tcp_timewait.c revision 162064
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 3050477Speter * $FreeBSD: head/sys/netinet/tcp_timewait.c 162064 2006-09-06 13:56:35Z glebius $ 311541Srgrimes */ 321541Srgrimes 3332752Seivind#include "opt_compat.h" 34125680Sbms#include "opt_inet.h" 3554263Sshin#include "opt_inet6.h" 3656041Sshin#include "opt_ipsec.h" 37101106Srwatson#include "opt_mac.h" 3829514Sjoerg#include "opt_tcpdebug.h" 39130989Sps#include "opt_tcp_sack.h" 4029514Sjoerg 411541Srgrimes#include <sys/param.h> 421541Srgrimes#include <sys/systm.h> 4350673Sjlemon#include <sys/callout.h> 4412172Sphk#include <sys/kernel.h> 4512172Sphk#include <sys/sysctl.h> 46101106Srwatson#include <sys/mac.h> 471541Srgrimes#include <sys/malloc.h> 481541Srgrimes#include <sys/mbuf.h> 4955679Sshin#ifdef INET6 5055679Sshin#include <sys/domain.h> 5155679Sshin#endif 5248758Sgreen#include <sys/proc.h> 531541Srgrimes#include <sys/socket.h> 541541Srgrimes#include <sys/socketvar.h> 551541Srgrimes#include <sys/protosw.h> 5675619Skris#include <sys/random.h> 5734923Sbde 5892760Sjeff#include <vm/uma.h> 591541Srgrimes 601541Srgrimes#include <net/route.h> 611541Srgrimes#include <net/if.h> 621541Srgrimes 631541Srgrimes#include <netinet/in.h> 641541Srgrimes#include <netinet/in_systm.h> 651541Srgrimes#include <netinet/ip.h> 6655679Sshin#ifdef INET6 6755679Sshin#include <netinet/ip6.h> 6855679Sshin#endif 691541Srgrimes#include <netinet/in_pcb.h> 7055679Sshin#ifdef INET6 7155679Sshin#include <netinet6/in6_pcb.h> 7255679Sshin#endif 737090Sbde#include <netinet/in_var.h> 741541Srgrimes#include <netinet/ip_var.h> 7555679Sshin#ifdef INET6 7655679Sshin#include <netinet6/ip6_var.h> 77148385Sume#include <netinet6/scope6_var.h> 78122922Sandre#include <netinet6/nd6.h> 7955679Sshin#endif 80145360Sandre#include <netinet/ip_icmp.h> 811541Srgrimes#include <netinet/tcp.h> 821541Srgrimes#include <netinet/tcp_fsm.h> 831541Srgrimes#include <netinet/tcp_seq.h> 841541Srgrimes#include <netinet/tcp_timer.h> 851541Srgrimes#include <netinet/tcp_var.h> 8655679Sshin#ifdef INET6 8755679Sshin#include <netinet6/tcp6_var.h> 8855679Sshin#endif 891541Srgrimes#include <netinet/tcpip.h> 906283Swollman#ifdef TCPDEBUG 916283Swollman#include <netinet/tcp_debug.h> 926283Swollman#endif 9355679Sshin#include <netinet6/ip6protosw.h> 941541Srgrimes 9555679Sshin#ifdef IPSEC 9655679Sshin#include <netinet6/ipsec.h> 9762587Sitojun#ifdef INET6 9862587Sitojun#include <netinet6/ipsec6.h> 9962587Sitojun#endif 100137396Ssuz#include <netkey/key.h> 10155679Sshin#endif /*IPSEC*/ 10255679Sshin 103105199Ssam#ifdef FAST_IPSEC 104105199Ssam#include <netipsec/ipsec.h> 105125680Sbms#include <netipsec/xform.h> 106105199Ssam#ifdef INET6 107105199Ssam#include <netipsec/ipsec6.h> 108105199Ssam#endif 109125680Sbms#include <netipsec/key.h> 110105199Ssam#define IPSEC 111105199Ssam#endif /*FAST_IPSEC*/ 112105199Ssam 11358698Sjlemon#include <machine/in_cksum.h> 11482122Ssilby#include <sys/md5.h> 11558698Sjlemon 116133874Srwatsonint tcp_mssdflt = TCP_MSS; 117133874SrwatsonSYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW, 11846381Sbillf &tcp_mssdflt , 0, "Default TCP Maximum Segment Size"); 11912296Sphk 12052904Sshin#ifdef INET6 12152904Sshinint tcp_v6mssdflt = TCP6_MSS; 12252904SshinSYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 12355679Sshin CTLFLAG_RW, &tcp_v6mssdflt , 0, 12455679Sshin "Default TCP Maximum Segment Size for IPv6"); 12552904Sshin#endif 12652904Sshin 127124258Sandre/* 128124258Sandre * Minimum MSS we accept and use. This prevents DoS attacks where 129124258Sandre * we are forced to a ridiculous low MSS like 20 and send hundreds 130124258Sandre * of packets instead of one. The effect scales with the available 131124258Sandre * bandwidth and quickly saturates the CPU and network interface 132124258Sandre * with packet generation and sending. Set to zero to disable MINMSS 133124258Sandre * checking. This setting prevents us from sending too small packets. 134124258Sandre */ 135124258Sandreint tcp_minmss = TCP_MINMSS; 136124258SandreSYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 137124258Sandre &tcp_minmss , 0, "Minmum TCP Maximum Segment Size"); 138124258Sandre/* 139124258Sandre * Number of TCP segments per second we accept from remote host 140124258Sandre * before we start to calculate average segment size. If average 141124258Sandre * segment size drops below the minimum TCP MSS we assume a DoS 142124258Sandre * attack and reset+drop the connection. Care has to be taken not to 143124258Sandre * set this value too small to not kill interactive type connections 144124258Sandre * (telnet, SSH) which send many small packets. 145124258Sandre */ 146124258Sandreint tcp_minmssoverload = TCP_MINMSSOVERLOAD; 147124258SandreSYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW, 148124258Sandre &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to" 149124258Sandre "be under the MINMSS Size"); 150124258Sandre 15150673Sjlemon#if 0 152133874Srwatsonstatic int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ; 153133874SrwatsonSYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW, 15446381Sbillf &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time"); 15550673Sjlemon#endif 15612296Sphk 15786764Sjlemonint tcp_do_rfc1323 = 1; 158133874SrwatsonSYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 15946381Sbillf &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions"); 16012296Sphk 16150426Sjlemonstatic int tcp_tcbhashsize = 0; 162121307SsilbySYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 16350426Sjlemon &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 16450426Sjlemon 16555198Smsmithstatic int do_tcpdrain = 1; 16666376SbmilekicSYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 16766376Sbmilekic "Enable tcp_drain routine for extra help when low on mbufs"); 16855198Smsmith 169133874SrwatsonSYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 17046381Sbillf &tcbinfo.ipi_count, 0, "Number of active PCBs"); 17136079Swollman 17272959Sjlemonstatic int icmp_may_rst = 1; 173133874SrwatsonSYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0, 17472959Sjlemon "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 17570103Sphk 17682122Ssilbystatic int tcp_isn_reseed_interval = 0; 17782122SsilbySYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 17882122Ssilby &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret"); 17982122Ssilby 180162035Sglebiusstatic uma_zone_t tcptw_zone; 181157478Sglebiusstatic int maxtcptw; 182162031Sglebiusstatic int 183162031Sglebiussysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 184162031Sglebius{ 185162031Sglebius int error, new; 186157478Sglebius 187162031Sglebius if (maxtcptw == 0) 188162031Sglebius new = maxsockets / 5; 189162031Sglebius else 190162031Sglebius new = maxtcptw; 191162031Sglebius error = sysctl_handle_int(oidp, &new, sizeof(int), req); 192162031Sglebius if (error == 0 && req->newptr) { 193162035Sglebius if (new > maxtcptw) { 194162031Sglebius maxtcptw = new; 195162035Sglebius uma_zone_set_max(tcptw_zone, maxtcptw); 196162035Sglebius } else 197162031Sglebius error = EINVAL; 198162031Sglebius } 199162031Sglebius return (error); 200162031Sglebius} 201162031SglebiusSYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 202162031Sglebius &maxtcptw, 0, sysctl_maxtcptw, "IU", 203162031Sglebius "Maximum number of compressed TCP TIME_WAIT entries"); 204162031Sglebius 205102017Sdillon/* 206133874Srwatson * TCP bandwidth limiting sysctls. Note that the default lower bound of 207133874Srwatson * 1024 exists only for debugging. A good production default would be 208102017Sdillon * something like 6100. 209102017Sdillon */ 210133072SandreSYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 211133072Sandre "TCP inflight data limiting"); 212133072Sandre 213124199Sandrestatic int tcp_inflight_enable = 1; 214133072SandreSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 215102017Sdillon &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 216102017Sdillon 217104825Sdillonstatic int tcp_inflight_debug = 0; 218133072SandreSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 219102017Sdillon &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 220102017Sdillon 221155767Sandrestatic int tcp_inflight_rttthresh; 222155767SandreSYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 223155767Sandre &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 224155767Sandre "RTT threshold below which inflight will deactivate itself"); 225155767Sandre 226107881Sdillonstatic int tcp_inflight_min = 6144; 227133072SandreSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 228102017Sdillon &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 229102017Sdillon 230102017Sdillonstatic int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 231133072SandreSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 232102017Sdillon &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 233133072Sandre 234107881Sdillonstatic int tcp_inflight_stab = 20; 235133072SandreSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 236107881Sdillon &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 237102017Sdillon 238130989Spsuma_zone_t sack_hole_zone; 239130989Sps 24098211Shsustatic struct inpcb *tcp_notify(struct inpcb *, int); 241128452Ssilbystatic void tcp_isn_tick(void *); 24212296Sphk 2437684Sdg/* 24432821Sdg * Target size of TCP PCB hash tables. Must be a power of two. 24543562Smsmith * 24643562Smsmith * Note that this can be overridden by the kernel environment 24743562Smsmith * variable net.inet.tcp.tcbhashsize 2487684Sdg */ 2497684Sdg#ifndef TCBHASHSIZE 25032821Sdg#define TCBHASHSIZE 512 2517684Sdg#endif 2521541Srgrimes 2531541Srgrimes/* 254111145Sjlemon * XXX 255111145Sjlemon * Callouts should be moved into struct tcp directly. They are currently 256123608Sjhb * separate because the tcpcb structure is exported to userland for sysctl 257111145Sjlemon * parsing purposes, which do not know about callouts. 25834881Swollman */ 259111145Sjlemonstruct tcpcb_mem { 26034881Swollman struct tcpcb tcb; 261111145Sjlemon struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 262111145Sjlemon struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 26334881Swollman}; 26434881Swollman 265111145Sjlemonstatic uma_zone_t tcpcb_zone; 266128452Ssilbystruct callout isn_callout; 267157967Srwatsonstatic struct mtx isn_mtx; 268111145Sjlemon 269157977Srwatson#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 270157977Srwatson#define ISN_LOCK() mtx_lock(&isn_mtx) 271157977Srwatson#define ISN_UNLOCK() mtx_unlock(&isn_mtx) 272157977Srwatson 27334881Swollman/* 274157431Srwatson * TCP initialization. 2751541Srgrimes */ 276157927Spsstatic void 277157927Spstcp_zone_change(void *tag) 278157927Sps{ 279157927Sps 280157927Sps uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 281157927Sps uma_zone_set_max(tcpcb_zone, maxsockets); 282162031Sglebius if (maxtcptw == 0) 283162031Sglebius uma_zone_set_max(tcptw_zone, maxsockets / 5); 284157927Sps} 285157927Sps 286160491Supsstatic int 287160491Supstcp_inpcb_init(void *mem, int size, int flags) 288160491Sups{ 289160491Sups struct inpcb *inp = (struct inpcb *) mem; 290160491Sups INP_LOCK_INIT(inp, "inp", "tcpinp"); 291160491Sups return (0); 292160491Sups} 293160491Sups 2941541Srgrimesvoid 295157431Srwatsontcp_init(void) 2961541Srgrimes{ 29777843Speter int hashsize = TCBHASHSIZE; 298133874Srwatson 29950673Sjlemon tcp_delacktime = TCPTV_DELACK; 30050673Sjlemon tcp_keepinit = TCPTV_KEEP_INIT; 30150673Sjlemon tcp_keepidle = TCPTV_KEEP_IDLE; 30250673Sjlemon tcp_keepintvl = TCPTV_KEEPINTVL; 30350673Sjlemon tcp_maxpersistidle = TCPTV_KEEP_IDLE; 30450673Sjlemon tcp_msl = TCPTV_MSL; 305100335Sdillon tcp_rexmit_min = TCPTV_MIN; 306100335Sdillon tcp_rexmit_slop = TCPTV_CPU_VAR; 307155767Sandre tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 30850673Sjlemon 30998102Shsu INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 3107684Sdg LIST_INIT(&tcb); 3117684Sdg tcbinfo.listhead = &tcb; 31277900Speter TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 31343576Smsmith if (!powerof2(hashsize)) { 31443562Smsmith printf("WARNING: TCB hash size not a power of 2\n"); 31543562Smsmith hashsize = 512; /* safe default */ 31643562Smsmith } 31750426Sjlemon tcp_tcbhashsize = hashsize; 31843562Smsmith tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 31943562Smsmith tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 32034923Sbde &tcbinfo.porthashmask); 321133874Srwatson tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 322160491Sups NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 32392760Sjeff uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 32455679Sshin#ifdef INET6 32555679Sshin#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 32655679Sshin#else /* INET6 */ 32755679Sshin#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 32855679Sshin#endif /* INET6 */ 32955679Sshin if (max_protohdr < TCP_MINPROTOHDR) 33055679Sshin max_protohdr = TCP_MINPROTOHDR; 33155679Sshin if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 3321541Srgrimes panic("tcp_init"); 33355679Sshin#undef TCP_MINPROTOHDR 334111145Sjlemon /* 335111145Sjlemon * These have to be type stable for the benefit of the timers. 