12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714271527162717271827192720272127222723272427252726272727282729273027312732273327342735273627372738273927402741274227432744274527462747274827492750275127522753275427552756275727582759276027612762276327642765276627672768276927702771277227732774277527762777277827792780278127822783278427852786278727882789279027912792279327942795279627972798279928002801280228032804280528062807280828092810281128122813281428152816281728182819282028212822282328242825282628272828282928302831283228332834283528362837283828392840284128422843284428452846284728482849285028512852285328542855285628572858285928602861286228632864286528662867286828692870287128722873287428752876287728782879288028812882288328842885288628872888288928902891289228932894289528962897289828992900290129022903290429052906290729082909291029112912291329142915291629172918291929202921292229232924292529262927292829292930293129322933293429352936293729382939294029412942294329442945294629472948294929502951295229532954295529562957295829592960296129622963296429652966296729682969297029712972297329742975297629772978297929802981298229832984298529862987298829892990299129922993299429952996299729982999300030013002300330043005300630073008300930103011301230133014301530163017301830193020302130223023302430253026302730283029303030313032303330343035303630373038303930403041304230433044304530463047304830493050305130523053305430553056305730583059306030613062306330643065306630673068306930703071307230733074307530763077307830793080308130823083308430853086308730883089309030913092309330943095309630973098309931003101310231033104310531063107310831093110311131123113311431153116311731183119312031213122312331243125312631273128312931303131313231333134313531363137313831393140314131423143314431453146314731483149315031513152315331543155315631573158315931603161316231633164316531663167 |
- #include "u.h"
- #include "../port/lib.h"
- #include "mem.h"
- #include "dat.h"
- #include "fns.h"
- #include "../port/error.h"
- #include "ip.h"
- enum
- {
- QMAX = 64*1024-1,
- IP_TCPPROTO = 6,
- TCP4_IPLEN = 8,
- TCP4_PHDRSIZE = 12,
- TCP4_HDRSIZE = 20,
- TCP4_TCBPHDRSZ = 40,
- TCP4_PKT = TCP4_IPLEN+TCP4_PHDRSIZE,
- TCP6_IPLEN = 0,
- TCP6_PHDRSIZE = 40,
- TCP6_HDRSIZE = 20,
- TCP6_TCBPHDRSZ = 60,
- TCP6_PKT = TCP6_IPLEN+TCP6_PHDRSIZE,
- TcptimerOFF = 0,
- TcptimerON = 1,
- TcptimerDONE = 2,
- MAX_TIME = (1<<20), /* Forever */
- TCP_ACK = 50, /* Timed ack sequence in ms */
- MAXBACKMS = 30000, /* longest backoff time (ms) before hangup */
- URG = 0x20, /* Data marked urgent */
- ACK = 0x10, /* Acknowledge is valid */
- PSH = 0x08, /* Whole data pipe is pushed */
- RST = 0x04, /* Reset connection */
- SYN = 0x02, /* Pkt. is synchronise */
- FIN = 0x01, /* Start close down */
- EOLOPT = 0,
- NOOPOPT = 1,
- MSSOPT = 2,
- MSS_LENGTH = 4, /* Mean segment size */
- WSOPT = 3,
- WS_LENGTH = 3, /* Bits to scale window size by */
- MSL2 = 10,
- MSPTICK = 50, /* Milliseconds per timer tick */
- DEF_MSS = 1460, /* Default mean segment */
- DEF_MSS6 = 1280, /* Default mean segment (min) for v6 */
- DEF_RTT = 500, /* Default round trip */
- DEF_KAT = 120000, /* Default time ms) between keep alives */
- TCP_LISTEN = 0, /* Listen connection */
- TCP_CONNECT = 1, /* Outgoing connection */
- SYNACK_RXTIMER = 250, /* ms between SYNACK retransmits */
- TCPREXMTTHRESH = 3, /* dupack threshhold for rxt */
- FORCE = 1,
- CLONE = 2,
- RETRAN = 4,
- ACTIVE = 8,
- SYNACK = 16,
- LOGAGAIN = 3,
- LOGDGAIN = 2,
- Closed = 0, /* Connection states */
- Listen,
- Syn_sent,
- Syn_received,
- Established,
- Finwait1,
- Finwait2,
- Close_wait,
- Closing,
- Last_ack,
- Time_wait,
- Maxlimbo = 1000, /* maximum procs waiting for response to SYN ACK */
- NLHT = 256, /* hash table size, must be a power of 2 */
- LHTMASK = NLHT-1,
- HaveWS = 1<<8,
- };
- /* Must correspond to the enumeration above */
- char *tcpstates[] =
- {
- "Closed", "Listen", "Syn_sent", "Syn_received",
- "Established", "Finwait1", "Finwait2", "Close_wait",
- "Closing", "Last_ack", "Time_wait"
- };
- typedef struct Tcptimer Tcptimer;
- struct Tcptimer
- {
- Tcptimer *next;
- Tcptimer *prev;
- Tcptimer *readynext;
- int state;
- int start;
- int count;
- void (*func)(void*);
- void *arg;
- };
- /*
- * v4 and v6 pseudo headers used for
- * checksuming tcp
- */
- typedef struct Tcp4hdr Tcp4hdr;
- struct Tcp4hdr
- {
- uchar vihl; /* Version and header length */
- uchar tos; /* Type of service */
- uchar length[2]; /* packet length */
- uchar id[2]; /* Identification */
- uchar frag[2]; /* Fragment information */
- uchar Unused;
- uchar proto;
- uchar tcplen[2];
- uchar tcpsrc[4];
- uchar tcpdst[4];
- uchar tcpsport[2];
- uchar tcpdport[2];
- uchar tcpseq[4];
- uchar tcpack[4];
- uchar tcpflag[2];
- uchar tcpwin[2];
- uchar tcpcksum[2];
- uchar tcpurg[2];
- /* Options segment */
- uchar tcpopt[1];
- };
- typedef struct Tcp6hdr Tcp6hdr;
- struct Tcp6hdr
- {
- uchar vcf[4];
- uchar ploadlen[2];
- uchar proto;
- uchar ttl;
- uchar tcpsrc[IPaddrlen];
- uchar tcpdst[IPaddrlen];
- uchar tcpsport[2];
- uchar tcpdport[2];
- uchar tcpseq[4];
- uchar tcpack[4];
- uchar tcpflag[2];
- uchar tcpwin[2];
- uchar tcpcksum[2];
- uchar tcpurg[2];
- /* Options segment */
- uchar tcpopt[1];
- };
- /*
- * this represents the control info
- * for a single packet. It is derived from
- * a packet in ntohtcp{4,6}() and stuck into
- * a packet in htontcp{4,6}().
- */
- typedef struct Tcp Tcp;
- struct Tcp
- {
- ushort source;
- ushort dest;
- ulong seq;
- ulong ack;
- uchar flags;
- ushort ws; /* window scale option (if not zero) */
- ulong wnd;
- ushort urg;
- ushort mss; /* max segment size option (if not zero) */
- ushort len; /* size of data */
- };
- /*
- * this header is malloc'd to thread together fragments
- * waiting to be coalesced
- */
- typedef struct Reseq Reseq;
- struct Reseq
- {
- Reseq *next;
- Tcp seg;
- Block *bp;
- ushort length;
- };
- /*
- * the qlock in the Conv locks this structure
- */
- typedef struct Tcpctl Tcpctl;
- struct Tcpctl
- {
- uchar state; /* Connection state */
- uchar type; /* Listening or active connection */
- uchar code; /* Icmp code */
- struct {
- ulong una; /* Unacked data pointer */
- ulong nxt; /* Next sequence expected */
- ulong ptr; /* Data pointer */
- ulong wnd; /* Tcp send window */
- ulong urg; /* Urgent data pointer */
- ulong wl2;
- int scale; /* how much to right shift window in xmitted packets */
- /* to implement tahoe and reno TCP */
- ulong dupacks; /* number of duplicate acks rcvd */
- int recovery; /* loss recovery flag */
- ulong rxt; /* right window marker for recovery */
- } snd;
- struct {
- ulong nxt; /* Receive pointer to next uchar slot */
- ulong wnd; /* Receive window incoming */
- ulong urg; /* Urgent pointer */
- int blocked;
- int una; /* unacked data segs */
- int scale; /* how much to left shift window in rcved packets */
- } rcv;
- ulong iss; /* Initial sequence number */
- int sawwsopt; /* true if we saw a wsopt on the incoming SYN */
- ulong cwind; /* Congestion window */
- int scale; /* desired snd.scale */
- ushort ssthresh; /* Slow start threshold */
- int resent; /* Bytes just resent */
- int irs; /* Initial received squence */
- ushort mss; /* Mean segment size */
- int rerecv; /* Overlap of data rerecevived */
- ulong window; /* Recevive window */
- uchar backoff; /* Exponential backoff counter */
- int backedoff; /* ms we've backed off for rexmits */
- uchar flags; /* State flags */
- Reseq *reseq; /* Resequencing queue */
- Tcptimer timer; /* Activity timer */
- Tcptimer acktimer; /* Acknowledge timer */
- Tcptimer rtt_timer; /* Round trip timer */
- Tcptimer katimer; /* keep alive timer */
- ulong rttseq; /* Round trip sequence */
- int srtt; /* Shortened round trip */
- int mdev; /* Mean deviation of round trip */
- int kacounter; /* count down for keep alive */
- uint sndsyntime; /* time syn sent */
- ulong time; /* time Finwait2 or Syn_received was sent */
- int nochecksum; /* non-zero means don't send checksums */
- int flgcnt; /* number of flags in the sequence (FIN,SEQ) */
- union {
- Tcp4hdr tcp4hdr;
- Tcp6hdr tcp6hdr;
- } protohdr; /* prototype header */
- };
- /*
- * New calls are put in limbo rather than having a conversation structure
- * allocated. Thus, a SYN attack results in lots of limbo'd calls but not
- * any real Conv structures mucking things up. Calls in limbo rexmit their
- * SYN ACK every SYNACK_RXTIMER ms up to 4 times, i.e., they disappear after 1 second.
- *
- * In particular they aren't on a listener's queue so that they don't figure
- * in the input queue limit.
- *
- * If 1/2 of a T3 was attacking SYN packets, we'ld have a permanent queue
- * of 70000 limbo'd calls. Not great for a linear list but doable. Therefore
- * there is no hashing of this list.
- */
- typedef struct Limbo Limbo;
- struct Limbo
- {
- Limbo *next;
- uchar laddr[IPaddrlen];
- uchar raddr[IPaddrlen];
- ushort lport;
- ushort rport;
- ulong irs; /* initial received sequence */
- ulong iss; /* initial sent sequence */
- ushort mss; /* mss from the other end */
- ushort rcvscale; /* how much to scale rcvd windows */
- ushort sndscale; /* how much to scale sent windows */
- ulong lastsend; /* last time we sent a synack */
- uchar version; /* v4 or v6 */
- uchar rexmits; /* number of retransmissions */
- };
- int tcp_irtt = DEF_RTT; /* Initial guess at round trip time */
- ushort tcp_mss = DEF_MSS; /* Maximum segment size to be sent */
- enum {
- /* MIB stats */
- MaxConn,
- ActiveOpens,
- PassiveOpens,
- EstabResets,
- CurrEstab,
- InSegs,
- OutSegs,
- RetransSegs,
- RetransTimeouts,
- InErrs,
- OutRsts,
- /* non-MIB stats */
- CsumErrs,
- HlenErrs,
- LenErrs,
- OutOfOrder,
- Nstats
- };
- static char *statnames[] =
- {
- [MaxConn] "MaxConn",
- [ActiveOpens] "ActiveOpens",
- [PassiveOpens] "PassiveOpens",
- [EstabResets] "EstabResets",
- [CurrEstab] "CurrEstab",
- [InSegs] "InSegs",
- [OutSegs] "OutSegs",
- [RetransSegs] "RetransSegs",
- [RetransTimeouts] "RetransTimeouts",
- [InErrs] "InErrs",
- [OutRsts] "OutRsts",
- [CsumErrs] "CsumErrs",
- [HlenErrs] "HlenErrs",
- [LenErrs] "LenErrs",
- [OutOfOrder] "OutOfOrder",
- };
- typedef struct Tcppriv Tcppriv;
- struct Tcppriv
- {
- /* List of active timers */
- QLock tl;
- Tcptimer *timers;
- /* hash table for matching conversations */
- Ipht ht;
- /* calls in limbo waiting for an ACK to our SYN ACK */
- int nlimbo;
- Limbo *lht[NLHT];
- /* for keeping track of tcpackproc */
- QLock apl;
- int ackprocstarted;
- ulong stats[Nstats];
- };
- /*
- * Setting tcpporthogdefense to non-zero enables Dong Lin's
- * solution to hijacked systems staking out port's as a form
- * of DoS attack.
