dcache.c 83 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714271527162717271827192720272127222723272427252726272727282729273027312732273327342735273627372738273927402741274227432744274527462747274827492750275127522753275427552756275727582759276027612762276327642765276627672768276927702771277227732774277527762777277827792780278127822783278427852786278727882789279027912792279327942795279627972798279928002801280228032804280528062807280828092810281128122813281428152816281728182819282028212822282328242825282628272828282928302831283228332834283528362837283828392840284128422843284428452846284728482849285028512852285328542855285628572858285928602861286228632864286528662867286828692870287128722873287428752876287728782879288028812882288328842885288628872888288928902891289228932894289528962897289828992900290129022903290429052906290729082909291029112912291329142915291629172918291929202921292229232924292529262927292829292930293129322933293429352936293729382939294029412942294329442945294629472948294929502951295229532954295529562957295829592960296129622963296429652966296729682969297029712972297329742975297629772978297929802981298229832984298529862987298829892990299129922993299429952996299729982999300030013002300330043005300630073008300930103011301230133014301530163017301830193020302130223023302430253026302730283029303030313032303330343035303630373038303930403041304230433044304530463047304830493050305130523053305430553056305730583059306030613062306330643065306630673068306930703071307230733074307530763077307830793080308130823083308430853086308730883089309030913092309330943095309630973098309931003101310231033104310531063107310831093110311131123113311431153116311731183119312031213122312331243125312631273128312931303131313231333134313531363137313831393140314131423143314431453146314731483149315031513152315331543155315631573158315931603161316231633164316531663167316831693170317131723173317431753176317731783179318031813182318331843185318631873188318931903191319231933194319531963197319831993200320132023203320432053206320732083209321032113212321332143215321632173218321932203221322232233224322532263227322832293230323132323233323432353236323732383239324032413242324332443245324632473248324932503251325232533254325532563257325832593260326132623263326432653266326732683269327032713272327332743275327632773278327932803281328232833284
  1. /*
  2. * fs/dcache.c
  3. *
  4. * Complete reimplementation
  5. * (C) 1997 Thomas Schoebel-Theuer,
  6. * with heavy changes by Linus Torvalds
  7. */
  8. /*
  9. * Notes on the allocation strategy:
  10. *
  11. * The dcache is a master of the icache - whenever a dcache entry
  12. * exists, the inode will always exist. "iput()" is done either when
  13. * the dcache entry is deleted or garbage collected.
  14. */
  15. #include <linux/syscalls.h>
  16. #include <linux/string.h>
  17. #include <linux/mm.h>
  18. #include <linux/fs.h>
  19. #include <linux/fsnotify.h>
  20. #include <linux/slab.h>
  21. #include <linux/init.h>
  22. #include <linux/hash.h>
  23. #include <linux/cache.h>
  24. #include <linux/export.h>
  25. #include <linux/mount.h>
  26. #include <linux/file.h>
  27. #include <asm/uaccess.h>
  28. #include <linux/security.h>
  29. #include <linux/seqlock.h>
  30. #include <linux/swap.h>
  31. #include <linux/bootmem.h>
  32. #include <linux/fs_struct.h>
  33. #include <linux/hardirq.h>
  34. #include <linux/bit_spinlock.h>
  35. #include <linux/rculist_bl.h>
  36. #include <linux/prefetch.h>
  37. #include <linux/ratelimit.h>
  38. #include "internal.h"
  39. #include "mount.h"
  40. /*
  41. * Usage:
  42. * dcache->d_inode->i_lock protects:
  43. * - i_dentry, d_alias, d_inode of aliases
  44. * dcache_hash_bucket lock protects:
  45. * - the dcache hash table
  46. * s_anon bl list spinlock protects:
  47. * - the s_anon list (see __d_drop)
  48. * dcache_lru_lock protects:
  49. * - the dcache lru lists and counters
  50. * d_lock protects:
  51. * - d_flags
  52. * - d_name
  53. * - d_lru
  54. * - d_count
  55. * - d_unhashed()
  56. * - d_parent and d_subdirs
  57. * - childrens' d_child and d_parent
  58. * - d_alias, d_inode
  59. *
  60. * Ordering:
  61. * dentry->d_inode->i_lock
  62. * dentry->d_lock
  63. * dcache_lru_lock
  64. * dcache_hash_bucket lock
  65. * s_anon lock
  66. *
  67. * If there is an ancestor relationship:
  68. * dentry->d_parent->...->d_parent->d_lock
  69. * ...
  70. * dentry->d_parent->d_lock
  71. * dentry->d_lock
  72. *
  73. * If no ancestor relationship:
  74. * if (dentry1 < dentry2)
  75. * dentry1->d_lock
  76. * dentry2->d_lock
  77. */
  78. int sysctl_vfs_cache_pressure __read_mostly = 100;
  79. EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
  80. static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
  81. __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
  82. EXPORT_SYMBOL(rename_lock);
  83. static struct kmem_cache *dentry_cache __read_mostly;
  84. /**
  85. * read_seqbegin_or_lock - begin a sequence number check or locking block
  86. * lock: sequence lock
  87. * seq : sequence number to be checked
  88. *
  89. * First try it once optimistically without taking the lock. If that fails,
  90. * take the lock. The sequence number is also used as a marker for deciding
  91. * whether to be a reader (even) or writer (odd).
  92. * N.B. seq must be initialized to an even number to begin with.
  93. */
  94. static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq)
  95. {
  96. if (!(*seq & 1)) /* Even */
  97. *seq = read_seqbegin(lock);
  98. else /* Odd */
  99. write_seqlock(lock);
  100. }
  101. static inline int need_seqretry(seqlock_t *lock, int seq)
  102. {
  103. return !(seq & 1) && read_seqretry(lock, seq);
  104. }
  105. static inline void done_seqretry(seqlock_t *lock, int seq)
  106. {
  107. if (seq & 1)
  108. write_sequnlock(lock);
  109. }
  110. /*
  111. * This is the single most critical data structure when it comes
  112. * to the dcache: the hashtable for lookups. Somebody should try
  113. * to make this good - I've just made it work.
  114. *
  115. * This hash-function tries to avoid losing too many bits of hash
  116. * information, yet avoid using a prime hash-size or similar.
  117. */
  118. #define D_HASHBITS d_hash_shift
  119. #define D_HASHMASK d_hash_mask
  120. static unsigned int d_hash_mask __read_mostly;
  121. static unsigned int d_hash_shift __read_mostly;
  122. static struct hlist_bl_head *dentry_hashtable __read_mostly;
  123. static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
  124. unsigned int hash)
  125. {
  126. hash += (unsigned long) parent / L1_CACHE_BYTES;
  127. hash = hash + (hash >> D_HASHBITS);
  128. return dentry_hashtable + (hash & D_HASHMASK);
  129. }
  130. /* Statistics gathering. */
  131. struct dentry_stat_t dentry_stat = {
  132. .age_limit = 45,
  133. };
  134. static DEFINE_PER_CPU(long, nr_dentry);
  135. static DEFINE_PER_CPU(long, nr_dentry_unused);
  136. #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  137. /*
  138. * Here we resort to our own counters instead of using generic per-cpu counters
  139. * for consistency with what the vfs inode code does. We are expected to harvest
  140. * better code and performance by having our own specialized counters.
  141. *
  142. * Please note that the loop is done over all possible CPUs, not over all online
  143. * CPUs. The reason for this is that we don't want to play games with CPUs going
  144. * on and off. If one of them goes off, we will just keep their counters.
  145. *
  146. * glommer: See cffbc8a for details, and if you ever intend to change this,
  147. * please update all vfs counters to match.
  148. */
  149. static long get_nr_dentry(void)
  150. {
  151. int i;
  152. long sum = 0;
  153. for_each_possible_cpu(i)
  154. sum += per_cpu(nr_dentry, i);
  155. return sum < 0 ? 0 : sum;
  156. }
  157. static long get_nr_dentry_unused(void)
  158. {
  159. int i;
  160. long sum = 0;
  161. for_each_possible_cpu(i)
  162. sum += per_cpu(nr_dentry_unused, i);
  163. return sum < 0 ? 0 : sum;
  164. }
  165. int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
  166. size_t *lenp, loff_t *ppos)
  167. {
  168. dentry_stat.nr_dentry = get_nr_dentry();
  169. dentry_stat.nr_unused = get_nr_dentry_unused();
  170. return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
  171. }
  172. #endif
  173. /*
  174. * Compare 2 name strings, return 0 if they match, otherwise non-zero.
  175. * The strings are both count bytes long, and count is non-zero.
  176. */
  177. #ifdef CONFIG_DCACHE_WORD_ACCESS
  178. #include <asm/word-at-a-time.h>
  179. /*
  180. * NOTE! 'cs' and 'scount' come from a dentry, so it has a
  181. * aligned allocation for this particular component. We don't
  182. * strictly need the load_unaligned_zeropad() safety, but it
  183. * doesn't hurt either.
  184. *
  185. * In contrast, 'ct' and 'tcount' can be from a pathname, and do
  186. * need the careful unaligned handling.
  187. */
  188. static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
  189. {
  190. unsigned long a,b,mask;
  191. for (;;) {
  192. a = *(unsigned long *)cs;
  193. b = load_unaligned_zeropad(ct);
  194. if (tcount < sizeof(unsigned long))
  195. break;
  196. if (unlikely(a != b))
  197. return 1;
  198. cs += sizeof(unsigned long);
  199. ct += sizeof(unsigned long);
  200. tcount -= sizeof(unsigned long);
  201. if (!tcount)
  202. return 0;
  203. }
  204. mask = ~(~0ul << tcount*8);
  205. return unlikely(!!((a ^ b) & mask));
  206. }
  207. #else
  208. static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
  209. {
  210. do {
  211. if (*cs != *ct)
  212. return 1;
  213. cs++;
  214. ct++;
  215. tcount--;
  216. } while (tcount);
  217. return 0;
  218. }
  219. #endif
  220. static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
  221. {
  222. const unsigned char *cs;
  223. /*
  224. * Be careful about RCU walk racing with rename:
  225. * use ACCESS_ONCE to fetch the name pointer.
  226. *
  227. * NOTE! Even if a rename will mean that the length
  228. * was not loaded atomically, we don't care. The
  229. * RCU walk will check the sequence count eventually,
  230. * and catch it. And we won't overrun the buffer,
  231. * because we're reading the name pointer atomically,
  232. * and a dentry name is guaranteed to be properly
  233. * terminated with a NUL byte.
  234. *
  235. * End result: even if 'len' is wrong, we'll exit
  236. * early because the data cannot match (there can
  237. * be no NUL in the ct/tcount data)
  238. */
  239. cs = ACCESS_ONCE(dentry->d_name.name);
  240. smp_read_barrier_depends();
  241. return dentry_string_cmp(cs, ct, tcount);
  242. }
  243. static void __d_free(struct rcu_head *head)
  244. {
  245. struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
  246. WARN_ON(!hlist_unhashed(&dentry->d_alias));
  247. if (dname_external(dentry))
  248. kfree(dentry->d_name.name);
  249. kmem_cache_free(dentry_cache, dentry);
  250. }
  251. /*
  252. * no locks, please.
  253. */
  254. static void d_free(struct dentry *dentry)
  255. {
  256. BUG_ON((int)dentry->d_lockref.count > 0);
  257. this_cpu_dec(nr_dentry);
  258. if (dentry->d_op && dentry->d_op->d_release)
  259. dentry->d_op->d_release(dentry);
  260. /* if dentry was never visible to RCU, immediate free is OK */
  261. if (!(dentry->d_flags & DCACHE_RCUACCESS))
  262. __d_free(&dentry->d_u.d_rcu);
  263. else
  264. call_rcu(&dentry->d_u.d_rcu, __d_free);
  265. }
  266. /**
  267. * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
  268. * @dentry: the target dentry
  269. * After this call, in-progress rcu-walk path lookup will fail. This
  270. * should be called after unhashing, and after changing d_inode (if
  271. * the dentry has not already been unhashed).
  272. */
  273. static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
  274. {
  275. assert_spin_locked(&dentry->d_lock);
  276. /* Go through a barrier */
  277. write_seqcount_barrier(&dentry->d_seq);
  278. }
  279. /*
  280. * Release the dentry's inode, using the filesystem
  281. * d_iput() operation if defined. Dentry has no refcount
  282. * and is unhashed.
  283. */
  284. static void dentry_iput(struct dentry * dentry)
  285. __releases(dentry->d_lock)
  286. __releases(dentry->d_inode->i_lock)
  287. {
  288. struct inode *inode = dentry->d_inode;
  289. if (inode) {
  290. dentry->d_inode = NULL;
  291. hlist_del_init(&dentry->d_alias);
  292. spin_unlock(&dentry->d_lock);
  293. spin_unlock(&inode->i_lock);
  294. if (!inode->i_nlink)
  295. fsnotify_inoderemove(inode);
  296. if (dentry->d_op && dentry->d_op->d_iput)
  297. dentry->d_op->d_iput(dentry, inode);
  298. else
  299. iput(inode);
  300. } else {
  301. spin_unlock(&dentry->d_lock);
  302. }
  303. }
  304. /*
  305. * Release the dentry's inode, using the filesystem
  306. * d_iput() operation if defined. dentry remains in-use.
  307. */
  308. static void dentry_unlink_inode(struct dentry * dentry)
  309. __releases(dentry->d_lock)
  310. __releases(dentry->d_inode->i_lock)
  311. {
  312. struct inode *inode = dentry->d_inode;
  313. dentry->d_inode = NULL;
  314. hlist_del_init(&dentry->d_alias);
  315. dentry_rcuwalk_barrier(dentry);
  316. spin_unlock(&dentry->d_lock);
  317. spin_unlock(&inode->i_lock);
  318. if (!inode->i_nlink)
  319. fsnotify_inoderemove(inode);
  320. if (dentry->d_op && dentry->d_op->d_iput)
  321. dentry->d_op->d_iput(dentry, inode);
  322. else
  323. iput(inode);
  324. }
  325. /*
  326. * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
  327. */
  328. static void dentry_lru_add(struct dentry *dentry)
  329. {
  330. if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) {
  331. spin_lock(&dcache_lru_lock);
  332. dentry->d_flags |= DCACHE_LRU_LIST;
  333. list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
  334. dentry->d_sb->s_nr_dentry_unused++;
  335. this_cpu_inc(nr_dentry_unused);
  336. spin_unlock(&dcache_lru_lock);
  337. }
  338. }
  339. static void __dentry_lru_del(struct dentry *dentry)
  340. {
  341. list_del_init(&dentry->d_lru);
  342. dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
  343. dentry->d_sb->s_nr_dentry_unused--;
  344. this_cpu_dec(nr_dentry_unused);
  345. }
  346. /*
  347. * Remove a dentry with references from the LRU.
