tcp_timewait.c revision 162768
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 162768 2006-09-29 06:46:31Z maxim $ 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; 182162767Ssilby 183162031Sglebiusstatic int 184162767Ssilbytcptw_auto_size(void) 185162767Ssilby{ 186162767Ssilby int halfrange; 187162767Ssilby 188162767Ssilby /* 189162768Smaxim * Max out at half the ephemeral port range so that TIME_WAIT 190162768Smaxim * sockets don't tie up too many ephemeral ports. 191162768Smaxim */ 192162767Ssilby if (ipport_lastauto > ipport_firstauto) 193162767Ssilby halfrange = (ipport_lastauto - ipport_firstauto) / 2; 194162767Ssilby else 195162767Ssilby halfrange = (ipport_firstauto - ipport_lastauto) / 2; 196162767Ssilby /* Protect against goofy port ranges smaller than 32. */ 197162767Ssilby return (imin(imax(halfrange, 32), maxsockets / 5)); 198162767Ssilby} 199162767Ssilby 200162767Ssilbystatic int 201162031Sglebiussysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 202162031Sglebius{ 203162031Sglebius int error, new; 204157478Sglebius 205162031Sglebius if (maxtcptw == 0) 206162767Ssilby new = tcptw_auto_size(); 207162031Sglebius else 208162031Sglebius new = maxtcptw; 209162031Sglebius error = sysctl_handle_int(oidp, &new, sizeof(int), req); 210162767Ssilby if (error == 0 && req->newptr) 211162767Ssilby if (new >= 32) { 212162031Sglebius maxtcptw = new; 213162035Sglebius uma_zone_set_max(tcptw_zone, maxtcptw); 214162767Ssilby } 215162031Sglebius return (error); 216162031Sglebius} 217162031SglebiusSYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 218162031Sglebius &maxtcptw, 0, sysctl_maxtcptw, "IU", 219162031Sglebius "Maximum number of compressed TCP TIME_WAIT entries"); 220162031Sglebius 221162151Sglebiusstatic int nolocaltimewait = 0; 222162151SglebiusSYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW, 223162151Sglebius &nolocaltimewait, 0, "Do not create compressed TCP TIME_WAIT entries" 224162151Sglebius "for local connections"); 225162151Sglebius 226102017Sdillon/* 227133874Srwatson * TCP bandwidth limiting sysctls. Note that the default lower bound of 228133874Srwatson * 1024 exists only for debugging. A good production default would be 229102017Sdillon * something like 6100. 230102017Sdillon */ 231133072SandreSYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0, 232133072Sandre "TCP inflight data limiting"); 233133072Sandre 234124199Sandrestatic int tcp_inflight_enable = 1; 235133072SandreSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW, 236102017Sdillon &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting"); 237102017Sdillon 238104825Sdillonstatic int tcp_inflight_debug = 0; 239133072SandreSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW, 240102017Sdillon &tcp_inflight_debug, 0, "Debug TCP inflight calculations"); 241102017Sdillon 242155767Sandrestatic int tcp_inflight_rttthresh; 243155767SandreSYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW, 244155767Sandre &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I", 245155767Sandre "RTT threshold below which inflight will deactivate itself"); 246155767Sandre 247107881Sdillonstatic int tcp_inflight_min = 6144; 248133072SandreSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW, 249102017Sdillon &tcp_inflight_min, 0, "Lower-bound for TCP inflight window"); 250102017Sdillon 251102017Sdillonstatic int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT; 252133072SandreSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW, 253102017Sdillon &tcp_inflight_max, 0, "Upper-bound for TCP inflight window"); 254133072Sandre 255107881Sdillonstatic int tcp_inflight_stab = 20; 256133072SandreSYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW, 257107881Sdillon &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets"); 258102017Sdillon 259130989Spsuma_zone_t sack_hole_zone; 260130989Sps 26198211Shsustatic struct inpcb *tcp_notify(struct inpcb *, int); 262128452Ssilbystatic void tcp_isn_tick(void *); 26312296Sphk 2647684Sdg/* 26532821Sdg * Target size of TCP PCB hash tables. Must be a power of two. 26643562Smsmith * 26743562Smsmith * Note that this can be overridden by the kernel environment 26843562Smsmith * variable net.inet.tcp.tcbhashsize 2697684Sdg */ 2707684Sdg#ifndef TCBHASHSIZE 27132821Sdg#define TCBHASHSIZE 512 2727684Sdg#endif 2731541Srgrimes 2741541Srgrimes/* 275111145Sjlemon * XXX 276111145Sjlemon * Callouts should be moved into struct tcp directly. They are currently 277123608Sjhb * separate because the tcpcb structure is exported to userland for sysctl 278111145Sjlemon * parsing purposes, which do not know about callouts. 27934881Swollman */ 280111145Sjlemonstruct tcpcb_mem { 28134881Swollman struct tcpcb tcb; 282111145Sjlemon struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep; 283111145Sjlemon struct callout tcpcb_mem_2msl, tcpcb_mem_delack; 28434881Swollman}; 28534881Swollman 286111145Sjlemonstatic uma_zone_t tcpcb_zone; 287128452Ssilbystruct callout isn_callout; 288157967Srwatsonstatic struct mtx isn_mtx; 289111145Sjlemon 290157977Srwatson#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 291157977Srwatson#define ISN_LOCK() mtx_lock(&isn_mtx) 292157977Srwatson#define ISN_UNLOCK() mtx_unlock(&isn_mtx) 293157977Srwatson 29434881Swollman/* 295157431Srwatson * TCP initialization. 2961541Srgrimes */ 297157927Spsstatic void 298157927Spstcp_zone_change(void *tag) 299157927Sps{ 300157927Sps 301157927Sps uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 302157927Sps uma_zone_set_max(tcpcb_zone, maxsockets); 303162031Sglebius if (maxtcptw == 0) 304162767Ssilby uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 305157927Sps} 306157927Sps 307160491Supsstatic int 308160491Supstcp_inpcb_init(void *mem, int size, int flags) 309160491Sups{ 310160491Sups struct inpcb *inp = (struct inpcb *) mem; 311160491Sups INP_LOCK_INIT(inp, "inp", "tcpinp"); 312160491Sups return (0); 313160491Sups} 314160491Sups 3151541Srgrimesvoid 316157431Srwatsontcp_init(void) 3171541Srgrimes{ 31877843Speter int hashsize = TCBHASHSIZE; 319133874Srwatson 32050673Sjlemon tcp_delacktime = TCPTV_DELACK; 32150673Sjlemon tcp_keepinit = TCPTV_KEEP_INIT; 32250673Sjlemon tcp_keepidle = TCPTV_KEEP_IDLE; 32350673Sjlemon tcp_keepintvl = TCPTV_KEEPINTVL; 32450673Sjlemon tcp_maxpersistidle = TCPTV_KEEP_IDLE; 32550673Sjlemon tcp_msl = TCPTV_MSL; 326100335Sdillon tcp_rexmit_min = TCPTV_MIN; 327100335Sdillon tcp_rexmit_slop = TCPTV_CPU_VAR; 328155767Sandre tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH; 32950673Sjlemon 33098102Shsu INP_INFO_LOCK_INIT(&tcbinfo, "tcp"); 3317684Sdg LIST_INIT(&tcb); 3327684Sdg tcbinfo.listhead = &tcb; 33377900Speter TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize); 33443576Smsmith if (!powerof2(hashsize)) { 33543562Smsmith printf("WARNING: TCB hash size not a power of 2\n"); 33643562Smsmith hashsize = 512; /* safe default */ 33743562Smsmith } 33850426Sjlemon tcp_tcbhashsize = hashsize; 33943562Smsmith tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask); 34043562Smsmith tcbinfo.porthashbase = hashinit(hashsize, M_PCB, 34134923Sbde &tcbinfo.porthashmask); 342133874Srwatson tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb), 343160491Sups NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 34492760Sjeff uma_zone_set_max(tcbinfo.ipi_zone, maxsockets); 34555679Sshin#ifdef INET6 34655679Sshin#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 34755679Sshin#else /* INET6 */ 34855679Sshin#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 34955679Sshin#endif /* INET6 */ 35055679Sshin if (max_protohdr < TCP_MINPROTOHDR) 35155679Sshin max_protohdr = TCP_MINPROTOHDR; 35255679Sshin if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 3531541Srgrimes panic("tcp_init"); 35455679Sshin#undef TCP_MINPROTOHDR 355111145Sjlemon /* 356111145Sjlemon * These have to be type stable for the benefit of the timers. 357111145Sjlemon */ 358133874Srwatson tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 359133517Sandre NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 360111145Sjlemon uma_zone_set_max(tcpcb_zone, maxsockets); 361162031Sglebius tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 362162031Sglebius NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 363157478Sglebius TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 364157478Sglebius if (maxtcptw == 0) 365162767Ssilby uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 366162031Sglebius else 367162031Sglebius uma_zone_set_max(tcptw_zone, maxtcptw); 368112009Sjlemon tcp_timer_init(); 36986764Sjlemon syncache_init(); 370122922Sandre tcp_hc_init(); 371126193Sandre tcp_reass_init(); 372157977Srwatson ISN_LOCK_INIT(); 373128452Ssilby callout_init(&isn_callout, CALLOUT_MPSAFE); 374128452Ssilby tcp_isn_tick(NULL); 375128452Ssilby EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 376128452Ssilby SHUTDOWN_PRI_DEFAULT); 377133874Srwatson sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 378133517Sandre NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 379157927Sps EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 380157927Sps EVENTHANDLER_PRI_ANY); 3811541Srgrimes} 3821541Srgrimes 383128452Ssilbyvoid 384157431Srwatsontcp_fini(void *xtp) 385128452Ssilby{ 386157431Srwatson 387128452Ssilby callout_stop(&isn_callout); 388128452Ssilby} 389128452Ssilby 3901541Srgrimes/* 39178642Ssilby * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 39278642Ssilby * tcp_template used to store this data in mbufs, but we now recopy it out 39378642Ssilby * of the tcpcb each time to conserve mbufs. 3941541Srgrimes */ 39578642Ssilbyvoid 396157431Srwatsontcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 3971541Srgrimes{ 398111144Sjlemon struct tcphdr *th = (struct tcphdr *)tcp_ptr; 3991541Srgrimes 400138410Srwatson INP_LOCK_ASSERT(inp); 401138410Srwatson 40255679Sshin#ifdef INET6 40355679Sshin if ((inp->inp_vflag & INP_IPV6) != 0) { 40478642Ssilby struct ip6_hdr *ip6; 40555679Sshin 40678642Ssilby ip6 = (struct ip6_hdr *)ip_ptr; 40755679Sshin ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 40855679Sshin (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK); 40955679Sshin ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 41055679Sshin (IPV6_VERSION & IPV6_VERSION_MASK); 41155679Sshin ip6->ip6_nxt = IPPROTO_TCP; 41255679Sshin ip6->ip6_plen = sizeof(struct tcphdr); 41355679Sshin ip6->ip6_src = inp->in6p_laddr; 41455679Sshin ip6->ip6_dst = inp->in6p_faddr; 41555679Sshin } else 41655679Sshin#endif 41778642Ssilby { 418111144Sjlemon struct ip *ip; 41955679Sshin 420111144Sjlemon ip = (struct ip *)ip_ptr; 421111144Sjlemon ip->ip_v = IPVERSION; 422111144Sjlemon ip->ip_hl = 5; 423111144Sjlemon ip->ip_tos = inp->inp_ip_tos; 424111144Sjlemon ip->ip_len = 0; 425111144Sjlemon ip->ip_id = 0; 426111144Sjlemon ip->ip_off = 0; 427111144Sjlemon ip->ip_ttl = inp->inp_ip_ttl; 428111144Sjlemon ip->ip_sum = 0; 429111144Sjlemon ip->ip_p = IPPROTO_TCP; 430111144Sjlemon ip->ip_src = inp->inp_laddr; 431111144Sjlemon ip->ip_dst = inp->inp_faddr; 43278642Ssilby } 433111144Sjlemon th->th_sport = inp->inp_lport; 434111144Sjlemon th->th_dport = inp->inp_fport; 435111144Sjlemon th->th_seq = 0; 436111144Sjlemon th->th_ack = 0; 437111144Sjlemon th->th_x2 = 0; 438111144Sjlemon th->th_off = 5; 439111144Sjlemon th->th_flags = 0; 440111144Sjlemon th->th_win = 0; 441111144Sjlemon th->th_urp = 0; 442111144Sjlemon th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 44378642Ssilby} 44478642Ssilby 44578642Ssilby/* 44678642Ssilby * Create template to be used to send tcp packets on a connection. 44778642Ssilby * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 44878642Ssilby * use for this function is in keepalives, which use tcp_respond. 