- *
- * To avoid stateless Conv hogs, we pick a sequence number at random. If
- * it that number gets acked by the other end, we shut down the connection.
- * Look for tcpporthogedefense in the code.
- */
- int tcpporthogdefense = 0;
- int addreseq(Tcpctl*, Tcppriv*, Tcp*, Block*, ushort);
- void getreseq(Tcpctl*, Tcp*, Block**, ushort*);
- void localclose(Conv*, char*);
- void procsyn(Conv*, Tcp*);
- void tcpiput(Proto*, Ipifc*, Block*);
- void tcpoutput(Conv*);
- int tcptrim(Tcpctl*, Tcp*, Block**, ushort*);
- void tcpstart(Conv*, int);
- void tcptimeout(void*);
- void tcpsndsyn(Conv*, Tcpctl*);
- void tcprcvwin(Conv*);
- void tcpacktimer(void*);
- void tcpkeepalive(void*);
- void tcpsetkacounter(Tcpctl*);
- void tcprxmit(Conv*);
- void tcpsettimer(Tcpctl*);
- void tcpsynackrtt(Conv*);
- void tcpsetscale(Conv*, Tcpctl*, ushort, ushort);
- static void limborexmit(Proto*);
- static void limbo(Conv*, uchar*, uchar*, Tcp*, int);
- void
- tcpsetstate(Conv *s, uchar newstate)
- {
- Tcpctl *tcb;
- uchar oldstate;
- Tcppriv *tpriv;
- tpriv = s->p->priv;
- tcb = (Tcpctl*)s->ptcl;
- oldstate = tcb->state;
- if(oldstate == newstate)
- return;
- if(oldstate == Established)
- tpriv->stats[CurrEstab]--;
- if(newstate == Established)
- tpriv->stats[CurrEstab]++;
- /**
- print( "%d/%d %s->%s CurrEstab=%d\n", s->lport, s->rport,
- tcpstates[oldstate], tcpstates[newstate], tpriv->tstats.tcpCurrEstab );
- **/
- switch(newstate) {
- case Closed:
- qclose(s->rq);
- qclose(s->wq);
- qclose(s->eq);
- break;
- case Close_wait: /* Remote closes */
- qhangup(s->rq, nil);
- break;
- }
- tcb->state = newstate;
- if(oldstate == Syn_sent && newstate != Closed)
- Fsconnected(s, nil);
- }
- static char*
- tcpconnect(Conv *c, char **argv, int argc)
- {
- char *e;
- e = Fsstdconnect(c, argv, argc);
- if(e != nil)
- return e;
- tcpstart(c, TCP_CONNECT);
- return nil;
- }
- static int
- tcpstate(Conv *c, char *state, int n)
- {
- Tcpctl *s;
- s = (Tcpctl*)(c->ptcl);
- return snprint(state, n,
- "%s srtt %d mdev %d cwin %lud swin %lud>>%d rwin %lud>>%d timer.start %d timer.count %d rerecv %d\n",
- tcpstates[s->state], s->srtt, s->mdev,
- s->cwind, s->snd.wnd, s->rcv.scale, s->rcv.wnd, s->snd.scale,
- s->timer.start, s->timer.count, s->rerecv);
- }
- static int
- tcpinuse(Conv *c)
- {
- Tcpctl *s;
- s = (Tcpctl*)(c->ptcl);
- return s->state != Closed;
- }
- static char*
- tcpannounce(Conv *c, char **argv, int argc)
- {
- char *e;
- e = Fsstdannounce(c, argv, argc);
- if(e != nil)
- return e;
- tcpstart(c, TCP_LISTEN);
- Fsconnected(c, nil);
- return nil;
- }
- /*
- * tcpclose is always called with the q locked
- */
- static void
- tcpclose(Conv *c)
- {
- Tcpctl *tcb;
- tcb = (Tcpctl*)c->ptcl;
- qhangup(c->rq, nil);
- qhangup(c->wq, nil);
- qhangup(c->eq, nil);
- qflush(c->rq);
- switch(tcb->state) {
- case Listen:
- /*
- * reset any incoming calls to this listener
- */
- Fsconnected(c, "Hangup");
- localclose(c, nil);
- break;
- case Closed:
- case Syn_sent:
- localclose(c, nil);
- break;
- case Syn_received:
- case Established:
- tcb->flgcnt++;
- tcb->snd.nxt++;
- tcpsetstate(c, Finwait1);
- tcpoutput(c);
- break;
- case Close_wait:
- tcb->flgcnt++;
- tcb->snd.nxt++;
- tcpsetstate(c, Last_ack);
- tcpoutput(c);
- break;
- }
- }
- void
- tcpkick(void *x)
- {
- Conv *s = x;
- Tcpctl *tcb;
- tcb = (Tcpctl*)s->ptcl;
- if(waserror()){
- qunlock(s);
- nexterror();
- }
- qlock(s);
- switch(tcb->state) {
- case Syn_sent:
- case Syn_received:
- case Established:
- case Close_wait:
- /*
- * Push data
- */
- tcprcvwin(s);
- tcpoutput(s);
- break;
- default:
- localclose(s, "Hangup");
- break;
- }
- qunlock(s);
- poperror();
- }
- void
- tcprcvwin(Conv *s) /* Call with tcb locked */
- {
- int w;
- Tcpctl *tcb;
- tcb = (Tcpctl*)s->ptcl;
- w = tcb->window - qlen(s->rq);
- if(w < 0)
- w = 0;
- tcb->rcv.wnd = w;
- if(w == 0)
- tcb->rcv.blocked = 1;
- }
- void
- tcpacktimer(void *v)
- {
- Tcpctl *tcb;
- Conv *s;
- s = v;
- tcb = (Tcpctl*)s->ptcl;
- if(waserror()){
- qunlock(s);
- nexterror();
- }
- qlock(s);
- if(tcb->state != Closed){
- tcb->flags |= FORCE;
- tcprcvwin(s);
- tcpoutput(s);
- }
- qunlock(s);
- poperror();
- }
- static void
- tcpcreate(Conv *c)
- {
- c->rq = qopen(QMAX, Qcoalesce, tcpacktimer, c);
- c->wq = qopen((3*QMAX)/2, Qkick, tcpkick, c);
- }
- static void
- timerstate(Tcppriv *priv, Tcptimer *t, int newstate)
- {
- if(newstate != TcptimerON){
- if(t->state == TcptimerON){
- // unchain
- if(priv->timers == t){
- priv->timers = t->next;
- if(t->prev != nil)
- panic("timerstate1");
- }
- if(t->next)
- t->next->prev = t->prev;
- if(t->prev)
- t->prev->next = t->next;
- t->next = t->prev = nil;
- }
- } else {
- if(t->state != TcptimerON){
- // chain
- if(t->prev != nil || t->next != nil)
- panic("timerstate2");
- t->prev = nil;
- t->next = priv->timers;
- if(t->next)
- t->next->prev = t;
- priv->timers = t;
- }
- }
- t->state = newstate;
- }
- void
- tcpackproc(void *a)
- {
- Tcptimer *t, *tp, *timeo;
- Proto *tcp;
- Tcppriv *priv;
- int loop;
- tcp = a;
- priv = tcp->priv;
- for(;;) {
- tsleep(&up->sleep, return0, 0, MSPTICK);
- qlock(&priv->tl);
- timeo = nil;
- loop = 0;
- for(t = priv->timers; t != nil; t = tp) {
- if(loop++ > 10000)
- panic("tcpackproc1");
- tp = t->next;
- if(t->state == TcptimerON) {
- t->count--;
- if(t->count == 0) {
- timerstate(priv, t, TcptimerDONE);
- t->readynext = timeo;
- timeo = t;
- }
- }
- }
- qunlock(&priv->tl);
- loop = 0;
- for(t = timeo; t != nil; t = t->readynext) {
- if(loop++ > 10000)
- panic("tcpackproc2");
- if(t->state == TcptimerDONE && t->func != nil && !waserror()){
- (*t->func)(t->arg);
- poperror();
- }
- }
- limborexmit(tcp);
- }
- }
- void
- tcpgo(Tcppriv *priv, Tcptimer *t)
- {
- if(t == nil || t->start == 0)
- return;
- qlock(&priv->tl);
- t->count = t->start;
- timerstate(priv, t, TcptimerON);
- qunlock(&priv->tl);
- }
- void
- tcphalt(Tcppriv *priv, Tcptimer *t)
- {
- if(t == nil)
- return;
- qlock(&priv->tl);
- timerstate(priv, t, TcptimerOFF);
- qunlock(&priv->tl);
- }
- int
- backoff(int n)
- {
- if(n < 5)
- return 1 << n;
- return 64;
- }
- void
- localclose(Conv *s, char *reason) /* called with tcb locked */
- {
- Tcpctl *tcb;
- Reseq *rp,*rp1;
- Tcppriv *tpriv;
- tpriv = s->p->priv;
- tcb = (Tcpctl*)s->ptcl;
- iphtrem(&tpriv->ht, s);
- tcphalt(tpriv, &tcb->timer);
- tcphalt(tpriv, &tcb->rtt_timer);
- tcphalt(tpriv, &tcb->acktimer);
- tcphalt(tpriv, &tcb->katimer);
- /* Flush reassembly queue; nothing more can arrive */
- for(rp = tcb->reseq; rp != nil; rp = rp1) {
- rp1 = rp->next;
- freeblist(rp->bp);
- free(rp);
- }
- tcb->reseq = nil;
- if(tcb->state == Syn_sent)
- Fsconnected(s, reason);
- if(s->state == Announced)
- wakeup(&s->listenr);
- qhangup(s->rq, reason);
- qhangup(s->wq, reason);
- tcpsetstate(s, Closed);
- }
- /* mtu (- TCP + IP hdr len) of 1st hop */
- int
- tcpmtu(Proto *tcp, uchar *addr, int version, int *scale)
- {
- Ipifc *ifc;
- int mtu;
- ifc = findipifc(tcp->f, addr, 0);
- switch(version){
- default:
- case V4:
- mtu = DEF_MSS;
- if(ifc != nil)
- mtu = ifc->maxtu - ifc->m->hsize - (TCP4_PKT + TCP4_HDRSIZE);
- break;
- case V6:
- mtu = DEF_MSS6;
- if(ifc != nil)
- mtu = ifc->maxtu - ifc->m->hsize - (TCP6_PKT + TCP6_HDRSIZE);
- break;
- }
- if(ifc != nil){
- if(ifc->mbps > 100)
- *scale = HaveWS | 3;
- else if(ifc->mbps > 10)
- *scale = HaveWS | 1;
- else
- *scale = HaveWS | 0;
- } else
- *scale = HaveWS | 0;
- return mtu;
- }
- void
- inittcpctl(Conv *s, int mode)
- {
- Tcpctl *tcb;
- Tcp4hdr* h4;
- Tcp6hdr* h6;
- int mss;
- tcb = (Tcpctl*)s->ptcl;
- memset(tcb, 0, sizeof(Tcpctl));
- tcb->ssthresh = 65535;
- tcb->srtt = tcp_irtt<<LOGAGAIN;
- tcb->mdev = 0;
- /* setup timers */
- tcb->timer.start = tcp_irtt / MSPTICK;
- tcb->timer.func = tcptimeout;
- tcb->timer.arg = s;