  348. */
  349. static void dentry_lru_del(struct dentry *dentry)
  350. {
  351. if (!list_empty(&dentry->d_lru)) {
  352. spin_lock(&dcache_lru_lock);
  353. __dentry_lru_del(dentry);
  354. spin_unlock(&dcache_lru_lock);
  355. }
  356. }
  357. static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
  358. {
  359. spin_lock(&dcache_lru_lock);
  360. if (list_empty(&dentry->d_lru)) {
  361. dentry->d_flags |= DCACHE_LRU_LIST;
  362. list_add_tail(&dentry->d_lru, list);
  363. dentry->d_sb->s_nr_dentry_unused++;
  364. this_cpu_inc(nr_dentry_unused);
  365. } else {
  366. list_move_tail(&dentry->d_lru, list);
  367. }
  368. spin_unlock(&dcache_lru_lock);
  369. }
  370. /**
  371. * d_kill - kill dentry and return parent
  372. * @dentry: dentry to kill
  373. * @parent: parent dentry
  374. *
  375. * The dentry must already be unhashed and removed from the LRU.
  376. *
  377. * If this is the root of the dentry tree, return NULL.
  378. *
  379. * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
  380. * d_kill.
  381. */
  382. static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
  383. __releases(dentry->d_lock)
  384. __releases(parent->d_lock)
  385. __releases(dentry->d_inode->i_lock)
  386. {
  387. list_del(&dentry->d_u.d_child);
  388. /*
  389. * Inform try_to_ascend() that we are no longer attached to the
  390. * dentry tree
  391. */
  392. dentry->d_flags |= DCACHE_DENTRY_KILLED;
  393. if (parent)
  394. spin_unlock(&parent->d_lock);
  395. dentry_iput(dentry);
  396. /*
  397. * dentry_iput drops the locks, at which point nobody (except
  398. * transient RCU lookups) can reach this dentry.
  399. */
  400. d_free(dentry);
  401. return parent;
  402. }
  403. /*
  404. * Unhash a dentry without inserting an RCU walk barrier or checking that
  405. * dentry->d_lock is locked. The caller must take care of that, if
  406. * appropriate.
  407. */
  408. static void __d_shrink(struct dentry *dentry)
  409. {
  410. if (!d_unhashed(dentry)) {
  411. struct hlist_bl_head *b;
  412. if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
  413. b = &dentry->d_sb->s_anon;
  414. else
  415. b = d_hash(dentry->d_parent, dentry->d_name.hash);
  416. hlist_bl_lock(b);
  417. __hlist_bl_del(&dentry->d_hash);
  418. dentry->d_hash.pprev = NULL;
  419. hlist_bl_unlock(b);
  420. }
  421. }
  422. /**
  423. * d_drop - drop a dentry
  424. * @dentry: dentry to drop
  425. *
  426. * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
  427. * be found through a VFS lookup any more. Note that this is different from
  428. * deleting the dentry - d_delete will try to mark the dentry negative if
  429. * possible, giving a successful _negative_ lookup, while d_drop will
  430. * just make the cache lookup fail.
  431. *
  432. * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
  433. * reason (NFS timeouts or autofs deletes).
  434. *
  435. * __d_drop requires dentry->d_lock.
  436. */
  437. void __d_drop(struct dentry *dentry)
  438. {
  439. if (!d_unhashed(dentry)) {
  440. __d_shrink(dentry);
  441. dentry_rcuwalk_barrier(dentry);
  442. }
  443. }
  444. EXPORT_SYMBOL(__d_drop);
  445. void d_drop(struct dentry *dentry)
  446. {
  447. spin_lock(&dentry->d_lock);
  448. __d_drop(dentry);
  449. spin_unlock(&dentry->d_lock);
  450. }
  451. EXPORT_SYMBOL(d_drop);
  452. /*
  453. * Finish off a dentry we've decided to kill.
  454. * dentry->d_lock must be held, returns with it unlocked.
  455. * If ref is non-zero, then decrement the refcount too.
  456. * Returns dentry requiring refcount drop, or NULL if we're done.
  457. */
  458. static inline struct dentry *dentry_kill(struct dentry *dentry)
  459. __releases(dentry->d_lock)
  460. {
  461. struct inode *inode;
  462. struct dentry *parent;
  463. inode = dentry->d_inode;
  464. if (inode && !spin_trylock(&inode->i_lock)) {
  465. relock:
  466. spin_unlock(&dentry->d_lock);
  467. cpu_relax();
  468. return dentry; /* try again with same dentry */
  469. }
  470. if (IS_ROOT(dentry))
  471. parent = NULL;
  472. else
  473. parent = dentry->d_parent;
  474. if (parent && !spin_trylock(&parent->d_lock)) {
  475. if (inode)
  476. spin_unlock(&inode->i_lock);
  477. goto relock;
  478. }
  479. /*
  480. * The dentry is now unrecoverably dead to the world.
  481. */
  482. lockref_mark_dead(&dentry->d_lockref);
  483. /*
  484. * inform the fs via d_prune that this dentry is about to be
  485. * unhashed and destroyed.
  486. */
  487. if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
  488. dentry->d_op->d_prune(dentry);
  489. dentry_lru_del(dentry);
  490. /* if it was on the hash then remove it */
  491. __d_drop(dentry);
  492. return d_kill(dentry, parent);
  493. }
  494. /*
  495. * This is dput
  496. *
  497. * This is complicated by the fact that we do not want to put
  498. * dentries that are no longer on any hash chain on the unused
  499. * list: we'd much rather just get rid of them immediately.
  500. *
  501. * However, that implies that we have to traverse the dentry
  502. * tree upwards to the parents which might _also_ now be
  503. * scheduled for deletion (it may have been only waiting for
  504. * its last child to go away).
  505. *
  506. * This tail recursion is done by hand as we don't want to depend
  507. * on the compiler to always get this right (gcc generally doesn't).
  508. * Real recursion would eat up our stack space.
  509. */
  510. /*
  511. * dput - release a dentry
  512. * @dentry: dentry to release
  513. *
  514. * Release a dentry. This will drop the usage count and if appropriate
  515. * call the dentry unlink method as well as removing it from the queues and
  516. * releasing its resources. If the parent dentries were scheduled for release
  517. * they too may now get deleted.
  518. */
  519. void dput(struct dentry *dentry)
  520. {
  521. if (unlikely(!dentry))
  522. return;
  523. repeat:
  524. if (lockref_put_or_lock(&dentry->d_lockref))
  525. return;
  526. /* Unreachable? Get rid of it */
  527. if (unlikely(d_unhashed(dentry)))
  528. goto kill_it;
  529. if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
  530. if (dentry->d_op->d_delete(dentry))
  531. goto kill_it;
  532. }
  533. dentry->d_flags |= DCACHE_REFERENCED;
  534. dentry_lru_add(dentry);
  535. dentry->d_lockref.count--;
  536. spin_unlock(&dentry->d_lock);
  537. return;
  538. kill_it:
  539. dentry = dentry_kill(dentry);
  540. if (dentry)
  541. goto repeat;
  542. }
  543. EXPORT_SYMBOL(dput);
  544. /**
  545. * d_invalidate - invalidate a dentry
  546. * @dentry: dentry to invalidate
  547. *
  548. * Try to invalidate the dentry if it turns out to be
  549. * possible. If there are other dentries that can be
  550. * reached through this one we can't delete it and we
  551. * return -EBUSY. On success we return 0.
  552. *
  553. * no dcache lock.
  554. */
  555. int d_invalidate(struct dentry * dentry)
  556. {
  557. /*
  558. * If it's already been dropped, return OK.
  559. */
  560. spin_lock(&dentry->d_lock);
  561. if (d_unhashed(dentry)) {
  562. spin_unlock(&dentry->d_lock);
  563. return 0;
  564. }
  565. /*
  566. * Check whether to do a partial shrink_dcache
  567. * to get rid of unused child entries.
  568. */
  569. if (!list_empty(&dentry->d_subdirs)) {
  570. spin_unlock(&dentry->d_lock);
  571. shrink_dcache_parent(dentry);
  572. spin_lock(&dentry->d_lock);
  573. }
  574. /*
  575. * Somebody else still using it?
  576. *
  577. * If it's a directory, we can't drop it
  578. * for fear of somebody re-populating it
  579. * with children (even though dropping it
  580. * would make it unreachable from the root,
  581. * we might still populate it if it was a
  582. * working directory or similar).
  583. * We also need to leave mountpoints alone,
  584. * directory or not.
  585. */
  586. if (dentry->d_lockref.count > 1 && dentry->d_inode) {
  587. if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
  588. spin_unlock(&dentry->d_lock);
  589. return -EBUSY;
  590. }
  591. }
  592. __d_drop(dentry);
  593. spin_unlock(&dentry->d_lock);
  594. return 0;
  595. }
  596. EXPORT_SYMBOL(d_invalidate);
  597. /* This must be called with d_lock held */
  598. static inline void __dget_dlock(struct dentry *dentry)
  599. {
  600. dentry->d_lockref.count++;
  601. }
  602. static inline void __dget(struct dentry *dentry)
  603. {
  604. lockref_get(&dentry->d_lockref);
  605. }
  606. struct dentry *dget_parent(struct dentry *dentry)
  607. {
  608. int gotref;
  609. struct dentry *ret;
  610. /*
  611. * Do optimistic parent lookup without any
  612. * locking.
  613. */
  614. rcu_read_lock();
  615. ret = ACCESS_ONCE(dentry->d_parent);
  616. gotref = lockref_get_not_zero(&ret->d_lockref);
  617. rcu_read_unlock();
  618. if (likely(gotref)) {
  619. if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
  620. return ret;
  621. dput(ret);
  622. }
  623. repeat:
  624. /*
  625. * Don't need rcu_dereference because we re-check it was correct under
  626. * the lock.
  627. */
  628. rcu_read_lock();
  629. ret = dentry->d_parent;
  630. spin_lock(&ret->d_lock);
  631. if (unlikely(ret != dentry->d_parent)) {
  632. spin_unlock(&ret->d_lock);
  633. rcu_read_unlock();
  634. goto repeat;
  635. }
  636. rcu_read_unlock();
  637. BUG_ON(!ret->d_lockref.count);
  638. ret->d_lockref.count++;
  639. spin_unlock(&ret->d_lock);
  640. return ret;
  641. }
  642. EXPORT_SYMBOL(dget_parent);
  643. /**
  644. * d_find_alias - grab a hashed alias of inode
  645. * @inode: inode in question
  646. * @want_discon: flag, used by d_splice_alias, to request
  647. * that only a DISCONNECTED alias be returned.
  648. *
  649. * If inode has a hashed alias, or is a directory and has any alias,
  650. * acquire the reference to alias and return it. Otherwise return NULL.
  651. * Notice that if inode is a directory there can be only one alias and
  652. * it can be unhashed only if it has no children, or if it is the root
  653. * of a filesystem.
  654. *
  655. * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
  656. * any other hashed alias over that one unless @want_discon is set,
  657. * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
  658. */
  659. static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
  660. {
  661. struct dentry *alias, *discon_alias;
  662. again:
  663. discon_alias = NULL;
  664. hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
  665. spin_lock(&alias->d_lock);
  666. if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
  667. if (IS_ROOT(alias) &&
  668. (alias->d_flags & DCACHE_DISCONNECTED)) {
  669. discon_alias = alias;
  670. } else if (!want_discon) {
  671. __dget_dlock(alias);
  672. spin_unlock(&alias->d_lock);
  673. return alias;
  674. }
  675. }
  676. spin_unlock(&alias->d_lock);
  677. }
  678. if (discon_alias) {
  679. alias = discon_alias;
  680. spin_lock(&alias->d_lock);
  681. if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
  682. if (IS_ROOT(alias) &&
  683. (alias->d_flags & DCACHE_DISCONNECTED)) {
  684. __dget_dlock(alias);
  685. spin_unlock(&alias->d_lock);
  686. return alias;
  687. }
  688. }
  689. spin_unlock(&alias->d_lock);
  690. goto again;
  691. }
  692. return NULL;
  693. }
  694. struct dentry *d_find_alias(struct inode *inode)
  695. {
  696. struct dentry *de = NULL;
  697. if (!hlist_empty(&inode->i_dentry)) {
  698. spin_lock(&inode->i_lock);
  699. de = __d_find_alias(inode, 0);
  700. spin_unlock(&inode->i_lock);
  701. }
  702. return de;
  703. }
  704. EXPORT_SYMBOL(d_find_alias);
  705. /*
  706. * Try to kill dentries associated with this inode.
  707. * WARNING: you must own a reference to inode.
  708. */
  709. void d_prune_aliases(struct inode *inode)
  710. {
  711. struct dentry *dentry;
  712. restart:
  713. spin_lock(&inode->i_lock);
  714. hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
  715. spin_lock(&dentry->d_lock);
  716. if (!dentry->d_lockref.count) {
  717. /*
  718. * inform the fs via d_prune that this dentry
  719. * is about to be unhashed and destroyed.
  720. */
  721. if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
  722. !d_unhashed(dentry))
  723. dentry->d_op->d_prune(dentry);
  724. __dget_dlock(dentry);
  725. __d_drop(dentry);
  726. spin_unlock(&dentry->d_lock);
  727. spin_unlock(&inode->i_lock);
  728. dput(dentry);
  729. goto restart;
  730. }
  731. spin_unlock(&dentry->d_lock);
  732. }
  733. spin_unlock(&inode->i_lock);
  734. }
  735. EXPORT_SYMBOL(d_prune_aliases);
  736. /*
  737. * Try to throw away a dentry - free the inode, dput the parent.