44978642Ssilby */ 45078642Ssilbystruct tcptemp * 451157431Srwatsontcpip_maketemplate(struct inpcb *inp) 45278642Ssilby{ 45378642Ssilby struct mbuf *m; 45478642Ssilby struct tcptemp *n; 45578642Ssilby 456149635Sandre m = m_get(M_DONTWAIT, MT_DATA); 45778642Ssilby if (m == NULL) 45878642Ssilby return (0); 45978642Ssilby m->m_len = sizeof(struct tcptemp); 46078642Ssilby n = mtod(m, struct tcptemp *); 46178642Ssilby 462111144Sjlemon tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t); 4631541Srgrimes return (n); 4641541Srgrimes} 4651541Srgrimes 4661541Srgrimes/* 4671541Srgrimes * Send a single message to the TCP at address specified by 468127870Srwatson * the given TCP/IP header. If m == NULL, then we make a copy 4691541Srgrimes * of the tcpiphdr at ti and send directly to the addressed host. 4701541Srgrimes * This is used to force keep alive messages out using the TCP 47178642Ssilby * template for a connection. If flags are given then we send 47278642Ssilby * a message back to the TCP which originated the * segment ti, 47378642Ssilby * and discard the mbuf containing it and any other attached mbufs. 4741541Srgrimes * 4751541Srgrimes * In any case the ack and sequence number of the transmitted 4761541Srgrimes * segment are as specified by the parameters. 47731848Sjulian * 47831848Sjulian * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 4791541Srgrimes */ 4801541Srgrimesvoid 481157431Srwatsontcp_respond(struct tcpcb *tp, void *ipgen, register struct tcphdr *th, 482157431Srwatson register struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) 4831541Srgrimes{ 4841541Srgrimes register int tlen; 4851541Srgrimes int win = 0; 48655679Sshin struct ip *ip; 48755679Sshin struct tcphdr *nth; 48855679Sshin#ifdef INET6 48955679Sshin struct ip6_hdr *ip6; 49055679Sshin int isipv6; 49155679Sshin#endif /* INET6 */ 49255679Sshin int ipflags = 0; 493128905Srwatson struct inpcb *inp; 4941541Srgrimes 495101137Srwatson KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 496101137Srwatson 49755679Sshin#ifdef INET6 498105586Sphk isipv6 = ((struct ip *)ipgen)->ip_v == 6; 49955679Sshin ip6 = ipgen; 50055679Sshin#endif /* INET6 */ 50155679Sshin ip = ipgen; 50255679Sshin 503127871Srwatson if (tp != NULL) { 504122327Ssam inp = tp->t_inpcb; 505122327Ssam KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 506122327Ssam INP_INFO_WLOCK_ASSERT(&tcbinfo); 507122327Ssam INP_LOCK_ASSERT(inp); 508128905Srwatson } else 509128905Srwatson inp = NULL; 510128905Srwatson 511128905Srwatson if (tp != NULL) { 51257576Sps if (!(flags & TH_RST)) { 513122327Ssam win = sbspace(&inp->inp_socket->so_rcv); 51457576Sps if (win > (long)TCP_MAXWIN << tp->rcv_scale) 51557576Sps win = (long)TCP_MAXWIN << tp->rcv_scale; 51657576Sps } 5171541Srgrimes } 518127870Srwatson if (m == NULL) { 519151967Sandre m = m_gethdr(M_DONTWAIT, MT_DATA); 5201541Srgrimes if (m == NULL) 5211541Srgrimes return; 5221541Srgrimes tlen = 0; 5231541Srgrimes m->m_data += max_linkhdr; 52455679Sshin#ifdef INET6 52555679Sshin if (isipv6) { 526133874Srwatson bcopy((caddr_t)ip6, mtod(m, caddr_t), 52755679Sshin sizeof(struct ip6_hdr)); 52855679Sshin ip6 = mtod(m, struct ip6_hdr *); 52955679Sshin nth = (struct tcphdr *)(ip6 + 1); 53055679Sshin } else 53155679Sshin#endif /* INET6 */ 53255679Sshin { 53355679Sshin bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 53455679Sshin ip = mtod(m, struct ip *); 53555679Sshin nth = (struct tcphdr *)(ip + 1); 53655679Sshin } 53755679Sshin bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 5381541Srgrimes flags = TH_ACK; 5391541Srgrimes } else { 5401541Srgrimes m_freem(m->m_next); 541127870Srwatson m->m_next = NULL; 54255679Sshin m->m_data = (caddr_t)ipgen; 54355679Sshin /* m_len is set later */ 5441541Srgrimes tlen = 0; 5451541Srgrimes#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 54655679Sshin#ifdef INET6 54755679Sshin if (isipv6) { 54855679Sshin xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 54955679Sshin nth = (struct tcphdr *)(ip6 + 1); 55055679Sshin } else 55155679Sshin#endif /* INET6 */ 55255679Sshin { 55355679Sshin xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long); 55455679Sshin nth = (struct tcphdr *)(ip + 1); 55555679Sshin } 55655679Sshin if (th != nth) { 55755679Sshin /* 55855679Sshin * this is usually a case when an extension header 55955679Sshin * exists between the IPv6 header and the 56055679Sshin * TCP header. 56155679Sshin */ 56255679Sshin nth->th_sport = th->th_sport; 56355679Sshin nth->th_dport = th->th_dport; 56455679Sshin } 56555679Sshin xchg(nth->th_dport, nth->th_sport, n_short); 5661541Srgrimes#undef xchg 5671541Srgrimes } 56855679Sshin#ifdef INET6 56955679Sshin if (isipv6) { 57090198Sume ip6->ip6_flow = 0; 57190198Sume ip6->ip6_vfc = IPV6_VERSION; 57290198Sume ip6->ip6_nxt = IPPROTO_TCP; 57355679Sshin ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) + 57455679Sshin tlen)); 57555679Sshin tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 57656039Sshin } else 57755679Sshin#endif 578133874Srwatson { 579133874Srwatson tlen += sizeof (struct tcpiphdr); 580133874Srwatson ip->ip_len = tlen; 581133874Srwatson ip->ip_ttl = ip_defttl; 582133874Srwatson if (path_mtu_discovery) 583133874Srwatson ip->ip_off |= IP_DF; 584133874Srwatson } 5851541Srgrimes m->m_len = tlen; 5861541Srgrimes m->m_pkthdr.len = tlen; 587127870Srwatson m->m_pkthdr.rcvif = NULL; 588101106Srwatson#ifdef MAC 589122327Ssam if (inp != NULL) { 590101106Srwatson /* 591101106Srwatson * Packet is associated with a socket, so allow the 592101106Srwatson * label of the response to reflect the socket label. 593101106Srwatson */ 594128905Srwatson INP_LOCK_ASSERT(inp); 595128905Srwatson mac_create_mbuf_from_inpcb(inp, m); 596101106Srwatson } else { 597101106Srwatson /* 598119245Srwatson * Packet is not associated with a socket, so possibly 599119245Srwatson * update the label in place. 600101106Srwatson */ 601119245Srwatson mac_reflect_mbuf_tcp(m); 602101106Srwatson } 603101106Srwatson#endif 60455679Sshin nth->th_seq = htonl(seq); 60555679Sshin nth->th_ack = htonl(ack); 60655679Sshin nth->th_x2 = 0; 60755679Sshin nth->th_off = sizeof (struct tcphdr) >> 2; 60855679Sshin nth->th_flags = flags; 609127870Srwatson if (tp != NULL) 61055679Sshin nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 6111541Srgrimes else 61255679Sshin nth->th_win = htons((u_short)win); 61355679Sshin nth->th_urp = 0; 61455679Sshin#ifdef INET6 61555679Sshin if (isipv6) { 61659392Sshin nth->th_sum = 0; 61755679Sshin nth->th_sum = in6_cksum(m, IPPROTO_TCP, 61855679Sshin sizeof(struct ip6_hdr), 61955679Sshin tlen - sizeof(struct ip6_hdr)); 620127870Srwatson ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 621127870Srwatson NULL, NULL); 62255679Sshin } else 62355679Sshin#endif /* INET6 */ 624133874Srwatson { 625133874Srwatson nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 626133874Srwatson htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 627133874Srwatson m->m_pkthdr.csum_flags = CSUM_TCP; 628133874Srwatson m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 629133874Srwatson } 6306283Swollman#ifdef TCPDEBUG 631122327Ssam if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 63255679Sshin tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 6336283Swollman#endif 63455679Sshin#ifdef INET6 635122922Sandre if (isipv6) 636122922Sandre (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 637122922Sandre else 63855679Sshin#endif /* INET6 */ 639122922Sandre (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 6401541Srgrimes} 6411541Srgrimes 6421541Srgrimes/* 6431541Srgrimes * Create a new TCP control block, making an 6441541Srgrimes * empty reassembly queue and hooking it to the argument 64534881Swollman * protocol control block. The `inp' parameter must have 64634881Swollman * come from the zone allocator set up in tcp_init(). 6471541Srgrimes */ 6481541Srgrimesstruct tcpcb * 649157431Srwatsontcp_newtcpcb(struct inpcb *inp) 6501541Srgrimes{ 651111145Sjlemon struct tcpcb_mem *tm; 652111145Sjlemon struct tcpcb *tp; 65355679Sshin#ifdef INET6 65455679Sshin int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 65555679Sshin#endif /* INET6 */ 6561541Srgrimes 657111145Sjlemon tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); 658111145Sjlemon if (tm == NULL) 659111145Sjlemon return (NULL); 660111145Sjlemon tp = &tm->tcb; 661111145Sjlemon /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 66255679Sshin tp->t_maxseg = tp->t_maxopd = 66355679Sshin#ifdef INET6 66455679Sshin isipv6 ? tcp_v6mssdflt : 66555679Sshin#endif /* INET6 */ 66655679Sshin tcp_mssdflt; 6671541Srgrimes 66850673Sjlemon /* Set up our timeouts. */ 669142906Sglebius callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, NET_CALLOUT_MPSAFE); 670142906Sglebius callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, NET_CALLOUT_MPSAFE); 671142906Sglebius callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, NET_CALLOUT_MPSAFE); 672142906Sglebius callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, NET_CALLOUT_MPSAFE); 673142906Sglebius callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, NET_CALLOUT_MPSAFE); 67450673Sjlemon 6756283Swollman if (tcp_do_rfc1323) 6766283Swollman tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 677130989Sps tp->sack_enable = tcp_do_sack; 678147735Sps TAILQ_INIT(&tp->snd_holes); 67934881Swollman tp->t_inpcb = inp; /* XXX */ 6801541Srgrimes /* 6811541Srgrimes * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 68216367Swollman * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 6831541Srgrimes * reasonable initial retransmit time. 6841541Srgrimes */ 6851541Srgrimes tp->t_srtt = TCPTV_SRTTBASE; 68616367Swollman tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 687100335Sdillon tp->t_rttmin = tcp_rexmit_min; 68816367Swollman tp->t_rxtcur = TCPTV_RTOBASE; 6891541Srgrimes tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 690102017Sdillon tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 6911541Srgrimes tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 69250673Sjlemon tp->t_rcvtime = ticks; 693102017Sdillon tp->t_bw_rtttime = ticks; 694133874Srwatson /* 69556564Sshin * IPv4 TTL initialization is necessary for an IPv6 socket as well, 69656564Sshin * because the socket may be bound to an IPv6 wildcard address, 69756564Sshin * which may match an IPv4-mapped IPv6 address. 69856564Sshin */ 69924570Sdg inp->inp_ip_ttl = ip_defttl; 700157432Srwatson inp->inp_ppcb = tp; 70134881Swollman return (tp); /* XXX */ 7021541Srgrimes} 7031541Srgrimes 7041541Srgrimes/* 7051541Srgrimes * Drop a TCP connection, reporting 7061541Srgrimes * the specified error. If connection is synchronized, 7071541Srgrimes * then send a RST to peer. 7081541Srgrimes */ 7091541Srgrimesstruct tcpcb * 710157431Srwatsontcp_drop(struct tcpcb *tp, int errno) 7111541Srgrimes{ 7121541Srgrimes struct socket *so = tp->t_inpcb->inp_socket; 7131541Srgrimes 714146864Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); 715138019Srwatson INP_LOCK_ASSERT(tp->t_inpcb); 716146864Srwatson 7171541Srgrimes if (TCPS_HAVERCVDSYN(tp->t_state)) { 7181541Srgrimes tp->t_state = TCPS_CLOSED; 7191541Srgrimes (void) tcp_output(tp); 7201541Srgrimes tcpstat.tcps_drops++; 7211541Srgrimes } else 7221541Srgrimes tcpstat.tcps_conndrops++; 7231541Srgrimes if (errno == ETIMEDOUT && tp->t_softerror) 7241541Srgrimes errno = tp->t_softerror; 7251541Srgrimes so->so_error = errno; 7261541Srgrimes return (tcp_close(tp)); 7271541Srgrimes} 7281541Srgrimes 729157376Srwatsonvoid 730157431Srwatsontcp_discardcb(struct tcpcb *tp) 7311541Srgrimes{ 732111145Sjlemon struct tseg_qent *q; 7331541Srgrimes struct inpcb *inp = tp->t_inpcb; 7341541Srgrimes struct socket *so = inp->inp_socket; 73555679Sshin#ifdef INET6 73655679Sshin int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 73755679Sshin#endif /* INET6 */ 7381541Srgrimes 739138410Srwatson INP_LOCK_ASSERT(inp); 740138410Srwatson 7411541Srgrimes /* 74250673Sjlemon * Make sure that all of our timers are stopped before we 74350673Sjlemon * delete the PCB. 74450673Sjlemon */ 74550673Sjlemon callout_stop(tp->tt_rexmt); 74650673Sjlemon callout_stop(tp->tt_persist); 74750673Sjlemon callout_stop(tp->tt_keep); 74850673Sjlemon callout_stop(tp->tt_2msl); 74950673Sjlemon callout_stop(tp->tt_delack); 75050673Sjlemon 75150673Sjlemon /* 7529373Swollman * If we got enough samples through the srtt filter, 7539373Swollman * save the rtt and rttvar in the routing entry. 