tcp_timewait.c revision 133192
1139823Simp/* 211150Swollman * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 31541Srgrimes * The Regents of the University of California. All rights reserved. 41541Srgrimes * 51541Srgrimes * Redistribution and use in source and binary forms, with or without 61541Srgrimes * modification, are permitted provided that the following conditions 71541Srgrimes * are met: 81541Srgrimes * 1. Redistributions of source code must retain the above copyright 91541Srgrimes * notice, this list of conditions and the following disclaimer. 101541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 111541Srgrimes * notice, this list of conditions and the following disclaimer in the 121541Srgrimes * documentation and/or other materials provided with the distribution. 131541Srgrimes * 4. Neither the name of the University nor the names of its contributors 141541Srgrimes * may be used to endorse or promote products derived from this software 151541Srgrimes * without specific prior written permission. 161541Srgrimes * 171541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 181541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 191541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 201541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 211541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 221541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 231541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 241541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 251541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 261541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 271541Srgrimes * SUCH DAMAGE. 281541Srgrimes * 2911150Swollman * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 301541Srgrimes * $FreeBSD: head/sys/netinet/tcp_timewait.c 133192 2004-08-06 03:45:45Z rwatson $ 311541Srgrimes */ 32172467Ssilby 33172467Ssilby#include "opt_compat.h" 34172467Ssilby#include "opt_inet.h" 35125680Sbms#include "opt_inet6.h" 3654263Sshin#include "opt_ipsec.h" 3729514Sjoerg#include "opt_mac.h" 3829514Sjoerg#include "opt_tcpdebug.h" 391541Srgrimes#include "opt_tcp_sack.h" 401541Srgrimes 4150673Sjlemon#include <sys/param.h> 4212172Sphk#include <sys/systm.h> 4312172Sphk#include <sys/callout.h> 441541Srgrimes#include <sys/kernel.h> 451541Srgrimes#include <sys/sysctl.h> 46164033Srwatson#include <sys/mac.h> 4748758Sgreen#include <sys/malloc.h> 481541Srgrimes#include <sys/mbuf.h> 491541Srgrimes#ifdef INET6 501541Srgrimes#include <sys/domain.h> 5175619Skris#endif 5234923Sbde#include <sys/proc.h> 5392760Sjeff#include <sys/socket.h> 541541Srgrimes#include <sys/socketvar.h> 551541Srgrimes#include <sys/protosw.h> 561541Srgrimes#include <sys/random.h> 57196019Srwatson 581541Srgrimes#include <vm/uma.h> 591541Srgrimes 60221250Sbz#include <net/route.h> 611541Srgrimes#include <net/if.h> 62221250Sbz 631541Srgrimes#include <netinet/in.h> 64221250Sbz#include <netinet/in_systm.h> 65221250Sbz#include <netinet/ip.h> 6655679Sshin#ifdef INET6 6755679Sshin#include <netinet/ip6.h> 6855679Sshin#endif 6955679Sshin#include <netinet/in_pcb.h> 70148385Sume#ifdef INET6 71122922Sandre#include <netinet6/in6_pcb.h> 7255679Sshin#endif 731541Srgrimes#include <netinet/in_var.h> 741541Srgrimes#include <netinet/ip_var.h> 751541Srgrimes#ifdef INET6 761541Srgrimes#include <netinet6/ip6_var.h> 771541Srgrimes#include <netinet6/nd6.h> 7855679Sshin#endif 7955679Sshin#include <netinet/tcp.h> 8055679Sshin#include <netinet/tcp_fsm.h> 811541Srgrimes#include <netinet/tcp_seq.h> 826283Swollman#include <netinet/tcp_timer.h> 836283Swollman#include <netinet/tcp_var.h> 846283Swollman#ifdef INET6 85221250Sbz#include <netinet6/tcp6_var.h> 8655679Sshin#endif 87221250Sbz#include <netinet/tcpip.h> 881541Srgrimes#ifdef TCPDEBUG 8958698Sjlemon#include <netinet/tcp_debug.h> 9058698Sjlemon#endif 91163606Srwatson#include <netinet6/ip6protosw.h> 92163606Srwatson 93215701Sdim#ifdef IPSEC 94207369Sbz#include <netinet6/ipsec.h> 95157478Sglebius#ifdef INET6 96162767Ssilby#include <netinet6/ipsec6.h> 97169608Sandre#endif 98169608Sandre#endif /*IPSEC*/ 99169608Sandre 100169608Sandre#ifdef FAST_IPSEC 101169608Sandre#include <netipsec/ipsec.h> 102169608Sandre#include <netipsec/xform.h> 103215701Sdim#ifdef INET6 104195727Srwatson#include <netipsec/ipsec6.h> 105195699Srwatson#endif 106169608Sandre#include <netipsec/key.h> 107169608Sandre#define IPSEC 108169608Sandre#endif /*FAST_IPSEC*/ 109162031Sglebius 110162767Ssilby#include <machine/in_cksum.h> 111162767Ssilby#include <sys/md5.h> 112162767Ssilby 113162767Ssilbyint tcp_mssdflt = TCP_MSS; 114162767SsilbySYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 115162768Smaxim &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 116162768Smaxim 117162768Smaxim#ifdef INET6 118181803Sbzint tcp_v6mssdflt = TCP6_MSS; 119181803SbzSYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 120162767Ssilby CTLFLAG_RW, &tcp_v6mssdflt , 0, 121181803Sbz "Default TCP Maximum Segment Size for IPv6"); 122162767Ssilby#endif 123162767Ssilby 124162767Ssilby/* 125162767Ssilby * Minimum MSS we accept and use. This prevents DoS attacks where 126162767Ssilby * we are forced to a ridiculous low MSS like 20 and send hundreds 127162031Sglebius * of packets instead of one. The effect scales with the available 128162031Sglebius * bandwidth and quickly saturates the CPU and network interface 129162031Sglebius * with packet generation and sending. Set to zero to disable MINMSS 130157478Sglebius * checking. This setting prevents us from sending too small packets. 131162031Sglebius */ 132162767Ssilbyint tcp_minmss = TCP_MINMSS; 133162031SglebiusSYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 134162031Sglebius &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 135170289Sdwmalone/* 136162767Ssilby * Number of TCP segments per second we accept from remote host 137162767Ssilby * before we start to calculate average segment size. If average 138162031Sglebius * segment size drops below the minimum TCP MSS we assume a DoS 139190787Szec * attack and reset+drop the connection. Care has to be taken not to 140162767Ssilby * set this value too small to not kill interactive type connections 141162031Sglebius * (telnet, SSH) which send many small packets. 142162031Sglebius */ 143169541Sandreint tcp_minmssoverload = TCP_MINMSSOVERLOAD; 144162031SglebiusSYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, 145162031Sglebius &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" 146162031Sglebius "be under the MINMSS Size"); 147162031Sglebius 148207369Sbz#if 0 149207369Sbzstatic int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 150195699SrwatsonSYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 151195699Srwatson &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 152167721Sandre#endif 153162151Sglebius 154169541Sandreint tcp_do_rfc1323 = 1; 155169541SandreSYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 156157927Sps &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 157157927Sps 158162031Sglebiusint tcp_do_rfc1644 = 0; 159190787SzecSYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW, 160157927Sps &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions"); 161157927Sps 1621541Srgrimesstatic int tcp_tcbhashsize = 0; 163169541SandreSYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 1641541Srgrimes &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 165169454Srwatson 166190787Szecstatic int do_tcpdrain = 1; 167162031SglebiusSYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 168157478Sglebius "Enable tcp_drain routine for extra help when low on mbufs"); 169157478Sglebius 170190787SzecSYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 171162031Sglebius &tcbinfo.ipi_count, 0, "Number of active PCBs"); 172190787Szec 173181803Sbzstatic int icmp_may_rst = 1; 1741541SrgrimesSYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 1751541Srgrimes "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 176193731Szec 177193731Szecstatic int tcp_isn_reseed_interval = 0; 178193731SzecSYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 179193731Szec &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 180193731Szec 181193731Szec/* 182193731Szec * TCP bandwidth limiting sysctls. Note that the default lower bound of 183193731Szec * 1024 exists only for debugging. A good production default would be 184193731Szec * something like 6100. 