336111145Sjlemon */ 337133874Srwatson tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 338133517Sandre NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 339111145Sjlemon uma_zone_set_max(tcpcb_zone, maxsockets); 340162031Sglebius tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 341162031Sglebius NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 342157478Sglebius TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 343157478Sglebius if (maxtcptw == 0) 344162031Sglebius uma_zone_set_max(tcptw_zone, maxsockets / 5); 345162031Sglebius else 346162031Sglebius uma_zone_set_max(tcptw_zone, maxtcptw); 347112009Sjlemon tcp_timer_init(); 34886764Sjlemon syncache_init(); 349122922Sandre tcp_hc_init(); 350126193Sandre tcp_reass_init(); 351157977Srwatson ISN_LOCK_INIT(); 352128452Ssilby callout_init(&isn_callout, CALLOUT_MPSAFE); 353128452Ssilby tcp_isn_tick(NULL); 354128452Ssilby EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 355128452Ssilby SHUTDOWN_PRI_DEFAULT); 356133874Srwatson sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 357133517Sandre NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 358157927Sps EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 359157927Sps EVENTHANDLER_PRI_ANY); 3601541Srgrimes} 3611541Srgrimes 362128452Ssilbyvoid 363157431Srwatsontcp_fini(void *xtp) 364128452Ssilby{ 365157431Srwatson 366128452Ssilby callout_stop(&isn_callout); 367128452Ssilby} 368128452Ssilby 3691541Srgrimes/* 37078642Ssilby * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 37178642Ssilby * tcp_template used to store this data in mbufs, but we now recopy it out 37278642Ssilby * of the tcpcb each time to conserve mbufs. 3731541Srgrimes */ 37478642Ssilbyvoid 375157431Srwatsontcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 3761541Srgrimes{ 377111144Sjlemon struct tcphdr *th = (struct tcphdr *)tcp_ptr; 3781541Srgrimes 379138410Srwatson INP_LOCK_ASSERT(inp); 380138410Srwatson 38155679Sshin#ifdef INET6 38255679Sshin if ((inp->inp_vflag & INP_IPV6) != 0) { 38378642Ssilby struct ip6_hdr *ip6; 38455679Sshin 38578642Ssilby ip6 = (struct ip6_hdr *)ip_ptr; 38655679Sshin ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 38755679Sshin (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 38855679Sshin ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 38955679Sshin (IPV6_VERSION & IPV6_VERSION_MASK); 39055679Sshin ip6->ip6_nxt = IPPROTO_TCP; 39155679Sshin ip6->ip6_plen = sizeof(struct tcphdr); 39255679Sshin ip6->ip6_src = inp->in6p_laddr; 39355679Sshin ip6->ip6_dst = inp->in6p_faddr; 39455679Sshin } else 39555679Sshin#endif 39678642Ssilby { 397111144Sjlemon struct ip *ip; 39855679Sshin 399111144Sjlemon ip = (struct ip *)ip_ptr; 400111144Sjlemon ip->ip_v = IPVERSION; 401111144Sjlemon ip->ip_hl = 5; 402111144Sjlemon ip->ip_tos = inp->inp_ip_tos; 403111144Sjlemon ip->ip_len = 0; 404111144Sjlemon ip->ip_id = 0; 405111144Sjlemon ip->ip_off = 0; 406111144Sjlemon ip->ip_ttl = inp->inp_ip_ttl; 407111144Sjlemon ip->ip_sum = 0; 408111144Sjlemon ip->ip_p = IPPROTO_TCP; 409111144Sjlemon ip->ip_src = inp->inp_laddr; 410111144Sjlemon ip->ip_dst = inp->inp_faddr; 41178642Ssilby } 412111144Sjlemon th->th_sport = inp->inp_lport; 413111144Sjlemon th->th_dport = inp->inp_fport; 414111144Sjlemon th->th_seq = 0; 415111144Sjlemon th->th_ack = 0; 416111144Sjlemon th->th_x2 = 0; 417111144Sjlemon th->th_off = 5; 418111144Sjlemon th->th_flags = 0; 419111144Sjlemon th->th_win = 0; 420111144Sjlemon th->th_urp = 0; 421111144Sjlemon th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 42278642Ssilby} 42378642Ssilby 42478642Ssilby/* 42578642Ssilby * Create template to be used to send tcp packets on a connection. 42678642Ssilby * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 42778642Ssilby * use for this function is in keepalives, which use tcp_respond. 42878642Ssilby */ 42978642Ssilbystruct tcptemp * 430157431Srwatsontcpip_maketemplate(struct inpcb *inp) 43178642Ssilby{ 43278642Ssilby struct mbuf *m; 43378642Ssilby struct tcptemp *n; 43478642Ssilby 435149635Sandre m = m_get(M_DONTWAIT, MT_DATA); 43678642Ssilby if (m == NULL) 43778642Ssilby return (0); 43878642Ssilby m->m_len = sizeof(struct tcptemp); 43978642Ssilby n = mtod(m, struct tcptemp *); 44078642Ssilby 441111144Sjlemon tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 4421541Srgrimes return (n); 4431541Srgrimes} 4441541Srgrimes 4451541Srgrimes/* 4461541Srgrimes * Send a single message to the TCP at address specified by 447127870Srwatson * the given TCP/IP header. If m == NULL, then we make a copy 4481541Srgrimes * of the tcpiphdr at ti and send directly to the addressed host. 4491541Srgrimes * This is used to force keep alive messages out using the TCP 45078642Ssilby * template for a connection. If flags are given then we send 45178642Ssilby * a message back to the TCP which originated the * segment ti, 45278642Ssilby * and discard the mbuf containing it and any other attached mbufs. 4531541Srgrimes * 4541541Srgrimes * In any case the ack and sequence number of the transmitted 4551541Srgrimes * segment are as specified by the parameters. 45631848Sjulian * 45731848Sjulian * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 4581541Srgrimes */ 4591541Srgrimesvoid 460157431Srwatsontcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th, 461157431Srwatson register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 4621541Srgrimes{ 4631541Srgrimes register int tlen; 4641541Srgrimes int win = 0; 46555679Sshin struct ip *ip; 46655679Sshin struct tcphdr *nth; 46755679Sshin#ifdef INET6 46855679Sshin struct ip6_hdr *ip6; 46955679Sshin int isipv6; 47055679Sshin#endif /* INET6 */ 47155679Sshin int ipflags = 0; 472128905Srwatson struct inpcb *inp; 4731541Srgrimes 474101137Srwatson KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 475101137Srwatson 47655679Sshin#ifdef INET6 477105586Sphk isipv6 = ((struct ip *)ipgen)->ip_v == 6; 47855679Sshin ip6 = ipgen; 47955679Sshin#endif /* INET6 */ 48055679Sshin ip = ipgen; 48155679Sshin 482127871Srwatson if (tp != NULL) { 483122327Ssam inp = tp->t_inpcb; 484122327Ssam KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 485122327Ssam INP_INFO_WLOCK_ASSERT(&tcbinfo); 486122327Ssam INP_LOCK_ASSERT(inp); 487128905Srwatson } else 488128905Srwatson inp = NULL; 489128905Srwatson 490128905Srwatson if (tp != NULL) { 49157576Sps if (!(flags & TH_RST)) { 492122327Ssam win = sbspace(&inp->inp_socket->so_rcv); 49357576Sps if (win > (long)TCP_MAXWIN << tp->rcv_scale) 49457576Sps win = (long)TCP_MAXWIN << tp->rcv_scale; 49557576Sps } 4961541Srgrimes } 497127870Srwatson if (m == NULL) { 498151967Sandre m = m_gethdr(M_DONTWAIT, MT_DATA); 4991541Srgrimes if (m == NULL) 5001541Srgrimes return; 5011541Srgrimes tlen = 0; 5021541Srgrimes m->m_data += max_linkhdr; 50355679Sshin#ifdef INET6 50455679Sshin if (isipv6) { 505133874Srwatson bcopy((caddr_t)ip6, mtod(m, caddr_t), 50655679Sshin sizeof(struct ip6_hdr)); 50755679Sshin ip6 = mtod(m, struct ip6_hdr *); 50855679Sshin nth = (struct tcphdr *)(ip6 + 1); 50955679Sshin } else 51055679Sshin#endif /* INET6 */ 51155679Sshin { 51255679Sshin bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 51355679Sshin ip = mtod(m, struct ip *); 51455679Sshin nth = (struct tcphdr *)(ip + 1); 51555679Sshin } 51655679Sshin bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 5171541Srgrimes flags = TH_ACK; 5181541Srgrimes } else { 5191541Srgrimes m_freem(m->m_next); 520127870Srwatson m->m_next = NULL; 52155679Sshin m->m_data = (caddr_t)ipgen; 52255679Sshin /* m_len is set later */ 5231541Srgrimes tlen = 0; 5241541Srgrimes#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 52555679Sshin#ifdef INET6 52655679Sshin if (isipv6) { 52755679Sshin xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 52855679Sshin nth = (struct tcphdr *)(ip6 + 1); 52955679Sshin } else 53055679Sshin#endif /* INET6 */ 53155679Sshin { 53255679Sshin xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 53355679Sshin nth = (struct tcphdr *)(ip + 1); 53455679Sshin } 53555679Sshin if (th != nth) { 53655679Sshin /* 53755679Sshin * this is usually a case when an extension header 53855679Sshin * exists between the IPv6 header and the 53955679Sshin * TCP header. 54055679Sshin */ 54155679Sshin nth->th_sport = th->th_sport; 54255679Sshin nth->th_dport = th->th_dport; 54355679Sshin } 54455679Sshin xchg(nth->th_dport, nth->th_sport, n_short); 5451541Srgrimes#undef xchg 5461541Srgrimes } 54755679Sshin#ifdef INET6 54855679Sshin if (isipv6) { 54990198Sume ip6->ip6_flow = 0; 55090198Sume ip6->ip6_vfc = IPV6_VERSION; 55190198Sume ip6->ip6_nxt = IPPROTO_TCP; 55255679Sshin ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 55355679Sshin tlen)); 55455679Sshin tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 55556039Sshin } else 55655679Sshin#endif 557133874Srwatson { 558133874Srwatson tlen += sizeof (struct tcpiphdr); 559133874Srwatson ip->ip_len = tlen; 560133874Srwatson ip->ip_ttl = ip_defttl; 561133874Srwatson if (path_mtu_discovery) 562133874Srwatson ip->ip_off |= IP_DF; 563133874Srwatson } 5641541Srgrimes m->m_len = tlen; 5651541Srgrimes m->m_pkthdr.len = tlen; 566127870Srwatson m->m_pkthdr.rcvif = NULL; 567101106Srwatson#ifdef MAC 568122327Ssam if (inp != NULL) { 569101106Srwatson /* 570101106Srwatson * Packet is associated with a socket, so allow the 571101106Srwatson * label of the response to reflect the socket label. 572101106Srwatson */ 573128905Srwatson INP_LOCK_ASSERT(inp); 574128905Srwatson mac_create_mbuf_from_inpcb(inp, m); 575101106Srwatson } else { 576101106Srwatson /* 577119245Srwatson * Packet is not associated with a socket, so possibly 578119245Srwatson * update the label in place. 579101106Srwatson */ 580119245Srwatson mac_reflect_mbuf_tcp(m); 581101106Srwatson } 582101106Srwatson#endif 58355679Sshin nth->th_seq = htonl(seq); 58455679Sshin nth->th_ack = htonl(ack); 58555679Sshin nth->th_x2 = 0; 58655679Sshin nth->th_off = sizeof (struct tcphdr) >> 2; 58755679Sshin nth->th_flags = flags; 588127870Srwatson if (tp != NULL) 58955679Sshin nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 5901541Srgrimes else 59155679Sshin nth->th_win = htons((u_short)win); 59255679Sshin nth->th_urp = 0; 59355679Sshin#ifdef INET6 59455679Sshin if (isipv6) { 59559392Sshin nth->th_sum = 0; 59655679Sshin nth->th_sum = in6_cksum(m, IPPROTO_TCP, 59755679Sshin sizeof(struct ip6_hdr), 59855679Sshin tlen - sizeof(struct ip6_hdr)); 599127870Srwatson ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 600127870Srwatson NULL, NULL); 60155679Sshin } else 60255679Sshin#endif /* INET6 */ 603133874Srwatson { 604133874Srwatson nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 605133874Srwatson htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 606133874Srwatson m->m_pkthdr.csum_flags = CSUM_TCP; 607133874Srwatson m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 608133874Srwatson } 6096283Swollman#ifdef TCPDEBUG 610122327Ssam if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 61155679Sshin tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 6126283Swollman#endif 61355679Sshin#ifdef INET6 614122922Sandre if (isipv6) 615122922Sandre (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 616122922Sandre else 61755679Sshin#endif /* INET6 */ 618122922Sandre (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 6191541Srgrimes} 6201541Srgrimes 6211541Srgrimes/* 6221541Srgrimes * Create a new TCP control block, making an 6231541Srgrimes * empty reassembly queue and hooking it to the argument 62434881Swollman * protocol control block. The `inp' parameter must have 62534881Swollman * come from the zone allocator set up in tcp_init(). 