dcache.c 44 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771
  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/config.h>
  16. #include <linux/syscalls.h>
  17. #include <linux/string.h>
  18. #include <linux/mm.h>
  19. #include <linux/fs.h>
  20. #include <linux/fsnotify.h>
  21. #include <linux/slab.h>
  22. #include <linux/init.h>
  23. #include <linux/smp_lock.h>
  24. #include <linux/hash.h>
  25. #include <linux/cache.h>
  26. #include <linux/module.h>
  27. #include <linux/mount.h>
  28. #include <linux/file.h>
  29. #include <asm/uaccess.h>
  30. #include <linux/security.h>
  31. #include <linux/seqlock.h>
  32. #include <linux/swap.h>
  33. #include <linux/bootmem.h>
  34. /* #define DCACHE_DEBUG 1 */
  35. int sysctl_vfs_cache_pressure = 100;
  36. EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
  37. __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
  38. static seqlock_t rename_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
  39. EXPORT_SYMBOL(dcache_lock);
  40. static kmem_cache_t *dentry_cache;
  41. #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
  42. /*
  43. * This is the single most critical data structure when it comes
  44. * to the dcache: the hashtable for lookups. Somebody should try
  45. * to make this good - I've just made it work.
  46. *
  47. * This hash-function tries to avoid losing too many bits of hash
  48. * information, yet avoid using a prime hash-size or similar.
  49. */
  50. #define D_HASHBITS d_hash_shift
  51. #define D_HASHMASK d_hash_mask
  52. static unsigned int d_hash_mask;
  53. static unsigned int d_hash_shift;
  54. static struct hlist_head *dentry_hashtable;
  55. static LIST_HEAD(dentry_unused);
  56. /* Statistics gathering. */
  57. struct dentry_stat_t dentry_stat = {
  58. .age_limit = 45,
  59. };
  60. static void d_callback(struct rcu_head *head)
  61. {
  62. struct dentry * dentry = container_of(head, struct dentry, d_rcu);
  63. if (dname_external(dentry))
  64. kfree(dentry->d_name.name);
  65. kmem_cache_free(dentry_cache, dentry);
  66. }
  67. /*
  68. * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
  69. * inside dcache_lock.
  70. */
  71. static void d_free(struct dentry *dentry)
  72. {
  73. if (dentry->d_op && dentry->d_op->d_release)
  74. dentry->d_op->d_release(dentry);
  75. call_rcu(&dentry->d_rcu, d_callback);
  76. }
  77. /*
  78. * Release the dentry's inode, using the filesystem
  79. * d_iput() operation if defined.
  80. * Called with dcache_lock and per dentry lock held, drops both.
  81. */
  82. static inline void dentry_iput(struct dentry * dentry)
  83. {
  84. struct inode *inode = dentry->d_inode;
  85. if (inode) {
  86. dentry->d_inode = NULL;
  87. list_del_init(&dentry->d_alias);
  88. spin_unlock(&dentry->d_lock);
  89. spin_unlock(&dcache_lock);
  90. fsnotify_inoderemove(inode);
  91. if (dentry->d_op && dentry->d_op->d_iput)
  92. dentry->d_op->d_iput(dentry, inode);
  93. else
  94. iput(inode);
  95. } else {
  96. spin_unlock(&dentry->d_lock);
  97. spin_unlock(&dcache_lock);
  98. }
  99. }
  100. /*
  101. * This is dput
  102. *
  103. * This is complicated by the fact that we do not want to put
  104. * dentries that are no longer on any hash chain on the unused
  105. * list: we'd much rather just get rid of them immediately.
  106. *
  107. * However, that implies that we have to traverse the dentry
  108. * tree upwards to the parents which might _also_ now be
  109. * scheduled for deletion (it may have been only waiting for
  110. * its last child to go away).
  111. *
  112. * This tail recursion is done by hand as we don't want to depend
  113. * on the compiler to always get this right (gcc generally doesn't).
  114. * Real recursion would eat up our stack space.
  115. */
  116. /*
  117. * dput - release a dentry
  118. * @dentry: dentry to release
  119. *
  120. * Release a dentry. This will drop the usage count and if appropriate
  121. * call the dentry unlink method as well as removing it from the queues and
  122. * releasing its resources. If the parent dentries were scheduled for release
  123. * they too may now get deleted.
  124. *
  125. * no dcache lock, please.
  126. */
  127. void dput(struct dentry *dentry)
  128. {
  129. if (!dentry)
  130. return;
  131. repeat:
  132. if (atomic_read(&dentry->d_count) == 1)
  133. might_sleep();
  134. if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
  135. return;
  136. spin_lock(&dentry->d_lock);
  137. if (atomic_read(&dentry->d_count)) {
  138. spin_unlock(&dentry->d_lock);
  139. spin_unlock(&dcache_lock);
  140. return;
  141. }
  142. /*
  143. * AV: ->d_delete() is _NOT_ allowed to block now.
  144. */
  145. if (dentry->d_op && dentry->d_op->d_delete) {
  146. if (dentry->d_op->d_delete(dentry))
  147. goto unhash_it;
  148. }
  149. /* Unreachable? Get rid of it */
  150. if (d_unhashed(dentry))
  151. goto kill_it;
  152. if (list_empty(&dentry->d_lru)) {
  153. dentry->d_flags |= DCACHE_REFERENCED;
  154. list_add(&dentry->d_lru, &dentry_unused);
  155. dentry_stat.nr_unused++;
  156. }
  157. spin_unlock(&dentry->d_lock);
  158. spin_unlock(&dcache_lock);
  159. return;
  160. unhash_it:
  161. __d_drop(dentry);
  162. kill_it: {
  163. struct dentry *parent;
  164. /* If dentry was on d_lru list
  165. * delete it from there
  166. */
  167. if (!list_empty(&dentry->d_lru)) {
  168. list_del(&dentry->d_lru);
  169. dentry_stat.nr_unused--;
  170. }
  171. list_del(&dentry->d_child);
  172. dentry_stat.nr_dentry--; /* For d_free, below */
  173. /*drops the locks, at that point nobody can reach this dentry */
  174. dentry_iput(dentry);
  175. parent = dentry->d_parent;
  176. d_free(dentry);
  177. if (dentry == parent)
  178. return;
  179. dentry = parent;
  180. goto repeat;
  181. }
  182. }
  183. /**
  184. * d_invalidate - invalidate a dentry
  185. * @dentry: dentry to invalidate
  186. *
  187. * Try to invalidate the dentry if it turns out to be
  188. * possible. If there are other dentries that can be
  189. * reached through this one we can't delete it and we
  190. * return -EBUSY. On success we return 0.
  191. *
  192. * no dcache lock.
  193. */
  194. int d_invalidate(struct dentry * dentry)
  195. {
  196. /*
  197. * If it's already been dropped, return OK.
  198. */
  199. spin_lock(&dcache_lock);
  200. if (d_unhashed(dentry)) {
  201. spin_unlock(&dcache_lock);
  202. return 0;
  203. }
  204. /*
  205. * Check whether to do a partial shrink_dcache
  206. * to get rid of unused child entries.
  207. */
  208. if (!list_empty(&dentry->d_subdirs)) {
  209. spin_unlock(&dcache_lock);
  210. shrink_dcache_parent(dentry);
  211. spin_lock(&dcache_lock);
  212. }
  213. /*
  214. * Somebody else still using it?