- tcb->rtt_timer.start = MAX_TIME;
- tcb->acktimer.start = TCP_ACK / MSPTICK;
- tcb->acktimer.func = tcpacktimer;
- tcb->acktimer.arg = s;
- tcb->katimer.start = DEF_KAT / MSPTICK;
- tcb->katimer.func = tcpkeepalive;
- tcb->katimer.arg = s;
- mss = DEF_MSS;
- /* create a prototype(pseudo) header */
- if(mode != TCP_LISTEN){
- if(ipcmp(s->laddr, IPnoaddr) == 0)
- findlocalip(s->p->f, s->laddr, s->raddr);
- switch(s->ipversion){
- case V4:
- h4 = &tcb->protohdr.tcp4hdr;
- memset(h4, 0, sizeof(*h4));
- h4->proto = IP_TCPPROTO;
- hnputs(h4->tcpsport, s->lport);
- hnputs(h4->tcpdport, s->rport);
- v6tov4(h4->tcpsrc, s->laddr);
- v6tov4(h4->tcpdst, s->raddr);
- break;
- case V6:
- h6 = &tcb->protohdr.tcp6hdr;
- memset(h6, 0, sizeof(*h6));
- h6->proto = IP_TCPPROTO;
- hnputs(h6->tcpsport, s->lport);
- hnputs(h6->tcpdport, s->rport);
- ipmove(h6->tcpsrc, s->laddr);
- ipmove(h6->tcpdst, s->raddr);
- mss = DEF_MSS6;
- break;
- default:
- panic("inittcpctl: version %d", s->ipversion);
- }
- }
- tcb->mss = tcb->cwind = mss;
- /* default is no window scaling */
- tcb->window = QMAX;
- tcb->rcv.wnd = QMAX;
- tcb->rcv.scale = 0;
- tcb->snd.scale = 0;
- qsetlimit(s->rq, QMAX);
- }
- /*
- * called with s qlocked
- */
- void
- tcpstart(Conv *s, int mode)
- {
- Tcpctl *tcb;
- Tcppriv *tpriv;
- char kpname[KNAMELEN];
- tpriv = s->p->priv;
- if(tpriv->ackprocstarted == 0){
- qlock(&tpriv->apl);
- if(tpriv->ackprocstarted == 0){
- sprint(kpname, "#I%dtcpack", s->p->f->dev);
- kproc(kpname, tcpackproc, s->p);
- tpriv->ackprocstarted = 1;
- }
- qunlock(&tpriv->apl);
- }
- tcb = (Tcpctl*)s->ptcl;
- inittcpctl(s, mode);
- iphtadd(&tpriv->ht, s);
- switch(mode) {
- case TCP_LISTEN:
- tpriv->stats[PassiveOpens]++;
- tcb->flags |= CLONE;
- tcpsetstate(s, Listen);
- break;
- case TCP_CONNECT:
- tpriv->stats[ActiveOpens]++;
- tcb->flags |= ACTIVE;
- tcpsndsyn(s, tcb);
- tcpsetstate(s, Syn_sent);
- tcpoutput(s);
- break;
- }
- }
- static char*
- tcpflag(ushort flag)
- {
- static char buf[128];
- sprint(buf, "%d", flag>>10); /* Head len */
- if(flag & URG)
- strcat(buf, " URG");
- if(flag & ACK)
- strcat(buf, " ACK");
- if(flag & PSH)
- strcat(buf, " PSH");
- if(flag & RST)
- strcat(buf, " RST");
- if(flag & SYN)
- strcat(buf, " SYN");
- if(flag & FIN)
- strcat(buf, " FIN");
- return buf;
- }
- Block *
- htontcp6(Tcp *tcph, Block *data, Tcp6hdr *ph, Tcpctl *tcb)
- {
- int dlen;
- Tcp6hdr *h;
- ushort csum;
- ushort hdrlen, optpad = 0;
- uchar *opt;
- hdrlen = TCP6_HDRSIZE;
- if(tcph->flags & SYN){
- if(tcph->mss)
- hdrlen += MSS_LENGTH;
- if(tcph->ws)
- hdrlen += WS_LENGTH;
- optpad = hdrlen & 3;
- if(optpad)
- optpad = 4 - optpad;
- hdrlen += optpad;
- }
- if(data) {
- dlen = blocklen(data);
- data = padblock(data, hdrlen + TCP6_PKT);
- if(data == nil)
- return nil;
- }
- else {
- dlen = 0;
- data = allocb(hdrlen + TCP6_PKT + 64); /* the 64 pad is to meet mintu's */
- if(data == nil)
- return nil;
- data->wp += hdrlen + TCP6_PKT;
- }
- /* copy in pseudo ip header plus port numbers */
- h = (Tcp6hdr *)(data->rp);
- memmove(h, ph, TCP6_TCBPHDRSZ);
- /* compose pseudo tcp header, do cksum calculation */
- hnputl(h->vcf, hdrlen + dlen);
- h->ploadlen[0] = h->ploadlen[1] = h->proto = 0;
- h->ttl = ph->proto;
- /* copy in variable bits */
- hnputl(h->tcpseq, tcph->seq);
- hnputl(h->tcpack, tcph->ack);
- hnputs(h->tcpflag, (hdrlen<<10) | tcph->flags);
- hnputs(h->tcpwin, tcph->wnd>>(tcb != nil ? tcb->snd.scale : 0));
- hnputs(h->tcpurg, tcph->urg);
- if(tcph->flags & SYN){
- opt = h->tcpopt;
- if(tcph->mss != 0){
- *opt++ = MSSOPT;
- *opt++ = MSS_LENGTH;
- hnputs(opt, tcph->mss);
- opt += 2;
- }
- if(tcph->ws != 0){
- *opt++ = WSOPT;
- *opt++ = WS_LENGTH;
- *opt++ = tcph->ws;
- }
- while(optpad-- > 0)
- *opt++ = NOOPOPT;
- }
- if(tcb != nil && tcb->nochecksum){
- h->tcpcksum[0] = h->tcpcksum[1] = 0;
- } else {
- csum = ptclcsum(data, TCP6_IPLEN, hdrlen+dlen+TCP6_PHDRSIZE);
- hnputs(h->tcpcksum, csum);
- }
- /* move from pseudo header back to normal ip header */
- memset(h->vcf, 0, 4);
- h->vcf[0] = IP_VER6;
- hnputs(h->ploadlen, hdrlen+dlen);
- h->proto = ph->proto;
- return data;
- }
- Block *
- htontcp4(Tcp *tcph, Block *data, Tcp4hdr *ph, Tcpctl *tcb)
- {
- int dlen;
- Tcp4hdr *h;
- ushort csum;
- ushort hdrlen, optpad = 0;
- uchar *opt;
- hdrlen = TCP4_HDRSIZE;
- if(tcph->flags & SYN){
- if(tcph->mss)
- hdrlen += MSS_LENGTH;
- if(tcph->ws)
- hdrlen += WS_LENGTH;
- optpad = hdrlen & 3;
- if(optpad)
- optpad = 4 - optpad;
- hdrlen += optpad;
- }
- if(data) {
- dlen = blocklen(data);
- data = padblock(data, hdrlen + TCP4_PKT);
- if(data == nil)
- return nil;
- }
- else {
- dlen = 0;
- data = allocb(hdrlen + TCP4_PKT + 64); /* the 64 pad is to meet mintu's */
- if(data == nil)
- return nil;
- data->wp += hdrlen + TCP4_PKT;
- }
- /* copy in pseudo ip header plus port numbers */
- h = (Tcp4hdr *)(data->rp);
- memmove(h, ph, TCP4_TCBPHDRSZ);
- /* copy in variable bits */
- hnputs(h->tcplen, hdrlen + dlen);
- hnputl(h->tcpseq, tcph->seq);
- hnputl(h->tcpack, tcph->ack);
- hnputs(h->tcpflag, (hdrlen<<10) | tcph->flags);
- hnputs(h->tcpwin, tcph->wnd>>(tcb != nil ? tcb->snd.scale : 0));
- hnputs(h->tcpurg, tcph->urg);
- if(tcph->flags & SYN){
- opt = h->tcpopt;
- if(tcph->mss != 0){
- *opt++ = MSSOPT;
- *opt++ = MSS_LENGTH;
- hnputs(opt, tcph->mss);
- opt += 2;
- }
- if(tcph->ws != 0){
- *opt++ = WSOPT;
- *opt++ = WS_LENGTH;
- *opt++ = tcph->ws;
- }
- while(optpad-- > 0)
- *opt++ = NOOPOPT;
- }
- if(tcb != nil && tcb->nochecksum){
- h->tcpcksum[0] = h->tcpcksum[1] = 0;
- } else {
- csum = ptclcsum(data, TCP4_IPLEN, hdrlen+dlen+TCP4_PHDRSIZE);
- hnputs(h->tcpcksum, csum);
- }
- return data;
- }
- int
- ntohtcp6(Tcp *tcph, Block **bpp)
- {
- Tcp6hdr *h;
- uchar *optr;
- ushort hdrlen;
- ushort optlen;
- int n;
- *bpp = pullupblock(*bpp, TCP6_PKT+TCP6_HDRSIZE);
- if(*bpp == nil)
- return -1;
- h = (Tcp6hdr *)((*bpp)->rp);
- tcph->source = nhgets(h->tcpsport);
- tcph->dest = nhgets(h->tcpdport);
- tcph->seq = nhgetl(h->tcpseq);
- tcph->ack = nhgetl(h->tcpack);
- hdrlen = (h->tcpflag[0]>>2) & ~3;
- if(hdrlen < TCP6_HDRSIZE) {
- freeblist(*bpp);
- return -1;
- }
- tcph->flags = h->tcpflag[1];
- tcph->wnd = nhgets(h->tcpwin);
- tcph->urg = nhgets(h->tcpurg);
- tcph->mss = 0;
- tcph->ws = 0;
- tcph->len = nhgets(h->ploadlen) - hdrlen;
- *bpp = pullupblock(*bpp, hdrlen+TCP6_PKT);
- if(*bpp == nil)
- return -1;
- optr = h->tcpopt;
- n = hdrlen - TCP6_HDRSIZE;
- while(n > 0 && *optr != EOLOPT) {
- if(*optr == NOOPOPT) {
- n--;
- optr++;
- continue;
- }
- optlen = optr[1];
- if(optlen < 2 || optlen > n)
- break;
- switch(*optr) {
- case MSSOPT:
- if(optlen == MSS_LENGTH)
- tcph->mss = nhgets(optr+2);
- break;
- case WSOPT:
- if(optlen == WS_LENGTH && *(optr+2) <= 14)
- tcph->ws = HaveWS | *(optr+2);
- break;
- }
- n -= optlen;
- optr += optlen;
- }
- return hdrlen;
- }
- int
- ntohtcp4(Tcp *tcph, Block **bpp)
- {
- Tcp4hdr *h;
- uchar *optr;
- ushort hdrlen;
- ushort optlen;
- int n;
- *bpp = pullupblock(*bpp, TCP4_PKT+TCP4_HDRSIZE);
- if(*bpp == nil)
- return -1;
- h = (Tcp4hdr *)((*bpp)->rp);
- tcph->source = nhgets(h->tcpsport);
- tcph->dest = nhgets(h->tcpdport);
- tcph->seq = nhgetl(h->tcpseq);
- tcph->ack = nhgetl(h->tcpack);
- hdrlen = (h->tcpflag[0]>>2) & ~3;
- if(hdrlen < TCP4_HDRSIZE) {
- freeblist(*bpp);
- return -1;
- }
- tcph->flags = h->tcpflag[1];
- tcph->wnd = nhgets(h->tcpwin);
- tcph->urg = nhgets(h->tcpurg);
- tcph->mss = 0;
- tcph->ws = 0;
- tcph->len = nhgets(h->length) - (hdrlen + TCP4_PKT);
- *bpp = pullupblock(*bpp, hdrlen+TCP4_PKT);
- if(*bpp == nil)
- return -1;
- optr = h->tcpopt;
- n = hdrlen - TCP4_HDRSIZE;
- while(n > 0 && *optr != EOLOPT) {
- if(*optr == NOOPOPT) {
- n--;
- optr++;
- continue;
- }
- optlen = optr[1];
- if(optlen < 2 || optlen > n)
- break;
- switch(*optr) {
