsize--; 776 } 777 tcp_free_sackholes(tp); 778 inp->inp_ppcb = NULL; 779 tp->t_inpcb = NULL; 780 uma_zfree(tcpcb_zone, tp); 781 782 /* 783 * XXXRW: This seems a bit unclean. 784 */ 785 if (so != NULL) 786 soisdisconnected(so); 787} 788 789/* 790 * Attempt to close a TCP control block, marking it as dropped, and freeing 791 * the socket if we hold the only reference. 792 */ 793struct tcpcb * 794tcp_close(struct tcpcb *tp) 795{ 796 struct inpcb *inp = tp->t_inpcb; 797 struct socket *so; 798 799 INP_INFO_WLOCK_ASSERT(&tcbinfo); 800 INP_LOCK_ASSERT(inp); 801 802 in_pcbdrop(inp); 803 tcpstat.tcps_closed++; 804 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 805 so = inp->inp_socket; 806 soisdisconnected(so); 807 if (inp->inp_vflag & INP_SOCKREF) { 808 KASSERT(so->so_state & SS_PROTOREF, 809 ("tcp_close: !SS_PROTOREF")); 810 inp->inp_vflag &= ~INP_SOCKREF; 811 INP_UNLOCK(inp); 812 ACCEPT_LOCK(); 813 SOCK_LOCK(so); 814 so->so_state &= ~SS_PROTOREF; 815 sofree(so); 816 return (NULL); 817 } 818 return (tp); 819} 820 821void 822tcp_drain(void) 823{ 824 825 if (do_tcpdrain) { 826 struct inpcb *inpb; 827 struct tcpcb *tcpb; 828 struct tseg_qent *te; 829 830 /* 831 * Walk the tcpbs, if existing, and flush the reassembly queue, 832 * if there is one... 833 * XXX: The "Net/3" implementation doesn't imply that the TCP 834 * reassembly queue should be flushed, but in a situation 835 * where we're really low on mbufs, this is potentially 836 * usefull. 837 */ 838 INP_INFO_RLOCK(&tcbinfo); 839 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 840 if (inpb->inp_vflag & INP_TIMEWAIT) 841 continue; 842 INP_LOCK(inpb); 843 if ((tcpb = intotcpcb(inpb)) != NULL) { 844 while ((te = LIST_FIRST(&tcpb->t_segq)) 845 != NULL) { 846 LIST_REMOVE(te, tqe_q); 847 m_freem(te->tqe_m); 848 uma_zfree(tcp_reass_zone, te); 849 tcpb->t_segqlen--; 850 tcp_reass_qsize--; 851 } 852 tcp_clean_sackreport(tcpb); 853 } 854 INP_UNLOCK(inpb); 855 } 856 INP_INFO_RUNLOCK(&tcbinfo); 857 } 858} 859 860/* 861 * Notify a tcp user of an asynchronous error; 862 * store error as soft error, but wake up user 863 * (for now, won't do anything until can select for soft error). 864 * 865 * Do not wake up user since there currently is no mechanism for 866 * reporting soft errors (yet - a kqueue filter may be added). 867 */ 868static struct inpcb * 869tcp_notify(struct inpcb *inp, int error) 870{ 871 struct tcpcb *tp; 872 873 INP_INFO_WLOCK_ASSERT(&tcbinfo); 874 INP_LOCK_ASSERT(inp); 875 876 if ((inp->inp_vflag & INP_TIMEWAIT) || 877 (inp->inp_vflag & INP_DROPPED)) 878 return (inp); 879 880 tp = intotcpcb(inp); 881 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 882 883 /* 884 * Ignore some errors if we are hooked up. 885 * If connection hasn't completed, has retransmitted several times, 886 * and receives a second error, give up now. This is better 887 * than waiting a long time to establish a connection that 888 * can never complete. 889 */ 890 if (tp->t_state == TCPS_ESTABLISHED && 891 (error == EHOSTUNREACH || error == ENETUNREACH || 892 error == EHOSTDOWN)) { 893 return (inp); 894 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 895 tp->t_softerror) { 896 tp = tcp_drop(tp, error); 897 if (tp != NULL) 898 return (inp); 899 else 900 return (NULL); 901 } else { 902 tp->t_softerror = error; 903 return (inp); 904 } 905#if 0 906 wakeup( &so->so_timeo); 907 sorwakeup(so); 908 sowwakeup(so); 909#endif 910} 911 912static int 913tcp_pcblist(SYSCTL_HANDLER_ARGS) 914{ 915 int error, i, n; 916 struct inpcb *inp, **inp_list; 917 inp_gen_t gencnt; 918 struct xinpgen xig; 919 920 /* 921 * The process of preparing the TCB list is too time-consuming and 922 * resource-intensive to repeat twice on every request. 923 */ 924 if (req->oldptr == NULL) { 925 n = tcbinfo.ipi_count; 926 req->oldidx = 2 * (sizeof xig) 927 + (n + n/8) * sizeof(struct xtcpcb); 928 return (0); 929 } 930 931 if (req->newptr != NULL) 932 return (EPERM); 933 934 /* 935 * OK, now we're committed to doing something. 936 */ 937 INP_INFO_RLOCK(&tcbinfo); 938 gencnt = tcbinfo.ipi_gencnt; 939 n = tcbinfo.ipi_count; 940 INP_INFO_RUNLOCK(&tcbinfo); 941 942 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 943 + n * sizeof(struct xtcpcb)); 944 if (error != 0) 945 return (error); 946 947 xig.xig_len = sizeof xig; 948 xig.xig_count = n; 949 xig.xig_gen = gencnt; 950 xig.xig_sogen = so_gencnt; 951 error = SYSCTL_OUT(req, &xig, sizeof xig); 952 if (error) 953 return (error); 954 955 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 956 if (inp_list == NULL) 957 return (ENOMEM); 958 959 INP_INFO_RLOCK(&tcbinfo); 960 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 961 inp = LIST_NEXT(inp, inp_list)) { 962 INP_LOCK(inp); 963 if (inp->inp_gencnt <= gencnt) { 964 /* 965 * XXX: This use of cr_cansee(), introduced with 966 * TCP state changes, is not quite right, but for 967 * now, better than nothing. 968 */ 969 if (inp->inp_vflag & INP_TIMEWAIT) { 970 if (intotw(inp) != NULL) 971 error = cr_cansee(req->td->td_ucred, 972 intotw(inp)->tw_cred); 973 else 974 error = EINVAL; /* Skip this inp. */ 975 } else 976 error = cr_canseesocket(req->td->td_ucred, 977 inp->inp_socket); 978 if (error == 0) 979 inp_list[i++] = inp; 980 } 981 INP_UNLOCK(inp); 982 } 983 INP_INFO_RUNLOCK(&tcbinfo); 984 n = i; 985 986 error = 0; 987 for (i = 0; i < n; i++) { 988 inp = inp_list[i]; 989 INP_LOCK(inp); 990 if (inp->inp_gencnt <= gencnt) { 991 struct xtcpcb xt; 992 void *inp_ppcb; 993 994 bzero(&xt, sizeof(xt)); 995 xt.xt_len = sizeof xt; 996 /* XXX should avoid extra copy */ 997 bcopy(inp, &xt.xt_inp, sizeof *inp); 998 inp_ppcb = inp->inp_ppcb; 999 if (inp_ppcb == NULL) 1000 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1001 else if (inp->inp_vflag & INP_TIMEWAIT) { 1002 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1003 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1004 } else 1005 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1006 if (inp->inp_socket != NULL) 1007 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1008 else { 1009 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1010 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1011 } 1012 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1013 INP_UNLOCK(inp); 1014 error = SYSCTL_OUT(req, &xt, sizeof xt); 1015 } else 1016 INP_UNLOCK(inp); 1017 1018 } 1019 if (!error) { 1020 /* 1021 * Give the user an updated idea of our state. 