inode.c 49 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901
  1. /*
  2. * (C) 1997 Linus Torvalds
  3. * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
  4. */
  5. #include <linux/export.h>
  6. #include <linux/fs.h>
  7. #include <linux/mm.h>
  8. #include <linux/backing-dev.h>
  9. #include <linux/hash.h>
  10. #include <linux/swap.h>
  11. #include <linux/security.h>
  12. #include <linux/cdev.h>
  13. #include <linux/bootmem.h>
  14. #include <linux/fsnotify.h>
  15. #include <linux/mount.h>
  16. #include <linux/posix_acl.h>
  17. #include <linux/prefetch.h>
  18. #include <linux/buffer_head.h> /* for inode_has_buffers */
  19. #include <linux/ratelimit.h>
  20. #include <linux/list_lru.h>
  21. #include "internal.h"
  22. /*
  23. * Inode locking rules:
  24. *
  25. * inode->i_lock protects:
  26. * inode->i_state, inode->i_hash, __iget()
  27. * Inode LRU list locks protect:
  28. * inode->i_sb->s_inode_lru, inode->i_lru
  29. * inode_sb_list_lock protects:
  30. * sb->s_inodes, inode->i_sb_list
  31. * bdi->wb.list_lock protects:
  32. * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
  33. * inode_hash_lock protects:
  34. * inode_hashtable, inode->i_hash
  35. *
  36. * Lock ordering:
  37. *
  38. * inode_sb_list_lock
  39. * inode->i_lock
  40. * Inode LRU list locks
  41. *
  42. * bdi->wb.list_lock
  43. * inode->i_lock
  44. *
  45. * inode_hash_lock
  46. * inode_sb_list_lock
  47. * inode->i_lock
  48. *
  49. * iunique_lock
  50. * inode_hash_lock
  51. */
  52. static unsigned int i_hash_mask __read_mostly;
  53. static unsigned int i_hash_shift __read_mostly;
  54. static struct hlist_head *inode_hashtable __read_mostly;
  55. static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
  56. __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
  57. /*
  58. * Empty aops. Can be used for the cases where the user does not
  59. * define any of the address_space operations.
  60. */
  61. const struct address_space_operations empty_aops = {
  62. };
  63. EXPORT_SYMBOL(empty_aops);
  64. /*
  65. * Statistics gathering..
  66. */
  67. struct inodes_stat_t inodes_stat;
  68. static DEFINE_PER_CPU(unsigned long, nr_inodes);
  69. static DEFINE_PER_CPU(unsigned long, nr_unused);
  70. static struct kmem_cache *inode_cachep __read_mostly;
  71. static long get_nr_inodes(void)
  72. {
  73. int i;
  74. long sum = 0;
  75. for_each_possible_cpu(i)
  76. sum += per_cpu(nr_inodes, i);
  77. return sum < 0 ? 0 : sum;
  78. }
  79. static inline long get_nr_inodes_unused(void)
  80. {
  81. int i;
  82. long sum = 0;
  83. for_each_possible_cpu(i)
  84. sum += per_cpu(nr_unused, i);
  85. return sum < 0 ? 0 : sum;
  86. }
  87. long get_nr_dirty_inodes(void)
  88. {
  89. /* not actually dirty inodes, but a wild approximation */
  90. long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
  91. return nr_dirty > 0 ? nr_dirty : 0;
  92. }
  93. /*
  94. * Handle nr_inode sysctl
  95. */
  96. #ifdef CONFIG_SYSCTL
  97. int proc_nr_inodes(ctl_table *table, int write,
  98. void __user *buffer, size_t *lenp, loff_t *ppos)
  99. {
  100. inodes_stat.nr_inodes = get_nr_inodes();
  101. inodes_stat.nr_unused = get_nr_inodes_unused();
  102. return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
  103. }
  104. #endif
  105. /**
  106. * inode_init_always - perform inode structure intialisation
  107. * @sb: superblock inode belongs to
  108. * @inode: inode to initialise
  109. *
  110. * These are initializations that need to be done on every inode
  111. * allocation as the fields are not initialised by slab allocation.
  112. */
  113. int inode_init_always(struct super_block *sb, struct inode *inode)
  114. {
  115. static const struct inode_operations empty_iops;
  116. static const struct file_operations empty_fops;
  117. struct address_space *const mapping = &inode->i_data;
  118. inode->i_sb = sb;
  119. inode->i_blkbits = sb->s_blocksize_bits;
  120. inode->i_flags = 0;
  121. atomic_set(&inode->i_count, 1);
  122. inode->i_op = &empty_iops;
  123. inode->i_fop = &empty_fops;
  124. inode->__i_nlink = 1;
  125. inode->i_opflags = 0;
  126. i_uid_write(inode, 0);
  127. i_gid_write(inode, 0);
  128. atomic_set(&inode->i_writecount, 0);
  129. inode->i_size = 0;
  130. inode->i_blocks = 0;
  131. inode->i_bytes = 0;
  132. inode->i_generation = 0;
  133. #ifdef CONFIG_QUOTA
  134. memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
  135. #endif
  136. inode->i_pipe = NULL;
  137. inode->i_bdev = NULL;
  138. inode->i_cdev = NULL;
  139. inode->i_rdev = 0;
  140. inode->dirtied_when = 0;
  141. if (security_inode_alloc(inode))
  142. goto out;
  143. spin_lock_init(&inode->i_lock);
  144. lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
  145. mutex_init(&inode->i_mutex);
  146. lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
  147. atomic_set(&inode->i_dio_count, 0);
  148. mapping->a_ops = &empty_aops;
  149. mapping->host = inode;
  150. mapping->flags = 0;
  151. mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
  152. mapping->private_data = NULL;
  153. mapping->backing_dev_info = &default_backing_dev_info;
  154. mapping->writeback_index = 0;
  155. /*
  156. * If the block_device provides a backing_dev_info for client
  157. * inodes then use that. Otherwise the inode share the bdev's
  158. * backing_dev_info.
  159. */
  160. if (sb->s_bdev) {
  161. struct backing_dev_info *bdi;
  162. bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
  163. mapping->backing_dev_info = bdi;
  164. }
  165. inode->i_private = NULL;
  166. inode->i_mapping = mapping;
  167. INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
  168. #ifdef CONFIG_FS_POSIX_ACL
  169. inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
  170. #endif
  171. #ifdef CONFIG_FSNOTIFY
  172. inode->i_fsnotify_mask = 0;
  173. #endif
  174. this_cpu_inc(nr_inodes);
  175. return 0;
  176. out:
  177. return -ENOMEM;
  178. }
  179. EXPORT_SYMBOL(inode_init_always);
  180. static struct inode *alloc_inode(struct super_block *sb)
  181. {
  182. struct inode *inode;
  183. if (sb->s_op->alloc_inode)
  184. inode = sb->s_op->alloc_inode(sb);
  185. else
  186. inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
  187. if (!inode)
  188. return NULL;
  189. if (unlikely(inode_init_always(sb, inode))) {
  190. if (inode->i_sb->s_op->destroy_inode)
  191. inode->i_sb->s_op->destroy_inode(inode);
  192. else
  193. kmem_cache_free(inode_cachep, inode);
  194. return NULL;
  195. }
  196. return inode;
  197. }
  198. void free_inode_nonrcu(struct inode *inode)
  199. {
  200. kmem_cache_free(inode_cachep, inode);
  201. }
  202. EXPORT_SYMBOL(free_inode_nonrcu);
  203. void __destroy_inode(struct inode *inode)
  204. {
  205. BUG_ON(inode_has_buffers(inode));
  206. security_inode_free(inode);
  207. fsnotify_inode_delete(inode);
  208. if (!inode->i_nlink) {
  209. WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
  210. atomic_long_dec(&inode->i_sb->s_remove_count);
  211. }
  212. #ifdef CONFIG_FS_POSIX_ACL
  213. if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
  214. posix_acl_release(inode->i_acl);
  215. if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
  216. posix_acl_release(inode->i_default_acl);
  217. #endif
  218. this_cpu_dec(nr_inodes);
  219. }
  220. EXPORT_SYMBOL(__destroy_inode);
  221. static void i_callback(struct rcu_head *head)
  222. {
  223. struct inode *inode = container_of(head, struct inode, i_rcu);
  224. kmem_cache_free(inode_cachep, inode);
  225. }
  226. static void destroy_inode(struct inode *inode)
  227. {
  228. BUG_ON(!list_empty(&inode->i_lru));
  229. __destroy_inode(inode);
  230. if (inode->i_sb->s_op->destroy_inode)
  231. inode->i_sb->s_op->destroy_inode(inode);
  232. else
  233. call_rcu(&inode->i_rcu, i_callback);
  234. }
  235. /**
  236. * drop_nlink - directly drop an inode's link count
  237. * @inode: inode
  238. *
  239. * This is a low-level filesystem helper to replace any
  240. * direct filesystem manipulation of i_nlink. In cases
  241. * where we are attempting to track writes to the
  242. * filesystem, a decrement to zero means an imminent
  243. * write when the file is truncated and actually unlinked
  244. * on the filesystem.