754122922Sandre * 'Enough' is arbitrarily defined as 4 rtt samples. 755122922Sandre * 4 samples is enough for the srtt filter to converge 756122922Sandre * to within enough % of the correct value; fewer samples 757122922Sandre * and we could save a bogus rtt. The danger is not high 758122922Sandre * as tcp quickly recovers from everything. 759122922Sandre * XXX: Works very well but needs some more statistics! 7601541Srgrimes */ 761122922Sandre if (tp->t_rttupdated >= 4) { 762122922Sandre struct hc_metrics_lite metrics; 763122922Sandre u_long ssthresh; 7641541Srgrimes 765122922Sandre bzero(&metrics, sizeof(metrics)); 7661541Srgrimes /* 767122922Sandre * Update the ssthresh always when the conditions below 768122922Sandre * are satisfied. This gives us better new start value 769122922Sandre * for the congestion avoidance for new connections. 770122922Sandre * ssthresh is only set if packet loss occured on a session. 771160925Srwatson * 772160925Srwatson * XXXRW: 'so' may be NULL here, and/or socket buffer may be 773160925Srwatson * being torn down. Ideally this code would not use 'so'. 7741541Srgrimes */ 775122922Sandre ssthresh = tp->snd_ssthresh; 776122922Sandre if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 7771541Srgrimes /* 7781541Srgrimes * convert the limit from user data bytes to 7791541Srgrimes * packets then to packet data bytes. 7801541Srgrimes */ 781122922Sandre ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 782122922Sandre if (ssthresh < 2) 783122922Sandre ssthresh = 2; 784122922Sandre ssthresh *= (u_long)(tp->t_maxseg + 78555679Sshin#ifdef INET6 78655679Sshin (isipv6 ? sizeof (struct ip6_hdr) + 78755679Sshin sizeof (struct tcphdr) : 78855679Sshin#endif 78955679Sshin sizeof (struct tcpiphdr) 79055679Sshin#ifdef INET6 79155679Sshin ) 79255679Sshin#endif 79355679Sshin ); 794122922Sandre } else 795122922Sandre ssthresh = 0; 796122922Sandre metrics.rmx_ssthresh = ssthresh; 797122922Sandre 798122922Sandre metrics.rmx_rtt = tp->t_srtt; 799122922Sandre metrics.rmx_rttvar = tp->t_rttvar; 800122922Sandre /* XXX: This wraps if the pipe is more than 4 Gbit per second */ 801122922Sandre metrics.rmx_bandwidth = tp->snd_bandwidth; 802122922Sandre metrics.rmx_cwnd = tp->snd_cwnd; 803133874Srwatson metrics.rmx_sendpipe = 0; 804122922Sandre metrics.rmx_recvpipe = 0; 805122922Sandre 806122922Sandre tcp_hc_update(&inp->inp_inc, &metrics); 8071541Srgrimes } 808122922Sandre 8091541Srgrimes /* free the reassembly queue, if any */ 810111145Sjlemon while ((q = LIST_FIRST(&tp->t_segq)) != NULL) { 81155679Sshin LIST_REMOVE(q, tqe_q); 81255679Sshin m_freem(q->tqe_m); 813126193Sandre uma_zfree(tcp_reass_zone, q); 814126193Sandre tp->t_segqlen--; 815126193Sandre tcp_reass_qsize--; 8161541Srgrimes } 817130989Sps tcp_free_sackholes(tp); 81832821Sdg inp->inp_ppcb = NULL; 819108265Shsu tp->t_inpcb = NULL; 820111145Sjlemon uma_zfree(tcpcb_zone, tp); 821111145Sjlemon} 822111145Sjlemon 823111145Sjlemon/* 824157430Srwatson * Attempt to close a TCP control block, marking it as dropped, and freeing 825157430Srwatson * the socket if we hold the only reference. 826111145Sjlemon */ 827111145Sjlemonstruct tcpcb * 828157431Srwatsontcp_close(struct tcpcb *tp) 829111145Sjlemon{ 830111145Sjlemon struct inpcb *inp = tp->t_inpcb; 831157376Srwatson struct socket *so; 832111145Sjlemon 833146864Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); 834138410Srwatson INP_LOCK_ASSERT(inp); 835138410Srwatson 836158009Srwatson in_pcbdrop(inp); 837157376Srwatson tcpstat.tcps_closed++; 838157376Srwatson KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 839157376Srwatson so = inp->inp_socket; 840157376Srwatson soisdisconnected(so); 841157376Srwatson if (inp->inp_vflag & INP_SOCKREF) { 842157376Srwatson KASSERT(so->so_state & SS_PROTOREF, 843157376Srwatson ("tcp_close: !SS_PROTOREF")); 844157376Srwatson inp->inp_vflag &= ~INP_SOCKREF; 845160549Srwatson INP_UNLOCK(inp); 846157376Srwatson ACCEPT_LOCK(); 847157376Srwatson SOCK_LOCK(so); 848157376Srwatson so->so_state &= ~SS_PROTOREF; 849157376Srwatson sofree(so); 850157376Srwatson return (NULL); 851157376Srwatson } 852157376Srwatson return (tp); 8531541Srgrimes} 8541541Srgrimes 8551541Srgrimesvoid 856157431Srwatsontcp_drain(void) 8571541Srgrimes{ 858157431Srwatson 859157431Srwatson if (do_tcpdrain) { 86055198Smsmith struct inpcb *inpb; 86155198Smsmith struct tcpcb *tcpb; 86255679Sshin struct tseg_qent *te; 8631541Srgrimes 86455198Smsmith /* 86555198Smsmith * Walk the tcpbs, if existing, and flush the reassembly queue, 86655198Smsmith * if there is one... 86755198Smsmith * XXX: The "Net/3" implementation doesn't imply that the TCP 86855198Smsmith * reassembly queue should be flushed, but in a situation 869133874Srwatson * where we're really low on mbufs, this is potentially 870133874Srwatson * usefull. 87155198Smsmith */ 87298102Shsu INP_INFO_RLOCK(&tcbinfo); 87374362Sphk LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) { 874111145Sjlemon if (inpb->inp_vflag & INP_TIMEWAIT) 875111145Sjlemon continue; 87698102Shsu INP_LOCK(inpb); 877127870Srwatson if ((tcpb = intotcpcb(inpb)) != NULL) { 87874362Sphk while ((te = LIST_FIRST(&tcpb->t_segq)) 87974362Sphk != NULL) { 88055679Sshin LIST_REMOVE(te, tqe_q); 88155679Sshin m_freem(te->tqe_m); 882126193Sandre uma_zfree(tcp_reass_zone, te); 883126193Sandre tcpb->t_segqlen--; 884126193Sandre tcp_reass_qsize--; 88555198Smsmith } 886144857Sps tcp_clean_sackreport(tcpb); 88755198Smsmith } 88898102Shsu INP_UNLOCK(inpb); 88955198Smsmith } 89098102Shsu INP_INFO_RUNLOCK(&tcbinfo); 89155198Smsmith } 8921541Srgrimes} 8931541Srgrimes 8941541Srgrimes/* 8951541Srgrimes * Notify a tcp user of an asynchronous error; 8961541Srgrimes * store error as soft error, but wake up user 8971541Srgrimes * (for now, won't do anything until can select for soft error). 89872960Sjlemon * 89972960Sjlemon * Do not wake up user since there currently is no mechanism for 90072960Sjlemon * reporting soft errors (yet - a kqueue filter may be added). 9011541Srgrimes */ 90298211Shsustatic struct inpcb * 903157431Srwatsontcp_notify(struct inpcb *inp, int error) 9041541Srgrimes{ 905157432Srwatson struct tcpcb *tp; 9061541Srgrimes 907146864Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); 908138410Srwatson INP_LOCK_ASSERT(inp); 909157433Srwatson 910157433Srwatson if ((inp->inp_vflag & INP_TIMEWAIT) || 911157433Srwatson (inp->inp_vflag & INP_DROPPED)) 912157433Srwatson return (inp); 913157433Srwatson 914157432Srwatson tp = intotcpcb(inp); 915157433Srwatson KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 916138410Srwatson 9171541Srgrimes /* 9181541Srgrimes * Ignore some errors if we are hooked up. 9191541Srgrimes * If connection hasn't completed, has retransmitted several times, 9201541Srgrimes * and receives a second error, give up now. This is better 9211541Srgrimes * than waiting a long time to establish a connection that 9221541Srgrimes * can never complete. 9231541Srgrimes */ 9241541Srgrimes if (tp->t_state == TCPS_ESTABLISHED && 925110896Shsu (error == EHOSTUNREACH || error == ENETUNREACH || 926110896Shsu error == EHOSTDOWN)) { 927139222Srwatson return (inp); 9281541Srgrimes } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 92998211Shsu tp->t_softerror) { 930157376Srwatson tp = tcp_drop(tp, error); 931157376Srwatson if (tp != NULL) 932157376Srwatson return (inp); 933157376Srwatson else 934157376Srwatson return (NULL); 93598211Shsu } else { 9361541Srgrimes tp->t_softerror = error; 937139222Srwatson return (inp); 93898211Shsu } 93972960Sjlemon#if 0 940111748Sdes wakeup( &so->so_timeo); 9411541Srgrimes sorwakeup(so); 9421541Srgrimes sowwakeup(so); 94372960Sjlemon#endif 9441541Srgrimes} 9451541Srgrimes 94636079Swollmanstatic int 94762573Sphktcp_pcblist(SYSCTL_HANDLER_ARGS) 94836079Swollman{ 949148156Srwatson int error, i, n; 95036079Swollman struct inpcb *inp, **inp_list; 95136079Swollman inp_gen_t gencnt; 95236079Swollman struct xinpgen xig; 95336079Swollman 95436079Swollman /* 95536079Swollman * The process of preparing the TCB list is too time-consuming and 95636079Swollman * resource-intensive to repeat twice on every request. 95736079Swollman */ 958127870Srwatson if (req->oldptr == NULL) { 95936079Swollman n = tcbinfo.ipi_count; 96036079Swollman req->oldidx = 2 * (sizeof xig) 96136079Swollman + (n + n/8) * sizeof(struct xtcpcb); 962139222Srwatson return (0); 96336079Swollman } 96436079Swollman 965127870Srwatson if (req->newptr != NULL) 966139222Srwatson return (EPERM); 96736079Swollman 96836079Swollman /* 96936079Swollman * OK, now we're committed to doing something. 97036079Swollman */ 97198102Shsu INP_INFO_RLOCK(&tcbinfo); 97236079Swollman gencnt = tcbinfo.ipi_gencnt; 97336079Swollman n = tcbinfo.ipi_count; 97498102Shsu INP_INFO_RUNLOCK(&tcbinfo); 97536079Swollman 976126253Struckman error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 977100831Struckman + n * sizeof(struct xtcpcb)); 978126253Struckman if (error != 0) 979126253Struckman return (error); 980100831Struckman 98136079Swollman xig.xig_len = sizeof xig; 98236079Swollman xig.xig_count = n; 98336079Swollman xig.xig_gen = gencnt; 98436079Swollman xig.xig_sogen = so_gencnt; 98536079Swollman error = SYSCTL_OUT(req, &xig, sizeof xig); 98636079Swollman if (error) 987139222Srwatson return (error); 98836079Swollman 989111119Simp inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 990127870Srwatson if (inp_list == NULL) 991139222Srwatson return (ENOMEM); 992133874Srwatson 99398102Shsu INP_INFO_RLOCK(&tcbinfo); 994127870Srwatson for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n; 99571999Sphk inp = LIST_NEXT(inp, inp_list)) { 99698102Shsu INP_LOCK(inp); 997113345Srwatson if (inp->inp_gencnt <= gencnt) { 998113345Srwatson /* 999113345Srwatson * XXX: This use of cr_cansee(), introduced with 1000113345Srwatson * TCP state changes, is not quite right, but for 1001113345Srwatson * now, better than nothing. 1002113345Srwatson */ 1003157474Srwatson if (inp->inp_vflag & INP_TIMEWAIT) { 1004157474Srwatson if (intotw(inp) != NULL) 1005157474Srwatson error = cr_cansee(req->td->td_ucred, 1006157474Srwatson intotw(inp)->tw_cred); 1007157474Srwatson else 1008157474Srwatson error = EINVAL; /* Skip this inp. */ 1009157474Srwatson } else 1010113345Srwatson error = cr_canseesocket(req->td->td_ucred, 1011113345Srwatson inp->inp_socket); 1012113345Srwatson if (error == 0) 1013113345Srwatson inp_list[i++] = inp; 1014113345Srwatson } 101598102Shsu INP_UNLOCK(inp); 101636079Swollman } 101798102Shsu INP_INFO_RUNLOCK(&tcbinfo); 101836079Swollman n = i; 101936079Swollman 102036079Swollman error = 0; 102136079Swollman for (i = 0; i < n; i++) { 102236079Swollman inp = inp_list[i]; 1023160491Sups INP_LOCK(inp); 102436079Swollman if (inp->inp_gencnt <= gencnt) { 102536079Swollman struct xtcpcb xt; 1026157432Srwatson void *inp_ppcb; 1027145978Scperciva 1028145978Scperciva bzero(&xt, sizeof(xt)); 102936079Swollman xt.xt_len = sizeof xt; 103036079Swollman /* XXX should avoid extra copy */ 103136079Swollman bcopy(inp, &xt.xt_inp, sizeof *inp); 103247960Stegge inp_ppcb = inp->inp_ppcb; 1033111145Sjlemon if (inp_ppcb == NULL) 1034111145Sjlemon bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1035111145Sjlemon else if (inp->inp_vflag & INP_TIMEWAIT) { 1036111145Sjlemon bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1037111145Sjlemon xt.xt_tp.t_state = TCPS_TIME_WAIT; 1038111145Sjlemon } else 103947960Stegge bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1040127870Srwatson if (inp->inp_socket != NULL) 104136079Swollman sotoxsocket(inp->inp_socket, &xt.xt_socket); 1042111145Sjlemon else { 1043111145Sjlemon bzero(&xt.xt_socket, sizeof xt.xt_socket); 1044111145Sjlemon xt.xt_socket.xso_protocol = IPPROTO_TCP; 1045111145Sjlemon } 1046110896Shsu xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1047160491Sups INP_UNLOCK(inp); 104836079Swollman error = SYSCTL_OUT(req, &xt, sizeof xt); 1049160491Sups } else 1050160491Sups INP_UNLOCK(inp); 1051160491Sups 105236079Swollman } 105336079Swollman if (!error) { 105436079Swollman /* 105536079Swollman * Give the user an updated idea of our state. 