  738. * Requires dentry->d_lock is held, and dentry->d_count == 0.
  739. * Releases dentry->d_lock.
  740. *
  741. * This may fail if locks cannot be acquired no problem, just try again.
  742. */
  743. static void try_prune_one_dentry(struct dentry *dentry)
  744. __releases(dentry->d_lock)
  745. {
  746. struct dentry *parent;
  747. parent = dentry_kill(dentry);
  748. /*
  749. * If dentry_kill returns NULL, we have nothing more to do.
  750. * if it returns the same dentry, trylocks failed. In either
  751. * case, just loop again.
  752. *
  753. * Otherwise, we need to prune ancestors too. This is necessary
  754. * to prevent quadratic behavior of shrink_dcache_parent(), but
  755. * is also expected to be beneficial in reducing dentry cache
  756. * fragmentation.
  757. */
  758. if (!parent)
  759. return;
  760. if (parent == dentry)
  761. return;
  762. /* Prune ancestors. */
  763. dentry = parent;
  764. while (dentry) {
  765. if (lockref_put_or_lock(&dentry->d_lockref))
  766. return;
  767. dentry = dentry_kill(dentry);
  768. }
  769. }
  770. static void shrink_dentry_list(struct list_head *list)
  771. {
  772. struct dentry *dentry;
  773. rcu_read_lock();
  774. for (;;) {
  775. dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
  776. if (&dentry->d_lru == list)
  777. break; /* empty */
  778. spin_lock(&dentry->d_lock);
  779. if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
  780. spin_unlock(&dentry->d_lock);
  781. continue;
  782. }
  783. /*
  784. * We found an inuse dentry which was not removed from
  785. * the LRU because of laziness during lookup. Do not free
  786. * it - just keep it off the LRU list.
  787. */
  788. if (dentry->d_lockref.count) {
  789. dentry_lru_del(dentry);
  790. spin_unlock(&dentry->d_lock);
  791. continue;
  792. }
  793. rcu_read_unlock();
  794. try_prune_one_dentry(dentry);
  795. rcu_read_lock();
  796. }
  797. rcu_read_unlock();
  798. }
  799. /**
  800. * prune_dcache_sb - shrink the dcache
  801. * @sb: superblock
  802. * @count: number of entries to try to free
  803. *
  804. * Attempt to shrink the superblock dcache LRU by @count entries. This is
  805. * done when we need more memory an called from the superblock shrinker
  806. * function.
  807. *
  808. * This function may fail to free any resources if all the dentries are in
  809. * use.
  810. */
  811. void prune_dcache_sb(struct super_block *sb, int count)
  812. {
  813. struct dentry *dentry;
  814. LIST_HEAD(referenced);
  815. LIST_HEAD(tmp);
  816. relock:
  817. spin_lock(&dcache_lru_lock);
  818. while (!list_empty(&sb->s_dentry_lru)) {
  819. dentry = list_entry(sb->s_dentry_lru.prev,
  820. struct dentry, d_lru);
  821. BUG_ON(dentry->d_sb != sb);
  822. if (!spin_trylock(&dentry->d_lock)) {
  823. spin_unlock(&dcache_lru_lock);
  824. cpu_relax();
  825. goto relock;
  826. }
  827. if (dentry->d_flags & DCACHE_REFERENCED) {
  828. dentry->d_flags &= ~DCACHE_REFERENCED;
  829. list_move(&dentry->d_lru, &referenced);
  830. spin_unlock(&dentry->d_lock);
  831. } else {
  832. list_move_tail(&dentry->d_lru, &tmp);
  833. dentry->d_flags |= DCACHE_SHRINK_LIST;
  834. spin_unlock(&dentry->d_lock);
  835. if (!--count)
  836. break;
  837. }
  838. cond_resched_lock(&dcache_lru_lock);
  839. }
  840. if (!list_empty(&referenced))
  841. list_splice(&referenced, &sb->s_dentry_lru);
  842. spin_unlock(&dcache_lru_lock);
  843. shrink_dentry_list(&tmp);
  844. }
  845. /**
  846. * shrink_dcache_sb - shrink dcache for a superblock
  847. * @sb: superblock
  848. *
  849. * Shrink the dcache for the specified super block. This is used to free
  850. * the dcache before unmounting a file system.
  851. */
  852. void shrink_dcache_sb(struct super_block *sb)
  853. {
  854. LIST_HEAD(tmp);
  855. spin_lock(&dcache_lru_lock);
  856. while (!list_empty(&sb->s_dentry_lru)) {
  857. list_splice_init(&sb->s_dentry_lru, &tmp);
  858. spin_unlock(&dcache_lru_lock);
  859. shrink_dentry_list(&tmp);
  860. spin_lock(&dcache_lru_lock);
  861. }
  862. spin_unlock(&dcache_lru_lock);
  863. }
  864. EXPORT_SYMBOL(shrink_dcache_sb);
  865. /*
  866. * destroy a single subtree of dentries for unmount
  867. * - see the comments on shrink_dcache_for_umount() for a description of the
  868. * locking
  869. */
  870. static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
  871. {
  872. struct dentry *parent;
  873. BUG_ON(!IS_ROOT(dentry));
  874. for (;;) {
  875. /* descend to the first leaf in the current subtree */
  876. while (!list_empty(&dentry->d_subdirs))
  877. dentry = list_entry(dentry->d_subdirs.next,
  878. struct dentry, d_u.d_child);
  879. /* consume the dentries from this leaf up through its parents
  880. * until we find one with children or run out altogether */
  881. do {
  882. struct inode *inode;
  883. /*
  884. * inform the fs that this dentry is about to be
  885. * unhashed and destroyed.
  886. */
  887. if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
  888. !d_unhashed(dentry))
  889. dentry->d_op->d_prune(dentry);
  890. dentry_lru_del(dentry);
  891. __d_shrink(dentry);
  892. if (dentry->d_lockref.count != 0) {
  893. printk(KERN_ERR
  894. "BUG: Dentry %p{i=%lx,n=%s}"
  895. " still in use (%d)"
  896. " [unmount of %s %s]\n",
  897. dentry,
  898. dentry->d_inode ?
  899. dentry->d_inode->i_ino : 0UL,
  900. dentry->d_name.name,
  901. dentry->d_lockref.count,
  902. dentry->d_sb->s_type->name,
  903. dentry->d_sb->s_id);
  904. BUG();
  905. }
  906. if (IS_ROOT(dentry)) {
  907. parent = NULL;
  908. list_del(&dentry->d_u.d_child);
  909. } else {
  910. parent = dentry->d_parent;
  911. parent->d_lockref.count--;
  912. list_del(&dentry->d_u.d_child);
  913. }
  914. inode = dentry->d_inode;
  915. if (inode) {
  916. dentry->d_inode = NULL;
  917. hlist_del_init(&dentry->d_alias);
  918. if (dentry->d_op && dentry->d_op->d_iput)
  919. dentry->d_op->d_iput(dentry, inode);
  920. else
  921. iput(inode);
  922. }
  923. d_free(dentry);
  924. /* finished when we fall off the top of the tree,
  925. * otherwise we ascend to the parent and move to the
  926. * next sibling if there is one */
  927. if (!parent)
  928. return;
  929. dentry = parent;
  930. } while (list_empty(&dentry->d_subdirs));
  931. dentry = list_entry(dentry->d_subdirs.next,
  932. struct dentry, d_u.d_child);
  933. }
  934. }
  935. /*
  936. * destroy the dentries attached to a superblock on unmounting
  937. * - we don't need to use dentry->d_lock because:
  938. * - the superblock is detached from all mountings and open files, so the
  939. * dentry trees will not be rearranged by the VFS
  940. * - s_umount is write-locked, so the memory pressure shrinker will ignore
  941. * any dentries belonging to this superblock that it comes across
  942. * - the filesystem itself is no longer permitted to rearrange the dentries
  943. * in this superblock
  944. */
  945. void shrink_dcache_for_umount(struct super_block *sb)
  946. {
  947. struct dentry *dentry;
  948. if (down_read_trylock(&sb->s_umount))
  949. BUG();
  950. dentry = sb->s_root;
  951. sb->s_root = NULL;
  952. dentry->d_lockref.count--;
  953. shrink_dcache_for_umount_subtree(dentry);
  954. while (!hlist_bl_empty(&sb->s_anon)) {
  955. dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
  956. shrink_dcache_for_umount_subtree(dentry);
  957. }
  958. }
  959. /*
  960. * This tries to ascend one level of parenthood, but
  961. * we can race with renaming, so we need to re-check
  962. * the parenthood after dropping the lock and check
  963. * that the sequence number still matches.
  964. */
  965. static struct dentry *try_to_ascend(struct dentry *old, unsigned seq)
  966. {
  967. struct dentry *new = old->d_parent;
  968. rcu_read_lock();
  969. spin_unlock(&old->d_lock);
  970. spin_lock(&new->d_lock);
  971. /*
  972. * might go back up the wrong parent if we have had a rename
  973. * or deletion
  974. */
  975. if (new != old->d_parent ||
  976. (old->d_flags & DCACHE_DENTRY_KILLED) ||
  977. need_seqretry(&rename_lock, seq)) {
  978. spin_unlock(&new->d_lock);
  979. new = NULL;
  980. }
  981. rcu_read_unlock();
  982. return new;
  983. }
  984. /**
  985. * enum d_walk_ret - action to talke during tree walk
  986. * @D_WALK_CONTINUE: contrinue walk
  987. * @D_WALK_QUIT: quit walk
  988. * @D_WALK_NORETRY: quit when retry is needed
  989. * @D_WALK_SKIP: skip this dentry and its children
  990. */
  991. enum d_walk_ret {
  992. D_WALK_CONTINUE,
  993. D_WALK_QUIT,
  994. D_WALK_NORETRY,
  995. D_WALK_SKIP,
  996. };
  997. /**
  998. * d_walk - walk the dentry tree
  999. * @parent: start of walk
  1000. * @data: data passed to @enter() and @finish()
  1001. * @enter: callback when first entering the dentry
  1002. * @finish: callback when successfully finished the walk
  1003. *
  1004. * The @enter() and @finish() callbacks are called with d_lock held.
  1005. */
  1006. static void d_walk(struct dentry *parent, void *data,
  1007. enum d_walk_ret (*enter)(void *, struct dentry *),
  1008. void (*finish)(void *))
  1009. {
  1010. struct dentry *this_parent;
  1011. struct list_head *next;
  1012. unsigned seq = 0;
  1013. enum d_walk_ret ret;
  1014. bool retry = true;
  1015. again:
  1016. read_seqbegin_or_lock(&rename_lock, &seq);
  1017. this_parent = parent;
  1018. spin_lock(&this_parent->d_lock);
  1019. ret = enter(data, this_parent);
  1020. switch (ret) {
  1021. case D_WALK_CONTINUE:
  1022. break;
  1023. case D_WALK_QUIT:
  1024. case D_WALK_SKIP:
  1025. goto out_unlock;
  1026. case D_WALK_NORETRY:
  1027. retry = false;
  1028. break;
  1029. }
  1030. repeat:
  1031. next = this_parent->d_subdirs.next;
  1032. resume:
  1033. while (next != &this_parent->d_subdirs) {
  1034. struct list_head *tmp = next;
  1035. struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
  1036. next = tmp->next;
  1037. spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
  1038. ret = enter(data, dentry);
  1039. switch (ret) {
  1040. case D_WALK_CONTINUE:
  1041. break;
  1042. case D_WALK_QUIT:
  1043. spin_unlock(&dentry->d_lock);
  1044. goto out_unlock;
  1045. case D_WALK_NORETRY:
  1046. retry = false;
  1047. break;
  1048. case D_WALK_SKIP:
  1049. spin_unlock(&dentry->d_lock);
  1050. continue;
  1051. }
  1052. if (!list_empty(&dentry->d_subdirs)) {
  1053. spin_unlock(&this_parent->d_lock);
  1054. spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
  1055. this_parent = dentry;
  1056. spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
  1057. goto repeat;
  1058. }
  1059. spin_unlock(&dentry->d_lock);
  1060. }
  1061. /*
  1062. * All done at this level ... ascend and resume the search.
  1063. */
  1064. if (this_parent != parent) {
  1065. struct dentry *child = this_parent;
  1066. this_parent = try_to_ascend(this_parent, seq);
  1067. if (!this_parent)
  1068. goto rename_retry;
  1069. next = child->d_u.d_child.next;
  1070. goto resume;
  1071. }
  1072. if (need_seqretry(&rename_lock, seq)) {
  1073. spin_unlock(&this_parent->d_lock);
  1074. goto rename_retry;
  1075. }
  1076. if (finish)
  1077. finish(data);
  1078. out_unlock:
  1079. spin_unlock(&this_parent->d_lock);
  1080. done_seqretry(&rename_lock, seq);
  1081. return;
  1082. rename_retry:
  1083. if (!retry)
  1084. return;
  1085. seq = 1;
  1086. goto again;
  1087. }
  1088. /*
  1089. * Search for at least 1 mount point in the dentry's subdirs.
  1090. * We descend to the next level whenever the d_subdirs
  1091. * list is non-empty and continue searching.
  1092. */
  1093. /**
  1094. * have_submounts - check for mounts over a dentry
  1095. * @parent: dentry to check.
  1096. *
  1097. * Return true if the parent or its subdirectories contain
  1098. * a mount point
  1099. */
  1100. static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
  1101. {
  1102. int *ret = data;
  1103. if (d_mountpoint(dentry)) {
  1104. *ret = 1;
  1105. return D_WALK_QUIT;
  1106. }
  1107. return D_WALK_CONTINUE;
  1108. }
  1109. int have_submounts(struct dentry *parent)
  1110. {
  1111. int ret = 0;
  1112. d_walk(parent, &ret, check_mount, NULL);
  1113. return ret;
  1114. }
  1115. EXPORT_SYMBOL(have_submounts);
  1116. /*
  1117. * Called by mount code to set a mountpoint and check if the mountpoint is
  1118. * reachable (e.g. NFS can unhash a directory dentry and then the complete
  1119. * subtree can become unreachable).
