tcp_timewait.c revision 157427
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 157427 2006-04-03 11:57:12Z 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 * Close a TCP control block: 773 * discard all space held by the tcp 774 * discard internet protocol block 775 * wake up any sleepers 776 */ 777struct tcpcb * 778tcp_close(tp) 779 struct tcpcb *tp; 780{ 781 struct inpcb *inp = tp->t_inpcb; 782 struct socket *so; 783 784 INP_INFO_WLOCK_ASSERT(&tcbinfo); 785 INP_LOCK_ASSERT(inp); 786 787 inp->inp_vflag |= INP_DROPPED; 788 789 tcpstat.tcps_closed++; 790 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 791 so = inp->inp_socket; 792 soisdisconnected(so); 793 if (inp->inp_vflag & INP_SOCKREF) { 794 KASSERT(so->so_state & SS_PROTOREF, 795 ("tcp_close: !SS_PROTOREF")); 796 inp->inp_vflag &= ~INP_SOCKREF; 797 tcp_discardcb(tp); 798#ifdef INET6 799 if (inp->inp_vflag & INP_IPV6PROTO) { 800 in6_pcbdetach(inp); 801 in6_pcbfree(inp); 802 } else { 803#endif 804 in_pcbdetach(inp); 805 in_pcbfree(inp); 806#ifdef INET6 807 } 808#endif 809 ACCEPT_LOCK(); 810 SOCK_LOCK(so); 811 so->so_state &= ~SS_PROTOREF; 812 sofree(so); 813 return (NULL); 814 } 815 return (tp); 816} 817 818void 819tcp_drain() 820{ 821 if (do_tcpdrain) 822 { 823 struct inpcb *inpb; 824 struct tcpcb *tcpb; 825 struct tseg_qent *te; 826 827 /* 828 * Walk the tcpbs, if existing, and flush the reassembly queue, 829 * if there is one... 830 * XXX: The "Net/3" implementation doesn't imply that the TCP 831 * reassembly queue should be flushed, but in a situation 832 * where we're really low on mbufs, this is potentially 833 * usefull. 834 */ 835 INP_INFO_RLOCK(&tcbinfo); 836 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 837 if (inpb->inp_vflag & INP_TIMEWAIT) 838 continue; 839 INP_LOCK(inpb); 840 if ((tcpb = intotcpcb(inpb)) != NULL) { 841 while ((te = LIST_FIRST(&tcpb->t_segq)) 842 != NULL) { 843 LIST_REMOVE(te, tqe_q); 844 m_freem(te->tqe_m); 845 uma_zfree(tcp_reass_zone, te); 846 tcpb->t_segqlen--; 847 tcp_reass_qsize--; 848 } 849 tcp_clean_sackreport(tcpb); 850 } 851 INP_UNLOCK(inpb); 852 } 853 INP_INFO_RUNLOCK(&tcbinfo); 854 } 855} 856 857/* 858 * Notify a tcp user of an asynchronous error; 859 * store error as soft error, but wake up user 860 * (for now, won't do anything until can select for soft error). 861 * 862 * Do not wake up user since there currently is no mechanism for 863 * reporting soft errors (yet - a kqueue filter may be added). 864 */ 865static struct inpcb * 866tcp_notify(inp, error) 867 struct inpcb *inp; 868 int error; 869{ 870 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 871 872 INP_INFO_WLOCK_ASSERT(&tcbinfo); 873 INP_LOCK_ASSERT(inp); 874 875 /* 876 * Ignore some errors if we are hooked up. 877 * If connection hasn't completed, has retransmitted several times, 878 * and receives a second error, give up now. This is better 879 * than waiting a long time to establish a connection that 880 * can never complete. 881 */ 882 if (tp->t_state == TCPS_ESTABLISHED && 883 (error == EHOSTUNREACH || error == ENETUNREACH || 884 error == EHOSTDOWN)) { 885 return (inp); 886 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 887 tp->t_softerror) { 888 tp = tcp_drop(tp, error); 889 if (tp != NULL) 890 return (inp); 891 else 892 return (NULL); 893 } else { 894 tp->t_softerror = error; 895 return (inp); 896 } 897#if 0 898 wakeup( &so->so_timeo); 899 sorwakeup(so); 900 sowwakeup(so); 901#endif 902} 903 904static int 905tcp_pcblist(SYSCTL_HANDLER_ARGS) 906{ 907 int error, i, n; 908 struct inpcb *inp, **inp_list; 909 inp_gen_t gencnt; 910 struct xinpgen xig; 911 912 /* 913 * The process of preparing the TCB list is too time-consuming and 914 * resource-intensive to repeat twice on every request. 915 */ 916 if (req->oldptr == NULL) { 917 n = tcbinfo.ipi_count; 918 req->oldidx = 2 * (sizeof xig) 919 + (n + n/8) * sizeof(struct xtcpcb); 920 return (0); 921 } 922 923 if (req->newptr != NULL) 924 return (EPERM); 925 926 /* 927 * OK, now we're committed to doing something. 928 */ 929 INP_INFO_RLOCK(&tcbinfo); 930 gencnt = tcbinfo.ipi_gencnt; 931 n = tcbinfo.ipi_count; 932 INP_INFO_RUNLOCK(&tcbinfo); 933 934 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 935 + n * sizeof(struct xtcpcb)); 936 if (error != 0) 937 return (error); 938 939 xig.xig_len = sizeof xig; 940 xig.xig_count = n; 941 xig.xig_gen = gencnt; 942 xig.xig_sogen = so_gencnt; 943 error = SYSCTL_OUT(req, &xig, sizeof xig); 944 if (error) 945 return (error); 946 947 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 948 if (inp_list == NULL) 949 return (ENOMEM); 950 951 INP_INFO_RLOCK(&tcbinfo); 952 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 953 inp = LIST_NEXT(inp, inp_list)) { 954 INP_LOCK(inp); 955 if (inp->inp_gencnt <= gencnt) { 956 /* 957 * XXX: This use of cr_cansee(), introduced with 958 * TCP state changes, is not quite right, but for 959 * now, better than nothing. 960 */ 961 if (inp->inp_vflag & INP_TIMEWAIT) 962 error = cr_cansee(req->td->td_ucred, 963 intotw(inp)->tw_cred); 964 else 965 error = cr_canseesocket(req->td->td_ucred, 966 inp->inp_socket); 967 if (error == 0) 968 inp_list[i++] = inp; 969 } 970 INP_UNLOCK(inp); 971 } 972 INP_INFO_RUNLOCK(&tcbinfo); 973 n = i; 974 975 error = 0; 976 for (i = 0; i < n; i++) { 977 inp = inp_list[i]; 978 if (inp->inp_gencnt <= gencnt) { 979 struct xtcpcb xt; 980 caddr_t inp_ppcb; 981 982 bzero(&xt, sizeof(xt)); 983 xt.xt_len = sizeof xt; 984 /* XXX should avoid extra copy */ 985 bcopy(inp, &xt.xt_inp, sizeof *inp); 986 inp_ppcb = inp->inp_ppcb; 987 if (inp_ppcb == NULL) 988 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 989 else if (inp->inp_vflag & INP_TIMEWAIT) { 990 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 991 xt.xt_tp.t_state = TCPS_TIME_WAIT; 992 } else 993 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 994 if (inp->inp_socket != NULL) 995 sotoxsocket(inp->inp_socket, &xt.xt_socket); 996 else { 997 bzero(&xt.xt_socket, sizeof xt.xt_socket); 998 xt.xt_socket.xso_protocol = IPPROTO_TCP; 999 } 1000 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1001 error = SYSCTL_OUT(req, &xt, sizeof xt); 1002 } 1003 } 1004 if (!error) { 1005 /* 1006 * Give the user an updated idea of our state. 