105636079Swollman * If the generation differs from what we told 105736079Swollman * her before, she knows that something happened 105836079Swollman * while we were processing this request, and it 105936079Swollman * might be necessary to retry. 106036079Swollman */ 106198102Shsu INP_INFO_RLOCK(&tcbinfo); 106236079Swollman xig.xig_gen = tcbinfo.ipi_gencnt; 106336079Swollman xig.xig_sogen = so_gencnt; 106436079Swollman xig.xig_count = tcbinfo.ipi_count; 106598102Shsu INP_INFO_RUNLOCK(&tcbinfo); 106636079Swollman error = SYSCTL_OUT(req, &xig, sizeof xig); 106736079Swollman } 106836079Swollman free(inp_list, M_TEMP); 1069139222Srwatson return (error); 107036079Swollman} 107136079Swollman 107236079SwollmanSYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, 107336079Swollman tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 107436079Swollman 107548758Sgreenstatic int 107662573Sphktcp_getcred(SYSCTL_HANDLER_ARGS) 107748758Sgreen{ 107872650Sgreen struct xucred xuc; 107948758Sgreen struct sockaddr_in addrs[2]; 108048758Sgreen struct inpcb *inp; 1081148156Srwatson int error; 108248758Sgreen 1083132653Scperciva error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 108448758Sgreen if (error) 108548758Sgreen return (error); 108648758Sgreen error = SYSCTL_IN(req, addrs, sizeof(addrs)); 108748758Sgreen if (error) 108848758Sgreen return (error); 108998102Shsu INP_INFO_RLOCK(&tcbinfo); 109048758Sgreen inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 109154263Sshin addrs[0].sin_addr, addrs[0].sin_port, 0, NULL); 109298102Shsu if (inp == NULL) { 109348758Sgreen error = ENOENT; 109498102Shsu goto outunlocked; 109548758Sgreen } 109699837Struckman INP_LOCK(inp); 109799837Struckman if (inp->inp_socket == NULL) { 109899837Struckman error = ENOENT; 109999837Struckman goto out; 110099837Struckman } 110192976Srwatson error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 110278697Sdwmalone if (error) 110378697Sdwmalone goto out; 110491354Sdd cru2x(inp->inp_socket->so_cred, &xuc); 110548758Sgreenout: 110698102Shsu INP_UNLOCK(inp); 110798102Shsuoutunlocked: 110898102Shsu INP_INFO_RUNLOCK(&tcbinfo); 110999838Struckman if (error == 0) 111099838Struckman error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 111148758Sgreen return (error); 111248758Sgreen} 111348758Sgreen 111478697SdwmaloneSYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 111578697Sdwmalone CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 111678697Sdwmalone tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 111748758Sgreen 111855679Sshin#ifdef INET6 111955679Sshinstatic int 112062573Sphktcp6_getcred(SYSCTL_HANDLER_ARGS) 112155679Sshin{ 112272650Sgreen struct xucred xuc; 112355679Sshin struct sockaddr_in6 addrs[2]; 112455679Sshin struct inpcb *inp; 1125148156Srwatson int error, mapped = 0; 112655679Sshin 1127132653Scperciva error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL); 112855679Sshin if (error) 112955679Sshin return (error); 113055679Sshin error = SYSCTL_IN(req, addrs, sizeof(addrs)); 113155679Sshin if (error) 113255679Sshin return (error); 1133148385Sume if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 || 1134148385Sume (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) { 1135148385Sume return (error); 1136148385Sume } 113755679Sshin if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 113855679Sshin if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 113955679Sshin mapped = 1; 114055679Sshin else 114155679Sshin return (EINVAL); 114255679Sshin } 1143148385Sume 114498102Shsu INP_INFO_RLOCK(&tcbinfo); 114555679Sshin if (mapped == 1) 114655679Sshin inp = in_pcblookup_hash(&tcbinfo, 114755679Sshin *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 114855679Sshin addrs[1].sin6_port, 114955679Sshin *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 115055679Sshin addrs[0].sin6_port, 115155679Sshin 0, NULL); 115255679Sshin else 1153151254Sphilip inp = in6_pcblookup_hash(&tcbinfo, 1154151254Sphilip &addrs[1].sin6_addr, addrs[1].sin6_port, 1155151254Sphilip &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL); 115698102Shsu if (inp == NULL) { 115755679Sshin error = ENOENT; 115898102Shsu goto outunlocked; 115955679Sshin } 116099837Struckman INP_LOCK(inp); 116199837Struckman if (inp->inp_socket == NULL) { 116299837Struckman error = ENOENT; 116399837Struckman goto out; 116499837Struckman } 116592976Srwatson error = cr_canseesocket(req->td->td_ucred, inp->inp_socket); 116678697Sdwmalone if (error) 116778697Sdwmalone goto out; 116891354Sdd cru2x(inp->inp_socket->so_cred, &xuc); 116955679Sshinout: 117098102Shsu INP_UNLOCK(inp); 117198102Shsuoutunlocked: 117298102Shsu INP_INFO_RUNLOCK(&tcbinfo); 117399838Struckman if (error == 0) 117499838Struckman error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 117555679Sshin return (error); 117655679Sshin} 117755679Sshin 117878697SdwmaloneSYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 117978697Sdwmalone CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 118078697Sdwmalone tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 118155679Sshin#endif 118255679Sshin 118355679Sshin 11841541Srgrimesvoid 1185157431Srwatsontcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 11861541Srgrimes{ 118772959Sjlemon struct ip *ip = vip; 118872959Sjlemon struct tcphdr *th; 118973109Sjlemon struct in_addr faddr; 119073109Sjlemon struct inpcb *inp; 119173109Sjlemon struct tcpcb *tp; 119298211Shsu struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1193145360Sandre struct icmp *icp; 1194145360Sandre struct in_conninfo inc; 1195145360Sandre tcp_seq icmp_tcp_seq; 1196148156Srwatson int mtu; 11971541Srgrimes 119873109Sjlemon faddr = ((struct sockaddr_in *)sa)->sin_addr; 119973109Sjlemon if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 120073109Sjlemon return; 120173109Sjlemon 1202145355Sandre if (cmd == PRC_MSGSIZE) 1203145355Sandre notify = tcp_mtudisc; 120474937Sjesper else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 120599156Sjesper cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 120672959Sjlemon notify = tcp_drop_syn_sent; 1207122922Sandre /* 1208122922Sandre * Redirects don't need to be handled up here. 1209122922Sandre */ 1210122922Sandre else if (PRC_IS_REDIRECT(cmd)) 1211122922Sandre return; 1212122922Sandre /* 1213145355Sandre * Source quench is depreciated. 1214145355Sandre */ 1215145355Sandre else if (cmd == PRC_QUENCH) 1216145355Sandre return; 1217145355Sandre /* 1218122922Sandre * Hostdead is ugly because it goes linearly through all PCBs. 1219122922Sandre * XXX: We never get this from ICMP, otherwise it makes an 1220122922Sandre * excellent DoS attack on machines with many connections. 1221122922Sandre */ 1222122922Sandre else if (cmd == PRC_HOSTDEAD) 1223127870Srwatson ip = NULL; 1224119995Sru else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 12251541Srgrimes return; 1226127870Srwatson if (ip != NULL) { 1227145360Sandre icp = (struct icmp *)((caddr_t)ip 1228145360Sandre - offsetof(struct icmp, icmp_ip)); 1229133874Srwatson th = (struct tcphdr *)((caddr_t)ip 1230105586Sphk + (ip->ip_hl << 2)); 123198596Shsu INP_INFO_WLOCK(&tcbinfo); 123273109Sjlemon inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport, 123373109Sjlemon ip->ip_src, th->th_sport, 0, NULL); 123498102Shsu if (inp != NULL) { 123598102Shsu INP_LOCK(inp); 1236157433Srwatson if (!(inp->inp_vflag & INP_TIMEWAIT) && 1237157433Srwatson !(inp->inp_vflag & INP_DROPPED) && 1238157433Srwatson !(inp->inp_socket == NULL)) { 1239145360Sandre icmp_tcp_seq = htonl(th->th_seq); 124098102Shsu tp = intotcpcb(inp); 1241145360Sandre if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1242145360Sandre SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1243145360Sandre if (cmd == PRC_MSGSIZE) { 1244145360Sandre /* 1245145360Sandre * MTU discovery: 1246145360Sandre * If we got a needfrag set the MTU 1247145360Sandre * in the route to the suggested new 1248145360Sandre * value (if given) and then notify. 1249145360Sandre */ 1250145360Sandre bzero(&inc, sizeof(inc)); 1251145360Sandre inc.inc_flags = 0; /* IPv4 */ 1252145360Sandre inc.inc_faddr = faddr; 1253145360Sandre 1254145360Sandre mtu = ntohs(icp->icmp_nextmtu); 1255145869Sandre /* 1256145869Sandre * If no alternative MTU was 1257145869Sandre * proposed, try the next smaller 1258149929Sandre * one. ip->ip_len has already 1259149929Sandre * been swapped in icmp_input(). 1260145869Sandre */ 1261145360Sandre if (!mtu) 1262149929Sandre mtu = ip_next_mtu(ip->ip_len, 1263145869Sandre 1); 1264145869Sandre if (mtu < max(296, (tcp_minmss) 1265145869Sandre + sizeof(struct tcpiphdr))) 1266145869Sandre mtu = 0; 1267145869Sandre if (!mtu) 1268145869Sandre mtu = tcp_mssdflt 1269145869Sandre + sizeof(struct tcpiphdr); 1270145869Sandre /* 1271145869Sandre * Only cache the the MTU if it 1272145869Sandre * is smaller than the interface 1273145869Sandre * or route MTU. tcp_mtudisc() 1274145869Sandre * will do right thing by itself. 1275145869Sandre */ 1276162084Sandre if (mtu <= tcp_maxmtu(&inc, NULL)) 1277145360Sandre tcp_hc_updatemtu(&inc, mtu); 1278145360Sandre } 1279145360Sandre 128098211Shsu inp = (*notify)(inp, inetctlerrmap[cmd]); 1281145360Sandre } 128298102Shsu } 1283127870Srwatson if (inp != NULL) 128498211Shsu INP_UNLOCK(inp); 128586764Sjlemon } else { 128686764Sjlemon inc.inc_fport = th->th_dport; 128786764Sjlemon inc.inc_lport = th->th_sport; 128886764Sjlemon inc.inc_faddr = faddr; 128986764Sjlemon inc.inc_laddr = ip->ip_src; 129086764Sjlemon#ifdef INET6 129186764Sjlemon inc.inc_isipv6 = 0; 129286764Sjlemon#endif 129386764Sjlemon syncache_unreach(&inc, th); 129473109Sjlemon } 129598596Shsu INP_INFO_WUNLOCK(&tcbinfo); 12961541Srgrimes } else 129798102Shsu in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify); 12981541Srgrimes} 12991541Srgrimes 130055679Sshin#ifdef INET6 130155679Sshinvoid 1302157431Srwatsontcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 130355679Sshin{ 130455679Sshin struct tcphdr th; 130598211Shsu struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 130655679Sshin struct ip6_hdr *ip6; 130755679Sshin struct mbuf *m; 130878064Sume struct ip6ctlparam *ip6cp = NULL; 130978064Sume const struct sockaddr_in6 *sa6_src = NULL; 131055679Sshin int off; 131178064Sume struct tcp_portonly { 131278064Sume u_int16_t th_sport; 131378064Sume u_int16_t th_dport; 131478064Sume } *thp; 131555679Sshin 131655679Sshin if (sa->sa_family != AF_INET6 || 131755679Sshin sa->sa_len != sizeof(struct sockaddr_in6)) 131855679Sshin return; 131955679Sshin 1320145355Sandre if (cmd == PRC_MSGSIZE) 132155679Sshin notify = tcp_mtudisc; 132255679Sshin else if (!PRC_IS_REDIRECT(cmd) && 1323119995Sru ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 132455679Sshin return; 1325145355Sandre /* Source quench is depreciated. */ 1326145355Sandre else if (cmd == PRC_QUENCH) 1327145355Sandre return; 132855679Sshin 132955679Sshin /* if the parameter is from icmp6, decode it. */ 133055679Sshin if (d != NULL) { 133178064Sume ip6cp = (struct ip6ctlparam *)d; 133255679Sshin m = ip6cp->ip6c_m; 133355679Sshin ip6 = ip6cp->ip6c_ip6; 133455679Sshin off = ip6cp->ip6c_off; 133578064Sume sa6_src = ip6cp->ip6c_src; 133655679Sshin } else { 133755679Sshin m = NULL; 133855679Sshin ip6 = NULL; 133967456Sitojun off = 0; /* fool gcc */ 134078064Sume sa6_src = &sa6_any; 134155679Sshin } 134255679Sshin 1343127870Srwatson if (ip6 != NULL) { 134486764Sjlemon struct in_conninfo inc; 134555679Sshin /* 134655679Sshin * XXX: We assume that when IPV6 is non NULL, 134755679Sshin * M and OFF are valid. 