123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897389838993900390139023903390439053906390739083909391039113912391339143915391639173918391939203921392239233924392539263927392839293930393139323933393439353936393739383939394039413942394339443945394639473948394939503951395239533954395539563957395839593960396139623963396439653966396739683969397039713972397339743975397639773978397939803981398239833984398539863987398839893990399139923993399439953996399739983999400040014002400340044005400640074008400940104011401240134014401540164017401840194020402140224023402440254026402740284029403040314032403340344035403640374038403940404041404240434044404540464047404840494050405140524053405440554056405740584059406040614062406340644065406640674068406940704071407240734074407540764077407840794080408140824083408440854086408740884089409040914092409340944095409640974098409941004101410241034104410541064107410841094110411141124113411441154116411741184119412041214122412341244125412641274128412941304131413241334134413541364137413841394140414141424143414441454146414741484149415041514152415341544155415641574158415941604161416241634164416541664167416841694170417141724173417441754176417741784179418041814182418341844185418641874188418941904191419241934194419541964197419841994200420142024203420442054206420742084209421042114212421342144215421642174218421942204221422242234224422542264227422842294230423142324233423442354236423742384239424042414242424342444245424642474248424942504251425242534254425542564257425842594260426142624263426442654266426742684269427042714272427342744275427642774278427942804281428242834284428542864287428842894290429142924293429442954296429742984299430043014302430343044305430643074308430943104311431243134314431543164317431843194320432143224323432443254326432743284329433043314332433343344335433643374338433943404341434243434344434543464347434843494350435143524353435443554356435743584359436043614362436343644365436643674368436943704371437243734374437543764377437843794380438143824383438443854386438743884389439043914392439343944395439643974398439944004401440244034404440544064407440844094410441144124413441444154416441744184419442044214422442344244425442644274428442944304431443244334434443544364437443844394440444144424443444444454446444744484449445044514452445344544455445644574458445944604461446244634464446544664467446844694470447144724473447444754476447744784479448044814482448344844485448644874488448944904491449244934494449544964497449844994500450145024503450445054506450745084509451045114512451345144515451645174518451945204521452245234524452545264527452845294530453145324533453445354536453745384539454045414542454345444545454645474548454945504551455245534554455545564557455845594560456145624563456445654566456745684569457045714572457345744575/*
 *   Copyright (C) International Business Machines Corp., 2000-2004
 *
 *   This program is free software;  you can redistribute it and/or modify
 *   it under the terms of the GNU General Public License as published by
 *   the Free Software Foundation; either version 2 of the License, or
 *   (at your option) any later version.
 *
 *   This program is distributed in the hope that it will be useful,
 *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
 *   the GNU General Public License for more details.
 *
 *   You should have received a copy of the GNU General Public License
 *   along with this program;  if not, write to the Free Software
 *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
 */

/*
 *	jfs_dtree.c: directory B+-tree manager
 *
 * B+-tree with variable length key directory:
 *
 * each directory page is structured as an array of 32-byte
 * directory entry slots initialized as a freelist
 * to avoid search/compaction of free space at insertion.
 * when an entry is inserted, a number of slots are allocated
 * from the freelist as required to store variable length data
 * of the entry; when the entry is deleted, slots of the entry
 * are returned to freelist.
 *
 * leaf entry stores full name as key and file serial number
 * (aka inode number) as data.
 * internal/router entry stores sufffix compressed name
 * as key and simple extent descriptor as data.
 *
 * each directory page maintains a sorted entry index table
 * which stores the start slot index of sorted entries
 * to allow binary search on the table.
 *
 * directory starts as a root/leaf page in on-disk inode
 * inline data area.
 * when it becomes full, it starts a leaf of a external extent
 * of length of 1 block. each time the first leaf becomes full,
 * it is extended rather than split (its size is doubled),
 * until its length becoms 4 KBytes, from then the extent is split
 * with new 4 Kbyte extent when it becomes full
 * to reduce external fragmentation of small directories.
 *
 * blah, blah, blah, for linear scan of directory in pieces by
 * readdir().
 *
 *
 *	case-insensitive directory file system
 *
 * names are stored in case-sensitive way in leaf entry.
 * but stored, searched and compared in case-insensitive (uppercase) order
 * (i.e., both search key and entry key are folded for search/compare):
 * (note that case-sensitive order is BROKEN in storage, e.g.,
 *  sensitive: Ad, aB, aC, aD -> insensitive: aB, aC, aD, Ad
 *
 *  entries which folds to the same key makes up a equivalent class
 *  whose members are stored as contiguous cluster (may cross page boundary)
 *  but whose order is arbitrary and acts as duplicate, e.g.,
 *  abc, Abc, aBc, abC)
 *
 * once match is found at leaf, requires scan forward/backward
 * either for, in case-insensitive search, duplicate
 * or for, in case-sensitive search, for exact match
 *
 * router entry must be created/stored in case-insensitive way
 * in internal entry:
 * (right most key of left page and left most key of right page
 * are folded, and its suffix compression is propagated as router
 * key in parent)
 * (e.g., if split occurs <abc> and <aBd>, <ABD> trather than <aB>
 * should be made the router key for the split)
 *
 * case-insensitive search:
 *
 *	fold search key;
 *
 *	case-insensitive search of B-tree:
 *	for internal entry, router key is already folded;
 *	for leaf entry, fold the entry key before comparison.
 *
 *	if (leaf entry case-insensitive match found)
 *		if (next entry satisfies case-insensitive match)
 *			return EDUPLICATE;
 *		if (prev entry satisfies case-insensitive match)
 *			return EDUPLICATE;
 *		return match;
 *	else
 *		return no match;
 *
 *	serialization:
 * target directory inode lock is being held on entry/exit
 * of all main directory service routines.
 *
 *	log based recovery:
 */

#include <linux/fs.h>
#include <linux/quotaops.h>
#include <linux/slab.h>
#include "jfs_incore.h"
#include "jfs_superblock.h"
#include "jfs_filsys.h"
#include "jfs_metapage.h"
#include "jfs_dmap.h"
#include "jfs_unicode.h"
#include "jfs_debug.h"

/* dtree split parameter */
struct dtsplit {
	struct metapage *mp;
	s16 index;
	s16 nslot;
	struct component_name *key;
	ddata_t *data;
	struct pxdlist *pxdlist;
};

#define DT_PAGE(IP, MP) BT_PAGE(IP, MP, dtpage_t, i_dtroot)

/* get page buffer for specified block address */
#define DT_GETPAGE(IP, BN, MP, SIZE, P, RC)				\
do {									\
	BT_GETPAGE(IP, BN, MP, dtpage_t, SIZE, P, RC, i_dtroot);	\
	if (!(RC)) {							\
		if (((P)->header.nextindex >				\
		     (((BN) == 0) ? DTROOTMAXSLOT : (P)->header.maxslot)) || \
		    ((BN) && ((P)->header.maxslot > DTPAGEMAXSLOT))) {	\
			BT_PUTPAGE(MP);					\
			jfs_error((IP)->i_sb,				\
				  "DT_GETPAGE: dtree page corrupt\n");	\
			MP = NULL;					\
			RC = -EIO;					\
		}							\
	}								\
} while (0)

/* for consistency */
#define DT_PUTPAGE(MP) BT_PUTPAGE(MP)

#define DT_GETSEARCH(IP, LEAF, BN, MP, P, INDEX) \
	BT_GETSEARCH(IP, LEAF, BN, MP, dtpage_t, P, INDEX, i_dtroot)

/*
 * forward references
 */
static int dtSplitUp(tid_t tid, struct inode *ip,
		     struct dtsplit * split, struct btstack * btstack);

static int dtSplitPage(tid_t tid, struct inode *ip, struct dtsplit * split,
		       struct metapage ** rmpp, dtpage_t ** rpp, pxd_t * rxdp);

static int dtExtendPage(tid_t tid, struct inode *ip,
			struct dtsplit * split, struct btstack * btstack);

static int dtSplitRoot(tid_t tid, struct inode *ip,
		       struct dtsplit * split, struct metapage ** rmpp);

static int dtDeleteUp(tid_t tid, struct inode *ip, struct metapage * fmp,
		      dtpage_t * fp, struct btstack * btstack);

static int dtRelink(tid_t tid, struct inode *ip, dtpage_t * p);

static int dtReadFirst(struct inode *ip, struct btstack * btstack);

static int dtReadNext(struct inode *ip,
		      loff_t * offset, struct btstack * btstack);

static int dtCompare(struct component_name * key, dtpage_t * p, int si);

static int ciCompare(struct component_name * key, dtpage_t * p, int si,
		     int flag);

static void dtGetKey(dtpage_t * p, int i, struct component_name * key,
		     int flag);

static int ciGetLeafPrefixKey(dtpage_t * lp, int li, dtpage_t * rp,
			      int ri, struct component_name * key, int flag);

static void dtInsertEntry(dtpage_t * p, int index, struct component_name * key,
			  ddata_t * data, struct dt_lock **);

static void dtMoveEntry(dtpage_t * sp, int si, dtpage_t * dp,
			struct dt_lock ** sdtlock, struct dt_lock ** ddtlock,
			int do_index);

static void dtDeleteEntry(dtpage_t * p, int fi, struct dt_lock ** dtlock);

static void dtTruncateEntry(dtpage_t * p, int ti, struct dt_lock ** dtlock);

static void dtLinelockFreelist(dtpage_t * p, int m, struct dt_lock ** dtlock);

#define ciToUpper(c)	UniStrupr((c)->name)

/*
 *	read_index_page()
 *
 *	Reads a page of a directory's index table.
 *	Having metadata mapped into the directory inode's address space
 *	presents a multitude of problems.  We avoid this by mapping to
 *	the absolute address space outside of the *_metapage routines
 */
static struct metapage *read_index_page(struct inode *inode, s64 blkno)
{
	int rc;
	s64 xaddr;
	int xflag;
	s32 xlen;

	rc = xtLookup(inode, blkno, 1, &xflag, &xaddr, &xlen, 1);
	if (rc || (xaddr == 0))
		return NULL;

	return read_metapage(inode, xaddr, PSIZE, 1);
}

/*
 *	get_index_page()
 *
 *	Same as get_index_page(), but get's a new page without reading
 */
static struct metapage *get_index_page(struct inode *inode, s64 blkno)
{
	int rc;
	s64 xaddr;
	int xflag;
	s32 xlen;

	rc = xtLookup(inode, blkno, 1, &xflag, &xaddr, &xlen, 1);
	if (rc || (xaddr == 0))
		return NULL;

	return get_metapage(inode, xaddr, PSIZE, 1);
}

/*
 *	find_index()
 *
 *	Returns dtree page containing directory table entry for specified
 *	index and pointer to its entry.
 *
 *	mp must be released by caller.
 */
static struct dir_table_slot *find_index(struct inode *ip, u32 index,
					 struct metapage ** mp, s64 *lblock)
{
	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
	s64 blkno;
	s64 offset;
	int page_offset;
	struct dir_table_slot *slot;
	static int maxWarnings = 10;

	if (index < 2) {
		if (maxWarnings) {
			jfs_warn("find_entry called with index = %d", index);
			maxWarnings--;
		}
		return NULL;
	}

	if (index >= jfs_ip->next_index) {
		jfs_warn("find_entry called with index >= next_index");
		return NULL;
	}

	if (jfs_dirtable_inline(ip)) {
		/*
		 * Inline directory table
		 */
		*mp = NULL;
		slot = &jfs_ip->i_dirtable[index - 2];
	} else {
		offset = (index - 2) * sizeof(struct dir_table_slot);
		page_offset = offset & (PSIZE - 1);
		blkno = ((offset + 1) >> L2PSIZE) <<
		    JFS_SBI(ip->i_sb)->l2nbperpage;

		if (*mp && (*lblock != blkno)) {
			release_metapage(*mp);
			*mp = NULL;
		}
		if (!(*mp)) {
			*lblock = blkno;
			*mp = read_index_page(ip, blkno);
		}
		if (!(*mp)) {
			jfs_err("free_index: error reading directory table");
			return NULL;
		}

		slot =
		    (struct dir_table_slot *) ((char *) (*mp)->data +
					       page_offset);
	}
	return slot;
}

static inline void lock_index(tid_t tid, struct inode *ip, struct metapage * mp,
			      u32 index)
{
	struct tlock *tlck;
	struct linelock *llck;
	struct lv *lv;

	tlck = txLock(tid, ip, mp, tlckDATA);
	llck = (struct linelock *) tlck->lock;

	if (llck->index >= llck->maxcnt)
		llck = txLinelock(llck);
	lv = &llck->lv[llck->index];

	/*
	 *	Linelock slot size is twice the size of directory table
	 *	slot size.  512 entries per page.
	 */
	lv->offset = ((index - 2) & 511) >> 1;
	lv->length = 1;
	llck->index++;
}

/*
 *	add_index()
 *
 *	Adds an entry to the directory index table.  This is used to provide
 *	each directory entry with a persistent index in which to resume
 *	directory traversals
 */
static u32 add_index(tid_t tid, struct inode *ip, s64 bn, int slot)
{
	struct super_block *sb = ip->i_sb;
	struct jfs_sb_info *sbi = JFS_SBI(sb);
	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
	u64 blkno;
	struct dir_table_slot *dirtab_slot;
	u32 index;
	struct linelock *llck;
	struct lv *lv;
	struct metapage *mp;
	s64 offset;
	uint page_offset;
	struct tlock *tlck;
	s64 xaddr;

	ASSERT(DO_INDEX(ip));

	if (jfs_ip->next_index < 2) {
		jfs_warn("add_index: next_index = %d.  Resetting!",
			   jfs_ip->next_index);
		jfs_ip->next_index = 2;
	}

	index = jfs_ip->next_index++;

	if (index <= MAX_INLINE_DIRTABLE_ENTRY) {
		/*
		 * i_size reflects size of index table, or 8 bytes per entry.
		 */
		ip->i_size = (loff_t) (index - 1) << 3;

		/*
		 * dir table fits inline within inode
		 */
		dirtab_slot = &jfs_ip->i_dirtable[index-2];
		dirtab_slot->flag = DIR_INDEX_VALID;
		dirtab_slot->slot = slot;
		DTSaddress(dirtab_slot, bn);

		set_cflag(COMMIT_Dirtable, ip);

		return index;
	}
	if (index == (MAX_INLINE_DIRTABLE_ENTRY + 1)) {
		struct dir_table_slot temp_table[12];

		/*
		 * It's time to move the inline table to an external
		 * page and begin to build the xtree
		 */
		if (dquot_alloc_block(ip, sbi->nbperpage))
			goto clean_up;
		if (dbAlloc(ip, 0, sbi->nbperpage, &xaddr)) {
			dquot_free_block(ip, sbi->nbperpage);
			goto clean_up;
		}

		/*
		 * Save the table, we're going to overwrite it with the
		 * xtree root
		 */
		memcpy(temp_table, &jfs_ip->i_dirtable, sizeof(temp_table));

		/*
		 * Initialize empty x-tree
		 */
		xtInitRoot(tid, ip);

		/*
		 * Add the first block to the xtree
		 */
		if (xtInsert(tid, ip, 0, 0, sbi->nbperpage, &xaddr, 0)) {
			/* This really shouldn't fail */
			jfs_warn("add_index: xtInsert failed!");
			memcpy(&jfs_ip->i_dirtable, temp_table,
			       sizeof (temp_table));
			dbFree(ip, xaddr, sbi->nbperpage);
			dquot_free_block(ip, sbi->nbperpage);
			goto clean_up;
		}
		ip->i_size = PSIZE;

		mp = get_index_page(ip, 0);
		if (!mp) {
			jfs_err("add_index: get_metapage failed!");
			xtTruncate(tid, ip, 0, COMMIT_PWMAP);
			memcpy(&jfs_ip->i_dirtable, temp_table,
			       sizeof (temp_table));
			goto clean_up;
		}
		tlck = txLock(tid, ip, mp, tlckDATA);
		llck = (struct linelock *) & tlck->lock;
		ASSERT(llck->index == 0);
		lv = &llck->lv[0];

		lv->offset = 0;
		lv->length = 6;	/* tlckDATA slot size is 16 bytes */
		llck->index++;

		memcpy(mp->data, temp_table, sizeof(temp_table));

		mark_metapage_dirty(mp);
		release_metapage(mp);

