tcp_timewait.c revision 157967
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 157967 2006-04-22 19:23:24Z 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/* 250 * TCP initialization. 251 */ 252static void 253tcp_zone_change(void *tag) 254{ 255 256 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 257 uma_zone_set_max(tcpcb_zone, maxsockets); 258 uma_zone_set_max(tcptw_zone, maxsockets / 5); 259} 260 261void 262tcp_init(void) 263{ 264 int hashsize = TCBHASHSIZE; 265 266 tcp_delacktime = TCPTV_DELACK; 267 tcp_keepinit = TCPTV_KEEP_INIT; 268 tcp_keepidle = TCPTV_KEEP_IDLE; 269 tcp_keepintvl = TCPTV_KEEPINTVL; 270 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 271 tcp_msl = TCPTV_MSL; 272 tcp_rexmit_min = TCPTV_MIN; 273 tcp_rexmit_slop = TCPTV_CPU_VAR; 274 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 275 276 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 277 LIST_INIT(&tcb); 278 tcbinfo.listhead = &tcb; 279 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 280 if (!powerof2(hashsize)) { 281 printf("WARNING: TCB hash size not a power of 2\n"); 282 hashsize = 512; /* safe default */ 283 } 284 tcp_tcbhashsize = hashsize; 285 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 286 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 287 &tcbinfo.porthashmask); 288 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 289 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 290 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 291#ifdef INET6 292#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 293#else /* INET6 */ 294#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 295#endif /* INET6 */ 296 if (max_protohdr < TCP_MINPROTOHDR) 297 max_protohdr = TCP_MINPROTOHDR; 298 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 299 panic("tcp_init"); 300#undef TCP_MINPROTOHDR 301 /* 302 * These have to be type stable for the benefit of the timers. 303 */ 304 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 305 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 306 uma_zone_set_max(tcpcb_zone, maxsockets); 307 TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 308 if (maxtcptw == 0) 309 maxtcptw = maxsockets / 5; 310 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 311 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 312 uma_zone_set_max(tcptw_zone, maxtcptw); 313 tcp_timer_init(); 314 syncache_init(); 315 tcp_hc_init(); 316 tcp_reass_init(); 317 mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF); 318 callout_init(&isn_callout, CALLOUT_MPSAFE); 319 tcp_isn_tick(NULL); 320 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 321 SHUTDOWN_PRI_DEFAULT); 322 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 323 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 324 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 325 EVENTHANDLER_PRI_ANY); 326} 327 328void 329tcp_fini(void *xtp) 330{ 331 332 callout_stop(&isn_callout); 333} 334 335/* 336 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 337 * tcp_template used to store this data in mbufs, but we now recopy it out 338 * of the tcpcb each time to conserve mbufs. 339 */ 340void 341tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 342{ 343 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 344 345 INP_LOCK_ASSERT(inp); 346 347#ifdef INET6 348 if ((inp->inp_vflag & INP_IPV6) != 0) { 349 struct ip6_hdr *ip6; 350 351 ip6 = (struct ip6_hdr *)ip_ptr; 352 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 353 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 354 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 355 (IPV6_VERSION & IPV6_VERSION_MASK); 356 ip6->ip6_nxt = IPPROTO_TCP; 357 ip6->ip6_plen = sizeof(struct tcphdr); 358 ip6->ip6_src = inp->in6p_laddr; 359 ip6->ip6_dst = inp->in6p_faddr; 360 } else 361#endif 362 { 363 struct ip *ip; 364 365 ip = (struct ip *)ip_ptr; 366 ip->ip_v = IPVERSION; 367 ip->ip_hl = 5; 368 ip->ip_tos = inp->inp_ip_tos; 369 ip->ip_len = 0; 370 ip->ip_id = 0; 371 ip->ip_off = 0; 372 ip->ip_ttl = inp->inp_ip_ttl; 373 ip->ip_sum = 0; 374 ip->ip_p = IPPROTO_TCP; 375 ip->ip_src = inp->inp_laddr; 376 ip->ip_dst = inp->inp_faddr; 377 } 378 th->th_sport = inp->inp_lport; 379 th->th_dport = inp->inp_fport; 380 th->th_seq = 0; 381 th->th_ack = 0; 382 th->th_x2 = 0; 383 th->th_off = 5; 384 th->th_flags = 0; 385 th->th_win = 0; 386 th->th_urp = 0; 387 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 388} 389 390/* 391 * Create template to be used to send tcp packets on a connection. 392 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 393 * use for this function is in keepalives, which use tcp_respond. 394 */ 395struct tcptemp * 396tcpip_maketemplate(struct inpcb *inp) 397{ 398 struct mbuf *m; 399 struct tcptemp *n; 400 401 m = m_get(M_DONTWAIT, MT_DATA); 402 if (m == NULL) 403 return (0); 404 m->m_len = sizeof(struct tcptemp); 405 n = mtod(m, struct tcptemp *); 406 407 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 408 return (n); 409} 410 411/* 412 * Send a single message to the TCP at address specified by 413 * the given TCP/IP header. If m == NULL, then we make a copy 414 * of the tcpiphdr at ti and send directly to the addressed host. 415 * This is used to force keep alive messages out using the TCP 416 * template for a connection. If flags are given then we send 417 * a message back to the TCP which originated the * segment ti, 418 * and discard the mbuf containing it and any other attached mbufs. 419 * 420 * In any case the ack and sequence number of the transmitted 421 * segment are as specified by the parameters. 422 * 423 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 424 */ 425void 426tcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th, 427 register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 428{ 429 register int tlen; 430 int win = 0; 431 struct ip *ip; 432 struct tcphdr *nth; 433#ifdef INET6 434 struct ip6_hdr *ip6; 435 int isipv6; 436#endif /* INET6 */ 437 int ipflags = 0; 438 struct inpcb *inp; 439 440 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 441 442#ifdef INET6 443 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 444 ip6 = ipgen; 445#endif /* INET6 */ 446 ip = ipgen; 447 448 if (tp != NULL) { 449 inp = tp->t_inpcb; 450 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 451 INP_INFO_WLOCK_ASSERT(&tcbinfo); 452 INP_LOCK_ASSERT(inp); 453 } else 454 inp = NULL; 455 456 if (tp != NULL) { 457 if (!(flags & TH_RST)) { 458 win = sbspace(&inp->inp_socket->so_rcv); 459 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 460 win = (long)TCP_MAXWIN << tp->rcv_scale; 461 } 462 } 463 if (m == NULL) { 464 m = m_gethdr(M_DONTWAIT, MT_DATA); 465 if (m == NULL) 466 return; 467 tlen = 0; 468 m->m_data += max_linkhdr; 469#ifdef INET6 470 if (isipv6) { 471 bcopy((caddr_t)ip6, mtod(m, caddr_t), 472 sizeof(struct ip6_hdr)); 473 ip6 = mtod(m, struct ip6_hdr *); 474 nth = (struct tcphdr *)(ip6 + 1); 475 } else 476#endif /* INET6 */ 477 { 478 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 479 ip = mtod(m, struct ip *); 480 nth = (struct tcphdr *)(ip + 1); 481 } 482 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 483 flags = TH_ACK; 484 } else { 485 m_freem(m->m_next); 486 m->m_next = NULL; 487 m->m_data = (caddr_t)ipgen; 488 /* m_len is set later */ 489 tlen = 0; 490#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 491#ifdef INET6 492 if (isipv6) { 493 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 494 nth = (struct tcphdr *)(ip6 + 1); 495 } else 496#endif /* INET6 */ 497 { 498 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 499 nth = (struct tcphdr *)(ip + 1); 500 } 501 if (th != nth) { 502 /* 503 * this is usually a case when an extension header 504 * exists between the IPv6 header and the 505 * TCP header. 