  215. *
  216. * If it's a directory, we can't drop it
  217. * for fear of somebody re-populating it
  218. * with children (even though dropping it
  219. * would make it unreachable from the root,
  220. * we might still populate it if it was a
  221. * working directory or similar).
  222. */
  223. spin_lock(&dentry->d_lock);
  224. if (atomic_read(&dentry->d_count) > 1) {
  225. if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
  226. spin_unlock(&dentry->d_lock);
  227. spin_unlock(&dcache_lock);
  228. return -EBUSY;
  229. }
  230. }
  231. __d_drop(dentry);
  232. spin_unlock(&dentry->d_lock);
  233. spin_unlock(&dcache_lock);
  234. return 0;
  235. }
  236. /* This should be called _only_ with dcache_lock held */
  237. static inline struct dentry * __dget_locked(struct dentry *dentry)
  238. {
  239. atomic_inc(&dentry->d_count);
  240. if (!list_empty(&dentry->d_lru)) {
  241. dentry_stat.nr_unused--;
  242. list_del_init(&dentry->d_lru);
  243. }
  244. return dentry;
  245. }
  246. struct dentry * dget_locked(struct dentry *dentry)
  247. {
  248. return __dget_locked(dentry);
  249. }
  250. /**
  251. * d_find_alias - grab a hashed alias of inode
  252. * @inode: inode in question
  253. * @want_discon: flag, used by d_splice_alias, to request
  254. * that only a DISCONNECTED alias be returned.
  255. *
  256. * If inode has a hashed alias, or is a directory and has any alias,
  257. * acquire the reference to alias and return it. Otherwise return NULL.
  258. * Notice that if inode is a directory there can be only one alias and
  259. * it can be unhashed only if it has no children, or if it is the root
  260. * of a filesystem.
  261. *
  262. * If the inode has a DCACHE_DISCONNECTED alias, then prefer
  263. * any other hashed alias over that one unless @want_discon is set,
  264. * in which case only return a DCACHE_DISCONNECTED alias.
  265. */
  266. static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
  267. {
  268. struct list_head *head, *next, *tmp;
  269. struct dentry *alias, *discon_alias=NULL;
  270. head = &inode->i_dentry;
  271. next = inode->i_dentry.next;
  272. while (next != head) {
  273. tmp = next;
  274. next = tmp->next;
  275. prefetch(next);
  276. alias = list_entry(tmp, struct dentry, d_alias);
  277. if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
  278. if (alias->d_flags & DCACHE_DISCONNECTED)
  279. discon_alias = alias;
  280. else if (!want_discon) {
  281. __dget_locked(alias);
  282. return alias;
  283. }
  284. }
  285. }
  286. if (discon_alias)
  287. __dget_locked(discon_alias);
  288. return discon_alias;
  289. }
  290. struct dentry * d_find_alias(struct inode *inode)
  291. {
  292. struct dentry *de;
  293. spin_lock(&dcache_lock);
  294. de = __d_find_alias(inode, 0);
  295. spin_unlock(&dcache_lock);
  296. return de;
  297. }
  298. /*
  299. * Try to kill dentries associated with this inode.
  300. * WARNING: you must own a reference to inode.
  301. */
  302. void d_prune_aliases(struct inode *inode)
  303. {
  304. struct list_head *tmp, *head = &inode->i_dentry;
  305. restart:
  306. spin_lock(&dcache_lock);
  307. tmp = head;
  308. while ((tmp = tmp->next) != head) {
  309. struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
  310. spin_lock(&dentry->d_lock);
  311. if (!atomic_read(&dentry->d_count)) {
  312. __dget_locked(dentry);
  313. __d_drop(dentry);
  314. spin_unlock(&dentry->d_lock);
  315. spin_unlock(&dcache_lock);
  316. dput(dentry);
  317. goto restart;
  318. }
  319. spin_unlock(&dentry->d_lock);
  320. }
  321. spin_unlock(&dcache_lock);
  322. }
  323. /*
  324. * Throw away a dentry - free the inode, dput the parent.
  325. * This requires that the LRU list has already been
  326. * removed.
  327. * Called with dcache_lock, drops it and then regains.
  328. */
  329. static inline void prune_one_dentry(struct dentry * dentry)
  330. {
  331. struct dentry * parent;
  332. __d_drop(dentry);
  333. list_del(&dentry->d_child);
  334. dentry_stat.nr_dentry--; /* For d_free, below */
  335. dentry_iput(dentry);
  336. parent = dentry->d_parent;
  337. d_free(dentry);
  338. if (parent != dentry)
  339. dput(parent);
  340. spin_lock(&dcache_lock);
  341. }
  342. /**
  343. * prune_dcache - shrink the dcache
  344. * @count: number of entries to try and free
  345. *
  346. * Shrink the dcache. This is done when we need
  347. * more memory, or simply when we need to unmount
  348. * something (at which point we need to unuse
  349. * all dentries).
  350. *
  351. * This function may fail to free any resources if
  352. * all the dentries are in use.
  353. */
  354. static void prune_dcache(int count)
  355. {
  356. spin_lock(&dcache_lock);
  357. for (; count ; count--) {
  358. struct dentry *dentry;
  359. struct list_head *tmp;
  360. cond_resched_lock(&dcache_lock);
  361. tmp = dentry_unused.prev;
  362. if (tmp == &dentry_unused)
  363. break;
  364. list_del_init(tmp);
  365. prefetch(dentry_unused.prev);
  366. dentry_stat.nr_unused--;
  367. dentry = list_entry(tmp, struct dentry, d_lru);
  368. spin_lock(&dentry->d_lock);
  369. /*
  370. * We found an inuse dentry which was not removed from
  371. * dentry_unused because of laziness during lookup. Do not free
  372. * it - just keep it off the dentry_unused list.
  373. */
  374. if (atomic_read(&dentry->d_count)) {
  375. spin_unlock(&dentry->d_lock);
  376. continue;
  377. }
  378. /* If the dentry was recently referenced, don't free it. */
  379. if (dentry->d_flags & DCACHE_REFERENCED) {
  380. dentry->d_flags &= ~DCACHE_REFERENCED;
  381. list_add(&dentry->d_lru, &dentry_unused);
  382. dentry_stat.nr_unused++;
  383. spin_unlock(&dentry->d_lock);
  384. continue;
  385. }
  386. prune_one_dentry(dentry);
  387. }
  388. spin_unlock(&dcache_lock);
  389. }
  390. /*
  391. * Shrink the dcache for the specified super block.
  392. * This allows us to unmount a device without disturbing
  393. * the dcache for the other devices.
  394. *
  395. * This implementation makes just two traversals of the
  396. * unused list. On the first pass we move the selected
  397. * dentries to the most recent end, and on the second
  398. * pass we free them. The second pass must restart after
  399. * each dput(), but since the target dentries are all at
  400. * the end, it's really just a single traversal.
  401. */
  402. /**
  403. * shrink_dcache_sb - shrink dcache for a superblock
  404. * @sb: superblock
  405. *
  406. * Shrink the dcache for the specified super block. This
  407. * is used to free the dcache before unmounting a file
  408. * system
  409. */
  410. void shrink_dcache_sb(struct super_block * sb)
  411. {
  412. struct list_head *tmp, *next;
  413. struct dentry *dentry;
  414. /*
  415. * Pass one ... move the dentries for the specified
  416. * superblock to the most recent end of the unused list.