		/*
		 * Logging is now directed by xtree tlocks
		 */
		clear_cflag(COMMIT_Dirtable, ip);
	}

	offset = (index - 2) * sizeof(struct dir_table_slot);
	page_offset = offset & (PSIZE - 1);
	blkno = ((offset + 1) >> L2PSIZE) << sbi->l2nbperpage;
	if (page_offset == 0) {
		/*
		 * This will be the beginning of a new page
		 */
		xaddr = 0;
		if (xtInsert(tid, ip, 0, blkno, sbi->nbperpage, &xaddr, 0)) {
			jfs_warn("add_index: xtInsert failed!");
			goto clean_up;
		}
		ip->i_size += PSIZE;

		if ((mp = get_index_page(ip, blkno)))
			memset(mp->data, 0, PSIZE);	/* Just looks better */
		else
			xtTruncate(tid, ip, offset, COMMIT_PWMAP);
	} else
		mp = read_index_page(ip, blkno);

	if (!mp) {
		jfs_err("add_index: get/read_metapage failed!");
		goto clean_up;
	}

	lock_index(tid, ip, mp, index);

	dirtab_slot =
	    (struct dir_table_slot *) ((char *) mp->data + page_offset);
	dirtab_slot->flag = DIR_INDEX_VALID;
	dirtab_slot->slot = slot;
	DTSaddress(dirtab_slot, bn);

	mark_metapage_dirty(mp);
	release_metapage(mp);

	return index;

      clean_up:

	jfs_ip->next_index--;

	return 0;
}

/*
 *	free_index()
 *
 *	Marks an entry to the directory index table as free.
 */
static void free_index(tid_t tid, struct inode *ip, u32 index, u32 next)
{
	struct dir_table_slot *dirtab_slot;
	s64 lblock;
	struct metapage *mp = NULL;

	dirtab_slot = find_index(ip, index, &mp, &lblock);

	if (!dirtab_slot)
		return;

	dirtab_slot->flag = DIR_INDEX_FREE;
	dirtab_slot->slot = dirtab_slot->addr1 = 0;
	dirtab_slot->addr2 = cpu_to_le32(next);

	if (mp) {
		lock_index(tid, ip, mp, index);
		mark_metapage_dirty(mp);
		release_metapage(mp);
	} else
		set_cflag(COMMIT_Dirtable, ip);
}

/*
 *	modify_index()
 *
 *	Changes an entry in the directory index table
 */
static void modify_index(tid_t tid, struct inode *ip, u32 index, s64 bn,
			 int slot, struct metapage ** mp, s64 *lblock)
{
	struct dir_table_slot *dirtab_slot;

	dirtab_slot = find_index(ip, index, mp, lblock);

	if (!dirtab_slot)
		return;

	DTSaddress(dirtab_slot, bn);
	dirtab_slot->slot = slot;

	if (*mp) {
		lock_index(tid, ip, *mp, index);
		mark_metapage_dirty(*mp);
	} else
		set_cflag(COMMIT_Dirtable, ip);
}

/*
 *	read_index()
 *
 *	reads a directory table slot
 */
static int read_index(struct inode *ip, u32 index,
		     struct dir_table_slot * dirtab_slot)
{
	s64 lblock;
	struct metapage *mp = NULL;
	struct dir_table_slot *slot;

	slot = find_index(ip, index, &mp, &lblock);
	if (!slot) {
		return -EIO;
	}

	memcpy(dirtab_slot, slot, sizeof(struct dir_table_slot));

	if (mp)
		release_metapage(mp);

	return 0;
}

/*
 *	dtSearch()
 *
 * function:
 *	Search for the entry with specified key
 *
 * parameter:
 *
 * return: 0 - search result on stack, leaf page pinned;
 *	   errno - I/O error
 */
int dtSearch(struct inode *ip, struct component_name * key, ino_t * data,
	     struct btstack * btstack, int flag)
{
	int rc = 0;
	int cmp = 1;		/* init for empty page */
	s64 bn;
	struct metapage *mp;
	dtpage_t *p;
	s8 *stbl;
	int base, index, lim;
	struct btframe *btsp;
	pxd_t *pxd;
	int psize = 288;	/* initial in-line directory */
	ino_t inumber;
	struct component_name ciKey;
	struct super_block *sb = ip->i_sb;

	ciKey.name = kmalloc_array(JFS_NAME_MAX + 1, sizeof(wchar_t),
				   GFP_NOFS);
	if (!ciKey.name) {
		rc = -ENOMEM;
		goto dtSearch_Exit2;
	}


	/* uppercase search key for c-i directory */
	UniStrcpy(ciKey.name, key->name);
	ciKey.namlen = key->namlen;

	/* only uppercase if case-insensitive support is on */
	if ((JFS_SBI(sb)->mntflag & JFS_OS2) == JFS_OS2) {
		ciToUpper(&ciKey);
	}
	BT_CLR(btstack);	/* reset stack */

	/* init level count for max pages to split */
	btstack->nsplit = 1;

	/*
	 *	search down tree from root:
	 *
	 * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of
	 * internal page, child page Pi contains entry with k, Ki <= K < Kj.
	 *
	 * if entry with search key K is not found
	 * internal page search find the entry with largest key Ki
	 * less than K which point to the child page to search;
	 * leaf page search find the entry with smallest key Kj
	 * greater than K so that the returned index is the position of
	 * the entry to be shifted right for insertion of new entry.
	 * for empty tree, search key is greater than any key of the tree.
	 *
	 * by convention, root bn = 0.
	 */
	for (bn = 0;;) {
		/* get/pin the page to search */
		DT_GETPAGE(ip, bn, mp, psize, p, rc);
		if (rc)
			goto dtSearch_Exit1;

		/* get sorted entry table of the page */
		stbl = DT_GETSTBL(p);

		/*
		 * binary search with search key K on the current page.
		 */
		for (base = 0, lim = p->header.nextindex; lim; lim >>= 1) {
			index = base + (lim >> 1);

			if (p->header.flag & BT_LEAF) {
				/* uppercase leaf name to compare */
				cmp =
				    ciCompare(&ciKey, p, stbl[index],
					      JFS_SBI(sb)->mntflag);
			} else {
				/* router key is in uppercase */

				cmp = dtCompare(&ciKey, p, stbl[index]);


			}
			if (cmp == 0) {
				/*
				 *	search hit
				 */
				/* search hit - leaf page:
				 * return the entry found
				 */
				if (p->header.flag & BT_LEAF) {
					inumber = le32_to_cpu(
			((struct ldtentry *) & p->slot[stbl[index]])->inumber);

					/*
					 * search for JFS_LOOKUP
					 */
					if (flag == JFS_LOOKUP) {
						*data = inumber;
						rc = 0;
						goto out;
					}

					/*
					 * search for JFS_CREATE
					 */
					if (flag == JFS_CREATE) {
						*data = inumber;
						rc = -EEXIST;
						goto out;
					}

					/*
					 * search for JFS_REMOVE or JFS_RENAME
					 */
					if ((flag == JFS_REMOVE ||
					     flag == JFS_RENAME) &&
					    *data != inumber) {
						rc = -ESTALE;
						goto out;
					}

					/*
					 * JFS_REMOVE|JFS_FINDDIR|JFS_RENAME
					 */
					/* save search result */
					*data = inumber;
					btsp = btstack->top;
					btsp->bn = bn;
					btsp->index = index;
					btsp->mp = mp;

					rc = 0;
					goto dtSearch_Exit1;
				}

				/* search hit - internal page:
				 * descend/search its child page
				 */
				goto getChild;
			}

			if (cmp > 0) {
				base = index + 1;
				--lim;
			}
		}

		/*
		 *	search miss
		 *
		 * base is the smallest index with key (Kj) greater than
		 * search key (K) and may be zero or (maxindex + 1) index.
		 */
		/*
		 * search miss - leaf page
		 *
		 * return location of entry (base) where new entry with
		 * search key K is to be inserted.
		 */
		if (p->header.flag & BT_LEAF) {
			/*
			 * search for JFS_LOOKUP, JFS_REMOVE, or JFS_RENAME
			 */
			if (flag == JFS_LOOKUP || flag == JFS_REMOVE ||
			    flag == JFS_RENAME) {
				rc = -ENOENT;
				goto out;
			}

			/*
			 * search for JFS_CREATE|JFS_FINDDIR:
			 *
			 * save search result
			 */
			*data = 0;
			btsp = btstack->top;
			btsp->bn = bn;
			btsp->index = base;
			btsp->mp = mp;

			rc = 0;
			goto dtSearch_Exit1;
		}

		/*
		 * search miss - internal page
		 *
		 * if base is non-zero, decrement base by one to get the parent
		 * entry of the child page to search.
		 */
		index = base ? base - 1 : base;

		/*
		 * go down to child page
		 */
	      getChild:
		/* update max. number of pages to split */
		if (BT_STACK_FULL(btstack)) {
			/* Something's corrupted, mark filesystem dirty so
			 * chkdsk will fix it.
			 */
			jfs_error(sb, "stack overrun!\n");
			BT_STACK_DUMP(btstack);
			rc = -EIO;
			goto out;
		}
		btstack->nsplit++;

		/* push (bn, index) of the parent page/entry */
		BT_PUSH(btstack, bn, index);

		/* get the child page block number */
		pxd = (pxd_t *) & p->slot[stbl[index]];
		bn = addressPXD(pxd);
		psize = lengthPXD(pxd) << JFS_SBI(ip->i_sb)->l2bsize;

		/* unpin the parent page */
		DT_PUTPAGE(mp);
	}

      out:
	DT_PUTPAGE(mp);

      dtSearch_Exit1:

	kfree(ciKey.name);

      dtSearch_Exit2:

	return rc;
}


/*
 *	dtInsert()
 *
 * function: insert an entry to directory tree
 *
 * parameter:
 *
 * return: 0 - success;
 *	   errno - failure;
 */
int dtInsert(tid_t tid, struct inode *ip,
	 struct component_name * name, ino_t * fsn, struct btstack * btstack)
{
	int rc = 0;
	struct metapage *mp;	/* meta-page buffer */
	dtpage_t *p;		/* base B+-tree index page */
	s64 bn;
	int index;
	struct dtsplit split;	/* split information */
	ddata_t data;
	struct dt_lock *dtlck;
	int n;
	struct tlock *tlck;
	struct lv *lv;

	/*
	 *	retrieve search result
	 *
	 * dtSearch() returns (leaf page pinned, index at which to insert).
	 * n.b. dtSearch() may return index of (maxindex + 1) of
	 * the full page.
	 */
	DT_GETSEARCH(ip, btstack->top, bn, mp, p, index);

	/*
	 *	insert entry for new key
	 */
	if (DO_INDEX(ip)) {
		if (JFS_IP(ip)->next_index == DIREND) {
			DT_PUTPAGE(mp);
			return -EMLINK;
		}
		n = NDTLEAF(name->namlen);
		data.leaf.tid = tid;
		data.leaf.ip = ip;
	} else {
		n = NDTLEAF_LEGACY(name->namlen);
		data.leaf.ip = NULL;	/* signifies legacy directory format */
	}
	data.leaf.ino = *fsn;

	/*
	 *	leaf page does not have enough room for new entry:
	 *
	 *	extend/split the leaf page;
	 *
	 * dtSplitUp() will insert the entry and unpin the leaf page.
	 */
	if (n > p->header.freecnt) {
		split.mp = mp;
		split.index = index;
		split.nslot = n;
		split.key = name;
		split.data = &data;
		rc = dtSplitUp(tid, ip, &split, btstack);
		return rc;
	}

	/*
	 *	leaf page does have enough room for new entry:
	 *
	 *	insert the new data entry into the leaf page;
	 */
	BT_MARK_DIRTY(mp, ip);
	/*
	 * acquire a transaction lock on the leaf page
	 */
	tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
	dtlck = (struct dt_lock *) & tlck->lock;
	ASSERT(dtlck->index == 0);
	lv = & dtlck->lv[0];

	/* linelock header */
	lv->offset = 0;
	lv->length = 1;
	dtlck->index++;

	dtInsertEntry(p, index, name, &data, &dtlck);

	/* linelock stbl of non-root leaf page */
	if (!(p->header.flag & BT_ROOT)) {
		if (dtlck->index >= dtlck->maxcnt)
			dtlck = (struct dt_lock *) txLinelock(dtlck);
		lv = & dtlck->lv[dtlck->index];
		n = index >> L2DTSLOTSIZE;
		lv->offset = p->header.stblindex + n;
		lv->length =
		    ((p->header.nextindex - 1) >> L2DTSLOTSIZE) - n + 1;
		dtlck->index++;
	}

	/* unpin the leaf page */
	DT_PUTPAGE(mp);

	return 0;
}


/*
 *	dtSplitUp()
 *
 * function: propagate insertion bottom up;
 *
 * parameter:
 *
 * return: 0 - success;
 *	   errno - failure;
 *	leaf page unpinned;
 */
static int dtSplitUp(tid_t tid,
	  struct inode *ip, struct dtsplit * split, struct btstack * btstack)
{
	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
	int rc = 0;
	struct metapage *smp;
	dtpage_t *sp;		/* split page */
	struct metapage *rmp;
	dtpage_t *rp;		/* new right page split from sp */
	pxd_t rpxd;		/* new right page extent descriptor */
	struct metapage *lmp;
	dtpage_t *lp;		/* left child page */
	int skip;		/* index of entry of insertion */
	struct btframe *parent;	/* parent page entry on traverse stack */
	s64 xaddr, nxaddr;
	int xlen, xsize;
	struct pxdlist pxdlist;
	pxd_t *pxd;
	struct component_name key = { 0, NULL };
	ddata_t *data = split->data;
	int n;
	struct dt_lock *dtlck;
	struct tlock *tlck;
	struct lv *lv;
	int quota_allocation = 0;

	/* get split page */
	smp = split->mp;
	sp = DT_PAGE(ip, smp);

	key.name = kmalloc_array(JFS_NAME_MAX + 2, sizeof(wchar_t), GFP_NOFS);
	if (!key.name) {
		DT_PUTPAGE(smp);
		rc = -ENOMEM;
		goto dtSplitUp_Exit;
	}

	/*
	 *	split leaf page
	 *
	 * The split routines insert the new entry, and
	 * acquire txLock as appropriate.
	 */
	/*
	 *	split root leaf page:
	 */
	if (sp->header.flag & BT_ROOT) {
		/*
		 * allocate a single extent child page
		 */
		xlen = 1;
		n = sbi->bsize >> L2DTSLOTSIZE;
		n -= (n + 31) >> L2DTSLOTSIZE;	/* stbl size */
		n -= DTROOTMAXSLOT - sp->header.freecnt; /* header + entries */
		if (n <= split->nslot)
			xlen++;
		if ((rc = dbAlloc(ip, 0, (s64) xlen, &xaddr))) {
			DT_PUTPAGE(smp);
			goto freeKeyName;
		}

		pxdlist.maxnpxd = 1;
		pxdlist.npxd = 0;
		pxd = &pxdlist.pxd[0];
		PXDaddress(pxd, xaddr);
		PXDlength(pxd, xlen);
		split->pxdlist = &pxdlist;
		rc = dtSplitRoot(tid, ip, split, &rmp);

		if (rc)
			dbFree(ip, xaddr, xlen);
		else
			DT_PUTPAGE(rmp);

		DT_PUTPAGE(smp);

		if (!DO_INDEX(ip))
			ip->i_size = xlen << sbi->l2bsize;

		goto freeKeyName;
	}

	/*
	 *	extend first leaf page
	 *
	 * extend the 1st extent if less than buffer page size
	 * (dtExtendPage() reurns leaf page unpinned)
	 */
	pxd = &sp->header.self;
	xlen = lengthPXD(pxd);
	xsize = xlen << sbi->l2bsize;
	if (xsize < PSIZE) {
		xaddr = addressPXD(pxd);
		n = xsize >> L2DTSLOTSIZE;
		n -= (n + 31) >> L2DTSLOTSIZE;	/* stbl size */
		if ((n + sp->header.freecnt) <= split->nslot)
			n = xlen + (xlen << 1);
		else
			n = xlen;