134855679Sshin */ 134955679Sshin 135067456Sitojun /* check if we can safely examine src and dst ports */ 135178064Sume if (m->m_pkthdr.len < off + sizeof(*thp)) 135267456Sitojun return; 135367456Sitojun 135478064Sume bzero(&th, sizeof(th)); 135578064Sume m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 135678064Sume 1357133192Srwatson in6_pcbnotify(&tcbinfo, sa, th.th_dport, 135878064Sume (struct sockaddr *)ip6cp->ip6c_src, 1359125776Sume th.th_sport, cmd, NULL, notify); 136086764Sjlemon 136186764Sjlemon inc.inc_fport = th.th_dport; 136286764Sjlemon inc.inc_lport = th.th_sport; 136386764Sjlemon inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 136486764Sjlemon inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 136586764Sjlemon inc.inc_isipv6 = 1; 1366133591Sdwmalone INP_INFO_WLOCK(&tcbinfo); 136786764Sjlemon syncache_unreach(&inc, &th); 1368133591Sdwmalone INP_INFO_WUNLOCK(&tcbinfo); 136955679Sshin } else 1370133192Srwatson in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1371125776Sume 0, cmd, NULL, notify); 137255679Sshin} 137355679Sshin#endif /* INET6 */ 137455679Sshin 137580428Speter 137682122Ssilby/* 137782122Ssilby * Following is where TCP initial sequence number generation occurs. 137882122Ssilby * 137982122Ssilby * There are two places where we must use initial sequence numbers: 138082122Ssilby * 1. In SYN-ACK packets. 138182122Ssilby * 2. In SYN packets. 138282122Ssilby * 138394390Ssilby * All ISNs for SYN-ACK packets are generated by the syncache. See 138494390Ssilby * tcp_syncache.c for details. 138582122Ssilby * 138682122Ssilby * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 138782122Ssilby * depends on this property. In addition, these ISNs should be 138882122Ssilby * unguessable so as to prevent connection hijacking. To satisfy 138982122Ssilby * the requirements of this situation, the algorithm outlined in 1390133874Srwatson * RFC 1948 is used, with only small modifications. 139182122Ssilby * 139282122Ssilby * Implementation details: 139382122Ssilby * 139482122Ssilby * Time is based off the system timer, and is corrected so that it 139582122Ssilby * increases by one megabyte per second. This allows for proper 139682122Ssilby * recycling on high speed LANs while still leaving over an hour 139782122Ssilby * before rollover. 139882122Ssilby * 1399128452Ssilby * As reading the *exact* system time is too expensive to be done 1400128452Ssilby * whenever setting up a TCP connection, we increment the time 1401128452Ssilby * offset in two ways. First, a small random positive increment 1402128452Ssilby * is added to isn_offset for each connection that is set up. 1403128452Ssilby * Second, the function tcp_isn_tick fires once per clock tick 1404128452Ssilby * and increments isn_offset as necessary so that sequence numbers 1405128452Ssilby * are incremented at approximately ISN_BYTES_PER_SECOND. The 1406128452Ssilby * random positive increments serve only to ensure that the same 1407128452Ssilby * exact sequence number is never sent out twice (as could otherwise 1408128452Ssilby * happen when a port is recycled in less than the system tick 1409128452Ssilby * interval.) 1410128452Ssilby * 141182122Ssilby * net.inet.tcp.isn_reseed_interval controls the number of seconds 141282122Ssilby * between seeding of isn_secret. This is normally set to zero, 141382122Ssilby * as reseeding should not be necessary. 141482122Ssilby * 1415138018Srwatson * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1416138018Srwatson * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1417138018Srwatson * general, this means holding an exclusive (write) lock. 141882122Ssilby */ 141979413Ssilby 142082122Ssilby#define ISN_BYTES_PER_SECOND 1048576 1421128452Ssilby#define ISN_STATIC_INCREMENT 4096 1422128452Ssilby#define ISN_RANDOM_INCREMENT (4096 - 1) 142379413Ssilby 1424138018Srwatsonstatic u_char isn_secret[32]; 1425138018Srwatsonstatic int isn_last_reseed; 1426138018Srwatsonstatic u_int32_t isn_offset, isn_offset_old; 1427138018Srwatsonstatic MD5_CTX isn_ctx; 142875619Skris 142975619Skristcp_seq 1430157431Srwatsontcp_new_isn(struct tcpcb *tp) 143175619Skris{ 143282122Ssilby u_int32_t md5_buffer[4]; 143382122Ssilby tcp_seq new_isn; 143475619Skris 1435138410Srwatson INP_LOCK_ASSERT(tp->t_inpcb); 1436138018Srwatson 1437157977Srwatson ISN_LOCK(); 143882122Ssilby /* Seed if this is the first use, reseed if requested. */ 143994390Ssilby if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) && 144082122Ssilby (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz) 144182122Ssilby < (u_int)ticks))) { 144282122Ssilby read_random(&isn_secret, sizeof(isn_secret)); 144382122Ssilby isn_last_reseed = ticks; 144482122Ssilby } 1445133874Srwatson 144682122Ssilby /* Compute the md5 hash and return the ISN. */ 144782122Ssilby MD5Init(&isn_ctx); 144882122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 144982122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 145082122Ssilby#ifdef INET6 145182122Ssilby if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 145282122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 145382122Ssilby sizeof(struct in6_addr)); 145482122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 145582122Ssilby sizeof(struct in6_addr)); 145682122Ssilby } else 145782122Ssilby#endif 145882122Ssilby { 145982122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 146082122Ssilby sizeof(struct in_addr)); 146182122Ssilby MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 146282122Ssilby sizeof(struct in_addr)); 146382122Ssilby } 146482122Ssilby MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret)); 146582122Ssilby MD5Final((u_char *) &md5_buffer, &isn_ctx); 146682122Ssilby new_isn = (tcp_seq) md5_buffer[0]; 1467128452Ssilby isn_offset += ISN_STATIC_INCREMENT + 1468128452Ssilby (arc4random() & ISN_RANDOM_INCREMENT); 1469128452Ssilby new_isn += isn_offset; 1470157977Srwatson ISN_UNLOCK(); 1471139222Srwatson return (new_isn); 147275619Skris} 147375619Skris 14741541Srgrimes/* 1475128452Ssilby * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary 1476128452Ssilby * to keep time flowing at a relatively constant rate. If the random 1477128452Ssilby * increments have already pushed us past the projected offset, do nothing. 1478128452Ssilby */ 1479128452Ssilbystatic void 1480157431Srwatsontcp_isn_tick(void *xtp) 1481128452Ssilby{ 1482128452Ssilby u_int32_t projected_offset; 1483133874Srwatson 1484157977Srwatson ISN_LOCK(); 1485141078Srwatson projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100; 1486128452Ssilby 1487128452Ssilby if (projected_offset > isn_offset) 1488128452Ssilby isn_offset = projected_offset; 1489128452Ssilby 1490128452Ssilby isn_offset_old = isn_offset; 1491141072Srwatson callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL); 1492157977Srwatson ISN_UNLOCK(); 1493128452Ssilby} 1494128452Ssilby 1495128452Ssilby/* 149672959Sjlemon * When a specific ICMP unreachable message is received and the 149772959Sjlemon * connection state is SYN-SENT, drop the connection. This behavior 149872959Sjlemon * is controlled by the icmp_may_rst sysctl. 149970103Sphk */ 150098211Shsustruct inpcb * 1501157431Srwatsontcp_drop_syn_sent(struct inpcb *inp, int errno) 150270103Sphk{ 1503157432Srwatson struct tcpcb *tp; 150470103Sphk 1505146864Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); 1506138019Srwatson INP_LOCK_ASSERT(inp); 1507157433Srwatson 1508157433Srwatson if ((inp->inp_vflag & INP_TIMEWAIT) || 1509157433Srwatson (inp->inp_vflag & INP_DROPPED)) 1510157433Srwatson return (inp); 1511157433Srwatson 1512157432Srwatson tp = intotcpcb(inp); 1513157433Srwatson if (tp->t_state != TCPS_SYN_SENT) 1514157433Srwatson return (inp); 1515146864Srwatson 1516157433Srwatson tp = tcp_drop(tp, errno); 1517157433Srwatson if (tp != NULL) 1518157433Srwatson return (inp); 1519157433Srwatson else 1520157433Srwatson return (NULL); 152172638Sphk} 152272638Sphk 152372638Sphk/* 152410881Swollman * When `need fragmentation' ICMP is received, update our idea of the MSS 152510881Swollman * based on the new value in the route. Also nudge TCP to send something, 152610881Swollman * since we know the packet we just sent was dropped. 152710930Swollman * This duplicates some code in the tcp_mss() function in tcp_input.c. 152810881Swollman */ 152998211Shsustruct inpcb * 1530157431Srwatsontcp_mtudisc(struct inpcb *inp, int errno) 153110881Swollman{ 1532157432Srwatson struct tcpcb *tp; 153310930Swollman struct socket *so = inp->inp_socket; 1534122922Sandre u_int maxmtu; 1535122922Sandre u_int romtu; 153610930Swollman int mss; 153755679Sshin#ifdef INET6 1538134041Sandre int isipv6; 153955679Sshin#endif /* INET6 */ 154010881Swollman 1541138019Srwatson INP_LOCK_ASSERT(inp); 1542157433Srwatson if ((inp->inp_vflag & INP_TIMEWAIT) || 1543157433Srwatson (inp->inp_vflag & INP_DROPPED)) 1544157433Srwatson return (inp); 1545157433Srwatson 1546157432Srwatson tp = intotcpcb(inp); 1547157433Srwatson KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1548157433Srwatson 1549134041Sandre#ifdef INET6 1550157433Srwatson isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0; 1551134041Sandre#endif 1552157433Srwatson maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */ 1553157433Srwatson romtu = 155455679Sshin#ifdef INET6 1555162084Sandre isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) : 1556122922Sandre#endif /* INET6 */ 1557162084Sandre tcp_maxmtu(&inp->inp_inc, NULL); 1558157433Srwatson if (!maxmtu) 1559157433Srwatson maxmtu = romtu; 1560157433Srwatson else 1561157433Srwatson maxmtu = min(maxmtu, romtu); 1562157433Srwatson if (!maxmtu) { 1563157433Srwatson tp->t_maxopd = tp->t_maxseg = 156455679Sshin#ifdef INET6 1565157433Srwatson isipv6 ? tcp_v6mssdflt : 156655679Sshin#endif /* INET6 */ 1567157433Srwatson tcp_mssdflt; 1568157433Srwatson return (inp); 1569157433Srwatson } 1570157433Srwatson mss = maxmtu - 157155679Sshin#ifdef INET6 1572157433Srwatson (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 157355679Sshin#endif /* INET6 */ 1574157433Srwatson sizeof(struct tcpiphdr) 157555679Sshin#ifdef INET6 1576157433Srwatson ) 157755679Sshin#endif /* INET6 */ 1578157433Srwatson ; 157955679Sshin 1580157433Srwatson /* 1581157433Srwatson * XXX - The above conditional probably violates the TCP 1582157433Srwatson * spec. The problem is that, since we don't know the 1583157433Srwatson * other end's MSS, we are supposed to use a conservative 1584157433Srwatson * default. But, if we do that, then MTU discovery will 1585157433Srwatson * never actually take place, because the conservative 1586157433Srwatson * default is much less than the MTUs typically seen 1587157433Srwatson * on the Internet today. For the moment, we'll sweep 1588157433Srwatson * this under the carpet. 1589157433Srwatson * 1590157433Srwatson * The conservative default might not actually be a problem 1591157433Srwatson * if the only case this occurs is when sending an initial 1592157433Srwatson * SYN with options and data to a host we've never talked 1593157433Srwatson * to before. Then, they will reply with an MSS value which 1594157433Srwatson * will get recorded and the new parameters should get 1595157433Srwatson * recomputed. For Further Study. 