185193731Szec */ 186204838SbzSYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 187204838Sbz "TCP inflight data limiting"); 188193731Szec 189193731Szecstatic int tcp_inflight_enable = 1; 190193731SzecSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 1911541Srgrimes &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 192111145Sjlemon 193138410Srwatsonstatic int tcp_inflight_debug = 0; 194111145SjlemonSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 195111145Sjlemon &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 196111145Sjlemon 197157431Srwatsonstatic int tcp_inflight_min = 6144; 198111145SjlemonSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 199111145Sjlemon &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 200162151Sglebius 201162111Srustatic int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 202111145SjlemonSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 203221250Sbz &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 204221250Sbz 205221250Sbzstatic int tcp_inflight_stab = 20; 206111145SjlemonSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 207181803Sbz &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 208178285Srwatson 209138020SrwatsonSYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 0, "TCP SACK"); 210221250Sbzint tcp_do_sack = 1; 211221250SbzSYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, enable, CTLFLAG_RW, 212221250Sbz &tcp_do_sack, 0, "Enable/Disable TCP SACK support"); 213221250Sbz 214221250Sbzuma_zone_t sack_hole_zone; 215221250Sbz 216221250Sbzstatic struct inpcb *tcp_notify(struct inpcb *, int); 217221250Sbzstatic void tcp_discardcb(struct tcpcb *); 218221250Sbzstatic void tcp_isn_tick(void *); 219221250Sbz 220221250Sbz/* 221221250Sbz * Target size of TCP PCB hash tables. Must be a power of two. 222221250Sbz * 223221250Sbz * Note that this can be overridden by the kernel environment 224221250Sbz * variable net.inet.tcp.tcbhashsize 225221250Sbz */ 226221250Sbz#ifndef TCBHASHSIZE 227221250Sbz#define TCBHASHSIZE 512 228162151Sglebius#endif 229162151Sglebius 230190787Szec/* 231112009Sjlemon * XXX 232169608Sandre * Callouts should be moved into struct tcp directly. They are currently 233112009Sjlemon * separate because the tcpcb structure is exported to userland for sysctl 234157376Srwatson * parsing purposes, which do not know about callouts. 235157376Srwatson */ 236178285Srwatsonstruct tcpcb_mem { 237112009Sjlemon struct tcpcb tcb; 238112009Sjlemon struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 239112009Sjlemon struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 240111145Sjlemon}; 241111145Sjlemon 242111145Sjlemonstatic uma_zone_t tcpcb_zone; 243111145Sjlemonstatic uma_zone_t tcptw_zone; 244111145Sjlemonstruct callout isn_callout; 245229665Sjhb 246229665Sjhb/* 247229665Sjhb * Tcp initialization 248229665Sjhb */ 249111145Sjlemonvoid 250111145Sjlemontcp_init() 251111145Sjlemon{ 252111145Sjlemon int hashsize = TCBHASHSIZE; 253133874Srwatson 254169477Sandre tcp_ccgen = 1; 255111145Sjlemon 256169477Sandre tcp_delacktime = TCPTV_DELACK; 257169477Sandre tcp_keepinit = TCPTV_KEEP_INIT; 258111145Sjlemon tcp_keepidle = TCPTV_KEEP_IDLE; 259169477Sandre tcp_keepintvl = TCPTV_KEEPINTVL; 260169477Sandre tcp_maxpersistidle = TCPTV_KEEP_IDLE; 261111145Sjlemon tcp_msl = TCPTV_MSL; 262111145Sjlemon tcp_rexmit_min = TCPTV_MIN; 263111145Sjlemon tcp_rexmit_slop = TCPTV_CPU_VAR; 264121850Ssilby 265121884Ssilby INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 266111145Sjlemon LIST_INIT(&tcb); 267112009Sjlemon tcbinfo.listhead = &tcb; 268111145Sjlemon TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 269111145Sjlemon if (!powerof2(hashsize)) { 270111145Sjlemon printf("WARNING: TCB hash size not a power of 2\n"); 271111145Sjlemon hashsize = 512; /* safe default */ 272111145Sjlemon } 273111145Sjlemon tcp_tcbhashsize = hashsize; 274137139Sandre tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 275157376Srwatson tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 276157376Srwatson &tcbinfo.porthashmask); 277157376Srwatson tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 278157376Srwatson NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 279157376Srwatson uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 280160925Srwatson#ifdef INET6 281160925Srwatson#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 282160925Srwatson#else /* INET6 */ 283157376Srwatson#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 284111145Sjlemon#endif /* INET6 */ 285111145Sjlemon if (max_protohdr < TCP_MINPROTOHDR) 286160925Srwatson max_protohdr = TCP_MINPROTOHDR; 287169347Srwatson if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 288130387Srwatson panic("tcp_init"); 289111145Sjlemon#undef TCP_MINPROTOHDR 290157376Srwatson /* 291114794Srwatson * These have to be type stable for the benefit of the timers. 292126351Srwatson */ 293157432Srwatson tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 294189848Srwatson NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 295169608Sandre uma_zone_set_max(tcpcb_zone, maxsockets); 296157376Srwatson tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 297157376Srwatson NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 298157376Srwatson uma_zone_set_max(tcptw_zone, maxsockets / 5); 299157376Srwatson tcp_timer_init(); 300157376Srwatson syncache_init(); 301189848Srwatson tcp_hc_init(); 302157376Srwatson tcp_reass_init(); 303157376Srwatson callout_init(&isn_callout, CALLOUT_MPSAFE); 304189848Srwatson tcp_isn_tick(NULL); 305178285Srwatson EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 306157376Srwatson SHUTDOWN_PRI_DEFAULT); 307157376Srwatson sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 308157376Srwatson NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 309157376Srwatson} 310157376Srwatson 311178285Srwatsonvoid 312111145Sjlemontcp_fini(xtp) 313111145Sjlemon void *xtp; 314162064Sglebius{ 315121850Ssilby callout_stop(&isn_callout); 316121884Ssilby 317121884Ssilby} 318121884Ssilby 319121884Ssilby/* 320121884Ssilby * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 321121884Ssilby * tcp_template used to store this data in mbufs, but we now recopy it out 322121884Ssilby * of the tcpcb each time to conserve mbufs. 323121884Ssilby */ 324121884Ssilbyvoid 325121884Ssilbytcpip_fillheaders(inp, ip_ptr, tcp_ptr) 326121850Ssilby struct inpcb *inp; 327121850Ssilby void *ip_ptr; 328121850Ssilby void *tcp_ptr; 329121850Ssilby{ 330121850Ssilby struct tcphdr *th = (struct tcphdr *)tcp_ptr; 331121850Ssilby 332121850Ssilby#ifdef INET6 333121884Ssilby if ((inp->inp_vflag & INP_IPV6) != 0) { 334121884Ssilby struct ip6_hdr *ip6; 335121850Ssilby 336181803Sbz ip6 = (struct ip6_hdr *)ip_ptr; 337121884Ssilby ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 338121884Ssilby (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 339133874Srwatson ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 340121884Ssilby (IPV6_VERSION & IPV6_VERSION_MASK); 341139222Srwatson ip6->ip6_nxt = IPPROTO_TCP; 342121850Ssilby ip6->ip6_plen = sizeof(struct tcphdr); 343139222Srwatson ip6->ip6_src = inp->in6p_laddr; 344121850Ssilby ip6->ip6_dst = inp->in6p_faddr; 345162064Sglebius } else 346121850Ssilby#endif 347169608Sandre { 348169608Sandre struct ip *ip; 349169608Sandre 350169608Sandre ip = (struct ip *)ip_ptr; 351169608Sandre ip->ip_v = IPVERSION; 352169608Sandre ip->ip_hl = 5; 353169608Sandre ip->ip_tos = inp->inp_ip_tos; 354169608Sandre ip->ip_len = 0; 355169608Sandre ip->ip_id = 0; 356169608Sandre ip->ip_off = 0; 357169608Sandre ip->ip_ttl = inp->inp_ip_ttl; 358169608Sandre ip->ip_sum = 0; 359169608Sandre ip->ip_p = IPPROTO_TCP; 360181803Sbz ip->ip_src = inp->inp_laddr; 361178285Srwatson ip->ip_dst = inp->inp_faddr; 362169608Sandre } 363169608Sandre th->th_sport = inp->inp_lport; 364169608Sandre th->th_dport = inp->inp_fport; 365169608Sandre th->th_seq = 0; 366169608Sandre th->th_ack = 0; 367169608Sandre th->th_x2 = 0; 368169608Sandre th->th_off = 5; 369169608Sandre th->th_flags = 0; 370169608Sandre th->th_win = 0; 371169608Sandre th->th_urp = 0; 372169608Sandre th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 373169608Sandre} 374169608Sandre 375169608Sandre/* 376169608Sandre * Create template to be used to send tcp packets on a connection. 377169608Sandre * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 378169608Sandre * use for this function is in keepalives, which use tcp_respond. 379169608Sandre */ 380169608Sandrestruct tcptemp * 381169608Sandretcpip_maketemplate(inp) 382169608Sandre struct inpcb *inp; 383169608Sandre{ 384169608Sandre struct mbuf *m; 385169608Sandre struct tcptemp *n; 386169608Sandre 387169608Sandre m = m_get(M_DONTWAIT, MT_HEADER); 388169608Sandre if (m == NULL) 389169608Sandre return (0); 390169608Sandre m->m_len = sizeof(struct tcptemp); 391169608Sandre n = mtod(m, struct tcptemp *); 392169608Sandre 393169608Sandre tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 394169608Sandre return (n); 395169608Sandre} 396169608Sandre 397169608Sandre/* 398169608Sandre * Send a single message to the TCP at address specified by 399169608Sandre * the given TCP/IP header. If m == NULL, then we make a copy 400169608Sandre * of the tcpiphdr at ti and send directly to the addressed host. 401169608Sandre * This is used to force keep alive messages out using the TCP 402169608Sandre * template for a connection. If flags are given then we send 403169608Sandre * a message back to the TCP which originated the * segment ti, 404169608Sandre * and discard the mbuf containing it and any other attached mbufs. 405169608Sandre * 406169608Sandre * In any case the ack and sequence number of the transmitted 407169608Sandre * segment are as specified by the parameters. 