		/* Allocate blocks to quota. */
		rc = dquot_alloc_block(ip, n);
		if (rc)
			goto extendOut;
		quota_allocation += n;

		if ((rc = dbReAlloc(sbi->ipbmap, xaddr, (s64) xlen,
				    (s64) n, &nxaddr)))
			goto extendOut;

		pxdlist.maxnpxd = 1;
		pxdlist.npxd = 0;
		pxd = &pxdlist.pxd[0];
		PXDaddress(pxd, nxaddr);
		PXDlength(pxd, xlen + n);
		split->pxdlist = &pxdlist;
		if ((rc = dtExtendPage(tid, ip, split, btstack))) {
			nxaddr = addressPXD(pxd);
			if (xaddr != nxaddr) {
				/* free relocated extent */
				xlen = lengthPXD(pxd);
				dbFree(ip, nxaddr, (s64) xlen);
			} else {
				/* free extended delta */
				xlen = lengthPXD(pxd) - n;
				xaddr = addressPXD(pxd) + xlen;
				dbFree(ip, xaddr, (s64) n);
			}
		} else if (!DO_INDEX(ip))
			ip->i_size = lengthPXD(pxd) << sbi->l2bsize;


	      extendOut:
		DT_PUTPAGE(smp);
		goto freeKeyName;
	}

	/*
	 *	split leaf page <sp> into <sp> and a new right page <rp>.
	 *
	 * return <rp> pinned and its extent descriptor <rpxd>
	 */
	/*
	 * allocate new directory page extent and
	 * new index page(s) to cover page split(s)
	 *
	 * allocation hint: ?
	 */
	n = btstack->nsplit;
	pxdlist.maxnpxd = pxdlist.npxd = 0;
	xlen = sbi->nbperpage;
	for (pxd = pxdlist.pxd; n > 0; n--, pxd++) {
		if ((rc = dbAlloc(ip, 0, (s64) xlen, &xaddr)) == 0) {
			PXDaddress(pxd, xaddr);
			PXDlength(pxd, xlen);
			pxdlist.maxnpxd++;
			continue;
		}

		DT_PUTPAGE(smp);

		/* undo allocation */
		goto splitOut;
	}

	split->pxdlist = &pxdlist;
	if ((rc = dtSplitPage(tid, ip, split, &rmp, &rp, &rpxd))) {
		DT_PUTPAGE(smp);

		/* undo allocation */
		goto splitOut;
	}

	if (!DO_INDEX(ip))
		ip->i_size += PSIZE;

	/*
	 * propagate up the router entry for the leaf page just split
	 *
	 * insert a router entry for the new page into the parent page,
	 * propagate the insert/split up the tree by walking back the stack
	 * of (bn of parent page, index of child page entry in parent page)
	 * that were traversed during the search for the page that split.
	 *
	 * the propagation of insert/split up the tree stops if the root
	 * splits or the page inserted into doesn't have to split to hold
	 * the new entry.
	 *
	 * the parent entry for the split page remains the same, and
	 * a new entry is inserted at its right with the first key and
	 * block number of the new right page.
	 *
	 * There are a maximum of 4 pages pinned at any time:
	 * two children, left parent and right parent (when the parent splits).
	 * keep the child pages pinned while working on the parent.
	 * make sure that all pins are released at exit.
	 */
	while ((parent = BT_POP(btstack)) != NULL) {
		/* parent page specified by stack frame <parent> */

		/* keep current child pages (<lp>, <rp>) pinned */
		lmp = smp;
		lp = sp;

		/*
		 * insert router entry in parent for new right child page <rp>
		 */
		/* get the parent page <sp> */
		DT_GETPAGE(ip, parent->bn, smp, PSIZE, sp, rc);
		if (rc) {
			DT_PUTPAGE(lmp);
			DT_PUTPAGE(rmp);
			goto splitOut;
		}

		/*
		 * The new key entry goes ONE AFTER the index of parent entry,
		 * because the split was to the right.
		 */
		skip = parent->index + 1;

		/*
		 * compute the key for the router entry
		 *
		 * key suffix compression:
		 * for internal pages that have leaf pages as children,
		 * retain only what's needed to distinguish between
		 * the new entry and the entry on the page to its left.
		 * If the keys compare equal, retain the entire key.
		 *
		 * note that compression is performed only at computing
		 * router key at the lowest internal level.
		 * further compression of the key between pairs of higher
		 * level internal pages loses too much information and
		 * the search may fail.
		 * (e.g., two adjacent leaf pages of {a, ..., x} {xx, ...,}
		 * results in two adjacent parent entries (a)(xx).
		 * if split occurs between these two entries, and
		 * if compression is applied, the router key of parent entry
		 * of right page (x) will divert search for x into right
		 * subtree and miss x in the left subtree.)
		 *
		 * the entire key must be retained for the next-to-leftmost
		 * internal key at any level of the tree, or search may fail
		 * (e.g., ?)
		 */
		switch (rp->header.flag & BT_TYPE) {
		case BT_LEAF:
			/*
			 * compute the length of prefix for suffix compression
			 * between last entry of left page and first entry
			 * of right page
			 */
			if ((sp->header.flag & BT_ROOT && skip > 1) ||
			    sp->header.prev != 0 || skip > 1) {
				/* compute uppercase router prefix key */
				rc = ciGetLeafPrefixKey(lp,
							lp->header.nextindex-1,
							rp, 0, &key,
							sbi->mntflag);
				if (rc) {
					DT_PUTPAGE(lmp);
					DT_PUTPAGE(rmp);
					DT_PUTPAGE(smp);
					goto splitOut;
				}
			} else {
				/* next to leftmost entry of
				   lowest internal level */

				/* compute uppercase router key */
				dtGetKey(rp, 0, &key, sbi->mntflag);
				key.name[key.namlen] = 0;

				if ((sbi->mntflag & JFS_OS2) == JFS_OS2)
					ciToUpper(&key);
			}

			n = NDTINTERNAL(key.namlen);
			break;

		case BT_INTERNAL:
			dtGetKey(rp, 0, &key, sbi->mntflag);
			n = NDTINTERNAL(key.namlen);
			break;

		default:
			jfs_err("dtSplitUp(): UFO!");
			break;
		}

		/* unpin left child page */
		DT_PUTPAGE(lmp);

		/*
		 * compute the data for the router entry
		 */
		data->xd = rpxd;	/* child page xd */

		/*
		 * parent page is full - split the parent page
		 */
		if (n > sp->header.freecnt) {
			/* init for parent page split */
			split->mp = smp;
			split->index = skip;	/* index at insert */
			split->nslot = n;
			split->key = &key;
			/* split->data = data; */

			/* unpin right child page */
			DT_PUTPAGE(rmp);

			/* The split routines insert the new entry,
			 * acquire txLock as appropriate.
			 * return <rp> pinned and its block number <rbn>.
			 */
			rc = (sp->header.flag & BT_ROOT) ?
			    dtSplitRoot(tid, ip, split, &rmp) :
			    dtSplitPage(tid, ip, split, &rmp, &rp, &rpxd);
			if (rc) {
				DT_PUTPAGE(smp);
				goto splitOut;
			}

			/* smp and rmp are pinned */
		}
		/*
		 * parent page is not full - insert router entry in parent page
		 */
		else {
			BT_MARK_DIRTY(smp, ip);
			/*
			 * acquire a transaction lock on the parent page
			 */
			tlck = txLock(tid, ip, smp, tlckDTREE | tlckENTRY);
			dtlck = (struct dt_lock *) & tlck->lock;
			ASSERT(dtlck->index == 0);
			lv = & dtlck->lv[0];

			/* linelock header */
			lv->offset = 0;
			lv->length = 1;
			dtlck->index++;

			/* linelock stbl of non-root parent page */
			if (!(sp->header.flag & BT_ROOT)) {
				lv++;
				n = skip >> L2DTSLOTSIZE;
				lv->offset = sp->header.stblindex + n;
				lv->length =
				    ((sp->header.nextindex -
				      1) >> L2DTSLOTSIZE) - n + 1;
				dtlck->index++;
			}

			dtInsertEntry(sp, skip, &key, data, &dtlck);

			/* exit propagate up */
			break;
		}
	}

	/* unpin current split and its right page */
	DT_PUTPAGE(smp);
	DT_PUTPAGE(rmp);

	/*
	 * free remaining extents allocated for split
	 */
      splitOut:
	n = pxdlist.npxd;
	pxd = &pxdlist.pxd[n];
	for (; n < pxdlist.maxnpxd; n++, pxd++)
		dbFree(ip, addressPXD(pxd), (s64) lengthPXD(pxd));

      freeKeyName:
	kfree(key.name);

	/* Rollback quota allocation */
	if (rc && quota_allocation)
		dquot_free_block(ip, quota_allocation);

      dtSplitUp_Exit:

	return rc;
}


/*
 *	dtSplitPage()
 *
 * function: Split a non-root page of a btree.
 *
 * parameter:
 *
 * return: 0 - success;
 *	   errno - failure;
 *	return split and new page pinned;
 */
static int dtSplitPage(tid_t tid, struct inode *ip, struct dtsplit * split,
	    struct metapage ** rmpp, dtpage_t ** rpp, pxd_t * rpxdp)
{
	int rc = 0;
	struct metapage *smp;
	dtpage_t *sp;
	struct metapage *rmp;
	dtpage_t *rp;		/* new right page allocated */
	s64 rbn;		/* new right page block number */
	struct metapage *mp;
	dtpage_t *p;
	s64 nextbn;
	struct pxdlist *pxdlist;
	pxd_t *pxd;
	int skip, nextindex, half, left, nxt, off, si;
	struct ldtentry *ldtentry;
	struct idtentry *idtentry;
	u8 *stbl;
	struct dtslot *f;
	int fsi, stblsize;
	int n;
	struct dt_lock *sdtlck, *rdtlck;
	struct tlock *tlck;
	struct dt_lock *dtlck;
	struct lv *slv, *rlv, *lv;

	/* get split page */
	smp = split->mp;
	sp = DT_PAGE(ip, smp);

	/*
	 * allocate the new right page for the split
	 */
	pxdlist = split->pxdlist;
	pxd = &pxdlist->pxd[pxdlist->npxd];
	pxdlist->npxd++;
	rbn = addressPXD(pxd);
	rmp = get_metapage(ip, rbn, PSIZE, 1);
	if (rmp == NULL)
		return -EIO;

	/* Allocate blocks to quota. */
	rc = dquot_alloc_block(ip, lengthPXD(pxd));
	if (rc) {
		release_metapage(rmp);
		return rc;
	}

	jfs_info("dtSplitPage: ip:0x%p smp:0x%p rmp:0x%p", ip, smp, rmp);

	BT_MARK_DIRTY(rmp, ip);
	/*
	 * acquire a transaction lock on the new right page
	 */
	tlck = txLock(tid, ip, rmp, tlckDTREE | tlckNEW);
	rdtlck = (struct dt_lock *) & tlck->lock;

	rp = (dtpage_t *) rmp->data;
	*rpp = rp;
	rp->header.self = *pxd;

	BT_MARK_DIRTY(smp, ip);
	/*
	 * acquire a transaction lock on the split page
	 *
	 * action:
	 */
	tlck = txLock(tid, ip, smp, tlckDTREE | tlckENTRY);
	sdtlck = (struct dt_lock *) & tlck->lock;

	/* linelock header of split page */
	ASSERT(sdtlck->index == 0);
	slv = & sdtlck->lv[0];
	slv->offset = 0;
	slv->length = 1;
	sdtlck->index++;

	/*
	 * initialize/update sibling pointers between sp and rp
	 */
	nextbn = le64_to_cpu(sp->header.next);
	rp->header.next = cpu_to_le64(nextbn);
	rp->header.prev = cpu_to_le64(addressPXD(&sp->header.self));
	sp->header.next = cpu_to_le64(rbn);

	/*
	 * initialize new right page
	 */
	rp->header.flag = sp->header.flag;

	/* compute sorted entry table at start of extent data area */
	rp->header.nextindex = 0;
	rp->header.stblindex = 1;

	n = PSIZE >> L2DTSLOTSIZE;
	rp->header.maxslot = n;
	stblsize = (n + 31) >> L2DTSLOTSIZE;	/* in unit of slot */

	/* init freelist */
	fsi = rp->header.stblindex + stblsize;
	rp->header.freelist = fsi;
	rp->header.freecnt = rp->header.maxslot - fsi;

	/*
	 *	sequential append at tail: append without split
	 *
	 * If splitting the last page on a level because of appending
	 * a entry to it (skip is maxentry), it's likely that the access is
	 * sequential. Adding an empty page on the side of the level is less
	 * work and can push the fill factor much higher than normal.
	 * If we're wrong it's no big deal, we'll just do the split the right
	 * way next time.
	 * (It may look like it's equally easy to do a similar hack for
	 * reverse sorted data, that is, split the tree left,
	 * but it's not. Be my guest.)
	 */
	if (nextbn == 0 && split->index == sp->header.nextindex) {
		/* linelock header + stbl (first slot) of new page */
		rlv = & rdtlck->lv[rdtlck->index];
		rlv->offset = 0;
		rlv->length = 2;
		rdtlck->index++;

		/*
		 * initialize freelist of new right page
		 */
		f = &rp->slot[fsi];
		for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
			f->next = fsi;
		f->next = -1;

		/* insert entry at the first entry of the new right page */
		dtInsertEntry(rp, 0, split->key, split->data, &rdtlck);

		goto out;
	}

	/*
	 *	non-sequential insert (at possibly middle page)
	 */

	/*
	 * update prev pointer of previous right sibling page;
	 */
	if (nextbn != 0) {
		DT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
		if (rc) {
			discard_metapage(rmp);
			return rc;
		}

		BT_MARK_DIRTY(mp, ip);
		/*
		 * acquire a transaction lock on the next page
		 */
		tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
		jfs_info("dtSplitPage: tlck = 0x%p, ip = 0x%p, mp=0x%p",
			tlck, ip, mp);
		dtlck = (struct dt_lock *) & tlck->lock;

		/* linelock header of previous right sibling page */
		lv = & dtlck->lv[dtlck->index];
		lv->offset = 0;
		lv->length = 1;
		dtlck->index++;

		p->header.prev = cpu_to_le64(rbn);

		DT_PUTPAGE(mp);
	}

	/*
	 * split the data between the split and right pages.
	 */
	skip = split->index;
	half = (PSIZE >> L2DTSLOTSIZE) >> 1;	/* swag */
	left = 0;

	/*
	 *	compute fill factor for split pages
	 *
	 * <nxt> traces the next entry to move to rp
	 * <off> traces the next entry to stay in sp
	 */
	stbl = (u8 *) & sp->slot[sp->header.stblindex];
	nextindex = sp->header.nextindex;
	for (nxt = off = 0; nxt < nextindex; ++off) {
		if (off == skip)
			/* check for fill factor with new entry size */
			n = split->nslot;
		else {
			si = stbl[nxt];
			switch (sp->header.flag & BT_TYPE) {
			case BT_LEAF:
				ldtentry = (struct ldtentry *) & sp->slot[si];
				if (DO_INDEX(ip))
					n = NDTLEAF(ldtentry->namlen);
				else
					n = NDTLEAF_LEGACY(ldtentry->
							   namlen);
				break;

			case BT_INTERNAL:
				idtentry = (struct idtentry *) & sp->slot[si];
				n = NDTINTERNAL(idtentry->namlen);
				break;

			default:
				break;
			}

			++nxt;	/* advance to next entry to move in sp */
		}

		left += n;
		if (left >= half)
			break;
	}

	/* <nxt> poins to the 1st entry to move */

	/*
	 *	move entries to right page
	 *
	 * dtMoveEntry() initializes rp and reserves entry for insertion
	 *
	 * split page moved out entries are linelocked;
	 * new/right page moved in entries are linelocked;
	 */
	/* linelock header + stbl of new right page */
	rlv = & rdtlck->lv[rdtlck->index];
	rlv->offset = 0;
	rlv->length = 5;
	rdtlck->index++;

	dtMoveEntry(sp, nxt, rp, &sdtlck, &rdtlck, DO_INDEX(ip));

	sp->header.nextindex = nxt;

	/*
	 * finalize freelist of new right page
	 */
	fsi = rp->header.freelist;
	f = &rp->slot[fsi];
	for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
		f->next = fsi;
	f->next = -1;

	/*
	 * Update directory index table for entries now in right page
	 */
	if ((rp->header.flag & BT_LEAF) && DO_INDEX(ip)) {
		s64 lblock;

		mp = NULL;
		stbl = DT_GETSTBL(rp);
		for (n = 0; n < rp->header.nextindex; n++) {
			ldtentry = (struct ldtentry *) & rp->slot[stbl[n]];
			modify_index(tid, ip, le32_to_cpu(ldtentry->index),
				     rbn, n, &mp, &lblock);
		}
		if (mp)
			release_metapage(mp);
	}