  245. */
  246. void drop_nlink(struct inode *inode)
  247. {
  248. WARN_ON(inode->i_nlink == 0);
  249. inode->__i_nlink--;
  250. if (!inode->i_nlink)
  251. atomic_long_inc(&inode->i_sb->s_remove_count);
  252. }
  253. EXPORT_SYMBOL(drop_nlink);
  254. /**
  255. * clear_nlink - directly zero an inode's link count
  256. * @inode: inode
  257. *
  258. * This is a low-level filesystem helper to replace any
  259. * direct filesystem manipulation of i_nlink. See
  260. * drop_nlink() for why we care about i_nlink hitting zero.
  261. */
  262. void clear_nlink(struct inode *inode)
  263. {
  264. if (inode->i_nlink) {
  265. inode->__i_nlink = 0;
  266. atomic_long_inc(&inode->i_sb->s_remove_count);
  267. }
  268. }
  269. EXPORT_SYMBOL(clear_nlink);
  270. /**
  271. * set_nlink - directly set an inode's link count
  272. * @inode: inode
  273. * @nlink: new nlink (should be non-zero)
  274. *
  275. * This is a low-level filesystem helper to replace any
  276. * direct filesystem manipulation of i_nlink.
  277. */
  278. void set_nlink(struct inode *inode, unsigned int nlink)
  279. {
  280. if (!nlink) {
  281. clear_nlink(inode);
  282. } else {
  283. /* Yes, some filesystems do change nlink from zero to one */
  284. if (inode->i_nlink == 0)
  285. atomic_long_dec(&inode->i_sb->s_remove_count);
  286. inode->__i_nlink = nlink;
  287. }
  288. }
  289. EXPORT_SYMBOL(set_nlink);
  290. /**
  291. * inc_nlink - directly increment an inode's link count
  292. * @inode: inode
  293. *
  294. * This is a low-level filesystem helper to replace any
  295. * direct filesystem manipulation of i_nlink. Currently,
  296. * it is only here for parity with dec_nlink().
  297. */
  298. void inc_nlink(struct inode *inode)
  299. {
  300. if (unlikely(inode->i_nlink == 0)) {
  301. WARN_ON(!(inode->i_state & I_LINKABLE));
  302. atomic_long_dec(&inode->i_sb->s_remove_count);
  303. }
  304. inode->__i_nlink++;
  305. }
  306. EXPORT_SYMBOL(inc_nlink);
  307. void address_space_init_once(struct address_space *mapping)
  308. {
  309. memset(mapping, 0, sizeof(*mapping));
  310. INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
  311. spin_lock_init(&mapping->tree_lock);
  312. mutex_init(&mapping->i_mmap_mutex);
  313. INIT_LIST_HEAD(&mapping->private_list);
  314. spin_lock_init(&mapping->private_lock);
  315. mapping->i_mmap = RB_ROOT;
  316. INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
  317. }
  318. EXPORT_SYMBOL(address_space_init_once);
  319. /*
  320. * These are initializations that only need to be done
  321. * once, because the fields are idempotent across use
  322. * of the inode, so let the slab aware of that.
  323. */
  324. void inode_init_once(struct inode *inode)
  325. {
  326. memset(inode, 0, sizeof(*inode));
  327. INIT_HLIST_NODE(&inode->i_hash);
  328. INIT_LIST_HEAD(&inode->i_devices);
  329. INIT_LIST_HEAD(&inode->i_wb_list);
  330. INIT_LIST_HEAD(&inode->i_lru);
  331. address_space_init_once(&inode->i_data);
  332. i_size_ordered_init(inode);
  333. #ifdef CONFIG_FSNOTIFY
  334. INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
  335. #endif
  336. }
  337. EXPORT_SYMBOL(inode_init_once);
  338. static void init_once(void *foo)
  339. {
  340. struct inode *inode = (struct inode *) foo;
  341. inode_init_once(inode);
  342. }
  343. /*
  344. * inode->i_lock must be held
  345. */
  346. void __iget(struct inode *inode)
  347. {
  348. atomic_inc(&inode->i_count);
  349. }
  350. /*
  351. * get additional reference to inode; caller must already hold one.
  352. */
  353. void ihold(struct inode *inode)
  354. {
  355. WARN_ON(atomic_inc_return(&inode->i_count) < 2);
  356. }
  357. EXPORT_SYMBOL(ihold);
  358. static void inode_lru_list_add(struct inode *inode)
  359. {
  360. if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
  361. this_cpu_inc(nr_unused);
  362. }
  363. /*
  364. * Add inode to LRU if needed (inode is unused and clean).
  365. *
  366. * Needs inode->i_lock held.
  367. */
  368. void inode_add_lru(struct inode *inode)
  369. {
  370. if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) &&
  371. !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
  372. inode_lru_list_add(inode);
  373. }
  374. static void inode_lru_list_del(struct inode *inode)
  375. {
  376. if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
  377. this_cpu_dec(nr_unused);
  378. }
  379. /**
  380. * inode_sb_list_add - add inode to the superblock list of inodes
  381. * @inode: inode to add
  382. */
  383. void inode_sb_list_add(struct inode *inode)
  384. {
  385. spin_lock(&inode_sb_list_lock);
  386. list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
  387. spin_unlock(&inode_sb_list_lock);
  388. }
  389. EXPORT_SYMBOL_GPL(inode_sb_list_add);
  390. static inline void inode_sb_list_del(struct inode *inode)
  391. {
  392. if (!list_empty(&inode->i_sb_list)) {
  393. spin_lock(&inode_sb_list_lock);
  394. list_del_init(&inode->i_sb_list);
  395. spin_unlock(&inode_sb_list_lock);
  396. }
  397. }
  398. static unsigned long hash(struct super_block *sb, unsigned long hashval)
  399. {
  400. unsigned long tmp;
  401. tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
  402. L1_CACHE_BYTES;
  403. tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
  404. return tmp & i_hash_mask;
  405. }
  406. /**
  407. * __insert_inode_hash - hash an inode
  408. * @inode: unhashed inode
  409. * @hashval: unsigned long value used to locate this object in the
  410. * inode_hashtable.
  411. *
  412. * Add an inode to the inode hash for this superblock.
  413. */
  414. void __insert_inode_hash(struct inode *inode, unsigned long hashval)
  415. {
  416. struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
  417. spin_lock(&inode_hash_lock);
  418. spin_lock(&inode->i_lock);
  419. hlist_add_head(&inode->i_hash, b);
  420. spin_unlock(&inode->i_lock);
  421. spin_unlock(&inode_hash_lock);
  422. }
  423. EXPORT_SYMBOL(__insert_inode_hash);
  424. /**
  425. * __remove_inode_hash - remove an inode from the hash
  426. * @inode: inode to unhash
  427. *
  428. * Remove an inode from the superblock.
  429. */
  430. void __remove_inode_hash(struct inode *inode)
  431. {
  432. spin_lock(&inode_hash_lock);
  433. spin_lock(&inode->i_lock);
  434. hlist_del_init(&inode->i_hash);
  435. spin_unlock(&inode->i_lock);
  436. spin_unlock(&inode_hash_lock);
  437. }
  438. EXPORT_SYMBOL(__remove_inode_hash);
  439. void clear_inode(struct inode *inode)
  440. {
  441. might_sleep();
  442. /*
  443. * We have to cycle tree_lock here because reclaim can be still in the
  444. * process of removing the last page (in __delete_from_page_cache())
  445. * and we must not free mapping under it.