408169608Sandre * 409169608Sandre * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 410169608Sandre */ 411169608Sandrevoid 412169608Sandretcp_respond(tp, ipgen, th, m, ack, seq, flags) 413169608Sandre struct tcpcb *tp; 414169608Sandre void *ipgen; 415169608Sandre register struct tcphdr *th; 416169608Sandre register struct mbuf *m; 417218909Sbrucec tcp_seq ack, seq; 418169608Sandre int flags; 419169608Sandre{ 420169608Sandre register int tlen; 421169608Sandre int win = 0; 422169608Sandre struct ip *ip; 423169608Sandre struct tcphdr *nth; 424169608Sandre#ifdef INET6 425169608Sandre struct ip6_hdr *ip6; 426169608Sandre int isipv6; 427169608Sandre#endif /* INET6 */ 428169608Sandre int ipflags = 0; 429169608Sandre struct inpcb *inp; 430169608Sandre 431169608Sandre KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 432169608Sandre 433169608Sandre#ifdef INET6 434169608Sandre isipv6 = ((struct ip *)ipgen)->ip_v == 6; 435169608Sandre ip6 = ipgen; 436169608Sandre#endif /* INET6 */ 437169608Sandre ip = ipgen; 438178285Srwatson 439169608Sandre if (tp != NULL) { 440169608Sandre inp = tp->t_inpcb; 441169608Sandre KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 442169608Sandre INP_INFO_WLOCK_ASSERT(&tcbinfo); 443157376Srwatson INP_LOCK_ASSERT(inp); 444112009Sjlemon } else 445111145Sjlemon inp = NULL; 446157376Srwatson 447111145Sjlemon if (tp != NULL) { 448111145Sjlemon if (!(flags & TH_RST)) { 449157376Srwatson win = sbspace(&inp->inp_socket->so_rcv); 450157386Srwatson if (win > (long)TCP_MAXWIN << tp->rcv_scale) 451157376Srwatson win = (long)TCP_MAXWIN << tp->rcv_scale; 452160549Srwatson } 453160549Srwatson } 454157386Srwatson if (m == NULL) { 455160549Srwatson m = m_gethdr(M_DONTWAIT, MT_HEADER); 456160549Srwatson if (m == NULL) 457157376Srwatson return; 458111145Sjlemon tlen = 0; 459189848Srwatson m->m_data += max_linkhdr; 460157432Srwatson#ifdef INET6 461181803Sbz if (isipv6) { 462178285Srwatson bcopy((caddr_t)ip6, mtod(m, caddr_t), 463138020Srwatson sizeof(struct ip6_hdr)); 464111145Sjlemon ip6 = mtod(m, struct ip6_hdr *); 465169608Sandre nth = (struct tcphdr *)(ip6 + 1); 466111145Sjlemon } else 467158009Srwatson#endif /* INET6 */ 468157376Srwatson { 469157376Srwatson bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 470157386Srwatson ip = mtod(m, struct ip *); 471160549Srwatson nth = (struct tcphdr *)(ip + 1); 472160549Srwatson } 473160549Srwatson bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 474160549Srwatson flags = TH_ACK; 475160549Srwatson } else { 476160549Srwatson m_freem(m->m_next); 477189848Srwatson m->m_next = NULL; 478189848Srwatson m->m_data = (caddr_t)ipgen; 479178285Srwatson /* m_len is set later */ 480157386Srwatson tlen = 0; 481157386Srwatson#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 482157386Srwatson#ifdef INET6 483157386Srwatson if (isipv6) { 484157386Srwatson xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 485157386Srwatson nth = (struct tcphdr *)(ip6 + 1); 486157376Srwatson } else 487157386Srwatson#endif /* INET6 */ 488157386Srwatson { 489157386Srwatson xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 490157386Srwatson nth = (struct tcphdr *)(ip + 1); 491157386Srwatson } 492178285Srwatson if (th != nth) { 493157376Srwatson /* 494185370Sbz * this is usually a case when an extension header 495185370Sbz * exists between the IPv6 header and the 496190948Srwatson * TCP header. 497126002Spjd */ 498126002Spjd nth->th_sport = th->th_sport; 499112009Sjlemon nth->th_dport = th->th_dport; 500157376Srwatson } 501190787Szec xchg(nth->th_dport, nth->th_sport, n_short); 502111145Sjlemon#undef xchg 503111145Sjlemon } 504111145Sjlemon#ifdef INET6 505126351Srwatson if (isipv6) { 506111145Sjlemon ip6->ip6_flow = 0; 507111145Sjlemon ip6->ip6_vfc = IPV6_VERSION; 508221250Sbz ip6->ip6_nxt = IPPROTO_TCP; 509221250Sbz ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 510221250Sbz tlen)); 511111145Sjlemon tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 512221250Sbz } else 513111145Sjlemon#endif 514221250Sbz { 515111145Sjlemon tlen += sizeof (struct tcpiphdr); 516221250Sbz ip->ip_len = tlen; 517168845Sandre ip->ip_ttl = ip_defttl; 518111145Sjlemon if (path_mtu_discovery) 519111145Sjlemon ip->ip_off |= IP_DF; 520186222Sbz } 521111145Sjlemon m->m_len = tlen; 522111145Sjlemon m->m_pkthdr.len = tlen; 523178285Srwatson m->m_pkthdr.rcvif = NULL; 524138020Srwatson#ifdef MAC 525151967Sandre if (inp != NULL) { 526111145Sjlemon /* 527111145Sjlemon * Packet is associated with a socket, so allow the 528111145Sjlemon * label of the response to reflect the socket label. 529111145Sjlemon */ 530114794Srwatson INP_LOCK_ASSERT(inp); 531172930Srwatson mac_create_mbuf_from_inpcb(inp, m); 532114794Srwatson } else { 533114794Srwatson /* 534111153Sjlemon * Packet is not associated with a socket, so possibly 535111145Sjlemon * update the label in place. 536111145Sjlemon */ 537111145Sjlemon mac_reflect_mbuf_tcp(m); 538111145Sjlemon } 539111145Sjlemon#endif 540221250Sbz nth->th_seq = htonl(seq); 541111153Sjlemon nth->th_ack = htonl(ack); 542221250Sbz nth->th_x2 = 0; 543221250Sbz nth->th_off = sizeof (struct tcphdr) >> 2; 544221250Sbz nth->th_flags = flags; 545221250Sbz if (tp != NULL) 546111153Sjlemon nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 547111145Sjlemon else 548111145Sjlemon nth->th_win = htons((u_short)win); 549111145Sjlemon nth->th_urp = 0; 550111145Sjlemon#ifdef INET6 551111145Sjlemon if (isipv6) { 552221250Sbz nth->th_sum = 0; 553168845Sandre nth->th_sum = in6_cksum(m, IPPROTO_TCP, 554133874Srwatson sizeof(struct ip6_hdr), 555133874Srwatson tlen - sizeof(struct ip6_hdr)); 556111145Sjlemon ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 557111145Sjlemon NULL, NULL); 558133874Srwatson } else 559111145Sjlemon#endif /* INET6 */ 560168845Sandre { 561169477Sandre nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 562168845Sandre htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 563133874Srwatson m->m_pkthdr.csum_flags = CSUM_TCP; 564168845Sandre m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 565111145Sjlemon } 566111145Sjlemon#ifdef TCPDEBUG 567111145Sjlemon if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 568111145Sjlemon tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 569111145Sjlemon#endif 570111145Sjlemon#ifdef INET6 571111145Sjlemon if (isipv6) 572111145Sjlemon (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 573111145Sjlemon else 574111145Sjlemon#endif /* INET6 */ 575111145Sjlemon (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 576111145Sjlemon} 577111153Sjlemon 578111145Sjlemon/* 579111145Sjlemon * Create a new TCP control block, making an 580111145Sjlemon * empty reassembly queue and hooking it to the argument 581122922Sandre * protocol control block. The `inp' parameter must have 582122922Sandre * come from the zone allocator set up in tcp_init(). 583111145Sjlemon */ 584221250Sbzstruct tcpcb * 585111153Sjlemontcp_newtcpcb(inp) 586221250Sbz struct inpcb *inp; 587221250Sbz{ 588221250Sbz struct tcpcb_mem *tm; 589221250Sbz struct tcpcb *tp; 590111153Sjlemon#ifdef INET6 591111145Sjlemon int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 592133874Srwatson#endif /* INET6 */ 593111145Sjlemon int callout_flag; 594111145Sjlemon 595111145Sjlemon tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 596181803Sbz if (tm == NULL) 597124248Sandre return (NULL); 598122922Sandre tp = &tm->tcb; 599134793Sjmg /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 600134793Sjmg tp->t_maxseg = tp->t_maxopd = 601111145Sjlemon#ifdef INET6 602221250Sbz isipv6 ? tcp_v6mssdflt : 603111145Sjlemon#endif /* INET6 */ 604190948Srwatson tcp_mssdflt; 605111145Sjlemon 606190948Srwatson /* Set up our timeouts. */ 607190948Srwatson /* 608111145Sjlemon * XXXRW: Are these actually MPSAFE? I think so, but need to 609111145Sjlemon * review the timed wait code, as it has some list variables, 610169608Sandre * etc, that are global. 611169608Sandre */ 612169608Sandre callout_flag = debug_mpsafenet ? CALLOUT_MPSAFE : 0; 613169608Sandre callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, callout_flag); 614169608Sandre callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, callout_flag); 615181803Sbz callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, callout_flag); 616178285Srwatson callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, callout_flag); 617169608Sandre callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, callout_flag); 618181803Sbz 619169608Sandre if (tcp_do_rfc1323) 620181803Sbz tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 621169608Sandre if (tcp_do_rfc1644) 622169608Sandre tp->t_flags |= TF_REQ_CC; 623169608Sandre tp->sack_enable = tcp_do_sack; 624169608Sandre tp->t_inpcb = inp; /* XXX */ 625169608Sandre /* 626169608Sandre * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 627181803Sbz * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 628181803Sbz * reasonable initial retransmit time. 629169608Sandre */ 630169608Sandre tp->t_srtt = TCPTV_SRTTBASE; 631169608Sandre tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 632169608Sandre tp->t_rttmin = tcp_rexmit_min; 633169608Sandre tp->t_rxtcur = TCPTV_RTOBASE; 634169608Sandre tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 635169608Sandre tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 636181803Sbz tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 637169608Sandre tp->t_rcvtime = ticks; 638181803Sbz tp->t_bw_rtttime = ticks; 639196410Speter /* 640169608Sandre * IPv4 TTL initialization is necessary for an IPv6 socket as well, 641178285Srwatson * because the socket may be bound to an IPv6 wildcard address, 642169608Sandre * which may match an IPv4-mapped IPv6 address. 