	/*
	 * the skipped index was on the left page,
	 */
	if (skip <= off) {
		/* insert the new entry in the split page */
		dtInsertEntry(sp, skip, split->key, split->data, &sdtlck);

		/* linelock stbl of split page */
		if (sdtlck->index >= sdtlck->maxcnt)
			sdtlck = (struct dt_lock *) txLinelock(sdtlck);
		slv = & sdtlck->lv[sdtlck->index];
		n = skip >> L2DTSLOTSIZE;
		slv->offset = sp->header.stblindex + n;
		slv->length =
		    ((sp->header.nextindex - 1) >> L2DTSLOTSIZE) - n + 1;
		sdtlck->index++;
	}
	/*
	 * the skipped index was on the right page,
	 */
	else {
		/* adjust the skip index to reflect the new position */
		skip -= nxt;

		/* insert the new entry in the right page */
		dtInsertEntry(rp, skip, split->key, split->data, &rdtlck);
	}

      out:
	*rmpp = rmp;
	*rpxdp = *pxd;

	return rc;
}


/*
 *	dtExtendPage()
 *
 * function: extend 1st/only directory leaf page
 *
 * parameter:
 *
 * return: 0 - success;
 *	   errno - failure;
 *	return extended page pinned;
 */
static int dtExtendPage(tid_t tid,
	     struct inode *ip, struct dtsplit * split, struct btstack * btstack)
{
	struct super_block *sb = ip->i_sb;
	int rc;
	struct metapage *smp, *pmp, *mp;
	dtpage_t *sp, *pp;
	struct pxdlist *pxdlist;
	pxd_t *pxd, *tpxd;
	int xlen, xsize;
	int newstblindex, newstblsize;
	int oldstblindex, oldstblsize;
	int fsi, last;
	struct dtslot *f;
	struct btframe *parent;
	int n;
	struct dt_lock *dtlck;
	s64 xaddr, txaddr;
	struct tlock *tlck;
	struct pxd_lock *pxdlock;
	struct lv *lv;
	uint type;
	struct ldtentry *ldtentry;
	u8 *stbl;

	/* get page to extend */
	smp = split->mp;
	sp = DT_PAGE(ip, smp);

	/* get parent/root page */
	parent = BT_POP(btstack);
	DT_GETPAGE(ip, parent->bn, pmp, PSIZE, pp, rc);
	if (rc)
		return (rc);

	/*
	 *	extend the extent
	 */
	pxdlist = split->pxdlist;
	pxd = &pxdlist->pxd[pxdlist->npxd];
	pxdlist->npxd++;

	xaddr = addressPXD(pxd);
	tpxd = &sp->header.self;
	txaddr = addressPXD(tpxd);
	/* in-place extension */
	if (xaddr == txaddr) {
		type = tlckEXTEND;
	}
	/* relocation */
	else {
		type = tlckNEW;

		/* save moved extent descriptor for later free */
		tlck = txMaplock(tid, ip, tlckDTREE | tlckRELOCATE);
		pxdlock = (struct pxd_lock *) & tlck->lock;
		pxdlock->flag = mlckFREEPXD;
		pxdlock->pxd = sp->header.self;
		pxdlock->index = 1;

		/*
		 * Update directory index table to reflect new page address
		 */
		if (DO_INDEX(ip)) {
			s64 lblock;

			mp = NULL;
			stbl = DT_GETSTBL(sp);
			for (n = 0; n < sp->header.nextindex; n++) {
				ldtentry =
				    (struct ldtentry *) & sp->slot[stbl[n]];
				modify_index(tid, ip,
					     le32_to_cpu(ldtentry->index),
					     xaddr, n, &mp, &lblock);
			}
			if (mp)
				release_metapage(mp);
		}
	}

	/*
	 *	extend the page
	 */
	sp->header.self = *pxd;

	jfs_info("dtExtendPage: ip:0x%p smp:0x%p sp:0x%p", ip, smp, sp);

	BT_MARK_DIRTY(smp, ip);
	/*
	 * acquire a transaction lock on the extended/leaf page
	 */
	tlck = txLock(tid, ip, smp, tlckDTREE | type);
	dtlck = (struct dt_lock *) & tlck->lock;
	lv = & dtlck->lv[0];

	/* update buffer extent descriptor of extended page */
	xlen = lengthPXD(pxd);
	xsize = xlen << JFS_SBI(sb)->l2bsize;

	/*
	 * copy old stbl to new stbl at start of extended area
	 */
	oldstblindex = sp->header.stblindex;
	oldstblsize = (sp->header.maxslot + 31) >> L2DTSLOTSIZE;
	newstblindex = sp->header.maxslot;
	n = xsize >> L2DTSLOTSIZE;
	newstblsize = (n + 31) >> L2DTSLOTSIZE;
	memcpy(&sp->slot[newstblindex], &sp->slot[oldstblindex],
	       sp->header.nextindex);

	/*
	 * in-line extension: linelock old area of extended page
	 */
	if (type == tlckEXTEND) {
		/* linelock header */
		lv->offset = 0;
		lv->length = 1;
		dtlck->index++;
		lv++;

		/* linelock new stbl of extended page */
		lv->offset = newstblindex;
		lv->length = newstblsize;
	}
	/*
	 * relocation: linelock whole relocated area
	 */
	else {
		lv->offset = 0;
		lv->length = sp->header.maxslot + newstblsize;
	}

	dtlck->index++;

	sp->header.maxslot = n;
	sp->header.stblindex = newstblindex;
	/* sp->header.nextindex remains the same */

	/*
	 * add old stbl region at head of freelist
	 */
	fsi = oldstblindex;
	f = &sp->slot[fsi];
	last = sp->header.freelist;
	for (n = 0; n < oldstblsize; n++, fsi++, f++) {
		f->next = last;
		last = fsi;
	}
	sp->header.freelist = last;
	sp->header.freecnt += oldstblsize;

	/*
	 * append free region of newly extended area at tail of freelist
	 */
	/* init free region of newly extended area */
	fsi = n = newstblindex + newstblsize;
	f = &sp->slot[fsi];
	for (fsi++; fsi < sp->header.maxslot; f++, fsi++)
		f->next = fsi;
	f->next = -1;

	/* append new free region at tail of old freelist */
	fsi = sp->header.freelist;
	if (fsi == -1)
		sp->header.freelist = n;
	else {
		do {
			f = &sp->slot[fsi];
			fsi = f->next;
		} while (fsi != -1);

		f->next = n;
	}

	sp->header.freecnt += sp->header.maxslot - n;

	/*
	 * insert the new entry
	 */
	dtInsertEntry(sp, split->index, split->key, split->data, &dtlck);

	BT_MARK_DIRTY(pmp, ip);
	/*
	 * linelock any freeslots residing in old extent
	 */
	if (type == tlckEXTEND) {
		n = sp->header.maxslot >> 2;
		if (sp->header.freelist < n)
			dtLinelockFreelist(sp, n, &dtlck);
	}

	/*
	 *	update parent entry on the parent/root page
	 */
	/*
	 * acquire a transaction lock on the parent/root page
	 */
	tlck = txLock(tid, ip, pmp, tlckDTREE | tlckENTRY);
	dtlck = (struct dt_lock *) & tlck->lock;
	lv = & dtlck->lv[dtlck->index];

	/* linelock parent entry - 1st slot */
	lv->offset = 1;
	lv->length = 1;
	dtlck->index++;

	/* update the parent pxd for page extension */
	tpxd = (pxd_t *) & pp->slot[1];
	*tpxd = *pxd;

	DT_PUTPAGE(pmp);
	return 0;
}


/*
 *	dtSplitRoot()
 *
 * function:
 *	split the full root page into
 *	original/root/split page and new right page
 *	i.e., root remains fixed in tree anchor (inode) and
 *	the root is copied to a single new right child page
 *	since root page << non-root page, and
 *	the split root page contains a single entry for the
 *	new right child page.
 *
 * parameter:
 *
 * return: 0 - success;
 *	   errno - failure;
 *	return new page pinned;
 */
static int dtSplitRoot(tid_t tid,
	    struct inode *ip, struct dtsplit * split, struct metapage ** rmpp)
{
	struct super_block *sb = ip->i_sb;
	struct metapage *smp;
	dtroot_t *sp;
	struct metapage *rmp;
	dtpage_t *rp;
	s64 rbn;
	int xlen;
	int xsize;
	struct dtslot *f;
	s8 *stbl;
	int fsi, stblsize, n;
	struct idtentry *s;
	pxd_t *ppxd;
	struct pxdlist *pxdlist;
	pxd_t *pxd;
	struct dt_lock *dtlck;
	struct tlock *tlck;
	struct lv *lv;
	int rc;

	/* get split root page */
	smp = split->mp;
	sp = &JFS_IP(ip)->i_dtroot;

	/*
	 *	allocate/initialize a single (right) child page
	 *
	 * N.B. at first split, a one (or two) block to fit new entry
	 * is allocated; at subsequent split, a full page is allocated;
	 */
	pxdlist = split->pxdlist;
	pxd = &pxdlist->pxd[pxdlist->npxd];
	pxdlist->npxd++;
	rbn = addressPXD(pxd);
	xlen = lengthPXD(pxd);
	xsize = xlen << JFS_SBI(sb)->l2bsize;
	rmp = get_metapage(ip, rbn, xsize, 1);
	if (!rmp)
		return -EIO;

	rp = rmp->data;

	/* Allocate blocks to quota. */
	rc = dquot_alloc_block(ip, lengthPXD(pxd));
	if (rc) {
		release_metapage(rmp);
		return rc;
	}

	BT_MARK_DIRTY(rmp, ip);
	/*
	 * acquire a transaction lock on the new right page
	 */
	tlck = txLock(tid, ip, rmp, tlckDTREE | tlckNEW);
	dtlck = (struct dt_lock *) & tlck->lock;

	rp->header.flag =
	    (sp->header.flag & BT_LEAF) ? BT_LEAF : BT_INTERNAL;
	rp->header.self = *pxd;

	/* initialize sibling pointers */
	rp->header.next = 0;
	rp->header.prev = 0;

	/*
	 *	move in-line root page into new right page extent
	 */
	/* linelock header + copied entries + new stbl (1st slot) in new page */
	ASSERT(dtlck->index == 0);
	lv = & dtlck->lv[0];
	lv->offset = 0;
	lv->length = 10;	/* 1 + 8 + 1 */
	dtlck->index++;

	n = xsize >> L2DTSLOTSIZE;
	rp->header.maxslot = n;
	stblsize = (n + 31) >> L2DTSLOTSIZE;

	/* copy old stbl to new stbl at start of extended area */
	rp->header.stblindex = DTROOTMAXSLOT;
	stbl = (s8 *) & rp->slot[DTROOTMAXSLOT];
	memcpy(stbl, sp->header.stbl, sp->header.nextindex);
	rp->header.nextindex = sp->header.nextindex;

	/* copy old data area to start of new data area */
	memcpy(&rp->slot[1], &sp->slot[1], IDATASIZE);

	/*
	 * append free region of newly extended area at tail of freelist
	 */
	/* init free region of newly extended area */
	fsi = n = DTROOTMAXSLOT + stblsize;
	f = &rp->slot[fsi];
	for (fsi++; fsi < rp->header.maxslot; f++, fsi++)
		f->next = fsi;
	f->next = -1;

	/* append new free region at tail of old freelist */
	fsi = sp->header.freelist;
	if (fsi == -1)
		rp->header.freelist = n;
	else {
		rp->header.freelist = fsi;

		do {
			f = &rp->slot[fsi];
			fsi = f->next;
		} while (fsi != -1);

		f->next = n;
	}

	rp->header.freecnt = sp->header.freecnt + rp->header.maxslot - n;

	/*
	 * Update directory index table for entries now in right page
	 */
	if ((rp->header.flag & BT_LEAF) && DO_INDEX(ip)) {
		s64 lblock;
		struct metapage *mp = NULL;
		struct ldtentry *ldtentry;

		stbl = DT_GETSTBL(rp);
		for (n = 0; n < rp->header.nextindex; n++) {
			ldtentry = (struct ldtentry *) & rp->slot[stbl[n]];
			modify_index(tid, ip, le32_to_cpu(ldtentry->index),
				     rbn, n, &mp, &lblock);
		}
		if (mp)
			release_metapage(mp);
	}
	/*
	 * insert the new entry into the new right/child page
	 * (skip index in the new right page will not change)
	 */
	dtInsertEntry(rp, split->index, split->key, split->data, &dtlck);

	/*
	 *	reset parent/root page
	 *
	 * set the 1st entry offset to 0, which force the left-most key
	 * at any level of the tree to be less than any search key.
	 *
	 * The btree comparison code guarantees that the left-most key on any
	 * level of the tree is never used, so it doesn't need to be filled in.
	 */
	BT_MARK_DIRTY(smp, ip);
	/*
	 * acquire a transaction lock on the root page (in-memory inode)
	 */
	tlck = txLock(tid, ip, smp, tlckDTREE | tlckNEW | tlckBTROOT);
	dtlck = (struct dt_lock *) & tlck->lock;

	/* linelock root */
	ASSERT(dtlck->index == 0);
	lv = & dtlck->lv[0];
	lv->offset = 0;
	lv->length = DTROOTMAXSLOT;
	dtlck->index++;

	/* update page header of root */
	if (sp->header.flag & BT_LEAF) {
		sp->header.flag &= ~BT_LEAF;
		sp->header.flag |= BT_INTERNAL;
	}

	/* init the first entry */
	s = (struct idtentry *) & sp->slot[DTENTRYSTART];
	ppxd = (pxd_t *) s;
	*ppxd = *pxd;
	s->next = -1;
	s->namlen = 0;

	stbl = sp->header.stbl;
	stbl[0] = DTENTRYSTART;
	sp->header.nextindex = 1;

	/* init freelist */
	fsi = DTENTRYSTART + 1;
	f = &sp->slot[fsi];

	/* init free region of remaining area */
	for (fsi++; fsi < DTROOTMAXSLOT; f++, fsi++)
		f->next = fsi;
	f->next = -1;

	sp->header.freelist = DTENTRYSTART + 1;
	sp->header.freecnt = DTROOTMAXSLOT - (DTENTRYSTART + 1);

	*rmpp = rmp;

	return 0;
}


/*
 *	dtDelete()
 *
 * function: delete the entry(s) referenced by a key.
 *
 * parameter:
 *
 * return:
 */
int dtDelete(tid_t tid,
	 struct inode *ip, struct component_name * key, ino_t * ino, int flag)
{
	int rc = 0;
	s64 bn;
	struct metapage *mp, *imp;
	dtpage_t *p;
	int index;
	struct btstack btstack;
	struct dt_lock *dtlck;
	struct tlock *tlck;
	struct lv *lv;
	int i;
	struct ldtentry *ldtentry;
	u8 *stbl;
	u32 table_index, next_index;
	struct metapage *nmp;
	dtpage_t *np;

	/*
	 *	search for the entry to delete:
	 *
	 * dtSearch() returns (leaf page pinned, index at which to delete).
	 */
	if ((rc = dtSearch(ip, key, ino, &btstack, flag)))
		return rc;

	/* retrieve search result */
	DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);

	/*
	 * We need to find put the index of the next entry into the
	 * directory index table in order to resume a readdir from this
	 * entry.
	 */
	if (DO_INDEX(ip)) {
		stbl = DT_GETSTBL(p);
		ldtentry = (struct ldtentry *) & p->slot[stbl[index]];
		table_index = le32_to_cpu(ldtentry->index);
		if (index == (p->header.nextindex - 1)) {
			/*
			 * Last entry in this leaf page
			 */
			if ((p->header.flag & BT_ROOT)
			    || (p->header.next == 0))
				next_index = -1;
			else {
				/* Read next leaf page */
				DT_GETPAGE(ip, le64_to_cpu(p->header.next),
					   nmp, PSIZE, np, rc);
				if (rc)
					next_index = -1;
				else {
					stbl = DT_GETSTBL(np);
					ldtentry =
					    (struct ldtentry *) & np->
					    slot[stbl[0]];
					next_index =
					    le32_to_cpu(ldtentry->index);
					DT_PUTPAGE(nmp);
				}
			}
		} else {
			ldtentry =
			    (struct ldtentry *) & p->slot[stbl[index + 1]];
			next_index = le32_to_cpu(ldtentry->index);
		}
		free_index(tid, ip, table_index, next_index);
	}
	/*
	 * the leaf page becomes empty, delete the page
	 */
	if (p->header.nextindex == 1) {
		/* delete empty page */
		rc = dtDeleteUp(tid, ip, mp, p, &btstack);
	}
	/*
	 * the leaf page has other entries remaining:
	 *
	 * delete the entry from the leaf page.
	 */
	else {
		BT_MARK_DIRTY(mp, ip);
		/*
		 * acquire a transaction lock on the leaf page
		 */
		tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
		dtlck = (struct dt_lock *) & tlck->lock;

