tcp_timewait.c revision 157430
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 157430 2006-04-03 12:52:13Z 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"); 179 180/* 181 * TCP bandwidth limiting sysctls. Note that the default lower bound of 182 * 1024 exists only for debugging. A good production default would be 183 * something like 6100. 184 */ 185SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 186 "TCP inflight data limiting"); 187 188static int tcp_inflight_enable = 1; 189SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 190 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 191 192static int tcp_inflight_debug = 0; 193SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 194 &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 195 196static int tcp_inflight_rttthresh; 197SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 198 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 199 "RTT threshold below which inflight will deactivate itself"); 200 201static int tcp_inflight_min = 6144; 202SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 203 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 204 205static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 206SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 207 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 208 209static int tcp_inflight_stab = 20; 210SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 211 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 212 213uma_zone_t sack_hole_zone; 214 215static struct inpcb *tcp_notify(struct inpcb *, int); 216static void tcp_isn_tick(void *); 217 218/* 219 * Target size of TCP PCB hash tables. Must be a power of two. 220 * 221 * Note that this can be overridden by the kernel environment 222 * variable net.inet.tcp.tcbhashsize 223 */ 224#ifndef TCBHASHSIZE 225#define TCBHASHSIZE 512 226#endif 227 228/* 229 * XXX 230 * Callouts should be moved into struct tcp directly. They are currently 231 * separate because the tcpcb structure is exported to userland for sysctl 232 * parsing purposes, which do not know about callouts. 233 */ 234struct tcpcb_mem { 235 struct tcpcb tcb; 236 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 237 struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 238}; 239 240static uma_zone_t tcpcb_zone; 241static uma_zone_t tcptw_zone; 242struct callout isn_callout; 243 244/* 245 * Tcp initialization 246 */ 247void 248tcp_init() 249{ 250 int hashsize = TCBHASHSIZE; 251 252 tcp_delacktime = TCPTV_DELACK; 253 tcp_keepinit = TCPTV_KEEP_INIT; 254 tcp_keepidle = TCPTV_KEEP_IDLE; 255 tcp_keepintvl = TCPTV_KEEPINTVL; 256 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 257 tcp_msl = TCPTV_MSL; 258 tcp_rexmit_min = TCPTV_MIN; 259 tcp_rexmit_slop = TCPTV_CPU_VAR; 260 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 261 262 INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 263 LIST_INIT(&tcb); 264 tcbinfo.listhead = &tcb; 265 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 266 if (!powerof2(hashsize)) { 267 printf("WARNING: TCB hash size not a power of 2\n"); 268 hashsize = 512; /* safe default */ 269 } 270 tcp_tcbhashsize = hashsize; 271 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 272 tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 273 &tcbinfo.porthashmask); 274 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 275 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 276 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 277#ifdef INET6 278#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 279#else /* INET6 */ 280#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 281#endif /* INET6 */ 282 if (max_protohdr < TCP_MINPROTOHDR) 283 max_protohdr = TCP_MINPROTOHDR; 284 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 285 panic("tcp_init"); 286#undef TCP_MINPROTOHDR 287 /* 288 * These have to be type stable for the benefit of the timers. 289 */ 290 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 291 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 292 uma_zone_set_max(tcpcb_zone, maxsockets); 293 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 294 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 295 uma_zone_set_max(tcptw_zone, maxsockets / 5); 296 tcp_timer_init(); 297 syncache_init(); 298 tcp_hc_init(); 299 tcp_reass_init(); 300 callout_init(&isn_callout, CALLOUT_MPSAFE); 301 tcp_isn_tick(NULL); 302 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 303 SHUTDOWN_PRI_DEFAULT); 304 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 305 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 306} 307 308void 309tcp_fini(xtp) 310 void *xtp; 311{ 312 callout_stop(&isn_callout); 313 314} 315 316/* 317 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 318 * tcp_template used to store this data in mbufs, but we now recopy it out 319 * of the tcpcb each time to conserve mbufs. 320 */ 321void 322tcpip_fillheaders(inp, ip_ptr, tcp_ptr) 323 struct inpcb *inp; 324 void *ip_ptr; 325 void *tcp_ptr; 326{ 327 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 328 329 INP_LOCK_ASSERT(inp); 330 331#ifdef INET6 332 if ((inp->inp_vflag & INP_IPV6) != 0) { 333 struct ip6_hdr *ip6; 334 335 ip6 = (struct ip6_hdr *)ip_ptr; 336 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 337 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 338 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 339 (IPV6_VERSION & IPV6_VERSION_MASK); 340 ip6->ip6_nxt = IPPROTO_TCP; 341 ip6->ip6_plen = sizeof(struct tcphdr); 342 ip6->ip6_src = inp->in6p_laddr; 343 ip6->ip6_dst = inp->in6p_faddr; 344 } else 345#endif 346 { 347 struct ip *ip; 348 349 ip = (struct ip *)ip_ptr; 350 ip->ip_v = IPVERSION; 351 ip->ip_hl = 5; 352 ip->ip_tos = inp->inp_ip_tos; 353 ip->ip_len = 0; 354 ip->ip_id = 0; 355 ip->ip_off = 0; 356 ip->ip_ttl = inp->inp_ip_ttl; 357 ip->ip_sum = 0; 358 ip->ip_p = IPPROTO_TCP; 359 ip->ip_src = inp->inp_laddr; 360 ip->ip_dst = inp->inp_faddr; 361 } 362 th->th_sport = inp->inp_lport; 363 th->th_dport = inp->inp_fport; 364 th->th_seq = 0; 365 th->th_ack = 0; 366 th->th_x2 = 0; 367 th->th_off = 5; 368 th->th_flags = 0; 369 th->th_win = 0; 370 th->th_urp = 0; 371 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 372} 373 374/* 375 * Create template to be used to send tcp packets on a connection. 376 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 377 * use for this function is in keepalives, which use tcp_respond. 378 */ 379struct tcptemp * 380tcpip_maketemplate(inp) 381 struct inpcb *inp; 382{ 383 struct mbuf *m; 384 struct tcptemp *n; 385 386 m = m_get(M_DONTWAIT, MT_DATA); 387 if (m == NULL) 388 return (0); 389 m->m_len = sizeof(struct tcptemp); 390 n = mtod(m, struct tcptemp *); 391 392 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 393 return (n); 394} 395 396/* 397 * Send a single message to the TCP at address specified by 398 * the given TCP/IP header. If m == NULL, then we make a copy 399 * of the tcpiphdr at ti and send directly to the addressed host. 400 * This is used to force keep alive messages out using the TCP 401 * template for a connection. If flags are given then we send 402 * a message back to the TCP which originated the * segment ti, 403 * and discard the mbuf containing it and any other attached mbufs. 404 * 405 * In any case the ack and sequence number of the transmitted 406 * segment are as specified by the parameters. 407 * 408 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 409 */ 410void 411tcp_respond(tp, ipgen, th, m, ack, seq, flags) 412 struct tcpcb *tp; 413 void *ipgen; 414 register struct tcphdr *th; 415 register struct mbuf *m; 416 tcp_seq ack, seq; 417 int flags; 418{ 419 register int tlen; 420 int win = 0; 421 struct ip *ip; 422 struct tcphdr *nth; 423#ifdef INET6 424 struct ip6_hdr *ip6; 425 int isipv6; 426#endif /* INET6 */ 427 int ipflags = 0; 428 struct inpcb *inp; 429 430 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 431 432#ifdef INET6 433 isipv6 = ((struct ip *)ipgen)->ip_v == 6; 434 ip6 = ipgen; 435#endif /* INET6 */ 436 ip = ipgen; 437 438 if (tp != NULL) { 439 inp = tp->t_inpcb; 440 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 441 INP_INFO_WLOCK_ASSERT(&tcbinfo); 442 INP_LOCK_ASSERT(inp); 443 } else 444 inp = NULL; 445 446 if (tp != NULL) { 447 if (!