6261541Srgrimes */ 6271541Srgrimesstruct tcpcb * 628157431Srwatsontcp_newtcpcb(struct inpcb *inp) 6291541Srgrimes{ 630111145Sjlemon struct tcpcb_mem *tm; 631111145Sjlemon struct tcpcb *tp; 63255679Sshin#ifdef INET6 63355679Sshin int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 63455679Sshin#endif /* INET6 */ 6351541Srgrimes 636111145Sjlemon tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 637111145Sjlemon if (tm == NULL) 638111145Sjlemon return (NULL); 639111145Sjlemon tp = &tm->tcb; 640111145Sjlemon /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 64155679Sshin tp->t_maxseg = tp->t_maxopd = 64255679Sshin#ifdef INET6 64355679Sshin isipv6 ? tcp_v6mssdflt : 64455679Sshin#endif /* INET6 */ 64555679Sshin tcp_mssdflt; 6461541Srgrimes 64750673Sjlemon /* Set up our timeouts. */ 648142906Sglebius callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 649142906Sglebius callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 650142906Sglebius callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 651142906Sglebius callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 652142906Sglebius callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 65350673Sjlemon 6546283Swollman if (tcp_do_rfc1323) 6556283Swollman tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 656130989Sps tp->sack_enable = tcp_do_sack; 657147735Sps TAILQ_INIT(&tp->snd_holes); 65834881Swollman tp->t_inpcb = inp; /* XXX */ 6591541Srgrimes /* 6601541Srgrimes * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 66116367Swollman * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 6621541Srgrimes * reasonable initial retransmit time. 6631541Srgrimes */ 6641541Srgrimes tp->t_srtt = TCPTV_SRTTBASE; 66516367Swollman tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 666100335Sdillon tp->t_rttmin = tcp_rexmit_min; 66716367Swollman tp->t_rxtcur = TCPTV_RTOBASE; 6681541Srgrimes tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 669102017Sdillon tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 6701541Srgrimes tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 67150673Sjlemon tp->t_rcvtime = ticks; 672102017Sdillon tp->t_bw_rtttime = ticks; 673133874Srwatson /* 67456564Sshin * IPv4 TTL initialization is necessary for an IPv6 socket as well, 67556564Sshin * because the socket may be bound to an IPv6 wildcard address, 67656564Sshin * which may match an IPv4-mapped IPv6 address. 67756564Sshin */ 67824570Sdg inp->inp_ip_ttl = ip_defttl; 679157432Srwatson inp->inp_ppcb = tp; 68034881Swollman return (tp); /* XXX */ 6811541Srgrimes} 6821541Srgrimes 6831541Srgrimes/* 6841541Srgrimes * Drop a TCP connection, reporting 6851541Srgrimes * the specified error. If connection is synchronized, 6861541Srgrimes * then send a RST to peer. 6871541Srgrimes */ 6881541Srgrimesstruct tcpcb * 689157431Srwatsontcp_drop(struct tcpcb *tp, int errno) 6901541Srgrimes{ 6911541Srgrimes struct socket *so = tp->t_inpcb->inp_socket; 6921541Srgrimes 693146864Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); 694138019Srwatson INP_LOCK_ASSERT(tp->t_inpcb); 695146864Srwatson 6961541Srgrimes if (TCPS_HAVERCVDSYN(tp->t_state)) { 6971541Srgrimes tp->t_state = TCPS_CLOSED; 6981541Srgrimes (void) tcp_output(tp); 6991541Srgrimes tcpstat.tcps_drops++; 7001541Srgrimes } else 7011541Srgrimes tcpstat.tcps_conndrops++; 7021541Srgrimes if (errno == ETIMEDOUT && tp->t_softerror) 7031541Srgrimes errno = tp->t_softerror; 7041541Srgrimes so->so_error = errno; 7051541Srgrimes return (tcp_close(tp)); 7061541Srgrimes} 7071541Srgrimes 708157376Srwatsonvoid 709157431Srwatsontcp_discardcb(struct tcpcb *tp) 7101541Srgrimes{ 711111145Sjlemon struct tseg_qent *q; 7121541Srgrimes struct inpcb *inp = tp->t_inpcb; 7131541Srgrimes struct socket *so = inp->inp_socket; 71455679Sshin#ifdef INET6 71555679Sshin int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 71655679Sshin#endif /* INET6 */ 7171541Srgrimes 718138410Srwatson INP_LOCK_ASSERT(inp); 719138410Srwatson 7201541Srgrimes /* 72150673Sjlemon * Make sure that all of our timers are stopped before we 72250673Sjlemon * delete the PCB. 72350673Sjlemon */ 72450673Sjlemon callout_stop(tp->tt_rexmt); 72550673Sjlemon callout_stop(tp->tt_persist); 72650673Sjlemon callout_stop(tp->tt_keep); 72750673Sjlemon callout_stop(tp->tt_2msl); 72850673Sjlemon callout_stop(tp->tt_delack); 72950673Sjlemon 73050673Sjlemon /* 7319373Swollman * If we got enough samples through the srtt filter, 7329373Swollman * save the rtt and rttvar in the routing entry. 733122922Sandre * 'Enough' is arbitrarily defined as 4 rtt samples. 734122922Sandre * 4 samples is enough for the srtt filter to converge 735122922Sandre * to within enough % of the correct value; fewer samples 736122922Sandre * and we could save a bogus rtt. The danger is not high 737122922Sandre * as tcp quickly recovers from everything. 738122922Sandre * XXX: Works very well but needs some more statistics! 7391541Srgrimes */ 740122922Sandre if (tp->t_rttupdated >= 4) { 741122922Sandre struct hc_metrics_lite metrics; 742122922Sandre u_long ssthresh; 7431541Srgrimes 744122922Sandre bzero(&metrics, sizeof(metrics)); 7451541Srgrimes /* 746122922Sandre * Update the ssthresh always when the conditions below 747122922Sandre * are satisfied. This gives us better new start value 748122922Sandre * for the congestion avoidance for new connections. 749122922Sandre * ssthresh is only set if packet loss occured on a session. 750160925Srwatson * 751160925Srwatson * XXXRW: 'so' may be NULL here, and/or socket buffer may be 752160925Srwatson * being torn down. Ideally this code would not use 'so'. 7531541Srgrimes */ 754122922Sandre ssthresh = tp->snd_ssthresh; 755122922Sandre if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 7561541Srgrimes /* 7571541Srgrimes * convert the limit from user data bytes to 7581541Srgrimes * packets then to packet data bytes. 7591541Srgrimes */ 760122922Sandre ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 761122922Sandre if (ssthresh < 2) 762122922Sandre ssthresh = 2; 763122922Sandre ssthresh *= (u_long)(tp->t_maxseg + 76455679Sshin#ifdef INET6 76555679Sshin (isipv6 ? sizeof (struct ip6_hdr) + 76655679Sshin sizeof (struct tcphdr) : 76755679Sshin#endif 76855679Sshin sizeof (struct tcpiphdr) 76955679Sshin#ifdef INET6 77055679Sshin ) 77155679Sshin#endif 77255679Sshin ); 773122922Sandre } else 774122922Sandre ssthresh = 0; 775122922Sandre metrics.rmx_ssthresh = ssthresh; 776122922Sandre 777122922Sandre metrics.rmx_rtt = tp->t_srtt; 778122922Sandre metrics.rmx_rttvar = tp->t_rttvar; 779122922Sandre /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 780122922Sandre metrics.rmx_bandwidth = tp->snd_bandwidth; 781122922Sandre metrics.rmx_cwnd = tp->snd_cwnd; 782133874Srwatson metrics.rmx_sendpipe = 0; 783122922Sandre metrics.rmx_recvpipe = 0; 784122922Sandre 785122922Sandre tcp_hc_update(&inp->inp_inc, &metrics); 7861541Srgrimes } 787122922Sandre 7881541Srgrimes /* free the reassembly queue, if any */ 789111145Sjlemon while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 79055679Sshin LIST_REMOVE(q, tqe_q); 79155679Sshin m_freem(q->tqe_m); 792126193Sandre uma_zfree(tcp_reass_zone, q); 793126193Sandre tp->t_segqlen--; 794126193Sandre tcp_reass_qsize--; 7951541Srgrimes } 796130989Sps tcp_free_sackholes(tp); 79732821Sdg inp->inp_ppcb = NULL; 798108265Shsu tp->t_inpcb = NULL; 799111145Sjlemon uma_zfree(tcpcb_zone, tp); 800111145Sjlemon} 801111145Sjlemon 802111145Sjlemon/* 803157430Srwatson * Attempt to close a TCP control block, marking it as dropped, and freeing 804157430Srwatson * the socket if we hold the only reference. 805111145Sjlemon */ 806111145Sjlemonstruct tcpcb * 807157431Srwatsontcp_close(struct tcpcb *tp) 808111145Sjlemon{ 809111145Sjlemon struct inpcb *inp = tp->t_inpcb; 810157376Srwatson struct socket *so; 811111145Sjlemon 812146864Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); 813138410Srwatson INP_LOCK_ASSERT(inp); 814138410Srwatson 815158009Srwatson in_pcbdrop(inp); 816157376Srwatson tcpstat.tcps_closed++; 817157376Srwatson KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 818157376Srwatson so = inp->inp_socket; 819157376Srwatson soisdisconnected(so); 820157376Srwatson if (inp->inp_vflag & INP_SOCKREF) { 821157376Srwatson KASSERT(so->so_state & SS_PROTOREF, 822157376Srwatson ("tcp_close: !SS_PROTOREF")); 823157376Srwatson inp->inp_vflag &= ~INP_SOCKREF; 824160549Srwatson INP_UNLOCK(inp); 825157376Srwatson ACCEPT_LOCK(); 826157376Srwatson SOCK_LOCK(so); 827157376Srwatson so->so_state &= ~SS_PROTOREF; 828157376Srwatson sofree(so); 829157376Srwatson return (NULL); 830157376Srwatson } 831157376Srwatson return (tp); 8321541Srgrimes} 8331541Srgrimes 8341541Srgrimesvoid 835157431Srwatsontcp_drain(void) 8361541Srgrimes{ 837157431Srwatson 838157431Srwatson if (do_tcpdrain) { 83955198Smsmith struct inpcb *inpb; 84055198Smsmith struct tcpcb *tcpb; 84155679Sshin struct tseg_qent *te; 8421541Srgrimes 84355198Smsmith /* 84455198Smsmith * Walk the tcpbs, if existing, and flush the reassembly queue, 84555198Smsmith * if there is one... 84655198Smsmith * XXX: The "Net/3" implementation doesn't imply that the TCP 84755198Smsmith * reassembly queue should be flushed, but in a situation 848133874Srwatson * where we're really low on mbufs, this is potentially 849133874Srwatson * usefull. 85055198Smsmith */ 85198102Shsu INP_INFO_RLOCK(&tcbinfo); 85274362Sphk LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 853111145Sjlemon if (inpb->inp_vflag & INP_TIMEWAIT) 854111145Sjlemon continue; 85598102Shsu INP_LOCK(inpb); 856127870Srwatson if ((tcpb = intotcpcb(inpb)) != NULL) { 85774362Sphk while ((te = LIST_FIRST(&tcpb->t_segq)) 85874362Sphk != NULL) { 85955679Sshin LIST_REMOVE(te, tqe_q); 86055679Sshin m_freem(te->tqe_m); 861126193Sandre uma_zfree(tcp_reass_zone, te); 862126193Sandre tcpb->t_segqlen--; 863126193Sandre tcp_reass_qsize--; 86455198Smsmith } 865144857Sps tcp_clean_sackreport(tcpb); 86655198Smsmith } 86798102Shsu INP_UNLOCK(inpb); 86855198Smsmith } 86998102Shsu INP_INFO_RUNLOCK(&tcbinfo); 87055198Smsmith } 8711541Srgrimes} 8721541Srgrimes 8731541Srgrimes/* 8741541Srgrimes * Notify a tcp user of an asynchronous error; 8751541Srgrimes * store error as soft error, but wake up user 8761541Srgrimes * (for now, won't do anything until can select for soft error). 87772960Sjlemon * 87872960Sjlemon * Do not wake up user since there currently is no mechanism for 87972960Sjlemon * reporting soft errors (yet - a kqueue filter may be added). 8801541Srgrimes */ 88198211Shsustatic struct inpcb * 882157431Srwatsontcp_notify(struct inpcb *inp, int error) 8831541Srgrimes{ 884157432Srwatson struct tcpcb *tp; 8851541Srgrimes 886146864Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); 887138410Srwatson INP_LOCK_ASSERT(inp); 888157433Srwatson 889157433Srwatson if ((inp->inp_vflag & INP_TIMEWAIT) || 890157433Srwatson (inp->inp_vflag & INP_DROPPED)) 891157433Srwatson return (inp); 892157433Srwatson 893157432Srwatson tp = intotcpcb(inp); 894157433Srwatson KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 895138410Srwatson 8961541Srgrimes /* 8971541Srgrimes * Ignore some errors if we are hooked up. 8981541Srgrimes * If connection hasn't completed, has retransmitted several times, 8991541Srgrimes * and receives a second error, give up now. This is better 9001541Srgrimes * than waiting a long time to establish a connection that 9011541Srgrimes * can never complete. 9021541Srgrimes */ 9031541Srgrimes if (tp->t_state == TCPS_ESTABLISHED && 904110896Shsu (error == EHOSTUNREACH || error == ENETUNREACH || 905110896Shsu error == EHOSTDOWN)) { 906139222Srwatson return (inp); 9071541Srgrimes } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 90898211Shsu tp->t_softerror) { 909157376Srwatson tp = tcp_drop(tp, error); 910157376Srwatson if (tp != NULL) 911157376Srwatson return (inp); 912157376Srwatson else 913157376Srwatson return (NULL); 91498211Shsu } else { 9151541Srgrimes tp->t_softerror = error; 916139222Srwatson return (inp); 91798211Shsu } 91872960Sjlemon#if 0 919111748Sdes wakeup( &so->so_timeo); 9201541Srgrimes sorwakeup(so); 9211541Srgrimes sowwakeup(so); 92272960Sjlemon#endif 9231541Srgrimes} 9241541Srgrimes 92536079Swollmanstatic int 92662573Sphktcp_pcblist(SYSCTL_HANDLER_ARGS) 92736079Swollman{ 928148156Srwatson int error, i, n; 92936079Swollman struct inpcb *inp, **inp_list; 93036079Swollman inp_gen_t gencnt; 93136079Swollman struct xinpgen xig; 93236079Swollman 93336079Swollman /* 93436079Swollman * The process of preparing the TCB list is too time-consuming and 93536079Swollman * resource-intensive to repeat twice on every request. 