1022 * If the generation differs from what we told 1023 * her before, she knows that something happened 1024 * while we were processing this request, and it 1025 * might be necessary to retry. 1026 */ 1027 INP_INFO_RLOCK(&tcbinfo); 1028 xig.xig_gen = tcbinfo.ipi_gencnt; 1029 xig.xig_sogen = so_gencnt; 1030 xig.xig_count = tcbinfo.ipi_count; 1031 INP_INFO_RUNLOCK(&tcbinfo); 1032 error = SYSCTL_OUT(req, &xig, sizeof xig); 1033 } 1034 free(inp_list, M_TEMP); 1035 return (error); 1036} 1037 1038SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1039 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1040 1041static int 1042tcp_getcred(SYSCTL_HANDLER_ARGS) 1043{ 1044 struct xucred xuc; 1045 struct sockaddr_in addrs[2]; 1046 struct inpcb *inp; 1047 int error; 1048 1049 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1050 if (error) 1051 return (error); 1052 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1053 if (error) 1054 return (error); 1055 INP_INFO_RLOCK(&tcbinfo); 1056 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1057 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1058 if (inp == NULL) { 1059 error = ENOENT; 1060 goto outunlocked; 1061 } 1062 INP_LOCK(inp); 1063 if (inp->inp_socket == NULL) { 1064 error = ENOENT; 1065 goto out; 1066 } 1067 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1068 if (error) 1069 goto out; 1070 cru2x(inp->inp_socket->so_cred, &xuc); 1071out: 1072 INP_UNLOCK(inp); 1073outunlocked: 1074 INP_INFO_RUNLOCK(&tcbinfo); 1075 if (error == 0) 1076 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1077 return (error); 1078} 1079 1080SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1081 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1082 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1083 1084#ifdef INET6 1085static int 1086tcp6_getcred(SYSCTL_HANDLER_ARGS) 1087{ 1088 struct xucred xuc; 1089 struct sockaddr_in6 addrs[2]; 1090 struct inpcb *inp; 1091 int error, mapped = 0; 1092 1093 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1094 if (error) 1095 return (error); 1096 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1097 if (error) 1098 return (error); 1099 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1100 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1101 return (error); 1102 } 1103 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1104 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1105 mapped = 1; 1106 else 1107 return (EINVAL); 1108 } 1109 1110 INP_INFO_RLOCK(&tcbinfo); 1111 if (mapped == 1) 1112 inp = in_pcblookup_hash(&tcbinfo, 1113 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1114 addrs[1].sin6_port, 1115 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1116 addrs[0].sin6_port, 1117 0, NULL); 1118 else 1119 inp = in6_pcblookup_hash(&tcbinfo, 1120 &addrs[1].sin6_addr, addrs[1].sin6_port, 1121 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1122 if (inp == NULL) { 1123 error = ENOENT; 1124 goto outunlocked; 1125 } 1126 INP_LOCK(inp); 1127 if (inp->inp_socket == NULL) { 1128 error = ENOENT; 1129 goto out; 1130 } 1131 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1132 if (error) 1133 goto out; 1134 cru2x(inp->inp_socket->so_cred, &xuc); 1135out: 1136 INP_UNLOCK(inp); 1137outunlocked: 1138 INP_INFO_RUNLOCK(&tcbinfo); 1139 if (error == 0) 1140 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1141 return (error); 1142} 1143 1144SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1145 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1146 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1147#endif 1148 1149 1150void 1151tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1152{ 1153 struct ip *ip = vip; 1154 struct tcphdr *th; 1155 struct in_addr faddr; 1156 struct inpcb *inp; 1157 struct tcpcb *tp; 1158 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1159 struct icmp *icp; 1160 struct in_conninfo inc; 1161 tcp_seq icmp_tcp_seq; 1162 int mtu; 1163 1164 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1165 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1166 return; 1167 1168 if (cmd == PRC_MSGSIZE) 1169 notify = tcp_mtudisc; 1170 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1171 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1172 notify = tcp_drop_syn_sent; 1173 /* 1174 * Redirects don't need to be handled up here. 1175 */ 1176 else if (PRC_IS_REDIRECT(cmd)) 1177 return; 1178 /* 1179 * Source quench is depreciated. 1180 */ 1181 else if (cmd == PRC_QUENCH) 1182 return; 1183 /* 1184 * Hostdead is ugly because it goes linearly through all PCBs. 1185 * XXX: We never get this from ICMP, otherwise it makes an 1186 * excellent DoS attack on machines with many connections. 1187 */ 1188 else if (cmd == PRC_HOSTDEAD) 1189 ip = NULL; 1190 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1191 return; 1192 if (ip != NULL) { 1193 icp = (struct icmp *)((caddr_t)ip 1194 - offsetof(struct icmp, icmp_ip)); 1195 th = (struct tcphdr *)((caddr_t)ip 1196 + (ip->ip_hl << 2)); 1197 INP_INFO_WLOCK(&tcbinfo); 1198 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1199 ip->ip_src, th->th_sport, 0, NULL); 1200 if (inp != NULL) { 1201 INP_LOCK(inp); 1202 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1203 !(inp->inp_vflag & INP_DROPPED) && 1204 !(inp->inp_socket == NULL)) { 1205 icmp_tcp_seq = htonl(th->th_seq); 1206 tp = intotcpcb(inp); 1207 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1208 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1209 if (cmd == PRC_MSGSIZE) { 1210 /* 1211 * MTU discovery: 1212 * If we got a needfrag set the MTU 1213 * in the route to the suggested new 1214 * value (if given) and then notify. 