- case MSSOPT:
- if(optlen == MSS_LENGTH)
- tcph->mss = nhgets(optr+2);
- break;
- case WSOPT:
- if(optlen == WS_LENGTH && *(optr+2) <= 14)
- tcph->ws = HaveWS | *(optr+2);
- break;
- }
- n -= optlen;
- optr += optlen;
- }
- return hdrlen;
- }
- /*
- * For outgiing calls, generate an initial sequence
- * number and put a SYN on the send queue
- */
- void
- tcpsndsyn(Conv *s, Tcpctl *tcb)
- {
- tcb->iss = (nrand(1<<16)<<16)|nrand(1<<16);
- tcb->rttseq = tcb->iss;
- tcb->snd.wl2 = tcb->iss;
- tcb->snd.una = tcb->iss;
- tcb->snd.ptr = tcb->rttseq;
- tcb->snd.nxt = tcb->rttseq;
- tcb->flgcnt++;
- tcb->flags |= FORCE;
- tcb->sndsyntime = NOW;
- /* set desired mss and scale */
- tcb->mss = tcpmtu(s->p, s->laddr, s->ipversion, &tcb->scale);
- }
- void
- sndrst(Proto *tcp, uchar *source, uchar *dest, ushort length, Tcp *seg, uchar version, char *reason)
- {
- Block *hbp;
- uchar rflags;
- Tcppriv *tpriv;
- Tcp4hdr ph4;
- Tcp6hdr ph6;
- netlog(tcp->f, Logtcp, "sndrst: %s", reason);
- tpriv = tcp->priv;
- if(seg->flags & RST)
- return;
- /* make pseudo header */
- switch(version) {
- case V4:
- memset(&ph4, 0, sizeof(ph4));
- ph4.vihl = IP_VER4;
- v6tov4(ph4.tcpsrc, dest);
- v6tov4(ph4.tcpdst, source);
- ph4.proto = IP_TCPPROTO;
- hnputs(ph4.tcplen, TCP4_HDRSIZE);
- hnputs(ph4.tcpsport, seg->dest);
- hnputs(ph4.tcpdport, seg->source);
- break;
- case V6:
- memset(&ph6, 0, sizeof(ph6));
- ph6.vcf[0] = IP_VER6;
- ipmove(ph6.tcpsrc, dest);
- ipmove(ph6.tcpdst, source);
- ph6.proto = IP_TCPPROTO;
- hnputs(ph6.ploadlen, TCP6_HDRSIZE);
- hnputs(ph6.tcpsport, seg->dest);
- hnputs(ph6.tcpdport, seg->source);
- break;
- default:
- panic("sndrst: version %d", version);
- }
- tpriv->stats[OutRsts]++;
- rflags = RST;
- /* convince the other end that this reset is in band */
- if(seg->flags & ACK) {
- seg->seq = seg->ack;
- seg->ack = 0;
- }
- else {
- rflags |= ACK;
- seg->ack = seg->seq;
- seg->seq = 0;
- if(seg->flags & SYN)
- seg->ack++;
- seg->ack += length;
- if(seg->flags & FIN)
- seg->ack++;
- }
- seg->flags = rflags;
- seg->wnd = 0;
- seg->urg = 0;
- seg->mss = 0;
- seg->ws = 0;
- switch(version) {
- case V4:
- hbp = htontcp4(seg, nil, &ph4, nil);
- if(hbp == nil)
- return;
- ipoput4(tcp->f, hbp, 0, MAXTTL, DFLTTOS, nil);
- break;
- case V6:
- hbp = htontcp6(seg, nil, &ph6, nil);
- if(hbp == nil)
- return;
- ipoput6(tcp->f, hbp, 0, MAXTTL, DFLTTOS, nil);
- break;
- default:
- panic("sndrst2: version %d", version);
- }
- }
- /*
- * send a reset to the remote side and close the conversation
- * called with s qlocked
- */
- char*
- tcphangup(Conv *s)
- {
- Tcp seg;
- Tcpctl *tcb;
- Block *hbp;
- tcb = (Tcpctl*)s->ptcl;
- if(waserror())
- return commonerror();
- if(s->raddr != 0) {
- if(!waserror()){
- seg.flags = RST | ACK;
- seg.ack = tcb->rcv.nxt;
- tcb->rcv.una = 0;
- seg.seq = tcb->snd.ptr;
- seg.wnd = 0;
- seg.urg = 0;
- seg.mss = 0;
- seg.ws = 0;
- switch(s->ipversion) {
- case V4:
- tcb->protohdr.tcp4hdr.vihl = IP_VER4;
- hbp = htontcp4(&seg, nil, &tcb->protohdr.tcp4hdr, tcb);
- ipoput4(s->p->f, hbp, 0, s->ttl, s->tos, s);
- break;
- case V6:
- tcb->protohdr.tcp6hdr.vcf[0] = IP_VER6;
- hbp = htontcp6(&seg, nil, &tcb->protohdr.tcp6hdr, tcb);
- ipoput6(s->p->f, hbp, 0, s->ttl, s->tos, s);
- break;
- default:
- panic("tcphangup: version %d", s->ipversion);
- }
- poperror();
- }
- }
- localclose(s, nil);
- poperror();
- return nil;
- }
- /*
- * (re)send a SYN ACK
- */
- int
- sndsynack(Proto *tcp, Limbo *lp)
- {
- Block *hbp;
- Tcp4hdr ph4;
- Tcp6hdr ph6;
- Tcp seg;
- int scale;
- /* make pseudo header */
- switch(lp->version) {
- case V4:
- memset(&ph4, 0, sizeof(ph4));
- ph4.vihl = IP_VER4;
- v6tov4(ph4.tcpsrc, lp->laddr);
- v6tov4(ph4.tcpdst, lp->raddr);
- ph4.proto = IP_TCPPROTO;
- hnputs(ph4.tcplen, TCP4_HDRSIZE);
- hnputs(ph4.tcpsport, lp->lport);
- hnputs(ph4.tcpdport, lp->rport);
- break;
- case V6:
- memset(&ph6, 0, sizeof(ph6));
- ph6.vcf[0] = IP_VER6;
- ipmove(ph6.tcpsrc, lp->laddr);
- ipmove(ph6.tcpdst, lp->raddr);
- ph6.proto = IP_TCPPROTO;
- hnputs(ph6.ploadlen, TCP6_HDRSIZE);
- hnputs(ph6.tcpsport, lp->lport);
- hnputs(ph6.tcpdport, lp->rport);
- break;
- default:
- panic("sndrst: version %d", lp->version);
- }
- seg.seq = lp->iss;
- seg.ack = lp->irs+1;
- seg.flags = SYN|ACK;
- seg.urg = 0;
- seg.mss = tcpmtu(tcp, lp->laddr, lp->version, &scale);
- seg.wnd = QMAX;
- /* if the other side set scale, we should too */
- if(lp->rcvscale){
- seg.ws = scale;
- lp->sndscale = scale;
- } else {
- seg.ws = 0;
- lp->sndscale = 0;
- }
- switch(lp->version) {
- case V4:
- hbp = htontcp4(&seg, nil, &ph4, nil);
- if(hbp == nil)
- return -1;
- ipoput4(tcp->f, hbp, 0, MAXTTL, DFLTTOS, nil);
- break;
- case V6:
- hbp = htontcp6(&seg, nil, &ph6, nil);
- if(hbp == nil)
- return -1;
- ipoput6(tcp->f, hbp, 0, MAXTTL, DFLTTOS, nil);
- break;
- default:
- panic("sndsnack: version %d", lp->version);
- }
- lp->lastsend = NOW;
- return 0;
- }
- #define hashipa(a, p) ( ( (a)[IPaddrlen-2] + (a)[IPaddrlen-1] + p )&LHTMASK )
- /*
- * put a call into limbo and respond with a SYN ACK
- *
- * called with proto locked
- */
- static void
- limbo(Conv *s, uchar *source, uchar *dest, Tcp *seg, int version)
- {
- Limbo *lp, **l;
- Tcppriv *tpriv;
- int h;
- tpriv = s->p->priv;
- h = hashipa(source, seg->source);
- for(l = &tpriv->lht[h]; *l != nil; l = &lp->next){
- lp = *l;
- if(lp->lport != seg->dest || lp->rport != seg->source || lp->version != version)
- continue;
- if(ipcmp(lp->raddr, source) != 0)
- continue;
- if(ipcmp(lp->laddr, dest) != 0)
- continue;
- /* each new SYN restarts the retransmits */
- lp->irs = seg->seq;
- break;
- }
- lp = *l;
- if(lp == nil){
- if(tpriv->nlimbo >= Maxlimbo && tpriv->lht[h]){
- lp = tpriv->lht[h];
- tpriv->lht[h] = lp->next;
- lp->next = nil;
- } else {
- lp = malloc(sizeof(*lp));
- if(lp == nil)
- return;
- tpriv->nlimbo++;
- }
- *l = lp;
- lp->version = version;
- ipmove(lp->laddr, dest);
- ipmove(lp->raddr, source);
- lp->lport = seg->dest;
- lp->rport = seg->source;
- lp->mss = seg->mss;
- lp->rcvscale = seg->ws;
- lp->irs = seg->seq;
- lp->iss = (nrand(1<<16)<<16)|nrand(1<<16);
- }
- if(sndsynack(s->p, lp) < 0){
- *l = lp->next;
- tpriv->nlimbo--;
- free(lp);
- }
- }
- /*
- * resend SYN ACK's once every SYNACK_RXTIMER ms.
- */
- static void
- limborexmit(Proto *tcp)
- {
- Tcppriv *tpriv;
- Limbo **l, *lp;
- int h;
- int seen;
- ulong now;
- tpriv = tcp->priv;
- if(!canqlock(tcp))
- return;
- seen = 0;
- now = NOW;
- for(h = 0; h < NLHT && seen < tpriv->nlimbo; h++){
- for(l = &tpriv->lht[h]; *l != nil && seen < tpriv->nlimbo; ){
- lp = *l;
- seen++;
- if(now - lp->lastsend < (lp->rexmits+1)*SYNACK_RXTIMER)
- continue;
- /* time it out after 1 second */
- if(++(lp->rexmits) > 5){
- tpriv->nlimbo--;
- *l = lp->next;
- free(lp);
- continue;
- }
- /* if we're being attacked, don't bother resending SYN ACK's */
- if(tpriv->nlimbo > 100)
- continue;
- if(sndsynack(tcp, lp) < 0){
- tpriv->nlimbo--;
- *l = lp->next;
- free(lp);
- continue;
- }
- l = &lp->next;
- }
- }
- qunlock(tcp);
- }
- /*
- * lookup call in limbo. if found, throw it out.
- *
- * called with proto locked
- */
- static void
- limborst(Conv *s, Tcp *segp, uchar *src, uchar *dst, uchar version)
- {
- Limbo *lp, **l;
- int h;
- Tcppriv *tpriv;
- tpriv = s->p->priv;
- /* find a call in limbo */
- h = hashipa(src, segp->source);
- for(l = &tpriv->lht[h]; *l != nil; l = &lp->next){
- lp = *l;
- if(lp->lport != segp->dest || lp->rport != segp->source || lp->version != version)
- continue;
- if(ipcmp(lp->laddr, dst) != 0)
- continue;
- if(ipcmp(lp->raddr, src) != 0)
- continue;
- /* RST can only follow the SYN */
- if(segp->seq == lp->irs+1){
- tpriv->nlimbo--;
- *l = lp->next;
- free(lp);
- }
- break;
- }
- }
- /*
- * come here when we finally get an ACK to our SYN-ACK.
