tcp_timewait.c revision 185370
1139823Simp/*- 211150Swollman * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 31541Srgrimes * The Regents of the University of California. All rights reserved. 41541Srgrimes * 51541Srgrimes * Redistribution and use in source and binary forms, with or without 61541Srgrimes * modification, are permitted provided that the following conditions 71541Srgrimes * are met: 81541Srgrimes * 1. Redistributions of source code must retain the above copyright 91541Srgrimes * notice, this list of conditions and the following disclaimer. 101541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 111541Srgrimes * notice, this list of conditions and the following disclaimer in the 121541Srgrimes * documentation and/or other materials provided with the distribution. 131541Srgrimes * 4. Neither the name of the University nor the names of its contributors 141541Srgrimes * may be used to endorse or promote products derived from this software 151541Srgrimes * without specific prior written permission. 161541Srgrimes * 171541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 181541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 191541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 201541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 211541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 221541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 231541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 241541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 251541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 261541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 271541Srgrimes * SUCH DAMAGE. 281541Srgrimes * 2911150Swollman * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 301541Srgrimes */ 311541Srgrimes 32172467Ssilby#include <sys/cdefs.h> 33172467Ssilby__FBSDID("$FreeBSD: head/sys/netinet/tcp_timewait.c 185370 2008-11-27 12:04:35Z bz $"); 34172467Ssilby 35125680Sbms#include "opt_inet.h" 3654263Sshin#include "opt_inet6.h" 37101106Srwatson#include "opt_mac.h" 3829514Sjoerg#include "opt_tcpdebug.h" 3929514Sjoerg 401541Srgrimes#include <sys/param.h> 411541Srgrimes#include <sys/systm.h> 4250673Sjlemon#include <sys/callout.h> 4312172Sphk#include <sys/kernel.h> 4412172Sphk#include <sys/sysctl.h> 451541Srgrimes#include <sys/malloc.h> 461541Srgrimes#include <sys/mbuf.h> 47164033Srwatson#include <sys/priv.h> 4848758Sgreen#include <sys/proc.h> 491541Srgrimes#include <sys/socket.h> 501541Srgrimes#include <sys/socketvar.h> 511541Srgrimes#include <sys/protosw.h> 5275619Skris#include <sys/random.h> 53181803Sbz#include <sys/vimage.h> 5434923Sbde 5592760Sjeff#include <vm/uma.h> 561541Srgrimes 571541Srgrimes#include <net/route.h> 581541Srgrimes#include <net/if.h> 591541Srgrimes 601541Srgrimes#include <netinet/in.h> 611541Srgrimes#include <netinet/in_systm.h> 621541Srgrimes#include <netinet/ip.h> 6355679Sshin#ifdef INET6 6455679Sshin#include <netinet/ip6.h> 6555679Sshin#endif 661541Srgrimes#include <netinet/in_pcb.h> 6755679Sshin#ifdef INET6 6855679Sshin#include <netinet6/in6_pcb.h> 6955679Sshin#endif 707090Sbde#include <netinet/in_var.h> 711541Srgrimes#include <netinet/ip_var.h> 7255679Sshin#ifdef INET6 7355679Sshin#include <netinet6/ip6_var.h> 74148385Sume#include <netinet6/scope6_var.h> 75122922Sandre#include <netinet6/nd6.h> 7655679Sshin#endif 77145360Sandre#include <netinet/ip_icmp.h> 781541Srgrimes#include <netinet/tcp.h> 791541Srgrimes#include <netinet/tcp_fsm.h> 801541Srgrimes#include <netinet/tcp_seq.h> 811541Srgrimes#include <netinet/tcp_timer.h> 821541Srgrimes#include <netinet/tcp_var.h> 8355679Sshin#ifdef INET6 8455679Sshin#include <netinet6/tcp6_var.h> 8555679Sshin#endif 861541Srgrimes#include <netinet/tcpip.h> 876283Swollman#ifdef TCPDEBUG 