1215 */ 1216 bzero(&inc, sizeof(inc)); 1217 inc.inc_flags = 0; /* IPv4 */ 1218 inc.inc_faddr = faddr; 1219 1220 mtu = ntohs(icp->icmp_nextmtu); 1221 /* 1222 * If no alternative MTU was 1223 * proposed, try the next smaller 1224 * one. ip->ip_len has already 1225 * been swapped in icmp_input(). 1226 */ 1227 if (!mtu) 1228 mtu = ip_next_mtu(ip->ip_len, 1229 1); 1230 if (mtu < max(296, (tcp_minmss) 1231 + sizeof(struct tcpiphdr))) 1232 mtu = 0; 1233 if (!mtu) 1234 mtu = tcp_mssdflt 1235 + sizeof(struct tcpiphdr); 1236 /* 1237 * Only cache the the MTU if it 1238 * is smaller than the interface 1239 * or route MTU. tcp_mtudisc() 1240 * will do right thing by itself. 1241 */ 1242 if (mtu <= tcp_maxmtu(&inc)) 1243 tcp_hc_updatemtu(&inc, mtu); 1244 } 1245 1246 inp = (*notify)(inp, inetctlerrmap[cmd]); 1247 } 1248 } 1249 if (inp != NULL) 1250 INP_UNLOCK(inp); 1251 } else { 1252 inc.inc_fport = th->th_dport; 1253 inc.inc_lport = th->th_sport; 1254 inc.inc_faddr = faddr; 1255 inc.inc_laddr = ip->ip_src; 1256#ifdef INET6 1257 inc.inc_isipv6 = 0; 1258#endif 1259 syncache_unreach(&inc, th); 1260 } 1261 INP_INFO_WUNLOCK(&tcbinfo); 1262 } else 1263 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1264} 1265 1266#ifdef INET6 1267void 1268tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1269{ 1270 struct tcphdr th; 1271 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1272 struct ip6_hdr *ip6; 1273 struct mbuf *m; 1274 struct ip6ctlparam *ip6cp = NULL; 1275 const struct sockaddr_in6 *sa6_src = NULL; 1276 int off; 1277 struct tcp_portonly { 1278 u_int16_t th_sport; 1279 u_int16_t th_dport; 1280 } *thp; 1281 1282 if (sa->sa_family != AF_INET6 || 1283 sa->sa_len != sizeof(struct sockaddr_in6)) 1284 return; 1285 1286 if (cmd == PRC_MSGSIZE) 1287 notify = tcp_mtudisc; 1288 else if (!PRC_IS_REDIRECT(cmd) && 1289 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1290 return; 1291 /* Source quench is depreciated. */ 1292 else if (cmd == PRC_QUENCH) 1293 return; 1294 1295 /* if the parameter is from icmp6, decode it. */ 1296 if (d != NULL) { 1297 ip6cp = (struct ip6ctlparam *)d; 1298 m = ip6cp->ip6c_m; 1299 ip6 = ip6cp->ip6c_ip6; 1300 off = ip6cp->ip6c_off; 1301 sa6_src = ip6cp->ip6c_src; 1302 } else { 1303 m = NULL; 1304 ip6 = NULL; 1305 off = 0; /* fool gcc */ 1306 sa6_src = &sa6_any; 1307 } 1308 1309 if (ip6 != NULL) { 1310 struct in_conninfo inc; 1311 /* 1312 * XXX: We assume that when IPV6 is non NULL, 1313 * M and OFF are valid. 1314 */ 1315 1316 /* check if we can safely examine src and dst ports */ 1317 if (m->m_pkthdr.len < off + sizeof(*thp)) 1318 return; 1319 1320 bzero(&th, sizeof(th)); 1321 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1322 1323 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1324 (struct sockaddr *)ip6cp->ip6c_src, 1325 th.th_sport, cmd, NULL, notify); 1326 1327 inc.inc_fport = th.th_dport; 1328 inc.inc_lport = th.th_sport; 1329 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1330 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1331 inc.inc_isipv6 = 1; 1332 INP_INFO_WLOCK(&tcbinfo); 1333 syncache_unreach(&inc, &th); 1334 INP_INFO_WUNLOCK(&tcbinfo); 1335 } else 1336 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1337 0, cmd, NULL, notify); 1338} 1339#endif /* INET6 */ 1340 1341 1342/* 1343 * Following is where TCP initial sequence number generation occurs. 1344 * 1345 * There are two places where we must use initial sequence numbers: 1346 * 1. In SYN-ACK packets. 1347 * 2. In SYN packets. 1348 * 1349 * All ISNs for SYN-ACK packets are generated by the syncache. See 1350 * tcp_syncache.c for details. 1351 * 1352 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1353 * depends on this property. In addition, these ISNs should be 1354 * unguessable so as to prevent connection hijacking. To satisfy 1355 * the requirements of this situation, the algorithm outlined in 1356 * RFC 1948 is used, with only small modifications. 1357 * 1358 * Implementation details: 1359 * 1360 * Time is based off the system timer, and is corrected so that it 1361 * increases by one megabyte per second. This allows for proper 1362 * recycling on high speed LANs while still leaving over an hour 1363 * before rollover. 1364 * 1365 * As reading the *exact* system time is too expensive to be done 1366 * whenever setting up a TCP connection, we increment the time 1367 * offset in two ways. First, a small random positive increment 1368 * is added to isn_offset for each connection that is set up. 1369 * Second, the function tcp_isn_tick fires once per clock tick 1370 * and increments isn_offset as necessary so that sequence numbers 1371 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1372 * random positive increments serve only to ensure that the same 1373 * exact sequence number is never sent out twice (as could otherwise 1374 * happen when a port is recycled in less than the system tick 1375 * interval.) 1376 * 1377 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1378 * between seeding of isn_secret. This is normally set to zero, 1379 * as reseeding should not be necessary. 1380 * 1381 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1382 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1383 * general, this means holding an exclusive (write) lock. 1384 */ 1385 1386#define ISN_BYTES_PER_SECOND 1048576 1387#define ISN_STATIC_INCREMENT 4096 1388#define ISN_RANDOM_INCREMENT (4096 - 1) 1389 1390static u_char isn_secret[32]; 1391static int isn_last_reseed; 1392static u_int32_t isn_offset, isn_offset_old; 1393static MD5_CTX isn_ctx; 1394 1395tcp_seq 1396tcp_new_isn(struct tcpcb *tp) 1397{ 1398 u_int32_t md5_buffer[4]; 1399 tcp_seq new_isn; 1400 1401 INP_LOCK_ASSERT(tp->t_inpcb); 1402 1403 ISN_LOCK(); 1404 /* Seed if this is the first use, reseed if requested. */ 1405 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1406 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1407 < (u_int)ticks))) { 1408 read_random(&isn_secret, sizeof(isn_secret)); 1409 