506 */ 507 nth->th_sport = th->th_sport; 508 nth->th_dport = th->th_dport; 509 } 510 xchg(nth->th_dport, nth->th_sport, n_short); 511#undef xchg 512 } 513#ifdef INET6 514 if (isipv6) { 515 ip6->ip6_flow = 0; 516 ip6->ip6_vfc = IPV6_VERSION; 517 ip6->ip6_nxt = IPPROTO_TCP; 518 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 519 tlen)); 520 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 521 } else 522#endif 523 { 524 tlen += sizeof (struct tcpiphdr); 525 ip->ip_len = tlen; 526 ip->ip_ttl = ip_defttl; 527 if (path_mtu_discovery) 528 ip->ip_off |= IP_DF; 529 } 530 m->m_len = tlen; 531 m->m_pkthdr.len = tlen; 532 m->m_pkthdr.rcvif = NULL; 533#ifdef MAC 534 if (inp != NULL) { 535 /* 536 * Packet is associated with a socket, so allow the 537 * label of the response to reflect the socket label. 538 */ 539 INP_LOCK_ASSERT(inp); 540 mac_create_mbuf_from_inpcb(inp, m); 541 } else { 542 /* 543 * Packet is not associated with a socket, so possibly 544 * update the label in place. 545 */ 546 mac_reflect_mbuf_tcp(m); 547 } 548#endif 549 nth->th_seq = htonl(seq); 550 nth->th_ack = htonl(ack); 551 nth->th_x2 = 0; 552 nth->th_off = sizeof (struct tcphdr) >> 2; 553 nth->th_flags = flags; 554 if (tp != NULL) 555 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 556 else 557 nth->th_win = htons((u_short)win); 558 nth->th_urp = 0; 559#ifdef INET6 560 if (isipv6) { 561 nth->th_sum = 0; 562 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 563 sizeof(struct ip6_hdr), 564 tlen - sizeof(struct ip6_hdr)); 565 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 566 NULL, NULL); 567 } else 568#endif /* INET6 */ 569 { 570 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 571 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 572 m->m_pkthdr.csum_flags = CSUM_TCP; 573 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 574 } 575#ifdef TCPDEBUG 576 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 577 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 578#endif 579#ifdef INET6 580 if (isipv6) 581 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 582 else 583#endif /* INET6 */ 584 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 585} 586 587/* 588 * Create a new TCP control block, making an 589 * empty reassembly queue and hooking it to the argument 590 * protocol control block. The `inp' parameter must have 591 * come from the zone allocator set up in tcp_init(). 592 */ 593struct tcpcb * 594tcp_newtcpcb(struct inpcb *inp) 595{ 596 struct tcpcb_mem *tm; 597 struct tcpcb *tp; 598#ifdef INET6 599 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 600#endif /* INET6 */ 601 602 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 603 if (tm == NULL) 604 return (NULL); 605 tp = &tm->tcb; 606 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 607 tp->t_maxseg = tp->t_maxopd = 608#ifdef INET6 609 isipv6 ? tcp_v6mssdflt : 610#endif /* INET6 */ 611 tcp_mssdflt; 612 613 /* Set up our timeouts. */ 614 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 615 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 616 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 617 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 618 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 619 620 if (tcp_do_rfc1323) 621 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 622 tp->sack_enable = tcp_do_sack; 623 TAILQ_INIT(&tp->snd_holes); 624 tp->t_inpcb = inp; /* XXX */ 625 /* 626 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 627 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 628 * reasonable initial retransmit time. 629 */ 630 tp->t_srtt = TCPTV_SRTTBASE; 631 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 632 tp->t_rttmin = tcp_rexmit_min; 633 tp->t_rxtcur = TCPTV_RTOBASE; 634 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 635 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 636 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 637 tp->t_rcvtime = ticks; 638 tp->t_bw_rtttime = ticks; 639 /* 640 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 641 * because the socket may be bound to an IPv6 wildcard address, 642 * which may match an IPv4-mapped IPv6 address. 643 */ 644 inp->inp_ip_ttl = ip_defttl; 645 inp->inp_ppcb = tp; 646 return (tp); /* XXX */ 647} 648 649/* 650 * Drop a TCP connection, reporting 651 * the specified error. If connection is synchronized, 652 * then send a RST to peer. 653 */ 654struct tcpcb * 655tcp_drop(struct tcpcb *tp, int errno) 656{ 657 struct socket *so = tp->t_inpcb->inp_socket; 658 659 INP_INFO_WLOCK_ASSERT(&tcbinfo); 660 INP_LOCK_ASSERT(tp->t_inpcb); 661 662 if (TCPS_HAVERCVDSYN(tp->t_state)) { 663 tp->t_state = TCPS_CLOSED; 664 (void) tcp_output(tp); 665 tcpstat.tcps_drops++; 666 } else 667 tcpstat.tcps_conndrops++; 668 if (errno == ETIMEDOUT && tp->t_softerror) 669 errno = tp->t_softerror; 670 so->so_error = errno; 671 return (tcp_close(tp)); 672} 673 674void 675tcp_discardcb(struct tcpcb *tp) 676{ 677 struct tseg_qent *q; 678 struct inpcb *inp = tp->t_inpcb; 679 struct socket *so = inp->inp_socket; 680#ifdef INET6 681 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 682#endif /* INET6 */ 683 684 /* 685 * XXXRW: This is all very well and good, but actually, we might be 686 * discarding the tcpcb after the socket is gone, so we can't do 687 * this: 688 KASSERT(so != NULL, ("tcp_discardcb: so == NULL")); 689 */ 690 INP_LOCK_ASSERT(inp); 691 692 /* 693 * Make sure that all of our timers are stopped before we 694 * delete the PCB. 695 */ 696 callout_stop(tp->tt_rexmt); 697 callout_stop(tp->tt_persist); 698 callout_stop(tp->tt_keep); 699 callout_stop(tp->tt_2msl); 700 callout_stop(tp->tt_delack); 701 702 /* 703 * If we got enough samples through the srtt filter, 704 * save the rtt and rttvar in the routing entry. 705 * 'Enough' is arbitrarily defined as 4 rtt samples. 706 * 4 samples is enough for the srtt filter to converge 707 * to within enough % of the correct value; fewer samples 708 * and we could save a bogus rtt. The danger is not high 709 * as tcp quickly recovers from everything. 710 * XXX: Works very well but needs some more statistics! 711 */ 712 if (tp->t_rttupdated >= 4) { 713 struct hc_metrics_lite metrics; 714 u_long ssthresh; 715 716 bzero(&metrics, sizeof(metrics)); 717 /* 718 * Update the ssthresh always when the conditions below 719 * are satisfied. This gives us better new start value 720 * for the congestion avoidance for new connections. 721 * ssthresh is only set if packet loss occured on a session. 722 */ 723 ssthresh = tp->snd_ssthresh; 724 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 725 /* 726 * convert the limit from user data bytes to 727 * packets then to packet data bytes. 