  417. */
  418. spin_lock(&dcache_lock);
  419. next = dentry_unused.next;
  420. while (next != &dentry_unused) {
  421. tmp = next;
  422. next = tmp->next;
  423. dentry = list_entry(tmp, struct dentry, d_lru);
  424. if (dentry->d_sb != sb)
  425. continue;
  426. list_del(tmp);
  427. list_add(tmp, &dentry_unused);
  428. }
  429. /*
  430. * Pass two ... free the dentries for this superblock.
  431. */
  432. repeat:
  433. next = dentry_unused.next;
  434. while (next != &dentry_unused) {
  435. tmp = next;
  436. next = tmp->next;
  437. dentry = list_entry(tmp, struct dentry, d_lru);
  438. if (dentry->d_sb != sb)
  439. continue;
  440. dentry_stat.nr_unused--;
  441. list_del_init(tmp);
  442. spin_lock(&dentry->d_lock);
  443. if (atomic_read(&dentry->d_count)) {
  444. spin_unlock(&dentry->d_lock);
  445. continue;
  446. }
  447. prune_one_dentry(dentry);
  448. goto repeat;
  449. }
  450. spin_unlock(&dcache_lock);
  451. }
  452. /*
  453. * Search for at least 1 mount point in the dentry's subdirs.
  454. * We descend to the next level whenever the d_subdirs
  455. * list is non-empty and continue searching.
  456. */
  457. /**
  458. * have_submounts - check for mounts over a dentry
  459. * @parent: dentry to check.
  460. *
  461. * Return true if the parent or its subdirectories contain
  462. * a mount point
  463. */
  464. int have_submounts(struct dentry *parent)
  465. {
  466. struct dentry *this_parent = parent;
  467. struct list_head *next;
  468. spin_lock(&dcache_lock);
  469. if (d_mountpoint(parent))
  470. goto positive;
  471. repeat:
  472. next = this_parent->d_subdirs.next;
  473. resume:
  474. while (next != &this_parent->d_subdirs) {
  475. struct list_head *tmp = next;
  476. struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
  477. next = tmp->next;
  478. /* Have we found a mount point ? */
  479. if (d_mountpoint(dentry))
  480. goto positive;
  481. if (!list_empty(&dentry->d_subdirs)) {
  482. this_parent = dentry;
  483. goto repeat;
  484. }
  485. }
  486. /*
  487. * All done at this level ... ascend and resume the search.
  488. */
  489. if (this_parent != parent) {
  490. next = this_parent->d_child.next;
  491. this_parent = this_parent->d_parent;
  492. goto resume;
  493. }
  494. spin_unlock(&dcache_lock);
  495. return 0; /* No mount points found in tree */
  496. positive:
  497. spin_unlock(&dcache_lock);
  498. return 1;
  499. }
  500. /*
  501. * Search the dentry child list for the specified parent,
  502. * and move any unused dentries to the end of the unused
  503. * list for prune_dcache(). We descend to the next level
  504. * whenever the d_subdirs list is non-empty and continue
  505. * searching.
  506. *
  507. * It returns zero iff there are no unused children,
  508. * otherwise it returns the number of children moved to
  509. * the end of the unused list. This may not be the total
  510. * number of unused children, because select_parent can
  511. * drop the lock and return early due to latency
  512. * constraints.
  513. */
  514. static int select_parent(struct dentry * parent)
  515. {
  516. struct dentry *this_parent = parent;
  517. struct list_head *next;
  518. int found = 0;
  519. spin_lock(&dcache_lock);
  520. repeat:
  521. next = this_parent->d_subdirs.next;
  522. resume:
  523. while (next != &this_parent->d_subdirs) {
  524. struct list_head *tmp = next;
  525. struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
  526. next = tmp->next;
  527. if (!list_empty(&dentry->d_lru)) {
  528. dentry_stat.nr_unused--;
  529. list_del_init(&dentry->d_lru);
  530. }
  531. /*
  532. * move only zero ref count dentries to the end
  533. * of the unused list for prune_dcache
  534. */
  535. if (!atomic_read(&dentry->d_count)) {
  536. list_add(&dentry->d_lru, dentry_unused.prev);
  537. dentry_stat.nr_unused++;
  538. found++;
  539. }
  540. /*
  541. * We can return to the caller if we have found some (this
  542. * ensures forward progress). We'll be coming back to find
  543. * the rest.
  544. */
  545. if (found && need_resched())
  546. goto out;
  547. /*
  548. * Descend a level if the d_subdirs list is non-empty.
  549. */
  550. if (!list_empty(&dentry->d_subdirs)) {
  551. this_parent = dentry;
  552. #ifdef DCACHE_DEBUG
  553. printk(KERN_DEBUG "select_parent: descending to %s/%s, found=%d\n",
  554. dentry->d_parent->d_name.name, dentry->d_name.name, found);
  555. #endif
  556. goto repeat;
  557. }
  558. }
  559. /*
  560. * All done at this level ... ascend and resume the search.
  561. */
  562. if (this_parent != parent) {
  563. next = this_parent->d_child.next;
  564. this_parent = this_parent->d_parent;
  565. #ifdef DCACHE_DEBUG
  566. printk(KERN_DEBUG "select_parent: ascending to %s/%s, found=%d\n",
  567. this_parent->d_parent->d_name.name, this_parent->d_name.name, found);
  568. #endif
  569. goto resume;
  570. }
  571. out:
  572. spin_unlock(&dcache_lock);
  573. return found;
  574. }
  575. /**
  576. * shrink_dcache_parent - prune dcache
  577. * @parent: parent of entries to prune
  578. *
  579. * Prune the dcache to remove unused children of the parent dentry.
  580. */
  581. void shrink_dcache_parent(struct dentry * parent)
  582. {
  583. int found;
  584. while ((found = select_parent(parent)) != 0)
  585. prune_dcache(found);
  586. }
  587. /**
  588. * shrink_dcache_anon - further prune the cache
  589. * @head: head of d_hash list of dentries to prune
  590. *
  591. * Prune the dentries that are anonymous
  592. *
  593. * parsing d_hash list does not hlist_for_each_rcu() as it
  594. * done under dcache_lock.
  595. *
  596. */
  597. void shrink_dcache_anon(struct hlist_head *head)
  598. {
  599. struct hlist_node *lp;
  600. int found;
  601. do {
  602. found = 0;
  603. spin_lock(&dcache_lock);
  604. hlist_for_each(lp, head) {
  605. struct dentry *this = hlist_entry(lp, struct dentry, d_hash);
  606. if (!list_empty(&this->d_lru)) {
  607. dentry_stat.nr_unused--;
  608. list_del_init(&this->d_lru);
  609. }
  610. /*
  611. * move only zero ref count dentries to the end
  612. * of the unused list for prune_dcache
  613. */
  614. if (!atomic_read(&this->d_count)) {
  615. list_add_tail(&this->d_lru, &dentry_unused);
  616. dentry_stat.nr_unused++;
  617. found++;
  618. }
  619. }
  620. spin_unlock(&dcache_lock);
  621. prune_dcache(found);
  622. } while(found);
  623. }
  624. /*
  625. * Scan `nr' dentries and return the number which remain.