  1120. *
  1121. * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
  1122. * this reason take rename_lock and d_lock on dentry and ancestors.
  1123. */
  1124. int d_set_mounted(struct dentry *dentry)
  1125. {
  1126. struct dentry *p;
  1127. int ret = -ENOENT;
  1128. write_seqlock(&rename_lock);
  1129. for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
  1130. /* Need exclusion wrt. check_submounts_and_drop() */
  1131. spin_lock(&p->d_lock);
  1132. if (unlikely(d_unhashed(p))) {
  1133. spin_unlock(&p->d_lock);
  1134. goto out;
  1135. }
  1136. spin_unlock(&p->d_lock);
  1137. }
  1138. spin_lock(&dentry->d_lock);
  1139. if (!d_unlinked(dentry)) {
  1140. dentry->d_flags |= DCACHE_MOUNTED;
  1141. ret = 0;
  1142. }
  1143. spin_unlock(&dentry->d_lock);
  1144. out:
  1145. write_sequnlock(&rename_lock);
  1146. return ret;
  1147. }
  1148. /*
  1149. * Search the dentry child list of the specified parent,
  1150. * and move any unused dentries to the end of the unused
  1151. * list for prune_dcache(). We descend to the next level
  1152. * whenever the d_subdirs list is non-empty and continue
  1153. * searching.
  1154. *
  1155. * It returns zero iff there are no unused children,
  1156. * otherwise it returns the number of children moved to
  1157. * the end of the unused list. This may not be the total
  1158. * number of unused children, because select_parent can
  1159. * drop the lock and return early due to latency
  1160. * constraints.
  1161. */
  1162. struct select_data {
  1163. struct dentry *start;
  1164. struct list_head dispose;
  1165. int found;
  1166. };
  1167. static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
  1168. {
  1169. struct select_data *data = _data;
  1170. enum d_walk_ret ret = D_WALK_CONTINUE;
  1171. if (data->start == dentry)
  1172. goto out;
  1173. /*
  1174. * move only zero ref count dentries to the dispose list.
  1175. *
  1176. * Those which are presently on the shrink list, being processed
  1177. * by shrink_dentry_list(), shouldn't be moved. Otherwise the
  1178. * loop in shrink_dcache_parent() might not make any progress
  1179. * and loop forever.
  1180. */
  1181. if (dentry->d_lockref.count) {
  1182. dentry_lru_del(dentry);
  1183. } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
  1184. dentry_lru_move_list(dentry, &data->dispose);
  1185. dentry->d_flags |= DCACHE_SHRINK_LIST;
  1186. data->found++;
  1187. ret = D_WALK_NORETRY;
  1188. }
  1189. /*
  1190. * We can return to the caller if we have found some (this
  1191. * ensures forward progress). We'll be coming back to find
  1192. * the rest.
  1193. */
  1194. if (data->found && need_resched())
  1195. ret = D_WALK_QUIT;
  1196. out:
  1197. return ret;
  1198. }
  1199. /**
  1200. * shrink_dcache_parent - prune dcache
  1201. * @parent: parent of entries to prune
  1202. *
  1203. * Prune the dcache to remove unused children of the parent dentry.
  1204. */
  1205. void shrink_dcache_parent(struct dentry *parent)
  1206. {
  1207. for (;;) {
  1208. struct select_data data;
  1209. INIT_LIST_HEAD(&data.dispose);
  1210. data.start = parent;
  1211. data.found = 0;
  1212. d_walk(parent, &data, select_collect, NULL);
  1213. if (!data.found)
  1214. break;
  1215. shrink_dentry_list(&data.dispose);
  1216. cond_resched();
  1217. }
  1218. }
  1219. EXPORT_SYMBOL(shrink_dcache_parent);
  1220. static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
  1221. {
  1222. struct select_data *data = _data;
  1223. if (d_mountpoint(dentry)) {
  1224. data->found = -EBUSY;
  1225. return D_WALK_QUIT;
  1226. }
  1227. return select_collect(_data, dentry);
  1228. }
  1229. static void check_and_drop(void *_data)
  1230. {
  1231. struct select_data *data = _data;
  1232. if (d_mountpoint(data->start))
  1233. data->found = -EBUSY;
  1234. if (!data->found)
  1235. __d_drop(data->start);
  1236. }
  1237. /**
  1238. * check_submounts_and_drop - prune dcache, check for submounts and drop
  1239. *
  1240. * All done as a single atomic operation relative to has_unlinked_ancestor().
  1241. * Returns 0 if successfully unhashed @parent. If there were submounts then
  1242. * return -EBUSY.
  1243. *
  1244. * @dentry: dentry to prune and drop
  1245. */
  1246. int check_submounts_and_drop(struct dentry *dentry)
  1247. {
  1248. int ret = 0;
  1249. /* Negative dentries can be dropped without further checks */
  1250. if (!dentry->d_inode) {
  1251. d_drop(dentry);
  1252. goto out;
  1253. }
  1254. for (;;) {
  1255. struct select_data data;
  1256. INIT_LIST_HEAD(&data.dispose);
  1257. data.start = dentry;
  1258. data.found = 0;
  1259. d_walk(dentry, &data, check_and_collect, check_and_drop);
  1260. ret = data.found;
  1261. if (!list_empty(&data.dispose))
  1262. shrink_dentry_list(&data.dispose);
  1263. if (ret <= 0)
  1264. break;
  1265. cond_resched();
  1266. }
  1267. out:
  1268. return ret;
  1269. }
  1270. EXPORT_SYMBOL(check_submounts_and_drop);
  1271. /**
  1272. * __d_alloc - allocate a dcache entry
  1273. * @sb: filesystem it will belong to
  1274. * @name: qstr of the name
  1275. *
  1276. * Allocates a dentry. It returns %NULL if there is insufficient memory
  1277. * available. On a success the dentry is returned. The name passed in is
  1278. * copied and the copy passed in may be reused after this call.
  1279. */
  1280. struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
  1281. {
  1282. struct dentry *dentry;
  1283. char *dname;
  1284. dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
  1285. if (!dentry)
  1286. return NULL;
  1287. /*
  1288. * We guarantee that the inline name is always NUL-terminated.
  1289. * This way the memcpy() done by the name switching in rename
  1290. * will still always have a NUL at the end, even if we might
  1291. * be overwriting an internal NUL character
  1292. */
  1293. dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
  1294. if (name->len > DNAME_INLINE_LEN-1) {
  1295. dname = kmalloc(name->len + 1, GFP_KERNEL);
  1296. if (!dname) {
  1297. kmem_cache_free(dentry_cache, dentry);
  1298. return NULL;
  1299. }
  1300. } else {
  1301. dname = dentry->d_iname;
  1302. }
  1303. dentry->d_name.len = name->len;
  1304. dentry->d_name.hash = name->hash;
  1305. memcpy(dname, name->name, name->len);
  1306. dname[name->len] = 0;
  1307. /* Make sure we always see the terminating NUL character */
  1308. smp_wmb();
  1309. dentry->d_name.name = dname;
  1310. dentry->d_lockref.count = 1;
  1311. dentry->d_flags = 0;
  1312. spin_lock_init(&dentry->d_lock);
  1313. seqcount_init(&dentry->d_seq);
  1314. dentry->d_inode = NULL;
  1315. dentry->d_parent = dentry;
  1316. dentry->d_sb = sb;
  1317. dentry->d_op = NULL;
  1318. dentry->d_fsdata = NULL;
  1319. INIT_HLIST_BL_NODE(&dentry->d_hash);
  1320. INIT_LIST_HEAD(&dentry->d_lru);
  1321. INIT_LIST_HEAD(&dentry->d_subdirs);
  1322. INIT_HLIST_NODE(&dentry->d_alias);
  1323. INIT_LIST_HEAD(&dentry->d_u.d_child);
  1324. d_set_d_op(dentry, dentry->d_sb->s_d_op);
  1325. this_cpu_inc(nr_dentry);
  1326. return dentry;
  1327. }
  1328. /**
  1329. * d_alloc - allocate a dcache entry
  1330. * @parent: parent of entry to allocate
  1331. * @name: qstr of the name
  1332. *
  1333. * Allocates a dentry. It returns %NULL if there is insufficient memory
  1334. * available. On a success the dentry is returned. The name passed in is
  1335. * copied and the copy passed in may be reused after this call.
  1336. */
  1337. struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
  1338. {
  1339. struct dentry *dentry = __d_alloc(parent->d_sb, name);
  1340. if (!dentry)
  1341. return NULL;
  1342. spin_lock(&parent->d_lock);
  1343. /*
  1344. * don't need child lock because it is not subject
  1345. * to concurrency here
  1346. */
  1347. __dget_dlock(parent);
  1348. dentry->d_parent = parent;
  1349. list_add(&dentry->d_u.d_child, &parent->d_subdirs);
  1350. spin_unlock(&parent->d_lock);
  1351. return dentry;
  1352. }
  1353. EXPORT_SYMBOL(d_alloc);
  1354. struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
  1355. {
  1356. struct dentry *dentry = __d_alloc(sb, name);
  1357. if (dentry)
  1358. dentry->d_flags |= DCACHE_DISCONNECTED;
  1359. return dentry;
  1360. }
  1361. EXPORT_SYMBOL(d_alloc_pseudo);
  1362. struct dentry *d_alloc_name(struct dentry *parent, const char *name)
  1363. {
  1364. struct qstr q;
  1365. q.name = name;
  1366. q.len = strlen(name);
  1367. q.hash = full_name_hash(q.name, q.len);
  1368. return d_alloc(parent, &q);
  1369. }
  1370. EXPORT_SYMBOL(d_alloc_name);
  1371. void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
  1372. {
  1373. WARN_ON_ONCE(dentry->d_op);
  1374. WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
  1375. DCACHE_OP_COMPARE |
  1376. DCACHE_OP_REVALIDATE |
  1377. DCACHE_OP_WEAK_REVALIDATE |
  1378. DCACHE_OP_DELETE ));
  1379. dentry->d_op = op;
  1380. if (!op)
  1381. return;
  1382. if (op->d_hash)
  1383. dentry->d_flags |= DCACHE_OP_HASH;
  1384. if (op->d_compare)
  1385. dentry->d_flags |= DCACHE_OP_COMPARE;
  1386. if (op->d_revalidate)
  1387. dentry->d_flags |= DCACHE_OP_REVALIDATE;
  1388. if (op->d_weak_revalidate)
  1389. dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
  1390. if (op->d_delete)
  1391. dentry->d_flags |= DCACHE_OP_DELETE;
  1392. if (op->d_prune)
  1393. dentry->d_flags |= DCACHE_OP_PRUNE;
  1394. }
  1395. EXPORT_SYMBOL(d_set_d_op);
  1396. static void __d_instantiate(struct dentry *dentry, struct inode *inode)
  1397. {
  1398. spin_lock(&dentry->d_lock);
  1399. if (inode) {
  1400. if (unlikely(IS_AUTOMOUNT(inode)))
  1401. dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
  1402. hlist_add_head(&dentry->d_alias, &inode->i_dentry);
  1403. }
  1404. dentry->d_inode = inode;
  1405. dentry_rcuwalk_barrier(dentry);
  1406. spin_unlock(&dentry->d_lock);
  1407. fsnotify_d_instantiate(dentry, inode);
  1408. }
  1409. /**
  1410. * d_instantiate - fill in inode information for a dentry
  1411. * @entry: dentry to complete
  1412. * @inode: inode to attach to this dentry
  1413. *
  1414. * Fill in inode information in the entry.
  1415. *
  1416. * This turns negative dentries into productive full members
  1417. * of society.
  1418. *
  1419. * NOTE! This assumes that the inode count has been incremented
  1420. * (or otherwise set) by the caller to indicate that it is now
  1421. * in use by the dcache.
  1422. */
  1423. void d_instantiate(struct dentry *entry, struct inode * inode)
  1424. {
  1425. BUG_ON(!hlist_unhashed(&entry->d_alias));
  1426. if (inode)
  1427. spin_lock(&inode->i_lock);
  1428. __d_instantiate(entry, inode);
  1429. if (inode)
  1430. spin_unlock(&inode->i_lock);
  1431. security_d_instantiate(entry, inode);
  1432. }
  1433. EXPORT_SYMBOL(d_instantiate);
  1434. /**
  1435. * d_instantiate_unique - instantiate a non-aliased dentry
  1436. * @entry: dentry to instantiate
  1437. * @inode: inode to attach to this dentry
  1438. *
  1439. * Fill in inode information in the entry. On success, it returns NULL.
  1440. * If an unhashed alias of "entry" already exists, then we return the
  1441. * aliased dentry instead and drop one reference to inode.
  1442. *
  1443. * Note that in order to avoid conflicts with rename() etc, the caller
  1444. * had better be holding the parent directory semaphore.
  1445. *
  1446. * This also assumes that the inode count has been incremented
  1447. * (or otherwise set) by the caller to indicate that it is now
  1448. * in use by the dcache.
  1449. */
  1450. static struct dentry *__d_instantiate_unique(struct dentry *entry,
  1451. struct inode *inode)
  1452. {
  1453. struct dentry *alias;
  1454. int len = entry->d_name.len;
  1455. const char *name = entry->d_name.name;
  1456. unsigned int hash = entry->d_name.hash;
  1457. if (!inode) {
  1458. __d_instantiate(entry, NULL);
  1459. return NULL;
  1460. }
  1461. hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
  1462. /*
  1463. * Don't need alias->d_lock here, because aliases with
  1464. * d_parent == entry->d_parent are not subject to name or
  1465. * parent changes, because the parent inode i_mutex is held.