728 */ 729 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 730 if (ssthresh < 2) 731 ssthresh = 2; 732 ssthresh *= (u_long)(tp->t_maxseg + 733#ifdef INET6 734 (isipv6 ? sizeof (struct ip6_hdr) + 735 sizeof (struct tcphdr) : 736#endif 737 sizeof (struct tcpiphdr) 738#ifdef INET6 739 ) 740#endif 741 ); 742 } else 743 ssthresh = 0; 744 metrics.rmx_ssthresh = ssthresh; 745 746 metrics.rmx_rtt = tp->t_srtt; 747 metrics.rmx_rttvar = tp->t_rttvar; 748 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 749 metrics.rmx_bandwidth = tp->snd_bandwidth; 750 metrics.rmx_cwnd = tp->snd_cwnd; 751 metrics.rmx_sendpipe = 0; 752 metrics.rmx_recvpipe = 0; 753 754 tcp_hc_update(&inp->inp_inc, &metrics); 755 } 756 757 /* free the reassembly queue, if any */ 758 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 759 LIST_REMOVE(q, tqe_q); 760 m_freem(q->tqe_m); 761 uma_zfree(tcp_reass_zone, q); 762 tp->t_segqlen--; 763 tcp_reass_qsize--; 764 } 765 tcp_free_sackholes(tp); 766 inp->inp_ppcb = NULL; 767 tp->t_inpcb = NULL; 768 uma_zfree(tcpcb_zone, tp); 769 770 /* 771 * XXXRW: This seems a bit unclean. 772 */ 773 if (so != NULL) 774 soisdisconnected(so); 775} 776 777/* 778 * Attempt to close a TCP control block, marking it as dropped, and freeing 779 * the socket if we hold the only reference. 780 */ 781struct tcpcb * 782tcp_close(struct tcpcb *tp) 783{ 784 struct inpcb *inp = tp->t_inpcb; 785 struct socket *so; 786 787 INP_INFO_WLOCK_ASSERT(&tcbinfo); 788 INP_LOCK_ASSERT(inp); 789 790 inp->inp_vflag |= INP_DROPPED; 791 792 tcpstat.tcps_closed++; 793 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 794 so = inp->inp_socket; 795 soisdisconnected(so); 796 if (inp->inp_vflag & INP_SOCKREF) { 797 KASSERT(so->so_state & SS_PROTOREF, 798 ("tcp_close: !SS_PROTOREF")); 799 inp->inp_vflag &= ~INP_SOCKREF; 800 tcp_discardcb(tp); 801#ifdef INET6 802 if (inp->inp_vflag & INP_IPV6PROTO) { 803 in6_pcbdetach(inp); 804 in6_pcbfree(inp); 805 } else { 806#endif 807 in_pcbdetach(inp); 808 in_pcbfree(inp); 809#ifdef INET6 810 } 811#endif 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 if (inp->inp_gencnt <= gencnt) { 990 struct xtcpcb xt; 991 void *inp_ppcb; 992 993 bzero(&xt, sizeof(xt)); 994 xt.xt_len = sizeof xt; 995 /* XXX should avoid extra copy */ 996 bcopy(inp, &xt.xt_inp, sizeof *inp); 997 inp_ppcb = inp->inp_ppcb; 998 if (inp_ppcb == NULL) 999 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1000 else if (inp->inp_vflag & INP_TIMEWAIT) { 1001 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1002 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1003 } else 1004 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1005 if (inp->inp_socket != NULL) 1006 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1007 else { 1008 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1009 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1010 } 1011 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1012 error = SYSCTL_OUT(req, &xt, sizeof xt); 1013 } 1014 } 1015 if (!error) { 1016 /* 1017 * Give the user an updated idea of our state. 1018 * If the generation differs from what we told 1019 * her before, she knows that something happened 1020 * while we were processing this request, and it 1021 * might be necessary to retry. 1022 */ 1023 INP_INFO_RLOCK(&tcbinfo); 1024 xig.xig_gen = tcbinfo.ipi_gencnt; 1025 xig.xig_sogen = so_gencnt; 1026 xig.xig_count = tcbinfo.ipi_count; 1027 INP_INFO_RUNLOCK(&tcbinfo); 1028 error = SYSCTL_OUT(req, &xig, sizeof xig); 1029 } 1030 free(inp_list, M_TEMP); 1031 return (error); 1032} 1033 1034SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1035 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1036 1037static int 1038tcp_getcred(SYSCTL_HANDLER_ARGS) 1039{ 1040 struct xucred xuc; 1041 struct sockaddr_in addrs[2]; 1042 struct inpcb *inp; 1043 int error; 1044 1045 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1046 if (error) 1047 return (error); 1048 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1049 if (error) 1050 return (error); 1051 INP_INFO_RLOCK(&tcbinfo); 1052 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1053 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1054 if (inp == NULL) { 1055 error = ENOENT; 1056 goto outunlocked; 1057 } 1058 INP_LOCK(inp); 1059 if (inp->inp_socket == NULL) { 1060 error = ENOENT; 1061 goto out; 1062 } 1063 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1064 if (error) 1065 goto out; 1066 cru2x(inp->inp_socket->so_cred, &xuc); 1067out: 1068 INP_UNLOCK(inp); 1069outunlocked: 1070 INP_INFO_RUNLOCK(&tcbinfo); 1071 if (error == 0) 1072 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1073 return (error); 1074} 1075 1076SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1077 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1078 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1079 1080#ifdef INET6 1081static int 1082tcp6_getcred(SYSCTL_HANDLER_ARGS) 1083{ 1084 struct xucred xuc; 1085 struct sockaddr_in6 addrs[2]; 1086 struct inpcb *inp; 1087 int error, mapped = 0; 1088 1089 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1090 if (error) 1091 return (error); 1092 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1093 if (error) 1094 return (error); 1095 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1096 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1097 return (error); 1098 } 1099 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1100 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1101 mapped = 1; 1102 else 1103 return (EINVAL); 1104 } 1105 1106 INP_INFO_RLOCK(&tcbinfo); 1107 if (mapped == 1) 1108 inp = in_pcblookup_hash(&tcbinfo, 1109 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1110 addrs[1].sin6_port, 1111 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1112 addrs[0].sin6_port, 1113 0, NULL); 1114 else 1115 inp = in6_pcblookup_hash(&tcbinfo, 1116 &addrs[1].sin6_addr, addrs[1].sin6_port, 1117 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1118 if (inp == NULL) { 1119 error = ENOENT; 1120 goto outunlocked; 1121 } 1122 INP_LOCK(inp); 1123 if (inp->inp_socket == NULL) { 1124 error = ENOENT; 1125 goto out; 1126 } 1127 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1128 if (error) 1129 goto out; 1130 cru2x(inp->inp_socket->so_cred, &xuc); 1131out: 1132 INP_UNLOCK(inp); 1133outunlocked: 1134 INP_INFO_RUNLOCK(&tcbinfo); 1135 if (error == 0) 1136 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1137 return (error); 1138} 1139 1140SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1141 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1142 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1143#endif 1144 1145 1146void 1147tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1148{ 1149 struct ip *ip = vip; 1150 struct tcphdr *th; 1151 struct in_addr faddr; 1152 struct inpcb *inp; 1153 struct tcpcb *tp; 1154 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1155 struct icmp *icp; 1156 struct in_conninfo inc; 1157 tcp_seq icmp_tcp_seq; 1158 int mtu; 1159 1160 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1161 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1162 return; 1163 1164 if (cmd == PRC_MSGSIZE) 1165 notify = tcp_mtudisc; 1166 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1167 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1168 notify = tcp_drop_syn_sent; 1169 /* 1170 * Redirects don't need to be handled up here. 1171 */ 1172 else if (PRC_IS_REDIRECT(cmd)) 1173 return; 1174 /* 1175 * Source quench is depreciated. 1176 */ 1177 else if (cmd == PRC_QUENCH) 1178 return; 1179 /* 1180 * Hostdead is ugly because it goes linearly through all PCBs. 1181 * XXX: We never get this from ICMP, otherwise it makes an 1182 * excellent DoS attack on machines with many connections. 