isn_last_reseed = ticks; 1410 } 1411 1412 /* Compute the md5 hash and return the ISN. */ 1413 MD5Init(&isn_ctx); 1414 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1415 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1416#ifdef INET6 1417 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1418 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1419 sizeof(struct in6_addr)); 1420 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1421 sizeof(struct in6_addr)); 1422 } else 1423#endif 1424 { 1425 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1426 sizeof(struct in_addr)); 1427 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1428 sizeof(struct in_addr)); 1429 } 1430 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1431 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1432 new_isn = (tcp_seq) md5_buffer[0]; 1433 isn_offset += ISN_STATIC_INCREMENT + 1434 (arc4random() & ISN_RANDOM_INCREMENT); 1435 new_isn += isn_offset; 1436 ISN_UNLOCK(); 1437 return (new_isn); 1438} 1439 1440/* 1441 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1442 * to keep time flowing at a relatively constant rate. If the random 1443 * increments have already pushed us past the projected offset, do nothing. 1444 */ 1445static void 1446tcp_isn_tick(void *xtp) 1447{ 1448 u_int32_t projected_offset; 1449 1450 ISN_LOCK(); 1451 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1452 1453 if (projected_offset > isn_offset) 1454 isn_offset = projected_offset; 1455 1456 isn_offset_old = isn_offset; 1457 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1458 ISN_UNLOCK(); 1459} 1460 1461/* 1462 * When a specific ICMP unreachable message is received and the 1463 * connection state is SYN-SENT, drop the connection. This behavior 1464 * is controlled by the icmp_may_rst sysctl. 1465 */ 1466struct inpcb * 1467tcp_drop_syn_sent(struct inpcb *inp, int errno) 1468{ 1469 struct tcpcb *tp; 1470 1471 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1472 INP_LOCK_ASSERT(inp); 1473 1474 if ((inp->inp_vflag & INP_TIMEWAIT) || 1475 (inp->inp_vflag & INP_DROPPED)) 1476 return (inp); 1477 1478 tp = intotcpcb(inp); 1479 if (tp->t_state != TCPS_SYN_SENT) 1480 return (inp); 1481 1482 tp = tcp_drop(tp, errno); 1483 if (tp != NULL) 1484 return (inp); 1485 else 1486 return (NULL); 1487} 1488 1489/* 1490 * When `need fragmentation' ICMP is received, update our idea of the MSS 1491 * based on the new value in the route. Also nudge TCP to send something, 1492 * since we know the packet we just sent was dropped. 1493 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1494 */ 1495struct inpcb * 1496tcp_mtudisc(struct inpcb *inp, int errno) 1497{ 1498 struct tcpcb *tp; 1499 struct socket *so = inp->inp_socket; 1500 u_int maxmtu; 1501 u_int romtu; 1502 int mss; 1503#ifdef INET6 1504 int isipv6; 1505#endif /* INET6 */ 1506 1507 INP_LOCK_ASSERT(inp); 1508 if ((inp->inp_vflag & INP_TIMEWAIT) || 1509 (inp->inp_vflag & INP_DROPPED)) 1510 return (inp); 1511 1512 tp = intotcpcb(inp); 1513 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1514 1515#ifdef INET6 1516 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1517#endif 1518 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1519 romtu = 1520#ifdef INET6 1521 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1522#endif /* INET6 */ 1523 tcp_maxmtu(&inp->inp_inc); 1524 if (!maxmtu) 1525 maxmtu = romtu; 1526 else 1527 maxmtu = min(maxmtu, romtu); 1528 if (!maxmtu) { 1529 tp->t_maxopd = tp->t_maxseg = 1530#ifdef INET6 1531 isipv6 ? tcp_v6mssdflt : 1532#endif /* INET6 */ 1533 tcp_mssdflt; 1534 return (inp); 1535 } 1536 mss = maxmtu - 1537#ifdef INET6 1538 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1539#endif /* INET6 */ 1540 sizeof(struct tcpiphdr) 1541#ifdef INET6 1542 ) 1543#endif /* INET6 */ 1544 ; 1545 1546 /* 1547 * XXX - The above conditional probably violates the TCP 1548 * spec. The problem is that, since we don't know the 1549 * other end's MSS, we are supposed to use a conservative 1550 * default. But, if we do that, then MTU discovery will 1551 * never actually take place, because the conservative 1552 * default is much less than the MTUs typically seen 1553 * on the Internet today. For the moment, we'll sweep 1554 * this under the carpet. 1555 * 1556 * The conservative default might not actually be a problem 1557 * if the only case this occurs is when sending an initial 1558 * SYN with options and data to a host we've never talked 1559 * to before. Then, they will reply with an MSS value which 1560 * will get recorded and the new parameters should get 1561 * recomputed. For Further Study. 1562 */ 1563 if (tp->t_maxopd <= mss) 1564 return (inp); 1565 tp->t_maxopd = mss; 1566 1567 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1568 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1569 mss -= TCPOLEN_TSTAMP_APPA; 1570#if (MCLBYTES & (MCLBYTES - 1)) == 0 1571 if (mss > MCLBYTES) 1572 mss &= ~(MCLBYTES-1); 1573#else 1574 if (mss > MCLBYTES) 1575 mss = mss / MCLBYTES * MCLBYTES; 1576#endif 1577 if (so->so_snd.sb_hiwat < mss) 1578 mss = so->so_snd.sb_hiwat; 1579 1580 tp->t_maxseg = mss; 1581 1582 tcpstat.tcps_mturesent++; 1583 tp->t_rtttime = 0; 1584 tp->snd_nxt = tp->snd_una; 1585 tcp_output(tp); 1586 return (inp); 1587} 1588 1589/* 1590 * Look-up the routing entry to the peer of this inpcb. If no route 1591 * is found and it cannot be allocated, then return NULL. This routine 1592 * is called by TCP routines that access the rmx structure and by tcp_mss 1593 * to get the interface MTU. 1594 */ 1595u_long 1596tcp_maxmtu(struct in_conninfo *inc) 1597{ 1598 struct route sro; 1599 struct sockaddr_in *dst; 1600 struct ifnet *ifp; 1601 u_long maxmtu = 0; 1602 1603 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1604 1605 bzero(&sro, sizeof(sro)); 1606 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1607 dst = (struct sockaddr_in *)&sro.ro_dst; 1608 dst->sin_family = AF_INET; 1609 dst->sin_len = sizeof(*dst); 1610 dst->sin_addr = inc->inc_faddr; 1611 rtalloc_ign(&sro, RTF_CLONING); 