		/*
		 * Do not assume that dtlck->index will be zero.  During a
		 * rename within a directory, this transaction may have
		 * modified this page already when adding the new entry.
		 */

		/* linelock header */
		if (dtlck->index >= dtlck->maxcnt)
			dtlck = (struct dt_lock *) txLinelock(dtlck);
		lv = & dtlck->lv[dtlck->index];
		lv->offset = 0;
		lv->length = 1;
		dtlck->index++;

		/* linelock stbl of non-root leaf page */
		if (!(p->header.flag & BT_ROOT)) {
			if (dtlck->index >= dtlck->maxcnt)
				dtlck = (struct dt_lock *) txLinelock(dtlck);
			lv = & dtlck->lv[dtlck->index];
			i = index >> L2DTSLOTSIZE;
			lv->offset = p->header.stblindex + i;
			lv->length =
			    ((p->header.nextindex - 1) >> L2DTSLOTSIZE) -
			    i + 1;
			dtlck->index++;
		}

		/* free the leaf entry */
		dtDeleteEntry(p, index, &dtlck);

		/*
		 * Update directory index table for entries moved in stbl
		 */
		if (DO_INDEX(ip) && index < p->header.nextindex) {
			s64 lblock;

			imp = NULL;
			stbl = DT_GETSTBL(p);
			for (i = index; i < p->header.nextindex; i++) {
				ldtentry =
				    (struct ldtentry *) & p->slot[stbl[i]];
				modify_index(tid, ip,
					     le32_to_cpu(ldtentry->index),
					     bn, i, &imp, &lblock);
			}
			if (imp)
				release_metapage(imp);
		}

		DT_PUTPAGE(mp);
	}

	return rc;
}


/*
 *	dtDeleteUp()
 *
 * function:
 *	free empty pages as propagating deletion up the tree
 *
 * parameter:
 *
 * return:
 */
static int dtDeleteUp(tid_t tid, struct inode *ip,
	   struct metapage * fmp, dtpage_t * fp, struct btstack * btstack)
{
	int rc = 0;
	struct metapage *mp;
	dtpage_t *p;
	int index, nextindex;
	int xlen;
	struct btframe *parent;
	struct dt_lock *dtlck;
	struct tlock *tlck;
	struct lv *lv;
	struct pxd_lock *pxdlock;
	int i;

	/*
	 *	keep the root leaf page which has become empty
	 */
	if (BT_IS_ROOT(fmp)) {
		/*
		 * reset the root
		 *
		 * dtInitRoot() acquires txlock on the root
		 */
		dtInitRoot(tid, ip, PARENT(ip));

		DT_PUTPAGE(fmp);

		return 0;
	}

	/*
	 *	free the non-root leaf page
	 */
	/*
	 * acquire a transaction lock on the page
	 *
	 * write FREEXTENT|NOREDOPAGE log record
	 * N.B. linelock is overlaid as freed extent descriptor, and
	 * the buffer page is freed;
	 */
	tlck = txMaplock(tid, ip, tlckDTREE | tlckFREE);
	pxdlock = (struct pxd_lock *) & tlck->lock;
	pxdlock->flag = mlckFREEPXD;
	pxdlock->pxd = fp->header.self;
	pxdlock->index = 1;

	/* update sibling pointers */
	if ((rc = dtRelink(tid, ip, fp))) {
		BT_PUTPAGE(fmp);
		return rc;
	}

	xlen = lengthPXD(&fp->header.self);

	/* Free quota allocation. */
	dquot_free_block(ip, xlen);

	/* free/invalidate its buffer page */
	discard_metapage(fmp);

	/*
	 *	propagate page deletion up the directory tree
	 *
	 * If the delete from the parent page makes it empty,
	 * continue all the way up the tree.
	 * stop if the root page is reached (which is never deleted) or
	 * if the entry deletion does not empty the page.
	 */
	while ((parent = BT_POP(btstack)) != NULL) {
		/* pin the parent page <sp> */
		DT_GETPAGE(ip, parent->bn, mp, PSIZE, p, rc);
		if (rc)
			return rc;

		/*
		 * free the extent of the child page deleted
		 */
		index = parent->index;

		/*
		 * delete the entry for the child page from parent
		 */
		nextindex = p->header.nextindex;

		/*
		 * the parent has the single entry being deleted:
		 *
		 * free the parent page which has become empty.
		 */
		if (nextindex == 1) {
			/*
			 * keep the root internal page which has become empty
			 */
			if (p->header.flag & BT_ROOT) {
				/*
				 * reset the root
				 *
				 * dtInitRoot() acquires txlock on the root
				 */
				dtInitRoot(tid, ip, PARENT(ip));

				DT_PUTPAGE(mp);

				return 0;
			}
			/*
			 * free the parent page
			 */
			else {
				/*
				 * acquire a transaction lock on the page
				 *
				 * write FREEXTENT|NOREDOPAGE log record
				 */
				tlck =
				    txMaplock(tid, ip,
					      tlckDTREE | tlckFREE);
				pxdlock = (struct pxd_lock *) & tlck->lock;
				pxdlock->flag = mlckFREEPXD;
				pxdlock->pxd = p->header.self;
				pxdlock->index = 1;

				/* update sibling pointers */
				if ((rc = dtRelink(tid, ip, p))) {
					DT_PUTPAGE(mp);
					return rc;
				}

				xlen = lengthPXD(&p->header.self);

				/* Free quota allocation */
				dquot_free_block(ip, xlen);

				/* free/invalidate its buffer page */
				discard_metapage(mp);

				/* propagate up */
				continue;
			}
		}

		/*
		 * the parent has other entries remaining:
		 *
		 * delete the router entry from the parent page.
		 */
		BT_MARK_DIRTY(mp, ip);
		/*
		 * acquire a transaction lock on the page
		 *
		 * action: router entry deletion
		 */
		tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
		dtlck = (struct dt_lock *) & tlck->lock;

		/* linelock header */
		if (dtlck->index >= dtlck->maxcnt)
			dtlck = (struct dt_lock *) txLinelock(dtlck);
		lv = & dtlck->lv[dtlck->index];
		lv->offset = 0;
		lv->length = 1;
		dtlck->index++;

		/* linelock stbl of non-root leaf page */
		if (!(p->header.flag & BT_ROOT)) {
			if (dtlck->index < dtlck->maxcnt)
				lv++;
			else {
				dtlck = (struct dt_lock *) txLinelock(dtlck);
				lv = & dtlck->lv[0];
			}
			i = index >> L2DTSLOTSIZE;
			lv->offset = p->header.stblindex + i;
			lv->length =
			    ((p->header.nextindex - 1) >> L2DTSLOTSIZE) -
			    i + 1;
			dtlck->index++;
		}

		/* free the router entry */
		dtDeleteEntry(p, index, &dtlck);

		/* reset key of new leftmost entry of level (for consistency) */
		if (index == 0 &&
		    ((p->header.flag & BT_ROOT) || p->header.prev == 0))
			dtTruncateEntry(p, 0, &dtlck);

		/* unpin the parent page */
		DT_PUTPAGE(mp);

		/* exit propagation up */
		break;
	}

	if (!DO_INDEX(ip))
		ip->i_size -= PSIZE;

	return 0;
}

#ifdef _NOTYET
/*
 * NAME:	dtRelocate()
 *
 * FUNCTION:	relocate dtpage (internal or leaf) of directory;
 *		This function is mainly used by defragfs utility.
 */
int dtRelocate(tid_t tid, struct inode *ip, s64 lmxaddr, pxd_t * opxd,
	       s64 nxaddr)
{
	int rc = 0;
	struct metapage *mp, *pmp, *lmp, *rmp;
	dtpage_t *p, *pp, *rp = 0, *lp= 0;
	s64 bn;
	int index;
	struct btstack btstack;
	pxd_t *pxd;
	s64 oxaddr, nextbn, prevbn;
	int xlen, xsize;
	struct tlock *tlck;
	struct dt_lock *dtlck;
	struct pxd_lock *pxdlock;
	s8 *stbl;
	struct lv *lv;

	oxaddr = addressPXD(opxd);
	xlen = lengthPXD(opxd);

	jfs_info("dtRelocate: lmxaddr:%Ld xaddr:%Ld:%Ld xlen:%d",
		   (long long)lmxaddr, (long long)oxaddr, (long long)nxaddr,
		   xlen);

	/*
	 *	1. get the internal parent dtpage covering
	 *	router entry for the tartget page to be relocated;
	 */
	rc = dtSearchNode(ip, lmxaddr, opxd, &btstack);
	if (rc)
		return rc;

	/* retrieve search result */
	DT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index);
	jfs_info("dtRelocate: parent router entry validated.");

	/*
	 *	2. relocate the target dtpage
	 */
	/* read in the target page from src extent */
	DT_GETPAGE(ip, oxaddr, mp, PSIZE, p, rc);
	if (rc) {
		/* release the pinned parent page */
		DT_PUTPAGE(pmp);
		return rc;
	}

	/*
	 * read in sibling pages if any to update sibling pointers;
	 */
	rmp = NULL;
	if (p->header.next) {
		nextbn = le64_to_cpu(p->header.next);
		DT_GETPAGE(ip, nextbn, rmp, PSIZE, rp, rc);
		if (rc) {
			DT_PUTPAGE(mp);
			DT_PUTPAGE(pmp);
			return (rc);
		}
	}

	lmp = NULL;
	if (p->header.prev) {
		prevbn = le64_to_cpu(p->header.prev);
		DT_GETPAGE(ip, prevbn, lmp, PSIZE, lp, rc);
		if (rc) {
			DT_PUTPAGE(mp);
			DT_PUTPAGE(pmp);
			if (rmp)
				DT_PUTPAGE(rmp);
			return (rc);
		}
	}

	/* at this point, all xtpages to be updated are in memory */

	/*
	 * update sibling pointers of sibling dtpages if any;
	 */
	if (lmp) {
		tlck = txLock(tid, ip, lmp, tlckDTREE | tlckRELINK);
		dtlck = (struct dt_lock *) & tlck->lock;
		/* linelock header */
		ASSERT(dtlck->index == 0);
		lv = & dtlck->lv[0];
		lv->offset = 0;
		lv->length = 1;
		dtlck->index++;

		lp->header.next = cpu_to_le64(nxaddr);
		DT_PUTPAGE(lmp);
	}

	if (rmp) {
		tlck = txLock(tid, ip, rmp, tlckDTREE | tlckRELINK);
		dtlck = (struct dt_lock *) & tlck->lock;
		/* linelock header */
		ASSERT(dtlck->index == 0);
		lv = & dtlck->lv[0];
		lv->offset = 0;
		lv->length = 1;
		dtlck->index++;

		rp->header.prev = cpu_to_le64(nxaddr);
		DT_PUTPAGE(rmp);
	}

	/*
	 * update the target dtpage to be relocated
	 *
	 * write LOG_REDOPAGE of LOG_NEW type for dst page
	 * for the whole target page (logredo() will apply
	 * after image and update bmap for allocation of the
	 * dst extent), and update bmap for allocation of
	 * the dst extent;
	 */
	tlck = txLock(tid, ip, mp, tlckDTREE | tlckNEW);
	dtlck = (struct dt_lock *) & tlck->lock;
	/* linelock header */
	ASSERT(dtlck->index == 0);
	lv = & dtlck->lv[0];

	/* update the self address in the dtpage header */
	pxd = &p->header.self;
	PXDaddress(pxd, nxaddr);

	/* the dst page is the same as the src page, i.e.,
	 * linelock for afterimage of the whole page;
	 */
	lv->offset = 0;
	lv->length = p->header.maxslot;
	dtlck->index++;

	/* update the buffer extent descriptor of the dtpage */
	xsize = xlen << JFS_SBI(ip->i_sb)->l2bsize;

	/* unpin the relocated page */
	DT_PUTPAGE(mp);
	jfs_info("dtRelocate: target dtpage relocated.");

	/* the moved extent is dtpage, then a LOG_NOREDOPAGE log rec
	 * needs to be written (in logredo(), the LOG_NOREDOPAGE log rec
	 * will also force a bmap update ).
	 */

	/*
	 *	3. acquire maplock for the source extent to be freed;
	 */
	/* for dtpage relocation, write a LOG_NOREDOPAGE record
	 * for the source dtpage (logredo() will init NoRedoPage
	 * filter and will also update bmap for free of the source
	 * dtpage), and upadte bmap for free of the source dtpage;
	 */
	tlck = txMaplock(tid, ip, tlckDTREE | tlckFREE);
	pxdlock = (struct pxd_lock *) & tlck->lock;
	pxdlock->flag = mlckFREEPXD;
	PXDaddress(&pxdlock->pxd, oxaddr);
	PXDlength(&pxdlock->pxd, xlen);
	pxdlock->index = 1;

	/*
	 *	4. update the parent router entry for relocation;
	 *
	 * acquire tlck for the parent entry covering the target dtpage;
	 * write LOG_REDOPAGE to apply after image only;
	 */
	jfs_info("dtRelocate: update parent router entry.");
	tlck = txLock(tid, ip, pmp, tlckDTREE | tlckENTRY);
	dtlck = (struct dt_lock *) & tlck->lock;
	lv = & dtlck->lv[dtlck->index];

	/* update the PXD with the new address */
	stbl = DT_GETSTBL(pp);
	pxd = (pxd_t *) & pp->slot[stbl[index]];
	PXDaddress(pxd, nxaddr);
	lv->offset = stbl[index];
	lv->length = 1;
	dtlck->index++;

	/* unpin the parent dtpage */
	DT_PUTPAGE(pmp);

	return rc;
}

/*
 * NAME:	dtSearchNode()
 *
 * FUNCTION:	Search for an dtpage containing a specified address
 *		This function is mainly used by defragfs utility.
 *
 * NOTE:	Search result on stack, the found page is pinned at exit.
 *		The result page must be an internal dtpage.
 *		lmxaddr give the address of the left most page of the
 *		dtree level, in which the required dtpage resides.
 */
static int dtSearchNode(struct inode *ip, s64 lmxaddr, pxd_t * kpxd,
			struct btstack * btstack)
{
	int rc = 0;
	s64 bn;
	struct metapage *mp;
	dtpage_t *p;
	int psize = 288;	/* initial in-line directory */
	s8 *stbl;
	int i;
	pxd_t *pxd;
	struct btframe *btsp;

	BT_CLR(btstack);	/* reset stack */

	/*
	 *	descend tree to the level with specified leftmost page
	 *
	 *  by convention, root bn = 0.
	 */
	for (bn = 0;;) {
		/* get/pin the page to search */
		DT_GETPAGE(ip, bn, mp, psize, p, rc);
		if (rc)
			return rc;

		/* does the xaddr of leftmost page of the levevl
		 * matches levevl search key ?
		 */
		if (p->header.flag & BT_ROOT) {
			if (lmxaddr == 0)
				break;
		} else if (addressPXD(&p->header.self) == lmxaddr)
			break;

		/*
		 * descend down to leftmost child page
		 */
		if (p->header.flag & BT_LEAF) {
			DT_PUTPAGE(mp);
			return -ESTALE;
		}

		/* get the leftmost entry */
		stbl = DT_GETSTBL(p);
		pxd = (pxd_t *) & p->slot[stbl[0]];

		/* get the child page block address */
		bn = addressPXD(pxd);
		psize = lengthPXD(pxd) << JFS_SBI(ip->i_sb)->l2bsize;
		/* unpin the parent page */
		DT_PUTPAGE(mp);
	}

	/*
	 *	search each page at the current levevl
	 */
      loop:
	stbl = DT_GETSTBL(p);
	for (i = 0; i < p->header.nextindex; i++) {
		pxd = (pxd_t *) & p->slot[stbl[i]];