  446. */
  447. spin_lock_irq(&inode->i_data.tree_lock);
  448. BUG_ON(inode->i_data.nrpages);
  449. spin_unlock_irq(&inode->i_data.tree_lock);
  450. BUG_ON(!list_empty(&inode->i_data.private_list));
  451. BUG_ON(!(inode->i_state & I_FREEING));
  452. BUG_ON(inode->i_state & I_CLEAR);
  453. /* don't need i_lock here, no concurrent mods to i_state */
  454. inode->i_state = I_FREEING | I_CLEAR;
  455. }
  456. EXPORT_SYMBOL(clear_inode);
  457. /*
  458. * Free the inode passed in, removing it from the lists it is still connected
  459. * to. We remove any pages still attached to the inode and wait for any IO that
  460. * is still in progress before finally destroying the inode.
  461. *
  462. * An inode must already be marked I_FREEING so that we avoid the inode being
  463. * moved back onto lists if we race with other code that manipulates the lists
  464. * (e.g. writeback_single_inode). The caller is responsible for setting this.
  465. *
  466. * An inode must already be removed from the LRU list before being evicted from
  467. * the cache. This should occur atomically with setting the I_FREEING state
  468. * flag, so no inodes here should ever be on the LRU when being evicted.
  469. */
  470. static void evict(struct inode *inode)
  471. {
  472. const struct super_operations *op = inode->i_sb->s_op;
  473. BUG_ON(!(inode->i_state & I_FREEING));
  474. BUG_ON(!list_empty(&inode->i_lru));
  475. if (!list_empty(&inode->i_wb_list))
  476. inode_wb_list_del(inode);
  477. inode_sb_list_del(inode);
  478. /*
  479. * Wait for flusher thread to be done with the inode so that filesystem
  480. * does not start destroying it while writeback is still running. Since
  481. * the inode has I_FREEING set, flusher thread won't start new work on
  482. * the inode. We just have to wait for running writeback to finish.
  483. */
  484. inode_wait_for_writeback(inode);
  485. if (op->evict_inode) {
  486. op->evict_inode(inode);
  487. } else {
  488. if (inode->i_data.nrpages)
  489. truncate_inode_pages(&inode->i_data, 0);
  490. clear_inode(inode);
  491. }
  492. if (S_ISBLK(inode->i_mode) && inode->i_bdev)
  493. bd_forget(inode);
  494. if (S_ISCHR(inode->i_mode) && inode->i_cdev)
  495. cd_forget(inode);
  496. remove_inode_hash(inode);
  497. spin_lock(&inode->i_lock);
  498. wake_up_bit(&inode->i_state, __I_NEW);
  499. BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
  500. spin_unlock(&inode->i_lock);
  501. destroy_inode(inode);
  502. }
  503. /*
  504. * dispose_list - dispose of the contents of a local list
  505. * @head: the head of the list to free
  506. *
  507. * Dispose-list gets a local list with local inodes in it, so it doesn't
  508. * need to worry about list corruption and SMP locks.
  509. */
  510. static void dispose_list(struct list_head *head)
  511. {
  512. while (!list_empty(head)) {
  513. struct inode *inode;
  514. inode = list_first_entry(head, struct inode, i_lru);
  515. list_del_init(&inode->i_lru);
  516. evict(inode);
  517. }
  518. }
  519. /**
  520. * evict_inodes - evict all evictable inodes for a superblock
  521. * @sb: superblock to operate on
  522. *
  523. * Make sure that no inodes with zero refcount are retained. This is
  524. * called by superblock shutdown after having MS_ACTIVE flag removed,
  525. * so any inode reaching zero refcount during or after that call will
  526. * be immediately evicted.
  527. */
  528. void evict_inodes(struct super_block *sb)
  529. {
  530. struct inode *inode, *next;
  531. LIST_HEAD(dispose);
  532. spin_lock(&inode_sb_list_lock);
  533. list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
  534. if (atomic_read(&inode->i_count))
  535. continue;
  536. spin_lock(&inode->i_lock);
  537. if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
  538. spin_unlock(&inode->i_lock);
  539. continue;
  540. }
  541. inode->i_state |= I_FREEING;
  542. inode_lru_list_del(inode);
  543. spin_unlock(&inode->i_lock);
  544. list_add(&inode->i_lru, &dispose);
  545. }
  546. spin_unlock(&inode_sb_list_lock);
  547. dispose_list(&dispose);
  548. }
  549. /**
  550. * invalidate_inodes - attempt to free all inodes on a superblock
  551. * @sb: superblock to operate on
  552. * @kill_dirty: flag to guide handling of dirty inodes
  553. *
  554. * Attempts to free all inodes for a given superblock. If there were any
  555. * busy inodes return a non-zero value, else zero.
  556. * If @kill_dirty is set, discard dirty inodes too, otherwise treat
  557. * them as busy.
  558. */
  559. int invalidate_inodes(struct super_block *sb, bool kill_dirty)
  560. {
  561. int busy = 0;
  562. struct inode *inode, *next;
  563. LIST_HEAD(dispose);
  564. spin_lock(&inode_sb_list_lock);
  565. list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
  566. spin_lock(&inode->i_lock);
  567. if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
  568. spin_unlock(&inode->i_lock);
  569. continue;
  570. }
  571. if (inode->i_state & I_DIRTY && !kill_dirty) {
  572. spin_unlock(&inode->i_lock);
  573. busy = 1;
  574. continue;
  575. }
  576. if (atomic_read(&inode->i_count)) {
  577. spin_unlock(&inode->i_lock);
  578. busy = 1;
  579. continue;
  580. }
  581. inode->i_state |= I_FREEING;
  582. inode_lru_list_del(inode);
  583. spin_unlock(&inode->i_lock);
  584. list_add(&inode->i_lru, &dispose);
  585. }
  586. spin_unlock(&inode_sb_list_lock);
  587. dispose_list(&dispose);
  588. return busy;
  589. }
  590. /*
  591. * Isolate the inode from the LRU in preparation for freeing it.
  592. *
  593. * Any inodes which are pinned purely because of attached pagecache have their
  594. * pagecache removed. If the inode has metadata buffers attached to
  595. * mapping->private_list then try to remove them.
  596. *
  597. * If the inode has the I_REFERENCED flag set, then it means that it has been
  598. * used recently - the flag is set in iput_final(). When we encounter such an
  599. * inode, clear the flag and move it to the back of the LRU so it gets another
  600. * pass through the LRU before it gets reclaimed. This is necessary because of
  601. * the fact we are doing lazy LRU updates to minimise lock contention so the
  602. * LRU does not have strict ordering. Hence we don't want to reclaim inodes
  603. * with this flag set because they are the inodes that are out of order.
  604. */
  605. static enum lru_status
  606. inode_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
  607. {
  608. struct list_head *freeable = arg;
  609. struct inode *inode = container_of(item, struct inode, i_lru);
  610. /*
  611. * we are inverting the lru lock/inode->i_lock here, so use a trylock.
  612. * If we fail to get the lock, just skip it.
  613. */
  614. if (!spin_trylock(&inode->i_lock))
  615. return LRU_SKIP;
  616. /*
  617. * Referenced or dirty inodes are still in use. Give them another pass
  618. * through the LRU as we canot reclaim them now.