643169608Sandre */ 644169608Sandre inp->inp_ip_ttl = ip_defttl; 645169608Sandre inp->inp_ppcb = (caddr_t)tp; 646169608Sandre return (tp); /* XXX */ 647169608Sandre} 648 649/* 650 * Drop a TCP connection, reporting 651 * the specified error. If connection is synchronized, 652 * then send a RST to peer. 653 */ 654struct tcpcb * 655tcp_drop(tp, errno) 656 register struct tcpcb *tp; 657 int errno; 658{ 659 struct socket *so = tp->t_inpcb->inp_socket; 660 661 if (TCPS_HAVERCVDSYN(tp->t_state)) { 662 tp->t_state = TCPS_CLOSED; 663 (void) tcp_output(tp); 664 tcpstat.tcps_drops++; 665 } else 666 tcpstat.tcps_conndrops++; 667 if (errno == ETIMEDOUT && tp->t_softerror) 668 errno = tp->t_softerror; 669 so->so_error = errno; 670 return (tcp_close(tp)); 671} 672 673static void 674tcp_discardcb(tp) 675 struct tcpcb *tp; 676{ 677 struct tseg_qent *q; 678 struct inpcb *inp = tp->t_inpcb; 679 struct socket *so = inp->inp_socket; 680#ifdef INET6 681 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 682#endif /* INET6 */ 683 684 /* 685 * Make sure that all of our timers are stopped before we 686 * delete the PCB. 687 */ 688 callout_stop(tp->tt_rexmt); 689 callout_stop(tp->tt_persist); 690 callout_stop(tp->tt_keep); 691 callout_stop(tp->tt_2msl); 692 callout_stop(tp->tt_delack); 693 694 /* 695 * If we got enough samples through the srtt filter, 696 * save the rtt and rttvar in the routing entry. 697 * 'Enough' is arbitrarily defined as 4 rtt samples. 698 * 4 samples is enough for the srtt filter to converge 699 * to within enough % of the correct value; fewer samples 700 * and we could save a bogus rtt. The danger is not high 701 * as tcp quickly recovers from everything. 702 * XXX: Works very well but needs some more statistics! 703 */ 704 if (tp->t_rttupdated >= 4) { 705 struct hc_metrics_lite metrics; 706 u_long ssthresh; 707 708 bzero(&metrics, sizeof(metrics)); 709 /* 710 * Update the ssthresh always when the conditions below 711 * are satisfied. This gives us better new start value 712 * for the congestion avoidance for new connections. 713 * ssthresh is only set if packet loss occured on a session. 714 */ 715 ssthresh = tp->snd_ssthresh; 716 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 717 /* 718 * convert the limit from user data bytes to 719 * packets then to packet data bytes. 720 */ 721 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 722 if (ssthresh < 2) 723 ssthresh = 2; 724 ssthresh *= (u_long)(tp->t_maxseg + 725#ifdef INET6 726 (isipv6 ? sizeof (struct ip6_hdr) + 727 sizeof (struct tcphdr) : 728#endif 729 sizeof (struct tcpiphdr) 730#ifdef INET6 731 ) 732#endif 733 ); 734 } else 735 ssthresh = 0; 736 metrics.rmx_ssthresh = ssthresh; 737 738 metrics.rmx_rtt = tp->t_srtt; 739 metrics.rmx_rttvar = tp->t_rttvar; 740 /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 741 metrics.rmx_bandwidth = tp->snd_bandwidth; 742 metrics.rmx_cwnd = tp->snd_cwnd; 743 metrics.rmx_sendpipe = 0; 744 metrics.rmx_recvpipe = 0; 745 746 tcp_hc_update(&inp->inp_inc, &metrics); 747 } 748 749 /* free the reassembly queue, if any */ 750 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 751 LIST_REMOVE(q, tqe_q); 752 m_freem(q->tqe_m); 753 uma_zfree(tcp_reass_zone, q); 754 tp->t_segqlen--; 755 tcp_reass_qsize--; 756 } 757 tcp_free_sackholes(tp); 758 inp->inp_ppcb = NULL; 759 tp->t_inpcb = NULL; 760 uma_zfree(tcpcb_zone, tp); 761 soisdisconnected(so); 762} 763 764/* 765 * Close a TCP control block: 766 * discard all space held by the tcp 767 * discard internet protocol block 768 * wake up any sleepers 769 */ 770struct tcpcb * 771tcp_close(tp) 772 struct tcpcb *tp; 773{ 774 struct inpcb *inp = tp->t_inpcb; 775#ifdef INET6 776 struct socket *so = inp->inp_socket; 777#endif 778 779 tcp_discardcb(tp); 780#ifdef INET6 781 if (INP_CHECK_SOCKAF(so, AF_INET6)) 782 in6_pcbdetach(inp); 783 else 784#endif 785 in_pcbdetach(inp); 786 tcpstat.tcps_closed++; 787 return (NULL); 788} 789 790void 791tcp_drain() 792{ 793 if (do_tcpdrain) 794 { 795 struct inpcb *inpb; 796 struct tcpcb *tcpb; 797 struct tseg_qent *te; 798 799 /* 800 * Walk the tcpbs, if existing, and flush the reassembly queue, 801 * if there is one... 802 * XXX: The "Net/3" implementation doesn't imply that the TCP 803 * reassembly queue should be flushed, but in a situation 804 * where we're really low on mbufs, this is potentially 805 * usefull. 806 */ 807 INP_INFO_RLOCK(&tcbinfo); 808 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 809 if (inpb->inp_vflag & INP_TIMEWAIT) 810 continue; 811 INP_LOCK(inpb); 812 if ((tcpb = intotcpcb(inpb)) != NULL) { 813 while ((te = LIST_FIRST(&tcpb->t_segq)) 814 != NULL) { 815 LIST_REMOVE(te, tqe_q); 816 m_freem(te->tqe_m); 817 uma_zfree(tcp_reass_zone, te); 818 tcpb->t_segqlen--; 819 tcp_reass_qsize--; 820 } 821 } 822 INP_UNLOCK(inpb); 823 } 824 INP_INFO_RUNLOCK(&tcbinfo); 825 } 826} 827 828/* 829 * Notify a tcp user of an asynchronous error; 830 * store error as soft error, but wake up user 831 * (for now, won't do anything until can select for soft error). 832 * 833 * Do not wake up user since there currently is no mechanism for 834 * reporting soft errors (yet - a kqueue filter may be added). 835 */ 836static struct inpcb * 837tcp_notify(inp, error) 838 struct inpcb *inp; 839 int error; 840{ 841 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb; 842 843 /* 844 * Ignore some errors if we are hooked up. 845 * If connection hasn't completed, has retransmitted several times, 846 * and receives a second error, give up now. This is better 847 * than waiting a long time to establish a connection that 848 * can never complete. 849 */ 850 if (tp->t_state == TCPS_ESTABLISHED && 851 (error == EHOSTUNREACH || error == ENETUNREACH || 852 error == EHOSTDOWN)) { 853 return inp; 854 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 855 tp->t_softerror) { 856 tcp_drop(tp, error); 857 return (struct inpcb *)0; 858 } else { 859 tp->t_softerror = error; 860 return inp; 861 } 862#if 0 863 wakeup( &so->so_timeo); 864 sorwakeup(so); 865 sowwakeup(so); 866#endif 867} 868 869static int 870tcp_pcblist(SYSCTL_HANDLER_ARGS) 871{ 872 int error, i, n, s; 873 struct inpcb *inp, **inp_list; 874 inp_gen_t gencnt; 875 struct xinpgen xig; 876 877 /* 878 * The process of preparing the TCB list is too time-consuming and 879 * resource-intensive to repeat twice on every request. 880 */ 881 if (req->oldptr == NULL) { 882 n = tcbinfo.ipi_count; 883 req->oldidx = 2 * (sizeof xig) 884 + (n + n/8) * sizeof(struct xtcpcb); 885 return 0; 886 } 887 888 if (req->newptr != NULL) 889 return EPERM; 890 891 /* 892 * OK, now we're committed to doing something. 893 */ 894 s = splnet(); 895 INP_INFO_RLOCK(&tcbinfo); 896 gencnt = tcbinfo.ipi_gencnt; 897 n = tcbinfo.ipi_count; 898 INP_INFO_RUNLOCK(&tcbinfo); 899 splx(s); 900 901 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 902 + n * sizeof(struct xtcpcb)); 903 if (error != 0) 904 return (error); 905 906 xig.xig_len = sizeof xig; 907 xig.xig_count = n; 908 xig.xig_gen = gencnt; 909 xig.xig_sogen = so_gencnt; 910 error = SYSCTL_OUT(req, &xig, sizeof xig); 911 if (error) 912 return error; 913 914 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 915 if (inp_list == NULL) 916 return ENOMEM; 917 918 s = splnet(); 919 INP_INFO_RLOCK(&tcbinfo); 920 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 921 inp = LIST_NEXT(inp, inp_list)) { 922 INP_LOCK(inp); 923 if (inp->inp_gencnt <= gencnt) { 924 /* 925 * XXX: This use of cr_cansee(), introduced with 926 * TCP state changes, is not quite right, but for 927 * now, better than nothing. 928 */ 929 if (inp->inp_vflag & INP_TIMEWAIT) 930 error = cr_cansee(req->td->td_ucred, 931 intotw(inp)->tw_cred); 932 else 933 error = cr_canseesocket(req->td->td_ucred, 934 inp->inp_socket); 935 if (error == 0) 936 inp_list[i++] = inp; 937 } 938 INP_UNLOCK(inp); 939 } 940 INP_INFO_RUNLOCK(&tcbinfo); 941 splx(s); 942 n = i; 943 944 error = 0; 945 for (i = 0; i < n; i++) { 946 inp = inp_list[i]; 947 if (inp->inp_gencnt <= gencnt) { 948 struct xtcpcb xt; 949 caddr_t inp_ppcb; 950 xt.xt_len = sizeof xt; 951 /* XXX should avoid extra copy */ 952 bcopy(inp, &xt.xt_inp, sizeof *inp); 953 inp_ppcb = inp->inp_ppcb; 954 if (inp_ppcb == NULL) 955 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 956 else if (inp->inp_vflag & INP_TIMEWAIT) { 957 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 958 xt.xt_tp.t_state = TCPS_TIME_WAIT; 959 } else 960 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 961 if (inp->inp_socket != NULL) 962 sotoxsocket(inp->inp_socket, &xt.xt_socket); 963 else { 964 bzero(&xt.xt_socket, sizeof xt.xt_socket); 965 xt.xt_socket.xso_protocol = IPPROTO_TCP; 966 } 967 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 968 error = SYSCTL_OUT(req, &xt, sizeof xt); 969 } 970 } 971 if (!error) { 972 /* 973 * Give the user an updated idea of our state. 