		/* found the specified router entry */
		if (addressPXD(pxd) == addressPXD(kpxd) &&
		    lengthPXD(pxd) == lengthPXD(kpxd)) {
			btsp = btstack->top;
			btsp->bn = bn;
			btsp->index = i;
			btsp->mp = mp;

			return 0;
		}
	}

	/* get the right sibling page if any */
	if (p->header.next)
		bn = le64_to_cpu(p->header.next);
	else {
		DT_PUTPAGE(mp);
		return -ESTALE;
	}

	/* unpin current page */
	DT_PUTPAGE(mp);

	/* get the right sibling page */
	DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
	if (rc)
		return rc;

	goto loop;
}
#endif /* _NOTYET */

/*
 *	dtRelink()
 *
 * function:
 *	link around a freed page.
 *
 * parameter:
 *	fp:	page to be freed
 *
 * return:
 */
static int dtRelink(tid_t tid, struct inode *ip, dtpage_t * p)
{
	int rc;
	struct metapage *mp;
	s64 nextbn, prevbn;
	struct tlock *tlck;
	struct dt_lock *dtlck;
	struct lv *lv;

	nextbn = le64_to_cpu(p->header.next);
	prevbn = le64_to_cpu(p->header.prev);

	/* update prev pointer of the next page */
	if (nextbn != 0) {
		DT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc);
		if (rc)
			return rc;

		BT_MARK_DIRTY(mp, ip);
		/*
		 * acquire a transaction lock on the next page
		 *
		 * action: update prev pointer;
		 */
		tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
		jfs_info("dtRelink nextbn: tlck = 0x%p, ip = 0x%p, mp=0x%p",
			tlck, ip, mp);
		dtlck = (struct dt_lock *) & tlck->lock;

		/* linelock header */
		if (dtlck->index >= dtlck->maxcnt)
			dtlck = (struct dt_lock *) txLinelock(dtlck);
		lv = & dtlck->lv[dtlck->index];
		lv->offset = 0;
		lv->length = 1;
		dtlck->index++;

		p->header.prev = cpu_to_le64(prevbn);
		DT_PUTPAGE(mp);
	}

	/* update next pointer of the previous page */
	if (prevbn != 0) {
		DT_GETPAGE(ip, prevbn, mp, PSIZE, p, rc);
		if (rc)
			return rc;

		BT_MARK_DIRTY(mp, ip);
		/*
		 * acquire a transaction lock on the prev page
		 *
		 * action: update next pointer;
		 */
		tlck = txLock(tid, ip, mp, tlckDTREE | tlckRELINK);
		jfs_info("dtRelink prevbn: tlck = 0x%p, ip = 0x%p, mp=0x%p",
			tlck, ip, mp);
		dtlck = (struct dt_lock *) & tlck->lock;

		/* linelock header */
		if (dtlck->index >= dtlck->maxcnt)
			dtlck = (struct dt_lock *) txLinelock(dtlck);
		lv = & dtlck->lv[dtlck->index];
		lv->offset = 0;
		lv->length = 1;
		dtlck->index++;

		p->header.next = cpu_to_le64(nextbn);
		DT_PUTPAGE(mp);
	}

	return 0;
}


/*
 *	dtInitRoot()
 *
 * initialize directory root (inline in inode)
 */
void dtInitRoot(tid_t tid, struct inode *ip, u32 idotdot)
{
	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
	dtroot_t *p;
	int fsi;
	struct dtslot *f;
	struct tlock *tlck;
	struct dt_lock *dtlck;
	struct lv *lv;
	u16 xflag_save;

	/*
	 * If this was previously an non-empty directory, we need to remove
	 * the old directory table.
	 */
	if (DO_INDEX(ip)) {
		if (!jfs_dirtable_inline(ip)) {
			struct tblock *tblk = tid_to_tblock(tid);
			/*
			 * We're playing games with the tid's xflag.  If
			 * we're removing a regular file, the file's xtree
			 * is committed with COMMIT_PMAP, but we always
			 * commit the directories xtree with COMMIT_PWMAP.
			 */
			xflag_save = tblk->xflag;
			tblk->xflag = 0;
			/*
			 * xtTruncate isn't guaranteed to fully truncate
			 * the xtree.  The caller needs to check i_size
			 * after committing the transaction to see if
			 * additional truncation is needed.  The
			 * COMMIT_Stale flag tells caller that we
			 * initiated the truncation.
			 */
			xtTruncate(tid, ip, 0, COMMIT_PWMAP);
			set_cflag(COMMIT_Stale, ip);

			tblk->xflag = xflag_save;
		} else
			ip->i_size = 1;

		jfs_ip->next_index = 2;
	} else
		ip->i_size = IDATASIZE;

	/*
	 * acquire a transaction lock on the root
	 *
	 * action: directory initialization;
	 */
	tlck = txLock(tid, ip, (struct metapage *) & jfs_ip->bxflag,
		      tlckDTREE | tlckENTRY | tlckBTROOT);
	dtlck = (struct dt_lock *) & tlck->lock;

	/* linelock root */
	ASSERT(dtlck->index == 0);
	lv = & dtlck->lv[0];
	lv->offset = 0;
	lv->length = DTROOTMAXSLOT;
	dtlck->index++;

	p = &jfs_ip->i_dtroot;

	p->header.flag = DXD_INDEX | BT_ROOT | BT_LEAF;

	p->header.nextindex = 0;

	/* init freelist */
	fsi = 1;
	f = &p->slot[fsi];

	/* init data area of root */
	for (fsi++; fsi < DTROOTMAXSLOT; f++, fsi++)
		f->next = fsi;
	f->next = -1;

	p->header.freelist = 1;
	p->header.freecnt = 8;

	/* init '..' entry */
	p->header.idotdot = cpu_to_le32(idotdot);

	return;
}

/*
 *	add_missing_indices()
 *
 * function: Fix dtree page in which one or more entries has an invalid index.
 *	     fsck.jfs should really fix this, but it currently does not.
 *	     Called from jfs_readdir when bad index is detected.
 */
static void add_missing_indices(struct inode *inode, s64 bn)
{
	struct ldtentry *d;
	struct dt_lock *dtlck;
	int i;
	uint index;
	struct lv *lv;
	struct metapage *mp;
	dtpage_t *p;
	int rc;
	s8 *stbl;
	tid_t tid;
	struct tlock *tlck;

	tid = txBegin(inode->i_sb, 0);

	DT_GETPAGE(inode, bn, mp, PSIZE, p, rc);

	if (rc) {
		printk(KERN_ERR "DT_GETPAGE failed!\n");
		goto end;
	}
	BT_MARK_DIRTY(mp, inode);

	ASSERT(p->header.flag & BT_LEAF);

	tlck = txLock(tid, inode, mp, tlckDTREE | tlckENTRY);
	if (BT_IS_ROOT(mp))
		tlck->type |= tlckBTROOT;

	dtlck = (struct dt_lock *) &tlck->lock;

	stbl = DT_GETSTBL(p);
	for (i = 0; i < p->header.nextindex; i++) {
		d = (struct ldtentry *) &p->slot[stbl[i]];
		index = le32_to_cpu(d->index);
		if ((index < 2) || (index >= JFS_IP(inode)->next_index)) {
			d->index = cpu_to_le32(add_index(tid, inode, bn, i));
			if (dtlck->index >= dtlck->maxcnt)
				dtlck = (struct dt_lock *) txLinelock(dtlck);
			lv = &dtlck->lv[dtlck->index];
			lv->offset = stbl[i];
			lv->length = 1;
			dtlck->index++;
		}
	}

	DT_PUTPAGE(mp);
	(void) txCommit(tid, 1, &inode, 0);
end:
	txEnd(tid);
}

/*
 * Buffer to hold directory entry info while traversing a dtree page
 * before being fed to the filldir function
 */
struct jfs_dirent {
	loff_t position;
	int ino;
	u16 name_len;
	char name[0];
};

/*
 * function to determine next variable-sized jfs_dirent in buffer
 */
static inline struct jfs_dirent *next_jfs_dirent(struct jfs_dirent *dirent)
{
	return (struct jfs_dirent *)
		((char *)dirent +
		 ((sizeof (struct jfs_dirent) + dirent->name_len + 1 +
		   sizeof (loff_t) - 1) &
		  ~(sizeof (loff_t) - 1)));
}

/*
 *	jfs_readdir()
 *
 * function: read directory entries sequentially
 *	from the specified entry offset
 *
 * parameter:
 *
 * return: offset = (pn, index) of start entry
 *	of next jfs_readdir()/dtRead()
 */
int jfs_readdir(struct file *file, struct dir_context *ctx)
{
	struct inode *ip = file_inode(file);
	struct nls_table *codepage = JFS_SBI(ip->i_sb)->nls_tab;
	int rc = 0;
	loff_t dtpos;	/* legacy OS/2 style position */
	struct dtoffset {
		s16 pn;
		s16 index;
		s32 unused;
	} *dtoffset = (struct dtoffset *) &dtpos;
	s64 bn;
	struct metapage *mp;
	dtpage_t *p;
	int index;
	s8 *stbl;
	struct btstack btstack;
	int i, next;
	struct ldtentry *d;
	struct dtslot *t;
	int d_namleft, len, outlen;
	unsigned long dirent_buf;
	char *name_ptr;
	u32 dir_index;
	int do_index = 0;
	uint loop_count = 0;
	struct jfs_dirent *jfs_dirent;
	int jfs_dirents;
	int overflow, fix_page, page_fixed = 0;
	static int unique_pos = 2;	/* If we can't fix broken index */

	if (ctx->pos == DIREND)
		return 0;

	if (DO_INDEX(ip)) {
		/*
		 * persistent index is stored in directory entries.
		 * Special cases:	 0 = .
		 *			 1 = ..
		 *			-1 = End of directory
		 */
		do_index = 1;

		dir_index = (u32) ctx->pos;

		/*
		 * NFSv4 reserves cookies 1 and 2 for . and .. so the value
		 * we return to the vfs is one greater than the one we use
		 * internally.
		 */
		if (dir_index)
			dir_index--;

		if (dir_index > 1) {
			struct dir_table_slot dirtab_slot;

			if (dtEmpty(ip) ||
			    (dir_index >= JFS_IP(ip)->next_index)) {
				/* Stale position.  Directory has shrunk */
				ctx->pos = DIREND;
				return 0;
			}
		      repeat:
			rc = read_index(ip, dir_index, &dirtab_slot);
			if (rc) {
				ctx->pos = DIREND;
				return rc;
			}
			if (dirtab_slot.flag == DIR_INDEX_FREE) {
				if (loop_count++ > JFS_IP(ip)->next_index) {
					jfs_err("jfs_readdir detected infinite loop!");
					ctx->pos = DIREND;
					return 0;
				}
				dir_index = le32_to_cpu(dirtab_slot.addr2);
				if (dir_index == -1) {
					ctx->pos = DIREND;
					return 0;
				}
				goto repeat;
			}
			bn = addressDTS(&dirtab_slot);
			index = dirtab_slot.slot;
			DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
			if (rc) {
				ctx->pos = DIREND;
				return 0;
			}
			if (p->header.flag & BT_INTERNAL) {
				jfs_err("jfs_readdir: bad index table");
				DT_PUTPAGE(mp);
				ctx->pos = DIREND;
				return 0;
			}
		} else {
			if (dir_index == 0) {
				/*
				 * self "."
				 */
				ctx->pos = 1;
				if (!dir_emit(ctx, ".", 1, ip->i_ino, DT_DIR))
					return 0;
			}
			/*
			 * parent ".."
			 */
			ctx->pos = 2;
			if (!dir_emit(ctx, "..", 2, PARENT(ip), DT_DIR))
				return 0;

			/*
			 * Find first entry of left-most leaf
			 */
			if (dtEmpty(ip)) {
				ctx->pos = DIREND;
				return 0;
			}

			if ((rc = dtReadFirst(ip, &btstack)))
				return rc;

			DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);
		}
	} else {
		/*
		 * Legacy filesystem - OS/2 & Linux JFS < 0.3.6
		 *
		 * pn = 0; index = 1:	First entry "."
		 * pn = 0; index = 2:	Second entry ".."
		 * pn > 0:		Real entries, pn=1 -> leftmost page
		 * pn = index = -1:	No more entries
		 */
		dtpos = ctx->pos;
		if (dtpos < 2) {
			/* build "." entry */
			ctx->pos = 1;
			if (!dir_emit(ctx, ".", 1, ip->i_ino, DT_DIR))
				return 0;
			dtoffset->index = 2;
			ctx->pos = dtpos;
		}

		if (dtoffset->pn == 0) {
			if (dtoffset->index == 2) {
				/* build ".." entry */
				if (!dir_emit(ctx, "..", 2, PARENT(ip), DT_DIR))
					return 0;
			} else {
				jfs_err("jfs_readdir called with invalid offset!");
			}
			dtoffset->pn = 1;
			dtoffset->index = 0;
			ctx->pos = dtpos;
		}

		if (dtEmpty(ip)) {
			ctx->pos = DIREND;
			return 0;
		}

		if ((rc = dtReadNext(ip, &ctx->pos, &btstack))) {
			jfs_err("jfs_readdir: unexpected rc = %d from dtReadNext",
				rc);
			ctx->pos = DIREND;
			return 0;
		}
		/* get start leaf page and index */
		DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);

		/* offset beyond directory eof ? */
		if (bn < 0) {
			ctx->pos = DIREND;
			return 0;
		}
	}

	dirent_buf = __get_free_page(GFP_KERNEL);
	if (dirent_buf == 0) {
		DT_PUTPAGE(mp);
		jfs_warn("jfs_readdir: __get_free_page failed!");
		ctx->pos = DIREND;
		return -ENOMEM;
	}

	while (1) {
		jfs_dirent = (struct jfs_dirent *) dirent_buf;
		jfs_dirents = 0;
		overflow = fix_page = 0;

		stbl = DT_GETSTBL(p);

		for (i = index; i < p->header.nextindex; i++) {
			d = (struct ldtentry *) & p->slot[stbl[i]];

			if (((long) jfs_dirent + d->namlen + 1) >
			    (dirent_buf + PAGE_SIZE)) {
				/* DBCS codepages could overrun dirent_buf */
				index = i;
				overflow = 1;
				break;
			}

			d_namleft = d->namlen;
			name_ptr = jfs_dirent->name;
			jfs_dirent->ino = le32_to_cpu(d->inumber);

			if (do_index) {
				len = min(d_namleft, DTLHDRDATALEN);
				jfs_dirent->position = le32_to_cpu(d->index);
				/*
				 * d->index should always be valid, but it
				 * isn't.  fsck.jfs doesn't create the
				 * directory index for the lost+found
				 * directory.  Rather than let it go,
				 * we can try to fix it.
				 */
				if ((jfs_dirent->position < 2) ||
				    (jfs_dirent->position >=
				     JFS_IP(ip)->next_index)) {
					if (!page_fixed && !isReadOnly(ip)) {
						fix_page = 1;
						/*
						 * setting overflow and setting
						 * index to i will cause the
						 * same page to be processed
						 * again starting here
						 */
						overflow = 1;
						index = i;
						break;
					}
					jfs_dirent->position = unique_pos++;
				}
				/*
				 * We add 1 to the index because we may
				 * use a value of 2 internally, and NFSv4
				 * doesn't like that.
				 */
				jfs_dirent->position++;
			} else {
				jfs_dirent->position = dtpos;
				len = min(d_namleft, DTLHDRDATALEN_LEGACY);
			}

			/* copy the name of head/only segment */
			outlen = jfs_strfromUCS_le(name_ptr, d->name, len,
						   codepage);
			jfs_dirent->name_len = outlen;

			/* copy name in the additional segment(s) */
			next = d->next;
			while (next >= 0) {
				t = (struct dtslot *) & p->slot[next];
				name_ptr += outlen;
				d_namleft -= len;
				/* Sanity Check */
				if (d_namleft == 0) {
					jfs_error(ip->i_sb,
						  "JFS:Dtree error: ino = %ld, bn=%lld, index = %d\n",
						  (long)ip->i_ino,
						  (long long)bn,
						  i);
					goto skip_one;
				}
				len = min(d_namleft, DTSLOTDATALEN);
				outlen = jfs_strfromUCS_le(name_ptr, t->name,
							   len, codepage);
				jfs_dirent->name_len += outlen;

				next = t->next;
			}

			jfs_dirents++;
			jfs_dirent = next_jfs_dirent(jfs_dirent);
skip_one:
			if (!do_index)
				dtoffset->index++;
		}

		if (!overflow) {
			/* Point to next leaf page */
			if (p->header.flag & BT_ROOT)
				bn = 0;
			else {
				bn = le64_to_cpu(p->header.next);
				index = 0;
				/* update offset (pn:index) for new page */
				if (!do_index) {
					dtoffset->pn++;
					dtoffset->index = 0;
				}
			}
			page_fixed = 0;
		}