  619. */
  620. if (atomic_read(&inode->i_count) ||
  621. (inode->i_state & ~I_REFERENCED)) {
  622. list_del_init(&inode->i_lru);
  623. spin_unlock(&inode->i_lock);
  624. this_cpu_dec(nr_unused);
  625. return LRU_REMOVED;
  626. }
  627. /* recently referenced inodes get one more pass */
  628. if (inode->i_state & I_REFERENCED) {
  629. inode->i_state &= ~I_REFERENCED;
  630. spin_unlock(&inode->i_lock);
  631. return LRU_ROTATE;
  632. }
  633. if (inode_has_buffers(inode) || inode->i_data.nrpages) {
  634. __iget(inode);
  635. spin_unlock(&inode->i_lock);
  636. spin_unlock(lru_lock);
  637. if (remove_inode_buffers(inode)) {
  638. unsigned long reap;
  639. reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
  640. if (current_is_kswapd())
  641. __count_vm_events(KSWAPD_INODESTEAL, reap);
  642. else
  643. __count_vm_events(PGINODESTEAL, reap);
  644. if (current->reclaim_state)
  645. current->reclaim_state->reclaimed_slab += reap;
  646. }
  647. iput(inode);
  648. spin_lock(lru_lock);
  649. return LRU_RETRY;
  650. }
  651. WARN_ON(inode->i_state & I_NEW);
  652. inode->i_state |= I_FREEING;
  653. list_move(&inode->i_lru, freeable);
  654. spin_unlock(&inode->i_lock);
  655. this_cpu_dec(nr_unused);
  656. return LRU_REMOVED;
  657. }
  658. /*
  659. * Walk the superblock inode LRU for freeable inodes and attempt to free them.
  660. * This is called from the superblock shrinker function with a number of inodes
  661. * to trim from the LRU. Inodes to be freed are moved to a temporary list and
  662. * then are freed outside inode_lock by dispose_list().
  663. */
  664. long prune_icache_sb(struct super_block *sb, unsigned long nr_to_scan,
  665. int nid)
  666. {
  667. LIST_HEAD(freeable);
  668. long freed;
  669. freed = list_lru_walk_node(&sb->s_inode_lru, nid, inode_lru_isolate,
  670. &freeable, &nr_to_scan);
  671. dispose_list(&freeable);
  672. return freed;
  673. }
  674. static void __wait_on_freeing_inode(struct inode *inode);
  675. /*
  676. * Called with the inode lock held.
  677. */
  678. static struct inode *find_inode(struct super_block *sb,
  679. struct hlist_head *head,
  680. int (*test)(struct inode *, void *),
  681. void *data)
  682. {
  683. struct inode *inode = NULL;
  684. repeat:
  685. hlist_for_each_entry(inode, head, i_hash) {
  686. if (inode->i_sb != sb)
  687. continue;
  688. if (!test(inode, data))
  689. continue;
  690. spin_lock(&inode->i_lock);
  691. if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
  692. __wait_on_freeing_inode(inode);
  693. goto repeat;
  694. }
  695. __iget(inode);
  696. spin_unlock(&inode->i_lock);
  697. return inode;
  698. }
  699. return NULL;
  700. }
  701. /*
  702. * find_inode_fast is the fast path version of find_inode, see the comment at
  703. * iget_locked for details.
  704. */
  705. static struct inode *find_inode_fast(struct super_block *sb,
  706. struct hlist_head *head, unsigned long ino)
  707. {
  708. struct inode *inode = NULL;
  709. repeat:
  710. hlist_for_each_entry(inode, head, i_hash) {
  711. if (inode->i_ino != ino)
  712. continue;
  713. if (inode->i_sb != sb)
  714. continue;
  715. spin_lock(&inode->i_lock);
  716. if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
  717. __wait_on_freeing_inode(inode);
  718. goto repeat;
  719. }
  720. __iget(inode);
  721. spin_unlock(&inode->i_lock);
  722. return inode;
  723. }
  724. return NULL;
  725. }
  726. /*
  727. * Each cpu owns a range of LAST_INO_BATCH numbers.
  728. * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
  729. * to renew the exhausted range.
  730. *
  731. * This does not significantly increase overflow rate because every CPU can
  732. * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
  733. * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
  734. * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
  735. * overflow rate by 2x, which does not seem too significant.
  736. *
  737. * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
  738. * error if st_ino won't fit in target struct field. Use 32bit counter
  739. * here to attempt to avoid that.
  740. */
  741. #define LAST_INO_BATCH 1024
  742. static DEFINE_PER_CPU(unsigned int, last_ino);
  743. unsigned int get_next_ino(void)
  744. {
  745. unsigned int *p = &get_cpu_var(last_ino);
  746. unsigned int res = *p;
  747. #ifdef CONFIG_SMP
  748. if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
  749. static atomic_t shared_last_ino;
  750. int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
  751. res = next - LAST_INO_BATCH;
  752. }
  753. #endif
  754. *p = ++res;
  755. put_cpu_var(last_ino);
  756. return res;
  757. }
  758. EXPORT_SYMBOL(get_next_ino);
  759. /**
  760. * new_inode_pseudo - obtain an inode
  761. * @sb: superblock
  762. *
  763. * Allocates a new inode for given superblock.
  764. * Inode wont be chained in superblock s_inodes list
  765. * This means :
  766. * - fs can't be unmount
  767. * - quotas, fsnotify, writeback can't work
  768. */
  769. struct inode *new_inode_pseudo(struct super_block *sb)
  770. {
  771. struct inode *inode = alloc_inode(sb);
  772. if (inode) {
  773. spin_lock(&inode->i_lock);
  774. inode->i_state = 0;
  775. spin_unlock(&inode->i_lock);
  776. INIT_LIST_HEAD(&inode->i_sb_list);
  777. }
  778. return inode;
  779. }
  780. /**
  781. * new_inode - obtain an inode
  782. * @sb: superblock
  783. *
  784. * Allocates a new inode for given superblock. The default gfp_mask
  785. * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
  786. * If HIGHMEM pages are unsuitable or it is known that pages allocated
  787. * for the page cache are not reclaimable or migratable,
  788. * mapping_set_gfp_mask() must be called with suitable flags on the
  789. * newly created inode's mapping
  790. *
  791. */
  792. struct inode *new_inode(struct super_block *sb)
  793. {
  794. struct inode *inode;
  795. spin_lock_prefetch(&inode_sb_list_lock);
  796. inode = new_inode_pseudo(sb);
  797. if (inode)
  798. inode_sb_list_add(inode);
  799. return inode;
  800. }
  801. EXPORT_SYMBOL(new_inode);
  802. #ifdef CONFIG_DEBUG_LOCK_ALLOC
  803. void lockdep_annotate_inode_mutex_key(struct inode *inode)
  804. {
  805. if (S_ISDIR(inode->i_mode)) {
  806. struct file_system_type *type = inode->i_sb->s_type;
  807. /* Set new key only if filesystem hasn't already changed it */
  808. if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
  809. /*
  810. * ensure nobody is actually holding i_mutex
  811. */
  812. mutex_destroy(&inode->i_mutex);
  813. mutex_init(&inode->i_mutex);
  814. lockdep_set_class(&inode->i_mutex,
  815. &type->i_mutex_dir_key);
  816. }
  817. }
  818. }
  819. EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
  820. #endif
  821. /**
  822. * unlock_new_inode - clear the I_NEW state and wake up any waiters
  823. * @inode: new inode to unlock
  824. *
  825. * Called when the inode is fully initialised to clear the new state of the
  826. * inode and wake up anyone waiting for the inode to finish initialisation.