1612 } 1613 if (sro.ro_rt != NULL) { 1614 ifp = sro.ro_rt->rt_ifp; 1615 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1616 maxmtu = ifp->if_mtu; 1617 else 1618 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1619 RTFREE(sro.ro_rt); 1620 } 1621 return (maxmtu); 1622} 1623 1624#ifdef INET6 1625u_long 1626tcp_maxmtu6(struct in_conninfo *inc) 1627{ 1628 struct route_in6 sro6; 1629 struct ifnet *ifp; 1630 u_long maxmtu = 0; 1631 1632 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1633 1634 bzero(&sro6, sizeof(sro6)); 1635 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1636 sro6.ro_dst.sin6_family = AF_INET6; 1637 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1638 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1639 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1640 } 1641 if (sro6.ro_rt != NULL) { 1642 ifp = sro6.ro_rt->rt_ifp; 1643 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1644 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1645 else 1646 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1647 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1648 RTFREE(sro6.ro_rt); 1649 } 1650 1651 return (maxmtu); 1652} 1653#endif /* INET6 */ 1654 1655#ifdef IPSEC 1656/* compute ESP/AH header size for TCP, including outer IP header. */ 1657size_t 1658ipsec_hdrsiz_tcp(struct tcpcb *tp) 1659{ 1660 struct inpcb *inp; 1661 struct mbuf *m; 1662 size_t hdrsiz; 1663 struct ip *ip; 1664#ifdef INET6 1665 struct ip6_hdr *ip6; 1666#endif 1667 struct tcphdr *th; 1668 1669 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1670 return (0); 1671 MGETHDR(m, M_DONTWAIT, MT_DATA); 1672 if (!m) 1673 return (0); 1674 1675#ifdef INET6 1676 if ((inp->inp_vflag & INP_IPV6) != 0) { 1677 ip6 = mtod(m, struct ip6_hdr *); 1678 th = (struct tcphdr *)(ip6 + 1); 1679 m->m_pkthdr.len = m->m_len = 1680 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1681 tcpip_fillheaders(inp, ip6, th); 1682 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1683 } else 1684#endif /* INET6 */ 1685 { 1686 ip = mtod(m, struct ip *); 1687 th = (struct tcphdr *)(ip + 1); 1688 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1689 tcpip_fillheaders(inp, ip, th); 1690 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1691 } 1692 1693 m_free(m); 1694 return (hdrsiz); 1695} 1696#endif /*IPSEC*/ 1697 1698/* 1699 * Move a TCP connection into TIME_WAIT state. 1700 * tcbinfo is locked. 1701 * inp is locked, and is unlocked before returning. 1702 */ 1703void 1704tcp_twstart(struct tcpcb *tp) 1705{ 1706 struct tcptw *tw; 1707 struct inpcb *inp; 1708 int tw_time, acknow; 1709 struct socket *so; 1710 1711 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1712 INP_LOCK_ASSERT(tp->t_inpcb); 1713 1714 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1715 if (tw == NULL) { 1716 tw = tcp_timer_2msl_tw(1); 1717 if (tw == NULL) { 1718 tp = tcp_close(tp); 1719 if (tp != NULL) 1720 INP_UNLOCK(tp->t_inpcb); 1721 return; 1722 } 1723 } 1724 inp = tp->t_inpcb; 1725 tw->tw_inpcb = inp; 1726 1727 /* 1728 * Recover last window size sent. 1729 */ 1730 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1731 1732 /* 1733 * Set t_recent if timestamps are used on the connection. 1734 */ 1735 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1736 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1737 tw->t_recent = tp->ts_recent; 1738 else 1739 tw->t_recent = 0; 1740 1741 tw->snd_nxt = tp->snd_nxt; 1742 tw->rcv_nxt = tp->rcv_nxt; 1743 tw->iss = tp->iss; 1744 tw->irs = tp->irs; 1745 tw->t_starttime = tp->t_starttime; 1746 tw->tw_time = 0; 1747 1748/* XXX 1749 * If this code will 1750 * be used for fin-wait-2 state also, then we may need 1751 * a ts_recent from the last segment. 1752 */ 1753 tw_time = 2 * tcp_msl; 1754 acknow = tp->t_flags & TF_ACKNOW; 1755 1756 /* 1757 * First, discard tcpcb state, which includes stopping its timers and 1758 * freeing it. tcp_discardcb() used to also release the inpcb, but 1759 * that work is now done in the caller. 1760 */ 1761 tcp_discardcb(tp); 1762 so = inp->inp_socket; 1763 SOCK_LOCK(so); 1764 tw->tw_cred = crhold(so->so_cred); 1765 tw->tw_so_options = so->so_options; 1766 SOCK_UNLOCK(so); 1767 if (acknow) 1768 tcp_twrespond(tw, TH_ACK); 1769 inp->inp_ppcb = tw; 1770 inp->inp_vflag |= INP_TIMEWAIT; 1771 tcp_timer_2msl_reset(tw, tw_time); 1772 1773 /* 1774 * If the inpcb owns the sole reference to the socket, then we can 1775 * detach and free the socket as it is not needed in time wait. 1776 */ 1777 if (inp->inp_vflag & INP_SOCKREF) { 1778 KASSERT(so->so_state & SS_PROTOREF, 1779 ("tcp_twstart: !SS_PROTOREF")); 1780 inp->inp_vflag &= ~INP_SOCKREF; 1781 INP_UNLOCK(inp); 1782 ACCEPT_LOCK(); 1783 SOCK_LOCK(so); 1784 so->so_state &= ~SS_PROTOREF; 1785 sofree(so); 1786 } else 1787 INP_UNLOCK(inp); 1788} 1789 1790/* 1791 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1792 * the actual rate is slightly higher due to the addition of 1793 * random positive increments. 1794 * 1795 * Most other new OSes use semi-randomized ISN values, so we 1796 * do not need to worry about them. 1797 */ 1798#define MS_ISN_BYTES_PER_SECOND 250000 1799 1800/* 1801 * Determine if the ISN we will generate has advanced beyond the last 1802 * sequence number used by the previous connection. If so, indicate 1803 * that it is safe to recycle this tw socket by returning 1. 1804 * 1805 * XXXRW: This function should assert the inpcb lock as it does multiple 1806 * non-atomic reads from the tcptw, but is currently called without it from 1807 * in_pcb.c:in_pcblookup_local(). 1808 */ 1809int 1810tcp_twrecycleable(struct tcptw *tw) 1811{ 1812 tcp_seq new_iss = tw->iss; 1813 tcp_seq new_irs = tw->irs; 1814 1815 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1816 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1817 1818 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1819 return (1); 1820 else 1821 return (0); 1822} 1823 1824void 1825tcp_twclose(struct tcptw *tw, int reuse) 1826{ 1827 struct socket *so; 1828 struct inpcb *inp; 1829 1830 /* 1831 * At this point, we are in one of two situations: 1832 * 1833 * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1834 * all state. 