1007 * If the generation differs from what we told 1008 * her before, she knows that something happened 1009 * while we were processing this request, and it 1010 * might be necessary to retry. 1011 */ 1012 INP_INFO_RLOCK(&tcbinfo); 1013 xig.xig_gen = tcbinfo.ipi_gencnt; 1014 xig.xig_sogen = so_gencnt; 1015 xig.xig_count = tcbinfo.ipi_count; 1016 INP_INFO_RUNLOCK(&tcbinfo); 1017 error = SYSCTL_OUT(req, &xig, sizeof xig); 1018 } 1019 free(inp_list, M_TEMP); 1020 return (error); 1021} 1022 1023SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 1024 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1025 1026static int 1027tcp_getcred(SYSCTL_HANDLER_ARGS) 1028{ 1029 struct xucred xuc; 1030 struct sockaddr_in addrs[2]; 1031 struct inpcb *inp; 1032 int error; 1033 1034 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1035 if (error) 1036 return (error); 1037 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1038 if (error) 1039 return (error); 1040 INP_INFO_RLOCK(&tcbinfo); 1041 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1042 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1043 if (inp == NULL) { 1044 error = ENOENT; 1045 goto outunlocked; 1046 } 1047 INP_LOCK(inp); 1048 if (inp->inp_socket == NULL) { 1049 error = ENOENT; 1050 goto out; 1051 } 1052 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1053 if (error) 1054 goto out; 1055 cru2x(inp->inp_socket->so_cred, &xuc); 1056out: 1057 INP_UNLOCK(inp); 1058outunlocked: 1059 INP_INFO_RUNLOCK(&tcbinfo); 1060 if (error == 0) 1061 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1062 return (error); 1063} 1064 1065SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1066 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1067 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1068 1069#ifdef INET6 1070static int 1071tcp6_getcred(SYSCTL_HANDLER_ARGS) 1072{ 1073 struct xucred xuc; 1074 struct sockaddr_in6 addrs[2]; 1075 struct inpcb *inp; 1076 int error, mapped = 0; 1077 1078 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1079 if (error) 1080 return (error); 1081 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1082 if (error) 1083 return (error); 1084 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1085 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1086 return (error); 1087 } 1088 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1089 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1090 mapped = 1; 1091 else 1092 return (EINVAL); 1093 } 1094 1095 INP_INFO_RLOCK(&tcbinfo); 1096 if (mapped == 1) 1097 inp = in_pcblookup_hash(&tcbinfo, 1098 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1099 addrs[1].sin6_port, 1100 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1101 addrs[0].sin6_port, 1102 0, NULL); 1103 else 1104 inp = in6_pcblookup_hash(&tcbinfo, 1105 &addrs[1].sin6_addr, addrs[1].sin6_port, 1106 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 1107 if (inp == NULL) { 1108 error = ENOENT; 1109 goto outunlocked; 1110 } 1111 INP_LOCK(inp); 1112 if (inp->inp_socket == NULL) { 1113 error = ENOENT; 1114 goto out; 1115 } 1116 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1117 if (error) 1118 goto out; 1119 cru2x(inp->inp_socket->so_cred, &xuc); 1120out: 1121 INP_UNLOCK(inp); 1122outunlocked: 1123 INP_INFO_RUNLOCK(&tcbinfo); 1124 if (error == 0) 1125 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1126 return (error); 1127} 1128 1129SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1130 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1131 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1132#endif 1133 1134 1135void 1136tcp_ctlinput(cmd, sa, vip) 1137 int cmd; 1138 struct sockaddr *sa; 1139 void *vip; 1140{ 1141 struct ip *ip = vip; 1142 struct tcphdr *th; 1143 struct in_addr faddr; 1144 struct inpcb *inp; 1145 struct tcpcb *tp; 1146 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1147 struct icmp *icp; 1148 struct in_conninfo inc; 1149 tcp_seq icmp_tcp_seq; 1150 int mtu; 1151 1152 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1153 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1154 return; 1155 1156 if (cmd == PRC_MSGSIZE) 1157 notify = tcp_mtudisc; 1158 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1159 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1160 notify = tcp_drop_syn_sent; 1161 /* 1162 * Redirects don't need to be handled up here. 1163 */ 1164 else if (PRC_IS_REDIRECT(cmd)) 1165 return; 1166 /* 1167 * Source quench is depreciated. 1168 */ 1169 else if (cmd == PRC_QUENCH) 1170 return; 1171 /* 1172 * Hostdead is ugly because it goes linearly through all PCBs. 1173 * XXX: We never get this from ICMP, otherwise it makes an 1174 * excellent DoS attack on machines with many connections. 1175 */ 1176 else if (cmd == PRC_HOSTDEAD) 1177 ip = NULL; 1178 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1179 return; 1180 if (ip != NULL) { 1181 icp = (struct icmp *)((caddr_t)ip 1182 - offsetof(struct icmp, icmp_ip)); 1183 th = (struct tcphdr *)((caddr_t)ip 1184 + (ip->ip_hl << 2)); 1185 INP_INFO_WLOCK(&tcbinfo); 1186 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1187 ip->ip_src, th->th_sport, 0, NULL); 1188 if (inp != NULL) { 1189 INP_LOCK(inp); 1190 if (inp->inp_socket != NULL) { 1191 icmp_tcp_seq = htonl(th->th_seq); 1192 tp = intotcpcb(inp); 1193 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1194 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1195 if (cmd == PRC_MSGSIZE) { 1196 /* 1197 * MTU discovery: 1198 * If we got a needfrag set the MTU 1199 * in the route to the suggested new 1200 * value (if given) and then notify. 1201 */ 1202 bzero(&inc, sizeof(inc)); 1203 inc.inc_flags = 0; /* IPv4 */ 1204 inc.inc_faddr = faddr; 1205 1206 mtu = ntohs(icp->icmp_nextmtu); 1207 /* 1208 * If no alternative MTU was 1209 * proposed, try the next smaller 1210 * one. ip->ip_len has already 1211 * been swapped in icmp_input(). 