  1466. */
  1467. if (alias->d_name.hash != hash)
  1468. continue;
  1469. if (alias->d_parent != entry->d_parent)
  1470. continue;
  1471. if (alias->d_name.len != len)
  1472. continue;
  1473. if (dentry_cmp(alias, name, len))
  1474. continue;
  1475. __dget(alias);
  1476. return alias;
  1477. }
  1478. __d_instantiate(entry, inode);
  1479. return NULL;
  1480. }
  1481. struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
  1482. {
  1483. struct dentry *result;
  1484. BUG_ON(!hlist_unhashed(&entry->d_alias));
  1485. if (inode)
  1486. spin_lock(&inode->i_lock);
  1487. result = __d_instantiate_unique(entry, inode);
  1488. if (inode)
  1489. spin_unlock(&inode->i_lock);
  1490. if (!result) {
  1491. security_d_instantiate(entry, inode);
  1492. return NULL;
  1493. }
  1494. BUG_ON(!d_unhashed(result));
  1495. iput(inode);
  1496. return result;
  1497. }
  1498. EXPORT_SYMBOL(d_instantiate_unique);
  1499. struct dentry *d_make_root(struct inode *root_inode)
  1500. {
  1501. struct dentry *res = NULL;
  1502. if (root_inode) {
  1503. static const struct qstr name = QSTR_INIT("/", 1);
  1504. res = __d_alloc(root_inode->i_sb, &name);
  1505. if (res)
  1506. d_instantiate(res, root_inode);
  1507. else
  1508. iput(root_inode);
  1509. }
  1510. return res;
  1511. }
  1512. EXPORT_SYMBOL(d_make_root);
  1513. static struct dentry * __d_find_any_alias(struct inode *inode)
  1514. {
  1515. struct dentry *alias;
  1516. if (hlist_empty(&inode->i_dentry))
  1517. return NULL;
  1518. alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
  1519. __dget(alias);
  1520. return alias;
  1521. }
  1522. /**
  1523. * d_find_any_alias - find any alias for a given inode
  1524. * @inode: inode to find an alias for
  1525. *
  1526. * If any aliases exist for the given inode, take and return a
  1527. * reference for one of them. If no aliases exist, return %NULL.
  1528. */
  1529. struct dentry *d_find_any_alias(struct inode *inode)
  1530. {
  1531. struct dentry *de;
  1532. spin_lock(&inode->i_lock);
  1533. de = __d_find_any_alias(inode);
  1534. spin_unlock(&inode->i_lock);
  1535. return de;
  1536. }
  1537. EXPORT_SYMBOL(d_find_any_alias);
  1538. /**
  1539. * d_obtain_alias - find or allocate a dentry for a given inode
  1540. * @inode: inode to allocate the dentry for
  1541. *
  1542. * Obtain a dentry for an inode resulting from NFS filehandle conversion or
  1543. * similar open by handle operations. The returned dentry may be anonymous,
  1544. * or may have a full name (if the inode was already in the cache).
  1545. *
  1546. * When called on a directory inode, we must ensure that the inode only ever
  1547. * has one dentry. If a dentry is found, that is returned instead of
  1548. * allocating a new one.
  1549. *
  1550. * On successful return, the reference to the inode has been transferred
  1551. * to the dentry. In case of an error the reference on the inode is released.
  1552. * To make it easier to use in export operations a %NULL or IS_ERR inode may
  1553. * be passed in and will be the error will be propagate to the return value,
  1554. * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
  1555. */
  1556. struct dentry *d_obtain_alias(struct inode *inode)
  1557. {
  1558. static const struct qstr anonstring = QSTR_INIT("/", 1);
  1559. struct dentry *tmp;
  1560. struct dentry *res;
  1561. if (!inode)
  1562. return ERR_PTR(-ESTALE);
  1563. if (IS_ERR(inode))
  1564. return ERR_CAST(inode);
  1565. res = d_find_any_alias(inode);
  1566. if (res)
  1567. goto out_iput;
  1568. tmp = __d_alloc(inode->i_sb, &anonstring);
  1569. if (!tmp) {
  1570. res = ERR_PTR(-ENOMEM);
  1571. goto out_iput;
  1572. }
  1573. spin_lock(&inode->i_lock);
  1574. res = __d_find_any_alias(inode);
  1575. if (res) {
  1576. spin_unlock(&inode->i_lock);
  1577. dput(tmp);
  1578. goto out_iput;
  1579. }
  1580. /* attach a disconnected dentry */
  1581. spin_lock(&tmp->d_lock);
  1582. tmp->d_inode = inode;
  1583. tmp->d_flags |= DCACHE_DISCONNECTED;
  1584. hlist_add_head(&tmp->d_alias, &inode->i_dentry);
  1585. hlist_bl_lock(&tmp->d_sb->s_anon);
  1586. hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
  1587. hlist_bl_unlock(&tmp->d_sb->s_anon);
  1588. spin_unlock(&tmp->d_lock);
  1589. spin_unlock(&inode->i_lock);
  1590. security_d_instantiate(tmp, inode);
  1591. return tmp;
  1592. out_iput:
  1593. if (res && !IS_ERR(res))
  1594. security_d_instantiate(res, inode);
  1595. iput(inode);
  1596. return res;
  1597. }
  1598. EXPORT_SYMBOL(d_obtain_alias);
  1599. /**
  1600. * d_splice_alias - splice a disconnected dentry into the tree if one exists
  1601. * @inode: the inode which may have a disconnected dentry
  1602. * @dentry: a negative dentry which we want to point to the inode.
  1603. *
  1604. * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
  1605. * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
  1606. * and return it, else simply d_add the inode to the dentry and return NULL.
  1607. *
  1608. * This is needed in the lookup routine of any filesystem that is exportable
  1609. * (via knfsd) so that we can build dcache paths to directories effectively.
  1610. *
  1611. * If a dentry was found and moved, then it is returned. Otherwise NULL
  1612. * is returned. This matches the expected return value of ->lookup.
  1613. *
  1614. * Cluster filesystems may call this function with a negative, hashed dentry.
  1615. * In that case, we know that the inode will be a regular file, and also this
  1616. * will only occur during atomic_open. So we need to check for the dentry
  1617. * being already hashed only in the final case.
  1618. */
  1619. struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
  1620. {
  1621. struct dentry *new = NULL;
  1622. if (IS_ERR(inode))
  1623. return ERR_CAST(inode);
  1624. if (inode && S_ISDIR(inode->i_mode)) {
  1625. spin_lock(&inode->i_lock);
  1626. new = __d_find_alias(inode, 1);
  1627. if (new) {
  1628. BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
  1629. spin_unlock(&inode->i_lock);
  1630. security_d_instantiate(new, inode);
  1631. d_move(new, dentry);
  1632. iput(inode);
  1633. } else {
  1634. /* already taking inode->i_lock, so d_add() by hand */
  1635. __d_instantiate(dentry, inode);
  1636. spin_unlock(&inode->i_lock);
  1637. security_d_instantiate(dentry, inode);
  1638. d_rehash(dentry);
  1639. }
  1640. } else {
  1641. d_instantiate(dentry, inode);
  1642. if (d_unhashed(dentry))
  1643. d_rehash(dentry);
  1644. }
  1645. return new;
  1646. }
  1647. EXPORT_SYMBOL(d_splice_alias);
  1648. /**
  1649. * d_add_ci - lookup or allocate new dentry with case-exact name
  1650. * @inode: the inode case-insensitive lookup has found
  1651. * @dentry: the negative dentry that was passed to the parent's lookup func
  1652. * @name: the case-exact name to be associated with the returned dentry
  1653. *
  1654. * This is to avoid filling the dcache with case-insensitive names to the
  1655. * same inode, only the actual correct case is stored in the dcache for
  1656. * case-insensitive filesystems.
  1657. *
  1658. * For a case-insensitive lookup match and if the the case-exact dentry
  1659. * already exists in in the dcache, use it and return it.
  1660. *
  1661. * If no entry exists with the exact case name, allocate new dentry with
  1662. * the exact case, and return the spliced entry.
  1663. */
  1664. struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
  1665. struct qstr *name)
  1666. {
  1667. struct dentry *found;
  1668. struct dentry *new;
  1669. /*
  1670. * First check if a dentry matching the name already exists,
  1671. * if not go ahead and create it now.
  1672. */
  1673. found = d_hash_and_lookup(dentry->d_parent, name);
  1674. if (unlikely(IS_ERR(found)))
  1675. goto err_out;
  1676. if (!found) {
  1677. new = d_alloc(dentry->d_parent, name);
  1678. if (!new) {
  1679. found = ERR_PTR(-ENOMEM);
  1680. goto err_out;
  1681. }
  1682. found = d_splice_alias(inode, new);
  1683. if (found) {
  1684. dput(new);
  1685. return found;
  1686. }
  1687. return new;
  1688. }
  1689. /*
  1690. * If a matching dentry exists, and it's not negative use it.
  1691. *
  1692. * Decrement the reference count to balance the iget() done
  1693. * earlier on.
  1694. */
  1695. if (found->d_inode) {
  1696. if (unlikely(found->d_inode != inode)) {
  1697. /* This can't happen because bad inodes are unhashed. */
  1698. BUG_ON(!is_bad_inode(inode));
  1699. BUG_ON(!is_bad_inode(found->d_inode));
  1700. }
  1701. iput(inode);
  1702. return found;
  1703. }
  1704. /*
  1705. * Negative dentry: instantiate it unless the inode is a directory and
  1706. * already has a dentry.
  1707. */
  1708. new = d_splice_alias(inode, found);
  1709. if (new) {
  1710. dput(found);
  1711. found = new;
  1712. }
  1713. return found;
  1714. err_out:
  1715. iput(inode);
  1716. return found;
  1717. }
  1718. EXPORT_SYMBOL(d_add_ci);
  1719. /*
  1720. * Do the slow-case of the dentry name compare.
  1721. *
  1722. * Unlike the dentry_cmp() function, we need to atomically
  1723. * load the name and length information, so that the
  1724. * filesystem can rely on them, and can use the 'name' and
  1725. * 'len' information without worrying about walking off the
  1726. * end of memory etc.
  1727. *
  1728. * Thus the read_seqcount_retry() and the "duplicate" info
  1729. * in arguments (the low-level filesystem should not look
  1730. * at the dentry inode or name contents directly, since
  1731. * rename can change them while we're in RCU mode).
  1732. */
  1733. enum slow_d_compare {
  1734. D_COMP_OK,
  1735. D_COMP_NOMATCH,
  1736. D_COMP_SEQRETRY,
  1737. };
  1738. static noinline enum slow_d_compare slow_dentry_cmp(
  1739. const struct dentry *parent,
  1740. struct dentry *dentry,
  1741. unsigned int seq,
  1742. const struct qstr *name)
  1743. {
  1744. int tlen = dentry->d_name.len;
  1745. const char *tname = dentry->d_name.name;
  1746. if (read_seqcount_retry(&dentry->d_seq, seq)) {
  1747. cpu_relax();
  1748. return D_COMP_SEQRETRY;
  1749. }
  1750. if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
  1751. return D_COMP_NOMATCH;
  1752. return D_COMP_OK;
  1753. }
  1754. /**
  1755. * __d_lookup_rcu - search for a dentry (racy, store-free)
  1756. * @parent: parent dentry
  1757. * @name: qstr of name we wish to find
  1758. * @seqp: returns d_seq value at the point where the dentry was found
  1759. * Returns: dentry, or NULL
  1760. *
  1761. * __d_lookup_rcu is the dcache lookup function for rcu-walk name
  1762. * resolution (store-free path walking) design described in
  1763. * Documentation/filesystems/path-lookup.txt.
  1764. *
  1765. * This is not to be used outside core vfs.
  1766. *
  1767. * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
  1768. * held, and rcu_read_lock held. The returned dentry must not be stored into
  1769. * without taking d_lock and checking d_seq sequence count against @seq
  1770. * returned here.
  1771. *
  1772. * A refcount may be taken on the found dentry with the d_rcu_to_refcount
  1773. * function.
  1774. *
  1775. * Alternatively, __d_lookup_rcu may be called again to look up the child of
  1776. * the returned dentry, so long as its parent's seqlock is checked after the
  1777. * child is looked up. Thus, an interlocking stepping of sequence lock checks
  1778. * is formed, giving integrity down the path walk.
  1779. *
  1780. * NOTE! The caller *has* to check the resulting dentry against the sequence
  1781. * number we've returned before using any of the resulting dentry state!
  1782. */
  1783. struct dentry *__d_lookup_rcu(const struct dentry *parent,
  1784. const struct qstr *name,
  1785. unsigned *seqp)
  1786. {
  1787. u64 hashlen = name->hash_len;
  1788. const unsigned char *str = name->name;
  1789. struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
  1790. struct hlist_bl_node *node;
  1791. struct dentry *dentry;
  1792. /*
  1793. * Note: There is significant duplication with __d_lookup_rcu which is
  1794. * required to prevent single threaded performance regressions
  1795. * especially on architectures where smp_rmb (in seqcounts) are costly.
  1796. * Keep the two functions in sync.
  1797. */
  1798. /*
  1799. * The hash list is protected using RCU.
  1800. *
  1801. * Carefully use d_seq when comparing a candidate dentry, to avoid
  1802. * races with d_move().
  1803. *
  1804. * It is possible that concurrent renames can mess up our list
  1805. * walk here and result in missing our dentry, resulting in the
  1806. * false-negative result. d_lookup() protects against concurrent
  1807. * renames using rename_lock seqlock.
  1808. *
  1809. * See Documentation/filesystems/path-lookup.txt for more details.
  1810. */
  1811. hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
  1812. unsigned seq;
  1813. seqretry:
  1814. /*
  1815. * The dentry sequence count protects us from concurrent
  1816. * renames, and thus protects parent and name fields.
  1817. *
  1818. * The caller must perform a seqcount check in order
  1819. * to do anything useful with the returned dentry.
  1820. *
  1821. * NOTE! We do a "raw" seqcount_begin here. That means that
  1822. * we don't wait for the sequence count to stabilize if it
  1823. * is in the middle of a sequence change. If we do the slow
  1824. * dentry compare, we will do seqretries until it is stable,
  1825. * and if we end up with a successful lookup, we actually
  1826. * want to exit RCU lookup anyway.