974 * If the generation differs from what we told 975 * her before, she knows that something happened 976 * while we were processing this request, and it 977 * might be necessary to retry. 978 */ 979 s = splnet(); 980 INP_INFO_RLOCK(&tcbinfo); 981 xig.xig_gen = tcbinfo.ipi_gencnt; 982 xig.xig_sogen = so_gencnt; 983 xig.xig_count = tcbinfo.ipi_count; 984 INP_INFO_RUNLOCK(&tcbinfo); 985 splx(s); 986 error = SYSCTL_OUT(req, &xig, sizeof xig); 987 } 988 free(inp_list, M_TEMP); 989 return error; 990} 991 992SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 993 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 994 995static int 996tcp_getcred(SYSCTL_HANDLER_ARGS) 997{ 998 struct xucred xuc; 999 struct sockaddr_in addrs[2]; 1000 struct inpcb *inp; 1001 int error, s; 1002 1003 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1004 if (error) 1005 return (error); 1006 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1007 if (error) 1008 return (error); 1009 s = splnet(); 1010 INP_INFO_RLOCK(&tcbinfo); 1011 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1012 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 1013 if (inp == NULL) { 1014 error = ENOENT; 1015 goto outunlocked; 1016 } 1017 INP_LOCK(inp); 1018 if (inp->inp_socket == NULL) { 1019 error = ENOENT; 1020 goto out; 1021 } 1022 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1023 if (error) 1024 goto out; 1025 cru2x(inp->inp_socket->so_cred, &xuc); 1026out: 1027 INP_UNLOCK(inp); 1028outunlocked: 1029 INP_INFO_RUNLOCK(&tcbinfo); 1030 splx(s); 1031 if (error == 0) 1032 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1033 return (error); 1034} 1035 1036SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1037 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1038 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1039 1040#ifdef INET6 1041static int 1042tcp6_getcred(SYSCTL_HANDLER_ARGS) 1043{ 1044 struct xucred xuc; 1045 struct sockaddr_in6 addrs[2]; 1046 struct inpcb *inp; 1047 int error, s, mapped = 0; 1048 1049 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 1050 if (error) 1051 return (error); 1052 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1053 if (error) 1054 return (error); 1055 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1056 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1057 mapped = 1; 1058 else 1059 return (EINVAL); 1060 } 1061 s = splnet(); 1062 INP_INFO_RLOCK(&tcbinfo); 1063 if (mapped == 1) 1064 inp = in_pcblookup_hash(&tcbinfo, 1065 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1066 addrs[1].sin6_port, 1067 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1068 addrs[0].sin6_port, 1069 0, NULL); 1070 else 1071 inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr, 1072 addrs[1].sin6_port, 1073 &addrs[0].sin6_addr, addrs[0].sin6_port, 1074 0, NULL); 1075 if (inp == NULL) { 1076 error = ENOENT; 1077 goto outunlocked; 1078 } 1079 INP_LOCK(inp); 1080 if (inp->inp_socket == NULL) { 1081 error = ENOENT; 1082 goto out; 1083 } 1084 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 1085 if (error) 1086 goto out; 1087 cru2x(inp->inp_socket->so_cred, &xuc); 1088out: 1089 INP_UNLOCK(inp); 1090outunlocked: 1091 INP_INFO_RUNLOCK(&tcbinfo); 1092 splx(s); 1093 if (error == 0) 1094 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1095 return (error); 1096} 1097 1098SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1099 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1100 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1101#endif 1102 1103 1104void 1105tcp_ctlinput(cmd, sa, vip) 1106 int cmd; 1107 struct sockaddr *sa; 1108 void *vip; 1109{ 1110 struct ip *ip = vip; 1111 struct tcphdr *th; 1112 struct in_addr faddr; 1113 struct inpcb *inp; 1114 struct tcpcb *tp; 1115 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1116 tcp_seq icmp_seq; 1117 int s; 1118 1119 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1120 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1121 return; 1122 1123 if (cmd == PRC_QUENCH) 1124 notify = tcp_quench; 1125 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1126 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1127 notify = tcp_drop_syn_sent; 1128 else if (cmd == PRC_MSGSIZE) 1129 notify = tcp_mtudisc; 1130 /* 1131 * Redirects don't need to be handled up here. 1132 */ 1133 else if (PRC_IS_REDIRECT(cmd)) 1134 return; 1135 /* 1136 * Hostdead is ugly because it goes linearly through all PCBs. 1137 * XXX: We never get this from ICMP, otherwise it makes an 1138 * excellent DoS attack on machines with many connections. 1139 */ 1140 else if (cmd == PRC_HOSTDEAD) 1141 ip = NULL; 1142 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1143 return; 1144 if (ip != NULL) { 1145 s = splnet(); 1146 th = (struct tcphdr *)((caddr_t)ip 1147 + (ip->ip_hl << 2)); 1148 INP_INFO_WLOCK(&tcbinfo); 1149 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 1150 ip->ip_src, th->th_sport, 0, NULL); 1151 if (inp != NULL) { 1152 INP_LOCK(inp); 1153 if (inp->inp_socket != NULL) { 1154 icmp_seq = htonl(th->th_seq); 1155 tp = intotcpcb(inp); 1156 if (SEQ_GEQ(icmp_seq, tp->snd_una) && 1157 SEQ_LT(icmp_seq, tp->snd_max)) 1158 inp = (*notify)(inp, inetctlerrmap[cmd]); 1159 } 1160 if (inp != NULL) 1161 INP_UNLOCK(inp); 1162 } else { 1163 struct in_conninfo inc; 1164 1165 inc.inc_fport = th->th_dport; 1166 inc.inc_lport = th->th_sport; 1167 inc.inc_faddr = faddr; 1168 inc.inc_laddr = ip->ip_src; 1169#ifdef INET6 1170 inc.inc_isipv6 = 0; 1171#endif 1172 syncache_unreach(&inc, th); 1173 } 1174 INP_INFO_WUNLOCK(&tcbinfo); 1175 splx(s); 1176 } else 1177 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 1178} 1179 1180#ifdef INET6 1181void 1182tcp6_ctlinput(cmd, sa, d) 1183 int cmd; 1184 struct sockaddr *sa; 1185 void *d; 1186{ 1187 struct tcphdr th; 1188 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1189 struct ip6_hdr *ip6; 1190 struct mbuf *m; 1191 struct ip6ctlparam *ip6cp = NULL; 1192 const struct sockaddr_in6 *sa6_src = NULL; 1193 int off; 1194 struct tcp_portonly { 1195 u_int16_t th_sport; 1196 u_int16_t th_dport; 1197 } *thp; 1198 1199 if (sa->sa_family != AF_INET6 || 1200 sa->sa_len != sizeof(struct sockaddr_in6)) 1201 return; 1202 1203 if (cmd == PRC_QUENCH) 1204 notify = tcp_quench; 1205 else if (cmd == PRC_MSGSIZE) 1206 notify = tcp_mtudisc; 1207 else if (!PRC_IS_REDIRECT(cmd) && 1208 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1209 return; 1210 1211 /* if the parameter is from icmp6, decode it. */ 1212 if (d != NULL) { 1213 ip6cp = (struct ip6ctlparam *)d; 1214 m = ip6cp->ip6c_m; 1215 ip6 = ip6cp->ip6c_ip6; 1216 off = ip6cp->ip6c_off; 1217 sa6_src = ip6cp->ip6c_src; 1218 } else { 1219 m = NULL; 1220 ip6 = NULL; 1221 off = 0; /* fool gcc */ 1222 sa6_src = &sa6_any; 1223 } 1224 1225 if (ip6 != NULL) { 1226 struct in_conninfo inc; 1227 /* 1228 * XXX: We assume that when IPV6 is non NULL, 1229 * M and OFF are valid. 1230 */ 1231 1232 /* check if we can safely examine src and dst ports */ 1233 if (m->m_pkthdr.len < off + sizeof(*thp)) 1234 return; 1235 1236 bzero(&th, sizeof(th)); 1237 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1238 1239 in6_pcbnotify(&tcbinfo, sa, th.th_dport, 1240 (struct sockaddr *)ip6cp->ip6c_src, 1241 th.th_sport, cmd, NULL, notify); 1242 1243 inc.inc_fport = th.th_dport; 1244 inc.inc_lport = th.th_sport; 1245 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1246 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1247 inc.inc_isipv6 = 1; 1248 syncache_unreach(&inc, &th); 1249 } else 1250 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1251 0, cmd, NULL, notify); 1252} 1253#endif /* INET6 */ 1254 1255 1256/* 1257 * Following is where TCP initial sequence number generation occurs. 1258 * 1259 * There are two places where we must use initial sequence numbers: 1260 * 1. In SYN-ACK packets. 1261 * 2. In SYN packets. 1262 * 1263 * All ISNs for SYN-ACK packets are generated by the syncache. See 1264 * tcp_syncache.c for details. 1265 * 1266 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1267 * depends on this property. In addition, these ISNs should be 1268 * unguessable so as to prevent connection hijacking. To satisfy 1269 * the requirements of this situation, the algorithm outlined in 1270 * RFC 1948 is used, with only small modifications. 1271 * 1272 * Implementation details: 1273 * 1274 * Time is based off the system timer, and is corrected so that it 1275 * increases by one megabyte per second. This allows for proper 1276 * recycling on high speed LANs while still leaving over an hour 1277 * before rollover. 