1596157433Srwatson */ 1597157433Srwatson if (tp->t_maxopd <= mss) 1598157433Srwatson return (inp); 1599157433Srwatson tp->t_maxopd = mss; 160010930Swollman 1601157433Srwatson if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 1602157433Srwatson (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 1603157433Srwatson mss -= TCPOLEN_TSTAMP_APPA; 160410930Swollman#if (MCLBYTES & (MCLBYTES - 1)) == 0 1605157433Srwatson if (mss > MCLBYTES) 1606157433Srwatson mss &= ~(MCLBYTES-1); 160710930Swollman#else 1608157433Srwatson if (mss > MCLBYTES) 1609157433Srwatson mss = mss / MCLBYTES * MCLBYTES; 161010881Swollman#endif 1611157433Srwatson if (so->so_snd.sb_hiwat < mss) 1612157433Srwatson mss = so->so_snd.sb_hiwat; 161310930Swollman 1614157433Srwatson tp->t_maxseg = mss; 161510930Swollman 1616157433Srwatson tcpstat.tcps_mturesent++; 1617157433Srwatson tp->t_rtttime = 0; 1618157433Srwatson tp->snd_nxt = tp->snd_una; 1619161645Smohans tcp_free_sackholes(tp); 1620161645Smohans tp->snd_recover = tp->snd_max; 1621161645Smohans if (tp->sack_enable) 1622161645Smohans EXIT_FASTRECOVERY(tp); 1623157433Srwatson tcp_output(tp); 1624139222Srwatson return (inp); 162510881Swollman} 162610881Swollman 162710881Swollman/* 16286283Swollman * Look-up the routing entry to the peer of this inpcb. If no route 1629108265Shsu * is found and it cannot be allocated, then return NULL. This routine 16306283Swollman * is called by TCP routines that access the rmx structure and by tcp_mss 16316283Swollman * to get the interface MTU. 16326283Swollman */ 1633133874Srwatsonu_long 1634162084Sandretcp_maxmtu(struct in_conninfo *inc, int *flags) 16356283Swollman{ 1636122922Sandre struct route sro; 1637122922Sandre struct sockaddr_in *dst; 1638122922Sandre struct ifnet *ifp; 1639122922Sandre u_long maxmtu = 0; 16406283Swollman 1641122922Sandre KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1642122922Sandre 1643122996Sandre bzero(&sro, sizeof(sro)); 1644122922Sandre if (inc->inc_faddr.s_addr != INADDR_ANY) { 1645122922Sandre dst = (struct sockaddr_in *)&sro.ro_dst; 1646122922Sandre dst->sin_family = AF_INET; 1647122922Sandre dst->sin_len = sizeof(*dst); 1648122922Sandre dst->sin_addr = inc->inc_faddr; 1649122922Sandre rtalloc_ign(&sro, RTF_CLONING); 16506283Swollman } 1651122922Sandre if (sro.ro_rt != NULL) { 1652122922Sandre ifp = sro.ro_rt->rt_ifp; 1653122922Sandre if (sro.ro_rt->rt_rmx.rmx_mtu == 0) 1654122922Sandre maxmtu = ifp->if_mtu; 1655122922Sandre else 1656122922Sandre maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 1657162084Sandre 1658162084Sandre /* Report additional interface capabilities. */ 1659162084Sandre if (flags != NULL) { 1660162084Sandre if (ifp->if_capenable & IFCAP_TSO4 && 1661162084Sandre ifp->if_hwassist & CSUM_TSO) 1662162084Sandre *flags |= CSUM_TSO; 1663162084Sandre } 1664122922Sandre RTFREE(sro.ro_rt); 1665122922Sandre } 1666122922Sandre return (maxmtu); 16676283Swollman} 16686283Swollman 166955679Sshin#ifdef INET6 1670122922Sandreu_long 1671162084Sandretcp_maxmtu6(struct in_conninfo *inc, int *flags) 167255679Sshin{ 1673122922Sandre struct route_in6 sro6; 1674122922Sandre struct ifnet *ifp; 1675122922Sandre u_long maxmtu = 0; 167655679Sshin 1677122922Sandre KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1678122922Sandre 1679122996Sandre bzero(&sro6, sizeof(sro6)); 1680122922Sandre if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1681122922Sandre sro6.ro_dst.sin6_family = AF_INET6; 1682122922Sandre sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1683122922Sandre sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1684122922Sandre rtalloc_ign((struct route *)&sro6, RTF_CLONING); 168555679Sshin } 1686122922Sandre if (sro6.ro_rt != NULL) { 1687122922Sandre ifp = sro6.ro_rt->rt_ifp; 1688122922Sandre if (sro6.ro_rt->rt_rmx.rmx_mtu == 0) 1689122922Sandre maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1690122922Sandre else 1691122922Sandre maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu, 1692122922Sandre IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1693162084Sandre 1694162084Sandre /* Report additional interface capabilities. */ 1695162084Sandre if (flags != NULL) { 1696162084Sandre if (ifp->if_capenable & IFCAP_TSO6 && 1697162084Sandre ifp->if_hwassist & CSUM_TSO) 1698162084Sandre *flags |= CSUM_TSO; 1699162084Sandre } 1700122922Sandre RTFREE(sro6.ro_rt); 1701122922Sandre } 1702122922Sandre 1703122922Sandre return (maxmtu); 170455679Sshin} 170555679Sshin#endif /* INET6 */ 170655679Sshin 170755679Sshin#ifdef IPSEC 170855679Sshin/* compute ESP/AH header size for TCP, including outer IP header. */ 170955679Sshinsize_t 1710157431Srwatsonipsec_hdrsiz_tcp(struct tcpcb *tp) 171155679Sshin{ 171255679Sshin struct inpcb *inp; 171355679Sshin struct mbuf *m; 171455679Sshin size_t hdrsiz; 171555679Sshin struct ip *ip; 171655679Sshin#ifdef INET6 171755679Sshin struct ip6_hdr *ip6; 1718111145Sjlemon#endif 171955679Sshin struct tcphdr *th; 172055679Sshin 172178642Ssilby if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) 1722139222Srwatson return (0); 1723111119Simp MGETHDR(m, M_DONTWAIT, MT_DATA); 172455679Sshin if (!m) 1725139222Srwatson return (0); 172655679Sshin 172755679Sshin#ifdef INET6 172855679Sshin if ((inp->inp_vflag & INP_IPV6) != 0) { 172955679Sshin ip6 = mtod(m, struct ip6_hdr *); 173055679Sshin th = (struct tcphdr *)(ip6 + 1); 173155679Sshin m->m_pkthdr.len = m->m_len = 173255679Sshin sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1733111144Sjlemon tcpip_fillheaders(inp, ip6, th); 173455679Sshin hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 173555679Sshin } else 173655679Sshin#endif /* INET6 */ 1737133874Srwatson { 1738133874Srwatson ip = mtod(m, struct ip *); 1739133874Srwatson th = (struct tcphdr *)(ip + 1); 1740133874Srwatson m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1741133874Srwatson tcpip_fillheaders(inp, ip, th); 1742133874Srwatson hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1743133874Srwatson } 174455679Sshin 174555679Sshin m_free(m); 1746139222Srwatson return (hdrsiz); 174755679Sshin} 174855679Sshin#endif /*IPSEC*/ 174955679Sshin 17506283Swollman/* 1751111145Sjlemon * Move a TCP connection into TIME_WAIT state. 1752138410Srwatson * tcbinfo is locked. 1753111145Sjlemon * inp is locked, and is unlocked before returning. 1754111145Sjlemon */ 1755111145Sjlemonvoid 1756157431Srwatsontcp_twstart(struct tcpcb *tp) 1757111145Sjlemon{ 1758111145Sjlemon struct tcptw *tw; 1759162151Sglebius struct inpcb *inp = tp->t_inpcb; 1760162111Sru int acknow; 1761111145Sjlemon struct socket *so; 1762111145Sjlemon 1763138025Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */ 1764162151Sglebius INP_LOCK_ASSERT(inp); 1765138020Srwatson 1766162151Sglebius if (nolocaltimewait && in_localip(inp->inp_faddr)) { 1767162151Sglebius tp = tcp_close(tp); 1768162151Sglebius if (tp != NULL) 1769162151Sglebius INP_UNLOCK(inp); 1770162151Sglebius return; 1771162151Sglebius } 1772162151Sglebius 1773112009Sjlemon tw = uma_zalloc(tcptw_zone, M_NOWAIT); 1774112009Sjlemon if (tw == NULL) { 1775112009Sjlemon tw = tcp_timer_2msl_tw(1); 1776112009Sjlemon if (tw == NULL) { 1777157376Srwatson tp = tcp_close(tp); 1778157376Srwatson if (tp != NULL) 1779162151Sglebius INP_UNLOCK(inp); 1780112009Sjlemon return; 1781112009Sjlemon } 1782112009Sjlemon } 1783111145Sjlemon tw->tw_inpcb = inp; 1784111145Sjlemon 1785111145Sjlemon /* 1786111145Sjlemon * Recover last window size sent. 1787111145Sjlemon */ 1788111145Sjlemon tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 1789111145Sjlemon 1790111145Sjlemon /* 1791111145Sjlemon * Set t_recent if timestamps are used on the connection. 1792111145Sjlemon */ 1793133874Srwatson if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 1794133874Srwatson (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 1795111145Sjlemon tw->t_recent = tp->ts_recent; 1796111145Sjlemon else 1797111145Sjlemon tw->t_recent = 0; 1798111145Sjlemon 1799111145Sjlemon tw->snd_nxt = tp->snd_nxt; 1800111145Sjlemon tw->rcv_nxt = tp->rcv_nxt; 1801121850Ssilby tw->iss = tp->iss; 1802121884Ssilby tw->irs = tp->irs; 1803111145Sjlemon tw->t_starttime = tp->t_starttime; 1804112009Sjlemon tw->tw_time = 0; 1805111145Sjlemon 1806111145Sjlemon/* XXX 1807111145Sjlemon * If this code will 1808111145Sjlemon * be used for fin-wait-2 state also, then we may need 1809111145Sjlemon * a ts_recent from the last segment. 1810111145Sjlemon */ 1811137139Sandre acknow = tp->t_flags & TF_ACKNOW; 1812157376Srwatson 1813157376Srwatson /* 1814157376Srwatson * First, discard tcpcb state, which includes stopping its timers and 1815157376Srwatson * freeing it. tcp_discardcb() used to also release the inpcb, but 1816157376Srwatson * that work is now done in the caller. 1817160925Srwatson * 1818160925Srwatson * Note: soisdisconnected() call used to be made in tcp_discardcb(), 1819160925Srwatson * and might not be needed here any longer. 1820157376Srwatson */ 1821111145Sjlemon tcp_discardcb(tp); 1822111145Sjlemon so = inp->inp_socket; 1823160925Srwatson soisdisconnected(so); 1824130387Srwatson SOCK_LOCK(so); 1825111145Sjlemon tw->tw_cred = crhold(so->so_cred); 1826111145Sjlemon tw->tw_so_options = so->so_options; 1827157376Srwatson SOCK_UNLOCK(so); 1828114794Srwatson if (acknow) 1829126351Srwatson tcp_twrespond(tw, TH_ACK); 1830157432Srwatson inp->inp_ppcb = tw; 1831111145Sjlemon inp->inp_vflag |= INP_TIMEWAIT; 1832162111Sru tcp_timer_2msl_reset(tw, 0); 1833157376Srwatson 1834157376Srwatson /* 1835157376Srwatson * If the inpcb owns the sole reference to the socket, then we can 1836157376Srwatson * detach and free the socket as it is not needed in time wait. 1837157376Srwatson */ 1838157376Srwatson if (inp->inp_vflag & INP_SOCKREF) { 1839157376Srwatson KASSERT(so->so_state & SS_PROTOREF, 1840157376Srwatson ("tcp_twstart: !SS_PROTOREF")); 1841157376Srwatson inp->inp_vflag &= ~INP_SOCKREF; 1842157376Srwatson INP_UNLOCK(inp); 1843157376Srwatson ACCEPT_LOCK(); 1844157376Srwatson SOCK_LOCK(so); 1845157376Srwatson so->so_state &= ~SS_PROTOREF; 1846157376Srwatson sofree(so); 1847157376Srwatson } else 1848157376Srwatson INP_UNLOCK(inp); 1849111145Sjlemon} 1850111145Sjlemon 1851162064Sglebius#if 0 1852121850Ssilby/* 1853121884Ssilby * The appromixate rate of ISN increase of Microsoft TCP stacks; 1854121884Ssilby * the actual rate is slightly higher due to the addition of 1855121884Ssilby * random positive increments. 1856121884Ssilby * 1857121884Ssilby * Most other new OSes use semi-randomized ISN values, so we 1858121884Ssilby * do not need to worry about them. 1859121884Ssilby */ 1860121884Ssilby#define MS_ISN_BYTES_PER_SECOND 250000 1861121884Ssilby 1862121884Ssilby/* 1863121850Ssilby * Determine if the ISN we will generate has advanced beyond the last 1864121850Ssilby * sequence number used by the previous connection. If so, indicate 1865121850Ssilby * that it is safe to recycle this tw socket by returning 1. 1866121850Ssilby */ 1867121850Ssilbyint 1868121850Ssilbytcp_twrecycleable(struct tcptw *tw) 1869121850Ssilby{ 1870121884Ssilby tcp_seq new_iss = tw->iss; 1871121884Ssilby tcp_seq new_irs = tw->irs; 1872121850Ssilby 1873162064Sglebius INP_INFO_WLOCK_ASSERT(&tcbinfo); 1874121884Ssilby new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 1875121884Ssilby new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 1876133874Srwatson 1877121884Ssilby if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 1878139222Srwatson return (1); 1879121850Ssilby else 1880139222Srwatson return (0); 1881121850Ssilby} 1882162064Sglebius#endif 1883121850Ssilby 1884157376Srwatsonvoid 1885112009Sjlemontcp_twclose(struct tcptw *tw, int reuse) 1886111145Sjlemon{ 1887157376Srwatson struct socket *so; 1888111145Sjlemon struct inpcb *inp; 1889111145Sjlemon 1890157376Srwatson /* 1891157386Srwatson * At this point, we are in one of two situations: 1892157376Srwatson * 1893160549Srwatson * (1) We have no socket, just an inpcb<->twtcp pair. We can free 1894160549Srwatson * all state. 1895157386Srwatson * 1896160549Srwatson * (2) We have a socket -- if we own a reference, release it and 1897160549Srwatson * notify the socket layer. 1898157376Srwatson */ 1899111145Sjlemon inp = tw->tw_inpcb; 1900157376Srwatson KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 1901157432Srwatson KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 1902138025Srwatson INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */ 1903138020Srwatson INP_LOCK_ASSERT(inp); 1904138020Srwatson 1905111145Sjlemon tw->tw_inpcb = NULL; 1906112009Sjlemon tcp_timer_2msl_stop(tw); 1907111145Sjlemon inp->inp_ppcb = NULL; 1908158009Srwatson in_pcbdrop(inp); 1909157376Srwatson 1910157376Srwatson so = inp->inp_socket; 1911157386Srwatson if (so != NULL) { 1912160549Srwatson /* 1913160549Srwatson * If there's a socket, handle two cases: first, we own a 1914160549Srwatson * strong reference, which we will now release, or we don't 1915160549Srwatson * in which case another reference exists (XXXRW: think 1916160549Srwatson * about this more), and we don't need to take action. 