1183 */ 1184 else if (cmd == PRC_HOSTDEAD) 1185 ip = NULL; 1186 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1187 return; 1188 if (ip != NULL) { 1189 icp = (struct icmp *)((caddr_t)ip 1190 - offsetof(struct icmp, icmp_ip)); 1191 th = (struct tcphdr *)((caddr_t)ip 1192 + (ip->ip_hl << 2)); 1193 INP_INFO_WLOCK(&tcbinfo); 1194 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1195 ip->ip_src, th->th_sport, 0, NULL); 1196 if (inp != NULL) { 1197 INP_LOCK(inp); 1198 if (!(inp->inp_vflag & INP_TIMEWAIT) && 1199 !(inp->inp_vflag & INP_DROPPED) && 1200 !(inp->inp_socket == NULL)) { 1201 icmp_tcp_seq = htonl(th->th_seq); 1202 tp = intotcpcb(inp); 1203 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1204 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1205 if (cmd == PRC_MSGSIZE) { 1206 /* 1207 * MTU discovery: 1208 * If we got a needfrag set the MTU 1209 * in the route to the suggested new 1210 * value (if given) and then notify. 1211 */ 1212 bzero(&inc, sizeof(inc)); 1213 inc.inc_flags = 0; /* IPv4 */ 1214 inc.inc_faddr = faddr; 1215 1216 mtu = ntohs(icp->icmp_nextmtu); 1217 /* 1218 * If no alternative MTU was 1219 * proposed, try the next smaller 1220 * one. ip->ip_len has already 1221 * been swapped in icmp_input(). 1222 */ 1223 if (!mtu) 1224 mtu = ip_next_mtu(ip->ip_len, 1225 1); 1226 if (mtu < max(296, (tcp_minmss) 1227 + sizeof(struct tcpiphdr))) 1228 mtu = 0; 1229 if (!mtu) 1230 mtu = tcp_mssdflt 1231 + sizeof(struct tcpiphdr); 1232 /* 1233 * Only cache the the MTU if it 1234 * is smaller than the interface 1235 * or route MTU. tcp_mtudisc() 1236 * will do right thing by itself. 1237 */ 1238 if (mtu <= tcp_maxmtu(&inc)) 1239 tcp_hc_updatemtu(&inc, mtu); 1240 } 1241 1242 inp = (*notify)(inp, inetctlerrmap[cmd]); 1243 } 1244 } 1245 if (inp != NULL) 1246 INP_UNLOCK(inp); 1247 } else { 1248 inc.inc_fport = th->th_dport; 1249 inc.inc_lport = th->th_sport; 1250 inc.inc_faddr = faddr; 1251 inc.inc_laddr = ip->ip_src; 1252#ifdef INET6 1253 inc.inc_isipv6 = 0; 1254#endif 1255 syncache_unreach(&inc, th); 1256 } 1257 INP_INFO_WUNLOCK(&tcbinfo); 1258 } else 1259 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1260} 1261 1262#ifdef INET6 1263void 1264tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1265{ 1266 struct tcphdr th; 1267 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1268 struct ip6_hdr *ip6; 1269 struct mbuf *m; 1270 struct ip6ctlparam *ip6cp = NULL; 1271 const struct sockaddr_in6 *sa6_src = NULL; 1272 int off; 1273 struct tcp_portonly { 1274 u_int16_t th_sport; 1275 u_int16_t th_dport; 1276 } *thp; 1277 1278 if (sa->sa_family != AF_INET6 || 1279 sa->sa_len != sizeof(struct sockaddr_in6)) 1280 return; 1281 1282 if (cmd == PRC_MSGSIZE) 1283 notify = tcp_mtudisc; 1284 else if (!PRC_IS_REDIRECT(cmd) && 1285 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1286 return; 1287 /* Source quench is depreciated. */ 1288 else if (cmd == PRC_QUENCH) 1289 return; 1290 1291 /* if the parameter is from icmp6, decode it. */ 1292 if (d != NULL) { 1293 ip6cp = (struct ip6ctlparam *)d; 1294 m = ip6cp->ip6c_m; 1295 ip6 = ip6cp->ip6c_ip6; 1296 off = ip6cp->ip6c_off; 1297 sa6_src = ip6cp->ip6c_src; 1298 } else { 1299 m = NULL; 1300 ip6 = NULL; 1301 off = 0; /* fool gcc */ 1302 sa6_src = &sa6_any; 1303 } 1304 1305 if (ip6 != NULL) { 1306 struct in_conninfo inc; 1307 /* 1308 * XXX: We assume that when IPV6 is non NULL, 1309 * M and OFF are valid. 1310 */ 1311 1312 /* check if we can safely examine src and dst ports */ 1313 if (m->m_pkthdr.len < off + sizeof(*thp)) 1314 return; 1315 1316 bzero(&th, sizeof(th)); 1317 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1318 1319 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1320 (struct sockaddr *)ip6cp->ip6c_src, 1321 th.th_sport, cmd, NULL, notify); 1322 1323 inc.inc_fport = th.th_dport; 1324 inc.inc_lport = th.th_sport; 1325 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1326 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1327 inc.inc_isipv6 = 1; 1328 INP_INFO_WLOCK(&tcbinfo); 1329 syncache_unreach(&inc, &th); 1330 INP_INFO_WUNLOCK(&tcbinfo); 1331 } else 1332 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1333 0, cmd, NULL, notify); 1334} 1335#endif /* INET6 */ 1336 1337 1338/* 1339 * Following is where TCP initial sequence number generation occurs. 1340 * 1341 * There are two places where we must use initial sequence numbers: 1342 * 1. In SYN-ACK packets. 1343 * 2. In SYN packets. 1344 * 1345 * All ISNs for SYN-ACK packets are generated by the syncache. See 1346 * tcp_syncache.c for details. 1347 * 1348 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1349 * depends on this property. In addition, these ISNs should be 1350 * unguessable so as to prevent connection hijacking. To satisfy 1351 * the requirements of this situation, the algorithm outlined in 1352 * RFC 1948 is used, with only small modifications. 1353 * 1354 * Implementation details: 1355 * 1356 * Time is based off the system timer, and is corrected so that it 1357 * increases by one megabyte per second. This allows for proper 1358 * recycling on high speed LANs while still leaving over an hour 1359 * before rollover. 1360 * 1361 * As reading the *exact* system time is too expensive to be done 1362 * whenever setting up a TCP connection, we increment the time 1363 * offset in two ways. First, a small random positive increment 1364 * is added to isn_offset for each connection that is set up. 1365 * Second, the function tcp_isn_tick fires once per clock tick 1366 * and increments isn_offset as necessary so that sequence numbers 1367 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1368 * random positive increments serve only to ensure that the same 1369 * exact sequence number is never sent out twice (as could otherwise 1370 * happen when a port is recycled in less than the system tick 1371 * interval.) 1372 * 1373 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1374 * between seeding of isn_secret. This is normally set to zero, 1375 * as reseeding should not be necessary. 1376 * 1377 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1378 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1379 * general, this means holding an exclusive (write) lock. 1380 */ 1381 1382#define ISN_BYTES_PER_SECOND 1048576 1383#define ISN_STATIC_INCREMENT 4096 1384#define ISN_RANDOM_INCREMENT (4096 - 1) 1385 1386static u_char isn_secret[32]; 1387static int isn_last_reseed; 1388static u_int32_t isn_offset, isn_offset_old; 1389static MD5_CTX isn_ctx; 1390 1391tcp_seq 1392tcp_new_isn(struct tcpcb *tp) 1393{ 1394 u_int32_t md5_buffer[4]; 1395 tcp_seq new_isn; 1396 1397 INP_LOCK_ASSERT(tp->t_inpcb); 1398 1399 mtx_lock(&isn_mtx); 1400 /* Seed if this is the first use, reseed if requested. */ 1401 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1402 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1403 < (u_int)ticks))) { 1404 read_random(&isn_secret, sizeof(isn_secret)); 1405 isn_last_reseed = ticks; 1406 } 1407 1408 /* Compute the md5 hash and return the ISN. */ 1409 MD5Init(&isn_ctx); 1410 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1411 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1412#ifdef INET6 1413 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1414 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1415 sizeof(struct in6_addr)); 1416 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1417 sizeof(struct in6_addr)); 1418 } else 1419#endif 1420 { 1421 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1422 sizeof(struct in_addr)); 1423 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1424 sizeof(struct in_addr)); 1425 } 1426 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1427 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1428 new_isn = (tcp_seq) md5_buffer[0]; 1429 isn_offset += ISN_STATIC_INCREMENT + 1430 (arc4random() & ISN_RANDOM_INCREMENT); 1431 new_isn += isn_offset; 1432 mtx_unlock(&isn_mtx); 1433 return (new_isn); 1434} 1435 1436/* 1437 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1438 * to keep time flowing at a relatively constant rate. If the random 1439 * increments have already pushed us past the projected offset, do nothing. 