1278 * 1279 * As reading the *exact* system time is too expensive to be done 1280 * whenever setting up a TCP connection, we increment the time 1281 * offset in two ways. First, a small random positive increment 1282 * is added to isn_offset for each connection that is set up. 1283 * Second, the function tcp_isn_tick fires once per clock tick 1284 * and increments isn_offset as necessary so that sequence numbers 1285 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1286 * random positive increments serve only to ensure that the same 1287 * exact sequence number is never sent out twice (as could otherwise 1288 * happen when a port is recycled in less than the system tick 1289 * interval.) 1290 * 1291 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1292 * between seeding of isn_secret. This is normally set to zero, 1293 * as reseeding should not be necessary. 1294 * 1295 */ 1296 1297#define ISN_BYTES_PER_SECOND 1048576 1298#define ISN_STATIC_INCREMENT 4096 1299#define ISN_RANDOM_INCREMENT (4096 - 1) 1300 1301u_char isn_secret[32]; 1302int isn_last_reseed; 1303u_int32_t isn_offset, isn_offset_old; 1304MD5_CTX isn_ctx; 1305 1306tcp_seq 1307tcp_new_isn(tp) 1308 struct tcpcb *tp; 1309{ 1310 u_int32_t md5_buffer[4]; 1311 tcp_seq new_isn; 1312 1313 /* Seed if this is the first use, reseed if requested. */ 1314 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 1315 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 1316 < (u_int)ticks))) { 1317 read_random(&isn_secret, sizeof(isn_secret)); 1318 isn_last_reseed = ticks; 1319 } 1320 1321 /* Compute the md5 hash and return the ISN. */ 1322 MD5Init(&isn_ctx); 1323 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1324 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1325#ifdef INET6 1326 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1327 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1328 sizeof(struct in6_addr)); 1329 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1330 sizeof(struct in6_addr)); 1331 } else 1332#endif 1333 { 1334 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1335 sizeof(struct in_addr)); 1336 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1337 sizeof(struct in_addr)); 1338 } 1339 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 1340 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1341 new_isn = (tcp_seq) md5_buffer[0]; 1342 isn_offset += ISN_STATIC_INCREMENT + 1343 (arc4random() & ISN_RANDOM_INCREMENT); 1344 new_isn += isn_offset; 1345 return new_isn; 1346} 1347 1348/* 1349 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1350 * to keep time flowing at a relatively constant rate. If the random 1351 * increments have already pushed us past the projected offset, do nothing. 1352 */ 1353static void 1354tcp_isn_tick(xtp) 1355 void *xtp; 1356{ 1357 u_int32_t projected_offset; 1358 1359 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / hz; 1360 1361 if (projected_offset > isn_offset) 1362 isn_offset = projected_offset; 1363 1364 isn_offset_old = isn_offset; 1365 callout_reset(&isn_callout, 1, tcp_isn_tick, NULL); 1366} 1367 1368/* 1369 * When a source quench is received, close congestion window 1370 * to one segment. We will gradually open it again as we proceed. 1371 */ 1372struct inpcb * 1373tcp_quench(inp, errno) 1374 struct inpcb *inp; 1375 int errno; 1376{ 1377 struct tcpcb *tp = intotcpcb(inp); 1378 1379 if (tp != NULL) 1380 tp->snd_cwnd = tp->t_maxseg; 1381 return (inp); 1382} 1383 1384/* 1385 * When a specific ICMP unreachable message is received and the 1386 * connection state is SYN-SENT, drop the connection. This behavior 1387 * is controlled by the icmp_may_rst sysctl. 1388 */ 1389struct inpcb * 1390tcp_drop_syn_sent(inp, errno) 1391 struct inpcb *inp; 1392 int errno; 1393{ 1394 struct tcpcb *tp = intotcpcb(inp); 1395 1396 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) { 1397 tcp_drop(tp, errno); 1398 return (struct inpcb *)0; 1399 } 1400 return inp; 1401} 1402 1403/* 1404 * When `need fragmentation' ICMP is received, update our idea of the MSS 1405 * based on the new value in the route. Also nudge TCP to send something, 1406 * since we know the packet we just sent was dropped. 1407 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1408 */ 1409struct inpcb * 1410tcp_mtudisc(inp, errno) 1411 struct inpcb *inp; 1412 int errno; 1413{ 1414 struct tcpcb *tp = intotcpcb(inp); 1415 struct rmxp_tao tao; 1416 struct socket *so = inp->inp_socket; 1417 u_int maxmtu; 1418 u_int romtu; 1419 int mss; 1420#ifdef INET6 1421 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1422#endif /* INET6 */ 1423 bzero(&tao, sizeof(tao)); 1424 1425 if (tp != NULL) { 1426 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1427 romtu = 1428#ifdef INET6 1429 isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1430#endif /* INET6 */ 1431 tcp_maxmtu(&inp->inp_inc); 1432 if (!maxmtu) 1433 maxmtu = romtu; 1434 else 1435 maxmtu = min(maxmtu, romtu); 1436 if (!maxmtu) { 1437 tp->t_maxopd = tp->t_maxseg = 1438#ifdef INET6 1439 isipv6 ? tcp_v6mssdflt : 1440#endif /* INET6 */ 1441 tcp_mssdflt; 1442 return inp; 1443 } 1444 mss = maxmtu - 1445#ifdef INET6 1446 (isipv6 ? 1447 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 1448#endif /* INET6 */ 1449 sizeof(struct tcpiphdr) 1450#ifdef INET6 1451 ) 1452#endif /* INET6 */ 1453 ; 1454 1455 if (tcp_do_rfc1644) { 1456 tcp_hc_gettao(&inp->inp_inc, &tao); 1457 if (tao.tao_mssopt) 1458 mss = min(mss, tao.tao_mssopt); 1459 } 1460 /* 1461 * XXX - The above conditional probably violates the TCP 1462 * spec. The problem is that, since we don't know the 1463 * other end's MSS, we are supposed to use a conservative 1464 * default. But, if we do that, then MTU discovery will 1465 * never actually take place, because the conservative 1466 * default is much less than the MTUs typically seen 1467 * on the Internet today. For the moment, we'll sweep 1468 * this under the carpet. 1469 * 1470 * The conservative default might not actually be a problem 1471 * if the only case this occurs is when sending an initial 1472 * SYN with options and data to a host we've never talked 1473 * to before. Then, they will reply with an MSS value which 1474 * will get recorded and the new parameters should get 1475 * recomputed. For Further Study. 1476 */ 1477 if (tp->t_maxopd <= mss) 1478 return inp; 1479 tp->t_maxopd = mss; 1480 1481 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1482 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1483 mss -= TCPOLEN_TSTAMP_APPA; 1484 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 1485 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC) 1486 mss -= TCPOLEN_CC_APPA; 1487#if (MCLBYTES & (MCLBYTES - 1)) == 0 1488 if (mss > MCLBYTES) 1489 mss &= ~(MCLBYTES-1); 1490#else 1491 if (mss > MCLBYTES) 1492 mss = mss / MCLBYTES * MCLBYTES; 1493#endif 1494 if (so->so_snd.sb_hiwat < mss) 1495 mss = so->so_snd.sb_hiwat; 1496 1497 tp->t_maxseg = mss; 1498 1499 tcpstat.tcps_mturesent++; 1500 tp->t_rtttime = 0; 1501 tp->snd_nxt = tp->snd_una; 1502 tcp_output(tp); 1503 } 1504 return inp; 1505} 1506 1507/* 1508 * Look-up the routing entry to the peer of this inpcb. If no route 1509 * is found and it cannot be allocated, then return NULL. This routine 1510 * is called by TCP routines that access the rmx structure and by tcp_mss 1511 * to get the interface MTU. 1512 */ 1513u_long 1514tcp_maxmtu(inc) 1515 struct in_conninfo *inc; 1516{ 1517 struct route sro; 1518 struct sockaddr_in *dst; 1519 struct ifnet *ifp; 1520 u_long maxmtu = 0; 1521 1522 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1523 1524 bzero(&sro, sizeof(sro)); 1525 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1526 dst = (struct sockaddr_in *)&sro.ro_dst; 1527 dst->sin_family = AF_INET; 1528 dst->sin_len = sizeof(*dst); 1529 dst->sin_addr = inc->inc_faddr; 1530 rtalloc_ign(&sro, RTF_CLONING); 1531 } 1532 if (sro.ro_rt != NULL) { 1533 ifp = sro.ro_rt->rt_ifp; 1534 if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1535 maxmtu = ifp->if_mtu; 1536 else 1537 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1538 RTFREE(sro.ro_rt); 1539 } 1540 return (maxmtu); 1541} 1542 1543#ifdef INET6 1544u_long 1545tcp_maxmtu6(inc) 1546 struct in_conninfo *inc; 1547{ 1548 struct route_in6 sro6; 1549 struct ifnet *ifp; 1550 u_long maxmtu = 0; 1551 1552 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1553 1554 bzero(&sro6, sizeof(sro6)); 1555 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1556 sro6.ro_dst.sin6_family = AF_INET6; 1557 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1558 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1559 rtalloc_ign((struct route *)&sro6, RTF_CLONING); 1560 } 1561 if (sro6.ro_rt != NULL) { 1562 ifp = sro6.ro_rt->rt_ifp; 1563 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1564 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1565 else 1566 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1567 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1568 RTFREE(sro6.ro_rt); 1569 } 1570 1571 return (maxmtu); 1572} 1573#endif /* INET6 */ 1574 1575#ifdef IPSEC 1576/* compute ESP/AH header size for TCP, including outer IP header. */ 1577size_t 1578ipsec_hdrsiz_tcp(tp) 1579 struct tcpcb *tp; 1580{ 1581 struct inpcb *inp; 1582 struct mbuf *m; 1583 size_t hdrsiz; 1584 struct ip *ip; 1585#ifdef INET6 1586 struct ip6_hdr *ip6; 1587#endif 1588 struct tcphdr *th; 1589 1590 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1591 return 0; 1592 MGETHDR(m, M_DONTWAIT, MT_DATA); 1593 if (!m) 1594 return 0; 1595 1596#ifdef INET6 1597 if ((inp->inp_vflag & INP_IPV6) != 0) { 1598 ip6 = mtod(m, struct ip6_hdr *); 1599 th = (struct tcphdr *)(ip6 + 1); 1600 m->m_pkthdr.len = m->m_len = 1601 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1602 tcpip_fillheaders(inp, ip6, th); 1603 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1604 } else 1605#endif /* INET6 */ 1606 { 1607 ip = mtod(m, struct ip *); 1608 th = (struct tcphdr *)(ip + 1); 1609 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1610 tcpip_fillheaders(inp, ip, th); 1611 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1612 } 1613 1614 m_free(m); 1615 return hdrsiz; 1616} 1617#endif /*IPSEC*/ 1618 1619/* 1620 * Move a TCP connection into TIME_WAIT state. 