(flags & TH_RST)) { 448 win = sbspace(&inp->inp_socket->so_rcv); 449 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 450 win = (long)TCP_MAXWIN << tp->rcv_scale; 451 } 452 } 453 if (m == NULL) { 454 m = m_gethdr(M_DONTWAIT, MT_DATA); 455 if (m == NULL) 456 return; 457 tlen = 0; 458 m->m_data += max_linkhdr; 459#ifdef INET6 460 if (isipv6) { 461 bcopy((caddr_t)ip6, mtod(m, caddr_t), 462 sizeof(struct ip6_hdr)); 463 ip6 = mtod(m, struct ip6_hdr *); 464 nth = (struct tcphdr *)(ip6 + 1); 465 } else 466#endif /* INET6 */ 467 { 468 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 469 ip = mtod(m, struct ip *); 470 nth = (struct tcphdr *)(ip + 1); 471 } 472 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 473 flags = TH_ACK; 474 } else { 475 m_freem(m->m_next); 476 m->m_next = NULL; 477 m->m_data = (caddr_t)ipgen; 478 /* m_len is set later */ 479 tlen = 0; 480#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 481#ifdef INET6 482 if (isipv6) { 483 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 484 nth = (struct tcphdr *)(ip6 + 1); 485 } else 486#endif /* INET6 */ 487 { 488 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 489 nth = (struct tcphdr *)(ip + 1); 490 } 491 if (th != nth) { 492 /* 493 * this is usually a case when an extension header 494 * exists between the IPv6 header and the 495 * TCP header. 496 */ 497 nth->th_sport = th->th_sport; 498 nth->th_dport = th->th_dport; 499 } 500 xchg(nth->th_dport, nth->th_sport, n_short); 501#undef xchg 502 } 503#ifdef INET6 504 if (isipv6) { 505 ip6->ip6_flow = 0; 506 ip6->ip6_vfc = IPV6_VERSION; 507 ip6->ip6_nxt = IPPROTO_TCP; 508 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 509 tlen)); 510 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 511 } else 512#endif 513 { 514 tlen += sizeof (struct tcpiphdr); 515 ip->ip_len = tlen; 516 ip->ip_ttl = ip_defttl; 517 if (path_mtu_discovery) 518 ip->ip_off |= IP_DF; 519 } 520 m->m_len = tlen; 521 m->m_pkthdr.len = tlen; 522 m->m_pkthdr.rcvif = NULL; 523#ifdef MAC 524 if (inp != NULL) { 525 /* 526 * Packet is associated with a socket, so allow the 527 * label of the response to reflect the socket label. 528 */ 529 INP_LOCK_ASSERT(inp); 530 mac_create_mbuf_from_inpcb(inp, m); 531 } else { 532 /* 533 * Packet is not associated with a socket, so possibly 534 * update the label in place. 535 */ 536 mac_reflect_mbuf_tcp(m); 537 } 538#endif 539 nth->th_seq = htonl(seq); 540 nth->th_ack = htonl(ack); 541 nth->th_x2 = 0; 542 nth->th_off = sizeof (struct tcphdr) >> 2; 543 nth->th_flags = flags; 544 if (tp != NULL) 545 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 546 else 547 nth->th_win = htons((u_short)win); 548 nth->th_urp = 0; 549#ifdef INET6 550 if (isipv6) { 551 nth->th_sum = 0; 552 nth->th_sum = in6_cksum(m, IPPROTO_TCP, 553 sizeof(struct ip6_hdr), 554 tlen - sizeof(struct ip6_hdr)); 555 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 556 NULL, NULL); 557 } else 558#endif /* INET6 */ 559 { 560 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 561 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 562 m->m_pkthdr.csum_flags = CSUM_TCP; 563 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 564 } 565#ifdef TCPDEBUG 566 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 567 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 568#endif 569#ifdef INET6 570 if (isipv6) 571 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 572 else 573#endif /* INET6 */ 574 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 575} 576 577/* 578 * Create a new TCP control block, making an 579 * empty reassembly queue and hooking it to the argument 580 * protocol control block. The `inp' parameter must have 581 * come from the zone allocator set up in tcp_init(). 582 */ 583struct tcpcb * 584tcp_newtcpcb(inp) 585 struct inpcb *inp; 586{ 587 struct tcpcb_mem *tm; 588 struct tcpcb *tp; 589#ifdef INET6 590 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 591#endif /* INET6 */ 592 593 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 594 if (tm == NULL) 595 return (NULL); 596 tp = &tm->tcb; 597 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 598 tp->t_maxseg = tp->t_maxopd = 599#ifdef INET6 600 isipv6 ? tcp_v6mssdflt : 601#endif /* INET6 */ 602 tcp_mssdflt; 603 604 /* Set up our timeouts. */ 605 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 606 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 607 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 608 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 609 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 610 611 if (tcp_do_rfc1323) 612 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 613 tp->sack_enable = tcp_do_sack; 614 TAILQ_INIT(&tp->snd_holes); 615 tp->t_inpcb = inp; /* XXX */ 616 /* 617 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 618 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 619 * reasonable initial retransmit time. 620 */ 621 tp->t_srtt = TCPTV_SRTTBASE; 622 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 623 tp->t_rttmin = tcp_rexmit_min; 624 tp->t_rxtcur = TCPTV_RTOBASE; 625 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 626 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 627 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 628 tp->t_rcvtime = ticks; 629 tp->t_bw_rtttime = ticks; 630 /* 631 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 632 * because the socket may be bound to an IPv6 wildcard address, 633 * which may match an IPv4-mapped IPv6 address. 634 */ 635 inp->inp_ip_ttl = ip_defttl; 636 inp->inp_ppcb = (caddr_t)tp; 637 return (tp); /* XXX */ 638} 639 640/* 641 * Drop a TCP connection, reporting 642 * the specified error. If connection is synchronized, 643 * then send a RST to peer. 644 */ 645struct tcpcb * 646tcp_drop(tp, errno) 647 register struct tcpcb *tp; 648 int errno; 649{ 650 struct socket *so = tp->t_inpcb->inp_socket; 651 652 INP_INFO_WLOCK_ASSERT(&tcbinfo); 653 INP_LOCK_ASSERT(tp->t_inpcb); 654 655 if (TCPS_HAVERCVDSYN(tp->t_state)) { 656 tp->t_state = TCPS_CLOSED; 657 (void) tcp_output(tp); 658 tcpstat.tcps_drops++; 659 } else 660 tcpstat.tcps_conndrops++; 661 if (errno == ETIMEDOUT && tp->t_softerror) 662 errno = tp->t_softerror; 663 so->so_error = errno; 664 return (tcp_close(tp)); 665} 666 667void 668tcp_discardcb(tp) 669 struct tcpcb *tp; 670{ 671 struct tseg_qent *q; 672 struct inpcb *inp = tp->t_inpcb; 673 struct socket *so = inp->inp_socket; 674#ifdef INET6 675 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 676#endif /* INET6 */ 677 678 /* 679 * XXXRW: This is all very well and good, but actually, we might be 680 * discarding the tcpcb after the socket is gone, so we can't do 681 * this: 682 KASSERT(so != NULL, ("tcp_discardcb: so == NULL")); 683 */ 684 INP_LOCK_ASSERT(inp); 685 686 /* 687 * Make sure that all of our timers are stopped before we 688 * delete the PCB. 689 */ 690 callout_stop(tp->tt_rexmt); 691 callout_stop(tp->tt_persist); 692 callout_stop(tp->tt_keep); 693 callout_stop(tp->tt_2msl); 694 callout_stop(tp->tt_delack); 695 696 /* 697 * If we got enough samples through the srtt filter, 698 * save the rtt and rttvar in the routing entry. 699 * 'Enough' is arbitrarily defined as 4 rtt samples. 700 * 4 samples is enough for the srtt filter to converge 701 * to within enough % of the correct value; fewer samples 702 * and we could save a bogus rtt. The danger is not high 703 * as tcp quickly recovers from everything. 704 * XXX: Works very well but needs some more statistics! 705 */ 706 if (tp->t_rttupdated >= 4) { 707 struct hc_metrics_lite metrics; 708 u_long ssthresh; 709 710 bzero(&metrics, sizeof(metrics)); 711 /* 712 * Update the ssthresh always when the conditions below 713 * are satisfied. This gives us better new start value 714 * for the congestion avoidance for new connections. 715 * ssthresh is only set if packet loss occured on a session. 716 */ 717 ssthresh = tp->snd_ssthresh; 718 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 719 /* 720 * convert the limit from user data bytes to 721 * packets then to packet data bytes. 