dquot.c 56 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999
  1. /*
  2. * Implementation of the diskquota system for the LINUX operating system. QUOTA
  3. * is implemented using the BSD system call interface as the means of
  4. * communication with the user level. This file contains the generic routines
  5. * called by the different filesystems on allocation of an inode or block.
  6. * These routines take care of the administration needed to have a consistent
  7. * diskquota tracking system. The ideas of both user and group quotas are based
  8. * on the Melbourne quota system as used on BSD derived systems. The internal
  9. * implementation is based on one of the several variants of the LINUX
  10. * inode-subsystem with added complexity of the diskquota system.
  11. *
  12. * Version: $Id: dquot.c,v 6.3 1996/11/17 18:35:34 mvw Exp mvw $
  13. *
  14. * Author: Marco van Wieringen <mvw@planets.elm.net>
  15. *
  16. * Fixes: Dmitry Gorodchanin <pgmdsg@ibi.com>, 11 Feb 96
  17. *
  18. * Revised list management to avoid races
  19. * -- Bill Hawes, <whawes@star.net>, 9/98
  20. *
  21. * Fixed races in dquot_transfer(), dqget() and dquot_alloc_...().
  22. * As the consequence the locking was moved from dquot_decr_...(),
  23. * dquot_incr_...() to calling functions.
  24. * invalidate_dquots() now writes modified dquots.
  25. * Serialized quota_off() and quota_on() for mount point.
  26. * Fixed a few bugs in grow_dquots().
  27. * Fixed deadlock in write_dquot() - we no longer account quotas on
  28. * quota files
  29. * remove_dquot_ref() moved to inode.c - it now traverses through inodes
  30. * add_dquot_ref() restarts after blocking
  31. * Added check for bogus uid and fixed check for group in quotactl.
  32. * Jan Kara, <jack@suse.cz>, sponsored by SuSE CR, 10-11/99
  33. *
  34. * Used struct list_head instead of own list struct
  35. * Invalidation of referenced dquots is no longer possible
  36. * Improved free_dquots list management
  37. * Quota and i_blocks are now updated in one place to avoid races
  38. * Warnings are now delayed so we won't block in critical section
  39. * Write updated not to require dquot lock
  40. * Jan Kara, <jack@suse.cz>, 9/2000
  41. *
  42. * Added dynamic quota structure allocation
  43. * Jan Kara <jack@suse.cz> 12/2000
  44. *
  45. * Rewritten quota interface. Implemented new quota format and
  46. * formats registering.
  47. * Jan Kara, <jack@suse.cz>, 2001,2002
  48. *
  49. * New SMP locking.
  50. * Jan Kara, <jack@suse.cz>, 10/2002
  51. *
  52. * Added journalled quota support, fix lock inversion problems
  53. * Jan Kara, <jack@suse.cz>, 2003,2004
  54. *
  55. * (C) Copyright 1994 - 1997 Marco van Wieringen
  56. */
  57. #include <linux/errno.h>
  58. #include <linux/kernel.h>
  59. #include <linux/fs.h>
  60. #include <linux/mount.h>
  61. #include <linux/mm.h>
  62. #include <linux/time.h>
  63. #include <linux/types.h>
  64. #include <linux/string.h>
  65. #include <linux/fcntl.h>
  66. #include <linux/stat.h>
  67. #include <linux/tty.h>
  68. #include <linux/file.h>
  69. #include <linux/slab.h>
  70. #include <linux/sysctl.h>
  71. #include <linux/init.h>
  72. #include <linux/module.h>
  73. #include <linux/proc_fs.h>
  74. #include <linux/security.h>
  75. #include <linux/kmod.h>
  76. #include <linux/namei.h>
  77. #include <linux/buffer_head.h>
  78. #include <linux/capability.h>
  79. #include <linux/quotaops.h>
  80. #include <linux/writeback.h> /* for inode_lock, oddly enough.. */
  81. #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
  82. #include <net/netlink.h>
  83. #include <net/genetlink.h>
  84. #endif
  85. #include <asm/uaccess.h>
  86. #define __DQUOT_PARANOIA
  87. /*
  88. * There are two quota SMP locks. dq_list_lock protects all lists with quotas
  89. * and quota formats and also dqstats structure containing statistics about the
  90. * lists. dq_data_lock protects data from dq_dqb and also mem_dqinfo structures
  91. * and also guards consistency of dquot->dq_dqb with inode->i_blocks, i_bytes.