  1827. */
  1828. seq = raw_seqcount_begin(&dentry->d_seq);
  1829. if (dentry->d_parent != parent)
  1830. continue;
  1831. if (d_unhashed(dentry))
  1832. continue;
  1833. if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
  1834. if (dentry->d_name.hash != hashlen_hash(hashlen))
  1835. continue;
  1836. *seqp = seq;
  1837. switch (slow_dentry_cmp(parent, dentry, seq, name)) {
  1838. case D_COMP_OK:
  1839. return dentry;
  1840. case D_COMP_NOMATCH:
  1841. continue;
  1842. default:
  1843. goto seqretry;
  1844. }
  1845. }
  1846. if (dentry->d_name.hash_len != hashlen)
  1847. continue;
  1848. *seqp = seq;
  1849. if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
  1850. return dentry;
  1851. }
  1852. return NULL;
  1853. }
  1854. /**
  1855. * d_lookup - search for a dentry
  1856. * @parent: parent dentry
  1857. * @name: qstr of name we wish to find
  1858. * Returns: dentry, or NULL
  1859. *
  1860. * d_lookup searches the children of the parent dentry for the name in
  1861. * question. If the dentry is found its reference count is incremented and the
  1862. * dentry is returned. The caller must use dput to free the entry when it has
  1863. * finished using it. %NULL is returned if the dentry does not exist.
  1864. */
  1865. struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
  1866. {
  1867. struct dentry *dentry;
  1868. unsigned seq;
  1869. do {
  1870. seq = read_seqbegin(&rename_lock);
  1871. dentry = __d_lookup(parent, name);
  1872. if (dentry)
  1873. break;
  1874. } while (read_seqretry(&rename_lock, seq));
  1875. return dentry;
  1876. }
  1877. EXPORT_SYMBOL(d_lookup);
  1878. /**
  1879. * __d_lookup - search for a dentry (racy)
  1880. * @parent: parent dentry
  1881. * @name: qstr of name we wish to find
  1882. * Returns: dentry, or NULL
  1883. *
  1884. * __d_lookup is like d_lookup, however it may (rarely) return a
  1885. * false-negative result due to unrelated rename activity.
  1886. *
  1887. * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
  1888. * however it must be used carefully, eg. with a following d_lookup in
  1889. * the case of failure.
  1890. *
  1891. * __d_lookup callers must be commented.
  1892. */
  1893. struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
  1894. {
  1895. unsigned int len = name->len;
  1896. unsigned int hash = name->hash;
  1897. const unsigned char *str = name->name;
  1898. struct hlist_bl_head *b = d_hash(parent, hash);
  1899. struct hlist_bl_node *node;
  1900. struct dentry *found = NULL;
  1901. struct dentry *dentry;
  1902. /*
  1903. * Note: There is significant duplication with __d_lookup_rcu which is
  1904. * required to prevent single threaded performance regressions
  1905. * especially on architectures where smp_rmb (in seqcounts) are costly.
  1906. * Keep the two functions in sync.
  1907. */
  1908. /*
  1909. * The hash list is protected using RCU.
  1910. *
  1911. * Take d_lock when comparing a candidate dentry, to avoid races
  1912. * with d_move().
  1913. *
  1914. * It is possible that concurrent renames can mess up our list
  1915. * walk here and result in missing our dentry, resulting in the
  1916. * false-negative result. d_lookup() protects against concurrent
  1917. * renames using rename_lock seqlock.
  1918. *
  1919. * See Documentation/filesystems/path-lookup.txt for more details.
  1920. */
  1921. rcu_read_lock();
  1922. hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
  1923. if (dentry->d_name.hash != hash)
  1924. continue;
  1925. spin_lock(&dentry->d_lock);
  1926. if (dentry->d_parent != parent)
  1927. goto next;
  1928. if (d_unhashed(dentry))
  1929. goto next;
  1930. /*
  1931. * It is safe to compare names since d_move() cannot
  1932. * change the qstr (protected by d_lock).
  1933. */
  1934. if (parent->d_flags & DCACHE_OP_COMPARE) {
  1935. int tlen = dentry->d_name.len;
  1936. const char *tname = dentry->d_name.name;
  1937. if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
  1938. goto next;
  1939. } else {
  1940. if (dentry->d_name.len != len)
  1941. goto next;
  1942. if (dentry_cmp(dentry, str, len))
  1943. goto next;
  1944. }
  1945. dentry->d_lockref.count++;
  1946. found = dentry;
  1947. spin_unlock(&dentry->d_lock);
  1948. break;
  1949. next:
  1950. spin_unlock(&dentry->d_lock);
  1951. }
  1952. rcu_read_unlock();
  1953. return found;
  1954. }
  1955. /**
  1956. * d_hash_and_lookup - hash the qstr then search for a dentry
  1957. * @dir: Directory to search in
  1958. * @name: qstr of name we wish to find
  1959. *
  1960. * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
  1961. */
  1962. struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
  1963. {
  1964. /*
  1965. * Check for a fs-specific hash function. Note that we must
  1966. * calculate the standard hash first, as the d_op->d_hash()
  1967. * routine may choose to leave the hash value unchanged.
  1968. */
  1969. name->hash = full_name_hash(name->name, name->len);
  1970. if (dir->d_flags & DCACHE_OP_HASH) {
  1971. int err = dir->d_op->d_hash(dir, name);
  1972. if (unlikely(err < 0))
  1973. return ERR_PTR(err);
  1974. }
  1975. return d_lookup(dir, name);
  1976. }
  1977. EXPORT_SYMBOL(d_hash_and_lookup);
  1978. /**
  1979. * d_validate - verify dentry provided from insecure source (deprecated)
  1980. * @dentry: The dentry alleged to be valid child of @dparent
  1981. * @dparent: The parent dentry (known to be valid)
  1982. *
  1983. * An insecure source has sent us a dentry, here we verify it and dget() it.
  1984. * This is used by ncpfs in its readdir implementation.
  1985. * Zero is returned in the dentry is invalid.
  1986. *
  1987. * This function is slow for big directories, and deprecated, do not use it.
  1988. */
  1989. int d_validate(struct dentry *dentry, struct dentry *dparent)
  1990. {
  1991. struct dentry *child;
  1992. spin_lock(&dparent->d_lock);
  1993. list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
  1994. if (dentry == child) {
  1995. spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
  1996. __dget_dlock(dentry);
  1997. spin_unlock(&dentry->d_lock);
  1998. spin_unlock(&dparent->d_lock);
  1999. return 1;
  2000. }
  2001. }
  2002. spin_unlock(&dparent->d_lock);
  2003. return 0;
  2004. }
  2005. EXPORT_SYMBOL(d_validate);
  2006. /*
  2007. * When a file is deleted, we have two options:
  2008. * - turn this dentry into a negative dentry
  2009. * - unhash this dentry and free it.
  2010. *
  2011. * Usually, we want to just turn this into
  2012. * a negative dentry, but if anybody else is
  2013. * currently using the dentry or the inode
  2014. * we can't do that and we fall back on removing
  2015. * it from the hash queues and waiting for
  2016. * it to be deleted later when it has no users
  2017. */
  2018. /**
  2019. * d_delete - delete a dentry
  2020. * @dentry: The dentry to delete
  2021. *
  2022. * Turn the dentry into a negative dentry if possible, otherwise
  2023. * remove it from the hash queues so it can be deleted later
  2024. */
  2025. void d_delete(struct dentry * dentry)
  2026. {
  2027. struct inode *inode;
  2028. int isdir = 0;
  2029. /*
  2030. * Are we the only user?
  2031. */
  2032. again:
  2033. spin_lock(&dentry->d_lock);
  2034. inode = dentry->d_inode;
  2035. isdir = S_ISDIR(inode->i_mode);
  2036. if (dentry->d_lockref.count == 1) {
  2037. if (!spin_trylock(&inode->i_lock)) {
  2038. spin_unlock(&dentry->d_lock);
  2039. cpu_relax();
  2040. goto again;
  2041. }
  2042. dentry->d_flags &= ~DCACHE_CANT_MOUNT;
  2043. dentry_unlink_inode(dentry);
  2044. fsnotify_nameremove(dentry, isdir);
  2045. return;
  2046. }
  2047. if (!d_unhashed(dentry))
  2048. __d_drop(dentry);
  2049. spin_unlock(&dentry->d_lock);
  2050. fsnotify_nameremove(dentry, isdir);
  2051. }
  2052. EXPORT_SYMBOL(d_delete);
  2053. static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
  2054. {
  2055. BUG_ON(!d_unhashed(entry));
  2056. hlist_bl_lock(b);
  2057. entry->d_flags |= DCACHE_RCUACCESS;
  2058. hlist_bl_add_head_rcu(&entry->d_hash, b);
  2059. hlist_bl_unlock(b);
  2060. }
  2061. static void _d_rehash(struct dentry * entry)
  2062. {
  2063. __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
  2064. }
  2065. /**
  2066. * d_rehash - add an entry back to the hash
  2067. * @entry: dentry to add to the hash
  2068. *
  2069. * Adds a dentry to the hash according to its name.
  2070. */
  2071. void d_rehash(struct dentry * entry)
  2072. {
  2073. spin_lock(&entry->d_lock);
  2074. _d_rehash(entry);
  2075. spin_unlock(&entry->d_lock);
  2076. }
  2077. EXPORT_SYMBOL(d_rehash);
  2078. /**
  2079. * dentry_update_name_case - update case insensitive dentry with a new name
  2080. * @dentry: dentry to be updated
  2081. * @name: new name
  2082. *
  2083. * Update a case insensitive dentry with new case of name.
  2084. *
  2085. * dentry must have been returned by d_lookup with name @name. Old and new
  2086. * name lengths must match (ie. no d_compare which allows mismatched name
  2087. * lengths).
  2088. *
  2089. * Parent inode i_mutex must be held over d_lookup and into this call (to
  2090. * keep renames and concurrent inserts, and readdir(2) away).
  2091. */
  2092. void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
  2093. {
  2094. BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
  2095. BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
  2096. spin_lock(&dentry->d_lock);
  2097. write_seqcount_begin(&dentry->d_seq);
  2098. memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
  2099. write_seqcount_end(&dentry->d_seq);
  2100. spin_unlock(&dentry->d_lock);
  2101. }
  2102. EXPORT_SYMBOL(dentry_update_name_case);
  2103. static void switch_names(struct dentry *dentry, struct dentry *target)
  2104. {
  2105. if (dname_external(target)) {
  2106. if (dname_external(dentry)) {
  2107. /*
  2108. * Both external: swap the pointers
  2109. */
  2110. swap(target->d_name.name, dentry->d_name.name);
  2111. } else {
  2112. /*
  2113. * dentry:internal, target:external. Steal target's
  2114. * storage and make target internal.
  2115. */
  2116. memcpy(target->d_iname, dentry->d_name.name,
  2117. dentry->d_name.len + 1);
  2118. dentry->d_name.name = target->d_name.name;
  2119. target->d_name.name = target->d_iname;
  2120. }
  2121. } else {
  2122. if (dname_external(dentry)) {
  2123. /*
  2124. * dentry:external, target:internal. Give dentry's
  2125. * storage to target and make dentry internal
  2126. */
  2127. memcpy(dentry->d_iname, target->d_name.name,
  2128. target->d_name.len + 1);
  2129. target->d_name.name = dentry->d_name.name;
  2130. dentry->d_name.name = dentry->d_iname;
  2131. } else {
  2132. /*
  2133. * Both are internal. Just copy target to dentry
  2134. */
  2135. memcpy(dentry->d_iname, target->d_name.name,
  2136. target->d_name.len + 1);
  2137. dentry->d_name.len = target->d_name.len;
  2138. return;
  2139. }
  2140. }
  2141. swap(dentry->d_name.len, target->d_name.len);
  2142. }
  2143. static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
  2144. {
  2145. /*
  2146. * XXXX: do we really need to take target->d_lock?
  2147. */
  2148. if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
  2149. spin_lock(&target->d_parent->d_lock);
  2150. else {
  2151. if (d_ancestor(dentry->d_parent, target->d_parent)) {
  2152. spin_lock(&dentry->d_parent->d_lock);
  2153. spin_lock_nested(&target->d_parent->d_lock,
  2154. DENTRY_D_LOCK_NESTED);
  2155. } else {
  2156. spin_lock(&target->d_parent->d_lock);
  2157. spin_lock_nested(&dentry->d_parent->d_lock,
  2158. DENTRY_D_LOCK_NESTED);
  2159. }
  2160. }
  2161. if (target < dentry) {
  2162. spin_lock_nested(&target->d_lock, 2);
  2163. spin_lock_nested(&dentry->d_lock, 3);
  2164. } else {
  2165. spin_lock_nested(&dentry->d_lock, 2);
  2166. spin_lock_nested(&target->d_lock, 3);
  2167. }
  2168. }
  2169. static void dentry_unlock_parents_for_move(struct dentry *dentry,
  2170. struct dentry *target)
  2171. {
  2172. if (target->d_parent != dentry->d_parent)
  2173. spin_unlock(&dentry->d_parent->d_lock);
  2174. if (target->d_parent != target)
  2175. spin_unlock(&target->d_parent->d_lock);
  2176. }
  2177. /*
  2178. * When switching names, the actual string doesn't strictly have to
  2179. * be preserved in the target - because we're dropping the target
  2180. * anyway. As such, we can just do a simple memcpy() to copy over
  2181. * the new name before we switch.
  2182. *
  2183. * Note that we have to be a lot more careful about getting the hash
  2184. * switched - we have to switch the hash value properly even if it
  2185. * then no longer matches the actual (corrupted) string of the target.
  2186. * The hash value has to match the hash queue that the dentry is on..