1440 */ 1441static void 1442tcp_isn_tick(void *xtp) 1443{ 1444 u_int32_t projected_offset; 1445 1446 mtx_lock(&isn_mtx); 1447 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1448 1449 if (projected_offset > isn_offset) 1450 isn_offset = projected_offset; 1451 1452 isn_offset_old = isn_offset; 1453 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1454 mtx_unlock(&isn_mtx); 1455} 1456 1457/* 1458 * When a specific ICMP unreachable message is received and the 1459 * connection state is SYN-SENT, drop the connection. This behavior 1460 * is controlled by the icmp_may_rst sysctl. 1461 */ 1462struct inpcb * 1463tcp_drop_syn_sent(struct inpcb *inp, int errno) 1464{ 1465 struct tcpcb *tp; 1466 1467 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1468 INP_LOCK_ASSERT(inp); 1469 1470 if ((inp->inp_vflag & INP_TIMEWAIT) || 1471 (inp->inp_vflag & INP_DROPPED)) 1472 return (inp); 1473 1474 tp = intotcpcb(inp); 1475 if (tp->t_state != TCPS_SYN_SENT) 1476 return (inp); 1477 1478 tp = tcp_drop(tp, errno); 1479 if (tp != NULL) 1480 return (inp); 1481 else 1482 return (NULL); 1483} 1484 1485/* 1486 * When `need fragmentation' ICMP is received, update our idea of the MSS 1487 * based on the new value in the route. Also nudge TCP to send something, 1488 * since we know the packet we just sent was dropped. 1489 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1490 */ 1491struct inpcb * 1492tcp_mtudisc(struct inpcb *inp, int errno) 1493{ 1494 struct tcpcb *tp; 1495 struct socket *so = inp->inp_socket; 1496 u_int maxmtu; 1497 u_int romtu; 1498 int mss; 1499#ifdef INET6 1500 int isipv6; 1501#endif /* INET6 */ 1502 1503 INP_LOCK_ASSERT(inp); 1504 if ((inp->inp_vflag & INP_TIMEWAIT) || 1505 (inp->inp_vflag & INP_DROPPED)) 1506 return (inp); 1507 1508 tp = intotcpcb(inp); 1509 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1510 1511#ifdef INET6 1512 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1513#endif 1514 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1515 romtu = 1516#ifdef INET6 1517 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1518#endif /* INET6 */ 1519 tcp_maxmtu(&inp->inp_inc); 1520 if (!maxmtu) 1521 maxmtu = romtu; 1522 else 1523 maxmtu = min(maxmtu, romtu); 1524 if (!maxmtu) { 1525 tp->t_maxopd = tp->t_maxseg = 1526#ifdef INET6 1527 isipv6 ? tcp_v6mssdflt : 1528#endif /* INET6 */ 1529 tcp_mssdflt; 1530 return (inp); 1531 } 1532 mss = maxmtu - 1533#ifdef INET6 1534 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1535#endif /* INET6 */ 1536 sizeof(struct tcpiphdr) 1537#ifdef INET6 1538 ) 1539#endif /* INET6 */ 1540 ; 1541 1542 /* 1543 * XXX - The above conditional probably violates the TCP 1544 * spec. The problem is that, since we don't know the 1545 * other end's MSS, we are supposed to use a conservative 1546 * default. But, if we do that, then MTU discovery will 1547 * never actually take place, because the conservative 1548 * default is much less than the MTUs typically seen 1549 * on the Internet today. For the moment, we'll sweep 1550 * this under the carpet. 1551 * 1552 * The conservative default might not actually be a problem 1553 * if the only case this occurs is when sending an initial 1554 * SYN with options and data to a host we've never talked 1555 * to before. Then, they will reply with an MSS value which 1556 * will get recorded and the new parameters should get 1557 * recomputed. For Further Study. 1558 */ 1559 if (tp->t_maxopd <= mss) 1560 return (inp); 1561 tp->t_maxopd = mss; 1562 1563 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1564 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1565 mss -= TCPOLEN_TSTAMP_APPA; 1566#if (MCLBYTES & (MCLBYTES - 1)) == 0 1567 if (mss > MCLBYTES) 1568 mss &= ~(MCLBYTES-1); 1569#else 1570 if (mss > MCLBYTES) 1571 mss = mss / MCLBYTES * MCLBYTES; 1572#endif 1573 if (so->so_snd.sb_hiwat < mss) 1574 mss = so->so_snd.sb_hiwat; 1575 1576 tp->t_maxseg = mss; 1577 1578 tcpstat.tcps_mturesent++; 1579 tp->t_rtttime = 0; 1580 tp->snd_nxt = tp->snd_una; 1581 tcp_output(tp); 1582 return (inp); 1583} 1584 1585/* 1586 * Look-up the routing entry to the peer of this inpcb. If no route 1587 * is found and it cannot be allocated, then return NULL. This routine 1588 * is called by TCP routines that access the rmx structure and by tcp_mss 1589 * to get the interface MTU. 1590 */ 1591u_long 1592tcp_maxmtu(struct in_conninfo *inc) 1593{ 1594 struct route sro; 1595 struct sockaddr_in *dst; 1596 struct ifnet *ifp; 1597 u_long maxmtu = 0; 1598 1599 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1600 1601 bzero(&sro, sizeof(sro)); 1602 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1603 dst = (struct sockaddr_in *)&sro.ro_dst; 1604 dst->sin_family = AF_INET; 1605 dst->sin_len = sizeof(*dst); 1606 dst->sin_addr = inc->inc_faddr; 1607 rtalloc_ign(&sro, RTF_CLONING); 1608 } 1609 if (sro.ro_rt != NULL) { 1610 ifp = sro.ro_rt->rt_ifp; 1611 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1612 maxmtu = ifp->if_mtu; 1613 else 1614 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1615 RTFREE(sro.ro_rt); 1616 } 1617 return (maxmtu); 1618} 1619 1620#ifdef INET6 1621u_long 1622tcp_maxmtu6(struct in_conninfo *inc) 1623{ 1624 struct route_in6 sro6; 1625 struct ifnet *ifp; 1626 u_long maxmtu = 0; 1627 1628 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1629 1630 bzero(&sro6, sizeof(sro6)); 1631 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1632 sro6.ro_dst.sin6_family = AF_INET6; 1633 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1634 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1635 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1636 } 1637 if (sro6.ro_rt != NULL) { 1638 ifp = sro6.ro_rt->rt_ifp; 1639 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1640 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1641 else 1642 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1643 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1644 RTFREE(sro6.ro_rt); 1645 } 1646 1647 return (maxmtu); 1648} 1649#endif /* INET6 */ 1650 1651#ifdef IPSEC 1652/* compute ESP/AH header size for TCP, including outer IP header. */ 1653size_t 1654ipsec_hdrsiz_tcp(struct tcpcb *tp) 1655{ 1656 struct inpcb *inp; 1657 struct mbuf *m; 1658 size_t hdrsiz; 1659 struct ip *ip; 1660#ifdef INET6 1661 struct ip6_hdr *ip6; 1662#endif 1663 struct tcphdr *th; 1664 1665 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1666 return (0); 1667 MGETHDR(m, M_DONTWAIT, MT_DATA); 1668 if (!m) 1669 return (0); 1670 1671#ifdef INET6 1672 if ((inp->inp_vflag & INP_IPV6) != 0) { 1673 ip6 = mtod(m, struct ip6_hdr *); 1674 th = (struct tcphdr *)(ip6 + 1); 1675 m->m_pkthdr.len = m->m_len = 1676 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1677 tcpip_fillheaders(inp, ip6, th); 1678 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1679 } else 1680#endif /* INET6 */ 1681 { 1682 ip = mtod(m, struct ip *); 1683 th = (struct tcphdr *)(ip + 1); 1684 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1685 tcpip_fillheaders(inp, ip, th); 1686 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1687 } 1688 1689 m_free(m); 1690 return (hdrsiz); 1691} 1692#endif /*IPSEC*/ 1693 1694/* 1695 * Move a TCP connection into TIME_WAIT state. 