		/* unpin previous leaf page */
		DT_PUTPAGE(mp);

		jfs_dirent = (struct jfs_dirent *) dirent_buf;
		while (jfs_dirents--) {
			ctx->pos = jfs_dirent->position;
			if (!dir_emit(ctx, jfs_dirent->name,
				    jfs_dirent->name_len,
				    jfs_dirent->ino, DT_UNKNOWN))
				goto out;
			jfs_dirent = next_jfs_dirent(jfs_dirent);
		}

		if (fix_page) {
			add_missing_indices(ip, bn);
			page_fixed = 1;
		}

		if (!overflow && (bn == 0)) {
			ctx->pos = DIREND;
			break;
		}

		DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
		if (rc) {
			free_page(dirent_buf);
			return rc;
		}
	}

      out:
	free_page(dirent_buf);

	return rc;
}


/*
 *	dtReadFirst()
 *
 * function: get the leftmost page of the directory
 */
static int dtReadFirst(struct inode *ip, struct btstack * btstack)
{
	int rc = 0;
	s64 bn;
	int psize = 288;	/* initial in-line directory */
	struct metapage *mp;
	dtpage_t *p;
	s8 *stbl;
	struct btframe *btsp;
	pxd_t *xd;

	BT_CLR(btstack);	/* reset stack */

	/*
	 *	descend leftmost path of the tree
	 *
	 * by convention, root bn = 0.
	 */
	for (bn = 0;;) {
		DT_GETPAGE(ip, bn, mp, psize, p, rc);
		if (rc)
			return rc;

		/*
		 * leftmost leaf page
		 */
		if (p->header.flag & BT_LEAF) {
			/* return leftmost entry */
			btsp = btstack->top;
			btsp->bn = bn;
			btsp->index = 0;
			btsp->mp = mp;

			return 0;
		}

		/*
		 * descend down to leftmost child page
		 */
		if (BT_STACK_FULL(btstack)) {
			DT_PUTPAGE(mp);
			jfs_error(ip->i_sb, "btstack overrun\n");
			BT_STACK_DUMP(btstack);
			return -EIO;
		}
		/* push (bn, index) of the parent page/entry */
		BT_PUSH(btstack, bn, 0);

		/* get the leftmost entry */
		stbl = DT_GETSTBL(p);
		xd = (pxd_t *) & p->slot[stbl[0]];

		/* get the child page block address */
		bn = addressPXD(xd);
		psize = lengthPXD(xd) << JFS_SBI(ip->i_sb)->l2bsize;

		/* unpin the parent page */
		DT_PUTPAGE(mp);
	}
}


/*
 *	dtReadNext()
 *
 * function: get the page of the specified offset (pn:index)
 *
 * return: if (offset > eof), bn = -1;
 *
 * note: if index > nextindex of the target leaf page,
 * start with 1st entry of next leaf page;
 */
static int dtReadNext(struct inode *ip, loff_t * offset,
		      struct btstack * btstack)
{
	int rc = 0;
	struct dtoffset {
		s16 pn;
		s16 index;
		s32 unused;
	} *dtoffset = (struct dtoffset *) offset;
	s64 bn;
	struct metapage *mp;
	dtpage_t *p;
	int index;
	int pn;
	s8 *stbl;
	struct btframe *btsp, *parent;
	pxd_t *xd;

	/*
	 * get leftmost leaf page pinned
	 */
	if ((rc = dtReadFirst(ip, btstack)))
		return rc;

	/* get leaf page */
	DT_GETSEARCH(ip, btstack->top, bn, mp, p, index);

	/* get the start offset (pn:index) */
	pn = dtoffset->pn - 1;	/* Now pn = 0 represents leftmost leaf */
	index = dtoffset->index;

	/* start at leftmost page ? */
	if (pn == 0) {
		/* offset beyond eof ? */
		if (index < p->header.nextindex)
			goto out;

		if (p->header.flag & BT_ROOT) {
			bn = -1;
			goto out;
		}

		/* start with 1st entry of next leaf page */
		dtoffset->pn++;
		dtoffset->index = index = 0;
		goto a;
	}

	/* start at non-leftmost page: scan parent pages for large pn */
	if (p->header.flag & BT_ROOT) {
		bn = -1;
		goto out;
	}

	/* start after next leaf page ? */
	if (pn > 1)
		goto b;

	/* get leaf page pn = 1 */
      a:
	bn = le64_to_cpu(p->header.next);

	/* unpin leaf page */
	DT_PUTPAGE(mp);

	/* offset beyond eof ? */
	if (bn == 0) {
		bn = -1;
		goto out;
	}

	goto c;

	/*
	 * scan last internal page level to get target leaf page
	 */
      b:
	/* unpin leftmost leaf page */
	DT_PUTPAGE(mp);

	/* get left most parent page */
	btsp = btstack->top;
	parent = btsp - 1;
	bn = parent->bn;
	DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
	if (rc)
		return rc;

	/* scan parent pages at last internal page level */
	while (pn >= p->header.nextindex) {
		pn -= p->header.nextindex;

		/* get next parent page address */
		bn = le64_to_cpu(p->header.next);

		/* unpin current parent page */
		DT_PUTPAGE(mp);

		/* offset beyond eof ? */
		if (bn == 0) {
			bn = -1;
			goto out;
		}

		/* get next parent page */
		DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
		if (rc)
			return rc;

		/* update parent page stack frame */
		parent->bn = bn;
	}

	/* get leaf page address */
	stbl = DT_GETSTBL(p);
	xd = (pxd_t *) & p->slot[stbl[pn]];
	bn = addressPXD(xd);

	/* unpin parent page */
	DT_PUTPAGE(mp);

	/*
	 * get target leaf page
	 */
      c:
	DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
	if (rc)
		return rc;

	/*
	 * leaf page has been completed:
	 * start with 1st entry of next leaf page
	 */
	if (index >= p->header.nextindex) {
		bn = le64_to_cpu(p->header.next);

		/* unpin leaf page */
		DT_PUTPAGE(mp);

		/* offset beyond eof ? */
		if (bn == 0) {
			bn = -1;
			goto out;
		}

		/* get next leaf page */
		DT_GETPAGE(ip, bn, mp, PSIZE, p, rc);
		if (rc)
			return rc;

		/* start with 1st entry of next leaf page */
		dtoffset->pn++;
		dtoffset->index = 0;
	}

      out:
	/* return target leaf page pinned */
	btsp = btstack->top;
	btsp->bn = bn;
	btsp->index = dtoffset->index;
	btsp->mp = mp;

	return 0;
}


/*
 *	dtCompare()
 *
 * function: compare search key with an internal entry
 *
 * return:
 *	< 0 if k is < record
 *	= 0 if k is = record
 *	> 0 if k is > record
 */
static int dtCompare(struct component_name * key,	/* search key */
		     dtpage_t * p,	/* directory page */
		     int si)
{				/* entry slot index */
	wchar_t *kname;
	__le16 *name;
	int klen, namlen, len, rc;
	struct idtentry *ih;
	struct dtslot *t;

	/*
	 * force the left-most key on internal pages, at any level of
	 * the tree, to be less than any search key.
	 * this obviates having to update the leftmost key on an internal
	 * page when the user inserts a new key in the tree smaller than
	 * anything that has been stored.
	 *
	 * (? if/when dtSearch() narrows down to 1st entry (index = 0),
	 * at any internal page at any level of the tree,
	 * it descends to child of the entry anyway -
	 * ? make the entry as min size dummy entry)
	 *
	 * if (e->index == 0 && h->prevpg == P_INVALID && !(h->flags & BT_LEAF))
	 * return (1);
	 */

	kname = key->name;
	klen = key->namlen;

	ih = (struct idtentry *) & p->slot[si];
	si = ih->next;
	name = ih->name;
	namlen = ih->namlen;
	len = min(namlen, DTIHDRDATALEN);

	/* compare with head/only segment */
	len = min(klen, len);
	if ((rc = UniStrncmp_le(kname, name, len)))
		return rc;

	klen -= len;
	namlen -= len;

	/* compare with additional segment(s) */
	kname += len;
	while (klen > 0 && namlen > 0) {
		/* compare with next name segment */
		t = (struct dtslot *) & p->slot[si];
		len = min(namlen, DTSLOTDATALEN);
		len = min(klen, len);
		name = t->name;
		if ((rc = UniStrncmp_le(kname, name, len)))
			return rc;

		klen -= len;
		namlen -= len;
		kname += len;
		si = t->next;
	}

	return (klen - namlen);
}




/*
 *	ciCompare()
 *
 * function: compare search key with an (leaf/internal) entry
 *
 * return:
 *	< 0 if k is < record
 *	= 0 if k is = record
 *	> 0 if k is > record
 */
static int ciCompare(struct component_name * key,	/* search key */
		     dtpage_t * p,	/* directory page */
		     int si,	/* entry slot index */
		     int flag)
{
	wchar_t *kname, x;
	__le16 *name;
	int klen, namlen, len, rc;
	struct ldtentry *lh;
	struct idtentry *ih;
	struct dtslot *t;
	int i;

	/*
	 * force the left-most key on internal pages, at any level of
	 * the tree, to be less than any search key.
	 * this obviates having to update the leftmost key on an internal
	 * page when the user inserts a new key in the tree smaller than
	 * anything that has been stored.
	 *
	 * (? if/when dtSearch() narrows down to 1st entry (index = 0),
	 * at any internal page at any level of the tree,
	 * it descends to child of the entry anyway -
	 * ? make the entry as min size dummy entry)
	 *
	 * if (e->index == 0 && h->prevpg == P_INVALID && !(h->flags & BT_LEAF))
	 * return (1);
	 */

	kname = key->name;
	klen = key->namlen;

	/*
	 * leaf page entry
	 */
	if (p->header.flag & BT_LEAF) {
		lh = (struct ldtentry *) & p->slot[si];
		si = lh->next;
		name = lh->name;
		namlen = lh->namlen;
		if (flag & JFS_DIR_INDEX)
			len = min(namlen, DTLHDRDATALEN);
		else
			len = min(namlen, DTLHDRDATALEN_LEGACY);
	}
	/*
	 * internal page entry
	 */
	else {
		ih = (struct idtentry *) & p->slot[si];
		si = ih->next;
		name = ih->name;
		namlen = ih->namlen;
		len = min(namlen, DTIHDRDATALEN);
	}

	/* compare with head/only segment */
	len = min(klen, len);
	for (i = 0; i < len; i++, kname++, name++) {
		/* only uppercase if case-insensitive support is on */
		if ((flag & JFS_OS2) == JFS_OS2)
			x = UniToupper(le16_to_cpu(*name));
		else
			x = le16_to_cpu(*name);
		if ((rc = *kname - x))
			return rc;
	}

	klen -= len;
	namlen -= len;

	/* compare with additional segment(s) */
	while (klen > 0 && namlen > 0) {
		/* compare with next name segment */
		t = (struct dtslot *) & p->slot[si];
		len = min(namlen, DTSLOTDATALEN);
		len = min(klen, len);
		name = t->name;
		for (i = 0; i < len; i++, kname++, name++) {
			/* only uppercase if case-insensitive support is on */
			if ((flag & JFS_OS2) == JFS_OS2)
				x = UniToupper(le16_to_cpu(*name));
			else
				x = le16_to_cpu(*name);

			if ((rc = *kname - x))
				return rc;
		}

		klen -= len;
		namlen -= len;
		si = t->next;
	}

	return (klen - namlen);
}


/*
 *	ciGetLeafPrefixKey()
 *
 * function: compute prefix of suffix compression
 *	     from two adjacent leaf entries
 *	     across page boundary
 *
 * return: non-zero on error
 *
 */
static int ciGetLeafPrefixKey(dtpage_t * lp, int li, dtpage_t * rp,
			       int ri, struct component_name * key, int flag)
{
	int klen, namlen;
	wchar_t *pl, *pr, *kname;
	struct component_name lkey;
	struct component_name rkey;

	lkey.name = kmalloc_array(JFS_NAME_MAX + 1, sizeof(wchar_t),
					GFP_KERNEL);
	if (lkey.name == NULL)
		return -ENOMEM;

	rkey.name = kmalloc_array(JFS_NAME_MAX + 1, sizeof(wchar_t),
					GFP_KERNEL);
	if (rkey.name == NULL) {
		kfree(lkey.name);
		return -ENOMEM;
	}

	/* get left and right key */
	dtGetKey(lp, li, &lkey, flag);
	lkey.name[lkey.namlen] = 0;

	if ((flag & JFS_OS2) == JFS_OS2)
		ciToUpper(&lkey);

	dtGetKey(rp, ri, &rkey, flag);
	rkey.name[rkey.namlen] = 0;


	if ((flag & JFS_OS2) == JFS_OS2)
		ciToUpper(&rkey);

	/* compute prefix */
	klen = 0;
	kname = key->name;
	namlen = min(lkey.namlen, rkey.namlen);
	for (pl = lkey.name, pr = rkey.name;
	     namlen; pl++, pr++, namlen--, klen++, kname++) {
		*kname = *pr;
		if (*pl != *pr) {
			key->namlen = klen + 1;
			goto free_names;
		}
	}

	/* l->namlen <= r->namlen since l <= r */
	if (lkey.namlen < rkey.namlen) {
		*kname = *pr;
		key->namlen = klen + 1;
	} else			/* l->namelen == r->namelen */
		key->namlen = klen;

free_names:
	kfree(lkey.name);
	kfree(rkey.name);
	return 0;
}



/*
 *	dtGetKey()
 *
 * function: get key of the entry
 */
static void dtGetKey(dtpage_t * p, int i,	/* entry index */
		     struct component_name * key, int flag)
{
	int si;
	s8 *stbl;
	struct ldtentry *lh;
	struct idtentry *ih;
	struct dtslot *t;
	int namlen, len;
	wchar_t *kname;
	__le16 *name;

	/* get entry */
	stbl = DT_GETSTBL(p);
	si = stbl[i];
	if (p->header.flag & BT_LEAF) {
		lh = (struct ldtentry *) & p->slot[si];
		si = lh->next;
		namlen = lh->namlen;
		name = lh->name;
		if (flag & JFS_DIR_INDEX)
			len = min(namlen, DTLHDRDATALEN);
		else
			len = min(namlen, DTLHDRDATALEN_LEGACY);
	} else {
		ih = (struct idtentry *) & p->slot[si];
		si = ih->next;
		namlen = ih->namlen;
		name = ih->name;
		len = min(namlen, DTIHDRDATALEN);
	}

	key->namlen = namlen;
	kname = key->name;

	/*
	 * move head/only segment
	 */
	UniStrncpy_from_le(kname, name, len);

	/*
	 * move additional segment(s)
	 */
	while (si >= 0) {
		/* get next segment */
		t = &p->slot[si];
		kname += len;
		namlen -= len;
		len = min(namlen, DTSLOTDATALEN);
		UniStrncpy_from_le(kname, t->name, len);

		si = t->next;
	}
}


/*
 *	dtInsertEntry()
 *
 * function: allocate free slot(s) and
 *	     write a leaf/internal entry
 *
 * return: entry slot index
 */
static void dtInsertEntry(dtpage_t * p, int index, struct component_name * key,
			  ddata_t * data, struct dt_lock ** dtlock)
{
	struct dtslot *h, *t;
	struct ldtentry *lh = NULL;
	struct idtentry *ih = NULL;
	int hsi, fsi, klen, len, nextindex;
	wchar_t *kname;
	__le16 *name;
	s8 *stbl;
	pxd_t *xd;
	struct dt_lock *dtlck = *dtlock;
	struct lv *lv;
	int xsi, n;
	s64 bn = 0;
	struct metapage *mp = NULL;

	klen = key->namlen;
	kname = key->name;

	/* allocate a free slot */
	hsi = fsi = p->header.freelist;
	h = &p->slot[fsi];
	p->header.freelist = h->next;
	--p->header.freecnt;

	/* open new linelock */
	if (dtlck->index >= dtlck->maxcnt)
		dtlck = (struct dt_lock *) txLinelock(dtlck);

	lv = & dtlck->lv[dtlck->index];
	lv->offset = hsi;