  626. *
  627. * We need to avoid reentering the filesystem if the caller is performing a
  628. * GFP_NOFS allocation attempt. One example deadlock is:
  629. *
  630. * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
  631. * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
  632. * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
  633. *
  634. * In this case we return -1 to tell the caller that we baled.
  635. */
  636. static int shrink_dcache_memory(int nr, unsigned int gfp_mask)
  637. {
  638. if (nr) {
  639. if (!(gfp_mask & __GFP_FS))
  640. return -1;
  641. prune_dcache(nr);
  642. }
  643. return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
  644. }
  645. /**
  646. * d_alloc - allocate a dcache entry
  647. * @parent: parent of entry to allocate
  648. * @name: qstr of the name
  649. *
  650. * Allocates a dentry. It returns %NULL if there is insufficient memory
  651. * available. On a success the dentry is returned. The name passed in is
  652. * copied and the copy passed in may be reused after this call.
  653. */
  654. struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
  655. {
  656. struct dentry *dentry;
  657. char *dname;
  658. dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
  659. if (!dentry)
  660. return NULL;
  661. if (name->len > DNAME_INLINE_LEN-1) {
  662. dname = kmalloc(name->len + 1, GFP_KERNEL);
  663. if (!dname) {
  664. kmem_cache_free(dentry_cache, dentry);
  665. return NULL;
  666. }
  667. } else {
  668. dname = dentry->d_iname;
  669. }
  670. dentry->d_name.name = dname;
  671. dentry->d_name.len = name->len;
  672. dentry->d_name.hash = name->hash;
  673. memcpy(dname, name->name, name->len);
  674. dname[name->len] = 0;
  675. atomic_set(&dentry->d_count, 1);
  676. dentry->d_flags = DCACHE_UNHASHED;
  677. spin_lock_init(&dentry->d_lock);
  678. dentry->d_inode = NULL;
  679. dentry->d_parent = NULL;
  680. dentry->d_sb = NULL;
  681. dentry->d_op = NULL;
  682. dentry->d_fsdata = NULL;
  683. dentry->d_mounted = 0;
  684. dentry->d_cookie = NULL;
  685. INIT_HLIST_NODE(&dentry->d_hash);
  686. INIT_LIST_HEAD(&dentry->d_lru);
  687. INIT_LIST_HEAD(&dentry->d_subdirs);
  688. INIT_LIST_HEAD(&dentry->d_alias);
  689. if (parent) {
  690. dentry->d_parent = dget(parent);
  691. dentry->d_sb = parent->d_sb;
  692. } else {
  693. INIT_LIST_HEAD(&dentry->d_child);
  694. }
  695. spin_lock(&dcache_lock);
  696. if (parent)
  697. list_add(&dentry->d_child, &parent->d_subdirs);
  698. dentry_stat.nr_dentry++;
  699. spin_unlock(&dcache_lock);
  700. return dentry;
  701. }
  702. struct dentry *d_alloc_name(struct dentry *parent, const char *name)
  703. {
  704. struct qstr q;
  705. q.name = name;
  706. q.len = strlen(name);
  707. q.hash = full_name_hash(q.name, q.len);
  708. return d_alloc(parent, &q);
  709. }
  710. /**
  711. * d_instantiate - fill in inode information for a dentry
  712. * @entry: dentry to complete
  713. * @inode: inode to attach to this dentry
  714. *
  715. * Fill in inode information in the entry.
  716. *
  717. * This turns negative dentries into productive full members
  718. * of society.
  719. *
  720. * NOTE! This assumes that the inode count has been incremented
  721. * (or otherwise set) by the caller to indicate that it is now
  722. * in use by the dcache.
  723. */
  724. void d_instantiate(struct dentry *entry, struct inode * inode)
  725. {
  726. if (!list_empty(&entry->d_alias)) BUG();
  727. spin_lock(&dcache_lock);
  728. if (inode)
  729. list_add(&entry->d_alias, &inode->i_dentry);
  730. entry->d_inode = inode;
  731. spin_unlock(&dcache_lock);
  732. security_d_instantiate(entry, inode);
  733. }
  734. /**
  735. * d_instantiate_unique - instantiate a non-aliased dentry
  736. * @entry: dentry to instantiate
  737. * @inode: inode to attach to this dentry
  738. *
  739. * Fill in inode information in the entry. On success, it returns NULL.
  740. * If an unhashed alias of "entry" already exists, then we return the
  741. * aliased dentry instead.
  742. *
  743. * Note that in order to avoid conflicts with rename() etc, the caller
  744. * had better be holding the parent directory semaphore.
  745. */
  746. struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
  747. {
  748. struct dentry *alias;
  749. int len = entry->d_name.len;
  750. const char *name = entry->d_name.name;
  751. unsigned int hash = entry->d_name.hash;
  752. BUG_ON(!list_empty(&entry->d_alias));
  753. spin_lock(&dcache_lock);
  754. if (!inode)
  755. goto do_negative;
  756. list_for_each_entry(alias, &inode->i_dentry, d_alias) {
  757. struct qstr *qstr = &alias->d_name;
  758. if (qstr->hash != hash)
  759. continue;
  760. if (alias->d_parent != entry->d_parent)
  761. continue;
  762. if (qstr->len != len)
  763. continue;
  764. if (memcmp(qstr->name, name, len))
  765. continue;
  766. dget_locked(alias);
  767. spin_unlock(&dcache_lock);
  768. BUG_ON(!d_unhashed(alias));
  769. return alias;
  770. }
  771. list_add(&entry->d_alias, &inode->i_dentry);
  772. do_negative:
  773. entry->d_inode = inode;
  774. spin_unlock(&dcache_lock);
  775. security_d_instantiate(entry, inode);
  776. return NULL;
  777. }
  778. EXPORT_SYMBOL(d_instantiate_unique);
  779. /**
  780. * d_alloc_root - allocate root dentry
  781. * @root_inode: inode to allocate the root for
  782. *
  783. * Allocate a root ("/") dentry for the inode given. The inode is
  784. * instantiated and returned. %NULL is returned if there is insufficient
  785. * memory or the inode passed is %NULL.
  786. */
  787. struct dentry * d_alloc_root(struct inode * root_inode)
  788. {
  789. struct dentry *res = NULL;
  790. if (root_inode) {
  791. static const struct qstr name = { .name = "/", .len = 1 };
  792. res = d_alloc(NULL, &name);
  793. if (res) {
  794. res->d_sb = root_inode->i_sb;
  795. res->d_parent = res;
  796. d_instantiate(res, root_inode);
  797. }
  798. }
  799. return res;
  800. }
  801. static inline struct hlist_head *d_hash(struct dentry *parent,
  802. unsigned long hash)
  803. {
  804. hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
  805. hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
  806. return dentry_hashtable + (hash & D_HASHMASK);
  807. }
  808. /**
  809. * d_alloc_anon - allocate an anonymous dentry
  810. * @inode: inode to allocate the dentry for
  811. *
  812. * This is similar to d_alloc_root. It is used by filesystems when
  813. * creating a dentry for a given inode, often in the process of
  814. * mapping a filehandle to a dentry. The returned dentry may be
  815. * anonymous, or may have a full name (if the inode was already
  816. * in the cache). The file system may need to make further
  817. * efforts to connect this dentry into the dcache properly.