93636079Swollman */ 937127870Srwatson if (req->oldptr == NULL) { 93836079Swollman n = tcbinfo.ipi_count; 93936079Swollman req->oldidx = 2 * (sizeof xig) 94036079Swollman + (n + n/8) * sizeof(struct xtcpcb); 941139222Srwatson return (0); 94236079Swollman } 94336079Swollman 944127870Srwatson if (req->newptr != NULL) 945139222Srwatson return (EPERM); 94636079Swollman 94736079Swollman /* 94836079Swollman * OK, now we're committed to doing something. 94936079Swollman */ 95098102Shsu INP_INFO_RLOCK(&tcbinfo); 95136079Swollman gencnt = tcbinfo.ipi_gencnt; 95236079Swollman n = tcbinfo.ipi_count; 95398102Shsu INP_INFO_RUNLOCK(&tcbinfo); 95436079Swollman 955126253Struckman error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 956100831Struckman + n * sizeof(struct xtcpcb)); 957126253Struckman if (error != 0) 958126253Struckman return (error); 959100831Struckman 96036079Swollman xig.xig_len = sizeof xig; 96136079Swollman xig.xig_count = n; 96236079Swollman xig.xig_gen = gencnt; 96336079Swollman xig.xig_sogen = so_gencnt; 96436079Swollman error = SYSCTL_OUT(req, &xig, sizeof xig); 96536079Swollman if (error) 966139222Srwatson return (error); 96736079Swollman 968111119Simp inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 969127870Srwatson if (inp_list == NULL) 970139222Srwatson return (ENOMEM); 971133874Srwatson 97298102Shsu INP_INFO_RLOCK(&tcbinfo); 973127870Srwatson for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 97471999Sphk inp = LIST_NEXT(inp, inp_list)) { 97598102Shsu INP_LOCK(inp); 976113345Srwatson if (inp->inp_gencnt <= gencnt) { 977113345Srwatson /* 978113345Srwatson * XXX: This use of cr_cansee(), introduced with 979113345Srwatson * TCP state changes, is not quite right, but for 980113345Srwatson * now, better than nothing. 981113345Srwatson */ 982157474Srwatson if (inp->inp_vflag & INP_TIMEWAIT) { 983157474Srwatson if (intotw(inp) != NULL) 984157474Srwatson error = cr_cansee(req->td->td_ucred, 985157474Srwatson intotw(inp)->tw_cred); 986157474Srwatson else 987157474Srwatson error = EINVAL; /* Skip this inp. */ 988157474Srwatson } else 989113345Srwatson error = cr_canseesocket(req->td->td_ucred, 990113345Srwatson inp->inp_socket); 991113345Srwatson if (error == 0) 992113345Srwatson inp_list[i++] = inp; 993113345Srwatson } 99498102Shsu INP_UNLOCK(inp); 99536079Swollman } 99698102Shsu INP_INFO_RUNLOCK(&tcbinfo); 99736079Swollman n = i; 99836079Swollman 99936079Swollman error = 0; 100036079Swollman for (i = 0; i < n; i++) { 100136079Swollman inp = inp_list[i]; 1002160491Sups INP_LOCK(inp); 100336079Swollman if (inp->inp_gencnt <= gencnt) { 100436079Swollman struct xtcpcb xt; 1005157432Srwatson void *inp_ppcb; 1006145978Scperciva 1007145978Scperciva bzero(&xt, sizeof(xt)); 100836079Swollman xt.xt_len = sizeof xt; 100936079Swollman /* XXX should avoid extra copy */ 101036079Swollman bcopy(inp, &xt.xt_inp, sizeof *inp); 101147960Stegge inp_ppcb = inp->inp_ppcb; 1012111145Sjlemon if (inp_ppcb == NULL) 1013111145Sjlemon bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1014111145Sjlemon else if (inp->inp_vflag & INP_TIMEWAIT) { 1015111145Sjlemon bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1016111145Sjlemon xt.xt_tp.t_state = TCPS_TIME_WAIT; 1017111145Sjlemon } else 101847960Stegge bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1019127870Srwatson if (inp->inp_socket != NULL) 102036079Swollman sotoxsocket(inp->inp_socket, &xt.xt_socket); 1021111145Sjlemon else { 1022111145Sjlemon bzero(&xt.xt_socket, sizeof xt.xt_socket); 1023111145Sjlemon xt.xt_socket.xso_protocol = IPPROTO_TCP; 1024111145Sjlemon } 1025110896Shsu xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1026160491Sups INP_UNLOCK(inp); 102736079Swollman error = SYSCTL_OUT(req, &xt, sizeof xt); 1028160491Sups } else 1029160491Sups INP_UNLOCK(inp); 1030160491Sups 103136079Swollman } 103236079Swollman if (!error) { 103336079Swollman /* 103436079Swollman * Give the user an updated idea of our state. 103536079Swollman * If the generation differs from what we told 103636079Swollman * her before, she knows that something happened 103736079Swollman * while we were processing this request, and it 103836079Swollman * might be necessary to retry. 103936079Swollman */ 104098102Shsu INP_INFO_RLOCK(&tcbinfo); 104136079Swollman xig.xig_gen = tcbinfo.ipi_gencnt; 104236079Swollman xig.xig_sogen = so_gencnt; 104336079Swollman xig.xig_count = tcbinfo.ipi_count; 104498102Shsu INP_INFO_RUNLOCK(&tcbinfo); 104536079Swollman error = SYSCTL_OUT(req, &xig, sizeof xig); 104636079Swollman } 104736079Swollman free(inp_list, M_TEMP); 1048139222Srwatson return (error); 104936079Swollman} 105036079Swollman 105136079SwollmanSYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 105236079Swollman tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 105336079Swollman 105448758Sgreenstatic int 105562573Sphktcp_getcred(SYSCTL_HANDLER_ARGS) 105648758Sgreen{ 105772650Sgreen struct xucred xuc; 105848758Sgreen struct sockaddr_in addrs[2]; 105948758Sgreen struct inpcb *inp; 1060148156Srwatson int error; 106148758Sgreen 1062132653Scperciva error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 106348758Sgreen if (error) 106448758Sgreen return (error); 106548758Sgreen error = SYSCTL_IN(req, addrs, sizeof(addrs)); 106648758Sgreen if (error) 106748758Sgreen return (error); 106898102Shsu INP_INFO_RLOCK(&tcbinfo); 106948758Sgreen inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 107054263Sshin addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 107198102Shsu if (inp == NULL) { 107248758Sgreen error = ENOENT; 107398102Shsu goto outunlocked; 107448758Sgreen } 107599837Struckman INP_LOCK(inp); 107699837Struckman if (inp->inp_socket == NULL) { 107799837Struckman error = ENOENT; 107899837Struckman goto out; 107999837Struckman } 108092976Srwatson error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 108178697Sdwmalone if (error) 108278697Sdwmalone goto out; 108391354Sdd cru2x(inp->inp_socket->so_cred, &xuc); 108448758Sgreenout: 108598102Shsu INP_UNLOCK(inp); 108698102Shsuoutunlocked: 108798102Shsu INP_INFO_RUNLOCK(&tcbinfo); 108899838Struckman if (error == 0) 108999838Struckman error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 109048758Sgreen return (error); 109148758Sgreen} 109248758Sgreen 109378697SdwmaloneSYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 109478697Sdwmalone CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 109578697Sdwmalone tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 109648758Sgreen 109755679Sshin#ifdef INET6 109855679Sshinstatic int 109962573Sphktcp6_getcred(SYSCTL_HANDLER_ARGS) 110055679Sshin{ 110172650Sgreen struct xucred xuc; 110255679Sshin struct sockaddr_in6 addrs[2]; 110355679Sshin struct inpcb *inp; 1104148156Srwatson int error, mapped = 0; 110555679Sshin 1106132653Scperciva error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 110755679Sshin if (error) 110855679Sshin return (error); 110955679Sshin error = SYSCTL_IN(req, addrs, sizeof(addrs)); 111055679Sshin if (error) 111155679Sshin return (error); 1112148385Sume if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1113148385Sume (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1114148385Sume return (error); 1115148385Sume } 111655679Sshin if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 111755679Sshin if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 111855679Sshin mapped = 1; 111955679Sshin else 112055679Sshin return (EINVAL); 112155679Sshin } 1122148385Sume 112398102Shsu INP_INFO_RLOCK(&tcbinfo); 112455679Sshin if (mapped == 1) 112555679Sshin inp = in_pcblookup_hash(&tcbinfo, 112655679Sshin *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 112755679Sshin addrs[1].sin6_port, 112855679Sshin *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 112955679Sshin addrs[0].sin6_port, 113055679Sshin 0, NULL); 113155679Sshin else 1132151254Sphilip inp = in6_pcblookup_hash(&tcbinfo, 1133151254Sphilip &addrs[1].sin6_addr, addrs[1].sin6_port, 1134151254Sphilip &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 113598102Shsu if (inp == NULL) { 113655679Sshin error = ENOENT; 113798102Shsu goto outunlocked; 113855679Sshin } 113999837Struckman INP_LOCK(inp); 114099837Struckman if (inp->inp_socket == NULL) { 114199837Struckman error = ENOENT; 114299837Struckman goto out; 114399837Struckman } 114492976Srwatson error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 114578697Sdwmalone if (error) 114678697Sdwmalone goto out; 114791354Sdd cru2x(inp->inp_socket->so_cred, &xuc); 114855679Sshinout: 114998102Shsu INP_UNLOCK(inp); 115098102Shsuoutunlocked: 115198102Shsu INP_INFO_RUNLOCK(&tcbinfo); 115299838Struckman if (error == 0) 115399838Struckman error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 115455679Sshin return (error); 115555679Sshin} 115655679Sshin 115778697SdwmaloneSYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 115878697Sdwmalone CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 115978697Sdwmalone tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 116055679Sshin#endif 116155679Sshin 116255679Sshin 11631541Srgrimesvoid 1164157431Srwatsontcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 11651541Srgrimes{ 116672959Sjlemon struct ip *ip = vip; 116772959Sjlemon struct tcphdr *th; 116873109Sjlemon struct in_addr faddr; 116973109Sjlemon struct inpcb *inp; 117073109Sjlemon struct tcpcb *tp; 117198211Shsu struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1172145360Sandre struct icmp *icp; 1173145360Sandre struct in_conninfo inc; 1174145360Sandre tcp_seq icmp_tcp_seq; 1175148156Srwatson int mtu; 11761541Srgrimes 117773109Sjlemon faddr = ((struct sockaddr_in *)sa)->sin_addr; 117873109Sjlemon if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 117973109Sjlemon return; 118073109Sjlemon 1181145355Sandre if (cmd == PRC_MSGSIZE) 1182145355Sandre notify = tcp_mtudisc; 118374937Sjesper else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 118499156Sjesper cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 118572959Sjlemon notify = tcp_drop_syn_sent; 1186122922Sandre /* 1187122922Sandre * Redirects don't need to be handled up here. 1188122922Sandre */ 1189122922Sandre else if (PRC_IS_REDIRECT(cmd)) 1190122922Sandre return; 1191122922Sandre /* 1192145355Sandre * Source quench is depreciated. 1193145355Sandre */ 1194145355Sandre else if (cmd == PRC_QUENCH) 1195145355Sandre return; 1196145355Sandre /* 1197122922Sandre * Hostdead is ugly because it goes linearly through all PCBs. 1198122922Sandre * XXX: We never get this from ICMP, otherwise it makes an 1199122922Sandre * excellent DoS attack on machines with many connections. 1200122922Sandre */ 1201122922Sandre else if (cmd == PRC_HOSTDEAD) 1202127870Srwatson ip = NULL; 1203119995Sru else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 12041541Srgrimes return; 1205127870Srwatson if (ip != NULL) { 1206145360Sandre icp = (struct icmp *)((caddr_t)ip 1207145360Sandre - offsetof(struct icmp, icmp_ip)); 1208133874Srwatson th = (struct tcphdr *)((caddr_t)ip 1209105586Sphk + (ip->ip_hl << 2)); 121098596Shsu INP_INFO_WLOCK(&tcbinfo); 121173109Sjlemon inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 121273109Sjlemon ip->ip_src, th->th_sport, 0, NULL); 121398102Shsu if (inp != NULL) { 121498102Shsu INP_LOCK(inp); 1215157433Srwatson if (!(inp->inp_vflag & INP_TIMEWAIT) && 1216157433Srwatson !(inp->inp_vflag & INP_DROPPED) && 1217157433Srwatson !(inp->inp_socket == NULL)) { 1218145360Sandre icmp_tcp_seq = htonl(th->th_seq); 121998102Shsu tp = intotcpcb(inp); 1220145360Sandre if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1221145360Sandre SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1222145360Sandre if (cmd == PRC_MSGSIZE) { 1223145360Sandre /* 1224145360Sandre * MTU discovery: 1225145360Sandre * If we got a needfrag set the MTU 1226145360Sandre * in the route to the suggested new 1227145360Sandre * value (if given) and then notify. 1228145360Sandre */ 1229145360Sandre bzero(&inc, sizeof(inc)); 1230145360Sandre inc.inc_flags = 0; /* IPv4 */ 1231145360Sandre inc.inc_faddr = faddr; 1232145360Sandre 1233145360Sandre mtu = ntohs(icp->icmp_nextmtu); 1234145869Sandre /* 1235145869Sandre * If no alternative MTU was 1236145869Sandre * proposed, try the next smaller 1237149929Sandre * one. ip->ip_len has already 1238149929Sandre * been swapped in icmp_input(). 