1621 * tcbinfo is unlocked. 1622 * inp is locked, and is unlocked before returning. 1623 */ 1624void 1625tcp_twstart(tp) 1626 struct tcpcb *tp; 1627{ 1628 struct tcptw *tw; 1629 struct inpcb *inp; 1630 int tw_time, acknow; 1631 struct socket *so; 1632 1633 tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1634 if (tw == NULL) { 1635 tw = tcp_timer_2msl_tw(1); 1636 if (tw == NULL) { 1637 tcp_close(tp); 1638 return; 1639 } 1640 } 1641 inp = tp->t_inpcb; 1642 tw->tw_inpcb = inp; 1643 1644 /* 1645 * Recover last window size sent. 1646 */ 1647 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1648 1649 /* 1650 * Set t_recent if timestamps are used on the connection. 1651 */ 1652 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1653 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1654 tw->t_recent = tp->ts_recent; 1655 else 1656 tw->t_recent = 0; 1657 1658 tw->snd_nxt = tp->snd_nxt; 1659 tw->rcv_nxt = tp->rcv_nxt; 1660 tw->iss = tp->iss; 1661 tw->irs = tp->irs; 1662 tw->cc_recv = tp->cc_recv; 1663 tw->cc_send = tp->cc_send; 1664 tw->t_starttime = tp->t_starttime; 1665 tw->tw_time = 0; 1666 1667/* XXX 1668 * If this code will 1669 * be used for fin-wait-2 state also, then we may need 1670 * a ts_recent from the last segment. 1671 */ 1672 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1673 if (tp->cc_recv != 0 && (ticks - tp->t_starttime) < tcp_msl) { 1674 tw_time = tp->t_rxtcur * TCPTV_TWTRUNC; 1675 /* For T/TCP client, force ACK now. */ 1676 acknow = 1; 1677 } else { 1678 tw_time = 2 * tcp_msl; 1679 acknow = tp->t_flags & TF_ACKNOW; 1680 } 1681 tcp_discardcb(tp); 1682 so = inp->inp_socket; 1683 SOCK_LOCK(so); 1684 so->so_pcb = NULL; 1685 tw->tw_cred = crhold(so->so_cred); 1686 tw->tw_so_options = so->so_options; 1687 sotryfree(so); 1688 inp->inp_socket = NULL; 1689 if (acknow) 1690 tcp_twrespond(tw, TH_ACK); 1691 inp->inp_ppcb = (caddr_t)tw; 1692 inp->inp_vflag |= INP_TIMEWAIT; 1693 tcp_timer_2msl_reset(tw, tw_time); 1694 INP_UNLOCK(inp); 1695} 1696 1697/* 1698 * The appromixate rate of ISN increase of Microsoft TCP stacks; 1699 * the actual rate is slightly higher due to the addition of 1700 * random positive increments. 1701 * 1702 * Most other new OSes use semi-randomized ISN values, so we 1703 * do not need to worry about them. 1704 */ 1705#define MS_ISN_BYTES_PER_SECOND 250000 1706 1707/* 1708 * Determine if the ISN we will generate has advanced beyond the last 1709 * sequence number used by the previous connection. If so, indicate 1710 * that it is safe to recycle this tw socket by returning 1. 1711 */ 1712int 1713tcp_twrecycleable(struct tcptw *tw) 1714{ 1715 tcp_seq new_iss = tw->iss; 1716 tcp_seq new_irs = tw->irs; 1717 1718 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1719 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1720 1721 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1722 return 1; 1723 else 1724 return 0; 1725} 1726 1727struct tcptw * 1728tcp_twclose(struct tcptw *tw, int reuse) 1729{ 1730 struct inpcb *inp; 1731 1732 inp = tw->tw_inpcb; 1733 tw->tw_inpcb = NULL; 1734 tcp_timer_2msl_stop(tw); 1735 inp->inp_ppcb = NULL; 1736#ifdef INET6 1737 if (inp->inp_vflag & INP_IPV6PROTO) 1738 in6_pcbdetach(inp); 1739 else 1740#endif 1741 in_pcbdetach(inp); 1742 tcpstat.tcps_closed++; 1743 crfree(tw->tw_cred); 1744 tw->tw_cred = NULL; 1745 if (reuse) 1746 return (tw); 1747 uma_zfree(tcptw_zone, tw); 1748 return (NULL); 1749} 1750 1751int 1752tcp_twrespond(struct tcptw *tw, int flags) 1753{ 1754 struct inpcb *inp = tw->tw_inpcb; 1755 struct tcphdr *th; 1756 struct mbuf *m; 1757 struct ip *ip = NULL; 1758 u_int8_t *optp; 1759 u_int hdrlen, optlen; 1760 int error; 1761#ifdef INET6 1762 struct ip6_hdr *ip6 = NULL; 1763 int isipv6 = inp->inp_inc.inc_isipv6; 1764#endif 1765 1766 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1767 if (m == NULL) 1768 return (ENOBUFS); 1769 m->m_data += max_linkhdr; 1770 1771#ifdef MAC 1772 mac_create_mbuf_from_inpcb(inp, m); 1773#endif 1774 1775#ifdef INET6 1776 if (isipv6) { 1777 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1778 ip6 = mtod(m, struct ip6_hdr *); 1779 th = (struct tcphdr *)(ip6 + 1); 1780 tcpip_fillheaders(inp, ip6, th); 1781 } else 1782#endif 1783 { 1784 hdrlen = sizeof(struct tcpiphdr); 1785 ip = mtod(m, struct ip *); 1786 th = (struct tcphdr *)(ip + 1); 1787 tcpip_fillheaders(inp, ip, th); 1788 } 1789 optp = (u_int8_t *)(th + 1); 1790 1791 /* 1792 * Send a timestamp and echo-reply if both our side and our peer 1793 * have sent timestamps in our SYN's and this is not a RST. 1794 */ 1795 if (tw->t_recent && flags == TH_ACK) { 1796 u_int32_t *lp = (u_int32_t *)optp; 1797 1798 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1799 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1800 *lp++ = htonl(ticks); 1801 *lp = htonl(tw->t_recent); 1802 optp += TCPOLEN_TSTAMP_APPA; 1803 } 1804 1805 /* 1806 * Send `CC-family' options if needed, and it's not a RST. 1807 */ 1808 if (tw->cc_recv != 0 && flags == TH_ACK) { 1809 u_int32_t *lp = (u_int32_t *)optp; 1810 1811 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1812 *lp = htonl(tw->cc_send); 1813 optp += TCPOLEN_CC_APPA; 1814 } 1815 optlen = optp - (u_int8_t *)(th + 1); 1816 1817 m->m_len = hdrlen + optlen; 1818 m->m_pkthdr.len = m->m_len; 1819 1820 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1821 1822 th->th_seq = htonl(tw->snd_nxt); 1823 th->th_ack = htonl(tw->rcv_nxt); 1824 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1825 th->th_flags = flags; 1826 th->th_win = htons(tw->last_win); 1827 1828#ifdef INET6 1829 if (isipv6) { 1830 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1831 sizeof(struct tcphdr) + optlen); 1832 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1833 error = ip6_output(m, inp->in6p_outputopts, NULL, 1834 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1835 } else 1836#endif 1837 { 1838 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1839 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1840 m->m_pkthdr.csum_flags = CSUM_TCP; 1841 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1842 ip->ip_len = m->m_pkthdr.len; 1843 if (path_mtu_discovery) 1844 ip->ip_off |= IP_DF; 1845 error = ip_output(m, inp->inp_options, NULL, 1846 (tw->tw_so_options & SO_DONTROUTE), NULL, inp); 1847 } 1848 if (flags & TH_ACK) 1849 tcpstat.tcps_sndacks++; 1850 else 1851 tcpstat.tcps_sndctrl++; 1852 tcpstat.tcps_sndtotal++; 1853 return (error); 1854} 1855 1856/* 1857 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 1858 * 1859 * This code attempts to calculate the bandwidth-delay product as a 1860 * means of determining the optimal window size to maximize bandwidth, 1861 * minimize RTT, and avoid the over-allocation of buffers on interfaces and 1862 * routers. This code also does a fairly good job keeping RTTs in check 1863 * across slow links like modems. We implement an algorithm which is very 1864 * similar (but not meant to be) TCP/Vegas. The code operates on the 1865 * transmitter side of a TCP connection and so only effects the transmit 1866 * side of the connection. 1867 * 1868 * BACKGROUND: TCP makes no provision for the management of buffer space 1869 * at the end points or at the intermediate routers and switches. A TCP 1870 * stream, whether using NewReno or not, will eventually buffer as 1871 * many packets as it is able and the only reason this typically works is 1872 * due to the fairly small default buffers made available for a connection 1873 * (typicaly 16K or 32K). As machines use larger windows and/or window 1874 * scaling it is now fairly easy for even a single TCP connection to blow-out 1875 * all available buffer space not only on the local interface, but on 1876 * intermediate routers and switches as well. NewReno makes a misguided 1877 * attempt to 'solve' this problem by waiting for an actual failure to occur, 1878 * then backing off, then steadily increasing the window again until another 1879 * failure occurs, ad-infinitum. This results in terrible oscillation that 1880 * is only made worse as network loads increase and the idea of intentionally 1881 * blowing out network buffers is, frankly, a terrible way to manage network 1882 * resources. 