- * lookup call in limbo. if found, create a new conversation
- *
- * called with proto locked
- */
- static Conv*
- tcpincoming(Conv *s, Tcp *segp, uchar *src, uchar *dst, uchar version)
- {
- Conv *new;
- Tcpctl *tcb;
- Tcppriv *tpriv;
- Tcp4hdr *h4;
- Tcp6hdr *h6;
- Limbo *lp, **l;
- int h;
- /* unless it's just an ack, it can't be someone coming out of limbo */
- if((segp->flags & SYN) || (segp->flags & ACK) == 0)
- return nil;
- tpriv = s->p->priv;
- /* find a call in limbo */
- h = hashipa(src, segp->source);
- for(l = &tpriv->lht[h]; (lp = *l) != nil; l = &lp->next){
- netlog(s->p->f, Logtcp, "tcpincoming s %I,%ux/%I,%ux d %I,%ux/%I,%ux v %d/%d",
- src, segp->source, lp->raddr, lp->rport,
- dst, segp->dest, lp->laddr, lp->lport,
- version, lp->version
- );
- if(lp->lport != segp->dest || lp->rport != segp->source || lp->version != version)
- continue;
- if(ipcmp(lp->laddr, dst) != 0)
- continue;
- if(ipcmp(lp->raddr, src) != 0)
- continue;
- /* we're assuming no data with the initial SYN */
- if(segp->seq != lp->irs+1 || segp->ack != lp->iss+1){
- netlog(s->p->f, Logtcp, "tcpincoming s %lux/%lux a %lux %lux",
- segp->seq, lp->irs+1, segp->ack, lp->iss+1);
- lp = nil;
- } else {
- tpriv->nlimbo--;
- *l = lp->next;
- }
- break;
- }
- if(lp == nil)
- return nil;
- new = Fsnewcall(s, src, segp->source, dst, segp->dest, version);
- if(new == nil)
- return nil;
- memmove(new->ptcl, s->ptcl, sizeof(Tcpctl));
- tcb = (Tcpctl*)new->ptcl;
- tcb->flags &= ~CLONE;
- tcb->timer.arg = new;
- tcb->timer.state = TcptimerOFF;
- tcb->acktimer.arg = new;
- tcb->acktimer.state = TcptimerOFF;
- tcb->katimer.arg = new;
- tcb->katimer.state = TcptimerOFF;
- tcb->rtt_timer.arg = new;
- tcb->rtt_timer.state = TcptimerOFF;
- tcb->irs = lp->irs;
- tcb->rcv.nxt = tcb->irs+1;
- tcb->rcv.urg = tcb->rcv.nxt;
- tcb->iss = lp->iss;
- tcb->rttseq = tcb->iss;
- tcb->snd.wl2 = tcb->iss;
- tcb->snd.una = tcb->iss+1;
- tcb->snd.ptr = tcb->iss+1;
- tcb->snd.nxt = tcb->iss+1;
- tcb->flgcnt = 0;
- tcb->flags |= SYNACK;
- /* our sending max segment size cannot be bigger than what he asked for */
- if(lp->mss != 0 && lp->mss < tcb->mss)
- tcb->mss = lp->mss;
- /* window scaling */
- tcpsetscale(new, tcb, lp->rcvscale, lp->sndscale);
- /* the congestion window always starts out as a single segment */
- tcb->snd.wnd = segp->wnd;
- tcb->cwind = tcb->mss;
- /* set initial round trip time */
- tcb->sndsyntime = lp->lastsend+lp->rexmits*SYNACK_RXTIMER;
- tcpsynackrtt(new);
- free(lp);
- /* set up proto header */
- switch(version){
- case V4:
- h4 = &tcb->protohdr.tcp4hdr;
- memset(h4, 0, sizeof(*h4));
- h4->proto = IP_TCPPROTO;
- hnputs(h4->tcpsport, new->lport);
- hnputs(h4->tcpdport, new->rport);
- v6tov4(h4->tcpsrc, dst);
- v6tov4(h4->tcpdst, src);
- break;
- case V6:
- h6 = &tcb->protohdr.tcp6hdr;
- memset(h6, 0, sizeof(*h6));
- h6->proto = IP_TCPPROTO;
- hnputs(h6->tcpsport, new->lport);
- hnputs(h6->tcpdport, new->rport);
- ipmove(h6->tcpsrc, dst);
- ipmove(h6->tcpdst, src);
- break;
- default:
- panic("tcpincoming: version %d", new->ipversion);
- }
- tcpsetstate(new, Established);
- iphtadd(&tpriv->ht, new);
- return new;
- }
- int
- seq_within(ulong x, ulong low, ulong high)
- {
- if(low <= high){
- if(low <= x && x <= high)
- return 1;
- }
- else {
- if(x >= low || x <= high)
- return 1;
- }
- return 0;
- }
- int
- seq_lt(ulong x, ulong y)
- {
- return (int)(x-y) < 0;
- }
- int
- seq_le(ulong x, ulong y)
- {
- return (int)(x-y) <= 0;
- }
- int
- seq_gt(ulong x, ulong y)
- {
- return (int)(x-y) > 0;
- }
- int
- seq_ge(ulong x, ulong y)
- {
- return (int)(x-y) >= 0;
- }
- /*
- * use the time between the first SYN and it's ack as the
- * initial round trip time
- */
- void
- tcpsynackrtt(Conv *s)
- {
- Tcpctl *tcb;
- int delta;
- Tcppriv *tpriv;
- tcb = (Tcpctl*)s->ptcl;
- tpriv = s->p->priv;
- delta = NOW - tcb->sndsyntime;
- tcb->srtt = delta<<LOGAGAIN;
- tcb->mdev = delta<<LOGDGAIN;
- /* halt round trip timer */
- tcphalt(tpriv, &tcb->rtt_timer);
- }
- void
- update(Conv *s, Tcp *seg)
- {
- int rtt, delta;
- Tcpctl *tcb;
- ulong acked;
- ulong expand;
- Tcppriv *tpriv;
- tpriv = s->p->priv;
- tcb = (Tcpctl*)s->ptcl;
- /* if everything has been acked, force output(?) */
- if(seq_gt(seg->ack, tcb->snd.nxt)) {
- tcb->flags |= FORCE;
- return;
- }
- /* added by Dong Lin for fast retransmission */
- if(seg->ack == tcb->snd.una
- && tcb->snd.una != tcb->snd.nxt
- && seg->len == 0
- && seg->wnd == tcb->snd.wnd) {
- /* this is a pure ack w/o window update */
- netlog(s->p->f, Logtcprxmt, "dupack %lud ack %lud sndwnd %d advwin %d\n",
- tcb->snd.dupacks, seg->ack, tcb->snd.wnd, seg->wnd);
- if(++tcb->snd.dupacks == TCPREXMTTHRESH) {
- /*
- * tahoe tcp rxt the packet, half sshthresh,
- * and set cwnd to one packet
- */
- tcb->snd.recovery = 1;
- tcb->snd.rxt = tcb->snd.nxt;
- netlog(s->p->f, Logtcprxmt, "fast rxt %lud, nxt %lud\n", tcb->snd.una, tcb->snd.nxt);
- tcprxmit(s);
- } else {
- /* do reno tcp here. */
- }
- }
- /*
- * update window
- */
- if( seq_gt(seg->ack, tcb->snd.wl2)
- || (tcb->snd.wl2 == seg->ack && seg->wnd > tcb->snd.wnd)){
- tcb->snd.wnd = seg->wnd;
- tcb->snd.wl2 = seg->ack;
- }
- if(!seq_gt(seg->ack, tcb->snd.una)){
- /*
- * don't let us hangup if sending into a closed window and
- * we're still getting acks
- */
- if((tcb->flags&RETRAN) && tcb->snd.wnd == 0){
- tcb->backedoff = MAXBACKMS/4;
- }
- return;
- }
- /*
- * any positive ack turns off fast rxt,
- * (should we do new-reno on partial acks?)
- */
- if(!tcb->snd.recovery || seq_ge(seg->ack, tcb->snd.rxt)) {
- tcb->snd.dupacks = 0;
- tcb->snd.recovery = 0;
- } else
- netlog(s->p->f, Logtcp, "rxt next %lud, cwin %ud\n", seg->ack, tcb->cwind);
- /* Compute the new send window size */
- acked = seg->ack - tcb->snd.una;
- /* avoid slow start and timers for SYN acks */
- if((tcb->flags & SYNACK) == 0) {
- tcb->flags |= SYNACK;
- acked--;
- tcb->flgcnt--;
- goto done;
- }
- /* slow start as long as we're not recovering from lost packets */
- if(tcb->cwind < tcb->snd.wnd && !tcb->snd.recovery) {
- if(tcb->cwind < tcb->ssthresh) {
- expand = tcb->mss;
- if(acked < expand)
- expand = acked;
- }
- else
- expand = ((int)tcb->mss * tcb->mss) / tcb->cwind;
- if(tcb->cwind + expand < tcb->cwind)
- expand = tcb->snd.wnd - tcb->cwind;
- if(tcb->cwind + expand > tcb->snd.wnd)
- expand = tcb->snd.wnd - tcb->cwind;
- if(expand != 0)
- tcb->cwind += expand;
- }
- /* Adjust the timers according to the round trip time */
- if(tcb->rtt_timer.state == TcptimerON && seq_ge(seg->ack, tcb->rttseq)) {
- tcphalt(tpriv, &tcb->rtt_timer);
- if((tcb->flags&RETRAN) == 0) {
- tcb->backoff = 0;
- tcb->backedoff = 0;
- rtt = tcb->rtt_timer.start - tcb->rtt_timer.count;
- if(rtt == 0)
- rtt = 1; /* otherwise all close systems will rexmit in 0 time */
- rtt *= MSPTICK;
- if(tcb->srtt == 0) {
- tcb->srtt = rtt << LOGAGAIN;
- tcb->mdev = rtt << LOGDGAIN;
- } else {
- delta = rtt - (tcb->srtt>>LOGAGAIN);
- tcb->srtt += delta;
- if(tcb->srtt <= 0)
- tcb->srtt = 1;
- delta = abs(delta) - (tcb->mdev>>LOGDGAIN);
- tcb->mdev += delta;
- if(tcb->mdev <= 0)
- tcb->mdev = 1;
- }
- tcpsettimer(tcb);
- }
- }
- done:
- if(qdiscard(s->wq, acked) < acked)
- tcb->flgcnt--;
- tcb->snd.una = seg->ack;
- if(seq_gt(seg->ack, tcb->snd.urg))
- tcb->snd.urg = seg->ack;
- if(tcb->snd.una != tcb->snd.nxt)
- tcpgo(tpriv, &tcb->timer);
- else
- tcphalt(tpriv, &tcb->timer);
- if(seq_lt(tcb->snd.ptr, tcb->snd.una))
- tcb->snd.ptr = tcb->snd.una;
- tcb->flags &= ~RETRAN;
- tcb->backoff = 0;
- tcb->backedoff = 0;
- }
- void
- tcpiput(Proto *tcp, Ipifc*, Block *bp)
- {
- Tcp seg;
- Tcp4hdr *h4;
- Tcp6hdr *h6;
- int hdrlen;
- Tcpctl *tcb;
- ushort length;
- uchar source[IPaddrlen], dest[IPaddrlen];
- Conv *s;
- Fs *f;
- Tcppriv *tpriv;
- uchar version;
- f = tcp->f;
- tpriv = tcp->priv;
-
- tpriv->stats[InSegs]++;
- h4 = (Tcp4hdr*)(bp->rp);
- h6 = (Tcp6hdr*)(bp->rp);
- if((h4->vihl&0xF0)==IP_VER4) {
- version = V4;
- length = nhgets(h4->length);
- v4tov6(dest, h4->tcpdst);
- v4tov6(source, h4->tcpsrc);
- h4->Unused = 0;
- hnputs(h4->tcplen, length-TCP4_PKT);
- if(!(bp->flag & Btcpck) && (h4->tcpcksum[0] || h4->tcpcksum[1]) &&