1212 */ 1213 if (!mtu) 1214 mtu = ip_next_mtu(ip->ip_len, 1215 1); 1216 if (mtu < max(296, (tcp_minmss) 1217 + sizeof(struct tcpiphdr))) 1218 mtu = 0; 1219 if (!mtu) 1220 mtu = tcp_mssdflt 1221 + sizeof(struct tcpiphdr); 1222 /* 1223 * Only cache the the MTU if it 1224 * is smaller than the interface 1225 * or route MTU. tcp_mtudisc() 1226 * will do right thing by itself. 1227 */ 1228 if (mtu <= tcp_maxmtu(&inc)) 1229 tcp_hc_updatemtu(&inc, mtu); 1230 } 1231 1232 inp = (*notify)(inp, inetctlerrmap[cmd]); 1233 } 1234 } 1235 if (inp != NULL) 1236 INP_UNLOCK(inp); 1237 } else { 1238 inc.inc_fport = th->th_dport; 1239 inc.inc_lport = th->th_sport; 1240 inc.inc_faddr = faddr; 1241 inc.inc_laddr = ip->ip_src; 1242#ifdef INET6 1243 inc.inc_isipv6 = 0; 1244#endif 1245 syncache_unreach(&inc, th); 1246 } 1247 INP_INFO_WUNLOCK(&tcbinfo); 1248 } else 1249 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1250} 1251 1252#ifdef INET6 1253void 1254tcp6_ctlinput(cmd, sa, d) 1255 int cmd; 1256 struct sockaddr *sa; 1257 void *d; 1258{ 1259 struct tcphdr th; 1260 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1261 struct ip6_hdr *ip6; 1262 struct mbuf *m; 1263 struct ip6ctlparam *ip6cp = NULL; 1264 const struct sockaddr_in6 *sa6_src = NULL; 1265 int off; 1266 struct tcp_portonly { 1267 u_int16_t th_sport; 1268 u_int16_t th_dport; 1269 } *thp; 1270 1271 if (sa->sa_family != AF_INET6 || 1272 sa->sa_len != sizeof(struct sockaddr_in6)) 1273 return; 1274 1275 if (cmd == PRC_MSGSIZE) 1276 notify = tcp_mtudisc; 1277 else if (!PRC_IS_REDIRECT(cmd) && 1278 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1279 return; 1280 /* Source quench is depreciated. */ 1281 else if (cmd == PRC_QUENCH) 1282 return; 1283 1284 /* if the parameter is from icmp6, decode it. */ 1285 if (d != NULL) { 1286 ip6cp = (struct ip6ctlparam *)d; 1287 m = ip6cp->ip6c_m; 1288 ip6 = ip6cp->ip6c_ip6; 1289 off = ip6cp->ip6c_off; 1290 sa6_src = ip6cp->ip6c_src; 1291 } else { 1292 m = NULL; 1293 ip6 = NULL; 1294 off = 0; /* fool gcc */ 1295 sa6_src = &sa6_any; 1296 } 1297 1298 if (ip6 != NULL) { 1299 struct in_conninfo inc; 1300 /* 1301 * XXX: We assume that when IPV6 is non NULL, 1302 * M and OFF are valid. 1303 */ 1304 1305 /* check if we can safely examine src and dst ports */ 1306 if (m->m_pkthdr.len < off + sizeof(*thp)) 1307 return; 1308 1309 bzero(&th, sizeof(th)); 1310 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1311 1312 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1313 (struct sockaddr *)ip6cp->ip6c_src, 1314 th.th_sport, cmd, NULL, notify); 1315 1316 inc.inc_fport = th.th_dport; 1317 inc.inc_lport = th.th_sport; 1318 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1319 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1320 inc.inc_isipv6 = 1; 1321 INP_INFO_WLOCK(&tcbinfo); 1322 syncache_unreach(&inc, &th); 1323 INP_INFO_WUNLOCK(&tcbinfo); 1324 } else 1325 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1326 0, cmd, NULL, notify); 1327} 1328#endif /* INET6 */ 1329 1330 1331/* 1332 * Following is where TCP initial sequence number generation occurs. 1333 * 1334 * There are two places where we must use initial sequence numbers: 1335 * 1. In SYN-ACK packets. 1336 * 2. In SYN packets. 1337 * 1338 * All ISNs for SYN-ACK packets are generated by the syncache. See 1339 * tcp_syncache.c for details. 1340 * 1341 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1342 * depends on this property. In addition, these ISNs should be 1343 * unguessable so as to prevent connection hijacking. To satisfy 1344 * the requirements of this situation, the algorithm outlined in 1345 * RFC 1948 is used, with only small modifications. 1346 * 1347 * Implementation details: 1348 * 1349 * Time is based off the system timer, and is corrected so that it 1350 * increases by one megabyte per second. This allows for proper 1351 * recycling on high speed LANs while still leaving over an hour 1352 * before rollover. 1353 * 1354 * As reading the *exact* system time is too expensive to be done 1355 * whenever setting up a TCP connection, we increment the time 1356 * offset in two ways. First, a small random positive increment 1357 * is added to isn_offset for each connection that is set up. 1358 * Second, the function tcp_isn_tick fires once per clock tick 1359 * and increments isn_offset as necessary so that sequence numbers 1360 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1361 * random positive increments serve only to ensure that the same 1362 * exact sequence number is never sent out twice (as could otherwise 1363 * happen when a port is recycled in less than the system tick 1364 * interval.) 1365 * 1366 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1367 * between seeding of isn_secret. This is normally set to zero, 1368 * as reseeding should not be necessary. 1369 * 1370 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1371 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1372 * general, this means holding an exclusive (write) lock. 1373 */ 1374 1375#define ISN_BYTES_PER_SECOND 1048576 1376#define ISN_STATIC_INCREMENT 4096 1377#define ISN_RANDOM_INCREMENT (4096 - 1) 1378 1379static u_char isn_secret[32]; 1380static int isn_last_reseed; 1381static u_int32_t isn_offset, isn_offset_old; 1382static MD5_CTX isn_ctx; 1383 1384tcp_seq 1385tcp_new_isn(tp) 1386 struct tcpcb *tp; 1387{ 1388 u_int32_t md5_buffer[4]; 1389 tcp_seq new_isn; 1390 1391 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1392 INP_LOCK_ASSERT(tp->t_inpcb); 1393 1394 /* Seed if this is the first use, reseed if requested. */ 1395 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1396 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1397 < (u_int)ticks))) { 1398 read_random(&isn_secret, sizeof(isn_secret)); 1399 isn_last_reseed = ticks; 1400 } 1401 1402 /* Compute the md5 hash and return the ISN. */ 1403 MD5Init(&isn_ctx); 1404 