  827. */
  828. void unlock_new_inode(struct inode *inode)
  829. {
  830. lockdep_annotate_inode_mutex_key(inode);
  831. spin_lock(&inode->i_lock);
  832. WARN_ON(!(inode->i_state & I_NEW));
  833. inode->i_state &= ~I_NEW;
  834. smp_mb();
  835. wake_up_bit(&inode->i_state, __I_NEW);
  836. spin_unlock(&inode->i_lock);
  837. }
  838. EXPORT_SYMBOL(unlock_new_inode);
  839. /**
  840. * lock_two_nondirectories - take two i_mutexes on non-directory objects
  841. * @inode1: first inode to lock
  842. * @inode2: second inode to lock
  843. */
  844. void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
  845. {
  846. WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
  847. if (inode1 == inode2 || !inode2) {
  848. mutex_lock(&inode1->i_mutex);
  849. return;
  850. }
  851. WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
  852. if (inode1 < inode2) {
  853. mutex_lock(&inode1->i_mutex);
  854. mutex_lock_nested(&inode2->i_mutex, I_MUTEX_NONDIR2);
  855. } else {
  856. mutex_lock(&inode2->i_mutex);
  857. mutex_lock_nested(&inode1->i_mutex, I_MUTEX_NONDIR2);
  858. }
  859. }
  860. EXPORT_SYMBOL(lock_two_nondirectories);
  861. /**
  862. * unlock_two_nondirectories - release locks from lock_two_nondirectories()
  863. * @inode1: first inode to unlock
  864. * @inode2: second inode to unlock
  865. */
  866. void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
  867. {
  868. mutex_unlock(&inode1->i_mutex);
  869. if (inode2 && inode2 != inode1)
  870. mutex_unlock(&inode2->i_mutex);
  871. }
  872. EXPORT_SYMBOL(unlock_two_nondirectories);
  873. /**
  874. * iget5_locked - obtain an inode from a mounted file system
  875. * @sb: super block of file system
  876. * @hashval: hash value (usually inode number) to get
  877. * @test: callback used for comparisons between inodes
  878. * @set: callback used to initialize a new struct inode
  879. * @data: opaque data pointer to pass to @test and @set
  880. *
  881. * Search for the inode specified by @hashval and @data in the inode cache,
  882. * and if present it is return it with an increased reference count. This is
  883. * a generalized version of iget_locked() for file systems where the inode
  884. * number is not sufficient for unique identification of an inode.
  885. *
  886. * If the inode is not in cache, allocate a new inode and return it locked,
  887. * hashed, and with the I_NEW flag set. The file system gets to fill it in
  888. * before unlocking it via unlock_new_inode().
  889. *
  890. * Note both @test and @set are called with the inode_hash_lock held, so can't
  891. * sleep.
  892. */
  893. struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
  894. int (*test)(struct inode *, void *),
  895. int (*set)(struct inode *, void *), void *data)
  896. {
  897. struct hlist_head *head = inode_hashtable + hash(sb, hashval);
  898. struct inode *inode;
  899. spin_lock(&inode_hash_lock);
  900. inode = find_inode(sb, head, test, data);
  901. spin_unlock(&inode_hash_lock);
  902. if (inode) {
  903. wait_on_inode(inode);
  904. return inode;
  905. }
  906. inode = alloc_inode(sb);
  907. if (inode) {
  908. struct inode *old;
  909. spin_lock(&inode_hash_lock);
  910. /* We released the lock, so.. */
  911. old = find_inode(sb, head, test, data);
  912. if (!old) {
  913. if (set(inode, data))
  914. goto set_failed;
  915. spin_lock(&inode->i_lock);
  916. inode->i_state = I_NEW;
  917. hlist_add_head(&inode->i_hash, head);
  918. spin_unlock(&inode->i_lock);
  919. inode_sb_list_add(inode);
  920. spin_unlock(&inode_hash_lock);
  921. /* Return the locked inode with I_NEW set, the
  922. * caller is responsible for filling in the contents
  923. */
  924. return inode;
  925. }
  926. /*
  927. * Uhhuh, somebody else created the same inode under
  928. * us. Use the old inode instead of the one we just
  929. * allocated.
  930. */
  931. spin_unlock(&inode_hash_lock);
  932. destroy_inode(inode);
  933. inode = old;
  934. wait_on_inode(inode);
  935. }
  936. return inode;
  937. set_failed:
  938. spin_unlock(&inode_hash_lock);
  939. destroy_inode(inode);
  940. return NULL;
  941. }
  942. EXPORT_SYMBOL(iget5_locked);
  943. /**
  944. * iget_locked - obtain an inode from a mounted file system
  945. * @sb: super block of file system
  946. * @ino: inode number to get
  947. *
  948. * Search for the inode specified by @ino in the inode cache and if present
  949. * return it with an increased reference count. This is for file systems
  950. * where the inode number is sufficient for unique identification of an inode.
  951. *
  952. * If the inode is not in cache, allocate a new inode and return it locked,
  953. * hashed, and with the I_NEW flag set. The file system gets to fill it in
  954. * before unlocking it via unlock_new_inode().
  955. */
  956. struct inode *iget_locked(struct super_block *sb, unsigned long ino)
  957. {
  958. struct hlist_head *head = inode_hashtable + hash(sb, ino);
  959. struct inode *inode;
  960. spin_lock(&inode_hash_lock);
  961. inode = find_inode_fast(sb, head, ino);
  962. spin_unlock(&inode_hash_lock);
  963. if (inode) {
  964. wait_on_inode(inode);
  965. return inode;
  966. }
  967. inode = alloc_inode(sb);
  968. if (inode) {
  969. struct inode *old;
  970. spin_lock(&inode_hash_lock);
  971. /* We released the lock, so.. */
  972. old = find_inode_fast(sb, head, ino);
  973. if (!old) {
  974. inode->i_ino = ino;
  975. spin_lock(&inode->i_lock);
  976. inode->i_state = I_NEW;
  977. hlist_add_head(&inode->i_hash, head);
  978. spin_unlock(&inode->i_lock);
  979. inode_sb_list_add(inode);
  980. spin_unlock(&inode_hash_lock);
  981. /* Return the locked inode with I_NEW set, the
  982. * caller is responsible for filling in the contents
  983. */
  984. return inode;
  985. }
  986. /*
  987. * Uhhuh, somebody else created the same inode under
  988. * us. Use the old inode instead of the one we just
  989. * allocated.
  990. */
  991. spin_unlock(&inode_hash_lock);
  992. destroy_inode(inode);
  993. inode = old;
  994. wait_on_inode(inode);
  995. }
  996. return inode;
  997. }
  998. EXPORT_SYMBOL(iget_locked);
  999. /*
  1000. * search the inode cache for a matching inode number.
  1001. * If we find one, then the inode number we are trying to
  1002. * allocate is not unique and so we should not use it.
  1003. *
  1004. * Returns 1 if the inode number is unique, 0 if it is not.
  1005. */
  1006. static int test_inode_iunique(struct super_block *sb, unsigned long ino)
  1007. {
  1008. struct hlist_head *b = inode_hashtable + hash(sb, ino);
  1009. struct inode *inode;
  1010. spin_lock(&inode_hash_lock);
  1011. hlist_for_each_entry(inode, b, i_hash) {
  1012. if (inode->i_ino == ino && inode->i_sb == sb) {
  1013. spin_unlock(&inode_hash_lock);
  1014. return 0;
  1015. }
  1016. }
  1017. spin_unlock(&inode_hash_lock);
  1018. return 1;
  1019. }
  1020. /**
  1021. * iunique - get a unique inode number
  1022. * @sb: superblock
  1023. * @max_reserved: highest reserved inode number
  1024. *
  1025. * Obtain an inode number that is unique on the system for a given
  1026. * superblock. This is used by file systems that have no natural
  1027. * permanent inode numbering system. An inode number is returned that
  1028. * is higher than the reserved limit but unique.
  1029. *
  1030. * BUGS:
  1031. * With a large number of inodes live on the file system this function
  1032. * currently becomes quite slow.
  1033. */
  1034. ino_t iunique(struct super_block *sb, ino_t max_reserved)
  1035. {
  1036. /*
  1037. * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
  1038. * error if st_ino won't fit in target struct field. Use 32bit counter
  1039. * here to attempt to avoid that.
  1040. */
  1041. static DEFINE_SPINLOCK(iunique_lock);
  1042. static unsigned int counter;
  1043. ino_t res;
  1044. spin_lock(&iunique_lock);
  1045. do {
  1046. if (counter <= max_reserved)
  1047. counter = max_reserved + 1;
  1048. res = counter++;
  1049. } while (!test_inode_iunique(sb, res));
  1050. spin_unlock(&iunique_lock);
  1051. return res;
  1052. }
  1053. EXPORT_SYMBOL(iunique);
  1054. struct inode *igrab(struct inode *inode)
  1055. {
  1056. spin_lock(&inode->i_lock);
  1057. if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
  1058. __iget(inode);
  1059. spin_unlock(&inode->i_lock);
  1060. } else {
  1061. spin_unlock(&inode->i_lock);
  1062. /*
  1063. * Handle the case where s_op->clear_inode is not been
  1064. * called yet, and somebody is calling igrab
  1065. * while the inode is getting freed.