886283Swollman#include <netinet/tcp_debug.h> 896283Swollman#endif 9055679Sshin#include <netinet6/ip6protosw.h> 911541Srgrimes 9258698Sjlemon#include <machine/in_cksum.h> 9358698Sjlemon 94163606Srwatson#include <security/mac/mac_framework.h> 95163606Srwatson 96162035Sglebiusstatic uma_zone_t tcptw_zone; 97157478Sglebiusstatic int maxtcptw; 98162767Ssilby 99169608Sandre/* 100169608Sandre * The timed wait queue contains references to each of the TCP sessions 101169608Sandre * currently in the TIME_WAIT state. The queue pointers, including the 102169608Sandre * queue pointers in each tcptw structure, are protected using the global 103169608Sandre * tcbinfo lock, which must be held over queue iteration and modification. 104169608Sandre */ 105185088Szec#ifdef VIMAGE_GLOBALS 106169608Sandrestatic TAILQ_HEAD(, tcptw) twq_2msl; 107185088Szecint nolocaltimewait; 108185088Szec#endif 109169608Sandre 110169608Sandrestatic void tcp_tw_2msl_reset(struct tcptw *, int); 111169608Sandrestatic void tcp_tw_2msl_stop(struct tcptw *); 112169608Sandre 113162031Sglebiusstatic int 114162767Ssilbytcptw_auto_size(void) 115162767Ssilby{ 116183550Szec INIT_VNET_INET(curvnet); 117162767Ssilby int halfrange; 118162767Ssilby 119162767Ssilby /* 120162768Smaxim * Max out at half the ephemeral port range so that TIME_WAIT 121162768Smaxim * sockets don't tie up too many ephemeral ports. 122162768Smaxim */ 123181803Sbz if (V_ipport_lastauto > V_ipport_firstauto) 124181803Sbz halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2; 125162767Ssilby else 126181803Sbz halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2; 127162767Ssilby /* Protect against goofy port ranges smaller than 32. */ 128162767Ssilby return (imin(imax(halfrange, 32), maxsockets / 5)); 129162767Ssilby} 130162767Ssilby 131162767Ssilbystatic int 132162031Sglebiussysctl_maxtcptw(SYSCTL_HANDLER_ARGS) 133162031Sglebius{ 134162031Sglebius int error, new; 135157478Sglebius 136162031Sglebius if (maxtcptw == 0) 137162767Ssilby new = tcptw_auto_size(); 138162031Sglebius else 139162031Sglebius new = maxtcptw; 140170289Sdwmalone error = sysctl_handle_int(oidp, &new, 0, req); 141162767Ssilby if (error == 0 && req->newptr) 142162767Ssilby if (new >= 32) { 143162031Sglebius maxtcptw = new; 144162035Sglebius uma_zone_set_max(tcptw_zone, maxtcptw); 145162767Ssilby } 146162031Sglebius return (error); 147162031Sglebius} 148169541Sandre 149162031SglebiusSYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, 150162031Sglebius &maxtcptw, 0, sysctl_maxtcptw, "IU", 151162031Sglebius "Maximum number of compressed TCP TIME_WAIT entries"); 152162031Sglebius 153185348SzecSYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, nolocaltimewait, 154185348Szec CTLFLAG_RW, nolocaltimewait, 0, 155167721Sandre "Do not create compressed TCP TIME_WAIT entries for local connections"); 156162151Sglebius 157169541Sandrevoid 158169541Sandretcp_tw_zone_change(void) 159157927Sps{ 160157927Sps 161162031Sglebius if (maxtcptw == 0) 162162767Ssilby uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 163157927Sps} 164157927Sps 1651541Srgrimesvoid 166169541Sandretcp_tw_init(void) 1671541Srgrimes{ 168183550Szec INIT_VNET_INET(curvnet); 169169454Srwatson 170162031Sglebius tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), 171162031Sglebius NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 172157478Sglebius TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); 173157478Sglebius if (maxtcptw == 0) 174162767Ssilby uma_zone_set_max(tcptw_zone, tcptw_auto_size()); 175162031Sglebius else 176162031Sglebius uma_zone_set_max(tcptw_zone, maxtcptw); 177181803Sbz TAILQ_INIT(&V_twq_2msl); 1781541Srgrimes} 1791541Srgrimes 1801541Srgrimes/* 181111145Sjlemon * Move a TCP connection into TIME_WAIT state. 182138410Srwatson * tcbinfo is locked. 