722 */ 723 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 724 if (ssthresh < 2) 725 ssthresh = 2; 726 ssthresh *= (u_long)(tp->t_maxseg + 727#ifdef INET6 728 (isipv6 ? sizeof (struct ip6_hdr) + 729 sizeof (struct tcphdr) : 730#endif 731 sizeof (struct tcpiphdr) 732#ifdef INET6 733 ) 734#endif 735 ); 736 } else 737 ssthresh = 0; 738 metrics.rmx_ssthresh = ssthresh; 739 740 metrics.rmx_rtt = tp->t_srtt; 741 metrics.rmx_rttvar = tp->t_rttvar; 742 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 743 metrics.rmx_bandwidth = tp->snd_bandwidth; 744 metrics.rmx_cwnd = tp->snd_cwnd; 745 metrics.rmx_sendpipe = 0; 746 metrics.rmx_recvpipe = 0; 747 748 tcp_hc_update(&inp->inp_inc, &metrics); 749 } 750 751 /* free the reassembly queue, if any */ 752 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 753 LIST_REMOVE(q, tqe_q); 754 m_freem(q->tqe_m); 755 uma_zfree(tcp_reass_zone, q); 756 tp->t_segqlen--; 757 tcp_reass_qsize--; 758 } 759 tcp_free_sackholes(tp); 760 inp->inp_ppcb = NULL; 761 tp->t_inpcb = NULL; 762 uma_zfree(tcpcb_zone, tp); 763 764 /* 765 * XXXRW: This seems a bit unclean. 766 */ 767 if (so != NULL) 768 soisdisconnected(so); 769} 770 771/* 772 * Attempt to close a TCP control block, marking it as dropped, and freeing 773 * the socket if we hold the only reference. 774 */ 775struct tcpcb * 776tcp_close(tp) 777 struct tcpcb *tp; 778{ 779 struct inpcb *inp = tp->t_inpcb; 780 struct socket *so; 781 782 INP_INFO_WLOCK_ASSERT(&tcbinfo); 783 INP_LOCK_ASSERT(inp); 784 785 inp->inp_vflag |= INP_DROPPED; 786 787 tcpstat.tcps_closed++; 788 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 789 so = inp->inp_socket; 790 soisdisconnected(so); 791 if (inp->inp_vflag & INP_SOCKREF) { 792 KASSERT(so->so_state & SS_PROTOREF, 793 ("tcp_close: !SS_PROTOREF")); 794 inp->inp_vflag &= ~INP_SOCKREF; 795 tcp_discardcb(tp); 796#ifdef INET6 797 if (inp->inp_vflag & INP_IPV6PROTO) { 798 in6_pcbdetach(inp); 799 in6_pcbfree(inp); 800 } else { 801#endif 802 in_pcbdetach(inp); 803 in_pcbfree(inp); 804#ifdef INET6 805 } 806#endif 807 ACCEPT_LOCK(); 808 SOCK_LOCK(so); 809 so->so_state &= ~SS_PROTOREF; 810 sofree(so); 811 return (NULL); 812 } 813 return (tp); 814} 815 816void 817tcp_drain() 818{ 819 if (do_tcpdrain) 820 { 821 struct inpcb *inpb; 822 struct tcpcb *tcpb; 823 struct tseg_qent *te; 824 825 /* 826 * Walk the tcpbs, if existing, and flush the reassembly queue, 827 * if there is one... 828 * XXX: The "Net/3" implementation doesn't imply that the TCP 829 * reassembly queue should be flushed, but in a situation 830 * where we're really low on mbufs, this is potentially 831 * usefull. 832 */ 833 INP_INFO_RLOCK(&tcbinfo); 834 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 835 if (inpb->inp_vflag & INP_TIMEWAIT) 836 continue; 837 INP_LOCK(inpb); 838 if ((tcpb = intotcpcb(inpb)) != NULL) { 839 while ((te = LIST_FIRST(&tcpb->t_segq)) 840 != NULL) { 841 LIST_REMOVE(te, tqe_q); 842 m_freem(te->tqe_m); 843 uma_zfree(tcp_reass_zone, te); 844 tcpb->t_segqlen--; 845 tcp_reass_qsize--; 846 } 847 tcp_clean_sackreport(tcpb); 848 } 849 INP_UNLOCK(inpb); 850 } 851 INP_INFO_RUNLOCK(&tcbinfo); 852 } 853} 854 855/* 856 * Notify a tcp user of an asynchronous error; 857 * store error as soft error, but wake up user 858 * (for now, won't do anything until can select for soft error). 859 * 860 * Do not wake up user since there currently is no mechanism for 861 * reporting soft errors (yet - a kqueue filter may be added). 862 */ 863static struct inpcb * 864tcp_notify(inp, error) 865 struct inpcb *inp; 866 int error; 867{ 868 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 869 870 INP_INFO_WLOCK_ASSERT(&tcbinfo); 871 INP_LOCK_ASSERT(inp); 872 873 /* 874 * Ignore some errors if we are hooked up. 875 * If connection hasn't completed, has retransmitted several times, 876 * and receives a second error, give up now. This is better 877 * than waiting a long time to establish a connection that 878 * can never complete. 879 */ 880 if (tp->t_state == TCPS_ESTABLISHED && 881 (error == EHOSTUNREACH || error == ENETUNREACH || 882 error == EHOSTDOWN)) { 883 return (inp); 884 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 885 tp->t_softerror) { 886 tp = tcp_drop(tp, error); 887 if (tp != NULL) 888 return (inp); 889 else 890 return (NULL); 891 } else { 892 tp->t_softerror = error; 893 return (inp); 894 } 895#if 0 896 wakeup( &so->so_timeo); 897 sorwakeup(so); 898 sowwakeup(so); 899#endif 900} 901 902static int 903tcp_pcblist(SYSCTL_HANDLER_ARGS) 904{ 905 int error, i, n; 906 struct inpcb *inp, **inp_list; 907 inp_gen_t gencnt; 908 struct xinpgen xig; 909 910 /* 911 * The process of preparing the TCB list is too time-consuming and 912 * resource-intensive to repeat twice on every request. 913 */ 914 if (req->oldptr == NULL) { 915 n = tcbinfo.ipi_count; 916 req->oldidx = 2 * (sizeof xig) 917 + (n + n/8) * sizeof(struct xtcpcb); 918 return (0); 919 } 920 921 if (req->newptr != NULL) 922 return (EPERM); 923 924 /* 925 * OK, now we're committed to doing something. 926 */ 927 INP_INFO_RLOCK(&tcbinfo); 928 gencnt = tcbinfo.ipi_gencnt; 929 n = tcbinfo.ipi_count; 930 INP_INFO_RUNLOCK(&tcbinfo); 931 932 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 933 + n * sizeof(struct xtcpcb)); 934 if (error != 0) 935 return (error); 936 937 xig.xig_len = sizeof xig; 938 xig.xig_count = n; 939 xig.xig_gen = gencnt; 940 xig.xig_sogen = so_gencnt; 941 error = SYSCTL_OUT(req, &xig, sizeof xig); 942 if (error) 943 return (error); 944 945 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 946 if (inp_list == NULL) 947 return (ENOMEM); 948 949 INP_INFO_RLOCK(&tcbinfo); 950 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 951 inp = LIST_NEXT(inp, inp_list)) { 952 INP_LOCK(inp); 953 if (inp->inp_gencnt <= gencnt) { 954 /* 955 * XXX: This use of cr_cansee(), introduced with 956 * TCP state changes, is not quite right, but for 957 * now, better than nothing. 958 */ 959 if (inp->inp_vflag & INP_TIMEWAIT) 960 error = cr_cansee(req->td->td_ucred, 961 intotw(inp)->tw_cred); 962 else 963 error = cr_canseesocket(req->td->td_ucred, 964 inp->inp_socket); 965 if (error == 0) 966 inp_list[i++] = inp; 967 } 968 INP_UNLOCK(inp); 969 } 970 INP_INFO_RUNLOCK(&tcbinfo); 971 n = i; 972 973 error = 0; 974 for (i = 0; i < n; i++) { 975 inp = inp_list[i]; 976 if (inp->inp_gencnt <= gencnt) { 977 struct xtcpcb xt; 978 caddr_t inp_ppcb; 979 980 bzero(&xt, sizeof(xt)); 981 xt.xt_len = sizeof xt; 982 /* XXX should avoid extra copy */ 983 bcopy(inp, &xt.xt_inp, sizeof *inp); 984 inp_ppcb = inp->inp_ppcb; 985 if (inp_ppcb == NULL) 986 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 987 else if (inp->inp_vflag & INP_TIMEWAIT) { 988 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 989 xt.xt_tp.t_state = TCPS_TIME_WAIT; 990 } else 991 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 992 if (inp->inp_socket != NULL) 993 sotoxsocket(inp->inp_socket, &xt.xt_socket); 994 else { 995 bzero(&xt.xt_socket, sizeof xt.xt_socket); 996 xt.xt_socket.xso_protocol = IPPROTO_TCP; 997 } 998 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 999 error = SYSCTL_OUT(req, &xt, sizeof xt); 1000 } 1001 } 1002 if (!error) { 1003 /* 1004 * Give the user an updated idea of our state. 1005 * If the generation differs from what we told 1006 * her before, she knows that something happened 1007 * while we were processing this request, and it 1008 * might be necessary to retry. 