  2187. */
  2188. /*
  2189. * __d_move - move a dentry
  2190. * @dentry: entry to move
  2191. * @target: new dentry
  2192. *
  2193. * Update the dcache to reflect the move of a file name. Negative
  2194. * dcache entries should not be moved in this way. Caller must hold
  2195. * rename_lock, the i_mutex of the source and target directories,
  2196. * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
  2197. */
  2198. static void __d_move(struct dentry * dentry, struct dentry * target)
  2199. {
  2200. if (!dentry->d_inode)
  2201. printk(KERN_WARNING "VFS: moving negative dcache entry\n");
  2202. BUG_ON(d_ancestor(dentry, target));
  2203. BUG_ON(d_ancestor(target, dentry));
  2204. dentry_lock_for_move(dentry, target);
  2205. write_seqcount_begin(&dentry->d_seq);
  2206. write_seqcount_begin(&target->d_seq);
  2207. /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
  2208. /*
  2209. * Move the dentry to the target hash queue. Don't bother checking
  2210. * for the same hash queue because of how unlikely it is.
  2211. */
  2212. __d_drop(dentry);
  2213. __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
  2214. /* Unhash the target: dput() will then get rid of it */
  2215. __d_drop(target);
  2216. list_del(&dentry->d_u.d_child);
  2217. list_del(&target->d_u.d_child);
  2218. /* Switch the names.. */
  2219. switch_names(dentry, target);
  2220. swap(dentry->d_name.hash, target->d_name.hash);
  2221. /* ... and switch the parents */
  2222. if (IS_ROOT(dentry)) {
  2223. dentry->d_parent = target->d_parent;
  2224. target->d_parent = target;
  2225. INIT_LIST_HEAD(&target->d_u.d_child);
  2226. } else {
  2227. swap(dentry->d_parent, target->d_parent);
  2228. /* And add them back to the (new) parent lists */
  2229. list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
  2230. }
  2231. list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
  2232. write_seqcount_end(&target->d_seq);
  2233. write_seqcount_end(&dentry->d_seq);
  2234. dentry_unlock_parents_for_move(dentry, target);
  2235. spin_unlock(&target->d_lock);
  2236. fsnotify_d_move(dentry);
  2237. spin_unlock(&dentry->d_lock);
  2238. }
  2239. /*
  2240. * d_move - move a dentry
  2241. * @dentry: entry to move
  2242. * @target: new dentry
  2243. *
  2244. * Update the dcache to reflect the move of a file name. Negative
  2245. * dcache entries should not be moved in this way. See the locking
  2246. * requirements for __d_move.
  2247. */
  2248. void d_move(struct dentry *dentry, struct dentry *target)
  2249. {
  2250. write_seqlock(&rename_lock);
  2251. __d_move(dentry, target);
  2252. write_sequnlock(&rename_lock);
  2253. }
  2254. EXPORT_SYMBOL(d_move);
  2255. /**
  2256. * d_ancestor - search for an ancestor
  2257. * @p1: ancestor dentry
  2258. * @p2: child dentry
  2259. *
  2260. * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
  2261. * an ancestor of p2, else NULL.
  2262. */
  2263. struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
  2264. {
  2265. struct dentry *p;
  2266. for (p = p2; !IS_ROOT(p); p = p->d_parent) {
  2267. if (p->d_parent == p1)
  2268. return p;
  2269. }
  2270. return NULL;
  2271. }
  2272. /*
  2273. * This helper attempts to cope with remotely renamed directories
  2274. *
  2275. * It assumes that the caller is already holding
  2276. * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
  2277. *
  2278. * Note: If ever the locking in lock_rename() changes, then please
  2279. * remember to update this too...
  2280. */
  2281. static struct dentry *__d_unalias(struct inode *inode,
  2282. struct dentry *dentry, struct dentry *alias)
  2283. {
  2284. struct mutex *m1 = NULL, *m2 = NULL;
  2285. struct dentry *ret = ERR_PTR(-EBUSY);
  2286. /* If alias and dentry share a parent, then no extra locks required */
  2287. if (alias->d_parent == dentry->d_parent)
  2288. goto out_unalias;
  2289. /* See lock_rename() */
  2290. if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
  2291. goto out_err;
  2292. m1 = &dentry->d_sb->s_vfs_rename_mutex;
  2293. if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
  2294. goto out_err;
  2295. m2 = &alias->d_parent->d_inode->i_mutex;
  2296. out_unalias:
  2297. if (likely(!d_mountpoint(alias))) {
  2298. __d_move(alias, dentry);
  2299. ret = alias;
  2300. }
  2301. out_err:
  2302. spin_unlock(&inode->i_lock);
  2303. if (m2)
  2304. mutex_unlock(m2);
  2305. if (m1)
  2306. mutex_unlock(m1);
  2307. return ret;
  2308. }
  2309. /*
  2310. * Prepare an anonymous dentry for life in the superblock's dentry tree as a
  2311. * named dentry in place of the dentry to be replaced.
  2312. * returns with anon->d_lock held!
  2313. */
  2314. static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
  2315. {
  2316. struct dentry *dparent;
  2317. dentry_lock_for_move(anon, dentry);
  2318. write_seqcount_begin(&dentry->d_seq);
  2319. write_seqcount_begin(&anon->d_seq);
  2320. dparent = dentry->d_parent;
  2321. switch_names(dentry, anon);
  2322. swap(dentry->d_name.hash, anon->d_name.hash);
  2323. dentry->d_parent = dentry;
  2324. list_del_init(&dentry->d_u.d_child);
  2325. anon->d_parent = dparent;
  2326. list_move(&anon->d_u.d_child, &dparent->d_subdirs);
  2327. write_seqcount_end(&dentry->d_seq);
  2328. write_seqcount_end(&anon->d_seq);
  2329. dentry_unlock_parents_for_move(anon, dentry);
  2330. spin_unlock(&dentry->d_lock);
  2331. /* anon->d_lock still locked, returns locked */
  2332. anon->d_flags &= ~DCACHE_DISCONNECTED;
  2333. }
  2334. /**
  2335. * d_materialise_unique - introduce an inode into the tree
  2336. * @dentry: candidate dentry
  2337. * @inode: inode to bind to the dentry, to which aliases may be attached
  2338. *
  2339. * Introduces an dentry into the tree, substituting an extant disconnected
  2340. * root directory alias in its place if there is one. Caller must hold the
  2341. * i_mutex of the parent directory.
  2342. */
  2343. struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
  2344. {
  2345. struct dentry *actual;
  2346. BUG_ON(!d_unhashed(dentry));
  2347. if (!inode) {
  2348. actual = dentry;
  2349. __d_instantiate(dentry, NULL);
  2350. d_rehash(actual);
  2351. goto out_nolock;
  2352. }
  2353. spin_lock(&inode->i_lock);
  2354. if (S_ISDIR(inode->i_mode)) {
  2355. struct dentry *alias;
  2356. /* Does an aliased dentry already exist? */
  2357. alias = __d_find_alias(inode, 0);
  2358. if (alias) {
  2359. actual = alias;
  2360. write_seqlock(&rename_lock);
  2361. if (d_ancestor(alias, dentry)) {
  2362. /* Check for loops */
  2363. actual = ERR_PTR(-ELOOP);
  2364. spin_unlock(&inode->i_lock);
  2365. } else if (IS_ROOT(alias)) {
  2366. /* Is this an anonymous mountpoint that we
  2367. * could splice into our tree? */
  2368. __d_materialise_dentry(dentry, alias);
  2369. write_sequnlock(&rename_lock);
  2370. __d_drop(alias);
  2371. goto found;
  2372. } else {
  2373. /* Nope, but we must(!) avoid directory
  2374. * aliasing. This drops inode->i_lock */
  2375. actual = __d_unalias(inode, dentry, alias);
  2376. }
  2377. write_sequnlock(&rename_lock);
  2378. if (IS_ERR(actual)) {
  2379. if (PTR_ERR(actual) == -ELOOP)
  2380. pr_warn_ratelimited(
  2381. "VFS: Lookup of '%s' in %s %s"
  2382. " would have caused loop\n",
  2383. dentry->d_name.name,
  2384. inode->i_sb->s_type->name,
  2385. inode->i_sb->s_id);
  2386. dput(alias);
  2387. }
  2388. goto out_nolock;
  2389. }
  2390. }
  2391. /* Add a unique reference */
  2392. actual = __d_instantiate_unique(dentry, inode);
  2393. if (!actual)
  2394. actual = dentry;
  2395. else
  2396. BUG_ON(!d_unhashed(actual));
  2397. spin_lock(&actual->d_lock);
  2398. found:
  2399. _d_rehash(actual);
  2400. spin_unlock(&actual->d_lock);
  2401. spin_unlock(&inode->i_lock);
  2402. out_nolock:
  2403. if (actual == dentry) {
  2404. security_d_instantiate(dentry, inode);
  2405. return NULL;
  2406. }
  2407. iput(inode);
  2408. return actual;
  2409. }
  2410. EXPORT_SYMBOL_GPL(d_materialise_unique);
  2411. static int prepend(char **buffer, int *buflen, const char *str, int namelen)
  2412. {
  2413. *buflen -= namelen;
  2414. if (*buflen < 0)
  2415. return -ENAMETOOLONG;
  2416. *buffer -= namelen;
  2417. memcpy(*buffer, str, namelen);
  2418. return 0;
  2419. }
  2420. /**
  2421. * prepend_name - prepend a pathname in front of current buffer pointer
  2422. * buffer: buffer pointer
  2423. * buflen: allocated length of the buffer
  2424. * name: name string and length qstr structure
  2425. *
  2426. * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
  2427. * make sure that either the old or the new name pointer and length are
  2428. * fetched. However, there may be mismatch between length and pointer.
  2429. * The length cannot be trusted, we need to copy it byte-by-byte until
  2430. * the length is reached or a null byte is found. It also prepends "/" at
  2431. * the beginning of the name. The sequence number check at the caller will
  2432. * retry it again when a d_move() does happen. So any garbage in the buffer
  2433. * due to mismatched pointer and length will be discarded.
  2434. */
  2435. static int prepend_name(char **buffer, int *buflen, struct qstr *name)
  2436. {
  2437. const char *dname = ACCESS_ONCE(name->name);
  2438. u32 dlen = ACCESS_ONCE(name->len);
  2439. char *p;
  2440. if (*buflen < dlen + 1)
  2441. return -ENAMETOOLONG;
  2442. *buflen -= dlen + 1;
  2443. p = *buffer -= dlen + 1;
  2444. *p++ = '/';
  2445. while (dlen--) {
  2446. char c = *dname++;
  2447. if (!c)
  2448. break;
  2449. *p++ = c;
  2450. }
  2451. return 0;
  2452. }
  2453. /**
  2454. * prepend_path - Prepend path string to a buffer
  2455. * @path: the dentry/vfsmount to report
  2456. * @root: root vfsmnt/dentry
  2457. * @buffer: pointer to the end of the buffer
  2458. * @buflen: pointer to buffer length
  2459. *
  2460. * The function tries to write out the pathname without taking any lock other
  2461. * than the RCU read lock to make sure that dentries won't go away. It only
  2462. * checks the sequence number of the global rename_lock as any change in the
  2463. * dentry's d_seq will be preceded by changes in the rename_lock sequence
  2464. * number. If the sequence number had been change, it will restart the whole
  2465. * pathname back-tracing sequence again. It performs a total of 3 trials of
  2466. * lockless back-tracing sequences before falling back to take the
  2467. * rename_lock.
  2468. */
  2469. static int prepend_path(const struct path *path,
  2470. const struct path *root,
  2471. char **buffer, int *buflen)
  2472. {
  2473. struct dentry *dentry = path->dentry;
  2474. struct vfsmount *vfsmnt = path->mnt;
  2475. struct mount *mnt = real_mount(vfsmnt);
  2476. int error = 0;
  2477. unsigned seq = 0;
  2478. char *bptr;
  2479. int blen;
  2480. rcu_read_lock();
  2481. restart:
  2482. bptr = *buffer;
  2483. blen = *buflen;
  2484. read_seqbegin_or_lock(&rename_lock, &seq);
  2485. while (dentry != root->dentry || vfsmnt != root->mnt) {
  2486. struct dentry * parent;
  2487. if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
  2488. /* Global root? */
  2489. if (mnt_has_parent(mnt)) {
  2490. dentry = mnt->mnt_mountpoint;
  2491. mnt = mnt->mnt_parent;
  2492. vfsmnt = &mnt->mnt;
  2493. continue;
  2494. }
  2495. /*
  2496. * Filesystems needing to implement special "root names"
  2497. * should do so with ->d_dname()
  2498. */
  2499. if (IS_ROOT(dentry) &&
  2500. (dentry->d_name.len != 1 ||
  2501. dentry->d_name.name[0] != '/')) {
  2502. WARN(1, "Root dentry has weird name <%.*s>\n",
  2503. (int) dentry->d_name.len,
  2504. dentry->d_name.name);
  2505. }
  2506. if (!error)
  2507. error = is_mounted(vfsmnt) ? 1 : 2;
  2508. break;
  2509. }
  2510. parent = dentry->d_parent;
  2511. prefetch(parent);
  2512. error = prepend_name(&bptr, &blen, &dentry->d_name);
  2513. if (error)
  2514. break;
  2515. dentry = parent;
  2516. }
  2517. if (!(seq & 1))
  2518. rcu_read_unlock();
  2519. if (need_seqretry(&rename_lock, seq)) {
  2520. seq = 1;
  2521. goto restart;
  2522. }
  2523. done_seqretry(&rename_lock, seq);
  2524. if (error >= 0 && bptr == *buffer) {
  2525. if (--blen < 0)
  2526. error = -ENAMETOOLONG;
  2527. else
  2528. *--bptr = '/';
  2529. }
  2530. *buffer = bptr;
  2531. *buflen = blen;
  2532. return error;
  2533. }
  2534. /**
  2535. * __d_path - return the path of a dentry
  2536. * @path: the dentry/vfsmount to report
  2537. * @root: root vfsmnt/dentry
  2538. * @buf: buffer to return value in
  2539. * @buflen: buffer length
  2540. *
  2541. * Convert a dentry into an ASCII path name.
  2542. *
  2543. * Returns a pointer into the buffer or an error code if the
  2544. * path was too long.
  2545. *
  2546. * "buflen" should be positive.
  2547. *
  2548. * If the path is not reachable from the supplied root, return %NULL.