183111145Sjlemon * inp is locked, and is unlocked before returning. 184111145Sjlemon */ 185111145Sjlemonvoid 186157431Srwatsontcp_twstart(struct tcpcb *tp) 187111145Sjlemon{ 188183550Szec#if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) 189183550Szec INIT_VNET_INET(tp->t_vnet); 190183550Szec#endif 191111145Sjlemon struct tcptw *tw; 192162151Sglebius struct inpcb *inp = tp->t_inpcb; 193162111Sru int acknow; 194111145Sjlemon struct socket *so; 195111145Sjlemon 196181803Sbz INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_reset(). */ 197178285Srwatson INP_WLOCK_ASSERT(inp); 198138020Srwatson 199181803Sbz if (V_nolocaltimewait && in_localip(inp->inp_faddr)) { 200162151Sglebius tp = tcp_close(tp); 201162151Sglebius if (tp != NULL) 202178285Srwatson INP_WUNLOCK(inp); 203162151Sglebius return; 204162151Sglebius } 205162151Sglebius 206112009Sjlemon tw = uma_zalloc(tcptw_zone, M_NOWAIT); 207112009Sjlemon if (tw == NULL) { 208169608Sandre tw = tcp_tw_2msl_scan(1); 209112009Sjlemon if (tw == NULL) { 210157376Srwatson tp = tcp_close(tp); 211157376Srwatson if (tp != NULL) 212178285Srwatson INP_WUNLOCK(inp); 213112009Sjlemon return; 214112009Sjlemon } 215112009Sjlemon } 216111145Sjlemon tw->tw_inpcb = inp; 217111145Sjlemon 218111145Sjlemon /* 219111145Sjlemon * Recover last window size sent. 220111145Sjlemon */ 221111145Sjlemon tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; 222111145Sjlemon 223111145Sjlemon /* 224111145Sjlemon * Set t_recent if timestamps are used on the connection. 225111145Sjlemon */ 226133874Srwatson if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == 227169477Sandre (TF_REQ_TSTMP|TF_RCVD_TSTMP)) { 228111145Sjlemon tw->t_recent = tp->ts_recent; 229169477Sandre tw->ts_offset = tp->ts_offset; 230169477Sandre } else { 231111145Sjlemon tw->t_recent = 0; 232169477Sandre tw->ts_offset = 0; 233169477Sandre } 234111145Sjlemon 235111145Sjlemon tw->snd_nxt = tp->snd_nxt; 236111145Sjlemon tw->rcv_nxt = tp->rcv_nxt; 237121850Ssilby tw->iss = tp->iss; 238121884Ssilby tw->irs = tp->irs; 239111145Sjlemon tw->t_starttime = tp->t_starttime; 240112009Sjlemon tw->tw_time = 0; 241111145Sjlemon 242111145Sjlemon/* XXX 243111145Sjlemon * If this code will 244111145Sjlemon * be used for fin-wait-2 state also, then we may need 245111145Sjlemon * a ts_recent from the last segment. 246111145Sjlemon */ 247137139Sandre acknow = tp->t_flags & TF_ACKNOW; 248157376Srwatson 249157376Srwatson /* 250157376Srwatson * First, discard tcpcb state, which includes stopping its timers and 251157376Srwatson * freeing it. tcp_discardcb() used to also release the inpcb, but 252157376Srwatson * that work is now done in the caller. 253160925Srwatson * 254160925Srwatson * Note: soisdisconnected() call used to be made in tcp_discardcb(), 255160925Srwatson * and might not be needed here any longer. 256157376Srwatson */ 257111145Sjlemon tcp_discardcb(tp); 258111145Sjlemon so = inp->inp_socket; 259160925Srwatson soisdisconnected(so); 260169347Srwatson tw->tw_cred = crhold(so->so_cred); 261130387Srwatson SOCK_LOCK(so); 262111145Sjlemon tw->tw_so_options = so->so_options; 263157376Srwatson SOCK_UNLOCK(so); 264114794Srwatson if (acknow) 265126351Srwatson tcp_twrespond(tw, TH_ACK); 266157432Srwatson inp->inp_ppcb = tw; 267111145Sjlemon inp->inp_vflag |= INP_TIMEWAIT; 268169608Sandre tcp_tw_2msl_reset(tw, 0); 269157376Srwatson 270157376Srwatson /* 271157376Srwatson * If the inpcb owns the sole reference to the socket, then we can 272157376Srwatson * detach and free the socket as it is not needed in time wait. 273157376Srwatson */ 274157376Srwatson if (inp->inp_vflag & INP_SOCKREF) { 275157376Srwatson KASSERT(so->so_state & SS_PROTOREF, 276157376Srwatson ("tcp_twstart: !SS_PROTOREF")); 277157376Srwatson inp->inp_vflag &= ~INP_SOCKREF; 278178285Srwatson INP_WUNLOCK(inp); 279157376Srwatson ACCEPT_LOCK(); 280157376Srwatson SOCK_LOCK(so); 281157376Srwatson so->so_state &= ~SS_PROTOREF; 282157376Srwatson sofree(so); 283157376Srwatson } else 284178285Srwatson INP_WUNLOCK(inp); 285111145Sjlemon} 286111145Sjlemon 287162064Sglebius#if 0 288121850Ssilby/* 289121884Ssilby * The appromixate rate of ISN increase of Microsoft TCP stacks; 290121884Ssilby * the actual rate is slightly higher due to the addition of 291121884Ssilby * random positive increments. 