1009 */ 1010 INP_INFO_RLOCK(&tcbinfo); 1011 xig.xig_gen = tcbinfo.ipi_gencnt; 1012 xig.xig_sogen = so_gencnt; 1013 xig.xig_count = tcbinfo.ipi_count; 1014 INP_INFO_RUNLOCK(&tcbinfo); 1015 error = SYSCTL_OUT(req, &xig, sizeof xig); 1016 } 1017 free(inp_list, M_TEMP); 1018 return (error); 1019} 1020 1021SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1022 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1023 1024static int 1025tcp_getcred(SYSCTL_HANDLER_ARGS) 1026{ 1027 struct xucred xuc; 1028 struct sockaddr_in addrs[2]; 1029 struct inpcb *inp; 1030 int error; 1031 1032 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1033 if (error) 1034 return (error); 1035 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1036 if (error) 1037 return (error); 1038 INP_INFO_RLOCK(&tcbinfo); 1039 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1040 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1041 if (inp == NULL) { 1042 error = ENOENT; 1043 goto outunlocked; 1044 } 1045 INP_LOCK(inp); 1046 if (inp->inp_socket == NULL) { 1047 error = ENOENT; 1048 goto out; 1049 } 1050 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1051 if (error) 1052 goto out; 1053 cru2x(inp->inp_socket->so_cred, &xuc); 1054out: 1055 INP_UNLOCK(inp); 1056outunlocked: 1057 INP_INFO_RUNLOCK(&tcbinfo); 1058 if (error == 0) 1059 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1060 return (error); 1061} 1062 1063SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1064 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1065 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1066 1067#ifdef INET6 1068static int 1069tcp6_getcred(SYSCTL_HANDLER_ARGS) 1070{ 1071 struct xucred xuc; 1072 struct sockaddr_in6 addrs[2]; 1073 struct inpcb *inp; 1074 int error, mapped = 0; 1075 1076 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1077 if (error) 1078 return (error); 1079 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1080 if (error) 1081 return (error); 1082 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1083 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1084 return (error); 1085 } 1086 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1087 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1088 mapped = 1; 1089 else 1090 return (EINVAL); 1091 } 1092 1093 INP_INFO_RLOCK(&tcbinfo); 1094 if (mapped == 1) 1095 inp = in_pcblookup_hash(&tcbinfo, 1096 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1097 addrs[1].sin6_port, 1098 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1099 addrs[0].sin6_port, 1100 0, NULL); 1101 else 1102 inp = in6_pcblookup_hash(&tcbinfo, 1103 &addrs[1].sin6_addr, addrs[1].sin6_port, 1104 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1105 if (inp == NULL) { 1106 error = ENOENT; 1107 goto outunlocked; 1108 } 1109 INP_LOCK(inp); 1110 if (inp->inp_socket == NULL) { 1111 error = ENOENT; 1112 goto out; 1113 } 1114 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1115 if (error) 1116 goto out; 1117 cru2x(inp->inp_socket->so_cred, &xuc); 1118out: 1119 INP_UNLOCK(inp); 1120outunlocked: 1121 INP_INFO_RUNLOCK(&tcbinfo); 1122 if (error == 0) 1123 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1124 return (error); 1125} 1126 1127SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1128 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1129 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1130#endif 1131 1132 1133void 1134tcp_ctlinput(cmd, sa, vip) 1135 int cmd; 1136 struct sockaddr *sa; 1137 void *vip; 1138{ 1139 struct ip *ip = vip; 1140 struct tcphdr *th; 1141 struct in_addr faddr; 1142 struct inpcb *inp; 1143 struct tcpcb *tp; 1144 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1145 struct icmp *icp; 1146 struct in_conninfo inc; 1147 tcp_seq icmp_tcp_seq; 1148 int mtu; 1149 1150 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1151 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1152 return; 1153 1154 if (cmd == PRC_MSGSIZE) 1155 notify = tcp_mtudisc; 1156 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1157 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1158 notify = tcp_drop_syn_sent; 1159 /* 1160 * Redirects don't need to be handled up here. 1161 */ 1162 else if (PRC_IS_REDIRECT(cmd)) 1163 return; 1164 /* 1165 * Source quench is depreciated. 1166 */ 1167 else if (cmd == PRC_QUENCH) 1168 return; 1169 /* 1170 * Hostdead is ugly because it goes linearly through all PCBs. 1171 * XXX: We never get this from ICMP, otherwise it makes an 1172 * excellent DoS attack on machines with many connections. 1173 */ 1174 else if (cmd == PRC_HOSTDEAD) 1175 ip = NULL; 1176 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1177 return; 1178 if (ip != NULL) { 1179 icp = (struct icmp *)((caddr_t)ip 1180 - offsetof(struct icmp, icmp_ip)); 1181 th = (struct tcphdr *)((caddr_t)ip 1182 + (ip->ip_hl << 2)); 1183 INP_INFO_WLOCK(&tcbinfo); 1184 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1185 ip->ip_src, th->th_sport, 0, NULL); 1186 if (inp != NULL) { 1187 INP_LOCK(inp); 1188 if (inp->inp_socket != NULL) { 1189 icmp_tcp_seq = htonl(th->th_seq); 1190 tp = intotcpcb(inp); 1191 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1192 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1193 if (cmd == PRC_MSGSIZE) { 1194 /* 1195 * MTU discovery: 1196 * If we got a needfrag set the MTU 1197 * in the route to the suggested new 1198 * value (if given) and then notify. 1199 */ 1200 bzero(&inc, sizeof(inc)); 1201 inc.inc_flags = 0; /* IPv4 */ 1202 inc.inc_faddr = faddr; 1203 1204 mtu = ntohs(icp->icmp_nextmtu); 1205 /* 1206 * If no alternative MTU was 1207 * proposed, try the next smaller 1208 * one. ip->ip_len has already 1209 * been swapped in icmp_input(). 1210 */ 1211 if (!mtu) 1212 mtu = ip_next_mtu(ip->ip_len, 1213 1); 1214 if (mtu < max(296, (tcp_minmss) 1215 + sizeof(struct tcpiphdr))) 1216 mtu = 0; 1217 if (!mtu) 1218 mtu = tcp_mssdflt 1219 + sizeof(struct tcpiphdr); 1220 /* 1221 * Only cache the the MTU if it 1222 * is smaller than the interface 1223 * or route MTU. tcp_mtudisc() 1224 * will do right thing by itself. 1225 */ 1226 if (mtu <= tcp_maxmtu(&inc)) 1227 tcp_hc_updatemtu(&inc, mtu); 1228 } 1229 1230 inp = (*notify)(inp, inetctlerrmap[cmd]); 1231 } 1232 } 1233 if (inp != NULL) 1234 INP_UNLOCK(inp); 1235 } else { 1236 inc.inc_fport = th->th_dport; 1237 inc.inc_lport = th->th_sport; 1238 inc.inc_faddr = faddr; 1239 inc.inc_laddr = ip->ip_src; 1240#ifdef INET6 1241 inc.inc_isipv6 = 0; 1242#endif 1243 syncache_unreach(&inc, th); 1244 } 1245 INP_INFO_WUNLOCK(&tcbinfo); 1246 } else 1247 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1248} 1249 1250#ifdef INET6 1251void 1252tcp6_ctlinput(cmd, sa, d) 1253 int cmd; 1254 struct sockaddr *sa; 1255 void *d; 1256{ 1257 struct tcphdr th; 1258 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1259 struct ip6_hdr *ip6; 1260 struct mbuf *m; 1261 struct ip6ctlparam *ip6cp = NULL; 1262 const struct sockaddr_in6 *sa6_src = NULL; 1263 int off; 1264 struct tcp_portonly { 1265 u_int16_t th_sport; 1266 u_int16_t th_dport; 1267 } *thp; 1268 1269 if (sa->sa_family != AF_INET6 || 1270 sa->sa_len != sizeof(struct sockaddr_in6)) 1271 return; 1272 1273 if (cmd == PRC_MSGSIZE) 1274 notify = tcp_mtudisc; 1275 else if (!PRC_IS_REDIRECT(cmd) && 1276 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1277 return; 1278 /* Source quench is depreciated. */ 1279 else if (cmd == PRC_QUENCH) 1280 return; 1281 1282 /* if the parameter is from icmp6, decode it. */ 1283 if (d != NULL) { 1284 ip6cp = (struct ip6ctlparam *)d; 1285 m = ip6cp->ip6c_m; 1286 ip6 = ip6cp->ip6c_ip6; 1287 off = ip6cp->ip6c_off; 1288 sa6_src = ip6cp->ip6c_src; 1289 } else { 1290 m = NULL; 1291 ip6 = NULL; 1292 off = 0; /* fool gcc */ 1293 sa6_src = &sa6_any; 1294 } 1295 1296 if (ip6 != NULL) { 1297 struct in_conninfo inc; 1298 /* 1299 * XXX: We assume that when IPV6 is non NULL, 1300 * M and OFF are valid. 1301 */ 1302 1303 /* check if we can safely examine src and dst ports */ 1304 if (m->m_pkthdr.len < off + sizeof(*thp)) 1305 return; 1306 1307 bzero(&th, sizeof(th)); 1308 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1309 1310 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1311 (struct sockaddr *)ip6cp->ip6c_src, 1312 th.th_sport, cmd, NULL, notify); 1313 1314 inc.inc_fport = th.th_dport; 1315 inc.inc_lport = th.th_sport; 1316 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1317 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1318 inc.inc_isipv6 = 1; 1319 INP_INFO_WLOCK(&tcbinfo); 1320 syncache_unreach(&inc, &th); 1321 INP_INFO_WUNLOCK(&tcbinfo); 1322 } else 1323 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1324 0, cmd, NULL, notify); 1325} 1326#endif /* INET6 */ 1327 1328 1329/* 1330 * Following is where TCP initial sequence number generation occurs. 1331 * 1332 * There are two places where we must use initial sequence numbers: 1333 * 1. In SYN-ACK packets. 