1917160549Srwatson */ 1918157386Srwatson if (inp->inp_vflag & INP_SOCKREF) { 1919157386Srwatson inp->inp_vflag &= ~INP_SOCKREF; 1920160549Srwatson INP_UNLOCK(inp); 1921157386Srwatson ACCEPT_LOCK(); 1922157386Srwatson SOCK_LOCK(so); 1923157386Srwatson KASSERT(so->so_state & SS_PROTOREF, 1924157386Srwatson ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 1925157386Srwatson so->so_state &= ~SS_PROTOREF; 1926157386Srwatson sofree(so); 1927157376Srwatson } else { 1928157386Srwatson /* 1929157386Srwatson * If we don't own the only reference, the socket and 1930157386Srwatson * inpcb need to be left around to be handled by 1931157386Srwatson * tcp_usr_detach() later. 1932157386Srwatson */ 1933157386Srwatson INP_UNLOCK(inp); 1934157376Srwatson } 1935157386Srwatson } else { 1936157376Srwatson#ifdef INET6 1937157376Srwatson if (inp->inp_vflag & INP_IPV6PROTO) 1938157376Srwatson in6_pcbfree(inp); 1939157376Srwatson else 1940157376Srwatson#endif 1941157376Srwatson in_pcbfree(inp); 1942157386Srwatson } 1943111145Sjlemon tcpstat.tcps_closed++; 1944126002Spjd crfree(tw->tw_cred); 1945126002Spjd tw->tw_cred = NULL; 1946112009Sjlemon if (reuse) 1947157376Srwatson return; 1948112009Sjlemon uma_zfree(tcptw_zone, tw); 1949111145Sjlemon} 1950111145Sjlemon 1951111145Sjlemonint 1952126351Srwatsontcp_twrespond(struct tcptw *tw, int flags) 1953111145Sjlemon{ 1954111145Sjlemon struct inpcb *inp = tw->tw_inpcb; 1955111145Sjlemon struct tcphdr *th; 1956111145Sjlemon struct mbuf *m; 1957111145Sjlemon struct ip *ip = NULL; 1958111145Sjlemon u_int8_t *optp; 1959111145Sjlemon u_int hdrlen, optlen; 1960111145Sjlemon int error; 1961111145Sjlemon#ifdef INET6 1962111145Sjlemon struct ip6_hdr *ip6 = NULL; 1963111145Sjlemon int isipv6 = inp->inp_inc.inc_isipv6; 1964111145Sjlemon#endif 1965111145Sjlemon 1966138020Srwatson INP_LOCK_ASSERT(inp); 1967138020Srwatson 1968151967Sandre m = m_gethdr(M_DONTWAIT, MT_DATA); 1969111145Sjlemon if (m == NULL) 1970111145Sjlemon return (ENOBUFS); 1971111145Sjlemon m->m_data += max_linkhdr; 1972111145Sjlemon 1973114794Srwatson#ifdef MAC 1974123607Srwatson mac_create_mbuf_from_inpcb(inp, m); 1975114794Srwatson#endif 1976114794Srwatson 1977111153Sjlemon#ifdef INET6 1978111145Sjlemon if (isipv6) { 1979111145Sjlemon hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1980111145Sjlemon ip6 = mtod(m, struct ip6_hdr *); 1981111145Sjlemon th = (struct tcphdr *)(ip6 + 1); 1982111145Sjlemon tcpip_fillheaders(inp, ip6, th); 1983111153Sjlemon } else 1984111153Sjlemon#endif 1985111153Sjlemon { 1986111145Sjlemon hdrlen = sizeof(struct tcpiphdr); 1987111145Sjlemon ip = mtod(m, struct ip *); 1988111145Sjlemon th = (struct tcphdr *)(ip + 1); 1989111145Sjlemon tcpip_fillheaders(inp, ip, th); 1990111145Sjlemon } 1991111145Sjlemon optp = (u_int8_t *)(th + 1); 1992133874Srwatson 1993133874Srwatson /* 1994111145Sjlemon * Send a timestamp and echo-reply if both our side and our peer 1995111145Sjlemon * have sent timestamps in our SYN's and this is not a RST. 1996133874Srwatson */ 1997111145Sjlemon if (tw->t_recent && flags == TH_ACK) { 1998111145Sjlemon u_int32_t *lp = (u_int32_t *)optp; 1999111145Sjlemon 2000133874Srwatson /* Form timestamp option as shown in appendix A of RFC 1323. */ 2001133874Srwatson *lp++ = htonl(TCPOPT_TSTAMP_HDR); 2002133874Srwatson *lp++ = htonl(ticks); 2003133874Srwatson *lp = htonl(tw->t_recent); 2004133874Srwatson optp += TCPOLEN_TSTAMP_APPA; 2005133874Srwatson } 2006111145Sjlemon 2007111145Sjlemon optlen = optp - (u_int8_t *)(th + 1); 2008111145Sjlemon 2009111145Sjlemon m->m_len = hdrlen + optlen; 2010111145Sjlemon m->m_pkthdr.len = m->m_len; 2011111145Sjlemon 2012111145Sjlemon KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 2013111145Sjlemon 2014111145Sjlemon th->th_seq = htonl(tw->snd_nxt); 2015111145Sjlemon th->th_ack = htonl(tw->rcv_nxt); 2016111145Sjlemon th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 2017111145Sjlemon th->th_flags = flags; 2018111145Sjlemon th->th_win = htons(tw->last_win); 2019111145Sjlemon 2020111153Sjlemon#ifdef INET6 2021111145Sjlemon if (isipv6) { 2022111145Sjlemon th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 2023111145Sjlemon sizeof(struct tcphdr) + optlen); 2024122922Sandre ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2025122922Sandre error = ip6_output(m, inp->in6p_outputopts, NULL, 2026111145Sjlemon (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 2027111153Sjlemon } else 2028111153Sjlemon#endif 2029111153Sjlemon { 2030111145Sjlemon th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2031133874Srwatson htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 2032111145Sjlemon m->m_pkthdr.csum_flags = CSUM_TCP; 2033111145Sjlemon m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2034111145Sjlemon ip->ip_len = m->m_pkthdr.len; 2035124248Sandre if (path_mtu_discovery) 2036124248Sandre ip->ip_off |= IP_DF; 2037122922Sandre error = ip_output(m, inp->inp_options, NULL, 2038134793Sjmg ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 2039134793Sjmg NULL, inp); 2040111145Sjlemon } 2041111145Sjlemon if (flags & TH_ACK) 2042111145Sjlemon tcpstat.tcps_sndacks++; 2043111145Sjlemon else 2044111145Sjlemon tcpstat.tcps_sndctrl++; 2045111145Sjlemon tcpstat.tcps_sndtotal++; 2046111145Sjlemon return (error); 2047111145Sjlemon} 2048111145Sjlemon 2049111145Sjlemon/* 2050102017Sdillon * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING 2051102017Sdillon * 2052102017Sdillon * This code attempts to calculate the bandwidth-delay product as a 2053102017Sdillon * means of determining the optimal window size to maximize bandwidth, 2054102017Sdillon * minimize RTT, and avoid the over-allocation of buffers on interfaces and 2055102017Sdillon * routers. This code also does a fairly good job keeping RTTs in check 2056102017Sdillon * across slow links like modems. We implement an algorithm which is very 2057102017Sdillon * similar (but not meant to be) TCP/Vegas. The code operates on the 2058102017Sdillon * transmitter side of a TCP connection and so only effects the transmit 2059102017Sdillon * side of the connection. 2060102017Sdillon * 2061102017Sdillon * BACKGROUND: TCP makes no provision for the management of buffer space 2062133874Srwatson * at the end points or at the intermediate routers and switches. A TCP 2063102017Sdillon * stream, whether using NewReno or not, will eventually buffer as 2064102017Sdillon * many packets as it is able and the only reason this typically works is 2065102017Sdillon * due to the fairly small default buffers made available for a connection 2066102017Sdillon * (typicaly 16K or 32K). As machines use larger windows and/or window 2067102017Sdillon * scaling it is now fairly easy for even a single TCP connection to blow-out 2068133874Srwatson * all available buffer space not only on the local interface, but on 2069102017Sdillon * intermediate routers and switches as well. NewReno makes a misguided 2070102017Sdillon * attempt to 'solve' this problem by waiting for an actual failure to occur, 2071102017Sdillon * then backing off, then steadily increasing the window again until another 2072102017Sdillon * failure occurs, ad-infinitum. This results in terrible oscillation that 2073102017Sdillon * is only made worse as network loads increase and the idea of intentionally 2074102017Sdillon * blowing out network buffers is, frankly, a terrible way to manage network 2075102017Sdillon * resources. 2076102017Sdillon * 2077102017Sdillon * It is far better to limit the transmit window prior to the failure 2078102017Sdillon * condition being achieved. There are two general ways to do this: First 2079102017Sdillon * you can 'scan' through different transmit window sizes and locate the 2080102017Sdillon * point where the RTT stops increasing, indicating that you have filled the 2081102017Sdillon * pipe, then scan backwards until you note that RTT stops decreasing, then 2082102017Sdillon * repeat ad-infinitum. This method works in principle but has severe 2083102017Sdillon * implementation issues due to RTT variances, timer granularity, and 2084102017Sdillon * instability in the algorithm which can lead to many false positives and 2085102017Sdillon * create oscillations as well as interact badly with other TCP streams 2086102017Sdillon * implementing the same algorithm. 2087102017Sdillon * 2088102017Sdillon * The second method is to limit the window to the bandwidth delay product 2089102017Sdillon * of the link. This is the method we implement. RTT variances and our 2090133874Srwatson * own manipulation of the congestion window, bwnd, can potentially 2091102017Sdillon * destabilize the algorithm. For this reason we have to stabilize the 2092102017Sdillon * elements used to calculate the window. We do this by using the minimum 2093102017Sdillon * observed RTT, the long term average of the observed bandwidth, and 2094102017Sdillon * by adding two segments worth of slop. It isn't perfect but it is able 2095102017Sdillon * to react to changing conditions and gives us a very stable basis on 2096102017Sdillon * which to extend the algorithm. 2097102017Sdillon */ 2098102017Sdillonvoid 2099102017Sdillontcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq) 2100102017Sdillon{ 2101102017Sdillon u_long bw; 2102102017Sdillon u_long bwnd; 2103102017Sdillon int save_ticks; 2104102017Sdillon 2105138410Srwatson INP_LOCK_ASSERT(tp->t_inpcb); 2106138410Srwatson 2107102017Sdillon /* 2108102017Sdillon * If inflight_enable is disabled in the middle of a tcp connection, 2109102017Sdillon * make sure snd_bwnd is effectively disabled. 2110102017Sdillon */ 2111155767Sandre if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) { 2112102017Sdillon tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2113102017Sdillon tp->snd_bandwidth = 0; 2114102017Sdillon return; 2115102017Sdillon } 2116102017Sdillon 2117102017Sdillon /* 2118102017Sdillon * Figure out the bandwidth. Due to the tick granularity this 2119102017Sdillon * is a very rough number and it MUST be averaged over a fairly 2120102017Sdillon * long period of time. XXX we need to take into account a link 2121102017Sdillon * that is not using all available bandwidth, but for now our 2122102017Sdillon * slop will ramp us up if this case occurs and the bandwidth later 2123102017Sdillon * increases. 2124102368Sdillon * 2125102368Sdillon * Note: if ticks rollover 'bw' may wind up negative. We must 2126102368Sdillon * effectively reset t_bw_rtttime for this case. 2127102017Sdillon */ 2128102017Sdillon save_ticks = ticks; 2129102017Sdillon if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1) 2130102017Sdillon return; 2131102017Sdillon 2132133874Srwatson bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / 2133102017Sdillon (save_ticks - tp->t_bw_rtttime); 2134102017Sdillon tp->t_bw_rtttime = save_ticks; 2135102017Sdillon tp->t_bw_rtseq = ack_seq; 2136102368Sdillon if (tp->t_bw_rtttime == 0 || (int)bw < 0) 2137102017Sdillon return; 2138102017Sdillon bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4; 2139102017Sdillon 2140102017Sdillon tp->snd_bandwidth = bw; 2141102017Sdillon 2142102017Sdillon /* 2143102017Sdillon * Calculate the semi-static bandwidth delay product, plus two maximal 2144102017Sdillon * segments. The additional slop puts us squarely in the sweet 2145107881Sdillon * spot and also handles the bandwidth run-up case and stabilization. 2146107881Sdillon * Without the slop we could be locking ourselves into a lower 2147107881Sdillon * bandwidth. 