292121884Ssilby * 293121884Ssilby * Most other new OSes use semi-randomized ISN values, so we 294121884Ssilby * do not need to worry about them. 295121884Ssilby */ 296121884Ssilby#define MS_ISN_BYTES_PER_SECOND 250000 297121884Ssilby 298121884Ssilby/* 299121850Ssilby * Determine if the ISN we will generate has advanced beyond the last 300121850Ssilby * sequence number used by the previous connection. If so, indicate 301121850Ssilby * that it is safe to recycle this tw socket by returning 1. 302121850Ssilby */ 303121850Ssilbyint 304121850Ssilbytcp_twrecycleable(struct tcptw *tw) 305121850Ssilby{ 306183550Szec INIT_VNET_INET(curvnet); 307121884Ssilby tcp_seq new_iss = tw->iss; 308121884Ssilby tcp_seq new_irs = tw->irs; 309121850Ssilby 310181803Sbz INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 311121884Ssilby new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz); 312121884Ssilby new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz); 313133874Srwatson 314121884Ssilby if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt)) 315139222Srwatson return (1); 316121850Ssilby else 317139222Srwatson return (0); 318121850Ssilby} 319162064Sglebius#endif 320121850Ssilby 321169608Sandre/* 322169608Sandre * Returns 1 if the TIME_WAIT state was killed and we should start over, 323169608Sandre * looking for a pcb in the listen state. Returns 0 otherwise. 324169608Sandre */ 325169608Sandreint 326169608Sandretcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th, 327169608Sandre struct mbuf *m, int tlen) 328169608Sandre{ 329183550Szec#if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) 330183550Szec INIT_VNET_INET(curvnet); 331183550Szec#endif 332169608Sandre struct tcptw *tw; 333169608Sandre int thflags; 334169608Sandre tcp_seq seq; 335169608Sandre#ifdef INET6 336169608Sandre int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 337169608Sandre#else 338169608Sandre const int isipv6 = 0; 339169608Sandre#endif 340169608Sandre 341169608Sandre /* tcbinfo lock required for tcp_twclose(), tcp_tw_2msl_reset(). */ 342181803Sbz INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 343178285Srwatson INP_WLOCK_ASSERT(inp); 344169608Sandre 345169608Sandre /* 346169608Sandre * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 347169608Sandre * still present. This is undesirable, but temporarily necessary 348169608Sandre * until we work out how to handle inpcb's who's timewait state has 349169608Sandre * been removed. 350169608Sandre */ 351169608Sandre tw = intotw(inp); 352169608Sandre if (tw == NULL) 353169608Sandre goto drop; 354169608Sandre 355169608Sandre thflags = th->th_flags; 356169608Sandre 357169608Sandre /* 358169608Sandre * NOTE: for FIN_WAIT_2 (to be added later), 359169608Sandre * must validate sequence number before accepting RST 360169608Sandre */ 361169608Sandre 362169608Sandre /* 363169608Sandre * If the segment contains RST: 364169608Sandre * Drop the segment - see Stevens, vol. 2, p. 964 and 365169608Sandre * RFC 1337. 