1334 * 2. In SYN packets. 1335 * 1336 * All ISNs for SYN-ACK packets are generated by the syncache. See 1337 * tcp_syncache.c for details. 1338 * 1339 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1340 * depends on this property. In addition, these ISNs should be 1341 * unguessable so as to prevent connection hijacking. To satisfy 1342 * the requirements of this situation, the algorithm outlined in 1343 * RFC 1948 is used, with only small modifications. 1344 * 1345 * Implementation details: 1346 * 1347 * Time is based off the system timer, and is corrected so that it 1348 * increases by one megabyte per second. This allows for proper 1349 * recycling on high speed LANs while still leaving over an hour 1350 * before rollover. 1351 * 1352 * As reading the *exact* system time is too expensive to be done 1353 * whenever setting up a TCP connection, we increment the time 1354 * offset in two ways. First, a small random positive increment 1355 * is added to isn_offset for each connection that is set up. 1356 * Second, the function tcp_isn_tick fires once per clock tick 1357 * and increments isn_offset as necessary so that sequence numbers 1358 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1359 * random positive increments serve only to ensure that the same 1360 * exact sequence number is never sent out twice (as could otherwise 1361 * happen when a port is recycled in less than the system tick 1362 * interval.) 1363 * 1364 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1365 * between seeding of isn_secret. This is normally set to zero, 1366 * as reseeding should not be necessary. 1367 * 1368 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1369 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1370 * general, this means holding an exclusive (write) lock. 1371 */ 1372 1373#define ISN_BYTES_PER_SECOND 1048576 1374#define ISN_STATIC_INCREMENT 4096 1375#define ISN_RANDOM_INCREMENT (4096 - 1) 1376 1377static u_char isn_secret[32]; 1378static int isn_last_reseed; 1379static u_int32_t isn_offset, isn_offset_old; 1380static MD5_CTX isn_ctx; 1381 1382tcp_seq 1383tcp_new_isn(tp) 1384 struct tcpcb *tp; 1385{ 1386 u_int32_t md5_buffer[4]; 1387 tcp_seq new_isn; 1388 1389 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1390 INP_LOCK_ASSERT(tp->t_inpcb); 1391 1392 /* Seed if this is the first use, reseed if requested. */ 1393 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1394 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1395 < (u_int)ticks))) { 1396 read_random(&isn_secret, sizeof(isn_secret)); 1397 isn_last_reseed = ticks; 1398 } 1399 1400 /* Compute the md5 hash and return the ISN. */ 1401 MD5Init(&isn_ctx); 1402 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1403 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1404#ifdef INET6 1405 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1406 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1407 sizeof(struct in6_addr)); 1408 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1409 sizeof(struct in6_addr)); 1410 } else 1411#endif 1412 { 1413 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1414 sizeof(struct in_addr)); 1415 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1416 sizeof(struct in_addr)); 1417 } 1418 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1419 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1420 new_isn = (tcp_seq) md5_buffer[0]; 1421 isn_offset += ISN_STATIC_INCREMENT + 1422 (arc4random() & ISN_RANDOM_INCREMENT); 1423 new_isn += isn_offset; 1424 return (new_isn); 1425} 1426 1427/* 1428 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1429 * to keep time flowing at a relatively constant rate. If the random 1430 * increments have already pushed us past the projected offset, do nothing. 1431 */ 1432static void 1433tcp_isn_tick(xtp) 1434 void *xtp; 1435{ 1436 u_int32_t projected_offset; 1437 1438 INP_INFO_WLOCK(&tcbinfo); 1439 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1440 1441 if (projected_offset > isn_offset) 1442 isn_offset = projected_offset; 1443 1444 isn_offset_old = isn_offset; 1445 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1446 INP_INFO_WUNLOCK(&tcbinfo); 1447} 1448 1449/* 1450 * When a specific ICMP unreachable message is received and the 1451 * connection state is SYN-SENT, drop the connection. This behavior 1452 * is controlled by the icmp_may_rst sysctl. 1453 */ 1454struct inpcb * 1455tcp_drop_syn_sent(inp, errno) 1456 struct inpcb *inp; 1457 int errno; 1458{ 1459 struct tcpcb *tp = intotcpcb(inp); 1460 1461 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1462 INP_LOCK_ASSERT(inp); 1463 1464 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) { 1465 tp = tcp_drop(tp, errno); 1466 if (tp != NULL) 1467 return (inp); 1468 else 1469 return (NULL); 1470 } 1471 return (inp); 1472} 1473 1474/* 1475 * When `need fragmentation' ICMP is received, update our idea of the MSS 1476 * based on the new value in the route. Also nudge TCP to send something, 1477 * since we know the packet we just sent was dropped. 1478 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1479 */ 1480struct inpcb * 1481tcp_mtudisc(inp, errno) 1482 struct inpcb *inp; 1483 int errno; 1484{ 1485 struct tcpcb *tp = intotcpcb(inp); 1486 struct socket *so = inp->inp_socket; 1487 u_int maxmtu; 1488 u_int romtu; 1489 int mss; 1490#ifdef INET6 1491 int isipv6; 1492#endif /* INET6 */ 1493 1494 INP_LOCK_ASSERT(inp); 1495 if (tp != NULL) { 1496#ifdef INET6 1497 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1498#endif 1499 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1500 romtu = 1501#ifdef INET6 1502 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1503#endif /* INET6 */ 1504 tcp_maxmtu(&inp->inp_inc); 1505 if (!maxmtu) 1506 maxmtu = romtu; 1507 else 1508 maxmtu = min(maxmtu, romtu); 1509 if (!maxmtu) { 1510 tp->t_maxopd = tp->t_maxseg = 1511#ifdef INET6 1512 isipv6 ? tcp_v6mssdflt : 1513#endif /* INET6 */ 1514 tcp_mssdflt; 1515 return (inp); 1516 } 1517 mss = maxmtu - 1518#ifdef INET6 1519 (isipv6 ? 1520 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1521#endif /* INET6 */ 1522 sizeof(struct tcpiphdr) 1523#ifdef INET6 1524 ) 1525#endif /* INET6 */ 1526 ; 1527 1528 /* 1529 * XXX - The above conditional probably violates the TCP 1530 * spec. The problem is that, since we don't know the 1531 * other end's MSS, we are supposed to use a conservative 1532 * default. But, if we do that, then MTU discovery will 1533 * never actually take place, because the conservative 1534 * default is much less than the MTUs typically seen 1535 * on the Internet today. For the moment, we'll sweep 1536 * this under the carpet. 1537 * 1538 * The conservative default might not actually be a problem 1539 * if the only case this occurs is when sending an initial 1540 * SYN with options and data to a host we've never talked 1541 * to before. Then, they will reply with an MSS value which 1542 * will get recorded and the new parameters should get 1543 * recomputed. For Further Study. 