1883 * 1884 * It is far better to limit the transmit window prior to the failure 1885 * condition being achieved. There are two general ways to do this: First 1886 * you can 'scan' through different transmit window sizes and locate the 1887 * point where the RTT stops increasing, indicating that you have filled the 1888 * pipe, then scan backwards until you note that RTT stops decreasing, then 1889 * repeat ad-infinitum. This method works in principle but has severe 1890 * implementation issues due to RTT variances, timer granularity, and 1891 * instability in the algorithm which can lead to many false positives and 1892 * create oscillations as well as interact badly with other TCP streams 1893 * implementing the same algorithm. 1894 * 1895 * The second method is to limit the window to the bandwidth delay product 1896 * of the link. This is the method we implement. RTT variances and our 1897 * own manipulation of the congestion window, bwnd, can potentially 1898 * destabilize the algorithm. For this reason we have to stabilize the 1899 * elements used to calculate the window. We do this by using the minimum 1900 * observed RTT, the long term average of the observed bandwidth, and 1901 * by adding two segments worth of slop. It isn't perfect but it is able 1902 * to react to changing conditions and gives us a very stable basis on 1903 * which to extend the algorithm. 1904 */ 1905void 1906tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 1907{ 1908 u_long bw; 1909 u_long bwnd; 1910 int save_ticks; 1911 1912 /* 1913 * If inflight_enable is disabled in the middle of a tcp connection, 1914 * make sure snd_bwnd is effectively disabled. 1915 */ 1916 if (tcp_inflight_enable == 0) { 1917 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1918 tp->snd_bandwidth = 0; 1919 return; 1920 } 1921 1922 /* 1923 * Figure out the bandwidth. Due to the tick granularity this 1924 * is a very rough number and it MUST be averaged over a fairly 1925 * long period of time. XXX we need to take into account a link 1926 * that is not using all available bandwidth, but for now our 1927 * slop will ramp us up if this case occurs and the bandwidth later 1928 * increases. 1929 * 1930 * Note: if ticks rollover 'bw' may wind up negative. We must 1931 * effectively reset t_bw_rtttime for this case. 1932 */ 1933 save_ticks = ticks; 1934 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 1935 return; 1936 1937 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 1938 (save_ticks - tp->t_bw_rtttime); 1939 tp->t_bw_rtttime = save_ticks; 1940 tp->t_bw_rtseq = ack_seq; 1941 if (tp->t_bw_rtttime == 0 || (int)bw < 0) 1942 return; 1943 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 1944 1945 tp->snd_bandwidth = bw; 1946 1947 /* 1948 * Calculate the semi-static bandwidth delay product, plus two maximal 1949 * segments. The additional slop puts us squarely in the sweet 1950 * spot and also handles the bandwidth run-up case and stabilization. 1951 * Without the slop we could be locking ourselves into a lower 1952 * bandwidth. 1953 * 1954 * Situations Handled: 1955 * (1) Prevents over-queueing of packets on LANs, especially on 1956 * high speed LANs, allowing larger TCP buffers to be 1957 * specified, and also does a good job preventing 1958 * over-queueing of packets over choke points like modems 1959 * (at least for the transmit side). 1960 * 1961 * (2) Is able to handle changing network loads (bandwidth 1962 * drops so bwnd drops, bandwidth increases so bwnd 1963 * increases). 1964 * 1965 * (3) Theoretically should stabilize in the face of multiple 1966 * connections implementing the same algorithm (this may need 1967 * a little work). 1968 * 1969 * (4) Stability value (defaults to 20 = 2 maximal packets) can 1970 * be adjusted with a sysctl but typically only needs to be 1971 * on very slow connections. A value no smaller then 5 1972 * should be used, but only reduce this default if you have 1973 * no other choice. 1974 */ 1975#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 1976 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 1977#undef USERTT 1978 1979 if (tcp_inflight_debug > 0) { 1980 static int ltime; 1981 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 1982 ltime = ticks; 1983 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 1984 tp, 1985 bw, 1986 tp->t_rttbest, 1987 tp->t_srtt, 1988 bwnd 1989 ); 1990 } 1991 } 1992 if ((long)bwnd < tcp_inflight_min) 1993 bwnd = tcp_inflight_min; 1994 if (bwnd > tcp_inflight_max) 1995 bwnd = tcp_inflight_max; 1996 if ((long)bwnd < tp->t_maxseg * 2) 1997 bwnd = tp->t_maxseg * 2; 1998 tp->snd_bwnd = bwnd; 1999} 2000 2001#ifdef TCP_SIGNATURE 2002/* 2003 * Callback function invoked by m_apply() to digest TCP segment data 2004 * contained within an mbuf chain. 2005 */ 2006static int 2007tcp_signature_apply(void *fstate, void *data, u_int len) 2008{ 2009 2010 MD5Update(fstate, (u_char *)data, len); 2011 return (0); 2012} 2013 2014/* 2015 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2016 * 2017 * Parameters: 2018 * m pointer to head of mbuf chain 2019 * off0 offset to TCP header within the mbuf chain 2020 * len length of TCP segment data, excluding options 2021 * optlen length of TCP segment options 2022 * buf pointer to storage for computed MD5 digest 2023 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2024 * 2025 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2026 * When called from tcp_input(), we can be sure that th_sum has been 2027 * zeroed out and verified already. 2028 * 2029 * This function is for IPv4 use only. Calling this function with an 2030 * IPv6 packet in the mbuf chain will yield undefined results. 2031 * 2032 * Return 0 if successful, otherwise return -1. 2033 * 2034 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2035 * search with the destination IP address, and a 'magic SPI' to be 2036 * determined by the application. This is hardcoded elsewhere to 1179 2037 * right now. Another branch of this code exists which uses the SPD to 2038 * specify per-application flows but it is unstable. 2039 */ 2040int 2041tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2042 u_char *buf, u_int direction) 2043{ 2044 union sockaddr_union dst; 2045 struct ippseudo ippseudo; 2046 MD5_CTX ctx; 2047 int doff; 2048 struct ip *ip; 2049 struct ipovly *ipovly; 2050 struct secasvar *sav; 2051 struct tcphdr *th; 2052 u_short savecsum; 2053 2054 KASSERT(m != NULL, ("NULL mbuf chain")); 2055 KASSERT(buf != NULL, ("NULL signature pointer")); 2056 2057 /* Extract the destination from the IP header in the mbuf. */ 2058 ip = mtod(m, struct ip *); 2059 bzero(&dst, sizeof(union sockaddr_union)); 2060 dst.sa.sa_len = sizeof(struct sockaddr_in); 2061 dst.sa.sa_family = AF_INET; 2062 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2063 ip->ip_src : ip->ip_dst; 2064 2065 /* Look up an SADB entry which matches the address of the peer. */ 2066 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2067 if (sav == NULL) { 2068 printf("%s: SADB lookup failed for %s\n", __func__, 2069 inet_ntoa(dst.sin.sin_addr)); 2070 return (EINVAL); 2071 } 2072 2073 MD5Init(&ctx); 2074 ipovly = (struct ipovly *)ip; 2075 th = (struct tcphdr *)((u_char *)ip + off0); 2076 doff = off0 + sizeof(struct tcphdr) + optlen; 2077 2078 /* 2079 * Step 1: Update MD5 hash with IP pseudo-header. 2080 * 2081 * XXX The ippseudo header MUST be digested in network byte order, 2082 * or else we'll fail the regression test. Assume all fields we've 2083 * been doing arithmetic on have been in host byte order. 2084 * XXX One cannot depend on ipovly->ih_len here. When called from 2085 * tcp_output(), the underlying ip_len member has not yet been set. 2086 */ 2087 ippseudo.ippseudo_src = ipovly->ih_src; 2088 ippseudo.ippseudo_dst = ipovly->ih_dst; 2089 ippseudo.ippseudo_pad = 0; 2090 ippseudo.ippseudo_p = IPPROTO_TCP; 2091 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2092 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2093 2094 /* 2095 * Step 2: Update MD5 hash with TCP header, excluding options. 2096 * The TCP checksum must be set to zero. 2097 */ 2098 savecsum = th->th_sum; 2099 th->th_sum = 0; 2100 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2101 th->th_sum = savecsum; 2102 2103 /* 2104 * Step 3: Update MD5 hash with TCP segment data. 2105 * Use m_apply() to avoid an early m_pullup(). 2106 */ 2107 if (len > 0) 2108 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2109 2110 /* 2111 * Step 4: Update MD5 hash with shared secret. 2112 */ 2113 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2114 MD5Final(buf, &ctx); 2115 2116 key_sa_recordxfer(sav, m); 2117 KEY_FREESAV(&sav); 2118 return (0); 2119} 2120#endif /* TCP_SIGNATURE */ 2121