1696 * tcbinfo is locked. 1697 * inp is locked, and is unlocked before returning. 1698 */ 1699void 1700tcp_twstart(struct tcpcb *tp) 1701{ 1702 struct tcptw *tw; 1703 struct inpcb *inp; 1704 int tw_time, acknow; 1705 struct socket *so; 1706 1707 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1708 INP_LOCK_ASSERT(tp->t_inpcb); 1709 1710 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1711 if (tw == NULL) { 1712 tw = tcp_timer_2msl_tw(1); 1713 if (tw == NULL) { 1714 tp = tcp_close(tp); 1715 if (tp != NULL) 1716 INP_UNLOCK(tp->t_inpcb); 1717 return; 1718 } 1719 } 1720 inp = tp->t_inpcb; 1721 tw->tw_inpcb = inp; 1722 1723 /* 1724 * Recover last window size sent. 1725 */ 1726 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1727 1728 /* 1729 * Set t_recent if timestamps are used on the connection. 1730 */ 1731 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1732 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1733 tw->t_recent = tp->ts_recent; 1734 else 1735 tw->t_recent = 0; 1736 1737 tw->snd_nxt = tp->snd_nxt; 1738 tw->rcv_nxt = tp->rcv_nxt; 1739 tw->iss = tp->iss; 1740 tw->irs = tp->irs; 1741 tw->t_starttime = tp->t_starttime; 1742 tw->tw_time = 0; 1743 1744/* XXX 1745 * If this code will 1746 * be used for fin-wait-2 state also, then we may need 1747 * a ts_recent from the last segment. 1748 */ 1749 tw_time = 2 * tcp_msl; 1750 acknow = tp->t_flags & TF_ACKNOW; 1751 1752 /* 1753 * First, discard tcpcb state, which includes stopping its timers and 1754 * freeing it. tcp_discardcb() used to also release the inpcb, but 1755 * that work is now done in the caller. 1756 */ 1757 tcp_discardcb(tp); 1758 so = inp->inp_socket; 1759 SOCK_LOCK(so); 1760 tw->tw_cred = crhold(so->so_cred); 1761 tw->tw_so_options = so->so_options; 1762 SOCK_UNLOCK(so); 1763 if (acknow) 1764 tcp_twrespond(tw, TH_ACK); 1765 inp->inp_ppcb = tw; 1766 inp->inp_vflag |= INP_TIMEWAIT; 1767 tcp_timer_2msl_reset(tw, tw_time); 1768 1769 /* 1770 * If the inpcb owns the sole reference to the socket, then we can 1771 * detach and free the socket as it is not needed in time wait. 1772 */ 1773 if (inp->inp_vflag & INP_SOCKREF) { 1774 KASSERT(so->so_state & SS_PROTOREF, 1775 ("tcp_twstart: !SS_PROTOREF")); 1776 inp->inp_vflag &= ~INP_SOCKREF; 1777#ifdef INET6 1778 if (inp->inp_vflag & INP_IPV6PROTO) 1779 in6_pcbdetach(inp); 1780 else 1781#endif 1782 in_pcbdetach(inp); 1783 INP_UNLOCK(inp); 1784 ACCEPT_LOCK(); 1785 SOCK_LOCK(so); 1786 so->so_state &= ~SS_PROTOREF; 1787 sofree(so); 1788 } else 1789 INP_UNLOCK(inp); 1790} 1791 1792/* 1793 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1794 * the actual rate is slightly higher due to the addition of 1795 * random positive increments. 1796 * 1797 * Most other new OSes use semi-randomized ISN values, so we 1798 * do not need to worry about them. 1799 */ 1800#define MS_ISN_BYTES_PER_SECOND 250000 1801 1802/* 1803 * Determine if the ISN we will generate has advanced beyond the last 1804 * sequence number used by the previous connection. If so, indicate 1805 * that it is safe to recycle this tw socket by returning 1. 1806 * 1807 * XXXRW: This function should assert the inpcb lock as it does multiple 1808 * non-atomic reads from the tcptw, but is currently called without it from 1809 * in_pcb.c:in_pcblookup_local(). 1810 */ 1811int 1812tcp_twrecycleable(struct tcptw *tw) 1813{ 1814 tcp_seq new_iss = tw->iss; 1815 tcp_seq new_irs = tw->irs; 1816 1817 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1818 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1819 1820 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1821 return (1); 1822 else 1823 return (0); 1824} 1825 1826void 1827tcp_twclose(struct tcptw *tw, int reuse) 1828{ 1829 struct socket *so; 1830 struct inpcb *inp; 1831 1832 /* 1833 * At this point, we are in one of two situations: 1834 * 1835 * (1) We have no socket, just an inpcb<->twtcp pair. Release it all 1836 * after validating. 1837 * 1838 * (2) We have a socket, which we may or may now own the reference 1839 * for. If we own the reference, release all the state after 1840 * validating. If not, leave it for the socket close to clean up. 1841 */ 1842 inp = tw->tw_inpcb; 1843 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1844 KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1845 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1846 INP_LOCK_ASSERT(inp); 1847 1848 tw->tw_inpcb = NULL; 1849 tcp_timer_2msl_stop(tw); 1850 inp->inp_ppcb = NULL; 1851 inp->inp_vflag |= INP_DROPPED; 1852 1853 so = inp->inp_socket; 1854 if (so != NULL) { 1855 if (inp->inp_vflag & INP_SOCKREF) { 1856 /* 1857 * If a socket is present, and we own the only 1858 * reference, we need to tear down the socket and the 1859 * inpcb. 1860 */ 1861 inp->inp_vflag &= ~INP_SOCKREF; 1862#ifdef INET6 1863 if (inp->inp_vflag & INP_IPV6PROTO) { 1864 in6_pcbdetach(inp); 1865 in6_pcbfree(inp); 1866 } else { 1867 in_pcbdetach(inp); 1868 in_pcbfree(inp); 1869 } 1870#endif 1871 ACCEPT_LOCK(); 1872 SOCK_LOCK(so); 1873 KASSERT(so->so_state & SS_PROTOREF, 1874 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1875 so->so_state &= ~SS_PROTOREF; 1876 sofree(so); 1877 } else { 1878 /* 1879 * If we don't own the only reference, the socket and 1880 * inpcb need to be left around to be handled by 1881 * tcp_usr_detach() later. 1882 */ 1883 INP_UNLOCK(inp); 1884 } 1885 } else { 1886#ifdef INET6 1887 if (inp->inp_vflag & INP_IPV6PROTO) 1888 in6_pcbfree(inp); 1889 else 1890#endif 1891 in_pcbfree(inp); 1892 } 1893 tcpstat.tcps_closed++; 1894 crfree(tw->tw_cred); 1895 tw->tw_cred = NULL; 1896 if (reuse) 1897 return; 1898 uma_zfree(tcptw_zone, tw); 1899} 1900 1901int 1902tcp_twrespond(struct tcptw *tw, int flags) 1903{ 1904 struct inpcb *inp = tw->tw_inpcb; 1905 struct tcphdr *th; 1906 struct mbuf *m; 1907 struct ip *ip = NULL; 1908 u_int8_t *optp; 1909 u_int hdrlen, optlen; 1910 int error; 1911#ifdef INET6 1912 struct ip6_hdr *ip6 = NULL; 1913 int isipv6 = inp->inp_inc.inc_isipv6; 1914#endif 1915 1916 INP_LOCK_ASSERT(inp); 1917 1918 m = m_gethdr(M_DONTWAIT, MT_DATA); 1919 if (m == NULL) 1920 return (ENOBUFS); 1921 m->m_data += max_linkhdr; 1922 1923#ifdef MAC 1924 mac_create_mbuf_from_inpcb(inp, m); 1925#endif 1926 1927#ifdef INET6 1928 if (isipv6) { 1929 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1930 ip6 = mtod(m, struct ip6_hdr *); 1931 th = (struct tcphdr *)(ip6 + 1); 1932 tcpip_fillheaders(inp, ip6, th); 1933 } else 1934#endif 1935 { 1936 hdrlen = sizeof(struct tcpiphdr); 1937 ip = mtod(m, struct ip *); 1938 th = (struct tcphdr *)(ip + 1); 1939 tcpip_fillheaders(inp, ip, th); 1940 } 1941 optp = (u_int8_t *)(th + 1); 1942 1943 /* 1944 * Send a timestamp and echo-reply if both our side and our peer 1945 * have sent timestamps in our SYN's and this is not a RST. 1946 */ 1947 if (tw->t_recent && flags == TH_ACK) { 1948 u_int32_t *lp = (u_int32_t *)optp; 1949 1950 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1951 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1952 *lp++ = htonl(ticks); 1953 *lp = htonl(tw->t_recent); 1954 optp += TCPOLEN_TSTAMP_APPA; 