1544 */ 1545 if (tp->t_maxopd <= mss) 1546 return (inp); 1547 tp->t_maxopd = mss; 1548 1549 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1550 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1551 mss -= TCPOLEN_TSTAMP_APPA; 1552#if (MCLBYTES & (MCLBYTES - 1)) == 0 1553 if (mss > MCLBYTES) 1554 mss &= ~(MCLBYTES-1); 1555#else 1556 if (mss > MCLBYTES) 1557 mss = mss / MCLBYTES * MCLBYTES; 1558#endif 1559 if (so->so_snd.sb_hiwat < mss) 1560 mss = so->so_snd.sb_hiwat; 1561 1562 tp->t_maxseg = mss; 1563 1564 tcpstat.tcps_mturesent++; 1565 tp->t_rtttime = 0; 1566 tp->snd_nxt = tp->snd_una; 1567 tcp_output(tp); 1568 } 1569 return (inp); 1570} 1571 1572/* 1573 * Look-up the routing entry to the peer of this inpcb. If no route 1574 * is found and it cannot be allocated, then return NULL. This routine 1575 * is called by TCP routines that access the rmx structure and by tcp_mss 1576 * to get the interface MTU. 1577 */ 1578u_long 1579tcp_maxmtu(inc) 1580 struct in_conninfo *inc; 1581{ 1582 struct route sro; 1583 struct sockaddr_in *dst; 1584 struct ifnet *ifp; 1585 u_long maxmtu = 0; 1586 1587 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1588 1589 bzero(&sro, sizeof(sro)); 1590 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1591 dst = (struct sockaddr_in *)&sro.ro_dst; 1592 dst->sin_family = AF_INET; 1593 dst->sin_len = sizeof(*dst); 1594 dst->sin_addr = inc->inc_faddr; 1595 rtalloc_ign(&sro, RTF_CLONING); 1596 } 1597 if (sro.ro_rt != NULL) { 1598 ifp = sro.ro_rt->rt_ifp; 1599 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1600 maxmtu = ifp->if_mtu; 1601 else 1602 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1603 RTFREE(sro.ro_rt); 1604 } 1605 return (maxmtu); 1606} 1607 1608#ifdef INET6 1609u_long 1610tcp_maxmtu6(inc) 1611 struct in_conninfo *inc; 1612{ 1613 struct route_in6 sro6; 1614 struct ifnet *ifp; 1615 u_long maxmtu = 0; 1616 1617 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1618 1619 bzero(&sro6, sizeof(sro6)); 1620 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1621 sro6.ro_dst.sin6_family = AF_INET6; 1622 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1623 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1624 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1625 } 1626 if (sro6.ro_rt != NULL) { 1627 ifp = sro6.ro_rt->rt_ifp; 1628 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1629 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1630 else 1631 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1632 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1633 RTFREE(sro6.ro_rt); 1634 } 1635 1636 return (maxmtu); 1637} 1638#endif /* INET6 */ 1639 1640#ifdef IPSEC 1641/* compute ESP/AH header size for TCP, including outer IP header. */ 1642size_t 1643ipsec_hdrsiz_tcp(tp) 1644 struct tcpcb *tp; 1645{ 1646 struct inpcb *inp; 1647 struct mbuf *m; 1648 size_t hdrsiz; 1649 struct ip *ip; 1650#ifdef INET6 1651 struct ip6_hdr *ip6; 1652#endif 1653 struct tcphdr *th; 1654 1655 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1656 return (0); 1657 MGETHDR(m, M_DONTWAIT, MT_DATA); 1658 if (!m) 1659 return (0); 1660 1661#ifdef INET6 1662 if ((inp->inp_vflag & INP_IPV6) != 0) { 1663 ip6 = mtod(m, struct ip6_hdr *); 1664 th = (struct tcphdr *)(ip6 + 1); 1665 m->m_pkthdr.len = m->m_len = 1666 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1667 tcpip_fillheaders(inp, ip6, th); 1668 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1669 } else 1670#endif /* INET6 */ 1671 { 1672 ip = mtod(m, struct ip *); 1673 th = (struct tcphdr *)(ip + 1); 1674 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1675 tcpip_fillheaders(inp, ip, th); 1676 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1677 } 1678 1679 m_free(m); 1680 return (hdrsiz); 1681} 1682#endif /*IPSEC*/ 1683 1684/* 1685 * Move a TCP connection into TIME_WAIT state. 1686 * tcbinfo is locked. 1687 * inp is locked, and is unlocked before returning. 1688 */ 1689void 1690tcp_twstart(tp) 1691 struct tcpcb *tp; 1692{ 1693 struct tcptw *tw; 1694 struct inpcb *inp; 1695 int tw_time, acknow; 1696 struct socket *so; 1697 1698 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1699 INP_LOCK_ASSERT(tp->t_inpcb); 1700 1701 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1702 if (tw == NULL) { 1703 tw = tcp_timer_2msl_tw(1); 1704 if (tw == NULL) { 1705 tp = tcp_close(tp); 1706 if (tp != NULL) 1707 INP_UNLOCK(tp->t_inpcb); 1708 return; 1709 } 1710 } 1711 inp = tp->t_inpcb; 1712 tw->tw_inpcb = inp; 1713 1714 /* 1715 * Recover last window size sent. 1716 */ 1717 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1718 1719 /* 1720 * Set t_recent if timestamps are used on the connection. 1721 */ 1722 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1723 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1724 tw->t_recent = tp->ts_recent; 1725 else 1726 tw->t_recent = 0; 1727 1728 tw->snd_nxt = tp->snd_nxt; 1729 tw->rcv_nxt = tp->rcv_nxt; 1730 tw->iss = tp->iss; 1731 tw->irs = tp->irs; 1732 tw->t_starttime = tp->t_starttime; 1733 tw->tw_time = 0; 1734 1735/* XXX 1736 * If this code will 1737 * be used for fin-wait-2 state also, then we may need 1738 * a ts_recent from the last segment. 1739 */ 1740 tw_time = 2 * tcp_msl; 1741 acknow = tp->t_flags & TF_ACKNOW; 1742 1743 /* 1744 * First, discard tcpcb state, which includes stopping its timers and 1745 * freeing it. tcp_discardcb() used to also release the inpcb, but 1746 * that work is now done in the caller. 1747 */ 1748 tcp_discardcb(tp); 1749 so = inp->inp_socket; 1750 SOCK_LOCK(so); 1751 tw->tw_cred = crhold(so->so_cred); 1752 tw->tw_so_options = so->so_options; 1753 SOCK_UNLOCK(so); 1754 if (acknow) 1755 tcp_twrespond(tw, TH_ACK); 1756 inp->inp_ppcb = (caddr_t)tw; 1757 inp->inp_vflag |= INP_TIMEWAIT; 1758 tcp_timer_2msl_reset(tw, tw_time); 1759 1760 /* 1761 * If the inpcb owns the sole reference to the socket, then we can 1762 * detach and free the socket as it is not needed in time wait. 1763 */ 1764 if (inp->inp_vflag & INP_SOCKREF) { 1765 KASSERT(so->so_state & SS_PROTOREF, 1766 ("tcp_twstart: !SS_PROTOREF")); 1767 inp->inp_vflag &= ~INP_SOCKREF; 1768#ifdef INET6 1769 if (inp->inp_vflag & INP_IPV6PROTO) 1770 in6_pcbdetach(inp); 1771 else 1772#endif 1773 in_pcbdetach(inp); 1774 INP_UNLOCK(inp); 1775 ACCEPT_LOCK(); 1776 SOCK_LOCK(so); 1777 so->so_state &= ~SS_PROTOREF; 1778 sofree(so); 1779 } else 1780 INP_UNLOCK(inp); 1781} 1782 1783/* 1784 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1785 * the actual rate is slightly higher due to the addition of 1786 * random positive increments. 1787 * 1788 * Most other new OSes use semi-randomized ISN values, so we 1789 * do not need to worry about them. 1790 */ 1791#define MS_ISN_BYTES_PER_SECOND 250000 1792 1793/* 1794 * Determine if the ISN we will generate has advanced beyond the last 1795 * sequence number used by the previous connection. If so, indicate 1796 * that it is safe to recycle this tw socket by returning 1. 1797 * 1798 * XXXRW: This function should assert the inpcb lock as it does multiple 1799 * non-atomic reads from the tcptw, but is currently called without it from 1800 * in_pcb.c:in_pcblookup_local(). 1801 */ 1802int 1803tcp_twrecycleable(struct tcptw *tw) 1804{ 1805 tcp_seq new_iss = tw->iss; 1806 tcp_seq new_irs = tw->irs; 1807 1808 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1809 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1810 1811 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1812 return (1); 1813 else 1814 return (0); 1815} 1816 1817void 1818tcp_twclose(struct tcptw *tw, int reuse) 1819{ 1820 struct socket *so; 1821 struct inpcb *inp; 1822 1823 /* 1824 * At this point, we are in one of two situations: 1825 * 1826 * (1) We have no socket, just an inpcb<->twtcp pair. Release it all 1827 * after validating. 1828 * 1829 * (2) We have a socket, which we may or may now own the reference 1830 * for. If we own the reference, release all the state after 1831 * validating. If not, leave it for the socket close to clean up. 1832 */ 1833 inp = tw->tw_inpcb; 1834 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1835 KASSERT(inp->inp_ppcb == (void *)tw, ("tcp_twclose: inp_ppcb != tw")); 1836 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1837 INP_LOCK_ASSERT(inp); 1838 1839 tw->tw_inpcb = NULL; 1840 tcp_timer_2msl_stop(tw); 1841 inp->inp_ppcb = NULL; 1842 inp->inp_vflag |= INP_DROPPED; 1843 1844 so = inp->inp_socket; 1845 if (so != NULL) { 1846 if (inp->inp_vflag & INP_SOCKREF) { 1847 /* 1848 * If a socket is present, and we own the only 1849 * reference, we need to tear down the socket and the 1850 * inpcb. 1851 */ 1852 inp->inp_vflag &= ~INP_SOCKREF; 1853#ifdef INET6 1854 if (inp->inp_vflag & INP_IPV6PROTO) { 1855 in6_pcbdetach(inp); 1856 in6_pcbfree(inp); 1857 } else { 1858 in_pcbdetach(inp); 1859 in_pcbfree(inp); 1860 } 1861#endif 1862 ACCEPT_LOCK(); 1863 SOCK_LOCK(so); 1864 KASSERT(so->so_state & SS_PROTOREF, 1865 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1866 so->so_state &= ~SS_PROTOREF; 1867 sofree(so); 1868 } else { 1869 /* 1870 * If we don't own the only reference, the socket and 1871 * inpcb need to be left around to be handled by 1872 * tcp_usr_detach() later. 