	/* write head/only segment */
	if (p->header.flag & BT_LEAF) {
		lh = (struct ldtentry *) h;
		lh->next = h->next;
		lh->inumber = cpu_to_le32(data->leaf.ino);
		lh->namlen = klen;
		name = lh->name;
		if (data->leaf.ip) {
			len = min(klen, DTLHDRDATALEN);
			if (!(p->header.flag & BT_ROOT))
				bn = addressPXD(&p->header.self);
			lh->index = cpu_to_le32(add_index(data->leaf.tid,
							  data->leaf.ip,
							  bn, index));
		} else
			len = min(klen, DTLHDRDATALEN_LEGACY);
	} else {
		ih = (struct idtentry *) h;
		ih->next = h->next;
		xd = (pxd_t *) ih;
		*xd = data->xd;
		ih->namlen = klen;
		name = ih->name;
		len = min(klen, DTIHDRDATALEN);
	}

	UniStrncpy_to_le(name, kname, len);

	n = 1;
	xsi = hsi;

	/* write additional segment(s) */
	t = h;
	klen -= len;
	while (klen) {
		/* get free slot */
		fsi = p->header.freelist;
		t = &p->slot[fsi];
		p->header.freelist = t->next;
		--p->header.freecnt;

		/* is next slot contiguous ? */
		if (fsi != xsi + 1) {
			/* close current linelock */
			lv->length = n;
			dtlck->index++;

			/* open new linelock */
			if (dtlck->index < dtlck->maxcnt)
				lv++;
			else {
				dtlck = (struct dt_lock *) txLinelock(dtlck);
				lv = & dtlck->lv[0];
			}

			lv->offset = fsi;
			n = 0;
		}

		kname += len;
		len = min(klen, DTSLOTDATALEN);
		UniStrncpy_to_le(t->name, kname, len);

		n++;
		xsi = fsi;
		klen -= len;
	}

	/* close current linelock */
	lv->length = n;
	dtlck->index++;

	*dtlock = dtlck;

	/* terminate last/only segment */
	if (h == t) {
		/* single segment entry */
		if (p->header.flag & BT_LEAF)
			lh->next = -1;
		else
			ih->next = -1;
	} else
		/* multi-segment entry */
		t->next = -1;

	/* if insert into middle, shift right succeeding entries in stbl */
	stbl = DT_GETSTBL(p);
	nextindex = p->header.nextindex;
	if (index < nextindex) {
		memmove(stbl + index + 1, stbl + index, nextindex - index);

		if ((p->header.flag & BT_LEAF) && data->leaf.ip) {
			s64 lblock;

			/*
			 * Need to update slot number for entries that moved
			 * in the stbl
			 */
			mp = NULL;
			for (n = index + 1; n <= nextindex; n++) {
				lh = (struct ldtentry *) & (p->slot[stbl[n]]);
				modify_index(data->leaf.tid, data->leaf.ip,
					     le32_to_cpu(lh->index), bn, n,
					     &mp, &lblock);
			}
			if (mp)
				release_metapage(mp);
		}
	}

	stbl[index] = hsi;

	/* advance next available entry index of stbl */
	++p->header.nextindex;
}


/*
 *	dtMoveEntry()
 *
 * function: move entries from split/left page to new/right page
 *
 *	nextindex of dst page and freelist/freecnt of both pages
 *	are updated.
 */
static void dtMoveEntry(dtpage_t * sp, int si, dtpage_t * dp,
			struct dt_lock ** sdtlock, struct dt_lock ** ddtlock,
			int do_index)
{
	int ssi, next;		/* src slot index */
	int di;			/* dst entry index */
	int dsi;		/* dst slot index */
	s8 *sstbl, *dstbl;	/* sorted entry table */
	int snamlen, len;
	struct ldtentry *slh, *dlh = NULL;
	struct idtentry *sih, *dih = NULL;
	struct dtslot *h, *s, *d;
	struct dt_lock *sdtlck = *sdtlock, *ddtlck = *ddtlock;
	struct lv *slv, *dlv;
	int xssi, ns, nd;
	int sfsi;

	sstbl = (s8 *) & sp->slot[sp->header.stblindex];
	dstbl = (s8 *) & dp->slot[dp->header.stblindex];

	dsi = dp->header.freelist;	/* first (whole page) free slot */
	sfsi = sp->header.freelist;

	/* linelock destination entry slot */
	dlv = & ddtlck->lv[ddtlck->index];
	dlv->offset = dsi;

	/* linelock source entry slot */
	slv = & sdtlck->lv[sdtlck->index];
	slv->offset = sstbl[si];
	xssi = slv->offset - 1;

	/*
	 * move entries
	 */
	ns = nd = 0;
	for (di = 0; si < sp->header.nextindex; si++, di++) {
		ssi = sstbl[si];
		dstbl[di] = dsi;

		/* is next slot contiguous ? */
		if (ssi != xssi + 1) {
			/* close current linelock */
			slv->length = ns;
			sdtlck->index++;

			/* open new linelock */
			if (sdtlck->index < sdtlck->maxcnt)
				slv++;
			else {
				sdtlck = (struct dt_lock *) txLinelock(sdtlck);
				slv = & sdtlck->lv[0];
			}

			slv->offset = ssi;
			ns = 0;
		}

		/*
		 * move head/only segment of an entry
		 */
		/* get dst slot */
		h = d = &dp->slot[dsi];

		/* get src slot and move */
		s = &sp->slot[ssi];
		if (sp->header.flag & BT_LEAF) {
			/* get source entry */
			slh = (struct ldtentry *) s;
			dlh = (struct ldtentry *) h;
			snamlen = slh->namlen;

			if (do_index) {
				len = min(snamlen, DTLHDRDATALEN);
				dlh->index = slh->index; /* little-endian */
			} else
				len = min(snamlen, DTLHDRDATALEN_LEGACY);

			memcpy(dlh, slh, 6 + len * 2);

			next = slh->next;

			/* update dst head/only segment next field */
			dsi++;
			dlh->next = dsi;
		} else {
			sih = (struct idtentry *) s;
			snamlen = sih->namlen;

			len = min(snamlen, DTIHDRDATALEN);
			dih = (struct idtentry *) h;
			memcpy(dih, sih, 10 + len * 2);
			next = sih->next;

			dsi++;
			dih->next = dsi;
		}

		/* free src head/only segment */
		s->next = sfsi;
		s->cnt = 1;
		sfsi = ssi;

		ns++;
		nd++;
		xssi = ssi;

		/*
		 * move additional segment(s) of the entry
		 */
		snamlen -= len;
		while ((ssi = next) >= 0) {
			/* is next slot contiguous ? */
			if (ssi != xssi + 1) {
				/* close current linelock */
				slv->length = ns;
				sdtlck->index++;

				/* open new linelock */
				if (sdtlck->index < sdtlck->maxcnt)
					slv++;
				else {
					sdtlck =
					    (struct dt_lock *)
					    txLinelock(sdtlck);
					slv = & sdtlck->lv[0];
				}

				slv->offset = ssi;
				ns = 0;
			}

			/* get next source segment */
			s = &sp->slot[ssi];

			/* get next destination free slot */
			d++;

			len = min(snamlen, DTSLOTDATALEN);
			UniStrncpy_le(d->name, s->name, len);

			ns++;
			nd++;
			xssi = ssi;

			dsi++;
			d->next = dsi;

			/* free source segment */
			next = s->next;
			s->next = sfsi;
			s->cnt = 1;
			sfsi = ssi;

			snamlen -= len;
		}		/* end while */

		/* terminate dst last/only segment */
		if (h == d) {
			/* single segment entry */
			if (dp->header.flag & BT_LEAF)
				dlh->next = -1;
			else
				dih->next = -1;
		} else
			/* multi-segment entry */
			d->next = -1;
	}			/* end for */

	/* close current linelock */
	slv->length = ns;
	sdtlck->index++;
	*sdtlock = sdtlck;

	dlv->length = nd;
	ddtlck->index++;
	*ddtlock = ddtlck;

	/* update source header */
	sp->header.freelist = sfsi;
	sp->header.freecnt += nd;

	/* update destination header */
	dp->header.nextindex = di;

	dp->header.freelist = dsi;
	dp->header.freecnt -= nd;
}


/*
 *	dtDeleteEntry()
 *
 * function: free a (leaf/internal) entry
 *
 * log freelist header, stbl, and each segment slot of entry
 * (even though last/only segment next field is modified,
 * physical image logging requires all segment slots of
 * the entry logged to avoid applying previous updates
 * to the same slots)
 */
static void dtDeleteEntry(dtpage_t * p, int fi, struct dt_lock ** dtlock)
{
	int fsi;		/* free entry slot index */
	s8 *stbl;
	struct dtslot *t;
	int si, freecnt;
	struct dt_lock *dtlck = *dtlock;
	struct lv *lv;
	int xsi, n;

	/* get free entry slot index */
	stbl = DT_GETSTBL(p);
	fsi = stbl[fi];

	/* open new linelock */
	if (dtlck->index >= dtlck->maxcnt)
		dtlck = (struct dt_lock *) txLinelock(dtlck);
	lv = & dtlck->lv[dtlck->index];

	lv->offset = fsi;

	/* get the head/only segment */
	t = &p->slot[fsi];
	if (p->header.flag & BT_LEAF)
		si = ((struct ldtentry *) t)->next;
	else
		si = ((struct idtentry *) t)->next;
	t->next = si;
	t->cnt = 1;

	n = freecnt = 1;
	xsi = fsi;

	/* find the last/only segment */
	while (si >= 0) {
		/* is next slot contiguous ? */
		if (si != xsi + 1) {
			/* close current linelock */
			lv->length = n;
			dtlck->index++;

			/* open new linelock */
			if (dtlck->index < dtlck->maxcnt)
				lv++;
			else {
				dtlck = (struct dt_lock *) txLinelock(dtlck);
				lv = & dtlck->lv[0];
			}

			lv->offset = si;
			n = 0;
		}

		n++;
		xsi = si;
		freecnt++;

		t = &p->slot[si];
		t->cnt = 1;
		si = t->next;
	}

	/* close current linelock */
	lv->length = n;
	dtlck->index++;

	*dtlock = dtlck;

	/* update freelist */
	t->next = p->header.freelist;
	p->header.freelist = fsi;
	p->header.freecnt += freecnt;

	/* if delete from middle,
	 * shift left the succedding entries in the stbl
	 */
	si = p->header.nextindex;
	if (fi < si - 1)
		memmove(&stbl[fi], &stbl[fi + 1], si - fi - 1);

	p->header.nextindex--;
}


/*
 *	dtTruncateEntry()
 *
 * function: truncate a (leaf/internal) entry
 *
 * log freelist header, stbl, and each segment slot of entry
 * (even though last/only segment next field is modified,
 * physical image logging requires all segment slots of
 * the entry logged to avoid applying previous updates
 * to the same slots)
 */
static void dtTruncateEntry(dtpage_t * p, int ti, struct dt_lock ** dtlock)
{
	int tsi;		/* truncate entry slot index */
	s8 *stbl;
	struct dtslot *t;
	int si, freecnt;
	struct dt_lock *dtlck = *dtlock;
	struct lv *lv;
	int fsi, xsi, n;

	/* get free entry slot index */
	stbl = DT_GETSTBL(p);
	tsi = stbl[ti];

	/* open new linelock */
	if (dtlck->index >= dtlck->maxcnt)
		dtlck = (struct dt_lock *) txLinelock(dtlck);
	lv = & dtlck->lv[dtlck->index];

	lv->offset = tsi;

	/* get the head/only segment */
	t = &p->slot[tsi];
	ASSERT(p->header.flag & BT_INTERNAL);
	((struct idtentry *) t)->namlen = 0;
	si = ((struct idtentry *) t)->next;
	((struct idtentry *) t)->next = -1;

	n = 1;
	freecnt = 0;
	fsi = si;
	xsi = tsi;

	/* find the last/only segment */
	while (si >= 0) {
		/* is next slot contiguous ? */
		if (si != xsi + 1) {
			/* close current linelock */
			lv->length = n;
			dtlck->index++;

			/* open new linelock */
			if (dtlck->index < dtlck->maxcnt)
				lv++;
			else {
				dtlck = (struct dt_lock *) txLinelock(dtlck);
				lv = & dtlck->lv[0];
			}

			lv->offset = si;
			n = 0;
		}

		n++;
		xsi = si;
		freecnt++;

		t = &p->slot[si];
		t->cnt = 1;
		si = t->next;
	}

	/* close current linelock */
	lv->length = n;
	dtlck->index++;

	*dtlock = dtlck;

	/* update freelist */
	if (freecnt == 0)
		return;
	t->next = p->header.freelist;
	p->header.freelist = fsi;
	p->header.freecnt += freecnt;
}


/*
 *	dtLinelockFreelist()
 */
static void dtLinelockFreelist(dtpage_t * p,	/* directory page */
			       int m,	/* max slot index */
			       struct dt_lock ** dtlock)
{
	int fsi;		/* free entry slot index */
	struct dtslot *t;
	int si;
	struct dt_lock *dtlck = *dtlock;
	struct lv *lv;
	int xsi, n;

	/* get free entry slot index */
	fsi = p->header.freelist;

	/* open new linelock */
	if (dtlck->index >= dtlck->maxcnt)
		dtlck = (struct dt_lock *) txLinelock(dtlck);
	lv = & dtlck->lv[dtlck->index];

	lv->offset = fsi;

	n = 1;
	xsi = fsi;

	t = &p->slot[fsi];
	si = t->next;

	/* find the last/only segment */
	while (si < m && si >= 0) {
		/* is next slot contiguous ? */
		if (si != xsi + 1) {
			/* close current linelock */
			lv->length = n;
			dtlck->index++;

			/* open new linelock */
			if (dtlck->index < dtlck->maxcnt)
				lv++;
			else {
				dtlck = (struct dt_lock *) txLinelock(dtlck);
				lv = & dtlck->lv[0];
			}

			lv->offset = si;
			n = 0;
		}

		n++;
		xsi = si;

		t = &p->slot[si];
		si = t->next;
	}

	/* close current linelock */
	lv->length = n;
	dtlck->index++;

	*dtlock = dtlck;
}


/*
 * NAME: dtModify
 *
 * FUNCTION: Modify the inode number part of a directory entry
 *
 * PARAMETERS:
 *	tid	- Transaction id
 *	ip	- Inode of parent directory
 *	key	- Name of entry to be modified
 *	orig_ino	- Original inode number expected in entry
 *	new_ino	- New inode number to put into entry
 *	flag	- JFS_RENAME
 *
 * RETURNS:
 *	-ESTALE	- If entry found does not match orig_ino passed in
 *	-ENOENT	- If no entry can be found to match key
 *	0	- If successfully modified entry
 */
int dtModify(tid_t tid, struct inode *ip,
	 struct component_name * key, ino_t * orig_ino, ino_t new_ino, int flag)
{
	int rc;
	s64 bn;
	struct metapage *mp;
	dtpage_t *p;
	int index;
	struct btstack btstack;
	struct tlock *tlck;
	struct dt_lock *dtlck;
	struct lv *lv;
	s8 *stbl;
	int entry_si;		/* entry slot index */
	struct ldtentry *entry;

	/*
	 *	search for the entry to modify:
	 *
	 * dtSearch() returns (leaf page pinned, index at which to modify).
	 */
	if ((rc = dtSearch(ip, key, orig_ino, &btstack, flag)))
		return rc;

	/* retrieve search result */
	DT_GETSEARCH(ip, btstack.top, bn, mp, p, index);

	BT_MARK_DIRTY(mp, ip);
	/*
	 * acquire a transaction lock on the leaf page of named entry
	 */
	tlck = txLock(tid, ip, mp, tlckDTREE | tlckENTRY);
	dtlck = (struct dt_lock *) & tlck->lock;

	/* get slot index of the entry */
	stbl = DT_GETSTBL(p);
	entry_si = stbl[index];

	/* linelock entry */
	ASSERT(dtlck->index == 0);
	lv = & dtlck->lv[0];
	lv->offset = entry_si;
	lv->length = 1;
	dtlck->index++;

	/* get the head/only segment */
	entry = (struct ldtentry *) & p->slot[entry_si];

	/* substitute the inode number of the entry */
	entry->inumber = cpu_to_le32(new_ino);

	/* unpin the leaf page */
	DT_PUTPAGE(mp);

	return 0;
}