- ptclcsum(bp, TCP4_IPLEN, length-TCP4_IPLEN)) {
- tpriv->stats[CsumErrs]++;
- tpriv->stats[InErrs]++;
- netlog(f, Logtcp, "bad tcp proto cksum\n");
- freeblist(bp);
- return;
- }
- hdrlen = ntohtcp4(&seg, &bp);
- if(hdrlen < 0){
- tpriv->stats[HlenErrs]++;
- tpriv->stats[InErrs]++;
- netlog(f, Logtcp, "bad tcp hdr len\n");
- return;
- }
- /* trim the packet to the size claimed by the datagram */
- length -= hdrlen+TCP4_PKT;
- bp = trimblock(bp, hdrlen+TCP4_PKT, length);
- if(bp == nil){
- tpriv->stats[LenErrs]++;
- tpriv->stats[InErrs]++;
- netlog(f, Logtcp, "tcp len < 0 after trim\n");
- return;
- }
- }
- else {
- int ttl = h6->ttl;
- int proto = h6->proto;
- version = V6;
- length = nhgets(h6->ploadlen);
- ipmove(dest, h6->tcpdst);
- ipmove(source, h6->tcpsrc);
- h6->ploadlen[0] = h6->ploadlen[1] = h6->proto = 0;
- h6->ttl = proto;
- hnputl(h6->vcf, length);
- if((h6->tcpcksum[0] || h6->tcpcksum[1]) &&
- ptclcsum(bp, TCP6_IPLEN, length+TCP6_PHDRSIZE)) {
- tpriv->stats[CsumErrs]++;
- tpriv->stats[InErrs]++;
- netlog(f, Logtcp, "bad tcp proto cksum\n");
- freeblist(bp);
- return;
- }
- h6->ttl = ttl;
- h6->proto = proto;
- hnputs(h6->ploadlen, length);
- hdrlen = ntohtcp6(&seg, &bp);
- if(hdrlen < 0){
- tpriv->stats[HlenErrs]++;
- tpriv->stats[InErrs]++;
- netlog(f, Logtcp, "bad tcp hdr len\n");
- return;
- }
- /* trim the packet to the size claimed by the datagram */
- length -= hdrlen;
- bp = trimblock(bp, hdrlen+TCP6_PKT, length);
- if(bp == nil){
- tpriv->stats[LenErrs]++;
- tpriv->stats[InErrs]++;
- netlog(f, Logtcp, "tcp len < 0 after trim\n");
- return;
- }
- }
- /* lock protocol while searching for a conversation */
- qlock(tcp);
- /* Look for a matching conversation */
- s = iphtlook(&tpriv->ht, source, seg.source, dest, seg.dest);
- if(s == nil){
- netlog(f, Logtcp, "iphtlook failed");
- reset:
- qunlock(tcp);
- sndrst(tcp, source, dest, length, &seg, version, "no conversation");
- freeblist(bp);
- return;
- }
- /* if it's a listener, look for the right flags and get a new conv */
- tcb = (Tcpctl*)s->ptcl;
- if(tcb->state == Listen){
- if(seg.flags & RST){
- limborst(s, &seg, source, dest, version);
- qunlock(tcp);
- freeblist(bp);
- return;
- }
- /* if this is a new SYN, put the call into limbo */
- if((seg.flags & SYN) && (seg.flags & ACK) == 0){
- limbo(s, source, dest, &seg, version);
- qunlock(tcp);
- freeblist(bp);
- return;
- }
- /*
- * if there's a matching call in limbo, tcpincoming will
- * return it in state Syn_received
- */
- s = tcpincoming(s, &seg, source, dest, version);
- if(s == nil)
- goto reset;
- }
- /* The rest of the input state machine is run with the control block
- * locked and implements the state machine directly out of the RFC.
- * Out-of-band data is ignored - it was always a bad idea.
- */
- tcb = (Tcpctl*)s->ptcl;
- if(waserror()){
- qunlock(s);
- nexterror();
- }
- qlock(s);
- qunlock(tcp);
- /* fix up window */
- seg.wnd <<= tcb->rcv.scale;
- if(tcb->kacounter > 0)
- tcpsetkacounter(tcb);
- switch(tcb->state) {
- case Closed:
- sndrst(tcp, source, dest, length, &seg, version, "sending to Closed");
- goto raise;
- case Syn_sent:
- if(seg.flags & ACK) {
- if(!seq_within(seg.ack, tcb->iss+1, tcb->snd.nxt)) {
- sndrst(tcp, source, dest, length, &seg, version,
- "bad seq in Syn_sent");
- goto raise;
- }
- }
- if(seg.flags & RST) {
- if(seg.flags & ACK)
- localclose(s, Econrefused);
- goto raise;
- }
- if(seg.flags & SYN) {
- procsyn(s, &seg);
- if(seg.flags & ACK){
- update(s, &seg);
- tcpsynackrtt(s);
- tcpsetstate(s, Established);
- tcpsetscale(s, tcb, seg.ws, tcb->scale);
- }
- else {
- tcb->time = NOW;
- tcpsetstate(s, Syn_received); /* DLP - shouldn't this be a reset? */
- }
- if(length != 0 || (seg.flags & FIN))
- break;
- freeblist(bp);
- goto output;
- }
- else
- freeblist(bp);
- qunlock(s);
- poperror();
- return;
- case Syn_received:
- /* doesn't matter if it's the correct ack, we're just trying to set timing */
- if(seg.flags & ACK)
- tcpsynackrtt(s);
- break;
- }
- /*
- * One DOS attack is to open connections to us and then forget about them,
- * thereby tying up a conv at no long term cost to the attacker.
- * This is an attempt to defeat these stateless DOS attacks. See
- * corresponding code in tcpsendka().
- */
- if(tcb->state != Syn_received && (seg.flags & RST) == 0){
- if(tcpporthogdefense
- && seq_within(seg.ack, tcb->snd.una-(1<<31), tcb->snd.una-(1<<29))){
- print("stateless hog %I.%d->%I.%d f %ux %lux - %lux - %lux\n",
- source, seg.source, dest, seg.dest, seg.flags,
- tcb->snd.una-(1<<31), seg.ack, tcb->snd.una-(1<<29));
- localclose(s, "stateless hog");
- }
- }
- /* Cut the data to fit the receive window */
- if(tcptrim(tcb, &seg, &bp, &length) == -1) {
- netlog(f, Logtcp, "tcp len < 0, %lux\n", seg.seq);
- update(s, &seg);
- if(qlen(s->wq)+tcb->flgcnt == 0 && tcb->state == Closing) {
- tcphalt(tpriv, &tcb->rtt_timer);
- tcphalt(tpriv, &tcb->acktimer);
- tcphalt(tpriv, &tcb->katimer);
- tcpsetstate(s, Time_wait);
- tcb->timer.start = MSL2*(1000 / MSPTICK);
- tcpgo(tpriv, &tcb->timer);
- }
- if(!(seg.flags & RST)) {
- tcb->flags |= FORCE;
- goto output;
- }
- qunlock(s);
- poperror();
- return;
- }
- /* Cannot accept so answer with a rst */
- if(length && tcb->state == Closed) {
- sndrst(tcp, source, dest, length, &seg, version, "sending to Closed");
- goto raise;
- }
- /* The segment is beyond the current receive pointer so
- * queue the data in the resequence queue
- */
- if(seg.seq != tcb->rcv.nxt)
- if(length != 0 || (seg.flags & (SYN|FIN))) {
- update(s, &seg);
- if(addreseq(tcb, tpriv, &seg, bp, length) < 0)
- print("reseq %I.%d -> %I.%d\n", s->raddr, s->rport, s->laddr, s->lport);
- tcb->flags |= FORCE;
- goto output;
- }
- /*
- * keep looping till we've processed this packet plus any
- * adjacent packets in the resequence queue
- */
- for(;;) {
- if(seg.flags & RST) {
- if(tcb->state == Established)
- tpriv->stats[EstabResets]++;
- localclose(s, Econrefused);
- goto raise;
- }
- if((seg.flags&ACK) == 0)
- goto raise;
- switch(tcb->state) {
- case Syn_received:
- if(!seq_within(seg.ack, tcb->snd.una+1, tcb->snd.nxt)){
- sndrst(tcp, source, dest, length, &seg, version,
- "bad seq in Syn_received");
- goto raise;
- }
- update(s, &seg);
- tcpsetstate(s, Established);
- case Established:
- case Close_wait:
- update(s, &seg);
- break;
- case Finwait1:
- update(s, &seg);
- if(qlen(s->wq)+tcb->flgcnt == 0){
- tcphalt(tpriv, &tcb->rtt_timer);
- tcphalt(tpriv, &tcb->acktimer);
- tcpsetkacounter(tcb);
- tcb->time = NOW;
- tcpsetstate(s, Finwait2);
- tcb->katimer.start = MSL2 * (1000 / MSPTICK);
- tcpgo(tpriv, &tcb->katimer);
- }
- break;
- case Finwait2:
- update(s, &seg);
- break;
- case Closing:
- update(s, &seg);
- if(qlen(s->wq)+tcb->flgcnt == 0) {
- tcphalt(tpriv, &tcb->rtt_timer);
- tcphalt(tpriv, &tcb->acktimer);
- tcphalt(tpriv, &tcb->katimer);
- tcpsetstate(s, Time_wait);
- tcb->timer.start = MSL2*(1000 / MSPTICK);
- tcpgo(tpriv, &tcb->timer);
- }
- break;
- case Last_ack:
- update(s, &seg);
- if(qlen(s->wq)+tcb->flgcnt == 0) {
- localclose(s, nil);
- goto raise;
- }
- case Time_wait:
- tcb->flags |= FORCE;
- if(tcb->timer.state != TcptimerON)
- tcpgo(tpriv, &tcb->timer);
- }
- if((seg.flags&URG) && seg.urg) {
- if(seq_gt(seg.urg + seg.seq, tcb->rcv.urg)) {
- tcb->rcv.urg = seg.urg + seg.seq;
- pullblock(&bp, seg.urg);
- }
- }
- else
- if(seq_gt(tcb->rcv.nxt, tcb->rcv.urg))
- tcb->rcv.urg = tcb->rcv.nxt;
- if(length == 0) {
- if(bp != nil)
- freeblist(bp);
- }
- else {
- switch(tcb->state){
- default:
- /* Ignore segment text */
- if(bp != nil)
- freeblist(bp);
- break;
- case Syn_received:
- case Established:
- case Finwait1:
- /* If we still have some data place on
- * receive queue
- */
- if(bp) {
- bp = packblock(bp);
- if(bp == nil)
- panic("tcp packblock");
- qpassnolim(s->rq, bp);
- bp = nil;
- /*
- * Force an ack every 2 data messages. This is
- * a hack for rob to make his home system run
- * faster.
- *
- * this also keeps the standard TCP congestion
- * control working since it needs an ack every
- * 2 max segs worth. This is not quite that,
- * but under a real stream is equivalent since
- * every packet has a max seg in it.