1873 */ 1874 INP_UNLOCK(inp); 1875 } 1876 } else { 1877#ifdef INET6 1878 if (inp->inp_vflag & INP_IPV6PROTO) 1879 in6_pcbfree(inp); 1880 else 1881#endif 1882 in_pcbfree(inp); 1883 } 1884 tcpstat.tcps_closed++; 1885 crfree(tw->tw_cred); 1886 tw->tw_cred = NULL; 1887 if (reuse) 1888 return; 1889 uma_zfree(tcptw_zone, tw); 1890} 1891 1892int 1893tcp_twrespond(struct tcptw *tw, int flags) 1894{ 1895 struct inpcb *inp = tw->tw_inpcb; 1896 struct tcphdr *th; 1897 struct mbuf *m; 1898 struct ip *ip = NULL; 1899 u_int8_t *optp; 1900 u_int hdrlen, optlen; 1901 int error; 1902#ifdef INET6 1903 struct ip6_hdr *ip6 = NULL; 1904 int isipv6 = inp->inp_inc.inc_isipv6; 1905#endif 1906 1907 INP_LOCK_ASSERT(inp); 1908 1909 m = m_gethdr(M_DONTWAIT, MT_DATA); 1910 if (m == NULL) 1911 return (ENOBUFS); 1912 m->m_data += max_linkhdr; 1913 1914#ifdef MAC 1915 mac_create_mbuf_from_inpcb(inp, m); 1916#endif 1917 1918#ifdef INET6 1919 if (isipv6) { 1920 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1921 ip6 = mtod(m, struct ip6_hdr *); 1922 th = (struct tcphdr *)(ip6 + 1); 1923 tcpip_fillheaders(inp, ip6, th); 1924 } else 1925#endif 1926 { 1927 hdrlen = sizeof(struct tcpiphdr); 1928 ip = mtod(m, struct ip *); 1929 th = (struct tcphdr *)(ip + 1); 1930 tcpip_fillheaders(inp, ip, th); 1931 } 1932 optp = (u_int8_t *)(th + 1); 1933 1934 /* 1935 * Send a timestamp and echo-reply if both our side and our peer 1936 * have sent timestamps in our SYN's and this is not a RST. 1937 */ 1938 if (tw->t_recent && flags == TH_ACK) { 1939 u_int32_t *lp = (u_int32_t *)optp; 1940 1941 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1942 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1943 *lp++ = htonl(ticks); 1944 *lp = htonl(tw->t_recent); 1945 optp += TCPOLEN_TSTAMP_APPA; 1946 } 1947 1948 optlen = optp - (u_int8_t *)(th + 1); 1949 1950 m->m_len = hdrlen + optlen; 1951 m->m_pkthdr.len = m->m_len; 1952 1953 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1954 1955 th->th_seq = htonl(tw->snd_nxt); 1956 th->th_ack = htonl(tw->rcv_nxt); 1957 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1958 th->th_flags = flags; 1959 th->th_win = htons(tw->last_win); 1960 1961#ifdef INET6 1962 if (isipv6) { 1963 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1964 sizeof(struct tcphdr) + optlen); 1965 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1966 error = ip6_output(m, inp->in6p_outputopts, NULL, 1967 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1968 } else 1969#endif 1970 { 1971 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1972 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1973 m->m_pkthdr.csum_flags = CSUM_TCP; 1974 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1975 ip->ip_len = m->m_pkthdr.len; 1976 if (path_mtu_discovery) 1977 ip->ip_off |= IP_DF; 1978 error = ip_output(m, inp->inp_options, NULL, 1979 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1980 NULL, inp); 1981 } 1982 if (flags & TH_ACK) 1983 tcpstat.tcps_sndacks++; 1984 else 1985 tcpstat.tcps_sndctrl++; 1986 tcpstat.tcps_sndtotal++; 1987 return (error); 1988} 1989 1990/* 1991 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1992 * 1993 * This code attempts to calculate the bandwidth-delay product as a 1994 * means of determining the optimal window size to maximize bandwidth, 1995 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1996 * routers. This code also does a fairly good job keeping RTTs in check 1997 * across slow links like modems. We implement an algorithm which is very 1998 * similar (but not meant to be) TCP/Vegas. The code operates on the 1999 * transmitter side of a TCP connection and so only effects the transmit 2000 * side of the connection. 2001 * 2002 * BACKGROUND: TCP makes no provision for the management of buffer space 2003 * at the end points or at the intermediate routers and switches. A TCP 2004 * stream, whether using NewReno or not, will eventually buffer as 2005 * many packets as it is able and the only reason this typically works is 2006 * due to the fairly small default buffers made available for a connection 2007 * (typicaly 16K or 32K). As machines use larger windows and/or window 2008 * scaling it is now fairly easy for even a single TCP connection to blow-out 2009 * all available buffer space not only on the local interface, but on 2010 * intermediate routers and switches as well. NewReno makes a misguided 2011 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2012 * then backing off, then steadily increasing the window again until another 2013 * failure occurs, ad-infinitum. This results in terrible oscillation that 2014 * is only made worse as network loads increase and the idea of intentionally 2015 * blowing out network buffers is, frankly, a terrible way to manage network 2016 * resources. 2017 * 2018 * It is far better to limit the transmit window prior to the failure 2019 * condition being achieved. There are two general ways to do this: First 2020 * you can 'scan' through different transmit window sizes and locate the 2021 * point where the RTT stops increasing, indicating that you have filled the 2022 * pipe, then scan backwards until you note that RTT stops decreasing, then 2023 * repeat ad-infinitum. This method works in principle but has severe 2024 * implementation issues due to RTT variances, timer granularity, and 2025 * instability in the algorithm which can lead to many false positives and 2026 * create oscillations as well as interact badly with other TCP streams 2027 * implementing the same algorithm. 2028 * 2029 * The second method is to limit the window to the bandwidth delay product 2030 * of the link. This is the method we implement. RTT variances and our 2031 * own manipulation of the congestion window, bwnd, can potentially 2032 * destabilize the algorithm. For this reason we have to stabilize the 2033 * elements used to calculate the window. We do this by using the minimum 2034 * observed RTT, the long term average of the observed bandwidth, and 2035 * by adding two segments worth of slop. It isn't perfect but it is able 2036 * to react to changing conditions and gives us a very stable basis on 2037 * which to extend the algorithm. 2038 */ 2039void 2040tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2041{ 2042 u_long bw; 2043 u_long bwnd; 2044 int save_ticks; 2045 2046 INP_LOCK_ASSERT(tp->t_inpcb); 2047 2048 /* 2049 * If inflight_enable is disabled in the middle of a tcp connection, 2050 * make sure snd_bwnd is effectively disabled. 2051 */ 2052 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2053 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2054 tp->snd_bandwidth = 0; 2055 return; 2056 } 2057 2058 /* 2059 * Figure out the bandwidth. Due to the tick granularity this 2060 * is a very rough number and it MUST be averaged over a fairly 2061 * long period of time. XXX we need to take into account a link 2062 * that is not using all available bandwidth, but for now our 2063 * slop will ramp us up if this case occurs and the bandwidth later 2064 * increases. 2065 * 2066 * Note: if ticks rollover 'bw' may wind up negative. We must 2067 * effectively reset t_bw_rtttime for this case. 2068 */ 2069 save_ticks = ticks; 2070 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2071 return; 2072 2073 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2074 (save_ticks - tp->t_bw_rtttime); 2075 tp->t_bw_rtttime = save_ticks; 2076 tp->t_bw_rtseq = ack_seq; 2077 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2078 return; 2079 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2080 2081 tp->snd_bandwidth = bw; 2082 2083 /* 2084 * Calculate the semi-static bandwidth delay product, plus two maximal 2085 * segments. The additional slop puts us squarely in the sweet 2086 * spot and also handles the bandwidth run-up case and stabilization. 2087 * Without the slop we could be locking ourselves into a lower 2088 * bandwidth. 2089 * 2090 * Situations Handled: 2091 * (1) Prevents over-queueing of packets on LANs, especially on 2092 * high speed LANs, allowing larger TCP buffers to be 2093 * specified, and also does a good job preventing 2094 * over-queueing of packets over choke points like modems 2095 * (at least for the transmit side). 