  2549. */
  2550. char *__d_path(const struct path *path,
  2551. const struct path *root,
  2552. char *buf, int buflen)
  2553. {
  2554. char *res = buf + buflen;
  2555. int error;
  2556. prepend(&res, &buflen, "\0", 1);
  2557. br_read_lock(&vfsmount_lock);
  2558. error = prepend_path(path, root, &res, &buflen);
  2559. br_read_unlock(&vfsmount_lock);
  2560. if (error < 0)
  2561. return ERR_PTR(error);
  2562. if (error > 0)
  2563. return NULL;
  2564. return res;
  2565. }
  2566. char *d_absolute_path(const struct path *path,
  2567. char *buf, int buflen)
  2568. {
  2569. struct path root = {};
  2570. char *res = buf + buflen;
  2571. int error;
  2572. prepend(&res, &buflen, "\0", 1);
  2573. br_read_lock(&vfsmount_lock);
  2574. error = prepend_path(path, &root, &res, &buflen);
  2575. br_read_unlock(&vfsmount_lock);
  2576. if (error > 1)
  2577. error = -EINVAL;
  2578. if (error < 0)
  2579. return ERR_PTR(error);
  2580. return res;
  2581. }
  2582. /*
  2583. * same as __d_path but appends "(deleted)" for unlinked files.
  2584. */
  2585. static int path_with_deleted(const struct path *path,
  2586. const struct path *root,
  2587. char **buf, int *buflen)
  2588. {
  2589. prepend(buf, buflen, "\0", 1);
  2590. if (d_unlinked(path->dentry)) {
  2591. int error = prepend(buf, buflen, " (deleted)", 10);
  2592. if (error)
  2593. return error;
  2594. }
  2595. return prepend_path(path, root, buf, buflen);
  2596. }
  2597. static int prepend_unreachable(char **buffer, int *buflen)
  2598. {
  2599. return prepend(buffer, buflen, "(unreachable)", 13);
  2600. }
  2601. /**
  2602. * d_path - return the path of a dentry
  2603. * @path: path to report
  2604. * @buf: buffer to return value in
  2605. * @buflen: buffer length
  2606. *
  2607. * Convert a dentry into an ASCII path name. If the entry has been deleted
  2608. * the string " (deleted)" is appended. Note that this is ambiguous.
  2609. *
  2610. * Returns a pointer into the buffer or an error code if the path was
  2611. * too long. Note: Callers should use the returned pointer, not the passed
  2612. * in buffer, to use the name! The implementation often starts at an offset
  2613. * into the buffer, and may leave 0 bytes at the start.
  2614. *
  2615. * "buflen" should be positive.
  2616. */
  2617. char *d_path(const struct path *path, char *buf, int buflen)
  2618. {
  2619. char *res = buf + buflen;
  2620. struct path root;
  2621. int error;
  2622. /*
  2623. * We have various synthetic filesystems that never get mounted. On
  2624. * these filesystems dentries are never used for lookup purposes, and
  2625. * thus don't need to be hashed. They also don't need a name until a
  2626. * user wants to identify the object in /proc/pid/fd/. The little hack
  2627. * below allows us to generate a name for these objects on demand:
  2628. */
  2629. if (path->dentry->d_op && path->dentry->d_op->d_dname)
  2630. return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
  2631. get_fs_root(current->fs, &root);
  2632. br_read_lock(&vfsmount_lock);
  2633. error = path_with_deleted(path, &root, &res, &buflen);
  2634. br_read_unlock(&vfsmount_lock);
  2635. if (error < 0)
  2636. res = ERR_PTR(error);
  2637. path_put(&root);
  2638. return res;
  2639. }
  2640. EXPORT_SYMBOL(d_path);
  2641. /*
  2642. * Helper function for dentry_operations.d_dname() members
  2643. */
  2644. char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
  2645. const char *fmt, ...)
  2646. {
  2647. va_list args;
  2648. char temp[64];
  2649. int sz;
  2650. va_start(args, fmt);
  2651. sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
  2652. va_end(args);
  2653. if (sz > sizeof(temp) || sz > buflen)
  2654. return ERR_PTR(-ENAMETOOLONG);
  2655. buffer += buflen - sz;
  2656. return memcpy(buffer, temp, sz);
  2657. }
  2658. char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
  2659. {
  2660. char *end = buffer + buflen;
  2661. /* these dentries are never renamed, so d_lock is not needed */
  2662. if (prepend(&end, &buflen, " (deleted)", 11) ||
  2663. prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
  2664. prepend(&end, &buflen, "/", 1))
  2665. end = ERR_PTR(-ENAMETOOLONG);
  2666. return end;
  2667. }
  2668. /*
  2669. * Write full pathname from the root of the filesystem into the buffer.
  2670. */
  2671. static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
  2672. {
  2673. char *end, *retval;
  2674. int len, seq = 0;
  2675. int error = 0;
  2676. rcu_read_lock();
  2677. restart:
  2678. end = buf + buflen;
  2679. len = buflen;
  2680. prepend(&end, &len, "\0", 1);
  2681. if (buflen < 1)
  2682. goto Elong;
  2683. /* Get '/' right */
  2684. retval = end-1;
  2685. *retval = '/';
  2686. read_seqbegin_or_lock(&rename_lock, &seq);
  2687. while (!IS_ROOT(dentry)) {
  2688. struct dentry *parent = dentry->d_parent;
  2689. int error;
  2690. prefetch(parent);
  2691. error = prepend_name(&end, &len, &dentry->d_name);
  2692. if (error)
  2693. break;
  2694. retval = end;
  2695. dentry = parent;
  2696. }
  2697. if (!(seq & 1))
  2698. rcu_read_unlock();
  2699. if (need_seqretry(&rename_lock, seq)) {
  2700. seq = 1;
  2701. goto restart;
  2702. }
  2703. done_seqretry(&rename_lock, seq);
  2704. if (error)
  2705. goto Elong;
  2706. return retval;
  2707. Elong:
  2708. return ERR_PTR(-ENAMETOOLONG);
  2709. }
  2710. char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
  2711. {
  2712. return __dentry_path(dentry, buf, buflen);
  2713. }
  2714. EXPORT_SYMBOL(dentry_path_raw);
  2715. char *dentry_path(struct dentry *dentry, char *buf, int buflen)
  2716. {
  2717. char *p = NULL;
  2718. char *retval;
  2719. if (d_unlinked(dentry)) {
  2720. p = buf + buflen;
  2721. if (prepend(&p, &buflen, "//deleted", 10) != 0)
  2722. goto Elong;
  2723. buflen++;
  2724. }
  2725. retval = __dentry_path(dentry, buf, buflen);
  2726. if (!IS_ERR(retval) && p)
  2727. *p = '/'; /* restore '/' overriden with '\0' */
  2728. return retval;
  2729. Elong:
  2730. return ERR_PTR(-ENAMETOOLONG);
  2731. }
  2732. /*
  2733. * NOTE! The user-level library version returns a
  2734. * character pointer. The kernel system call just
  2735. * returns the length of the buffer filled (which
  2736. * includes the ending '\0' character), or a negative
  2737. * error value. So libc would do something like
  2738. *
  2739. * char *getcwd(char * buf, size_t size)
  2740. * {
  2741. * int retval;
  2742. *
  2743. * retval = sys_getcwd(buf, size);
  2744. * if (retval >= 0)
  2745. * return buf;
  2746. * errno = -retval;
  2747. * return NULL;
  2748. * }
  2749. */
  2750. SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
  2751. {
  2752. int error;
  2753. struct path pwd, root;
  2754. char *page = (char *) __get_free_page(GFP_USER);
  2755. if (!page)
  2756. return -ENOMEM;
  2757. get_fs_root_and_pwd(current->fs, &root, &pwd);
  2758. error = -ENOENT;
  2759. br_read_lock(&vfsmount_lock);
  2760. if (!d_unlinked(pwd.dentry)) {
  2761. unsigned long len;
  2762. char *cwd = page + PAGE_SIZE;
  2763. int buflen = PAGE_SIZE;
  2764. prepend(&cwd, &buflen, "\0", 1);
  2765. error = prepend_path(&pwd, &root, &cwd, &buflen);
  2766. br_read_unlock(&vfsmount_lock);
  2767. if (error < 0)
  2768. goto out;
  2769. /* Unreachable from current root */
  2770. if (error > 0) {
  2771. error = prepend_unreachable(&cwd, &buflen);
  2772. if (error)
  2773. goto out;
  2774. }
  2775. error = -ERANGE;
  2776. len = PAGE_SIZE + page - cwd;
  2777. if (len <= size) {
  2778. error = len;
  2779. if (copy_to_user(buf, cwd, len))
  2780. error = -EFAULT;
  2781. }
  2782. } else {
  2783. br_read_unlock(&vfsmount_lock);
  2784. }
  2785. out:
  2786. path_put(&pwd);
  2787. path_put(&root);
  2788. free_page((unsigned long) page);
  2789. return error;
  2790. }
  2791. /*
  2792. * Test whether new_dentry is a subdirectory of old_dentry.
  2793. *
  2794. * Trivially implemented using the dcache structure
  2795. */
  2796. /**
  2797. * is_subdir - is new dentry a subdirectory of old_dentry
  2798. * @new_dentry: new dentry
  2799. * @old_dentry: old dentry
  2800. *
  2801. * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
  2802. * Returns 0 otherwise.
  2803. * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
  2804. */
  2805. int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
  2806. {
  2807. int result;
  2808. unsigned seq;
  2809. if (new_dentry == old_dentry)
  2810. return 1;
  2811. do {
  2812. /* for restarting inner loop in case of seq retry */
  2813. seq = read_seqbegin(&rename_lock);
  2814. /*
  2815. * Need rcu_readlock to protect against the d_parent trashing
  2816. * due to d_move
  2817. */
  2818. rcu_read_lock();
  2819. if (d_ancestor(old_dentry, new_dentry))
  2820. result = 1;
  2821. else
  2822. result = 0;
  2823. rcu_read_unlock();
  2824. } while (read_seqretry(&rename_lock, seq));
  2825. return result;
  2826. }
  2827. static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
  2828. {
  2829. struct dentry *root = data;
  2830. if (dentry != root) {
  2831. if (d_unhashed(dentry) || !dentry->d_inode)
  2832. return D_WALK_SKIP;
  2833. if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
  2834. dentry->d_flags |= DCACHE_GENOCIDE;
  2835. dentry->d_lockref.count--;
  2836. }
  2837. }
  2838. return D_WALK_CONTINUE;
  2839. }
  2840. void d_genocide(struct dentry *parent)
  2841. {
  2842. d_walk(parent, parent, d_genocide_kill, NULL);
  2843. }
  2844. void d_tmpfile(struct dentry *dentry, struct inode *inode)
  2845. {
  2846. inode_dec_link_count(inode);
  2847. BUG_ON(dentry->d_name.name != dentry->d_iname ||
  2848. !hlist_unhashed(&dentry->d_alias) ||
  2849. !d_unlinked(dentry));
  2850. spin_lock(&dentry->d_parent->d_lock);
  2851. spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
  2852. dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
  2853. (unsigned long long)inode->i_ino);
  2854. spin_unlock(&dentry->d_lock);
  2855. spin_unlock(&dentry->d_parent->d_lock);
  2856. d_instantiate(dentry, inode);
  2857. }
  2858. EXPORT_SYMBOL(d_tmpfile);
  2859. static __initdata unsigned long dhash_entries;
  2860. static int __init set_dhash_entries(char *str)
  2861. {
  2862. if (!str)
  2863. return 0;
  2864. dhash_entries = simple_strtoul(str, &str, 0);
  2865. return 1;
  2866. }
  2867. __setup("dhash_entries=", set_dhash_entries);
  2868. static void __init dcache_init_early(void)
  2869. {
  2870. unsigned int loop;
  2871. /* If hashes are distributed across NUMA nodes, defer
  2872. * hash allocation until vmalloc space is available.
  2873. */
  2874. if (hashdist)
  2875. return;
  2876. dentry_hashtable =
  2877. alloc_large_system_hash("Dentry cache",
  2878. sizeof(struct hlist_bl_head),
  2879. dhash_entries,
  2880. 13,
  2881. HASH_EARLY,
  2882. &d_hash_shift,
  2883. &d_hash_mask,
  2884. 0,
  2885. 0);
  2886. for (loop = 0; loop < (1U << d_hash_shift); loop++)
  2887. INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
  2888. }
  2889. static void __init dcache_init(void)
  2890. {
  2891. unsigned int loop;
  2892. /*
  2893. * A constructor could be added for stable state like the lists,
  2894. * but it is probably not worth it because of the cache nature
  2895. * of the dcache.
  2896. */
  2897. dentry_cache = KMEM_CACHE(dentry,
  2898. SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
  2899. /* Hash may have been set up in dcache_init_early */
  2900. if (!hashdist)
  2901. return;
  2902. dentry_hashtable =
  2903. alloc_large_system_hash("Dentry cache",
  2904. sizeof(struct hlist_bl_head),
  2905. dhash_entries,
  2906. 13,
  2907. 0,
  2908. &d_hash_shift,
  2909. &d_hash_mask,
  2910. 0,
  2911. 0);
  2912. for (loop = 0; loop < (1U << d_hash_shift); loop++)
  2913. INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
  2914. }
  2915. /* SLAB cache for __getname() consumers */
  2916. struct kmem_cache *names_cachep __read_mostly;
  2917. EXPORT_SYMBOL(names_cachep);
  2918. EXPORT_SYMBOL(d_genocide);
  2919. void __init vfs_caches_init_early(void)
  2920. {
  2921. dcache_init_early();
  2922. inode_init_early();
  2923. }
  2924. void __init vfs_caches_init(unsigned long mempages)
  2925. {
  2926. unsigned long reserve;
  2927. /* Base hash sizes on available memory, with a reserve equal to
  2928. 150% of current kernel size */
  2929. reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
  2930. mempages -= reserve;
  2931. names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
  2932. SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
  2933. dcache_init();
  2934. inode_init();
  2935. files_init(mempages);
  2936. mnt_init();
  2937. bdev_cache_init();
  2938. chrdev_init();
  2939. }