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1405 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1406#ifdef INET6 1407 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1408 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1409 sizeof(struct in6_addr)); 1410 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1411 sizeof(struct in6_addr)); 1412 } else 1413#endif 1414 { 1415 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1416 sizeof(struct in_addr)); 1417 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1418 sizeof(struct in_addr)); 1419 } 1420 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1421 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1422 new_isn = (tcp_seq) md5_buffer[0]; 1423 isn_offset += ISN_STATIC_INCREMENT + 1424 (arc4random() & ISN_RANDOM_INCREMENT); 1425 new_isn += isn_offset; 1426 return (new_isn); 1427} 1428 1429/* 1430 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1431 * to keep time flowing at a relatively constant rate. If the random 1432 * increments have already pushed us past the projected offset, do nothing. 1433 */ 1434static void 1435tcp_isn_tick(xtp) 1436 void *xtp; 1437{ 1438 u_int32_t projected_offset; 1439 1440 INP_INFO_WLOCK(&tcbinfo); 1441 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1442 1443 if (projected_offset > isn_offset) 1444 isn_offset = projected_offset; 1445 1446 isn_offset_old = isn_offset; 1447 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1448 INP_INFO_WUNLOCK(&tcbinfo); 1449} 1450 1451/* 1452 * When a specific ICMP unreachable message is received and the 1453 * connection state is SYN-SENT, drop the connection. This behavior 1454 * is controlled by the icmp_may_rst sysctl. 1455 */ 1456struct inpcb * 1457tcp_drop_syn_sent(inp, errno) 1458 struct inpcb *inp; 1459 int errno; 1460{ 1461 struct tcpcb *tp = intotcpcb(inp); 1462 1463 INP_INFO_WLOCK_ASSERT(&tcbinfo); 1464 INP_LOCK_ASSERT(inp); 1465 1466 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) { 1467 tp = tcp_drop(tp, errno); 1468 if (tp != NULL) 1469 return (inp); 1470 else 1471 return (NULL); 1472 } 1473 return (inp); 1474} 1475 1476/* 1477 * When `need fragmentation' ICMP is received, update our idea of the MSS 1478 * based on the new value in the route. Also nudge TCP to send something, 1479 * since we know the packet we just sent was dropped. 1480 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1481 */ 1482struct inpcb * 1483tcp_mtudisc(inp, errno) 1484 struct inpcb *inp; 1485 int errno; 1486{ 1487 struct tcpcb *tp = intotcpcb(inp); 1488 struct socket *so = inp->inp_socket; 1489 u_int maxmtu; 1490 u_int romtu; 1491 int mss; 1492#ifdef INET6 1493 int isipv6; 1494#endif /* INET6 */ 1495 1496 INP_LOCK_ASSERT(inp); 1497 if (tp != NULL) { 1498#ifdef INET6 1499 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1500#endif 1501 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1502 romtu = 1503#ifdef INET6 1504 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1505#endif /* INET6 */ 1506 tcp_maxmtu(&inp->inp_inc); 1507 if (!maxmtu) 1508 maxmtu = romtu; 1509 else 1510 maxmtu = min(maxmtu, romtu); 1511 if (!maxmtu) { 1512 tp->t_maxopd = tp->t_maxseg = 1513#ifdef INET6 1514 isipv6 ? tcp_v6mssdflt : 1515#endif /* INET6 */ 1516 tcp_mssdflt; 1517 return (inp); 1518 } 1519 mss = maxmtu - 1520#ifdef INET6 1521 (isipv6 ? 1522 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1523#endif /* INET6 */ 1524 sizeof(struct tcpiphdr) 1525#ifdef INET6 1526 ) 1527#endif /* INET6 */ 1528 ; 1529 1530 /* 1531 * XXX - The above conditional probably violates the TCP 1532 * spec. The problem is that, since we don't know the 1533 * other end's MSS, we are supposed to use a conservative 1534 * default. But, if we do that, then MTU discovery will 1535 * never actually take place, because the conservative 1536 * default is much less than the MTUs typically seen 1537 * on the Internet today. For the moment, we'll sweep 1538 * this under the carpet. 1539 * 1540 * The conservative default might not actually be a problem 1541 * if the only case this occurs is when sending an initial 1542 * SYN with options and data to a host we've never talked 1543 * to before. Then, they will reply with an MSS value which 1544 * will get recorded and the new parameters should get 1545 * recomputed. For Further Study. 1546 */ 1547 if (tp->t_maxopd <= mss) 1548 return (inp); 1549 tp->t_maxopd = mss; 1550 1551 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1552 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1553 mss -= TCPOLEN_TSTAMP_APPA; 1554#if (MCLBYTES & (MCLBYTES - 1)) == 0 1555 if (mss > MCLBYTES) 1556 mss &= ~(MCLBYTES-1); 1557#else 1558 if (mss > MCLBYTES) 1559 mss = mss / MCLBYTES * MCLBYTES; 1560#endif 1561 if (so->so_snd.sb_hiwat < mss) 1562 mss = so->so_snd.sb_hiwat; 1563 1564 tp->t_maxseg = mss; 1565 1566 tcpstat.tcps_mturesent++; 1567 tp->t_rtttime = 0; 1568 tp->snd_nxt = tp->snd_una; 1569 tcp_output(tp); 1570 } 1571 return (inp); 1572} 1573 1574/* 1575 * Look-up the routing entry to the peer of this inpcb. If no route 1576 * is found and it cannot be allocated, then return NULL. This routine 1577 * is called by TCP routines that access the rmx structure and by tcp_mss 1578 * to get the interface MTU. 1579 */ 1580u_long 1581tcp_maxmtu(inc) 1582 struct in_conninfo *inc; 1583{ 1584 struct route sro; 1585 struct sockaddr_in *dst; 1586 struct ifnet *ifp; 1587 u_long maxmtu = 0; 1588 1589 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1590 1591 bzero(&sro, sizeof(sro)); 1592 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1593 dst = (struct sockaddr_in *)&sro.ro_dst; 1594 dst->sin_family = AF_INET; 1595 dst->sin_len = sizeof(*dst); 1596 dst->sin_addr = inc->inc_faddr; 1597 rtalloc_ign(&sro, RTF_CLONING); 1598 } 1599 if (sro.ro_rt != NULL) { 1600 ifp = sro.ro_rt->rt_ifp; 1601 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1602 maxmtu = ifp->if_mtu; 1603 else 1604 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1605 RTFREE(sro.ro_rt); 