366169608Sandre */ 367169608Sandre if (thflags & TH_RST) 368169608Sandre goto drop; 369169608Sandre 370169608Sandre#if 0 371169608Sandre/* PAWS not needed at the moment */ 372169608Sandre /* 373169608Sandre * RFC 1323 PAWS: If we have a timestamp reply on this segment 374169608Sandre * and it's less than ts_recent, drop it. 375169608Sandre */ 376169608Sandre if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 377169608Sandre TSTMP_LT(to.to_tsval, tp->ts_recent)) { 378169608Sandre if ((thflags & TH_ACK) == 0) 379169608Sandre goto drop; 380169608Sandre goto ack; 381169608Sandre } 382169608Sandre /* 383169608Sandre * ts_recent is never updated because we never accept new segments. 384169608Sandre */ 385169608Sandre#endif 386169608Sandre 387169608Sandre /* 388169608Sandre * If a new connection request is received 389169608Sandre * while in TIME_WAIT, drop the old connection 390169608Sandre * and start over if the sequence numbers 391169608Sandre * are above the previous ones. 392169608Sandre */ 393169608Sandre if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 394169608Sandre tcp_twclose(tw, 0); 395169608Sandre return (1); 396169608Sandre } 397169608Sandre 398169608Sandre /* 399169608Sandre * Drop the the segment if it does not contain an ACK. 400169608Sandre */ 401169608Sandre if ((thflags & TH_ACK) == 0) 402169608Sandre goto drop; 403169608Sandre 404169608Sandre /* 405169608Sandre * Reset the 2MSL timer if this is a duplicate FIN. 406169608Sandre */ 407169608Sandre if (thflags & TH_FIN) { 408169608Sandre seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 409169608Sandre if (seq + 1 == tw->rcv_nxt) 410169608Sandre tcp_tw_2msl_reset(tw, 1); 411169608Sandre } 412169608Sandre 413169608Sandre /* 414169608Sandre * Acknowledge the segment if it has data or is not a duplicate ACK. 415169608Sandre */ 416169608Sandre if (thflags != TH_ACK || tlen != 0 || 417169608Sandre th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 418169608Sandre tcp_twrespond(tw, TH_ACK); 419169608Sandre goto drop; 420169608Sandre 421169608Sandre /* 422169608Sandre * Generate a RST, dropping incoming segment. 423169608Sandre * Make ACK acceptable to originator of segment. 424169608Sandre * Don't bother to respond if destination was broadcast/multicast. 425169608Sandre */ 426169608Sandre if (m->m_flags & (M_BCAST|M_MCAST)) 427169608Sandre goto drop; 428169608Sandre if (isipv6) { 429169635Soleg#ifdef INET6 430169608Sandre struct ip6_hdr *ip6; 431169608Sandre 432169608Sandre /* IPv6 anycast check is done at tcp6_input() */ 433169608Sandre ip6 = mtod(m, struct ip6_hdr *); 434169608Sandre if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 435169608Sandre IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 436169608Sandre goto drop; 437169635Soleg#endif 438169608Sandre } else { 439169608Sandre struct ip *ip; 440169608Sandre 441169608Sandre ip = mtod(m, struct ip *); 442169608Sandre if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 443169608Sandre IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 444169608Sandre ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 445169608Sandre in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 446169608Sandre goto drop; 447169608Sandre } 448169608Sandre if (thflags & TH_ACK) { 449169608Sandre tcp_respond(NULL, 450169608Sandre mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 451169608Sandre } else { 452169608Sandre seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 453169608Sandre tcp_respond(NULL, 454169608Sandre mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 455169608Sandre } 456178285Srwatson INP_WUNLOCK(inp); 457169608Sandre return (0); 458169608Sandre 459169608Sandredrop: 460178285Srwatson INP_WUNLOCK(inp); 461169608Sandre m_freem(m); 462169608Sandre return (0); 463169608Sandre} 464169608Sandre 465157376Srwatsonvoid 466112009Sjlemontcp_twclose(struct tcptw *tw, int reuse) 467111145Sjlemon{ 468183550Szec INIT_VNET_INET(curvnet); 469157376Srwatson struct socket *so; 470111145Sjlemon struct inpcb *inp; 471111145Sjlemon 472157376Srwatson /* 473157386Srwatson * At this point, we are in one of two situations: 474157376Srwatson * 475160549Srwatson * (1) We have no socket, just an inpcb<->twtcp pair. We can free 476160549Srwatson * all state. 