1239145869Sandre */ 1240145360Sandre if (!mtu) 1241149929Sandre mtu = ip_next_mtu(ip->ip_len, 1242145869Sandre 1); 1243145869Sandre if (mtu < max(296, (tcp_minmss) 1244145869Sandre + sizeof(struct tcpiphdr))) 1245145869Sandre mtu = 0; 1246145869Sandre if (!mtu) 1247145869Sandre mtu = tcp_mssdflt 1248145869Sandre + sizeof(struct tcpiphdr); 1249145869Sandre /* 1250145869Sandre * Only cache the the MTU if it 1251145869Sandre * is smaller than the interface 1252145869Sandre * or route MTU. tcp_mtudisc() 1253145869Sandre * will do right thing by itself. 1254145869Sandre */ 1255145869Sandre if (mtu <= tcp_maxmtu(&inc)) 1256145360Sandre tcp_hc_updatemtu(&inc, mtu); 1257145360Sandre } 1258145360Sandre 125998211Shsu inp = (*notify)(inp, inetctlerrmap[cmd]); 1260145360Sandre } 126198102Shsu } 1262127870Srwatson if (inp != NULL) 126398211Shsu INP_UNLOCK(inp); 126486764Sjlemon } else { 126586764Sjlemon inc.inc_fport = th->th_dport; 126686764Sjlemon inc.inc_lport = th->th_sport; 126786764Sjlemon inc.inc_faddr = faddr; 126886764Sjlemon inc.inc_laddr = ip->ip_src; 126986764Sjlemon#ifdef INET6 127086764Sjlemon inc.inc_isipv6 = 0; 127186764Sjlemon#endif 127286764Sjlemon syncache_unreach(&inc, th); 127373109Sjlemon } 127498596Shsu INP_INFO_WUNLOCK(&tcbinfo); 12751541Srgrimes } else 127698102Shsu in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 12771541Srgrimes} 12781541Srgrimes 127955679Sshin#ifdef INET6 128055679Sshinvoid 1281157431Srwatsontcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 128255679Sshin{ 128355679Sshin struct tcphdr th; 128498211Shsu struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 128555679Sshin struct ip6_hdr *ip6; 128655679Sshin struct mbuf *m; 128778064Sume struct ip6ctlparam *ip6cp = NULL; 128878064Sume const struct sockaddr_in6 *sa6_src = NULL; 128955679Sshin int off; 129078064Sume struct tcp_portonly { 129178064Sume u_int16_t th_sport; 129278064Sume u_int16_t th_dport; 129378064Sume } *thp; 129455679Sshin 129555679Sshin if (sa->sa_family != AF_INET6 || 129655679Sshin sa->sa_len != sizeof(struct sockaddr_in6)) 129755679Sshin return; 129855679Sshin 1299145355Sandre if (cmd == PRC_MSGSIZE) 130055679Sshin notify = tcp_mtudisc; 130155679Sshin else if (!PRC_IS_REDIRECT(cmd) && 1302119995Sru ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 130355679Sshin return; 1304145355Sandre /* Source quench is depreciated. */ 1305145355Sandre else if (cmd == PRC_QUENCH) 1306145355Sandre return; 130755679Sshin 130855679Sshin /* if the parameter is from icmp6, decode it. */ 130955679Sshin if (d != NULL) { 131078064Sume ip6cp = (struct ip6ctlparam *)d; 131155679Sshin m = ip6cp->ip6c_m; 131255679Sshin ip6 = ip6cp->ip6c_ip6; 131355679Sshin off = ip6cp->ip6c_off; 131478064Sume sa6_src = ip6cp->ip6c_src; 131555679Sshin } else { 131655679Sshin m = NULL; 131755679Sshin ip6 = NULL; 131867456Sitojun off = 0; /* fool gcc */ 131978064Sume sa6_src = &sa6_any; 132055679Sshin } 132155679Sshin 1322127870Srwatson if (ip6 != NULL) { 132386764Sjlemon struct in_conninfo inc; 132455679Sshin /* 132555679Sshin * XXX: We assume that when IPV6 is non NULL, 132655679Sshin * M and OFF are valid. 132755679Sshin */ 132855679Sshin 132967456Sitojun /* check if we can safely examine src and dst ports */ 133078064Sume if (m->m_pkthdr.len < off + sizeof(*thp)) 133167456Sitojun return; 133267456Sitojun 133378064Sume bzero(&th, sizeof(th)); 133478064Sume m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 133578064Sume 1336133192Srwatson in6_pcbnotify(&tcbinfo, sa, th.th_dport, 133778064Sume (struct sockaddr *)ip6cp->ip6c_src, 1338125776Sume th.th_sport, cmd, NULL, notify); 133986764Sjlemon 134086764Sjlemon inc.inc_fport = th.th_dport; 134186764Sjlemon inc.inc_lport = th.th_sport; 134286764Sjlemon inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 134386764Sjlemon inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 134486764Sjlemon inc.inc_isipv6 = 1; 1345133591Sdwmalone INP_INFO_WLOCK(&tcbinfo); 134686764Sjlemon syncache_unreach(&inc, &th); 1347133591Sdwmalone INP_INFO_WUNLOCK(&tcbinfo); 134855679Sshin } else 1349133192Srwatson in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1350125776Sume 0, cmd, NULL, notify); 135155679Sshin} 135255679Sshin#endif /* INET6 */ 135355679Sshin 135480428Speter 135582122Ssilby/* 135682122Ssilby * Following is where TCP initial sequence number generation occurs. 135782122Ssilby * 135882122Ssilby * There are two places where we must use initial sequence numbers: 135982122Ssilby * 1. In SYN-ACK packets. 136082122Ssilby * 2. In SYN packets. 136182122Ssilby * 136294390Ssilby * All ISNs for SYN-ACK packets are generated by the syncache. See 136394390Ssilby * tcp_syncache.c for details. 136482122Ssilby * 136582122Ssilby * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 136682122Ssilby * depends on this property. In addition, these ISNs should be 136782122Ssilby * unguessable so as to prevent connection hijacking. To satisfy 136882122Ssilby * the requirements of this situation, the algorithm outlined in 1369133874Srwatson * RFC 1948 is used, with only small modifications. 137082122Ssilby * 137182122Ssilby * Implementation details: 137282122Ssilby * 137382122Ssilby * Time is based off the system timer, and is corrected so that it 137482122Ssilby * increases by one megabyte per second. This allows for proper 137582122Ssilby * recycling on high speed LANs while still leaving over an hour 137682122Ssilby * before rollover. 137782122Ssilby * 1378128452Ssilby * As reading the *exact* system time is too expensive to be done 1379128452Ssilby * whenever setting up a TCP connection, we increment the time 1380128452Ssilby * offset in two ways. First, a small random positive increment 1381128452Ssilby * is added to isn_offset for each connection that is set up. 1382128452Ssilby * Second, the function tcp_isn_tick fires once per clock tick 1383128452Ssilby * and increments isn_offset as necessary so that sequence numbers 1384128452Ssilby * are incremented at approximately ISN_BYTES_PER_SECOND. The 1385128452Ssilby * random positive increments serve only to ensure that the same 1386128452Ssilby * exact sequence number is never sent out twice (as could otherwise 1387128452Ssilby * happen when a port is recycled in less than the system tick 1388128452Ssilby * interval.) 1389128452Ssilby * 139082122Ssilby * net.inet.tcp.isn_reseed_interval controls the number of seconds 139182122Ssilby * between seeding of isn_secret. This is normally set to zero, 139282122Ssilby * as reseeding should not be necessary. 139382122Ssilby * 1394138018Srwatson * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1395138018Srwatson * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1396138018Srwatson * general, this means holding an exclusive (write) lock. 139782122Ssilby */ 139879413Ssilby 139982122Ssilby#define ISN_BYTES_PER_SECOND 1048576 1400128452Ssilby#define ISN_STATIC_INCREMENT 4096 1401128452Ssilby#define ISN_RANDOM_INCREMENT (4096 - 1) 140279413Ssilby 1403138018Srwatsonstatic u_char isn_secret[32]; 1404138018Srwatsonstatic int isn_last_reseed; 1405138018Srwatsonstatic u_int32_t isn_offset, isn_offset_old; 1406138018Srwatsonstatic MD5_CTX isn_ctx; 140775619Skris 140875619Skristcp_seq 1409157431Srwatsontcp_new_isn(struct tcpcb *tp) 141075619Skris{ 141182122Ssilby u_int32_t md5_buffer[4]; 141282122Ssilby tcp_seq new_isn; 141375619Skris 1414138410Srwatson INP_LOCK_ASSERT(tp->t_inpcb); 1415138018Srwatson 1416157977Srwatson ISN_LOCK(); 141782122Ssilby /* Seed if this is the first use, reseed if requested. */ 141894390Ssilby if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 141982122Ssilby (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 142082122Ssilby < (u_int)ticks))) { 142182122Ssilby read_random(&isn_secret, sizeof(isn_secret)); 142282122Ssilby isn_last_reseed = ticks; 142382122Ssilby } 1424133874Srwatson 142582122Ssilby /* Compute the md5 hash and return the ISN. */ 142682122Ssilby MD5Init(&isn_ctx); 142782122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 142882122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 142982122Ssilby#ifdef INET6 143082122Ssilby if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 143182122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 143282122Ssilby sizeof(struct in6_addr)); 143382122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 143482122Ssilby sizeof(struct in6_addr)); 143582122Ssilby } else 143682122Ssilby#endif 143782122Ssilby { 143882122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 143982122Ssilby sizeof(struct in_addr)); 144082122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 144182122Ssilby sizeof(struct in_addr)); 144282122Ssilby } 144382122Ssilby MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 144482122Ssilby MD5Final((u_char *) &md5_buffer, &isn_ctx); 144582122Ssilby new_isn = (tcp_seq) md5_buffer[0]; 1446128452Ssilby isn_offset += ISN_STATIC_INCREMENT + 1447128452Ssilby (arc4random() & ISN_RANDOM_INCREMENT); 1448128452Ssilby new_isn += isn_offset; 1449157977Srwatson ISN_UNLOCK(); 1450139222Srwatson return (new_isn); 145175619Skris} 145275619Skris 14531541Srgrimes/* 1454128452Ssilby * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1455128452Ssilby * to keep time flowing at a relatively constant rate. If the random 1456128452Ssilby * increments have already pushed us past the projected offset, do nothing. 1457128452Ssilby */ 1458128452Ssilbystatic void 1459157431Srwatsontcp_isn_tick(void *xtp) 1460128452Ssilby{ 1461128452Ssilby u_int32_t projected_offset; 1462133874Srwatson 1463157977Srwatson ISN_LOCK(); 1464141078Srwatson projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1465128452Ssilby 1466128452Ssilby if (projected_offset > isn_offset) 1467128452Ssilby isn_offset = projected_offset; 1468128452Ssilby 1469128452Ssilby isn_offset_old = isn_offset; 1470141072Srwatson callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1471157977Srwatson ISN_UNLOCK(); 1472128452Ssilby} 1473128452Ssilby 1474128452Ssilby/* 147572959Sjlemon * When a specific ICMP unreachable message is received and the 147672959Sjlemon * connection state is SYN-SENT, drop the connection. This behavior 147772959Sjlemon * is controlled by the icmp_may_rst sysctl. 147870103Sphk */ 147998211Shsustruct inpcb * 1480157431Srwatsontcp_drop_syn_sent(struct inpcb *inp, int errno) 148170103Sphk{ 1482157432Srwatson struct tcpcb *tp; 148370103Sphk 1484146864Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); 1485138019Srwatson INP_LOCK_ASSERT(inp); 1486157433Srwatson 1487157433Srwatson if ((inp->inp_vflag & INP_TIMEWAIT) || 1488157433Srwatson (inp->inp_vflag & INP_DROPPED)) 1489157433Srwatson return (inp); 1490157433Srwatson 1491157432Srwatson tp = intotcpcb(inp); 1492157433Srwatson if (tp->t_state != TCPS_SYN_SENT) 1493157433Srwatson return (inp); 1494146864Srwatson 1495157433Srwatson tp = tcp_drop(tp, errno); 1496157433Srwatson if (tp != NULL) 1497157433Srwatson return (inp); 1498157433Srwatson else 1499157433Srwatson return (NULL); 150072638Sphk} 150172638Sphk 150272638Sphk/* 150310881Swollman * When `need fragmentation' ICMP is received, update our idea of the MSS 150410881Swollman * based on the new value in the route. Also nudge TCP to send something, 150510881Swollman * since we know the packet we just sent was dropped. 150610930Swollman * This duplicates some code in the tcp_mss() function in tcp_input.c. 150710881Swollman */ 150898211Shsustruct inpcb * 1509157431Srwatsontcp_mtudisc(struct inpcb *inp, int errno) 151010881Swollman{ 1511157432Srwatson struct tcpcb *tp; 151210930Swollman struct socket *so = inp->inp_socket; 1513122922Sandre u_int maxmtu; 1514122922Sandre u_int romtu; 151510930Swollman int mss; 151655679Sshin#ifdef INET6 1517134041Sandre int isipv6; 151855679Sshin#endif /* INET6 */ 151910881Swollman 1520138019Srwatson INP_LOCK_ASSERT(inp); 1521157433Srwatson if ((inp->inp_vflag & INP_TIMEWAIT) || 1522157433Srwatson (inp->inp_vflag & INP_DROPPED)) 1523157433Srwatson return (inp); 1524157433Srwatson 1525157432Srwatson tp = intotcpcb(inp); 1526157433Srwatson KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1527157433Srwatson 1528134041Sandre#ifdef INET6 1529157433Srwatson isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1530134041Sandre#endif 1531157433Srwatson maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1532157433Srwatson romtu = 153355679Sshin#ifdef INET6 1534157433Srwatson isipv6 ? tcp_maxmtu6(&inp->inp_inc) : 1535122922Sandre#endif /* INET6 */ 1536157433Srwatson tcp_maxmtu(&inp->inp_inc); 1537157433Srwatson if (!maxmtu) 1538157433Srwatson maxmtu = romtu; 1539157433Srwatson else 1540157433Srwatson maxmtu = min(maxmtu, romtu); 1541157433Srwatson if (!maxmtu) { 1542157433Srwatson tp->t_maxopd = tp->t_maxseg = 154355679Sshin#ifdef INET6 1544157433Srwatson isipv6 ? tcp_v6mssdflt : 154555679Sshin#endif /* INET6 */ 1546157433Srwatson tcp_mssdflt; 1547157433Srwatson return (inp); 1548157433Srwatson } 1549157433Srwatson mss = maxmtu - 155055679Sshin#ifdef INET6 1551157433Srwatson (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 155255679Sshin#endif /* INET6 */ 1553157433Srwatson sizeof(struct tcpiphdr) 155455679Sshin#ifdef INET6 1555157433Srwatson ) 155655679Sshin#endif /* INET6 */ 1557157433Srwatson ; 155855679Sshin 1559157433Srwatson /* 1560157433Srwatson * XXX - The above conditional probably violates the TCP 1561157433Srwatson * spec. The problem is that, since we don't know the 1562157433Srwatson * other end's MSS, we are supposed to use a conservative 1563157433Srwatson * default. But, if we do that, then MTU discovery will 1564157433Srwatson * never actually take place, because the conservative 1565157433Srwatson * default is much less than the MTUs typically seen 1566157433Srwatson * on the Internet today. For the moment, we'll sweep 1567157433Srwatson * this under the carpet. 