1606 } 1607 return (maxmtu); 1608} 1609 1610#ifdef INET6 1611u_long 1612tcp_maxmtu6(inc) 1613 struct in_conninfo *inc; 1614{ 1615 struct route_in6 sro6; 1616 struct ifnet *ifp; 1617 u_long maxmtu = 0; 1618 1619 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1620 1621 bzero(&sro6, sizeof(sro6)); 1622 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1623 sro6.ro_dst.sin6_family = AF_INET6; 1624 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1625 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1626 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1627 } 1628 if (sro6.ro_rt != NULL) { 1629 ifp = sro6.ro_rt->rt_ifp; 1630 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1631 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1632 else 1633 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1634 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1635 RTFREE(sro6.ro_rt); 1636 } 1637 1638 return (maxmtu); 1639} 1640#endif /* INET6 */ 1641 1642#ifdef IPSEC 1643/* compute ESP/AH header size for TCP, including outer IP header. */ 1644size_t 1645ipsec_hdrsiz_tcp(tp) 1646 struct tcpcb *tp; 1647{ 1648 struct inpcb *inp; 1649 struct mbuf *m; 1650 size_t hdrsiz; 1651 struct ip *ip; 1652#ifdef INET6 1653 struct ip6_hdr *ip6; 1654#endif 1655 struct tcphdr *th; 1656 1657 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1658 return (0); 1659 MGETHDR(m, M_DONTWAIT, MT_DATA); 1660 if (!m) 1661 return (0); 1662 1663#ifdef INET6 1664 if ((inp->inp_vflag & INP_IPV6) != 0) { 1665 ip6 = mtod(m, struct ip6_hdr *); 1666 th = (struct tcphdr *)(ip6 + 1); 1667 m->m_pkthdr.len = m->m_len = 1668 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1669 tcpip_fillheaders(inp, ip6, th); 1670 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1671 } else 1672#endif /* INET6 */ 1673 { 1674 ip = mtod(m, struct ip *); 1675 th = (struct tcphdr *)(ip + 1); 1676 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1677 tcpip_fillheaders(inp, ip, th); 1678 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1679 } 1680 1681 m_free(m); 1682 return (hdrsiz); 1683} 1684#endif /*IPSEC*/ 1685 1686/* 1687 * Move a TCP connection into TIME_WAIT state. 1688 * tcbinfo is locked. 1689 * inp is locked, and is unlocked before returning. 1690 */ 1691void 1692tcp_twstart(tp) 1693 struct tcpcb *tp; 1694{ 1695 struct tcptw *tw; 1696 struct inpcb *inp; 1697 int tw_time, acknow; 1698 struct socket *so; 1699 1700 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1701 INP_LOCK_ASSERT(tp->t_inpcb); 1702 1703 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1704 if (tw == NULL) { 1705 tw = tcp_timer_2msl_tw(1); 1706 if (tw == NULL) { 1707 tp = tcp_close(tp); 1708 if (tp != NULL) 1709 INP_UNLOCK(tp->t_inpcb); 1710 return; 1711 } 1712 } 1713 inp = tp->t_inpcb; 1714 tw->tw_inpcb = inp; 1715 1716 /* 1717 * Recover last window size sent. 1718 */ 1719 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1720 1721 /* 1722 * Set t_recent if timestamps are used on the connection. 1723 */ 1724 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1725 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1726 tw->t_recent = tp->ts_recent; 1727 else 1728 tw->t_recent = 0; 1729 1730 tw->snd_nxt = tp->snd_nxt; 1731 tw->rcv_nxt = tp->rcv_nxt; 1732 tw->iss = tp->iss; 1733 tw->irs = tp->irs; 1734 tw->t_starttime = tp->t_starttime; 1735 tw->tw_time = 0; 1736 1737/* XXX 1738 * If this code will 1739 * be used for fin-wait-2 state also, then we may need 1740 * a ts_recent from the last segment. 1741 */ 1742 tw_time = 2 * tcp_msl; 1743 acknow = tp->t_flags & TF_ACKNOW; 1744 1745 /* 1746 * First, discard tcpcb state, which includes stopping its timers and 1747 * freeing it. tcp_discardcb() used to also release the inpcb, but 1748 * that work is now done in the caller. 1749 */ 1750 tcp_discardcb(tp); 1751 so = inp->inp_socket; 1752 SOCK_LOCK(so); 1753 tw->tw_cred = crhold(so->so_cred); 1754 tw->tw_so_options = so->so_options; 1755 SOCK_UNLOCK(so); 1756 if (acknow) 1757 tcp_twrespond(tw, TH_ACK); 1758 inp->inp_ppcb = (caddr_t)tw; 1759 inp->inp_vflag |= INP_TIMEWAIT; 1760 tcp_timer_2msl_reset(tw, tw_time); 1761 1762 /* 1763 * If the inpcb owns the sole reference to the socket, then we can 1764 * detach and free the socket as it is not needed in time wait. 1765 */ 1766 if (inp->inp_vflag & INP_SOCKREF) { 1767 KASSERT(so->so_state & SS_PROTOREF, 1768 ("tcp_twstart: !SS_PROTOREF")); 1769 inp->inp_vflag &= ~INP_SOCKREF; 1770#ifdef INET6 1771 if (inp->inp_vflag & INP_IPV6PROTO) 1772 in6_pcbdetach(inp); 1773 else 1774#endif 1775 in_pcbdetach(inp); 1776 INP_UNLOCK(inp); 1777 ACCEPT_LOCK(); 1778 SOCK_LOCK(so); 1779 so->so_state &= ~SS_PROTOREF; 1780 sofree(so); 1781 } else 1782 INP_UNLOCK(inp); 1783} 1784 1785/* 1786 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1787 * the actual rate is slightly higher due to the addition of 1788 * random positive increments. 1789 * 1790 * Most other new OSes use semi-randomized ISN values, so we 1791 * do not need to worry about them. 1792 */ 1793#define MS_ISN_BYTES_PER_SECOND 250000 1794 1795/* 1796 * Determine if the ISN we will generate has advanced beyond the last 1797 * sequence number used by the previous connection. If so, indicate 1798 * that it is safe to recycle this tw socket by returning 1. 1799 * 1800 * XXXRW: This function should assert the inpcb lock as it does multiple 1801 * non-atomic reads from the tcptw, but is currently called without it from 1802 * in_pcb.c:in_pcblookup_local(). 1803 */ 1804int 1805tcp_twrecycleable(struct tcptw *tw) 1806{ 1807 tcp_seq new_iss = tw->iss; 1808 tcp_seq new_irs = tw->irs; 1809 1810 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1811 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1812 1813 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1814 return (1); 1815 else 1816 return (0); 1817} 1818 1819void 1820tcp_twclose(struct tcptw *tw, int reuse) 1821{ 1822 struct socket *so; 1823 struct inpcb *inp; 1824 1825 /* 1826 * At this point, we are in one of two situations: 1827 * 1828 * (1) We have no socket, just an inpcb<->twtcp pair. Release it all 1829 * after validating. 