2096 * 2097 * (2) Is able to handle changing network loads (bandwidth 2098 * drops so bwnd drops, bandwidth increases so bwnd 2099 * increases). 2100 * 2101 * (3) Theoretically should stabilize in the face of multiple 2102 * connections implementing the same algorithm (this may need 2103 * a little work). 2104 * 2105 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2106 * be adjusted with a sysctl but typically only needs to be 2107 * on very slow connections. A value no smaller then 5 2108 * should be used, but only reduce this default if you have 2109 * no other choice. 2110 */ 2111#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2112 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2113#undef USERTT 2114 2115 if (tcp_inflight_debug > 0) { 2116 static int ltime; 2117 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2118 ltime = ticks; 2119 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2120 tp, 2121 bw, 2122 tp->t_rttbest, 2123 tp->t_srtt, 2124 bwnd 2125 ); 2126 } 2127 } 2128 if ((long)bwnd < tcp_inflight_min) 2129 bwnd = tcp_inflight_min; 2130 if (bwnd > tcp_inflight_max) 2131 bwnd = tcp_inflight_max; 2132 if ((long)bwnd < tp->t_maxseg * 2) 2133 bwnd = tp->t_maxseg * 2; 2134 tp->snd_bwnd = bwnd; 2135} 2136 2137#ifdef TCP_SIGNATURE 2138/* 2139 * Callback function invoked by m_apply() to digest TCP segment data 2140 * contained within an mbuf chain. 2141 */ 2142static int 2143tcp_signature_apply(void *fstate, void *data, u_int len) 2144{ 2145 2146 MD5Update(fstate, (u_char *)data, len); 2147 return (0); 2148} 2149 2150/* 2151 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2152 * 2153 * Parameters: 2154 * m pointer to head of mbuf chain 2155 * off0 offset to TCP header within the mbuf chain 2156 * len length of TCP segment data, excluding options 2157 * optlen length of TCP segment options 2158 * buf pointer to storage for computed MD5 digest 2159 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2160 * 2161 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2162 * When called from tcp_input(), we can be sure that th_sum has been 2163 * zeroed out and verified already. 2164 * 2165 * This function is for IPv4 use only. Calling this function with an 2166 * IPv6 packet in the mbuf chain will yield undefined results. 2167 * 2168 * Return 0 if successful, otherwise return -1. 2169 * 2170 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2171 * search with the destination IP address, and a 'magic SPI' to be 2172 * determined by the application. This is hardcoded elsewhere to 1179 2173 * right now. Another branch of this code exists which uses the SPD to 2174 * specify per-application flows but it is unstable. 2175 */ 2176int 2177tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2178 u_char *buf, u_int direction) 2179{ 2180 union sockaddr_union dst; 2181 struct ippseudo ippseudo; 2182 MD5_CTX ctx; 2183 int doff; 2184 struct ip *ip; 2185 struct ipovly *ipovly; 2186 struct secasvar *sav; 2187 struct tcphdr *th; 2188 u_short savecsum; 2189 2190 KASSERT(m != NULL, ("NULL mbuf chain")); 2191 KASSERT(buf != NULL, ("NULL signature pointer")); 2192 2193 /* Extract the destination from the IP header in the mbuf. */ 2194 ip = mtod(m, struct ip *); 2195 bzero(&dst, sizeof(union sockaddr_union)); 2196 dst.sa.sa_len = sizeof(struct sockaddr_in); 2197 dst.sa.sa_family = AF_INET; 2198 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2199 ip->ip_src : ip->ip_dst; 2200 2201 /* Look up an SADB entry which matches the address of the peer. */ 2202 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2203 if (sav == NULL) { 2204 printf("%s: SADB lookup failed for %s\n", __func__, 2205 inet_ntoa(dst.sin.sin_addr)); 2206 return (EINVAL); 2207 } 2208 2209 MD5Init(&ctx); 2210 ipovly = (struct ipovly *)ip; 2211 th = (struct tcphdr *)((u_char *)ip + off0); 2212 doff = off0 + sizeof(struct tcphdr) + optlen; 2213 2214 /* 2215 * Step 1: Update MD5 hash with IP pseudo-header. 2216 * 2217 * XXX The ippseudo header MUST be digested in network byte order, 2218 * or else we'll fail the regression test. Assume all fields we've 2219 * been doing arithmetic on have been in host byte order. 2220 * XXX One cannot depend on ipovly->ih_len here. When called from 2221 * tcp_output(), the underlying ip_len member has not yet been set. 2222 */ 2223 ippseudo.ippseudo_src = ipovly->ih_src; 2224 ippseudo.ippseudo_dst = ipovly->ih_dst; 2225 ippseudo.ippseudo_pad = 0; 2226 ippseudo.ippseudo_p = IPPROTO_TCP; 2227 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2228 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2229 2230 /* 2231 * Step 2: Update MD5 hash with TCP header, excluding options. 2232 * The TCP checksum must be set to zero. 2233 */ 2234 savecsum = th->th_sum; 2235 th->th_sum = 0; 2236 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2237 th->th_sum = savecsum; 2238 2239 /* 2240 * Step 3: Update MD5 hash with TCP segment data. 2241 * Use m_apply() to avoid an early m_pullup(). 2242 */ 2243 if (len > 0) 2244 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2245 2246 /* 2247 * Step 4: Update MD5 hash with shared secret. 2248 */ 2249 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2250 MD5Final(buf, &ctx); 2251 2252 key_sa_recordxfer(sav, m); 2253 KEY_FREESAV(&sav); 2254 return (0); 2255} 2256#endif /* TCP_SIGNATURE */ 2257 2258static int 2259sysctl_drop(SYSCTL_HANDLER_ARGS) 2260{ 2261 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2262 struct sockaddr_storage addrs[2]; 2263 struct inpcb *inp; 2264 struct tcpcb *tp; 2265 struct tcptw *tw; 2266 struct sockaddr_in *fin, *lin; 2267#ifdef INET6 2268 struct sockaddr_in6 *fin6, *lin6; 2269 struct in6_addr f6, l6; 2270#endif 2271 int error; 2272 2273 inp = NULL; 2274 fin = lin = NULL; 2275#ifdef INET6 2276 fin6 = lin6 = NULL; 2277#endif 2278 error = 0; 2279 2280 if (req->oldptr != NULL || req->oldlen != 0) 2281 return (EINVAL); 2282 if (req->newptr == NULL) 2283 return (EPERM); 2284 if (req->newlen < sizeof(addrs)) 2285 return (ENOMEM); 2286 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2287 if (error) 2288 return (error); 2289 2290 switch (addrs[0].ss_family) { 2291#ifdef INET6 2292 case AF_INET6: 2293 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2294 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2295 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2296 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2297 return (EINVAL); 2298 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2299 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2300 return (EINVAL); 2301 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2302 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2303 fin = (struct sockaddr_in *)&addrs[0]; 2304 lin = (struct sockaddr_in *)&addrs[1]; 2305 break; 2306 } 2307 error = sa6_embedscope(fin6, ip6_use_defzone); 2308 if (error) 2309 return (error); 2310 error = sa6_embedscope(lin6, ip6_use_defzone); 2311 if (error) 2312 return (error); 2313 break; 2314#endif 2315 case AF_INET: 2316 fin = (struct sockaddr_in *)&addrs[0]; 2317 lin = (struct sockaddr_in *)&addrs[1]; 2318 if (fin->sin_len != sizeof(struct sockaddr_in) || 2319 lin->sin_len != sizeof(struct sockaddr_in)) 2320 return (EINVAL); 2321 break; 2322 default: 2323 return (EINVAL); 2324 } 2325 INP_INFO_WLOCK(&tcbinfo); 2326 switch (addrs[0].ss_family) { 2327#ifdef INET6 2328 case AF_INET6: 2329 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2330 &l6, lin6->sin6_port, 0, NULL); 2331 break; 2332#endif 2333 case AF_INET: 2334 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2335 lin->sin_addr, lin->sin_port, 0, NULL); 2336 break; 2337 } 2338 if (inp != NULL) { 2339 INP_LOCK(inp); 2340 if (inp->inp_vflag & INP_TIMEWAIT) { 2341 tw = intotw(inp); 2342 tcp_twclose(tw, 0); 2343 } else if (!(inp->inp_vflag & INP_DROPPED) && 2344 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2345 tp = intotcpcb(inp); 2346 tcp_drop(tp, ECONNABORTED); 2347 } 2348 INP_UNLOCK(inp); 2349 } else 2350 error = ESRCH; 2351 INP_INFO_WUNLOCK(&tcbinfo); 2352 return (error); 2353} 2354 2355SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2356 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2357 0, sysctl_drop, "", "Drop TCP connection"); 2358