1955 } 1956 1957 optlen = optp - (u_int8_t *)(th + 1); 1958 1959 m->m_len = hdrlen + optlen; 1960 m->m_pkthdr.len = m->m_len; 1961 1962 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1963 1964 th->th_seq = htonl(tw->snd_nxt); 1965 th->th_ack = htonl(tw->rcv_nxt); 1966 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1967 th->th_flags = flags; 1968 th->th_win = htons(tw->last_win); 1969 1970#ifdef INET6 1971 if (isipv6) { 1972 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1973 sizeof(struct tcphdr) + optlen); 1974 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1975 error = ip6_output(m, inp->in6p_outputopts, NULL, 1976 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1977 } else 1978#endif 1979 { 1980 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1981 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1982 m->m_pkthdr.csum_flags = CSUM_TCP; 1983 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1984 ip->ip_len = m->m_pkthdr.len; 1985 if (path_mtu_discovery) 1986 ip->ip_off |= IP_DF; 1987 error = ip_output(m, inp->inp_options, NULL, 1988 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1989 NULL, inp); 1990 } 1991 if (flags & TH_ACK) 1992 tcpstat.tcps_sndacks++; 1993 else 1994 tcpstat.tcps_sndctrl++; 1995 tcpstat.tcps_sndtotal++; 1996 return (error); 1997} 1998 1999/* 2000 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 2001 * 2002 * This code attempts to calculate the bandwidth-delay product as a 2003 * means of determining the optimal window size to maximize bandwidth, 2004 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 2005 * routers. This code also does a fairly good job keeping RTTs in check 2006 * across slow links like modems. We implement an algorithm which is very 2007 * similar (but not meant to be) TCP/Vegas. The code operates on the 2008 * transmitter side of a TCP connection and so only effects the transmit 2009 * side of the connection. 2010 * 2011 * BACKGROUND: TCP makes no provision for the management of buffer space 2012 * at the end points or at the intermediate routers and switches. A TCP 2013 * stream, whether using NewReno or not, will eventually buffer as 2014 * many packets as it is able and the only reason this typically works is 2015 * due to the fairly small default buffers made available for a connection 2016 * (typicaly 16K or 32K). As machines use larger windows and/or window 2017 * scaling it is now fairly easy for even a single TCP connection to blow-out 2018 * all available buffer space not only on the local interface, but on 2019 * intermediate routers and switches as well. NewReno makes a misguided 2020 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2021 * then backing off, then steadily increasing the window again until another 2022 * failure occurs, ad-infinitum. This results in terrible oscillation that 2023 * is only made worse as network loads increase and the idea of intentionally 2024 * blowing out network buffers is, frankly, a terrible way to manage network 2025 * resources. 2026 * 2027 * It is far better to limit the transmit window prior to the failure 2028 * condition being achieved. There are two general ways to do this: First 2029 * you can 'scan' through different transmit window sizes and locate the 2030 * point where the RTT stops increasing, indicating that you have filled the 2031 * pipe, then scan backwards until you note that RTT stops decreasing, then 2032 * repeat ad-infinitum. This method works in principle but has severe 2033 * implementation issues due to RTT variances, timer granularity, and 2034 * instability in the algorithm which can lead to many false positives and 2035 * create oscillations as well as interact badly with other TCP streams 2036 * implementing the same algorithm. 2037 * 2038 * The second method is to limit the window to the bandwidth delay product 2039 * of the link. This is the method we implement. RTT variances and our 2040 * own manipulation of the congestion window, bwnd, can potentially 2041 * destabilize the algorithm. For this reason we have to stabilize the 2042 * elements used to calculate the window. We do this by using the minimum 2043 * observed RTT, the long term average of the observed bandwidth, and 2044 * by adding two segments worth of slop. It isn't perfect but it is able 2045 * to react to changing conditions and gives us a very stable basis on 2046 * which to extend the algorithm. 2047 */ 2048void 2049tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2050{ 2051 u_long bw; 2052 u_long bwnd; 2053 int save_ticks; 2054 2055 INP_LOCK_ASSERT(tp->t_inpcb); 2056 2057 /* 2058 * If inflight_enable is disabled in the middle of a tcp connection, 2059 * make sure snd_bwnd is effectively disabled. 2060 */ 2061 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2062 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2063 tp->snd_bandwidth = 0; 2064 return; 2065 } 2066 2067 /* 2068 * Figure out the bandwidth. Due to the tick granularity this 2069 * is a very rough number and it MUST be averaged over a fairly 2070 * long period of time. XXX we need to take into account a link 2071 * that is not using all available bandwidth, but for now our 2072 * slop will ramp us up if this case occurs and the bandwidth later 2073 * increases. 2074 * 2075 * Note: if ticks rollover 'bw' may wind up negative. We must 2076 * effectively reset t_bw_rtttime for this case. 2077 */ 2078 save_ticks = ticks; 2079 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2080 return; 2081 2082 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2083 (save_ticks - tp->t_bw_rtttime); 2084 tp->t_bw_rtttime = save_ticks; 2085 tp->t_bw_rtseq = ack_seq; 2086 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2087 return; 2088 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2089 2090 tp->snd_bandwidth = bw; 2091 2092 /* 2093 * Calculate the semi-static bandwidth delay product, plus two maximal 2094 * segments. The additional slop puts us squarely in the sweet 2095 * spot and also handles the bandwidth run-up case and stabilization. 2096 * Without the slop we could be locking ourselves into a lower 2097 * bandwidth. 2098 * 2099 * Situations Handled: 2100 * (1) Prevents over-queueing of packets on LANs, especially on 2101 * high speed LANs, allowing larger TCP buffers to be 2102 * specified, and also does a good job preventing 2103 * over-queueing of packets over choke points like modems 2104 * (at least for the transmit side). 2105 * 2106 * (2) Is able to handle changing network loads (bandwidth 2107 * drops so bwnd drops, bandwidth increases so bwnd 2108 * increases). 2109 * 2110 * (3) Theoretically should stabilize in the face of multiple 2111 * connections implementing the same algorithm (this may need 2112 * a little work). 2113 * 2114 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2115 * be adjusted with a sysctl but typically only needs to be 2116 * on very slow connections. A value no smaller then 5 2117 * should be used, but only reduce this default if you have 2118 * no other choice. 2119 */ 2120#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2121 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2122#undef USERTT 2123 2124 if (tcp_inflight_debug > 0) { 2125 static int ltime; 2126 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2127 ltime = ticks; 2128 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2129 tp, 2130 bw, 2131 tp->t_rttbest, 2132 tp->t_srtt, 2133 bwnd 2134 ); 2135 } 2136 } 2137 if ((long)bwnd < tcp_inflight_min) 2138 bwnd = tcp_inflight_min; 2139 if (bwnd > tcp_inflight_max) 2140 bwnd = tcp_inflight_max; 2141 if ((long)bwnd < tp->t_maxseg * 2) 2142 bwnd = tp->t_maxseg * 2; 2143 tp->snd_bwnd = bwnd; 2144} 2145 2146#ifdef TCP_SIGNATURE 2147/* 2148 * Callback function invoked by m_apply() to digest TCP segment data 2149 * contained within an mbuf chain. 