1830 * 1831 * (2) We have a socket, which we may or may now own the reference 1832 * for. If we own the reference, release all the state after 1833 * validating. If not, leave it for the socket close to clean up. 1834 */ 1835 inp = tw->tw_inpcb; 1836 KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1837 KASSERT(inp->inp_ppcb == (void *)tw, ("tcp_twclose: inp_ppcb != tw")); 1838 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1839 INP_LOCK_ASSERT(inp); 1840 1841 tw->tw_inpcb = NULL; 1842 tcp_timer_2msl_stop(tw); 1843 inp->inp_ppcb = NULL; 1844 inp->inp_vflag |= INP_DROPPED; 1845 1846 so = inp->inp_socket; 1847 if (so != NULL) { 1848 if (inp->inp_vflag & INP_SOCKREF) { 1849 /* 1850 * If a socket is present, and we own the only 1851 * reference, we need to tear down the socket and the 1852 * inpcb. 1853 */ 1854 inp->inp_vflag &= ~INP_SOCKREF; 1855#ifdef INET6 1856 if (inp->inp_vflag & INP_IPV6PROTO) { 1857 in6_pcbdetach(inp); 1858 in6_pcbfree(inp); 1859 } else { 1860 in_pcbdetach(inp); 1861 in_pcbfree(inp); 1862 } 1863#endif 1864 ACCEPT_LOCK(); 1865 SOCK_LOCK(so); 1866 KASSERT(so->so_state & SS_PROTOREF, 1867 ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1868 so->so_state &= ~SS_PROTOREF; 1869 sofree(so); 1870 } else { 1871 /* 1872 * If we don't own the only reference, the socket and 1873 * inpcb need to be left around to be handled by 1874 * tcp_usr_detach() later. 1875 */ 1876 INP_UNLOCK(inp); 1877 } 1878 } else { 1879#ifdef INET6 1880 if (inp->inp_vflag & INP_IPV6PROTO) 1881 in6_pcbfree(inp); 1882 else 1883#endif 1884 in_pcbfree(inp); 1885 } 1886 tcpstat.tcps_closed++; 1887 crfree(tw->tw_cred); 1888 tw->tw_cred = NULL; 1889 if (reuse) 1890 return; 1891 uma_zfree(tcptw_zone, tw); 1892} 1893 1894int 1895tcp_twrespond(struct tcptw *tw, int flags) 1896{ 1897 struct inpcb *inp = tw->tw_inpcb; 1898 struct tcphdr *th; 1899 struct mbuf *m; 1900 struct ip *ip = NULL; 1901 u_int8_t *optp; 1902 u_int hdrlen, optlen; 1903 int error; 1904#ifdef INET6 1905 struct ip6_hdr *ip6 = NULL; 1906 int isipv6 = inp->inp_inc.inc_isipv6; 1907#endif 1908 1909 INP_LOCK_ASSERT(inp); 1910 1911 m = m_gethdr(M_DONTWAIT, MT_DATA); 1912 if (m == NULL) 1913 return (ENOBUFS); 1914 m->m_data += max_linkhdr; 1915 1916#ifdef MAC 1917 mac_create_mbuf_from_inpcb(inp, m); 1918#endif 1919 1920#ifdef INET6 1921 if (isipv6) { 1922 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1923 ip6 = mtod(m, struct ip6_hdr *); 1924 th = (struct tcphdr *)(ip6 + 1); 1925 tcpip_fillheaders(inp, ip6, th); 1926 } else 1927#endif 1928 { 1929 hdrlen = sizeof(struct tcpiphdr); 1930 ip = mtod(m, struct ip *); 1931 th = (struct tcphdr *)(ip + 1); 1932 tcpip_fillheaders(inp, ip, th); 1933 } 1934 optp = (u_int8_t *)(th + 1); 1935 1936 /* 1937 * Send a timestamp and echo-reply if both our side and our peer 1938 * have sent timestamps in our SYN's and this is not a RST. 1939 */ 1940 if (tw->t_recent && flags == TH_ACK) { 1941 u_int32_t *lp = (u_int32_t *)optp; 1942 1943 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1944 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1945 *lp++ = htonl(ticks); 1946 *lp = htonl(tw->t_recent); 1947 optp += TCPOLEN_TSTAMP_APPA; 1948 } 1949 1950 optlen = optp - (u_int8_t *)(th + 1); 1951 1952 m->m_len = hdrlen + optlen; 1953 m->m_pkthdr.len = m->m_len; 1954 1955 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1956 1957 th->th_seq = htonl(tw->snd_nxt); 1958 th->th_ack = htonl(tw->rcv_nxt); 1959 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1960 th->th_flags = flags; 1961 th->th_win = htons(tw->last_win); 1962 1963#ifdef INET6 1964 if (isipv6) { 1965 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1966 sizeof(struct tcphdr) + optlen); 1967 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1968 error = ip6_output(m, inp->in6p_outputopts, NULL, 1969 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1970 } else 1971#endif 1972 { 1973 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1974 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1975 m->m_pkthdr.csum_flags = CSUM_TCP; 1976 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1977 ip->ip_len = m->m_pkthdr.len; 1978 if (path_mtu_discovery) 1979 ip->ip_off |= IP_DF; 1980 error = ip_output(m, inp->inp_options, NULL, 1981 ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1982 NULL, inp); 1983 } 1984 if (flags & TH_ACK) 1985 tcpstat.tcps_sndacks++; 1986 else 1987 tcpstat.tcps_sndctrl++; 1988 tcpstat.tcps_sndtotal++; 1989 return (error); 1990} 1991 1992/* 1993 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1994 * 1995 * This code attempts to calculate the bandwidth-delay product as a 1996 * means of determining the optimal window size to maximize bandwidth, 1997 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1998 * routers. This code also does a fairly good job keeping RTTs in check 1999 * across slow links like modems. We implement an algorithm which is very 2000 * similar (but not meant to be) TCP/Vegas. The code operates on the 2001 * transmitter side of a TCP connection and so only effects the transmit 2002 * side of the connection. 2003 * 2004 * BACKGROUND: TCP makes no provision for the management of buffer space 2005 * at the end points or at the intermediate routers and switches. A TCP 2006 * stream, whether using NewReno or not, will eventually buffer as 2007 * many packets as it is able and the only reason this typically works is 2008 * due to the fairly small default buffers made available for a connection 2009 * (typicaly 16K or 32K). As machines use larger windows and/or window 2010 * scaling it is now fairly easy for even a single TCP connection to blow-out 2011 * all available buffer space not only on the local interface, but on 2012 * intermediate routers and switches as well. NewReno makes a misguided 2013 * attempt to 'solve' this problem by waiting for an actual failure to occur, 2014 * then backing off, then steadily increasing the window again until another 2015 * failure occurs, ad-infinitum. This results in terrible oscillation that 2016 * is only made worse as network loads increase and the idea of intentionally 2017 * blowing out network buffers is, frankly, a terrible way to manage network 2018 * resources. 