  818. *
  819. * When called on a directory inode, we must ensure that
  820. * the inode only ever has one dentry. If a dentry is
  821. * found, that is returned instead of allocating a new one.
  822. *
  823. * On successful return, the reference to the inode has been transferred
  824. * to the dentry. If %NULL is returned (indicating kmalloc failure),
  825. * the reference on the inode has not been released.
  826. */
  827. struct dentry * d_alloc_anon(struct inode *inode)
  828. {
  829. static const struct qstr anonstring = { .name = "" };
  830. struct dentry *tmp;
  831. struct dentry *res;
  832. if ((res = d_find_alias(inode))) {
  833. iput(inode);
  834. return res;
  835. }
  836. tmp = d_alloc(NULL, &anonstring);
  837. if (!tmp)
  838. return NULL;
  839. tmp->d_parent = tmp; /* make sure dput doesn't croak */
  840. spin_lock(&dcache_lock);
  841. res = __d_find_alias(inode, 0);
  842. if (!res) {
  843. /* attach a disconnected dentry */
  844. res = tmp;
  845. tmp = NULL;
  846. spin_lock(&res->d_lock);
  847. res->d_sb = inode->i_sb;
  848. res->d_parent = res;
  849. res->d_inode = inode;
  850. res->d_flags |= DCACHE_DISCONNECTED;
  851. res->d_flags &= ~DCACHE_UNHASHED;
  852. list_add(&res->d_alias, &inode->i_dentry);
  853. hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
  854. spin_unlock(&res->d_lock);
  855. inode = NULL; /* don't drop reference */
  856. }
  857. spin_unlock(&dcache_lock);
  858. if (inode)
  859. iput(inode);
  860. if (tmp)
  861. dput(tmp);
  862. return res;
  863. }
  864. /**
  865. * d_splice_alias - splice a disconnected dentry into the tree if one exists
  866. * @inode: the inode which may have a disconnected dentry
  867. * @dentry: a negative dentry which we want to point to the inode.
  868. *
  869. * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
  870. * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
  871. * and return it, else simply d_add the inode to the dentry and return NULL.
  872. *
  873. * This is needed in the lookup routine of any filesystem that is exportable
  874. * (via knfsd) so that we can build dcache paths to directories effectively.
  875. *
  876. * If a dentry was found and moved, then it is returned. Otherwise NULL
  877. * is returned. This matches the expected return value of ->lookup.
  878. *
  879. */
  880. struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
  881. {
  882. struct dentry *new = NULL;
  883. if (inode) {
  884. spin_lock(&dcache_lock);
  885. new = __d_find_alias(inode, 1);
  886. if (new) {
  887. BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
  888. spin_unlock(&dcache_lock);
  889. security_d_instantiate(new, inode);
  890. d_rehash(dentry);
  891. d_move(new, dentry);
  892. iput(inode);
  893. } else {
  894. /* d_instantiate takes dcache_lock, so we do it by hand */
  895. list_add(&dentry->d_alias, &inode->i_dentry);
  896. dentry->d_inode = inode;
  897. spin_unlock(&dcache_lock);
  898. security_d_instantiate(dentry, inode);
  899. d_rehash(dentry);
  900. }
  901. } else
  902. d_add(dentry, inode);
  903. return new;
  904. }
  905. /**
  906. * d_lookup - search for a dentry
  907. * @parent: parent dentry
  908. * @name: qstr of name we wish to find
  909. *
  910. * Searches the children of the parent dentry for the name in question. If
  911. * the dentry is found its reference count is incremented and the dentry
  912. * is returned. The caller must use d_put to free the entry when it has
  913. * finished using it. %NULL is returned on failure.
  914. *
  915. * __d_lookup is dcache_lock free. The hash list is protected using RCU.
  916. * Memory barriers are used while updating and doing lockless traversal.
  917. * To avoid races with d_move while rename is happening, d_lock is used.
  918. *
  919. * Overflows in memcmp(), while d_move, are avoided by keeping the length
  920. * and name pointer in one structure pointed by d_qstr.
  921. *
  922. * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
  923. * lookup is going on.
  924. *
  925. * dentry_unused list is not updated even if lookup finds the required dentry
  926. * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
  927. * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
  928. * acquisition.
  929. *
  930. * d_lookup() is protected against the concurrent renames in some unrelated
  931. * directory using the seqlockt_t rename_lock.
  932. */
  933. struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
  934. {
  935. struct dentry * dentry = NULL;
  936. unsigned long seq;
  937. do {
  938. seq = read_seqbegin(&rename_lock);
  939. dentry = __d_lookup(parent, name);
  940. if (dentry)
  941. break;
  942. } while (read_seqretry(&rename_lock, seq));
  943. return dentry;
  944. }
  945. struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
  946. {
  947. unsigned int len = name->len;
  948. unsigned int hash = name->hash;
  949. const unsigned char *str = name->name;
  950. struct hlist_head *head = d_hash(parent,hash);
  951. struct dentry *found = NULL;
  952. struct hlist_node *node;
  953. rcu_read_lock();
  954. hlist_for_each_rcu(node, head) {
  955. struct dentry *dentry;
  956. struct qstr *qstr;
  957. dentry = hlist_entry(node, struct dentry, d_hash);
  958. if (dentry->d_name.hash != hash)
  959. continue;
  960. if (dentry->d_parent != parent)
  961. continue;
  962. spin_lock(&dentry->d_lock);
  963. /*
  964. * Recheck the dentry after taking the lock - d_move may have
  965. * changed things. Don't bother checking the hash because we're
  966. * about to compare the whole name anyway.
  967. */
  968. if (dentry->d_parent != parent)
  969. goto next;
  970. /*
  971. * It is safe to compare names since d_move() cannot
  972. * change the qstr (protected by d_lock).
  973. */
  974. qstr = &dentry->d_name;
  975. if (parent->d_op && parent->d_op->d_compare) {
  976. if (parent->d_op->d_compare(parent, qstr, name))
  977. goto next;
  978. } else {
  979. if (qstr->len != len)
  980. goto next;
  981. if (memcmp(qstr->name, str, len))
  982. goto next;
  983. }
  984. if (!d_unhashed(dentry)) {
  985. atomic_inc(&dentry->d_count);
  986. found = dentry;
  987. }
  988. spin_unlock(&dentry->d_lock);
  989. break;
  990. next:
  991. spin_unlock(&dentry->d_lock);
  992. }
  993. rcu_read_unlock();
  994. return found;
  995. }
  996. /**
  997. * d_validate - verify dentry provided from insecure source
  998. * @dentry: The dentry alleged to be valid child of @dparent
  999. * @dparent: The parent dentry (known to be valid)
  1000. * @hash: Hash of the dentry
  1001. * @len: Length of the name
  1002. *
  1003. * An insecure source has sent us a dentry, here we verify it and dget() it.
  1004. * This is used by ncpfs in its readdir implementation.
  1005. * Zero is returned in the dentry is invalid.