2150 */ 2151static int 2152tcp_signature_apply(void *fstate, void *data, u_int len) 2153{ 2154 2155 MD5Update(fstate, (u_char *)data, len); 2156 return (0); 2157} 2158 2159/* 2160 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2161 * 2162 * Parameters: 2163 * m pointer to head of mbuf chain 2164 * off0 offset to TCP header within the mbuf chain 2165 * len length of TCP segment data, excluding options 2166 * optlen length of TCP segment options 2167 * buf pointer to storage for computed MD5 digest 2168 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2169 * 2170 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2171 * When called from tcp_input(), we can be sure that th_sum has been 2172 * zeroed out and verified already. 2173 * 2174 * This function is for IPv4 use only. Calling this function with an 2175 * IPv6 packet in the mbuf chain will yield undefined results. 2176 * 2177 * Return 0 if successful, otherwise return -1. 2178 * 2179 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2180 * search with the destination IP address, and a 'magic SPI' to be 2181 * determined by the application. This is hardcoded elsewhere to 1179 2182 * right now. Another branch of this code exists which uses the SPD to 2183 * specify per-application flows but it is unstable. 2184 */ 2185int 2186tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2187 u_char *buf, u_int direction) 2188{ 2189 union sockaddr_union dst; 2190 struct ippseudo ippseudo; 2191 MD5_CTX ctx; 2192 int doff; 2193 struct ip *ip; 2194 struct ipovly *ipovly; 2195 struct secasvar *sav; 2196 struct tcphdr *th; 2197 u_short savecsum; 2198 2199 KASSERT(m != NULL, ("NULL mbuf chain")); 2200 KASSERT(buf != NULL, ("NULL signature pointer")); 2201 2202 /* Extract the destination from the IP header in the mbuf. */ 2203 ip = mtod(m, struct ip *); 2204 bzero(&dst, sizeof(union sockaddr_union)); 2205 dst.sa.sa_len = sizeof(struct sockaddr_in); 2206 dst.sa.sa_family = AF_INET; 2207 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2208 ip->ip_src : ip->ip_dst; 2209 2210 /* Look up an SADB entry which matches the address of the peer. */ 2211 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2212 if (sav == NULL) { 2213 printf("%s: SADB lookup failed for %s\n", __func__, 2214 inet_ntoa(dst.sin.sin_addr)); 2215 return (EINVAL); 2216 } 2217 2218 MD5Init(&ctx); 2219 ipovly = (struct ipovly *)ip; 2220 th = (struct tcphdr *)((u_char *)ip + off0); 2221 doff = off0 + sizeof(struct tcphdr) + optlen; 2222 2223 /* 2224 * Step 1: Update MD5 hash with IP pseudo-header. 2225 * 2226 * XXX The ippseudo header MUST be digested in network byte order, 2227 * or else we'll fail the regression test. Assume all fields we've 2228 * been doing arithmetic on have been in host byte order. 2229 * XXX One cannot depend on ipovly->ih_len here. When called from 2230 * tcp_output(), the underlying ip_len member has not yet been set. 2231 */ 2232 ippseudo.ippseudo_src = ipovly->ih_src; 2233 ippseudo.ippseudo_dst = ipovly->ih_dst; 2234 ippseudo.ippseudo_pad = 0; 2235 ippseudo.ippseudo_p = IPPROTO_TCP; 2236 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2237 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2238 2239 /* 2240 * Step 2: Update MD5 hash with TCP header, excluding options. 2241 * The TCP checksum must be set to zero. 2242 */ 2243 savecsum = th->th_sum; 2244 th->th_sum = 0; 2245 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2246 th->th_sum = savecsum; 2247 2248 /* 2249 * Step 3: Update MD5 hash with TCP segment data. 2250 * Use m_apply() to avoid an early m_pullup(). 2251 */ 2252 if (len > 0) 2253 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2254 2255 /* 2256 * Step 4: Update MD5 hash with shared secret. 2257 */ 2258 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2259 MD5Final(buf, &ctx); 2260 2261 key_sa_recordxfer(sav, m); 2262 KEY_FREESAV(&sav); 2263 return (0); 2264} 2265#endif /* TCP_SIGNATURE */ 2266 2267static int 2268sysctl_drop(SYSCTL_HANDLER_ARGS) 2269{ 2270 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2271 struct sockaddr_storage addrs[2]; 2272 struct inpcb *inp; 2273 struct tcpcb *tp; 2274 struct tcptw *tw; 2275 struct sockaddr_in *fin, *lin; 2276#ifdef INET6 2277 struct sockaddr_in6 *fin6, *lin6; 2278 struct in6_addr f6, l6; 2279#endif 2280 int error; 2281 2282 inp = NULL; 2283 fin = lin = NULL; 2284#ifdef INET6 2285 fin6 = lin6 = NULL; 2286#endif 2287 error = 0; 2288 2289 if (req->oldptr != NULL || req->oldlen != 0) 2290 return (EINVAL); 2291 if (req->newptr == NULL) 2292 return (EPERM); 2293 if (req->newlen < sizeof(addrs)) 2294 return (ENOMEM); 2295 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2296 if (error) 2297 return (error); 2298 2299 switch (addrs[0].ss_family) { 2300#ifdef INET6 2301 case AF_INET6: 2302 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2303 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2304 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2305 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2306 return (EINVAL); 2307 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2308 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2309 return (EINVAL); 2310 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2311 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2312 fin = (struct sockaddr_in *)&addrs[0]; 2313 lin = (struct sockaddr_in *)&addrs[1]; 2314 break; 2315 } 2316 error = sa6_embedscope(fin6, ip6_use_defzone); 2317 if (error) 2318 return (error); 2319 error = sa6_embedscope(lin6, ip6_use_defzone); 2320 if (error) 2321 return (error); 2322 break; 2323#endif 2324 case AF_INET: 2325 fin = (struct sockaddr_in *)&addrs[0]; 2326 lin = (struct sockaddr_in *)&addrs[1]; 2327 if (fin->sin_len != sizeof(struct sockaddr_in) || 2328 lin->sin_len != sizeof(struct sockaddr_in)) 2329 return (EINVAL); 2330 break; 2331 default: 2332 return (EINVAL); 2333 } 2334 INP_INFO_WLOCK(&tcbinfo); 2335 switch (addrs[0].ss_family) { 2336#ifdef INET6 2337 case AF_INET6: 2338 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2339 &l6, lin6->sin6_port, 0, NULL); 2340 break; 2341#endif 2342 case AF_INET: 2343 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2344 lin->sin_addr, lin->sin_port, 0, NULL); 2345 break; 2346 } 2347 if (inp != NULL) { 2348 INP_LOCK(inp); 2349 if (inp->inp_vflag & INP_TIMEWAIT) { 2350 /* 2351 * XXXRW: There currently exists a state where an 2352 * inpcb is present, but its timewait state has been 2353 * discarded. For now, don't allow dropping of this 2354 * type of inpcb. 2355 */ 2356 tw = intotw(inp); 2357 if (tw != NULL) 2358 tcp_twclose(tw, 0); 2359 } else if (!(inp->inp_vflag & INP_DROPPED) && 2360 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2361 tp = intotcpcb(inp); 2362 tcp_drop(tp, ECONNABORTED); 2363 } 2364 INP_UNLOCK(inp); 2365 } else 2366 error = ESRCH; 2367 INP_INFO_WUNLOCK(&tcbinfo); 2368 return (error); 2369} 2370 2371SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2372 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2373 0, sysctl_drop, "", "Drop TCP connection"); 2374