2148102017Sdillon * 2149102017Sdillon * Situations Handled: 2150102017Sdillon * (1) Prevents over-queueing of packets on LANs, especially on 2151102017Sdillon * high speed LANs, allowing larger TCP buffers to be 2152133874Srwatson * specified, and also does a good job preventing 2153102017Sdillon * over-queueing of packets over choke points like modems 2154102017Sdillon * (at least for the transmit side). 2155102017Sdillon * 2156102017Sdillon * (2) Is able to handle changing network loads (bandwidth 2157102017Sdillon * drops so bwnd drops, bandwidth increases so bwnd 2158102017Sdillon * increases). 2159102017Sdillon * 2160102017Sdillon * (3) Theoretically should stabilize in the face of multiple 2161102017Sdillon * connections implementing the same algorithm (this may need 2162102017Sdillon * a little work). 2163107881Sdillon * 2164107881Sdillon * (4) Stability value (defaults to 20 = 2 maximal packets) can 2165107881Sdillon * be adjusted with a sysctl but typically only needs to be 2166107881Sdillon * on very slow connections. A value no smaller then 5 2167107881Sdillon * should be used, but only reduce this default if you have 2168107881Sdillon * no other choice. 2169102017Sdillon */ 2170102017Sdillon#define USERTT ((tp->t_srtt + tp->t_rttbest) / 2) 2171107881Sdillon bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10; 2172102368Sdillon#undef USERTT 2173102017Sdillon 2174102017Sdillon if (tcp_inflight_debug > 0) { 2175102017Sdillon static int ltime; 2176102017Sdillon if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) { 2177102017Sdillon ltime = ticks; 2178102017Sdillon printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n", 2179102017Sdillon tp, 2180102017Sdillon bw, 2181102017Sdillon tp->t_rttbest, 2182102017Sdillon tp->t_srtt, 2183102017Sdillon bwnd 2184102017Sdillon ); 2185102017Sdillon } 2186102017Sdillon } 2187102017Sdillon if ((long)bwnd < tcp_inflight_min) 2188102017Sdillon bwnd = tcp_inflight_min; 2189102017Sdillon if (bwnd > tcp_inflight_max) 2190102017Sdillon bwnd = tcp_inflight_max; 2191102017Sdillon if ((long)bwnd < tp->t_maxseg * 2) 2192102017Sdillon bwnd = tp->t_maxseg * 2; 2193102017Sdillon tp->snd_bwnd = bwnd; 2194102017Sdillon} 2195102017Sdillon 2196125680Sbms#ifdef TCP_SIGNATURE 2197125680Sbms/* 2198125783Sbms * Callback function invoked by m_apply() to digest TCP segment data 2199125783Sbms * contained within an mbuf chain. 2200125783Sbms */ 2201125783Sbmsstatic int 2202125783Sbmstcp_signature_apply(void *fstate, void *data, u_int len) 2203125783Sbms{ 2204125783Sbms 2205125819Sbms MD5Update(fstate, (u_char *)data, len); 2206125783Sbms return (0); 2207125783Sbms} 2208125783Sbms 2209125783Sbms/* 2210125680Sbms * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385) 2211125680Sbms * 2212125741Sbms * Parameters: 2213125741Sbms * m pointer to head of mbuf chain 2214125741Sbms * off0 offset to TCP header within the mbuf chain 2215125741Sbms * len length of TCP segment data, excluding options 2216125741Sbms * optlen length of TCP segment options 2217125741Sbms * buf pointer to storage for computed MD5 digest 2218125741Sbms * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2219125741Sbms * 2220125680Sbms * We do this over ip, tcphdr, segment data, and the key in the SADB. 2221125680Sbms * When called from tcp_input(), we can be sure that th_sum has been 2222125680Sbms * zeroed out and verified already. 2223125680Sbms * 2224125680Sbms * This function is for IPv4 use only. Calling this function with an 2225125680Sbms * IPv6 packet in the mbuf chain will yield undefined results. 2226125680Sbms * 2227125680Sbms * Return 0 if successful, otherwise return -1. 2228125680Sbms * 2229125680Sbms * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2230125680Sbms * search with the destination IP address, and a 'magic SPI' to be 2231125680Sbms * determined by the application. This is hardcoded elsewhere to 1179 2232125680Sbms * right now. Another branch of this code exists which uses the SPD to 2233125680Sbms * specify per-application flows but it is unstable. 2234125680Sbms */ 2235125680Sbmsint 2236125783Sbmstcp_signature_compute(struct mbuf *m, int off0, int len, int optlen, 2237125741Sbms u_char *buf, u_int direction) 2238125680Sbms{ 2239125680Sbms union sockaddr_union dst; 2240125680Sbms struct ippseudo ippseudo; 2241125680Sbms MD5_CTX ctx; 2242125680Sbms int doff; 2243125680Sbms struct ip *ip; 2244125680Sbms struct ipovly *ipovly; 2245125680Sbms struct secasvar *sav; 2246125680Sbms struct tcphdr *th; 2247125680Sbms u_short savecsum; 2248125680Sbms 2249125741Sbms KASSERT(m != NULL, ("NULL mbuf chain")); 2250125741Sbms KASSERT(buf != NULL, ("NULL signature pointer")); 2251125741Sbms 2252125741Sbms /* Extract the destination from the IP header in the mbuf. */ 2253125680Sbms ip = mtod(m, struct ip *); 2254125680Sbms bzero(&dst, sizeof(union sockaddr_union)); 2255125680Sbms dst.sa.sa_len = sizeof(struct sockaddr_in); 2256125680Sbms dst.sa.sa_family = AF_INET; 2257125680Sbms dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2258125680Sbms ip->ip_src : ip->ip_dst; 2259125741Sbms 2260125741Sbms /* Look up an SADB entry which matches the address of the peer. */ 2261125680Sbms sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2262125680Sbms if (sav == NULL) { 2263125680Sbms printf("%s: SADB lookup failed for %s\n", __func__, 2264125680Sbms inet_ntoa(dst.sin.sin_addr)); 2265125680Sbms return (EINVAL); 2266125680Sbms } 2267125741Sbms 2268125680Sbms MD5Init(&ctx); 2269125680Sbms ipovly = (struct ipovly *)ip; 2270125680Sbms th = (struct tcphdr *)((u_char *)ip + off0); 2271125680Sbms doff = off0 + sizeof(struct tcphdr) + optlen; 2272125741Sbms 2273125680Sbms /* 2274125680Sbms * Step 1: Update MD5 hash with IP pseudo-header. 2275125680Sbms * 2276125680Sbms * XXX The ippseudo header MUST be digested in network byte order, 2277125680Sbms * or else we'll fail the regression test. Assume all fields we've 2278125680Sbms * been doing arithmetic on have been in host byte order. 2279125680Sbms * XXX One cannot depend on ipovly->ih_len here. When called from 2280125680Sbms * tcp_output(), the underlying ip_len member has not yet been set. 2281125680Sbms */ 2282125680Sbms ippseudo.ippseudo_src = ipovly->ih_src; 2283125680Sbms ippseudo.ippseudo_dst = ipovly->ih_dst; 2284125680Sbms ippseudo.ippseudo_pad = 0; 2285125680Sbms ippseudo.ippseudo_p = IPPROTO_TCP; 2286125680Sbms ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); 2287125680Sbms MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2288125741Sbms 2289125680Sbms /* 2290125680Sbms * Step 2: Update MD5 hash with TCP header, excluding options. 2291125680Sbms * The TCP checksum must be set to zero. 2292125680Sbms */ 2293125680Sbms savecsum = th->th_sum; 2294125680Sbms th->th_sum = 0; 2295125680Sbms MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2296125680Sbms th->th_sum = savecsum; 2297125741Sbms 2298125680Sbms /* 2299125680Sbms * Step 3: Update MD5 hash with TCP segment data. 2300125680Sbms * Use m_apply() to avoid an early m_pullup(). 2301125680Sbms */ 2302125680Sbms if (len > 0) 2303125783Sbms m_apply(m, doff, len, tcp_signature_apply, &ctx); 2304125741Sbms 2305125680Sbms /* 2306125680Sbms * Step 4: Update MD5 hash with shared secret. 2307125680Sbms */ 2308125680Sbms MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth)); 2309125680Sbms MD5Final(buf, &ctx); 2310125741Sbms 2311125680Sbms key_sa_recordxfer(sav, m); 2312125680Sbms KEY_FREESAV(&sav); 2313125680Sbms return (0); 2314125680Sbms} 2315125680Sbms#endif /* TCP_SIGNATURE */ 2316141886Smaxim 2317141886Smaximstatic int 2318141886Smaximsysctl_drop(SYSCTL_HANDLER_ARGS) 2319141886Smaxim{ 2320141886Smaxim /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2321141886Smaxim struct sockaddr_storage addrs[2]; 2322141886Smaxim struct inpcb *inp; 2323141886Smaxim struct tcpcb *tp; 2324150804Smaxim struct tcptw *tw; 2325141886Smaxim struct sockaddr_in *fin, *lin; 2326141886Smaxim#ifdef INET6 2327141886Smaxim struct sockaddr_in6 *fin6, *lin6; 2328141886Smaxim struct in6_addr f6, l6; 2329141886Smaxim#endif 2330141886Smaxim int error; 2331141886Smaxim 2332141886Smaxim inp = NULL; 2333141886Smaxim fin = lin = NULL; 2334141886Smaxim#ifdef INET6 2335141886Smaxim fin6 = lin6 = NULL; 2336141886Smaxim#endif 2337141886Smaxim error = 0; 2338141886Smaxim 2339141886Smaxim if (req->oldptr != NULL || req->oldlen != 0) 2340141886Smaxim return (EINVAL); 2341141886Smaxim if (req->newptr == NULL) 2342141886Smaxim return (EPERM); 2343141886Smaxim if (req->newlen < sizeof(addrs)) 2344141886Smaxim return (ENOMEM); 2345141886Smaxim error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2346141886Smaxim if (error) 2347141886Smaxim return (error); 2348141886Smaxim 2349141886Smaxim switch (addrs[0].ss_family) { 2350141886Smaxim#ifdef INET6 2351141886Smaxim case AF_INET6: 2352141886Smaxim fin6 = (struct sockaddr_in6 *)&addrs[0]; 2353141886Smaxim lin6 = (struct sockaddr_in6 *)&addrs[1]; 2354141886Smaxim if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2355141886Smaxim lin6->sin6_len != sizeof(struct sockaddr_in6)) 2356141886Smaxim return (EINVAL); 2357141886Smaxim if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2358141886Smaxim if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2359141886Smaxim return (EINVAL); 2360141886Smaxim in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2361141886Smaxim in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2362141886Smaxim fin = (struct sockaddr_in *)&addrs[0]; 2363141886Smaxim lin = (struct sockaddr_in *)&addrs[1]; 2364141886Smaxim break; 2365141886Smaxim } 2366148385Sume error = sa6_embedscope(fin6, ip6_use_defzone); 2367141886Smaxim if (error) 2368148385Sume return (error); 2369148385Sume error = sa6_embedscope(lin6, ip6_use_defzone); 2370148616Sume if (error) 2371148385Sume return (error); 2372141886Smaxim break; 2373141886Smaxim#endif 2374141886Smaxim case AF_INET: 2375141886Smaxim fin = (struct sockaddr_in *)&addrs[0]; 2376141886Smaxim lin = (struct sockaddr_in *)&addrs[1]; 2377141886Smaxim if (fin->sin_len != sizeof(struct sockaddr_in) || 2378141886Smaxim lin->sin_len != sizeof(struct sockaddr_in)) 2379141886Smaxim return (EINVAL); 2380141886Smaxim break; 2381141886Smaxim default: 2382141886Smaxim return (EINVAL); 2383141886Smaxim } 2384141886Smaxim INP_INFO_WLOCK(&tcbinfo); 2385141886Smaxim switch (addrs[0].ss_family) { 2386141886Smaxim#ifdef INET6 2387141886Smaxim case AF_INET6: 2388141886Smaxim inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port, 2389141886Smaxim &l6, lin6->sin6_port, 0, NULL); 2390141886Smaxim break; 2391141886Smaxim#endif 2392141886Smaxim case AF_INET: 2393141886Smaxim inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port, 2394141886Smaxim lin->sin_addr, lin->sin_port, 0, NULL); 2395141886Smaxim break; 2396141886Smaxim } 2397141886Smaxim if (inp != NULL) { 2398141886Smaxim INP_LOCK(inp); 2399157427Srwatson if (inp->inp_vflag & INP_TIMEWAIT) { 2400157474Srwatson /* 2401157474Srwatson * XXXRW: There currently exists a state where an 2402157474Srwatson * inpcb is present, but its timewait state has been 2403157474Srwatson * discarded. For now, don't allow dropping of this 2404157474Srwatson * type of inpcb. 2405157474Srwatson */ 2406157427Srwatson tw = intotw(inp); 2407157474Srwatson if (tw != NULL) 2408157474Srwatson tcp_twclose(tw, 0); 2409157427Srwatson } else if (!(inp->inp_vflag & INP_DROPPED) && 2410157427Srwatson !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2411157427Srwatson tp = intotcpcb(inp); 2412157427Srwatson tcp_drop(tp, ECONNABORTED); 2413157427Srwatson } 2414157427Srwatson INP_UNLOCK(inp); 2415141886Smaxim } else 2416141886Smaxim error = ESRCH; 2417141886Smaxim INP_INFO_WUNLOCK(&tcbinfo); 2418141886Smaxim return (error); 2419141886Smaxim} 2420141886Smaxim 2421141886SmaximSYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2422141886Smaxim CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2423141886Smaxim 0, sysctl_drop, "", "Drop TCP connection"); 2424