  1006. */
  1007. int d_validate(struct dentry *dentry, struct dentry *dparent)
  1008. {
  1009. struct hlist_head *base;
  1010. struct hlist_node *lhp;
  1011. /* Check whether the ptr might be valid at all.. */
  1012. if (!kmem_ptr_validate(dentry_cache, dentry))
  1013. goto out;
  1014. if (dentry->d_parent != dparent)
  1015. goto out;
  1016. spin_lock(&dcache_lock);
  1017. base = d_hash(dparent, dentry->d_name.hash);
  1018. hlist_for_each(lhp,base) {
  1019. /* hlist_for_each_rcu() not required for d_hash list
  1020. * as it is parsed under dcache_lock
  1021. */
  1022. if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
  1023. __dget_locked(dentry);
  1024. spin_unlock(&dcache_lock);
  1025. return 1;
  1026. }
  1027. }
  1028. spin_unlock(&dcache_lock);
  1029. out:
  1030. return 0;
  1031. }
  1032. /*
  1033. * When a file is deleted, we have two options:
  1034. * - turn this dentry into a negative dentry
  1035. * - unhash this dentry and free it.
  1036. *
  1037. * Usually, we want to just turn this into
  1038. * a negative dentry, but if anybody else is
  1039. * currently using the dentry or the inode
  1040. * we can't do that and we fall back on removing
  1041. * it from the hash queues and waiting for
  1042. * it to be deleted later when it has no users
  1043. */
  1044. /**
  1045. * d_delete - delete a dentry
  1046. * @dentry: The dentry to delete
  1047. *
  1048. * Turn the dentry into a negative dentry if possible, otherwise
  1049. * remove it from the hash queues so it can be deleted later
  1050. */
  1051. void d_delete(struct dentry * dentry)
  1052. {
  1053. int isdir = 0;
  1054. /*
  1055. * Are we the only user?
  1056. */
  1057. spin_lock(&dcache_lock);
  1058. spin_lock(&dentry->d_lock);
  1059. isdir = S_ISDIR(dentry->d_inode->i_mode);
  1060. if (atomic_read(&dentry->d_count) == 1) {
  1061. dentry_iput(dentry);
  1062. fsnotify_nameremove(dentry, isdir);
  1063. return;
  1064. }
  1065. if (!d_unhashed(dentry))
  1066. __d_drop(dentry);
  1067. spin_unlock(&dentry->d_lock);
  1068. spin_unlock(&dcache_lock);
  1069. fsnotify_nameremove(dentry, isdir);
  1070. }
  1071. static void __d_rehash(struct dentry * entry, struct hlist_head *list)
  1072. {
  1073. entry->d_flags &= ~DCACHE_UNHASHED;
  1074. hlist_add_head_rcu(&entry->d_hash, list);
  1075. }
  1076. /**
  1077. * d_rehash - add an entry back to the hash
  1078. * @entry: dentry to add to the hash
  1079. *
  1080. * Adds a dentry to the hash according to its name.
  1081. */
  1082. void d_rehash(struct dentry * entry)
  1083. {
  1084. struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
  1085. spin_lock(&dcache_lock);
  1086. spin_lock(&entry->d_lock);
  1087. __d_rehash(entry, list);
  1088. spin_unlock(&entry->d_lock);
  1089. spin_unlock(&dcache_lock);
  1090. }
  1091. #define do_switch(x,y) do { \
  1092. __typeof__ (x) __tmp = x; \
  1093. x = y; y = __tmp; } while (0)
  1094. /*
  1095. * When switching names, the actual string doesn't strictly have to
  1096. * be preserved in the target - because we're dropping the target
  1097. * anyway. As such, we can just do a simple memcpy() to copy over
  1098. * the new name before we switch.
  1099. *
  1100. * Note that we have to be a lot more careful about getting the hash
  1101. * switched - we have to switch the hash value properly even if it
  1102. * then no longer matches the actual (corrupted) string of the target.
  1103. * The hash value has to match the hash queue that the dentry is on..
  1104. */
  1105. static void switch_names(struct dentry *dentry, struct dentry *target)
  1106. {
  1107. if (dname_external(target)) {
  1108. if (dname_external(dentry)) {
  1109. /*
  1110. * Both external: swap the pointers
  1111. */
  1112. do_switch(target->d_name.name, dentry->d_name.name);
  1113. } else {
  1114. /*
  1115. * dentry:internal, target:external. Steal target's
  1116. * storage and make target internal.
  1117. */
  1118. dentry->d_name.name = target->d_name.name;
  1119. target->d_name.name = target->d_iname;
  1120. }
  1121. } else {
  1122. if (dname_external(dentry)) {
  1123. /*
  1124. * dentry:external, target:internal. Give dentry's
  1125. * storage to target and make dentry internal
  1126. */
  1127. memcpy(dentry->d_iname, target->d_name.name,
  1128. target->d_name.len + 1);
  1129. target->d_name.name = dentry->d_name.name;
  1130. dentry->d_name.name = dentry->d_iname;
  1131. } else {
  1132. /*
  1133. * Both are internal. Just copy target to dentry
  1134. */
  1135. memcpy(dentry->d_iname, target->d_name.name,
  1136. target->d_name.len + 1);
  1137. }
  1138. }
  1139. }
  1140. /*
  1141. * We cannibalize "target" when moving dentry on top of it,
  1142. * because it's going to be thrown away anyway. We could be more
  1143. * polite about it, though.
  1144. *
  1145. * This forceful removal will result in ugly /proc output if
  1146. * somebody holds a file open that got deleted due to a rename.
  1147. * We could be nicer about the deleted file, and let it show
  1148. * up under the name it got deleted rather than the name that
  1149. * deleted it.
  1150. */
  1151. /**
  1152. * d_move - move a dentry
  1153. * @dentry: entry to move
  1154. * @target: new dentry
  1155. *
  1156. * Update the dcache to reflect the move of a file name. Negative
  1157. * dcache entries should not be moved in this way.
  1158. */
  1159. void d_move(struct dentry * dentry, struct dentry * target)
  1160. {
  1161. struct hlist_head *list;
  1162. if (!dentry->d_inode)
  1163. printk(KERN_WARNING "VFS: moving negative dcache entry\n");
  1164. spin_lock(&dcache_lock);
  1165. write_seqlock(&rename_lock);
  1166. /*
  1167. * XXXX: do we really need to take target->d_lock?
  1168. */
  1169. if (target < dentry) {
  1170. spin_lock(&target->d_lock);
  1171. spin_lock(&dentry->d_lock);
  1172. } else {
  1173. spin_lock(&dentry->d_lock);
  1174. spin_lock(&target->d_lock);
  1175. }
  1176. /* Move the dentry to the target hash queue, if on different bucket */
  1177. if (dentry->d_flags & DCACHE_UNHASHED)
  1178. goto already_unhashed;
  1179. hlist_del_rcu(&dentry->d_hash);
  1180. already_unhashed:
  1181. list = d_hash(target->d_parent, target->d_name.hash);
  1182. __d_rehash(dentry, list);
  1183. /* Unhash the target: dput() will then get rid of it */
  1184. __d_drop(target);
  1185. list_del(&dentry->d_child);
  1186. list_del(&target->d_child);
  1187. /* Switch the names.. */
  1188. switch_names(dentry, target);
  1189. do_switch(dentry->d_name.len, target->d_name.len);
  1190. do_switch(dentry->d_name.hash, target->d_name.hash);
  1191. /* ... and switch the parents */
  1192. if (IS_ROOT(dentry)) {
  1193. dentry->d_parent = target->d_parent;
  1194. target->d_parent = target;
  1195. INIT_LIST_HEAD(&target->d_child);
  1196. } else {
  1197. do_switch(dentry->d_parent, target->d_parent);
  1198. /* And add them back to the (new) parent lists */
  1199. list_add(&target->d_child, &target->d_parent->d_subdirs);
  1200. }
  1201. list_add(&dentry->d_child, &dentry->d_parent->d_subdirs);
  1202. spin_unlock(&target->d_lock);
  1203. spin_unlock(&dentry->d_lock);
  1204. write_sequnlock(&rename_lock);
  1205. spin_unlock(&dcache_lock);
  1206. }
  1207. /**
  1208. * d_path - return the path of a dentry
  1209. * @dentry: dentry to report
  1210. * @vfsmnt: vfsmnt to which the dentry belongs
  1211. * @root: root dentry
  1212. * @rootmnt: vfsmnt to which the root dentry belongs
  1213. * @buffer: buffer to return value in
  1214. * @buflen: buffer length
  1215. *
  1216. * Convert a dentry into an ASCII path name. If the entry has been deleted
  1217. * the string " (deleted)" is appended. Note that this is ambiguous.