  92. * i_blocks and i_bytes updates itself are guarded by i_lock acquired directly
  93. * in inode_add_bytes() and inode_sub_bytes().
  94. *
  95. * The spinlock ordering is hence: dq_data_lock > dq_list_lock > i_lock
  96. *
  97. * Note that some things (eg. sb pointer, type, id) doesn't change during
  98. * the life of the dquot structure and so needn't to be protected by a lock
  99. *
  100. * Any operation working on dquots via inode pointers must hold dqptr_sem. If
  101. * operation is just reading pointers from inode (or not using them at all) the
  102. * read lock is enough. If pointers are altered function must hold write lock
  103. * (these locking rules also apply for S_NOQUOTA flag in the inode - note that
  104. * for altering the flag i_mutex is also needed). If operation is holding
  105. * reference to dquot in other way (e.g. quotactl ops) it must be guarded by
  106. * dqonoff_mutex.
  107. * This locking assures that:
  108. * a) update/access to dquot pointers in inode is serialized
  109. * b) everyone is guarded against invalidate_dquots()
  110. *
  111. * Each dquot has its dq_lock mutex. Locked dquots might not be referenced
  112. * from inodes (dquot_alloc_space() and such don't check the dq_lock).
  113. * Currently dquot is locked only when it is being read to memory (or space for
  114. * it is being allocated) on the first dqget() and when it is being released on
  115. * the last dqput(). The allocation and release oparations are serialized by
  116. * the dq_lock and by checking the use count in dquot_release(). Write
  117. * operations on dquots don't hold dq_lock as they copy data under dq_data_lock
  118. * spinlock to internal buffers before writing.
  119. *
  120. * Lock ordering (including related VFS locks) is the following:
  121. * i_mutex > dqonoff_sem > journal_lock > dqptr_sem > dquot->dq_lock >
  122. * dqio_mutex
  123. * i_mutex on quota files is special (it's below dqio_mutex)
  124. */
  125. static DEFINE_SPINLOCK(dq_list_lock);
  126. DEFINE_SPINLOCK(dq_data_lock);
  127. static char *quotatypes[] = INITQFNAMES;
  128. static struct quota_format_type *quota_formats; /* List of registered formats */
  129. static struct quota_module_name module_names[] = INIT_QUOTA_MODULE_NAMES;
  130. /* SLAB cache for dquot structures */
  131. static struct kmem_cache *dquot_cachep;
  132. int register_quota_format(struct quota_format_type *fmt)
  133. {
  134. spin_lock(&dq_list_lock);
  135. fmt->qf_next = quota_formats;
  136. quota_formats = fmt;
  137. spin_unlock(&dq_list_lock);
  138. return 0;
  139. }
  140. void unregister_quota_format(struct quota_format_type *fmt)
  141. {
  142. struct quota_format_type **actqf;
  143. spin_lock(&dq_list_lock);
  144. for (actqf = &quota_formats; *actqf && *actqf != fmt; actqf = &(*actqf)->qf_next);
  145. if (*actqf)
  146. *actqf = (*actqf)->qf_next;
  147. spin_unlock(&dq_list_lock);
  148. }
  149. static struct quota_format_type *find_quota_format(int id)
  150. {
  151. struct quota_format_type *actqf;
  152. spin_lock(&dq_list_lock);
  153. for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
  154. if (!actqf || !try_module_get(actqf->qf_owner)) {
  155. int qm;
  156. spin_unlock(&dq_list_lock);
  157. for (qm = 0; module_names[qm].qm_fmt_id && module_names[qm].qm_fmt_id != id; qm++);
  158. if (!module_names[qm].qm_fmt_id || request_module(module_names[qm].qm_mod_name))
  159. return NULL;
  160. spin_lock(&dq_list_lock);
  161. for (actqf = quota_formats; actqf && actqf->qf_fmt_id != id; actqf = actqf->qf_next);
  162. if (actqf && !try_module_get(actqf->qf_owner))
  163. actqf = NULL;
  164. }
  165. spin_unlock(&dq_list_lock);
  166. return actqf;
  167. }
  168. static void put_quota_format(struct quota_format_type *fmt)
  169. {
  170. module_put(fmt->qf_owner);
  171. }
  172. /*
  173. * Dquot List Management:
  174. * The quota code uses three lists for dquot management: the inuse_list,
  175. * free_dquots, and dquot_hash[] array. A single dquot structure may be
  176. * on all three lists, depending on its current state.