- */
- if(++(tcb->rcv.una) >= 2)
- tcb->flags |= FORCE;
- }
- tcb->rcv.nxt += length;
- /*
- * update our rcv window
- */
- tcprcvwin(s);
- /*
- * turn on the acktimer if there's something
- * to ack
- */
- if(tcb->acktimer.state != TcptimerON)
- tcpgo(tpriv, &tcb->acktimer);
- break;
- case Finwait2:
- /* no process to read the data, send a reset */
- if(bp != nil)
- freeblist(bp);
- sndrst(tcp, source, dest, length, &seg, version,
- "send to Finwait2");
- qunlock(s);
- poperror();
- return;
- }
- }
- if(seg.flags & FIN) {
- tcb->flags |= FORCE;
- switch(tcb->state) {
- case Syn_received:
- case Established:
- tcb->rcv.nxt++;
- tcpsetstate(s, Close_wait);
- break;
- case Finwait1:
- tcb->rcv.nxt++;
- if(qlen(s->wq)+tcb->flgcnt == 0) {
- tcphalt(tpriv, &tcb->rtt_timer);
- tcphalt(tpriv, &tcb->acktimer);
- tcphalt(tpriv, &tcb->katimer);
- tcpsetstate(s, Time_wait);
- tcb->timer.start = MSL2*(1000/MSPTICK);
- tcpgo(tpriv, &tcb->timer);
- }
- else
- tcpsetstate(s, Closing);
- break;
- case Finwait2:
- tcb->rcv.nxt++;
- tcphalt(tpriv, &tcb->rtt_timer);
- tcphalt(tpriv, &tcb->acktimer);
- tcphalt(tpriv, &tcb->katimer);
- tcpsetstate(s, Time_wait);
- tcb->timer.start = MSL2 * (1000/MSPTICK);
- tcpgo(tpriv, &tcb->timer);
- break;
- case Close_wait:
- case Closing:
- case Last_ack:
- break;
- case Time_wait:
- tcpgo(tpriv, &tcb->timer);
- break;
- }
- }
- /*
- * get next adjacent segment from the resequence queue.
- * dump/trim any overlapping segments
- */
- for(;;) {
- if(tcb->reseq == nil)
- goto output;
- if(seq_ge(tcb->rcv.nxt, tcb->reseq->seg.seq) == 0)
- goto output;
- getreseq(tcb, &seg, &bp, &length);
- if(tcptrim(tcb, &seg, &bp, &length) == 0)
- break;
- }
- }
- output:
- tcpoutput(s);
- qunlock(s);
- poperror();
- return;
- raise:
- qunlock(s);
- poperror();
- freeblist(bp);
- tcpkick(s);
- }
- /*
- * always enters and exits with the s locked. We drop
- * the lock to ipoput the packet so some care has to be
- * taken by callers.
- */
- void
- tcpoutput(Conv *s)
- {
- Tcp seg;
- int msgs;
- Tcpctl *tcb;
- Block *hbp, *bp;
- int sndcnt, n;
- ulong ssize, dsize, usable, sent;
- Fs *f;
- Tcppriv *tpriv;
- uchar version;
- f = s->p->f;
- tpriv = s->p->priv;
- version = s->ipversion;
- for(msgs = 0; msgs < 100; msgs++) {
- tcb = (Tcpctl*)s->ptcl;
-
- switch(tcb->state) {
- case Listen:
- case Closed:
- case Finwait2:
- return;
- }
-
- /* force an ack when a window has opened up */
- if(tcb->rcv.blocked && tcb->rcv.wnd > 0){
- tcb->rcv.blocked = 0;
- tcb->flags |= FORCE;
- }
-
- sndcnt = qlen(s->wq)+tcb->flgcnt;
- sent = tcb->snd.ptr - tcb->snd.una;
- /* Don't send anything else until our SYN has been acked */
- if(tcb->snd.ptr != tcb->iss && (tcb->flags & SYNACK) == 0)
- break;
- /* Compute usable segment based on offered window and limit
- * window probes to one
- */
- if(tcb->snd.wnd == 0){
- if(sent != 0) {
- if((tcb->flags&FORCE) == 0)
- break;
- // tcb->snd.ptr = tcb->snd.una;
- }
- usable = 1;
- }
- else {
- usable = tcb->cwind;
- if(tcb->snd.wnd < usable)
- usable = tcb->snd.wnd;
- usable -= sent;
- }
- ssize = sndcnt-sent;
- if(ssize && usable < 2)
- netlog(s->p->f, Logtcp, "throttled snd.wnd %lud cwind %lud\n",
- tcb->snd.wnd, tcb->cwind);
- if(usable < ssize)
- ssize = usable;
- if(tcb->mss < ssize)
- ssize = tcb->mss;
- dsize = ssize;
- seg.urg = 0;
- if(ssize == 0)
- if((tcb->flags&FORCE) == 0)
- break;
- tcb->flags &= ~FORCE;
- tcprcvwin(s);
- /* By default we will generate an ack */
- tcphalt(tpriv, &tcb->acktimer);
- tcb->rcv.una = 0;
- seg.source = s->lport;
- seg.dest = s->rport;
- seg.flags = ACK;
- seg.mss = 0;
- seg.ws = 0;
- switch(tcb->state){
- case Syn_sent:
- seg.flags = 0;
- if(tcb->snd.ptr == tcb->iss){
- seg.flags |= SYN;
- dsize--;
- seg.mss = tcb->mss;
- seg.ws = tcb->scale;
- }
- break;
- case Syn_received:
- /*
- * don't send any data with a SYN/ACK packet
- * because Linux rejects the packet in its
- * attempt to solve the SYN attack problem
- */
- if(tcb->snd.ptr == tcb->iss){
- seg.flags |= SYN;
- dsize = 0;
- ssize = 1;
- seg.mss = tcb->mss;
- seg.ws = tcb->scale;
- }
- break;
- }
- seg.seq = tcb->snd.ptr;
- seg.ack = tcb->rcv.nxt;
- seg.wnd = tcb->rcv.wnd;
- /* Pull out data to send */
- bp = nil;
- if(dsize != 0) {
- bp = qcopy(s->wq, dsize, sent);
- if(BLEN(bp) != dsize) {
- seg.flags |= FIN;
- dsize--;
- }
- }
- if(sent+dsize == sndcnt)
- seg.flags |= PSH;
- /* keep track of balance of resent data */
- if(seq_lt(tcb->snd.ptr, tcb->snd.nxt)) {
- n = tcb->snd.nxt - tcb->snd.ptr;
- if(ssize < n)
- n = ssize;
- tcb->resent += n;
- netlog(f, Logtcp, "rexmit: %I.%d -> %I.%d ptr %lux nxt %lux\n",
- s->raddr, s->rport, s->laddr, s->lport, tcb->snd.ptr, tcb->snd.nxt);
- tpriv->stats[RetransSegs]++;
- }
- tcb->snd.ptr += ssize;
- /* Pull up the send pointer so we can accept acks
- * for this window
- */
- if(seq_gt(tcb->snd.ptr,tcb->snd.nxt))
- tcb->snd.nxt = tcb->snd.ptr;
- /* Build header, link data and compute cksum */
- switch(version){
- case V4:
- tcb->protohdr.tcp4hdr.vihl = IP_VER4;
- hbp = htontcp4(&seg, bp, &tcb->protohdr.tcp4hdr, tcb);
- if(hbp == nil) {
- freeblist(bp);
- return;
- }
- break;
- case V6:
- tcb->protohdr.tcp6hdr.vcf[0] = IP_VER6;
- hbp = htontcp6(&seg, bp, &tcb->protohdr.tcp6hdr, tcb);
- if(hbp == nil) {
- freeblist(bp);
- return;
- }
- break;
- default:
- hbp = nil; /* to suppress a warning */
- panic("tcpoutput: version %d", version);
- }
- /* Start the transmission timers if there is new data and we
- * expect acknowledges
- */
- if(ssize != 0){
- if(tcb->timer.state != TcptimerON)
- tcpgo(tpriv, &tcb->timer);
- /* If round trip timer isn't running, start it.
- * measure the longest packet only in case the
- * transmission time dominates RTT
- */
- if(tcb->rtt_timer.state != TcptimerON)
- if(ssize == tcb->mss) {
- tcpgo(tpriv, &tcb->rtt_timer);
- tcb->rttseq = tcb->snd.ptr;
- }
- }
- tpriv->stats[OutSegs]++;
- if(tcb->kacounter > 0)
- tcpgo(tpriv, &tcb->katimer);
- switch(version){
- case V4:
- if(ipoput4(f, hbp, 0, s->ttl, s->tos, s) < 0){
- /* a negative return means no route */
- localclose(s, "no route");
- }
- break;
- case V6:
- if(ipoput6(f, hbp, 0, s->ttl, s->tos, s) < 0){
- /* a negative return means no route */
- localclose(s, "no route");
- }
- break;
- default:
- panic("tcpoutput2: version %d", version);
- }
- if((msgs%4) == 1){
- qunlock(s);
- sched();
- qlock(s);
- }
- }
- }
- /*
- * the BSD convention (hack?) for keep alives. resend last uchar acked.
- */
- void
- tcpsendka(Conv *s)
- {
- Tcp seg;
- Tcpctl *tcb;
- Block *hbp,*dbp;
- tcb = (Tcpctl*)s->ptcl;
- dbp = nil;
- seg.urg = 0;
- seg.source = s->lport;
- seg.dest = s->rport;
- seg.flags = ACK|PSH;
- seg.mss = 0;
- seg.ws = 0;
- if(tcpporthogdefense)
- seg.seq = tcb->snd.una-(1<<30)-nrand(1<<20);
- else
- seg.seq = tcb->snd.una-1;
- seg.ack = tcb->rcv.nxt;
- tcb->rcv.una = 0;
- seg.wnd = tcb->rcv.wnd;
- if(tcb->state == Finwait2){
- seg.flags |= FIN;
- } else {
- dbp = allocb(1);
- dbp->wp++;
- }
- if(isv4(s->raddr)) {
- /* Build header, link data and compute cksum */
- tcb->protohdr.tcp4hdr.vihl = IP_VER4;
- hbp = htontcp4(&seg, dbp, &tcb->protohdr.tcp4hdr, tcb);
- if(hbp == nil) {
- freeblist(dbp);
- return;
- }
- ipoput4(s->p->f, hbp, 0, s->ttl, s->tos, s);
- }
- else {
- /* Build header, link data and compute cksum */
- tcb->protohdr.tcp6hdr.vcf[0] = IP_VER6;
- hbp = htontcp6(&seg, dbp, &tcb->protohdr.tcp6hdr, tcb);
- if(hbp == nil) {
- freeblist(dbp);
- return;
- }
- ipoput6(s->p->f, hbp, 0, s->ttl, s->tos, s);
- }
- }
- /*
- * if we've timed out, close the connection
- * otherwise, send a keepalive and restart the timer
- */
- void
- tcpsetkacounter(Tcpctl *tcb)
- {
- tcb->kacounter = (12 * 60 * 1000) / (tcb->katimer.start*MSPTICK);
- if(tcb->kacounter < 3)
- tcb->kacounter = 3;
- }
- void
- tcpkeepalive(void *v)
- {
- Tcpctl *tcb;
- Conv *s;
- s = v;
- tcb = (Tcpctl*)s->ptcl;
- if(waserror()){
- qunlock(s);
- nexterror();
- }
- qlock(s);
- if(tcb->state != Closed){
- if(--(tcb->kacounter) <= 0) {
- localclose(s, Etimedout);
- } else {
- tcpsendka(s);
- tcpgo(s->p->priv, &tcb->katimer);
- }
- }
- qunlock(s);
- poperror();
- }
- /*
- * start keepalive timer
- */
- char*
- tcpstartka(Conv *s, char **f, int n)
- {
- Tcpctl *tcb;
- int x;
- tcb = (Tcpctl*)s->ptcl;
- if(n > 1){
- x = atoi(f[1]);
- if(x >= MSPTICK)
- tcb->katimer.start = x/MSPTICK;
- }
- tcpsetkacounter(tcb);
- tcpgo(s->p->priv, &tcb->katimer);
- return nil;
- }
- /*
- * turn checksums on/off
- */
- char*
- tcpsetchecksum(Conv *s, char **f, int)
- {
- Tcpctl *tcb;