  1066. */
  1067. inode = NULL;
  1068. }
  1069. return inode;
  1070. }
  1071. EXPORT_SYMBOL(igrab);
  1072. /**
  1073. * ilookup5_nowait - search for an inode in the inode cache
  1074. * @sb: super block of file system to search
  1075. * @hashval: hash value (usually inode number) to search for
  1076. * @test: callback used for comparisons between inodes
  1077. * @data: opaque data pointer to pass to @test
  1078. *
  1079. * Search for the inode specified by @hashval and @data in the inode cache.
  1080. * If the inode is in the cache, the inode is returned with an incremented
  1081. * reference count.
  1082. *
  1083. * Note: I_NEW is not waited upon so you have to be very careful what you do
  1084. * with the returned inode. You probably should be using ilookup5() instead.
  1085. *
  1086. * Note2: @test is called with the inode_hash_lock held, so can't sleep.
  1087. */
  1088. struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
  1089. int (*test)(struct inode *, void *), void *data)
  1090. {
  1091. struct hlist_head *head = inode_hashtable + hash(sb, hashval);
  1092. struct inode *inode;
  1093. spin_lock(&inode_hash_lock);
  1094. inode = find_inode(sb, head, test, data);
  1095. spin_unlock(&inode_hash_lock);
  1096. return inode;
  1097. }
  1098. EXPORT_SYMBOL(ilookup5_nowait);
  1099. /**
  1100. * ilookup5 - search for an inode in the inode cache
  1101. * @sb: super block of file system to search
  1102. * @hashval: hash value (usually inode number) to search for
  1103. * @test: callback used for comparisons between inodes
  1104. * @data: opaque data pointer to pass to @test
  1105. *
  1106. * Search for the inode specified by @hashval and @data in the inode cache,
  1107. * and if the inode is in the cache, return the inode with an incremented
  1108. * reference count. Waits on I_NEW before returning the inode.
  1109. * returned with an incremented reference count.
  1110. *
  1111. * This is a generalized version of ilookup() for file systems where the
  1112. * inode number is not sufficient for unique identification of an inode.
  1113. *
  1114. * Note: @test is called with the inode_hash_lock held, so can't sleep.
  1115. */
  1116. struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
  1117. int (*test)(struct inode *, void *), void *data)
  1118. {
  1119. struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
  1120. if (inode)
  1121. wait_on_inode(inode);
  1122. return inode;
  1123. }
  1124. EXPORT_SYMBOL(ilookup5);
  1125. /**
  1126. * ilookup - search for an inode in the inode cache
  1127. * @sb: super block of file system to search
  1128. * @ino: inode number to search for
  1129. *
  1130. * Search for the inode @ino in the inode cache, and if the inode is in the
  1131. * cache, the inode is returned with an incremented reference count.
  1132. */
  1133. struct inode *ilookup(struct super_block *sb, unsigned long ino)
  1134. {
  1135. struct hlist_head *head = inode_hashtable + hash(sb, ino);
  1136. struct inode *inode;
  1137. spin_lock(&inode_hash_lock);
  1138. inode = find_inode_fast(sb, head, ino);
  1139. spin_unlock(&inode_hash_lock);
  1140. if (inode)
  1141. wait_on_inode(inode);
  1142. return inode;
  1143. }
  1144. EXPORT_SYMBOL(ilookup);
  1145. int insert_inode_locked(struct inode *inode)
  1146. {
  1147. struct super_block *sb = inode->i_sb;
  1148. ino_t ino = inode->i_ino;
  1149. struct hlist_head *head = inode_hashtable + hash(sb, ino);
  1150. while (1) {
  1151. struct inode *old = NULL;
  1152. spin_lock(&inode_hash_lock);
  1153. hlist_for_each_entry(old, head, i_hash) {
  1154. if (old->i_ino != ino)
  1155. continue;
  1156. if (old->i_sb != sb)
  1157. continue;
  1158. spin_lock(&old->i_lock);
  1159. if (old->i_state & (I_FREEING|I_WILL_FREE)) {
  1160. spin_unlock(&old->i_lock);
  1161. continue;
  1162. }
  1163. break;
  1164. }
  1165. if (likely(!old)) {
  1166. spin_lock(&inode->i_lock);
  1167. inode->i_state |= I_NEW;
  1168. hlist_add_head(&inode->i_hash, head);
  1169. spin_unlock(&inode->i_lock);
  1170. spin_unlock(&inode_hash_lock);
  1171. return 0;
  1172. }
  1173. __iget(old);
  1174. spin_unlock(&old->i_lock);
  1175. spin_unlock(&inode_hash_lock);
  1176. wait_on_inode(old);
  1177. if (unlikely(!inode_unhashed(old))) {
  1178. iput(old);
  1179. return -EBUSY;
  1180. }
  1181. iput(old);
  1182. }
  1183. }
  1184. EXPORT_SYMBOL(insert_inode_locked);
  1185. int insert_inode_locked4(struct inode *inode, unsigned long hashval,
  1186. int (*test)(struct inode *, void *), void *data)
  1187. {
  1188. struct super_block *sb = inode->i_sb;
  1189. struct hlist_head *head = inode_hashtable + hash(sb, hashval);
  1190. while (1) {
  1191. struct inode *old = NULL;
  1192. spin_lock(&inode_hash_lock);
  1193. hlist_for_each_entry(old, head, i_hash) {
  1194. if (old->i_sb != sb)
  1195. continue;
  1196. if (!test(old, data))
  1197. continue;
  1198. spin_lock(&old->i_lock);
  1199. if (old->i_state & (I_FREEING|I_WILL_FREE)) {
  1200. spin_unlock(&old->i_lock);
  1201. continue;
  1202. }
  1203. break;
  1204. }
  1205. if (likely(!old)) {
  1206. spin_lock(&inode->i_lock);
  1207. inode->i_state |= I_NEW;
  1208. hlist_add_head(&inode->i_hash, head);
  1209. spin_unlock(&inode->i_lock);
  1210. spin_unlock(&inode_hash_lock);
  1211. return 0;
  1212. }
  1213. __iget(old);
  1214. spin_unlock(&old->i_lock);
  1215. spin_unlock(&inode_hash_lock);
  1216. wait_on_inode(old);
  1217. if (unlikely(!inode_unhashed(old))) {
  1218. iput(old);
  1219. return -EBUSY;
  1220. }
  1221. iput(old);
  1222. }
  1223. }
  1224. EXPORT_SYMBOL(insert_inode_locked4);
  1225. int generic_delete_inode(struct inode *inode)
  1226. {
  1227. return 1;
  1228. }
  1229. EXPORT_SYMBOL(generic_delete_inode);
  1230. /*
  1231. * Called when we're dropping the last reference
  1232. * to an inode.
  1233. *
  1234. * Call the FS "drop_inode()" function, defaulting to
  1235. * the legacy UNIX filesystem behaviour. If it tells
  1236. * us to evict inode, do so. Otherwise, retain inode
  1237. * in cache if fs is alive, sync and evict if fs is
  1238. * shutting down.
  1239. */
  1240. static void iput_final(struct inode *inode)
  1241. {
  1242. struct super_block *sb = inode->i_sb;
  1243. const struct super_operations *op = inode->i_sb->s_op;
  1244. int drop;
  1245. WARN_ON(inode->i_state & I_NEW);
  1246. if (op->drop_inode)
  1247. drop = op->drop_inode(inode);
  1248. else
  1249. drop = generic_drop_inode(inode);
  1250. if (!drop && (sb->s_flags & MS_ACTIVE)) {
  1251. inode->i_state |= I_REFERENCED;
  1252. inode_add_lru(inode);
  1253. spin_unlock(&inode->i_lock);
  1254. return;
  1255. }
  1256. if (!drop) {
  1257. inode->i_state |= I_WILL_FREE;
  1258. spin_unlock(&inode->i_lock);
  1259. write_inode_now(inode, 1);
  1260. spin_lock(&inode->i_lock);
  1261. WARN_ON(inode->i_state & I_NEW);
  1262. inode->i_state &= ~I_WILL_FREE;
  1263. }
  1264. inode->i_state |= I_FREEING;
  1265. if (!list_empty(&inode->i_lru))
  1266. inode_lru_list_del(inode);
  1267. spin_unlock(&inode->i_lock);
  1268. evict(inode);
  1269. }
  1270. /**
  1271. * iput - put an inode
  1272. * @inode: inode to put
  1273. *
  1274. * Puts an inode, dropping its usage count. If the inode use count hits
  1275. * zero, the inode is then freed and may also be destroyed.