1568157433Srwatson * 1569157433Srwatson * The conservative default might not actually be a problem 1570157433Srwatson * if the only case this occurs is when sending an initial 1571157433Srwatson * SYN with options and data to a host we've never talked 1572157433Srwatson * to before. Then, they will reply with an MSS value which 1573157433Srwatson * will get recorded and the new parameters should get 1574157433Srwatson * recomputed. For Further Study. 1575157433Srwatson */ 1576157433Srwatson if (tp->t_maxopd <= mss) 1577157433Srwatson return (inp); 1578157433Srwatson tp->t_maxopd = mss; 157910930Swollman 1580157433Srwatson if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1581157433Srwatson (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1582157433Srwatson mss -= TCPOLEN_TSTAMP_APPA; 158310930Swollman#if (MCLBYTES & (MCLBYTES - 1)) == 0 1584157433Srwatson if (mss > MCLBYTES) 1585157433Srwatson mss &= ~(MCLBYTES-1); 158610930Swollman#else 1587157433Srwatson if (mss > MCLBYTES) 1588157433Srwatson mss = mss / MCLBYTES * MCLBYTES; 158910881Swollman#endif 1590157433Srwatson if (so->so_snd.sb_hiwat < mss) 1591157433Srwatson mss = so->so_snd.sb_hiwat; 159210930Swollman 1593157433Srwatson tp->t_maxseg = mss; 159410930Swollman 1595157433Srwatson tcpstat.tcps_mturesent++; 1596157433Srwatson tp->t_rtttime = 0; 1597157433Srwatson tp->snd_nxt = tp->snd_una; 1598161645Smohans tcp_free_sackholes(tp); 1599161645Smohans tp->snd_recover = tp->snd_max; 1600161645Smohans if (tp->sack_enable) 1601161645Smohans EXIT_FASTRECOVERY(tp); 1602157433Srwatson tcp_output(tp); 1603139222Srwatson return (inp); 160410881Swollman} 160510881Swollman 160610881Swollman/* 16076283Swollman * Look-up the routing entry to the peer of this inpcb. If no route 1608108265Shsu * is found and it cannot be allocated, then return NULL. This routine 16096283Swollman * is called by TCP routines that access the rmx structure and by tcp_mss 16106283Swollman * to get the interface MTU. 16116283Swollman */ 1612133874Srwatsonu_long 1613157431Srwatsontcp_maxmtu(struct in_conninfo *inc) 16146283Swollman{ 1615122922Sandre struct route sro; 1616122922Sandre struct sockaddr_in *dst; 1617122922Sandre struct ifnet *ifp; 1618122922Sandre u_long maxmtu = 0; 16196283Swollman 1620122922Sandre KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1621122922Sandre 1622122996Sandre bzero(&sro, sizeof(sro)); 1623122922Sandre if (inc->inc_faddr.s_addr != INADDR_ANY) { 1624122922Sandre dst = (struct sockaddr_in *)&sro.ro_dst; 1625122922Sandre dst->sin_family = AF_INET; 1626122922Sandre dst->sin_len = sizeof(*dst); 1627122922Sandre dst->sin_addr = inc->inc_faddr; 1628122922Sandre rtalloc_ign(&sro, RTF_CLONING); 16296283Swollman } 1630122922Sandre if (sro.ro_rt != NULL) { 1631122922Sandre ifp = sro.ro_rt->rt_ifp; 1632122922Sandre if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1633122922Sandre maxmtu = ifp->if_mtu; 1634122922Sandre else 1635122922Sandre maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1636122922Sandre RTFREE(sro.ro_rt); 1637122922Sandre } 1638122922Sandre return (maxmtu); 16396283Swollman} 16406283Swollman 164155679Sshin#ifdef INET6 1642122922Sandreu_long 1643157431Srwatsontcp_maxmtu6(struct in_conninfo *inc) 164455679Sshin{ 1645122922Sandre struct route_in6 sro6; 1646122922Sandre struct ifnet *ifp; 1647122922Sandre u_long maxmtu = 0; 164855679Sshin 1649122922Sandre KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1650122922Sandre 1651122996Sandre bzero(&sro6, sizeof(sro6)); 1652122922Sandre if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1653122922Sandre sro6.ro_dst.sin6_family = AF_INET6; 1654122922Sandre sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1655122922Sandre sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1656122922Sandre rtalloc_ign((struct route *)&sro6, RTF_CLONING); 165755679Sshin } 1658122922Sandre if (sro6.ro_rt != NULL) { 1659122922Sandre ifp = sro6.ro_rt->rt_ifp; 1660122922Sandre if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1661122922Sandre maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1662122922Sandre else 1663122922Sandre maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1664122922Sandre IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1665122922Sandre RTFREE(sro6.ro_rt); 1666122922Sandre } 1667122922Sandre 1668122922Sandre return (maxmtu); 166955679Sshin} 167055679Sshin#endif /* INET6 */ 167155679Sshin 167255679Sshin#ifdef IPSEC 167355679Sshin/* compute ESP/AH header size for TCP, including outer IP header. */ 167455679Sshinsize_t 1675157431Srwatsonipsec_hdrsiz_tcp(struct tcpcb *tp) 167655679Sshin{ 167755679Sshin struct inpcb *inp; 167855679Sshin struct mbuf *m; 167955679Sshin size_t hdrsiz; 168055679Sshin struct ip *ip; 168155679Sshin#ifdef INET6 168255679Sshin struct ip6_hdr *ip6; 1683111145Sjlemon#endif 168455679Sshin struct tcphdr *th; 168555679Sshin 168678642Ssilby if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1687139222Srwatson return (0); 1688111119Simp MGETHDR(m, M_DONTWAIT, MT_DATA); 168955679Sshin if (!m) 1690139222Srwatson return (0); 169155679Sshin 169255679Sshin#ifdef INET6 169355679Sshin if ((inp->inp_vflag & INP_IPV6) != 0) { 169455679Sshin ip6 = mtod(m, struct ip6_hdr *); 169555679Sshin th = (struct tcphdr *)(ip6 + 1); 169655679Sshin m->m_pkthdr.len = m->m_len = 169755679Sshin sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1698111144Sjlemon tcpip_fillheaders(inp, ip6, th); 169955679Sshin hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 170055679Sshin } else 170155679Sshin#endif /* INET6 */ 1702133874Srwatson { 1703133874Srwatson ip = mtod(m, struct ip *); 1704133874Srwatson th = (struct tcphdr *)(ip + 1); 1705133874Srwatson m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1706133874Srwatson tcpip_fillheaders(inp, ip, th); 1707133874Srwatson hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1708133874Srwatson } 170955679Sshin 171055679Sshin m_free(m); 1711139222Srwatson return (hdrsiz); 171255679Sshin} 171355679Sshin#endif /*IPSEC*/ 171455679Sshin 17156283Swollman/* 1716111145Sjlemon * Move a TCP connection into TIME_WAIT state. 1717138410Srwatson * tcbinfo is locked. 1718111145Sjlemon * inp is locked, and is unlocked before returning. 1719111145Sjlemon */ 1720111145Sjlemonvoid 1721157431Srwatsontcp_twstart(struct tcpcb *tp) 1722111145Sjlemon{ 1723111145Sjlemon struct tcptw *tw; 1724111145Sjlemon struct inpcb *inp; 1725111145Sjlemon int tw_time, acknow; 1726111145Sjlemon struct socket *so; 1727111145Sjlemon 1728138025Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1729138020Srwatson INP_LOCK_ASSERT(tp->t_inpcb); 1730138020Srwatson 1731112009Sjlemon tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1732112009Sjlemon if (tw == NULL) { 1733112009Sjlemon tw = tcp_timer_2msl_tw(1); 1734112009Sjlemon if (tw == NULL) { 1735157376Srwatson tp = tcp_close(tp); 1736157376Srwatson if (tp != NULL) 1737157376Srwatson INP_UNLOCK(tp->t_inpcb); 1738112009Sjlemon return; 1739112009Sjlemon } 1740112009Sjlemon } 1741111145Sjlemon inp = tp->t_inpcb; 1742111145Sjlemon tw->tw_inpcb = inp; 1743111145Sjlemon 1744111145Sjlemon /* 1745111145Sjlemon * Recover last window size sent. 1746111145Sjlemon */ 1747111145Sjlemon tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1748111145Sjlemon 1749111145Sjlemon /* 1750111145Sjlemon * Set t_recent if timestamps are used on the connection. 1751111145Sjlemon */ 1752133874Srwatson if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1753133874Srwatson (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1754111145Sjlemon tw->t_recent = tp->ts_recent; 1755111145Sjlemon else 1756111145Sjlemon tw->t_recent = 0; 1757111145Sjlemon 1758111145Sjlemon tw->snd_nxt = tp->snd_nxt; 1759111145Sjlemon tw->rcv_nxt = tp->rcv_nxt; 1760121850Ssilby tw->iss = tp->iss; 1761121884Ssilby tw->irs = tp->irs; 1762111145Sjlemon tw->t_starttime = tp->t_starttime; 1763112009Sjlemon tw->tw_time = 0; 1764111145Sjlemon 1765111145Sjlemon/* XXX 1766111145Sjlemon * If this code will 1767111145Sjlemon * be used for fin-wait-2 state also, then we may need 1768111145Sjlemon * a ts_recent from the last segment. 1769111145Sjlemon */ 1770137139Sandre tw_time = 2 * tcp_msl; 1771137139Sandre acknow = tp->t_flags & TF_ACKNOW; 1772157376Srwatson 1773157376Srwatson /* 1774157376Srwatson * First, discard tcpcb state, which includes stopping its timers and 1775157376Srwatson * freeing it. tcp_discardcb() used to also release the inpcb, but 1776157376Srwatson * that work is now done in the caller. 1777160925Srwatson * 1778160925Srwatson * Note: soisdisconnected() call used to be made in tcp_discardcb(), 1779160925Srwatson * and might not be needed here any longer. 1780157376Srwatson */ 1781111145Sjlemon tcp_discardcb(tp); 1782111145Sjlemon so = inp->inp_socket; 1783160925Srwatson soisdisconnected(so); 1784130387Srwatson SOCK_LOCK(so); 1785111145Sjlemon tw->tw_cred = crhold(so->so_cred); 1786111145Sjlemon tw->tw_so_options = so->so_options; 1787157376Srwatson SOCK_UNLOCK(so); 1788114794Srwatson if (acknow) 1789126351Srwatson tcp_twrespond(tw, TH_ACK); 1790157432Srwatson inp->inp_ppcb = tw; 1791111145Sjlemon inp->inp_vflag |= INP_TIMEWAIT; 1792161226Smohans tcp_timer_2msl_reset(tw, tw_time, 0); 1793157376Srwatson 1794157376Srwatson /* 1795157376Srwatson * If the inpcb owns the sole reference to the socket, then we can 1796157376Srwatson * detach and free the socket as it is not needed in time wait. 1797157376Srwatson */ 1798157376Srwatson if (inp->inp_vflag & INP_SOCKREF) { 1799157376Srwatson KASSERT(so->so_state & SS_PROTOREF, 1800157376Srwatson ("tcp_twstart: !SS_PROTOREF")); 1801157376Srwatson inp->inp_vflag &= ~INP_SOCKREF; 1802157376Srwatson INP_UNLOCK(inp); 1803157376Srwatson ACCEPT_LOCK(); 1804157376Srwatson SOCK_LOCK(so); 1805157376Srwatson so->so_state &= ~SS_PROTOREF; 1806157376Srwatson sofree(so); 1807157376Srwatson } else 1808157376Srwatson INP_UNLOCK(inp); 1809111145Sjlemon} 1810111145Sjlemon 1811162064Sglebius#if 0 1812121850Ssilby/* 1813121884Ssilby * The appromixate rate of ISN increase of Microsoft TCP stacks; 1814121884Ssilby * the actual rate is slightly higher due to the addition of 1815121884Ssilby * random positive increments. 1816121884Ssilby * 1817121884Ssilby * Most other new OSes use semi-randomized ISN values, so we 1818121884Ssilby * do not need to worry about them. 1819121884Ssilby */ 1820121884Ssilby#define MS_ISN_BYTES_PER_SECOND 250000 1821121884Ssilby 1822121884Ssilby/* 1823121850Ssilby * Determine if the ISN we will generate has advanced beyond the last 1824121850Ssilby * sequence number used by the previous connection. If so, indicate 1825121850Ssilby * that it is safe to recycle this tw socket by returning 1. 1826121850Ssilby */ 1827121850Ssilbyint 1828121850Ssilbytcp_twrecycleable(struct tcptw *tw) 1829121850Ssilby{ 1830121884Ssilby tcp_seq new_iss = tw->iss; 1831121884Ssilby tcp_seq new_irs = tw->irs; 1832121850Ssilby 1833162064Sglebius INP_INFO_WLOCK_ASSERT(&tcbinfo); 1834121884Ssilby new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1835121884Ssilby new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1836133874Srwatson 1837121884Ssilby if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1838139222Srwatson return (1); 1839121850Ssilby else 1840139222Srwatson return (0); 1841121850Ssilby} 1842162064Sglebius#endif 1843121850Ssilby 1844157376Srwatsonvoid 1845112009Sjlemontcp_twclose(struct tcptw *tw, int reuse) 1846111145Sjlemon{ 1847157376Srwatson struct socket *so; 1848111145Sjlemon struct inpcb *inp; 1849111145Sjlemon 1850157376Srwatson /* 1851157386Srwatson * At this point, we are in one of two situations: 1852157376Srwatson * 1853160549Srwatson * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1854160549Srwatson * all state. 