2019 * 2020 * It is far better to limit the transmit window prior to the failure 2021 * condition being achieved. There are two general ways to do this: First 2022 * you can 'scan' through different transmit window sizes and locate the 2023 * point where the RTT stops increasing, indicating that you have filled the 2024 * pipe, then scan backwards until you note that RTT stops decreasing, then 2025 * repeat ad-infinitum. This method works in principle but has severe 2026 * implementation issues due to RTT variances, timer granularity, and 2027 * instability in the algorithm which can lead to many false positives and 2028 * create oscillations as well as interact badly with other TCP streams 2029 * implementing the same algorithm. 2030 * 2031 * The second method is to limit the window to the bandwidth delay product 2032 * of the link. This is the method we implement. RTT variances and our 2033 * own manipulation of the congestion window, bwnd, can potentially 2034 * destabilize the algorithm. For this reason we have to stabilize the 2035 * elements used to calculate the window. We do this by using the minimum 2036 * observed RTT, the long term average of the observed bandwidth, and 2037 * by adding two segments worth of slop. It isn't perfect but it is able 2038 * to react to changing conditions and gives us a very stable basis on 2039 * which to extend the algorithm. 2040 */ 2041void 2042tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2043{ 2044 u_long bw; 2045 u_long bwnd; 2046 int save_ticks; 2047 2048 INP_LOCK_ASSERT(tp->t_inpcb); 2049 2050 /* 2051 * If inflight_enable is disabled in the middle of a tcp connection, 2052 * make sure snd_bwnd is effectively disabled. 2053 */ 2054 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2055 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2056 tp->snd_bandwidth = 0; 2057 return; 2058 } 2059 2060 /* 2061 * Figure out the bandwidth. Due to the tick granularity this 2062 * is a very rough number and it MUST be averaged over a fairly 2063 * long period of time. XXX we need to take into account a link 2064 * that is not using all available bandwidth, but for now our 2065 * slop will ramp us up if this case occurs and the bandwidth later 2066 * increases. 2067 * 2068 * Note: if ticks rollover 'bw' may wind up negative. We must 2069 * effectively reset t_bw_rtttime for this case. 2070 */ 2071 save_ticks = ticks; 2072 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2073 return; 2074 2075 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2076 (save_ticks - tp->t_bw_rtttime); 2077 tp->t_bw_rtttime = save_ticks; 2078 tp->t_bw_rtseq = ack_seq; 2079 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2080 return; 2081 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2082 2083 tp->snd_bandwidth = bw; 2084 2085 /* 2086 * Calculate the semi-static bandwidth delay product, plus two maximal 2087 * segments. The additional slop puts us squarely in the sweet 2088 * spot and also handles the bandwidth run-up case and stabilization. 2089 * Without the slop we could be locking ourselves into a lower 2090 * bandwidth. 2091 * 2092 * Situations Handled: 2093 * (1) Prevents over-queueing of packets on LANs, especially on 2094 * high speed LANs, allowing larger TCP buffers to be 2095 * specified, and also does a good job preventing 2096 * over-queueing of packets over choke points like modems 2097 * (at least for the transmit side). 2098 * 2099 * (2) Is able to handle changing network loads (bandwidth 2100 * drops so bwnd drops, bandwidth increases so bwnd 2101 * increases). 2102 * 2103 * (3) Theoretically should stabilize in the face of multiple 2104 * connections implementing the same algorithm (this may need 2105 * a little work). 2106 * 2107 * (4) Stability value (defaults to 20 = 2 maximal packets) can 2108 * be adjusted with a sysctl but typically only needs to be 2109 * on very slow connections. A value no smaller then 5 2110 * should be used, but only reduce this default if you have 2111 * no other choice. 2112 */ 2113#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2114 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2115#undef USERTT 2116 2117 if (tcp_inflight_debug > 0) { 2118 static int ltime; 2119 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2120 ltime = ticks; 2121 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2122 tp, 2123 bw, 2124 tp->t_rttbest, 2125 tp->t_srtt, 2126 bwnd 2127 ); 2128 } 2129 } 2130 if ((long)bwnd < tcp_inflight_min) 2131 bwnd = tcp_inflight_min; 2132 if (bwnd > tcp_inflight_max) 2133 bwnd = tcp_inflight_max; 2134 if ((long)bwnd < tp->t_maxseg * 2) 2135 bwnd = tp->t_maxseg * 2; 2136 tp->snd_bwnd = bwnd; 2137} 2138 2139#ifdef TCP_SIGNATURE 2140/* 2141 * Callback function invoked by m_apply() to digest TCP segment data 2142 * contained within an mbuf chain. 2143 */ 2144static int 2145tcp_signature_apply(void *fstate, void *data, u_int len) 2146{ 2147 2148 MD5Update(fstate, (u_char *)data, len); 2149 return (0); 2150} 2151 2152/* 2153 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2154 * 2155 * Parameters: 2156 * m pointer to head of mbuf chain 2157 * off0 offset to TCP header within the mbuf chain 2158 * len length of TCP segment data, excluding options 2159 * optlen length of TCP segment options 2160 * buf pointer to storage for computed MD5 digest 2161 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2162 * 2163 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2164 * When called from tcp_input(), we can be sure that th_sum has been 2165 * zeroed out and verified already. 2166 * 2167 * This function is for IPv4 use only. Calling this function with an 2168 * IPv6 packet in the mbuf chain will yield undefined results. 