  1276. *
  1277. * Consequently, iput() can sleep.
  1278. */
  1279. void iput(struct inode *inode)
  1280. {
  1281. if (inode) {
  1282. BUG_ON(inode->i_state & I_CLEAR);
  1283. if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
  1284. iput_final(inode);
  1285. }
  1286. }
  1287. EXPORT_SYMBOL(iput);
  1288. /**
  1289. * bmap - find a block number in a file
  1290. * @inode: inode of file
  1291. * @block: block to find
  1292. *
  1293. * Returns the block number on the device holding the inode that
  1294. * is the disk block number for the block of the file requested.
  1295. * That is, asked for block 4 of inode 1 the function will return the
  1296. * disk block relative to the disk start that holds that block of the
  1297. * file.
  1298. */
  1299. sector_t bmap(struct inode *inode, sector_t block)
  1300. {
  1301. sector_t res = 0;
  1302. if (inode->i_mapping->a_ops->bmap)
  1303. res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
  1304. return res;
  1305. }
  1306. EXPORT_SYMBOL(bmap);
  1307. /*
  1308. * With relative atime, only update atime if the previous atime is
  1309. * earlier than either the ctime or mtime or if at least a day has
  1310. * passed since the last atime update.
  1311. */
  1312. static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
  1313. struct timespec now)
  1314. {
  1315. if (!(mnt->mnt_flags & MNT_RELATIME))
  1316. return 1;
  1317. /*
  1318. * Is mtime younger than atime? If yes, update atime:
  1319. */
  1320. if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
  1321. return 1;
  1322. /*
  1323. * Is ctime younger than atime? If yes, update atime:
  1324. */
  1325. if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
  1326. return 1;
  1327. /*
  1328. * Is the previous atime value older than a day? If yes,
  1329. * update atime:
  1330. */
  1331. if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
  1332. return 1;
  1333. /*
  1334. * Good, we can skip the atime update:
  1335. */
  1336. return 0;
  1337. }
  1338. /*
  1339. * This does the actual work of updating an inodes time or version. Must have
  1340. * had called mnt_want_write() before calling this.
  1341. */
  1342. static int update_time(struct inode *inode, struct timespec *time, int flags)
  1343. {
  1344. if (inode->i_op->update_time)
  1345. return inode->i_op->update_time(inode, time, flags);
  1346. if (flags & S_ATIME)
  1347. inode->i_atime = *time;
  1348. if (flags & S_VERSION)
  1349. inode_inc_iversion(inode);
  1350. if (flags & S_CTIME)
  1351. inode->i_ctime = *time;
  1352. if (flags & S_MTIME)
  1353. inode->i_mtime = *time;
  1354. mark_inode_dirty_sync(inode);
  1355. return 0;
  1356. }
  1357. /**
  1358. * touch_atime - update the access time
  1359. * @path: the &struct path to update
  1360. *
  1361. * Update the accessed time on an inode and mark it for writeback.
  1362. * This function automatically handles read only file systems and media,
  1363. * as well as the "noatime" flag and inode specific "noatime" markers.
  1364. */
  1365. void touch_atime(const struct path *path)
  1366. {
  1367. struct vfsmount *mnt = path->mnt;
  1368. struct inode *inode = path->dentry->d_inode;
  1369. struct timespec now;
  1370. if (inode->i_flags & S_NOATIME)
  1371. return;
  1372. if (IS_NOATIME(inode))
  1373. return;
  1374. if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
  1375. return;
  1376. if (mnt->mnt_flags & MNT_NOATIME)
  1377. return;
  1378. if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
  1379. return;
  1380. now = current_fs_time(inode->i_sb);
  1381. if (!relatime_need_update(mnt, inode, now))
  1382. return;
  1383. if (timespec_equal(&inode->i_atime, &now))
  1384. return;
  1385. if (!sb_start_write_trylock(inode->i_sb))
  1386. return;
  1387. if (__mnt_want_write(mnt))
  1388. goto skip_update;
  1389. /*
  1390. * File systems can error out when updating inodes if they need to
  1391. * allocate new space to modify an inode (such is the case for
  1392. * Btrfs), but since we touch atime while walking down the path we
  1393. * really don't care if we failed to update the atime of the file,
  1394. * so just ignore the return value.
  1395. * We may also fail on filesystems that have the ability to make parts
  1396. * of the fs read only, e.g. subvolumes in Btrfs.
  1397. */
  1398. update_time(inode, &now, S_ATIME);
  1399. __mnt_drop_write(mnt);
  1400. skip_update:
  1401. sb_end_write(inode->i_sb);
  1402. }
  1403. EXPORT_SYMBOL(touch_atime);
  1404. /*
  1405. * The logic we want is
  1406. *
  1407. * if suid or (sgid and xgrp)
  1408. * remove privs
  1409. */
  1410. int should_remove_suid(struct dentry *dentry)
  1411. {
  1412. umode_t mode = dentry->d_inode->i_mode;
  1413. int kill = 0;
  1414. /* suid always must be killed */
  1415. if (unlikely(mode & S_ISUID))
  1416. kill = ATTR_KILL_SUID;
  1417. /*
  1418. * sgid without any exec bits is just a mandatory locking mark; leave
  1419. * it alone. If some exec bits are set, it's a real sgid; kill it.
  1420. */
  1421. if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
  1422. kill |= ATTR_KILL_SGID;
  1423. if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
  1424. return kill;
  1425. return 0;
  1426. }
  1427. EXPORT_SYMBOL(should_remove_suid);
  1428. static int __remove_suid(struct dentry *dentry, int kill)
  1429. {
  1430. struct iattr newattrs;
  1431. newattrs.ia_valid = ATTR_FORCE | kill;
  1432. /*
  1433. * Note we call this on write, so notify_change will not
  1434. * encounter any conflicting delegations:
  1435. */
  1436. return notify_change(dentry, &newattrs, NULL);
  1437. }
  1438. int file_remove_suid(struct file *file)
  1439. {
  1440. struct dentry *dentry = file->f_path.dentry;
  1441. struct inode *inode = dentry->d_inode;
  1442. int killsuid;
  1443. int killpriv;
  1444. int error = 0;
  1445. /* Fast path for nothing security related */
  1446. if (IS_NOSEC(inode))
  1447. return 0;
  1448. killsuid = should_remove_suid(dentry);
  1449. killpriv = security_inode_need_killpriv(dentry);
  1450. if (killpriv < 0)
  1451. return killpriv;
  1452. if (killpriv)
  1453. error = security_inode_killpriv(dentry);
  1454. if (!error && killsuid)
  1455. error = __remove_suid(dentry, killsuid);
  1456. if (!error && (inode->i_sb->s_flags & MS_NOSEC))
  1457. inode->i_flags |= S_NOSEC;
  1458. return error;
  1459. }
  1460. EXPORT_SYMBOL(file_remove_suid);
  1461. /**
  1462. * file_update_time - update mtime and ctime time
  1463. * @file: file accessed
  1464. *
  1465. * Update the mtime and ctime members of an inode and mark the inode
  1466. * for writeback. Note that this function is meant exclusively for
  1467. * usage in the file write path of filesystems, and filesystems may
  1468. * choose to explicitly ignore update via this function with the
  1469. * S_NOCMTIME inode flag, e.g. for network filesystem where these
  1470. * timestamps are handled by the server. This can return an error for
  1471. * file systems who need to allocate space in order to update an inode.