- tcb = (Tcpctl*)s->ptcl;
- tcb->nochecksum = !atoi(f[1]);
- return nil;
- }
- void
- tcprxmit(Conv *s)
- {
- Tcpctl *tcb;
- tcb = (Tcpctl*)s->ptcl;
- tcb->flags |= RETRAN|FORCE;
- tcb->snd.ptr = tcb->snd.una;
- /*
- * We should be halving the slow start threshhold (down to one
- * mss) but leaving it at mss seems to work well enough
- */
- tcb->ssthresh = tcb->mss;
- /*
- * pull window down to a single packet
- */
- tcb->cwind = tcb->mss;
- tcpoutput(s);
- }
- void
- tcptimeout(void *arg)
- {
- Conv *s;
- Tcpctl *tcb;
- int maxback;
- Tcppriv *tpriv;
- s = (Conv*)arg;
- tpriv = s->p->priv;
- tcb = (Tcpctl*)s->ptcl;
- if(waserror()){
- qunlock(s);
- nexterror();
- }
- qlock(s);
- switch(tcb->state){
- default:
- tcb->backoff++;
- if(tcb->state == Syn_sent)
- maxback = MAXBACKMS/2;
- else
- maxback = MAXBACKMS;
- tcb->backedoff += tcb->timer.start * MSPTICK;
- if(tcb->backedoff >= maxback) {
- localclose(s, Etimedout);
- break;
- }
- netlog(s->p->f, Logtcprxmt, "timeout rexmit 0x%lux %d/%d\n", tcb->snd.una, tcb->timer.start, NOW);
- tcpsettimer(tcb);
- tcprxmit(s);
- tpriv->stats[RetransTimeouts]++;
- tcb->snd.dupacks = 0;
- break;
- case Time_wait:
- localclose(s, nil);
- break;
- case Closed:
- break;
- }
- qunlock(s);
- poperror();
- }
- int
- inwindow(Tcpctl *tcb, int seq)
- {
- return seq_within(seq, tcb->rcv.nxt, tcb->rcv.nxt+tcb->rcv.wnd-1);
- }
- /*
- * set up state for a received SYN (or SYN ACK) packet
- */
- void
- procsyn(Conv *s, Tcp *seg)
- {
- Tcpctl *tcb;
- tcb = (Tcpctl*)s->ptcl;
- tcb->flags |= FORCE;
- tcb->rcv.nxt = seg->seq + 1;
- tcb->rcv.urg = tcb->rcv.nxt;
- tcb->irs = seg->seq;
- /* our sending max segment size cannot be bigger than what he asked for */
- if(seg->mss != 0 && seg->mss < tcb->mss)
- tcb->mss = seg->mss;
- /* the congestion window always starts out as a single segment */
- tcb->snd.wnd = seg->wnd;
- tcb->cwind = tcb->mss;
- }
- int
- addreseq(Tcpctl *tcb, Tcppriv *tpriv, Tcp *seg, Block *bp, ushort length)
- {
- Reseq *rp, *rp1;
- int i;
- static int once;
- rp = malloc(sizeof(Reseq));
- if(rp == nil){
- freeblist(bp); /* bp always consumed by add_reseq */
- return 0;
- }
- rp->seg = *seg;
- rp->bp = bp;
- rp->length = length;
- /* Place on reassembly list sorting by starting seq number */
- rp1 = tcb->reseq;
- if(rp1 == nil || seq_lt(seg->seq, rp1->seg.seq)) {
- rp->next = rp1;
- tcb->reseq = rp;
- if(rp->next != nil)
- tpriv->stats[OutOfOrder]++;
- return 0;
- }
- length = 0;
- for(i = 0;; i++) {
- length += rp1->length;
- if(rp1->next == nil || seq_lt(seg->seq, rp1->next->seg.seq)) {
- rp->next = rp1->next;
- rp1->next = rp;
- if(rp->next != nil)
- tpriv->stats[OutOfOrder]++;
- break;
- }
- rp1 = rp1->next;
- }
- if(length > QMAX && once++ == 0){
- print("very long tcp resequence queue: %d\n", length);
- for(rp1 = tcb->reseq, i = 0; i < 10 && rp1 != nil; rp1 = rp1->next, i++)
- print("0x%lux 0x%lux 0x%ux\n", rp1->seg.seq, rp1->seg.ack,
- rp1->seg.flags);
- return -1;
- }
- return 0;
- }
- void
- getreseq(Tcpctl *tcb, Tcp *seg, Block **bp, ushort *length)
- {
- Reseq *rp;
- rp = tcb->reseq;
- if(rp == nil)
- return;
- tcb->reseq = rp->next;
- *seg = rp->seg;
- *bp = rp->bp;
- *length = rp->length;
- free(rp);
- }
- int
- tcptrim(Tcpctl *tcb, Tcp *seg, Block **bp, ushort *length)
- {
- ushort len;
- uchar accept;
- int dupcnt, excess;
- accept = 0;
- len = *length;
- if(seg->flags & SYN)
- len++;
- if(seg->flags & FIN)
- len++;
- if(tcb->rcv.wnd == 0) {
- if(len == 0 && seg->seq == tcb->rcv.nxt)
- return 0;
- }
- else {
- /* Some part of the segment should be in the window */
- if(inwindow(tcb,seg->seq))
- accept++;
- else
- if(len != 0) {
- if(inwindow(tcb, seg->seq+len-1) ||
- seq_within(tcb->rcv.nxt, seg->seq,seg->seq+len-1))
- accept++;
- }
- }
- if(!accept) {
- freeblist(*bp);
- return -1;
- }
- dupcnt = tcb->rcv.nxt - seg->seq;
- if(dupcnt > 0){
- tcb->rerecv += dupcnt;
- if(seg->flags & SYN){
- seg->flags &= ~SYN;
- seg->seq++;
- if(seg->urg > 1)
- seg->urg--;
- else
- seg->flags &= ~URG;
- dupcnt--;
- }
- if(dupcnt > 0){
- pullblock(bp, (ushort)dupcnt);
- seg->seq += dupcnt;
- *length -= dupcnt;
- if(seg->urg > dupcnt)
- seg->urg -= dupcnt;
- else {
- seg->flags &= ~URG;
- seg->urg = 0;
- }
- }
- }
- excess = seg->seq + *length - (tcb->rcv.nxt + tcb->rcv.wnd);
- if(excess > 0) {
- tcb->rerecv += excess;
- *length -= excess;
- *bp = trimblock(*bp, 0, *length);
- if(*bp == nil)
- panic("presotto is a boofhead");
- seg->flags &= ~FIN;
- }
- return 0;
- }
- void
- tcpadvise(Proto *tcp, Block *bp, char *msg)
- {
- Tcp4hdr *h4;
- Tcp6hdr *h6;
- Tcpctl *tcb;
- uchar source[IPaddrlen];
- uchar dest[IPaddrlen];
- ushort psource, pdest;
- Conv *s, **p;
- h4 = (Tcp4hdr*)(bp->rp);
- h6 = (Tcp6hdr*)(bp->rp);
- if((h4->vihl&0xF0)==IP_VER4) {
- v4tov6(dest, h4->tcpdst);
- v4tov6(source, h4->tcpsrc);
- psource = nhgets(h4->tcpsport);
- pdest = nhgets(h4->tcpdport);
- }
- else {
- ipmove(dest, h6->tcpdst);
- ipmove(source, h6->tcpsrc);
- psource = nhgets(h6->tcpsport);
- pdest = nhgets(h6->tcpdport);
- }
- /* Look for a connection */
- qlock(tcp);
- for(p = tcp->conv; *p; p++) {
- s = *p;
- tcb = (Tcpctl*)s->ptcl;
- if(s->rport == pdest)
- if(s->lport == psource)
- if(tcb->state != Closed)
- if(ipcmp(s->raddr, dest) == 0)
- if(ipcmp(s->laddr, source) == 0){
- qlock(s);
- qunlock(tcp);
- switch(tcb->state){
- case Syn_sent:
- localclose(s, msg);
- break;
- }
- qunlock(s);
- freeblist(bp);
- return;
- }
- }
- qunlock(tcp);
- freeblist(bp);
- }
- static char*
- tcpporthogdefensectl(char *val)
- {
- if(strcmp(val, "on") == 0)
- tcpporthogdefense = 1;
- else if(strcmp(val, "off") == 0)
- tcpporthogdefense = 0;
- else
- return "unknown value for tcpporthogdefense";
- return nil;
- }
- /* called with c qlocked */
- char*
- tcpctl(Conv* c, char** f, int n)
- {
- if(n == 1 && strcmp(f[0], "hangup") == 0)
- return tcphangup(c);
- if(n >= 1 && strcmp(f[0], "keepalive") == 0)
- return tcpstartka(c, f, n);
- if(n >= 1 && strcmp(f[0], "checksum") == 0)
- return tcpsetchecksum(c, f, n);
- if(n >= 1 && strcmp(f[0], "tcpporthogdefense") == 0)
- return tcpporthogdefensectl(f[1]);
- return "unknown control request";
- }
- int
- tcpstats(Proto *tcp, char *buf, int len)
- {
- Tcppriv *priv;
- char *p, *e;
- int i;
- priv = tcp->priv;
- p = buf;
- e = p+len;
- for(i = 0; i < Nstats; i++)
- p = seprint(p, e, "%s: %lud\n", statnames[i], priv->stats[i]);
- return p - buf;
- }
- /*
- * garbage collect any stale conversations:
- * - SYN received but no SYN-ACK after 5 seconds (could be the SYN attack)
- * - Finwait2 after 5 minutes
- *
- * this is called whenever we run out of channels. Both checks are
- * of questionable validity so we try to use them only when we're
- * up against the wall.
- */
- int
- tcpgc(Proto *tcp)
- {
- Conv *c, **pp, **ep;
- int n;
- Tcpctl *tcb;
- n = 0;
- ep = &tcp->conv[tcp->nc];
- for(pp = tcp->conv; pp < ep; pp++) {
- c = *pp;
- if(c == nil)
- break;
- if(!canqlock(c))
- continue;
- tcb = (Tcpctl*)c->ptcl;
- switch(tcb->state){
- case Syn_received:
- if(NOW - tcb->time > 5000){
- localclose(c, "timed out");
- n++;
- }
- break;
- case Finwait2:
- if(NOW - tcb->time > 5*60*1000){
- localclose(c, "timed out");
- n++;
- }
- break;
- }
- qunlock(c);
- }
- return n;
- }
- void
- tcpsettimer(Tcpctl *tcb)
- {
- int x;
- /* round trip depenency */
- x = backoff(tcb->backoff) *
- (tcb->mdev + (tcb->srtt>>LOGAGAIN) + MSPTICK) / MSPTICK;
- /* bounded twixt 1/2 and 10 seconds */
- if(x < 500/MSPTICK)
- x = 500/MSPTICK;
- else if(x > (10000/MSPTICK))
- x = 10000/MSPTICK;
- tcb->timer.start = x;
- }
- void
- tcpinit(Fs *fs)
- {
- Proto *tcp;
- Tcppriv *tpriv;
- tcp = smalloc(sizeof(Proto));
- tpriv = tcp->priv = smalloc(sizeof(Tcppriv));
- tcp->name = "tcp";
- tcp->connect = tcpconnect;
- tcp->announce = tcpannounce;
- tcp->ctl = tcpctl;
- tcp->state = tcpstate;
- tcp->create = tcpcreate;
- tcp->close = tcpclose;
- tcp->rcv = tcpiput;
- tcp->advise = tcpadvise;
- tcp->stats = tcpstats;
- tcp->inuse = tcpinuse;
- tcp->gc = tcpgc;
- tcp->ipproto = IP_TCPPROTO;
- tcp->nc = scalednconv();
- tcp->ptclsize = sizeof(Tcpctl);
- tpriv->stats[MaxConn] = tcp->nc;
- Fsproto(fs, tcp);
- }
- void
- tcpsetscale(Conv *s, Tcpctl *tcb, ushort rcvscale, ushort sndscale)
- {
- if(rcvscale){
- tcb->rcv.scale = rcvscale & 0xff;
- tcb->snd.scale = sndscale & 0xff;
- tcb->window = QMAX<<tcb->snd.scale;
- qsetlimit(s->rq, tcb->window);
- } else {
- tcb->rcv.scale = 0;
- tcb->snd.scale = 0;
- tcb->window = QMAX;
- qsetlimit(s->rq, tcb->window);
- }
- }
|