  1218. *
  1219. * Returns the buffer or an error code if the path was too long.
  1220. *
  1221. * "buflen" should be positive. Caller holds the dcache_lock.
  1222. */
  1223. static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
  1224. struct dentry *root, struct vfsmount *rootmnt,
  1225. char *buffer, int buflen)
  1226. {
  1227. char * end = buffer+buflen;
  1228. char * retval;
  1229. int namelen;
  1230. *--end = '\0';
  1231. buflen--;
  1232. if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
  1233. buflen -= 10;
  1234. end -= 10;
  1235. if (buflen < 0)
  1236. goto Elong;
  1237. memcpy(end, " (deleted)", 10);
  1238. }
  1239. if (buflen < 1)
  1240. goto Elong;
  1241. /* Get '/' right */
  1242. retval = end-1;
  1243. *retval = '/';
  1244. for (;;) {
  1245. struct dentry * parent;
  1246. if (dentry == root && vfsmnt == rootmnt)
  1247. break;
  1248. if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
  1249. /* Global root? */
  1250. spin_lock(&vfsmount_lock);
  1251. if (vfsmnt->mnt_parent == vfsmnt) {
  1252. spin_unlock(&vfsmount_lock);
  1253. goto global_root;
  1254. }
  1255. dentry = vfsmnt->mnt_mountpoint;
  1256. vfsmnt = vfsmnt->mnt_parent;
  1257. spin_unlock(&vfsmount_lock);
  1258. continue;
  1259. }
  1260. parent = dentry->d_parent;
  1261. prefetch(parent);
  1262. namelen = dentry->d_name.len;
  1263. buflen -= namelen + 1;
  1264. if (buflen < 0)
  1265. goto Elong;
  1266. end -= namelen;
  1267. memcpy(end, dentry->d_name.name, namelen);
  1268. *--end = '/';
  1269. retval = end;
  1270. dentry = parent;
  1271. }
  1272. return retval;
  1273. global_root:
  1274. namelen = dentry->d_name.len;
  1275. buflen -= namelen;
  1276. if (buflen < 0)
  1277. goto Elong;
  1278. retval -= namelen-1; /* hit the slash */
  1279. memcpy(retval, dentry->d_name.name, namelen);
  1280. return retval;
  1281. Elong:
  1282. return ERR_PTR(-ENAMETOOLONG);
  1283. }
  1284. /* write full pathname into buffer and return start of pathname */
  1285. char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
  1286. char *buf, int buflen)
  1287. {
  1288. char *res;
  1289. struct vfsmount *rootmnt;
  1290. struct dentry *root;
  1291. read_lock(&current->fs->lock);
  1292. rootmnt = mntget(current->fs->rootmnt);
  1293. root = dget(current->fs->root);
  1294. read_unlock(&current->fs->lock);
  1295. spin_lock(&dcache_lock);
  1296. res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
  1297. spin_unlock(&dcache_lock);
  1298. dput(root);
  1299. mntput(rootmnt);
  1300. return res;
  1301. }
  1302. /*
  1303. * NOTE! The user-level library version returns a
  1304. * character pointer. The kernel system call just
  1305. * returns the length of the buffer filled (which
  1306. * includes the ending '\0' character), or a negative
  1307. * error value. So libc would do something like
  1308. *
  1309. * char *getcwd(char * buf, size_t size)
  1310. * {
  1311. * int retval;
  1312. *
  1313. * retval = sys_getcwd(buf, size);
  1314. * if (retval >= 0)
  1315. * return buf;
  1316. * errno = -retval;
  1317. * return NULL;
  1318. * }
  1319. */
  1320. asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
  1321. {
  1322. int error;
  1323. struct vfsmount *pwdmnt, *rootmnt;
  1324. struct dentry *pwd, *root;
  1325. char *page = (char *) __get_free_page(GFP_USER);
  1326. if (!page)
  1327. return -ENOMEM;
  1328. read_lock(&current->fs->lock);
  1329. pwdmnt = mntget(current->fs->pwdmnt);
  1330. pwd = dget(current->fs->pwd);
  1331. rootmnt = mntget(current->fs->rootmnt);
  1332. root = dget(current->fs->root);
  1333. read_unlock(&current->fs->lock);
  1334. error = -ENOENT;
  1335. /* Has the current directory has been unlinked? */
  1336. spin_lock(&dcache_lock);
  1337. if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
  1338. unsigned long len;
  1339. char * cwd;
  1340. cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
  1341. spin_unlock(&dcache_lock);
  1342. error = PTR_ERR(cwd);
  1343. if (IS_ERR(cwd))
  1344. goto out;
  1345. error = -ERANGE;
  1346. len = PAGE_SIZE + page - cwd;
  1347. if (len <= size) {
  1348. error = len;
  1349. if (copy_to_user(buf, cwd, len))
  1350. error = -EFAULT;
  1351. }
  1352. } else
  1353. spin_unlock(&dcache_lock);
  1354. out:
  1355. dput(pwd);
  1356. mntput(pwdmnt);
  1357. dput(root);
  1358. mntput(rootmnt);
  1359. free_page((unsigned long) page);
  1360. return error;
  1361. }
  1362. /*
  1363. * Test whether new_dentry is a subdirectory of old_dentry.
  1364. *
  1365. * Trivially implemented using the dcache structure
  1366. */
  1367. /**
  1368. * is_subdir - is new dentry a subdirectory of old_dentry
  1369. * @new_dentry: new dentry
  1370. * @old_dentry: old dentry
  1371. *
  1372. * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
  1373. * Returns 0 otherwise.