1835 * 1836 * (2) We have a socket -- if we own a reference, release it and 1837 * notify the socket layer. 1838 */ 1839 inp = tw->tw_inpcb; 1840 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1841 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1842 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1843 INP_LOCK_ASSERT(inp); 1844 1845 tw->tw_inpcb = NULL; 1846 tcp_timer_2msl_stop(tw); 1847 inp->inp_ppcb = NULL; 1848 in_pcbdrop(inp); 1849 1850 so = inp->inp_socket; 1851 if (so != NULL) { 1852 /* 1853 * If there's a socket, handle two cases: first, we own a 1854 * strong reference, which we will now release, or we don't 1855 * in which case another reference exists (XXXRW: think 1856 * about this more), and we don't need to take action. 1857 */ 1858 if (inp->inp_vflag & INP_SOCKREF) { 1859 inp->inp_vflag &= ~INP_SOCKREF; 1860 INP_UNLOCK(inp); 1861 ACCEPT_LOCK(); 1862 SOCK_LOCK(so); 1863 KASSERT(so->so_state & SS_PROTOREF, 1864 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1865 so->so_state &= ~SS_PROTOREF; 1866 sofree(so); 1867 } else { 1868 /* 1869 * If we don't own the only reference, the socket and 1870 * inpcb need to be left around to be handled by 1871 * tcp_usr_detach() later. 1872 */ 1873 INP_UNLOCK(inp); 1874 } 1875 } else { 1876#ifdef INET6 1877 if (inp->inp_vflag & INP_IPV6PROTO) 1878 in6_pcbfree(inp); 1879 else 1880#endif 1881 in_pcbfree(inp); 1882 } 1883 tcpstat.tcps_closed++; 1884 crfree(tw->tw_cred); 1885 tw->tw_cred = NULL; 1886 if (reuse) 1887 return; 1888 uma_zfree(tcptw_zone, tw); 1889} 1890 1891int 1892tcp_twrespond(struct tcptw *tw, int flags) 1893{ 1894 struct inpcb *inp = tw->tw_inpcb; 1895 struct tcphdr *th; 1896 struct mbuf *m; 1897 struct ip *ip = NULL; 1898 u_int8_t *optp; 1899 u_int hdrlen, optlen; 1900 int error; 1901#ifdef INET6 1902 struct ip6_hdr *ip6 = NULL; 1903 int isipv6 = inp->inp_inc.inc_isipv6; 1904#endif 1905 1906 INP_LOCK_ASSERT(inp); 1907 1908 m = m_gethdr(M_DONTWAIT, MT_DATA); 1909 if (m == NULL) 1910 return (ENOBUFS); 1911 m->m_data += max_linkhdr; 1912 1913#ifdef MAC 1914 mac_create_mbuf_from_inpcb(inp, m); 1915#endif 1916 1917#ifdef INET6 1918 if (isipv6) { 1919 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1920 ip6 = mtod(m, struct ip6_hdr *); 1921 th = (struct tcphdr *)(ip6 + 1); 1922 tcpip_fillheaders(inp, ip6, th); 1923 } else 1924#endif 1925 { 1926 hdrlen = sizeof(struct tcpiphdr); 1927 ip = mtod(m, struct ip *); 1928 th = (struct tcphdr *)(ip + 1); 1929 tcpip_fillheaders(inp, ip, th); 1930 } 1931 optp = (u_int8_t *)(th + 1); 1932 1933 /* 1934 * Send a timestamp and echo-reply if both our side and our peer 1935 * have sent timestamps in our SYN's and this is not a RST. 1936 */ 1937 if (tw->t_recent && flags == TH_ACK) { 1938 u_int32_t *lp = (u_int32_t *)optp; 1939 1940 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1941 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1942 *lp++ = htonl(ticks); 1943 *lp = htonl(tw->t_recent); 1944 optp += TCPOLEN_TSTAMP_APPA; 1945 } 1946 1947 optlen = optp - (u_int8_t *)(th + 1); 1948 1949 m->m_len = hdrlen + optlen; 1950 m->m_pkthdr.len = m->m_len; 1951 1952 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1953 1954 th->th_seq = htonl(tw->snd_nxt); 1955 th->th_ack = htonl(tw->rcv_nxt); 1956 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1957 th->th_flags = flags; 1958 th->th_win = htons(tw->last_win); 1959 1960#ifdef INET6 1961 if (isipv6) { 1962 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1963 sizeof(struct tcphdr) + optlen); 1964 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1965 error = ip6_output(m, inp->in6p_outputopts, NULL, 1966 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1967 } else 1968#endif 1969 { 1970 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1971 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1972 m->m_pkthdr.csum_flags = CSUM_TCP; 1973 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1974 ip->ip_len = m->m_pkthdr.len; 1975 if (path_mtu_discovery) 1976 ip->ip_off |= IP_DF; 1977 error = ip_output(m, inp->inp_options, NULL, 1978 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1979 NULL, inp); 1980 } 1981 if (flags & TH_ACK) 1982 tcpstat.tcps_sndacks++; 1983 else 1984 tcpstat.tcps_sndctrl++; 1985 tcpstat.tcps_sndtotal++; 1986 return (error); 1987} 1988 1989/* 1990 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1991 * 1992 * This code attempts to calculate the bandwidth-delay product as a 1993 * means of determining the optimal window size to maximize bandwidth, 1994 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1995 * routers. This code also does a fairly good job keeping RTTs in check 1996 * across slow links like modems. We implement an algorithm which is very 1997 * similar (but not meant to be) TCP/Vegas. The code operates on the 1998 * transmitter side of a TCP connection and so only effects the transmit 1999 * side of the connection. 2000 * 2001 * BACKGROUND: TCP makes no provision for the management of buffer space 2002 * at the end points or at the intermediate routers and switches. A TCP 2003 * stream, whether using NewReno or not, will eventually buffer as 2004 * many packets as it is able and the only reason this typically works is 2005 * due to the fairly small default buffers made available for a connection 2006 * (typicaly 16K or 32K). As machines use larger windows and/or window 2007 * scaling it is now fairly easy for even a single TCP connection to blow-out 2008 * all available buffer space not only on the local interface, but on 2009 * intermediate routers and switches as well. NewReno makes a misguided 2010 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2011 * then backing off, then steadily increasing the window again until another 2012 * failure occurs, ad-infinitum. This results in terrible oscillation that 2013 * is only made worse as network loads increase and the idea of intentionally 2014 * blowing out network buffers is, frankly, a terrible way to manage network 2015 * resources. 