477157386Srwatson * 478160549Srwatson * (2) We have a socket -- if we own a reference, release it and 479160549Srwatson * notify the socket layer. 480157376Srwatson */ 481111145Sjlemon inp = tw->tw_inpcb; 482157376Srwatson KASSERT((inp->inp_vflag & INP_TIMEWAIT), ("tcp_twclose: !timewait")); 483157432Srwatson KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); 484181803Sbz INP_INFO_WLOCK_ASSERT(&V_tcbinfo); /* tcp_tw_2msl_stop(). */ 485178285Srwatson INP_WLOCK_ASSERT(inp); 486138020Srwatson 487111145Sjlemon tw->tw_inpcb = NULL; 488169608Sandre tcp_tw_2msl_stop(tw); 489111145Sjlemon inp->inp_ppcb = NULL; 490158009Srwatson in_pcbdrop(inp); 491157376Srwatson 492157376Srwatson so = inp->inp_socket; 493157386Srwatson if (so != NULL) { 494160549Srwatson /* 495160549Srwatson * If there's a socket, handle two cases: first, we own a 496160549Srwatson * strong reference, which we will now release, or we don't 497160549Srwatson * in which case another reference exists (XXXRW: think 498160549Srwatson * about this more), and we don't need to take action. 499160549Srwatson */ 500157386Srwatson if (inp->inp_vflag & INP_SOCKREF) { 501157386Srwatson inp->inp_vflag &= ~INP_SOCKREF; 502178285Srwatson INP_WUNLOCK(inp); 503157386Srwatson ACCEPT_LOCK(); 504157386Srwatson SOCK_LOCK(so); 505157386Srwatson KASSERT(so->so_state & SS_PROTOREF, 506157386Srwatson ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); 507157386Srwatson so->so_state &= ~SS_PROTOREF; 508157386Srwatson sofree(so); 509157376Srwatson } else { 510157386Srwatson /* 511157386Srwatson * If we don't own the only reference, the socket and 512157386Srwatson * inpcb need to be left around to be handled by 513157386Srwatson * tcp_usr_detach() later. 514157386Srwatson */ 515178285Srwatson INP_WUNLOCK(inp); 516157376Srwatson } 517185370Sbz } else 518185370Sbz in_pcbfree(inp); 519181803Sbz V_tcpstat.tcps_closed++; 520126002Spjd crfree(tw->tw_cred); 521126002Spjd tw->tw_cred = NULL; 522112009Sjlemon if (reuse) 523157376Srwatson return; 524112009Sjlemon uma_zfree(tcptw_zone, tw); 525111145Sjlemon} 526111145Sjlemon 527111145Sjlemonint 528126351Srwatsontcp_twrespond(struct tcptw *tw, int flags) 529111145Sjlemon{ 530183550Szec INIT_VNET_INET(curvnet); 531111145Sjlemon struct inpcb *inp = tw->tw_inpcb; 532111145Sjlemon struct tcphdr *th; 533111145Sjlemon struct mbuf *m; 534111145Sjlemon struct ip *ip = NULL; 535111145Sjlemon u_int hdrlen, optlen; 536111145Sjlemon int error; 537168845Sandre struct tcpopt to; 538111145Sjlemon#ifdef INET6 539111145Sjlemon struct ip6_hdr *ip6 = NULL; 540111145Sjlemon int isipv6 = inp->inp_inc.inc_isipv6; 541111145Sjlemon#endif 542111145Sjlemon 543178285Srwatson INP_WLOCK_ASSERT(inp); 544138020Srwatson 545151967Sandre m = m_gethdr(M_DONTWAIT, MT_DATA); 546111145Sjlemon if (m == NULL) 547111145Sjlemon return (ENOBUFS); 548111145Sjlemon m->m_data += max_linkhdr; 549111145Sjlemon 550114794Srwatson#ifdef MAC 551172930Srwatson mac_inpcb_create_mbuf(inp, m); 552114794Srwatson#endif 553114794Srwatson 554111153Sjlemon#ifdef INET6 555111145Sjlemon if (isipv6) { 556111145Sjlemon hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 557111145Sjlemon ip6 = mtod(m, struct ip6_hdr *); 558111145Sjlemon th = (struct tcphdr *)(ip6 + 1); 559111145Sjlemon tcpip_fillheaders(inp, ip6, th); 560111153Sjlemon } else 561111153Sjlemon#endif 562111153Sjlemon { 563111145Sjlemon hdrlen = sizeof(struct tcpiphdr); 564111145Sjlemon ip = mtod(m, struct ip *); 565111145Sjlemon th = (struct tcphdr *)(ip + 1); 566111145Sjlemon tcpip_fillheaders(inp, ip, th); 