1855157386Srwatson * 1856160549Srwatson * (2) We have a socket -- if we own a reference, release it and 1857160549Srwatson * notify the socket layer. 1858157376Srwatson */ 1859111145Sjlemon inp = tw->tw_inpcb; 1860157376Srwatson KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1861157432Srwatson KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1862138025Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1863138020Srwatson INP_LOCK_ASSERT(inp); 1864138020Srwatson 1865111145Sjlemon tw->tw_inpcb = NULL; 1866112009Sjlemon tcp_timer_2msl_stop(tw); 1867111145Sjlemon inp->inp_ppcb = NULL; 1868158009Srwatson in_pcbdrop(inp); 1869157376Srwatson 1870157376Srwatson so = inp->inp_socket; 1871157386Srwatson if (so != NULL) { 1872160549Srwatson /* 1873160549Srwatson * If there's a socket, handle two cases: first, we own a 1874160549Srwatson * strong reference, which we will now release, or we don't 1875160549Srwatson * in which case another reference exists (XXXRW: think 1876160549Srwatson * about this more), and we don't need to take action. 1877160549Srwatson */ 1878157386Srwatson if (inp->inp_vflag & INP_SOCKREF) { 1879157386Srwatson inp->inp_vflag &= ~INP_SOCKREF; 1880160549Srwatson INP_UNLOCK(inp); 1881157386Srwatson ACCEPT_LOCK(); 1882157386Srwatson SOCK_LOCK(so); 1883157386Srwatson KASSERT(so->so_state & SS_PROTOREF, 1884157386Srwatson ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1885157386Srwatson so->so_state &= ~SS_PROTOREF; 1886157386Srwatson sofree(so); 1887157376Srwatson } else { 1888157386Srwatson /* 1889157386Srwatson * If we don't own the only reference, the socket and 1890157386Srwatson * inpcb need to be left around to be handled by 1891157386Srwatson * tcp_usr_detach() later. 1892157386Srwatson */ 1893157386Srwatson INP_UNLOCK(inp); 1894157376Srwatson } 1895157386Srwatson } else { 1896157376Srwatson#ifdef INET6 1897157376Srwatson if (inp->inp_vflag & INP_IPV6PROTO) 1898157376Srwatson in6_pcbfree(inp); 1899157376Srwatson else 1900157376Srwatson#endif 1901157376Srwatson in_pcbfree(inp); 1902157386Srwatson } 1903111145Sjlemon tcpstat.tcps_closed++; 1904126002Spjd crfree(tw->tw_cred); 1905126002Spjd tw->tw_cred = NULL; 1906112009Sjlemon if (reuse) 1907157376Srwatson return; 1908112009Sjlemon uma_zfree(tcptw_zone, tw); 1909111145Sjlemon} 1910111145Sjlemon 1911111145Sjlemonint 1912126351Srwatsontcp_twrespond(struct tcptw *tw, int flags) 1913111145Sjlemon{ 1914111145Sjlemon struct inpcb *inp = tw->tw_inpcb; 1915111145Sjlemon struct tcphdr *th; 1916111145Sjlemon struct mbuf *m; 1917111145Sjlemon struct ip *ip = NULL; 1918111145Sjlemon u_int8_t *optp; 1919111145Sjlemon u_int hdrlen, optlen; 1920111145Sjlemon int error; 1921111145Sjlemon#ifdef INET6 1922111145Sjlemon struct ip6_hdr *ip6 = NULL; 1923111145Sjlemon int isipv6 = inp->inp_inc.inc_isipv6; 1924111145Sjlemon#endif 1925111145Sjlemon 1926138020Srwatson INP_LOCK_ASSERT(inp); 1927138020Srwatson 1928151967Sandre m = m_gethdr(M_DONTWAIT, MT_DATA); 1929111145Sjlemon if (m == NULL) 1930111145Sjlemon return (ENOBUFS); 1931111145Sjlemon m->m_data += max_linkhdr; 1932111145Sjlemon 1933114794Srwatson#ifdef MAC 1934123607Srwatson mac_create_mbuf_from_inpcb(inp, m); 1935114794Srwatson#endif 1936114794Srwatson 1937111153Sjlemon#ifdef INET6 1938111145Sjlemon if (isipv6) { 1939111145Sjlemon hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1940111145Sjlemon ip6 = mtod(m, struct ip6_hdr *); 1941111145Sjlemon th = (struct tcphdr *)(ip6 + 1); 1942111145Sjlemon tcpip_fillheaders(inp, ip6, th); 1943111153Sjlemon } else 1944111153Sjlemon#endif 1945111153Sjlemon { 1946111145Sjlemon hdrlen = sizeof(struct tcpiphdr); 1947111145Sjlemon ip = mtod(m, struct ip *); 1948111145Sjlemon th = (struct tcphdr *)(ip + 1); 1949111145Sjlemon tcpip_fillheaders(inp, ip, th); 1950111145Sjlemon } 1951111145Sjlemon optp = (u_int8_t *)(th + 1); 1952133874Srwatson 1953133874Srwatson /* 1954111145Sjlemon * Send a timestamp and echo-reply if both our side and our peer 1955111145Sjlemon * have sent timestamps in our SYN's and this is not a RST. 1956133874Srwatson */ 1957111145Sjlemon if (tw->t_recent && flags == TH_ACK) { 1958111145Sjlemon u_int32_t *lp = (u_int32_t *)optp; 1959111145Sjlemon 1960133874Srwatson /* Form timestamp option as shown in appendix A of RFC 1323. */ 1961133874Srwatson *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1962133874Srwatson *lp++ = htonl(ticks); 1963133874Srwatson *lp = htonl(tw->t_recent); 1964133874Srwatson optp += TCPOLEN_TSTAMP_APPA; 1965133874Srwatson } 1966111145Sjlemon 1967111145Sjlemon optlen = optp - (u_int8_t *)(th + 1); 1968111145Sjlemon 1969111145Sjlemon m->m_len = hdrlen + optlen; 1970111145Sjlemon m->m_pkthdr.len = m->m_len; 1971111145Sjlemon 1972111145Sjlemon KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 1973111145Sjlemon 1974111145Sjlemon th->th_seq = htonl(tw->snd_nxt); 1975111145Sjlemon th->th_ack = htonl(tw->rcv_nxt); 1976111145Sjlemon th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1977111145Sjlemon th->th_flags = flags; 1978111145Sjlemon th->th_win = htons(tw->last_win); 1979111145Sjlemon 1980111153Sjlemon#ifdef INET6 1981111145Sjlemon if (isipv6) { 1982111145Sjlemon th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 1983111145Sjlemon sizeof(struct tcphdr) + optlen); 1984122922Sandre ip6->ip6_hlim = in6_selecthlim(inp, NULL); 1985122922Sandre error = ip6_output(m, inp->in6p_outputopts, NULL, 1986111145Sjlemon (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 1987111153Sjlemon } else 1988111153Sjlemon#endif 1989111153Sjlemon { 1990111145Sjlemon th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1991133874Srwatson htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 1992111145Sjlemon m->m_pkthdr.csum_flags = CSUM_TCP; 1993111145Sjlemon m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1994111145Sjlemon ip->ip_len = m->m_pkthdr.len; 1995124248Sandre if (path_mtu_discovery) 1996124248Sandre ip->ip_off |= IP_DF; 1997122922Sandre error = ip_output(m, inp->inp_options, NULL, 1998134793Sjmg ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 1999134793Sjmg NULL, inp); 2000111145Sjlemon } 2001111145Sjlemon if (flags & TH_ACK) 2002111145Sjlemon tcpstat.tcps_sndacks++; 2003111145Sjlemon else 2004111145Sjlemon tcpstat.tcps_sndctrl++; 2005111145Sjlemon tcpstat.tcps_sndtotal++; 2006111145Sjlemon return (error); 2007111145Sjlemon} 2008111145Sjlemon 2009111145Sjlemon/* 2010102017Sdillon * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 2011102017Sdillon * 2012102017Sdillon * This code attempts to calculate the bandwidth-delay product as a 2013102017Sdillon * means of determining the optimal window size to maximize bandwidth, 2014102017Sdillon * minimize RTT, and avoid the over-allocation of buffers on interfaces and 2015102017Sdillon * routers. This code also does a fairly good job keeping RTTs in check 2016102017Sdillon * across slow links like modems. We implement an algorithm which is very 2017102017Sdillon * similar (but not meant to be) TCP/Vegas. The code operates on the 2018102017Sdillon * transmitter side of a TCP connection and so only effects the transmit 2019102017Sdillon * side of the connection. 2020102017Sdillon * 2021102017Sdillon * BACKGROUND: TCP makes no provision for the management of buffer space 2022133874Srwatson * at the end points or at the intermediate routers and switches. A TCP 2023102017Sdillon * stream, whether using NewReno or not, will eventually buffer as 2024102017Sdillon * many packets as it is able and the only reason this typically works is 2025102017Sdillon * due to the fairly small default buffers made available for a connection 2026102017Sdillon * (typicaly 16K or 32K). As machines use larger windows and/or window 2027102017Sdillon * scaling it is now fairly easy for even a single TCP connection to blow-out 2028133874Srwatson * all available buffer space not only on the local interface, but on 2029102017Sdillon * intermediate routers and switches as well. NewReno makes a misguided 2030102017Sdillon * attempt to 'solve' this problem by waiting for an actual failure to occur, 2031102017Sdillon * then backing off, then steadily increasing the window again until another 2032102017Sdillon * failure occurs, ad-infinitum. This results in terrible oscillation that 2033102017Sdillon * is only made worse as network loads increase and the idea of intentionally 2034102017Sdillon * blowing out network buffers is, frankly, a terrible way to manage network 2035102017Sdillon * resources. 2036102017Sdillon * 2037102017Sdillon * It is far better to limit the transmit window prior to the failure 2038102017Sdillon * condition being achieved. There are two general ways to do this: First 2039102017Sdillon * you can 'scan' through different transmit window sizes and locate the 2040102017Sdillon * point where the RTT stops increasing, indicating that you have filled the 2041102017Sdillon * pipe, then scan backwards until you note that RTT stops decreasing, then 2042102017Sdillon * repeat ad-infinitum. This method works in principle but has severe 2043102017Sdillon * implementation issues due to RTT variances, timer granularity, and 2044102017Sdillon * instability in the algorithm which can lead to many false positives and 2045102017Sdillon * create oscillations as well as interact badly with other TCP streams 2046102017Sdillon * implementing the same algorithm. 2047102017Sdillon * 2048102017Sdillon * The second method is to limit the window to the bandwidth delay product 2049102017Sdillon * of the link. This is the method we implement. RTT variances and our 2050133874Srwatson * own manipulation of the congestion window, bwnd, can potentially 2051102017Sdillon * destabilize the algorithm. For this reason we have to stabilize the 2052102017Sdillon * elements used to calculate the window. We do this by using the minimum 2053102017Sdillon * observed RTT, the long term average of the observed bandwidth, and 2054102017Sdillon * by adding two segments worth of slop. It isn't perfect but it is able 2055102017Sdillon * to react to changing conditions and gives us a very stable basis on 2056102017Sdillon * which to extend the algorithm. 2057102017Sdillon */ 2058102017Sdillonvoid 2059102017Sdillontcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2060102017Sdillon{ 2061102017Sdillon u_long bw; 2062102017Sdillon u_long bwnd; 2063102017Sdillon int save_ticks; 2064102017Sdillon 2065138410Srwatson INP_LOCK_ASSERT(tp->t_inpcb); 2066138410Srwatson 2067102017Sdillon /* 2068102017Sdillon * If inflight_enable is disabled in the middle of a tcp connection, 2069102017Sdillon * make sure snd_bwnd is effectively disabled. 2070102017Sdillon */ 2071155767Sandre if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2072102017Sdillon tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2073102017Sdillon tp->snd_bandwidth = 0; 2074102017Sdillon return; 2075102017Sdillon } 2076102017Sdillon 2077102017Sdillon /* 2078102017Sdillon * Figure out the bandwidth. Due to the tick granularity this 2079102017Sdillon * is a very rough number and it MUST be averaged over a fairly 2080102017Sdillon * long period of time. XXX we need to take into account a link 2081102017Sdillon * that is not using all available bandwidth, but for now our 2082102017Sdillon * slop will ramp us up if this case occurs and the bandwidth later 2083102017Sdillon * increases. 2084102368Sdillon * 2085102368Sdillon * Note: if ticks rollover 'bw' may wind up negative. We must 2086102368Sdillon * effectively reset t_bw_rtttime for this case. 2087102017Sdillon */ 2088102017Sdillon save_ticks = ticks; 2089102017Sdillon if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2090102017Sdillon return; 2091102017Sdillon 2092133874Srwatson bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2093102017Sdillon (save_ticks - tp->t_bw_rtttime); 2094102017Sdillon tp->t_bw_rtttime = save_ticks; 2095102017Sdillon tp->t_bw_rtseq = ack_seq; 2096102368Sdillon if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2097102017Sdillon return; 2098102017Sdillon bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2099102017Sdillon 2100102017Sdillon tp->snd_bandwidth = bw; 2101102017Sdillon 2102102017Sdillon /* 2103102017Sdillon * Calculate the semi-static bandwidth delay product, plus two maximal 2104102017Sdillon * segments. The additional slop puts us squarely in the sweet 2105107881Sdillon * spot and also handles the bandwidth run-up case and stabilization. 