2016 * 2017 * It is far better to limit the transmit window prior to the failure 2018 * condition being achieved. There are two general ways to do this: First 2019 * you can 'scan' through different transmit window sizes and locate the 2020 * point where the RTT stops increasing, indicating that you have filled the 2021 * pipe, then scan backwards until you note that RTT stops decreasing, then 2022 * repeat ad-infinitum. This method works in principle but has severe 2023 * implementation issues due to RTT variances, timer granularity, and 2024 * instability in the algorithm which can lead to many false positives and 2025 * create oscillations as well as interact badly with other TCP streams 2026 * implementing the same algorithm. 2027 * 2028 * The second method is to limit the window to the bandwidth delay product 2029 * of the link. This is the method we implement. RTT variances and our 2030 * own manipulation of the congestion window, bwnd, can potentially 2031 * destabilize the algorithm. For this reason we have to stabilize the 2032 * elements used to calculate the window. We do this by using the minimum 2033 * observed RTT, the long term average of the observed bandwidth, and 2034 * by adding two segments worth of slop. It isn't perfect but it is able 2035 * to react to changing conditions and gives us a very stable basis on 2036 * which to extend the algorithm. 2037 */ 2038void 2039tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2040{ 2041 u_long bw; 2042 u_long bwnd; 2043 int save_ticks; 2044 2045 INP_LOCK_ASSERT(tp->t_inpcb); 2046 2047 /* 2048 * If inflight_enable is disabled in the middle of a tcp connection, 2049 * make sure snd_bwnd is effectively disabled. 2050 */ 2051 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2052 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2053 tp->snd_bandwidth = 0; 2054 return; 2055 } 2056 2057 /* 2058 * Figure out the bandwidth. Due to the tick granularity this 2059 * is a very rough number and it MUST be averaged over a fairly 2060 * long period of time. XXX we need to take into account a link 2061 * that is not using all available bandwidth, but for now our 2062 * slop will ramp us up if this case occurs and the bandwidth later 2063 * increases. 2064 * 2065 * Note: if ticks rollover 'bw' may wind up negative. We must 2066 * effectively reset t_bw_rtttime for this case. 2067 */ 2068 save_ticks = ticks; 2069 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2070 return; 2071 2072 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2073 (save_ticks - tp->t_bw_rtttime); 2074 tp->t_bw_rtttime = save_ticks; 2075 tp->t_bw_rtseq = ack_seq; 2076 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2077 return; 2078 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2079 2080 tp->snd_bandwidth = bw; 2081 2082 /* 2083 * Calculate the semi-static bandwidth delay product, plus two maximal 2084 * segments. The additional slop puts us squarely in the sweet 2085 * spot and also handles the bandwidth run-up case and stabilization. 2086 * Without the slop we could be locking ourselves into a lower 2087 * bandwidth. 2088 * 2089 * Situations Handled: 2090 * (1) Prevents over-queueing of packets on LANs, especially on 2091 * high speed LANs, allowing larger TCP buffers to be 2092 * specified, and also does a good job preventing 2093 * over-queueing of packets over choke points like modems 2094 * (at least for the transmit side). 2095 * 2096 * (2) Is able to handle changing network loads (bandwidth 2097 * drops so bwnd drops, bandwidth increases so bwnd 2098 * increases). 2099 * 2100 * (3) Theoretically should stabilize in the face of multiple 2101 * connections implementing the same algorithm (this may need 2102 * a little work). 2103 * 2104 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2105 * be adjusted with a sysctl but typically only needs to be 2106 * on very slow connections. A value no smaller then 5 2107 * should be used, but only reduce this default if you have 2108 * no other choice. 2109 */ 2110#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2111 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2112#undef USERTT 2113 2114 if (tcp_inflight_debug > 0) { 2115 static int ltime; 2116 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2117 ltime = ticks; 2118 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2119 tp, 2120 bw, 2121 tp->t_rttbest, 2122 tp->t_srtt, 2123 bwnd 2124 ); 2125 } 2126 } 2127 if ((long)bwnd < tcp_inflight_min) 2128 bwnd = tcp_inflight_min; 2129 if (bwnd > tcp_inflight_max) 2130 bwnd = tcp_inflight_max; 2131 if ((long)bwnd < tp->t_maxseg * 2) 2132 bwnd = tp->t_maxseg * 2; 2133 tp->snd_bwnd = bwnd; 2134} 2135 2136#ifdef TCP_SIGNATURE 2137/* 2138 * Callback function invoked by m_apply() to digest TCP segment data 2139 * contained within an mbuf chain. 2140 */ 2141static int 2142tcp_signature_apply(void *fstate, void *data, u_int len) 2143{ 2144 2145 MD5Update(fstate, (u_char *)data, len); 2146 return (0); 2147} 2148 2149/* 2150 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2151 * 2152 * Parameters: 2153 * m pointer to head of mbuf chain 2154 * off0 offset to TCP header within the mbuf chain 2155 * len length of TCP segment data, excluding options 2156 * optlen length of TCP segment options 2157 * buf pointer to storage for computed MD5 digest 2158 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2159 * 2160 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2161 * When called from tcp_input(), we can be sure that th_sum has been 2162 * zeroed out and verified already. 2163 * 2164 * This function is for IPv4 use only. Calling this function with an 2165 * IPv6 packet in the mbuf chain will yield undefined results. 2166 * 2167 * Return 0 if successful, otherwise return -1. 