567111145Sjlemon } 568168845Sandre to.to_flags = 0; 569133874Srwatson 570133874Srwatson /* 571111145Sjlemon * Send a timestamp and echo-reply if both our side and our peer 572111145Sjlemon * have sent timestamps in our SYN's and this is not a RST. 573133874Srwatson */ 574111145Sjlemon if (tw->t_recent && flags == TH_ACK) { 575168845Sandre to.to_flags |= TOF_TS; 576169477Sandre to.to_tsval = ticks + tw->ts_offset; 577168845Sandre to.to_tsecr = tw->t_recent; 578133874Srwatson } 579168845Sandre optlen = tcp_addoptions(&to, (u_char *)(th + 1)); 580111145Sjlemon 581111145Sjlemon m->m_len = hdrlen + optlen; 582111145Sjlemon m->m_pkthdr.len = m->m_len; 583111145Sjlemon 584111145Sjlemon KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); 585111145Sjlemon 586111145Sjlemon th->th_seq = htonl(tw->snd_nxt); 587111145Sjlemon th->th_ack = htonl(tw->rcv_nxt); 588111145Sjlemon th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 589111145Sjlemon th->th_flags = flags; 590111145Sjlemon th->th_win = htons(tw->last_win); 591111145Sjlemon 592111153Sjlemon#ifdef INET6 593111145Sjlemon if (isipv6) { 594111145Sjlemon th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), 595111145Sjlemon sizeof(struct tcphdr) + optlen); 596122922Sandre ip6->ip6_hlim = in6_selecthlim(inp, NULL); 597122922Sandre error = ip6_output(m, inp->in6p_outputopts, NULL, 598111145Sjlemon (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); 599111153Sjlemon } else 600111153Sjlemon#endif 601111153Sjlemon { 602111145Sjlemon th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 603133874Srwatson htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); 604111145Sjlemon m->m_pkthdr.csum_flags = CSUM_TCP; 605111145Sjlemon m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 606111145Sjlemon ip->ip_len = m->m_pkthdr.len; 607181803Sbz if (V_path_mtu_discovery) 608124248Sandre ip->ip_off |= IP_DF; 609122922Sandre error = ip_output(m, inp->inp_options, NULL, 610134793Sjmg ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 611134793Sjmg NULL, inp); 612111145Sjlemon } 613111145Sjlemon if (flags & TH_ACK) 614181803Sbz V_tcpstat.tcps_sndacks++; 615111145Sjlemon else 616181803Sbz V_tcpstat.tcps_sndctrl++; 617181803Sbz V_tcpstat.tcps_sndtotal++; 618111145Sjlemon return (error); 619111145Sjlemon} 620169608Sandre 621169608Sandrestatic void 622169608Sandretcp_tw_2msl_reset(struct tcptw *tw, int rearm) 623169608Sandre{ 624183550Szec INIT_VNET_INET(curvnet); 625169608Sandre 626181803Sbz INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 627178285Srwatson INP_WLOCK_ASSERT(tw->tw_inpcb); 628169608Sandre if (rearm) 629181803Sbz TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 630169608Sandre tw->tw_time = ticks + 2 * tcp_msl; 631181803Sbz TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl); 632169608Sandre} 633169608Sandre 634169608Sandrestatic void 635169608Sandretcp_tw_2msl_stop(struct tcptw *tw) 636169608Sandre{ 637183550Szec INIT_VNET_INET(curvnet); 638169608Sandre 639181803Sbz INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 640181803Sbz TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); 641169608Sandre} 642169608Sandre 643169608Sandrestruct tcptw * 644169608Sandretcp_tw_2msl_scan(int reuse) 645169608Sandre{ 646183550Szec INIT_VNET_INET(curvnet); 647169608Sandre struct tcptw *tw; 648169608Sandre 649181803Sbz INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 650169608Sandre for (;;) { 651181803Sbz tw = TAILQ_FIRST(&V_twq_2msl); 652169608Sandre if (tw == NULL || (!reuse && tw->tw_time > ticks)) 653169608Sandre break; 654178285Srwatson INP_WLOCK(tw->tw_inpcb); 655169608Sandre tcp_twclose(tw, reuse); 656169608Sandre if (reuse) 657169608Sandre return (tw); 658169608Sandre } 659169608Sandre return (NULL); 660169608Sandre} 661