  1472. */
  1473. int file_update_time(struct file *file)
  1474. {
  1475. struct inode *inode = file_inode(file);
  1476. struct timespec now;
  1477. int sync_it = 0;
  1478. int ret;
  1479. /* First try to exhaust all avenues to not sync */
  1480. if (IS_NOCMTIME(inode))
  1481. return 0;
  1482. now = current_fs_time(inode->i_sb);
  1483. if (!timespec_equal(&inode->i_mtime, &now))
  1484. sync_it = S_MTIME;
  1485. if (!timespec_equal(&inode->i_ctime, &now))
  1486. sync_it |= S_CTIME;
  1487. if (IS_I_VERSION(inode))
  1488. sync_it |= S_VERSION;
  1489. if (!sync_it)
  1490. return 0;
  1491. /* Finally allowed to write? Takes lock. */
  1492. if (__mnt_want_write_file(file))
  1493. return 0;
  1494. ret = update_time(inode, &now, sync_it);
  1495. __mnt_drop_write_file(file);
  1496. return ret;
  1497. }
  1498. EXPORT_SYMBOL(file_update_time);
  1499. int inode_needs_sync(struct inode *inode)
  1500. {
  1501. if (IS_SYNC(inode))
  1502. return 1;
  1503. if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
  1504. return 1;
  1505. return 0;
  1506. }
  1507. EXPORT_SYMBOL(inode_needs_sync);
  1508. int inode_wait(void *word)
  1509. {
  1510. schedule();
  1511. return 0;
  1512. }
  1513. EXPORT_SYMBOL(inode_wait);
  1514. /*
  1515. * If we try to find an inode in the inode hash while it is being
  1516. * deleted, we have to wait until the filesystem completes its
  1517. * deletion before reporting that it isn't found. This function waits
  1518. * until the deletion _might_ have completed. Callers are responsible
  1519. * to recheck inode state.
  1520. *
  1521. * It doesn't matter if I_NEW is not set initially, a call to
  1522. * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
  1523. * will DTRT.
  1524. */
  1525. static void __wait_on_freeing_inode(struct inode *inode)
  1526. {
  1527. wait_queue_head_t *wq;
  1528. DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
  1529. wq = bit_waitqueue(&inode->i_state, __I_NEW);
  1530. prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
  1531. spin_unlock(&inode->i_lock);
  1532. spin_unlock(&inode_hash_lock);
  1533. schedule();
  1534. finish_wait(wq, &wait.wait);
  1535. spin_lock(&inode_hash_lock);
  1536. }
  1537. static __initdata unsigned long ihash_entries;
  1538. static int __init set_ihash_entries(char *str)
  1539. {
  1540. if (!str)
  1541. return 0;
  1542. ihash_entries = simple_strtoul(str, &str, 0);
  1543. return 1;
  1544. }
  1545. __setup("ihash_entries=", set_ihash_entries);
  1546. /*
  1547. * Initialize the waitqueues and inode hash table.
  1548. */
  1549. void __init inode_init_early(void)
  1550. {
  1551. unsigned int loop;
  1552. /* If hashes are distributed across NUMA nodes, defer
  1553. * hash allocation until vmalloc space is available.
  1554. */
  1555. if (hashdist)
  1556. return;
  1557. inode_hashtable =
  1558. alloc_large_system_hash("Inode-cache",
  1559. sizeof(struct hlist_head),
  1560. ihash_entries,
  1561. 14,
  1562. HASH_EARLY,
  1563. &i_hash_shift,
  1564. &i_hash_mask,
  1565. 0,
  1566. 0);
  1567. for (loop = 0; loop < (1U << i_hash_shift); loop++)
  1568. INIT_HLIST_HEAD(&inode_hashtable[loop]);
  1569. }
  1570. void __init inode_init(void)
  1571. {
  1572. unsigned int loop;
  1573. /* inode slab cache */
  1574. inode_cachep = kmem_cache_create("inode_cache",
  1575. sizeof(struct inode),
  1576. 0,
  1577. (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
  1578. SLAB_MEM_SPREAD),
  1579. init_once);
  1580. /* Hash may have been set up in inode_init_early */
  1581. if (!hashdist)
  1582. return;
  1583. inode_hashtable =
  1584. alloc_large_system_hash("Inode-cache",
  1585. sizeof(struct hlist_head),
  1586. ihash_entries,
  1587. 14,
  1588. 0,
  1589. &i_hash_shift,
  1590. &i_hash_mask,
  1591. 0,
  1592. 0);
  1593. for (loop = 0; loop < (1U << i_hash_shift); loop++)
  1594. INIT_HLIST_HEAD(&inode_hashtable[loop]);
  1595. }
  1596. void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
  1597. {
  1598. inode->i_mode = mode;
  1599. if (S_ISCHR(mode)) {
  1600. inode->i_fop = &def_chr_fops;
  1601. inode->i_rdev = rdev;
  1602. } else if (S_ISBLK(mode)) {
  1603. inode->i_fop = &def_blk_fops;
  1604. inode->i_rdev = rdev;
  1605. } else if (S_ISFIFO(mode))
  1606. inode->i_fop = &pipefifo_fops;
  1607. else if (S_ISSOCK(mode))
  1608. inode->i_fop = &bad_sock_fops;
  1609. else
  1610. printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
  1611. " inode %s:%lu\n", mode, inode->i_sb->s_id,
  1612. inode->i_ino);
  1613. }
  1614. EXPORT_SYMBOL(init_special_inode);
  1615. /**
  1616. * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
  1617. * @inode: New inode
  1618. * @dir: Directory inode
  1619. * @mode: mode of the new inode
  1620. */
  1621. void inode_init_owner(struct inode *inode, const struct inode *dir,
  1622. umode_t mode)
  1623. {
  1624. inode->i_uid = current_fsuid();
  1625. if (dir && dir->i_mode & S_ISGID) {
  1626. inode->i_gid = dir->i_gid;
  1627. if (S_ISDIR(mode))
  1628. mode |= S_ISGID;
  1629. } else
  1630. inode->i_gid = current_fsgid();
  1631. inode->i_mode = mode;
  1632. }
  1633. EXPORT_SYMBOL(inode_init_owner);
  1634. /**
  1635. * inode_owner_or_capable - check current task permissions to inode
  1636. * @inode: inode being checked
  1637. *
  1638. * Return true if current either has CAP_FOWNER to the inode, or
  1639. * owns the file.
  1640. */
  1641. bool inode_owner_or_capable(const struct inode *inode)
  1642. {
  1643. if (uid_eq(current_fsuid(), inode->i_uid))
  1644. return true;
  1645. if (inode_capable(inode, CAP_FOWNER))
  1646. return true;
  1647. return false;
  1648. }
  1649. EXPORT_SYMBOL(inode_owner_or_capable);
  1650. /*
  1651. * Direct i/o helper functions
  1652. */
  1653. static void __inode_dio_wait(struct inode *inode)
  1654. {
  1655. wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
  1656. DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
  1657. do {
  1658. prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
  1659. if (atomic_read(&inode->i_dio_count))
  1660. schedule();
  1661. } while (atomic_read(&inode->i_dio_count));
  1662. finish_wait(wq, &q.wait);
  1663. }
  1664. /**
  1665. * inode_dio_wait - wait for outstanding DIO requests to finish
  1666. * @inode: inode to wait for
  1667. *
  1668. * Waits for all pending direct I/O requests to finish so that we can
  1669. * proceed with a truncate or equivalent operation.
  1670. *
  1671. * Must be called under a lock that serializes taking new references
  1672. * to i_dio_count, usually by inode->i_mutex.
  1673. */
  1674. void inode_dio_wait(struct inode *inode)
  1675. {
  1676. if (atomic_read(&inode->i_dio_count))
  1677. __inode_dio_wait(inode);
  1678. }
  1679. EXPORT_SYMBOL(inode_dio_wait);
  1680. /*
  1681. * inode_dio_done - signal finish of a direct I/O requests
  1682. * @inode: inode the direct I/O happens on
  1683. *
  1684. * This is called once we've finished processing a direct I/O request,
  1685. * and is used to wake up callers waiting for direct I/O to be quiesced.
  1686. */
  1687. void inode_dio_done(struct inode *inode)
  1688. {
  1689. if (atomic_dec_and_test(&inode->i_dio_count))
  1690. wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
  1691. }
  1692. EXPORT_SYMBOL(inode_dio_done);