2168 * 2169 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2170 * search with the destination IP address, and a 'magic SPI' to be 2171 * determined by the application. This is hardcoded elsewhere to 1179 2172 * right now. Another branch of this code exists which uses the SPD to 2173 * specify per-application flows but it is unstable. 2174 */ 2175int 2176tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2177 u_char *buf, u_int direction) 2178{ 2179 union sockaddr_union dst; 2180 struct ippseudo ippseudo; 2181 MD5_CTX ctx; 2182 int doff; 2183 struct ip *ip; 2184 struct ipovly *ipovly; 2185 struct secasvar *sav; 2186 struct tcphdr *th; 2187 u_short savecsum; 2188 2189 KASSERT(m != NULL, ("NULL mbuf chain")); 2190 KASSERT(buf != NULL, ("NULL signature pointer")); 2191 2192 /* Extract the destination from the IP header in the mbuf. */ 2193 ip = mtod(m, struct ip *); 2194 bzero(&dst, sizeof(union sockaddr_union)); 2195 dst.sa.sa_len = sizeof(struct sockaddr_in); 2196 dst.sa.sa_family = AF_INET; 2197 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2198 ip->ip_src : ip->ip_dst; 2199 2200 /* Look up an SADB entry which matches the address of the peer. */ 2201 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2202 if (sav == NULL) { 2203 printf("%s: SADB lookup failed for %s\n", __func__, 2204 inet_ntoa(dst.sin.sin_addr)); 2205 return (EINVAL); 2206 } 2207 2208 MD5Init(&ctx); 2209 ipovly = (struct ipovly *)ip; 2210 th = (struct tcphdr *)((u_char *)ip + off0); 2211 doff = off0 + sizeof(struct tcphdr) + optlen; 2212 2213 /* 2214 * Step 1: Update MD5 hash with IP pseudo-header. 2215 * 2216 * XXX The ippseudo header MUST be digested in network byte order, 2217 * or else we'll fail the regression test. Assume all fields we've 2218 * been doing arithmetic on have been in host byte order. 2219 * XXX One cannot depend on ipovly->ih_len here. When called from 2220 * tcp_output(), the underlying ip_len member has not yet been set. 2221 */ 2222 ippseudo.ippseudo_src = ipovly->ih_src; 2223 ippseudo.ippseudo_dst = ipovly->ih_dst; 2224 ippseudo.ippseudo_pad = 0; 2225 ippseudo.ippseudo_p = IPPROTO_TCP; 2226 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2227 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2228 2229 /* 2230 * Step 2: Update MD5 hash with TCP header, excluding options. 2231 * The TCP checksum must be set to zero. 2232 */ 2233 savecsum = th->th_sum; 2234 th->th_sum = 0; 2235 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2236 th->th_sum = savecsum; 2237 2238 /* 2239 * Step 3: Update MD5 hash with TCP segment data. 2240 * Use m_apply() to avoid an early m_pullup(). 2241 */ 2242 if (len > 0) 2243 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2244 2245 /* 2246 * Step 4: Update MD5 hash with shared secret. 2247 */ 2248 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2249 MD5Final(buf, &ctx); 2250 2251 key_sa_recordxfer(sav, m); 2252 KEY_FREESAV(&sav); 2253 return (0); 2254} 2255#endif /* TCP_SIGNATURE */ 2256 2257static int 2258sysctl_drop(SYSCTL_HANDLER_ARGS) 2259{ 2260 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2261 struct sockaddr_storage addrs[2]; 2262 struct inpcb *inp; 2263 struct tcpcb *tp; 2264 struct tcptw *tw; 2265 struct sockaddr_in *fin, *lin; 2266#ifdef INET6 2267 struct sockaddr_in6 *fin6, *lin6; 2268 struct in6_addr f6, l6; 2269#endif 2270 int error; 2271 2272 inp = NULL; 2273 fin = lin = NULL; 2274#ifdef INET6 2275 fin6 = lin6 = NULL; 2276#endif 2277 error = 0; 2278 2279 if (req->oldptr != NULL || req->oldlen != 0) 2280 return (EINVAL); 2281 if (req->newptr == NULL) 2282 return (EPERM); 2283 if (req->newlen < sizeof(addrs)) 2284 return (ENOMEM); 2285 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2286 if (error) 2287 return (error); 2288 2289 switch (addrs[0].ss_family) { 2290#ifdef INET6 2291 case AF_INET6: 2292 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2293 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2294 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2295 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2296 return (EINVAL); 2297 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2298 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2299 return (EINVAL); 2300 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2301 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2302 fin = (struct sockaddr_in *)&addrs[0]; 2303 lin = (struct sockaddr_in *)&addrs[1]; 2304 break; 2305 } 2306 error = sa6_embedscope(fin6, ip6_use_defzone); 2307 if (error) 2308 return (error); 2309 error = sa6_embedscope(lin6, ip6_use_defzone); 2310 if (error) 2311 return (error); 2312 break; 2313#endif 2314 case AF_INET: 2315 fin = (struct sockaddr_in *)&addrs[0]; 2316 lin = (struct sockaddr_in *)&addrs[1]; 2317 if (fin->sin_len != sizeof(struct sockaddr_in) || 2318 lin->sin_len != sizeof(struct sockaddr_in)) 2319 return (EINVAL); 2320 break; 2321 default: 2322 return (EINVAL); 2323 } 2324 INP_INFO_WLOCK(&tcbinfo); 2325 switch (addrs[0].ss_family) { 2326#ifdef INET6 2327 case AF_INET6: 2328 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2329 &l6, lin6->sin6_port, 0, NULL); 2330 break; 2331#endif 2332 case AF_INET: 2333 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2334 lin->sin_addr, lin->sin_port, 0, NULL); 2335 break; 2336 } 2337 if (inp != NULL) { 2338 INP_LOCK(inp); 2339 if (inp->inp_vflag & INP_TIMEWAIT) { 2340 /* 2341 * XXXRW: There currently exists a state where an 2342 * inpcb is present, but its timewait state has been 2343 * discarded. For now, don't allow dropping of this 2344 * type of inpcb. 2345 */ 2346 tw = intotw(inp); 2347 if (tw != NULL) 2348 tcp_twclose(tw, 0); 2349 } else if (!(inp->inp_vflag & INP_DROPPED) && 2350 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2351 tp = intotcpcb(inp); 2352 tcp_drop(tp, ECONNABORTED); 2353 } 2354 INP_UNLOCK(inp); 2355 } else 2356 error = ESRCH; 2357 INP_INFO_WUNLOCK(&tcbinfo); 2358 return (error); 2359} 2360 2361SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2362 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2363 0, sysctl_drop, "", "Drop TCP connection"); 2364