2106107881Sdillon * Without the slop we could be locking ourselves into a lower 2107107881Sdillon * bandwidth. 2108102017Sdillon * 2109102017Sdillon * Situations Handled: 2110102017Sdillon * (1) Prevents over-queueing of packets on LANs, especially on 2111102017Sdillon * high speed LANs, allowing larger TCP buffers to be 2112133874Srwatson * specified, and also does a good job preventing 2113102017Sdillon * over-queueing of packets over choke points like modems 2114102017Sdillon * (at least for the transmit side). 2115102017Sdillon * 2116102017Sdillon * (2) Is able to handle changing network loads (bandwidth 2117102017Sdillon * drops so bwnd drops, bandwidth increases so bwnd 2118102017Sdillon * increases). 2119102017Sdillon * 2120102017Sdillon * (3) Theoretically should stabilize in the face of multiple 2121102017Sdillon * connections implementing the same algorithm (this may need 2122102017Sdillon * a little work). 2123107881Sdillon * 2124107881Sdillon * (4) Stability value (defaults to 20 = 2 maximal packets) can 2125107881Sdillon * be adjusted with a sysctl but typically only needs to be 2126107881Sdillon * on very slow connections. A value no smaller then 5 2127107881Sdillon * should be used, but only reduce this default if you have 2128107881Sdillon * no other choice. 2129102017Sdillon */ 2130102017Sdillon#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2131107881Sdillon bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2132102368Sdillon#undef USERTT 2133102017Sdillon 2134102017Sdillon if (tcp_inflight_debug > 0) { 2135102017Sdillon static int ltime; 2136102017Sdillon if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2137102017Sdillon ltime = ticks; 2138102017Sdillon printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2139102017Sdillon tp, 2140102017Sdillon bw, 2141102017Sdillon tp->t_rttbest, 2142102017Sdillon tp->t_srtt, 2143102017Sdillon bwnd 2144102017Sdillon ); 2145102017Sdillon } 2146102017Sdillon } 2147102017Sdillon if ((long)bwnd < tcp_inflight_min) 2148102017Sdillon bwnd = tcp_inflight_min; 2149102017Sdillon if (bwnd > tcp_inflight_max) 2150102017Sdillon bwnd = tcp_inflight_max; 2151102017Sdillon if ((long)bwnd < tp->t_maxseg * 2) 2152102017Sdillon bwnd = tp->t_maxseg * 2; 2153102017Sdillon tp->snd_bwnd = bwnd; 2154102017Sdillon} 2155102017Sdillon 2156125680Sbms#ifdef TCP_SIGNATURE 2157125680Sbms/* 2158125783Sbms * Callback function invoked by m_apply() to digest TCP segment data 2159125783Sbms * contained within an mbuf chain. 2160125783Sbms */ 2161125783Sbmsstatic int 2162125783Sbmstcp_signature_apply(void *fstate, void *data, u_int len) 2163125783Sbms{ 2164125783Sbms 2165125819Sbms MD5Update(fstate, (u_char *)data, len); 2166125783Sbms return (0); 2167125783Sbms} 2168125783Sbms 2169125783Sbms/* 2170125680Sbms * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2171125680Sbms * 2172125741Sbms * Parameters: 2173125741Sbms * m pointer to head of mbuf chain 2174125741Sbms * off0 offset to TCP header within the mbuf chain 2175125741Sbms * len length of TCP segment data, excluding options 2176125741Sbms * optlen length of TCP segment options 2177125741Sbms * buf pointer to storage for computed MD5 digest 2178125741Sbms * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2179125741Sbms * 2180125680Sbms * We do this over ip, tcphdr, segment data, and the key in the SADB. 2181125680Sbms * When called from tcp_input(), we can be sure that th_sum has been 2182125680Sbms * zeroed out and verified already. 2183125680Sbms * 2184125680Sbms * This function is for IPv4 use only. Calling this function with an 2185125680Sbms * IPv6 packet in the mbuf chain will yield undefined results. 2186125680Sbms * 2187125680Sbms * Return 0 if successful, otherwise return -1. 2188125680Sbms * 2189125680Sbms * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2190125680Sbms * search with the destination IP address, and a 'magic SPI' to be 2191125680Sbms * determined by the application. This is hardcoded elsewhere to 1179 2192125680Sbms * right now. Another branch of this code exists which uses the SPD to 2193125680Sbms * specify per-application flows but it is unstable. 2194125680Sbms */ 2195125680Sbmsint 2196125783Sbmstcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2197125741Sbms u_char *buf, u_int direction) 2198125680Sbms{ 2199125680Sbms union sockaddr_union dst; 2200125680Sbms struct ippseudo ippseudo; 2201125680Sbms MD5_CTX ctx; 2202125680Sbms int doff; 2203125680Sbms struct ip *ip; 2204125680Sbms struct ipovly *ipovly; 2205125680Sbms struct secasvar *sav; 2206125680Sbms struct tcphdr *th; 2207125680Sbms u_short savecsum; 2208125680Sbms 2209125741Sbms KASSERT(m != NULL, ("NULL mbuf chain")); 2210125741Sbms KASSERT(buf != NULL, ("NULL signature pointer")); 2211125741Sbms 2212125741Sbms /* Extract the destination from the IP header in the mbuf. */ 2213125680Sbms ip = mtod(m, struct ip *); 2214125680Sbms bzero(&dst, sizeof(union sockaddr_union)); 2215125680Sbms dst.sa.sa_len = sizeof(struct sockaddr_in); 2216125680Sbms dst.sa.sa_family = AF_INET; 2217125680Sbms dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2218125680Sbms ip->ip_src : ip->ip_dst; 2219125741Sbms 2220125741Sbms /* Look up an SADB entry which matches the address of the peer. */ 2221125680Sbms sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2222125680Sbms if (sav == NULL) { 2223125680Sbms printf("%s: SADB lookup failed for %s\n", __func__, 2224125680Sbms inet_ntoa(dst.sin.sin_addr)); 2225125680Sbms return (EINVAL); 2226125680Sbms } 2227125741Sbms 2228125680Sbms MD5Init(&ctx); 2229125680Sbms ipovly = (struct ipovly *)ip; 2230125680Sbms th = (struct tcphdr *)((u_char *)ip + off0); 2231125680Sbms doff = off0 + sizeof(struct tcphdr) + optlen; 2232125741Sbms 2233125680Sbms /* 2234125680Sbms * Step 1: Update MD5 hash with IP pseudo-header. 2235125680Sbms * 2236125680Sbms * XXX The ippseudo header MUST be digested in network byte order, 2237125680Sbms * or else we'll fail the regression test. Assume all fields we've 2238125680Sbms * been doing arithmetic on have been in host byte order. 2239125680Sbms * XXX One cannot depend on ipovly->ih_len here. When called from 2240125680Sbms * tcp_output(), the underlying ip_len member has not yet been set. 2241125680Sbms */ 2242125680Sbms ippseudo.ippseudo_src = ipovly->ih_src; 2243125680Sbms ippseudo.ippseudo_dst = ipovly->ih_dst; 2244125680Sbms ippseudo.ippseudo_pad = 0; 2245125680Sbms ippseudo.ippseudo_p = IPPROTO_TCP; 2246125680Sbms ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2247125680Sbms MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2248125741Sbms 2249125680Sbms /* 2250125680Sbms * Step 2: Update MD5 hash with TCP header, excluding options. 2251125680Sbms * The TCP checksum must be set to zero. 2252125680Sbms */ 2253125680Sbms savecsum = th->th_sum; 2254125680Sbms th->th_sum = 0; 2255125680Sbms MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2256125680Sbms th->th_sum = savecsum; 2257125741Sbms 2258125680Sbms /* 2259125680Sbms * Step 3: Update MD5 hash with TCP segment data. 2260125680Sbms * Use m_apply() to avoid an early m_pullup(). 2261125680Sbms */ 2262125680Sbms if (len > 0) 2263125783Sbms m_apply(m, doff, len, tcp_signature_apply, &ctx); 2264125741Sbms 2265125680Sbms /* 2266125680Sbms * Step 4: Update MD5 hash with shared secret. 2267125680Sbms */ 2268125680Sbms MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2269125680Sbms MD5Final(buf, &ctx); 2270125741Sbms 2271125680Sbms key_sa_recordxfer(sav, m); 2272125680Sbms KEY_FREESAV(&sav); 2273125680Sbms return (0); 2274125680Sbms} 2275125680Sbms#endif /* TCP_SIGNATURE */ 2276141886Smaxim 2277141886Smaximstatic int 2278141886Smaximsysctl_drop(SYSCTL_HANDLER_ARGS) 2279141886Smaxim{ 2280141886Smaxim /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2281141886Smaxim struct sockaddr_storage addrs[2]; 2282141886Smaxim struct inpcb *inp; 2283141886Smaxim struct tcpcb *tp; 2284150804Smaxim struct tcptw *tw; 2285141886Smaxim struct sockaddr_in *fin, *lin; 2286141886Smaxim#ifdef INET6 2287141886Smaxim struct sockaddr_in6 *fin6, *lin6; 2288141886Smaxim struct in6_addr f6, l6; 2289141886Smaxim#endif 2290141886Smaxim int error; 2291141886Smaxim 2292141886Smaxim inp = NULL; 2293141886Smaxim fin = lin = NULL; 2294141886Smaxim#ifdef INET6 2295141886Smaxim fin6 = lin6 = NULL; 2296141886Smaxim#endif 2297141886Smaxim error = 0; 2298141886Smaxim 2299141886Smaxim if (req->oldptr != NULL || req->oldlen != 0) 2300141886Smaxim return (EINVAL); 2301141886Smaxim if (req->newptr == NULL) 2302141886Smaxim return (EPERM); 2303141886Smaxim if (req->newlen < sizeof(addrs)) 2304141886Smaxim return (ENOMEM); 2305141886Smaxim error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2306141886Smaxim if (error) 2307141886Smaxim return (error); 2308141886Smaxim 2309141886Smaxim switch (addrs[0].ss_family) { 2310141886Smaxim#ifdef INET6 2311141886Smaxim case AF_INET6: 2312141886Smaxim fin6 = (struct sockaddr_in6 *)&addrs[0]; 2313141886Smaxim lin6 = (struct sockaddr_in6 *)&addrs[1]; 2314141886Smaxim if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2315141886Smaxim lin6->sin6_len != sizeof(struct sockaddr_in6)) 2316141886Smaxim return (EINVAL); 2317141886Smaxim if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2318141886Smaxim if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2319141886Smaxim return (EINVAL); 2320141886Smaxim in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2321141886Smaxim in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2322141886Smaxim fin = (struct sockaddr_in *)&addrs[0]; 2323141886Smaxim lin = (struct sockaddr_in *)&addrs[1]; 2324141886Smaxim break; 2325141886Smaxim } 2326148385Sume error = sa6_embedscope(fin6, ip6_use_defzone); 2327141886Smaxim if (error) 2328148385Sume return (error); 2329148385Sume error = sa6_embedscope(lin6, ip6_use_defzone); 2330148616Sume if (error) 2331148385Sume return (error); 2332141886Smaxim break; 2333141886Smaxim#endif 2334141886Smaxim case AF_INET: 2335141886Smaxim fin = (struct sockaddr_in *)&addrs[0]; 2336141886Smaxim lin = (struct sockaddr_in *)&addrs[1]; 2337141886Smaxim if (fin->sin_len != sizeof(struct sockaddr_in) || 2338141886Smaxim lin->sin_len != sizeof(struct sockaddr_in)) 2339141886Smaxim return (EINVAL); 2340141886Smaxim break; 2341141886Smaxim default: 2342141886Smaxim return (EINVAL); 2343141886Smaxim } 2344141886Smaxim INP_INFO_WLOCK(&tcbinfo); 2345141886Smaxim switch (addrs[0].ss_family) { 2346141886Smaxim#ifdef INET6 2347141886Smaxim case AF_INET6: 2348141886Smaxim inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2349141886Smaxim &l6, lin6->sin6_port, 0, NULL); 2350141886Smaxim break; 2351141886Smaxim#endif 2352141886Smaxim case AF_INET: 2353141886Smaxim inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2354141886Smaxim lin->sin_addr, lin->sin_port, 0, NULL); 2355141886Smaxim break; 2356141886Smaxim } 2357141886Smaxim if (inp != NULL) { 2358141886Smaxim INP_LOCK(inp); 2359157427Srwatson if (inp->inp_vflag & INP_TIMEWAIT) { 2360157474Srwatson /* 2361157474Srwatson * XXXRW: There currently exists a state where an 2362157474Srwatson * inpcb is present, but its timewait state has been 2363157474Srwatson * discarded. For now, don't allow dropping of this 2364157474Srwatson * type of inpcb. 2365157474Srwatson */ 2366157427Srwatson tw = intotw(inp); 2367157474Srwatson if (tw != NULL) 2368157474Srwatson tcp_twclose(tw, 0); 2369157427Srwatson } else if (!(inp->inp_vflag & INP_DROPPED) && 2370157427Srwatson !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2371157427Srwatson tp = intotcpcb(inp); 2372157427Srwatson tcp_drop(tp, ECONNABORTED); 2373157427Srwatson } 2374157427Srwatson INP_UNLOCK(inp); 2375141886Smaxim } else 2376141886Smaxim error = ESRCH; 2377141886Smaxim INP_INFO_WUNLOCK(&tcbinfo); 2378141886Smaxim return (error); 2379141886Smaxim} 2380141886Smaxim 2381141886SmaximSYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2382141886Smaxim CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2383141886Smaxim 0, sysctl_drop, "", "Drop TCP connection"); 2384