  1374. * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
  1375. */
  1376. int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
  1377. {
  1378. int result;
  1379. struct dentry * saved = new_dentry;
  1380. unsigned long seq;
  1381. /* need rcu_readlock to protect against the d_parent trashing due to
  1382. * d_move
  1383. */
  1384. rcu_read_lock();
  1385. do {
  1386. /* for restarting inner loop in case of seq retry */
  1387. new_dentry = saved;
  1388. result = 0;
  1389. seq = read_seqbegin(&rename_lock);
  1390. for (;;) {
  1391. if (new_dentry != old_dentry) {
  1392. struct dentry * parent = new_dentry->d_parent;
  1393. if (parent == new_dentry)
  1394. break;
  1395. new_dentry = parent;
  1396. continue;
  1397. }
  1398. result = 1;
  1399. break;
  1400. }
  1401. } while (read_seqretry(&rename_lock, seq));
  1402. rcu_read_unlock();
  1403. return result;
  1404. }
  1405. void d_genocide(struct dentry *root)
  1406. {
  1407. struct dentry *this_parent = root;
  1408. struct list_head *next;
  1409. spin_lock(&dcache_lock);
  1410. repeat:
  1411. next = this_parent->d_subdirs.next;
  1412. resume:
  1413. while (next != &this_parent->d_subdirs) {
  1414. struct list_head *tmp = next;
  1415. struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
  1416. next = tmp->next;
  1417. if (d_unhashed(dentry)||!dentry->d_inode)
  1418. continue;
  1419. if (!list_empty(&dentry->d_subdirs)) {
  1420. this_parent = dentry;
  1421. goto repeat;
  1422. }
  1423. atomic_dec(&dentry->d_count);
  1424. }
  1425. if (this_parent != root) {
  1426. next = this_parent->d_child.next;
  1427. atomic_dec(&this_parent->d_count);
  1428. this_parent = this_parent->d_parent;
  1429. goto resume;
  1430. }
  1431. spin_unlock(&dcache_lock);
  1432. }
  1433. /**
  1434. * find_inode_number - check for dentry with name
  1435. * @dir: directory to check
  1436. * @name: Name to find.
  1437. *
  1438. * Check whether a dentry already exists for the given name,
  1439. * and return the inode number if it has an inode. Otherwise
  1440. * 0 is returned.
  1441. *
  1442. * This routine is used to post-process directory listings for
  1443. * filesystems using synthetic inode numbers, and is necessary
  1444. * to keep getcwd() working.
  1445. */
  1446. ino_t find_inode_number(struct dentry *dir, struct qstr *name)
  1447. {
  1448. struct dentry * dentry;
  1449. ino_t ino = 0;
  1450. /*
  1451. * Check for a fs-specific hash function. Note that we must
  1452. * calculate the standard hash first, as the d_op->d_hash()
  1453. * routine may choose to leave the hash value unchanged.
  1454. */
  1455. name->hash = full_name_hash(name->name, name->len);
  1456. if (dir->d_op && dir->d_op->d_hash)
  1457. {
  1458. if (dir->d_op->d_hash(dir, name) != 0)
  1459. goto out;
  1460. }
  1461. dentry = d_lookup(dir, name);
  1462. if (dentry)
  1463. {
  1464. if (dentry->d_inode)
  1465. ino = dentry->d_inode->i_ino;
  1466. dput(dentry);
  1467. }
  1468. out:
  1469. return ino;
  1470. }
  1471. static __initdata unsigned long dhash_entries;
  1472. static int __init set_dhash_entries(char *str)
  1473. {
  1474. if (!str)
  1475. return 0;
  1476. dhash_entries = simple_strtoul(str, &str, 0);
  1477. return 1;
  1478. }
  1479. __setup("dhash_entries=", set_dhash_entries);
  1480. static void __init dcache_init_early(void)
  1481. {
  1482. int loop;
  1483. /* If hashes are distributed across NUMA nodes, defer
  1484. * hash allocation until vmalloc space is available.
  1485. */
  1486. if (hashdist)
  1487. return;
  1488. dentry_hashtable =
  1489. alloc_large_system_hash("Dentry cache",
  1490. sizeof(struct hlist_head),
  1491. dhash_entries,
  1492. 13,
  1493. HASH_EARLY,
  1494. &d_hash_shift,
  1495. &d_hash_mask,
  1496. 0);
  1497. for (loop = 0; loop < (1 << d_hash_shift); loop++)
  1498. INIT_HLIST_HEAD(&dentry_hashtable[loop]);
  1499. }
  1500. static void __init dcache_init(unsigned long mempages)
  1501. {
  1502. int loop;
  1503. /*
  1504. * A constructor could be added for stable state like the lists,
  1505. * but it is probably not worth it because of the cache nature
  1506. * of the dcache.
  1507. */
  1508. dentry_cache = kmem_cache_create("dentry_cache",
  1509. sizeof(struct dentry),
  1510. 0,
  1511. SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
  1512. NULL, NULL);
  1513. set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
  1514. /* Hash may have been set up in dcache_init_early */
  1515. if (!hashdist)
  1516. return;
  1517. dentry_hashtable =
  1518. alloc_large_system_hash("Dentry cache",
  1519. sizeof(struct hlist_head),
  1520. dhash_entries,
  1521. 13,
  1522. 0,
  1523. &d_hash_shift,
  1524. &d_hash_mask,
  1525. 0);
  1526. for (loop = 0; loop < (1 << d_hash_shift); loop++)
  1527. INIT_HLIST_HEAD(&dentry_hashtable[loop]);
  1528. }
  1529. /* SLAB cache for __getname() consumers */
  1530. kmem_cache_t *names_cachep;
  1531. /* SLAB cache for file structures */
  1532. kmem_cache_t *filp_cachep;
  1533. EXPORT_SYMBOL(d_genocide);
  1534. extern void bdev_cache_init(void);
  1535. extern void chrdev_init(void);
  1536. void __init vfs_caches_init_early(void)
  1537. {
  1538. dcache_init_early();
  1539. inode_init_early();
  1540. }
  1541. void __init vfs_caches_init(unsigned long mempages)
  1542. {
  1543. unsigned long reserve;
  1544. /* Base hash sizes on available memory, with a reserve equal to
  1545. 150% of current kernel size */
  1546. reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
  1547. mempages -= reserve;
  1548. names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
  1549. SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
  1550. filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
  1551. SLAB_HWCACHE_ALIGN|SLAB_PANIC, filp_ctor, filp_dtor);
  1552. dcache_init(mempages);
  1553. inode_init(mempages);
  1554. files_init(mempages);
  1555. mnt_init(mempages);
  1556. bdev_cache_init();
  1557. chrdev_init();
  1558. }
  1559. EXPORT_SYMBOL(d_alloc);
  1560. EXPORT_SYMBOL(d_alloc_anon);
  1561. EXPORT_SYMBOL(d_alloc_root);
  1562. EXPORT_SYMBOL(d_delete);
  1563. EXPORT_SYMBOL(d_find_alias);
  1564. EXPORT_SYMBOL(d_instantiate);
  1565. EXPORT_SYMBOL(d_invalidate);
  1566. EXPORT_SYMBOL(d_lookup);
  1567. EXPORT_SYMBOL(d_move);
  1568. EXPORT_SYMBOL(d_path);
  1569. EXPORT_SYMBOL(d_prune_aliases);
  1570. EXPORT_SYMBOL(d_rehash);
  1571. EXPORT_SYMBOL(d_splice_alias);
  1572. EXPORT_SYMBOL(d_validate);
  1573. EXPORT_SYMBOL(dget_locked);
  1574. EXPORT_SYMBOL(dput);
  1575. EXPORT_SYMBOL(find_inode_number);
  1576. EXPORT_SYMBOL(have_submounts);
  1577. EXPORT_SYMBOL(names_cachep);
  1578. EXPORT_SYMBOL(shrink_dcache_parent);
  1579. EXPORT_SYMBOL(shrink_dcache_sb);