2169 * 2170 * Return 0 if successful, otherwise return -1. 2171 * 2172 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2173 * search with the destination IP address, and a 'magic SPI' to be 2174 * determined by the application. This is hardcoded elsewhere to 1179 2175 * right now. Another branch of this code exists which uses the SPD to 2176 * specify per-application flows but it is unstable. 2177 */ 2178int 2179tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2180 u_char *buf, u_int direction) 2181{ 2182 union sockaddr_union dst; 2183 struct ippseudo ippseudo; 2184 MD5_CTX ctx; 2185 int doff; 2186 struct ip *ip; 2187 struct ipovly *ipovly; 2188 struct secasvar *sav; 2189 struct tcphdr *th; 2190 u_short savecsum; 2191 2192 KASSERT(m != NULL, ("NULL mbuf chain")); 2193 KASSERT(buf != NULL, ("NULL signature pointer")); 2194 2195 /* Extract the destination from the IP header in the mbuf. */ 2196 ip = mtod(m, struct ip *); 2197 bzero(&dst, sizeof(union sockaddr_union)); 2198 dst.sa.sa_len = sizeof(struct sockaddr_in); 2199 dst.sa.sa_family = AF_INET; 2200 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2201 ip->ip_src : ip->ip_dst; 2202 2203 /* Look up an SADB entry which matches the address of the peer. */ 2204 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2205 if (sav == NULL) { 2206 printf("%s: SADB lookup failed for %s\n", __func__, 2207 inet_ntoa(dst.sin.sin_addr)); 2208 return (EINVAL); 2209 } 2210 2211 MD5Init(&ctx); 2212 ipovly = (struct ipovly *)ip; 2213 th = (struct tcphdr *)((u_char *)ip + off0); 2214 doff = off0 + sizeof(struct tcphdr) + optlen; 2215 2216 /* 2217 * Step 1: Update MD5 hash with IP pseudo-header. 2218 * 2219 * XXX The ippseudo header MUST be digested in network byte order, 2220 * or else we'll fail the regression test. Assume all fields we've 2221 * been doing arithmetic on have been in host byte order. 2222 * XXX One cannot depend on ipovly->ih_len here. When called from 2223 * tcp_output(), the underlying ip_len member has not yet been set. 2224 */ 2225 ippseudo.ippseudo_src = ipovly->ih_src; 2226 ippseudo.ippseudo_dst = ipovly->ih_dst; 2227 ippseudo.ippseudo_pad = 0; 2228 ippseudo.ippseudo_p = IPPROTO_TCP; 2229 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2230 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2231 2232 /* 2233 * Step 2: Update MD5 hash with TCP header, excluding options. 2234 * The TCP checksum must be set to zero. 2235 */ 2236 savecsum = th->th_sum; 2237 th->th_sum = 0; 2238 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2239 th->th_sum = savecsum; 2240 2241 /* 2242 * Step 3: Update MD5 hash with TCP segment data. 2243 * Use m_apply() to avoid an early m_pullup(). 2244 */ 2245 if (len > 0) 2246 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2247 2248 /* 2249 * Step 4: Update MD5 hash with shared secret. 2250 */ 2251 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2252 MD5Final(buf, &ctx); 2253 2254 key_sa_recordxfer(sav, m); 2255 KEY_FREESAV(&sav); 2256 return (0); 2257} 2258#endif /* TCP_SIGNATURE */ 2259 2260static int 2261sysctl_drop(SYSCTL_HANDLER_ARGS) 2262{ 2263 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2264 struct sockaddr_storage addrs[2]; 2265 struct inpcb *inp; 2266 struct tcpcb *tp; 2267 struct tcptw *tw; 2268 struct sockaddr_in *fin, *lin; 2269#ifdef INET6 2270 struct sockaddr_in6 *fin6, *lin6; 2271 struct in6_addr f6, l6; 2272#endif 2273 int error; 2274 2275 inp = NULL; 2276 fin = lin = NULL; 2277#ifdef INET6 2278 fin6 = lin6 = NULL; 2279#endif 2280 error = 0; 2281 2282 if (req->oldptr != NULL || req->oldlen != 0) 2283 return (EINVAL); 2284 if (req->newptr == NULL) 2285 return (EPERM); 2286 if (req->newlen < sizeof(addrs)) 2287 return (ENOMEM); 2288 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2289 if (error) 2290 return (error); 2291 2292 switch (addrs[0].ss_family) { 2293#ifdef INET6 2294 case AF_INET6: 2295 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2296 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2297 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2298 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2299 return (EINVAL); 2300 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2301 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2302 return (EINVAL); 2303 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2304 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2305 fin = (struct sockaddr_in *)&addrs[0]; 2306 lin = (struct sockaddr_in *)&addrs[1]; 2307 break; 2308 } 2309 error = sa6_embedscope(fin6, ip6_use_defzone); 2310 if (error) 2311 return (error); 2312 error = sa6_embedscope(lin6, ip6_use_defzone); 2313 if (error) 2314 return (error); 2315 break; 2316#endif 2317 case AF_INET: 2318 fin = (struct sockaddr_in *)&addrs[0]; 2319 lin = (struct sockaddr_in *)&addrs[1]; 2320 if (fin->sin_len != sizeof(struct sockaddr_in) || 2321 lin->sin_len != sizeof(struct sockaddr_in)) 2322 return (EINVAL); 2323 break; 2324 default: 2325 return (EINVAL); 2326 } 2327 INP_INFO_WLOCK(&tcbinfo); 2328 switch (addrs[0].ss_family) { 2329#ifdef INET6 2330 case AF_INET6: 2331 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2332 &l6, lin6->sin6_port, 0, NULL); 2333 break; 2334#endif 2335 case AF_INET: 2336 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2337 lin->sin_addr, lin->sin_port, 0, NULL); 2338 break; 2339 } 2340 if (inp != NULL) { 2341 INP_LOCK(inp); 2342 if (inp->inp_vflag & INP_TIMEWAIT) { 2343 tw = intotw(inp); 2344 tcp_twclose(tw, 0); 2345 } else if (!(inp->inp_vflag & INP_DROPPED) && 2346 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2347 tp = intotcpcb(inp); 2348 tcp_drop(tp, ECONNABORTED); 2349 } 2350 INP_UNLOCK(inp); 2351 } else 2352 error = ESRCH; 2353 INP_INFO_WUNLOCK(&tcbinfo); 2354 return (error); 2355} 2356 2357SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2358 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2359 0, sysctl_drop, "", "Drop TCP connection"); 2360