  177. *
  178. * All dquots are placed to the end of inuse_list when first created, and this
  179. * list is used for invalidate operation, which must look at every dquot.
  180. *
  181. * Unused dquots (dq_count == 0) are added to the free_dquots list when freed,
  182. * and this list is searched whenever we need an available dquot. Dquots are
  183. * removed from the list as soon as they are used again, and
  184. * dqstats.free_dquots gives the number of dquots on the list. When
  185. * dquot is invalidated it's completely released from memory.
  186. *
  187. * Dquots with a specific identity (device, type and id) are placed on
  188. * one of the dquot_hash[] hash chains. The provides an efficient search
  189. * mechanism to locate a specific dquot.
  190. */
  191. static LIST_HEAD(inuse_list);
  192. static LIST_HEAD(free_dquots);
  193. static unsigned int dq_hash_bits, dq_hash_mask;
  194. static struct hlist_head *dquot_hash;
  195. struct dqstats dqstats;
  196. static void dqput(struct dquot *dquot);
  197. static inline unsigned int
  198. hashfn(const struct super_block *sb, unsigned int id, int type)
  199. {
  200. unsigned long tmp;
  201. tmp = (((unsigned long)sb>>L1_CACHE_SHIFT) ^ id) * (MAXQUOTAS - type);
  202. return (tmp + (tmp >> dq_hash_bits)) & dq_hash_mask;
  203. }
  204. /*
  205. * Following list functions expect dq_list_lock to be held
  206. */
  207. static inline void insert_dquot_hash(struct dquot *dquot)
  208. {
  209. struct hlist_head *head = dquot_hash + hashfn(dquot->dq_sb, dquot->dq_id, dquot->dq_type);
  210. hlist_add_head(&dquot->dq_hash, head);
  211. }
  212. static inline void remove_dquot_hash(struct dquot *dquot)
  213. {
  214. hlist_del_init(&dquot->dq_hash);
  215. }
  216. static inline struct dquot *find_dquot(unsigned int hashent, struct super_block *sb, unsigned int id, int type)
  217. {
  218. struct hlist_node *node;
  219. struct dquot *dquot;
  220. hlist_for_each (node, dquot_hash+hashent) {
  221. dquot = hlist_entry(node, struct dquot, dq_hash);
  222. if (dquot->dq_sb == sb && dquot->dq_id == id && dquot->dq_type == type)
  223. return dquot;
  224. }
  225. return NODQUOT;
  226. }
  227. /* Add a dquot to the tail of the free list */
  228. static inline void put_dquot_last(struct dquot *dquot)
  229. {
  230. list_add_tail(&dquot->dq_free, &free_dquots);
  231. dqstats.free_dquots++;
  232. }
  233. static inline void remove_free_dquot(struct dquot *dquot)
  234. {
  235. if (list_empty(&dquot->dq_free))
  236. return;
  237. list_del_init(&dquot->dq_free);
  238. dqstats.free_dquots--;
  239. }
  240. static inline void put_inuse(struct dquot *dquot)
  241. {
  242. /* We add to the back of inuse list so we don't have to restart
  243. * when traversing this list and we block */
  244. list_add_tail(&dquot->dq_inuse, &inuse_list);
  245. dqstats.allocated_dquots++;
  246. }
  247. static inline void remove_inuse(struct dquot *dquot)
  248. {
  249. dqstats.allocated_dquots--;
  250. list_del(&dquot->dq_inuse);
  251. }
  252. /*
  253. * End of list functions needing dq_list_lock
  254. */
  255. static void wait_on_dquot(struct dquot *dquot)
  256. {
  257. mutex_lock(&dquot->dq_lock);
  258. mutex_unlock(&dquot->dq_lock);
  259. }
  260. #define mark_dquot_dirty(dquot) ((dquot)->dq_sb->dq_op->mark_dirty(dquot))
  261. int dquot_mark_dquot_dirty(struct dquot *dquot)
  262. {
  263. spin_lock(&dq_list_lock);
  264. if (!test_and_set_bit(DQ_MOD_B, &dquot->dq_flags))
  265. list_add(&dquot->dq_dirty, &sb_dqopt(dquot->dq_sb)->
  266. info[dquot->dq_type].dqi_dirty_list);
  267. spin_unlock(&dq_list_lock);
  268. return 0;
  269. }
  270. /* This function needs dq_list_lock */
  271. static inline int clear_dquot_dirty(struct dquot *dquot)
  272. {
  273. if (!test_and_clear_bit(DQ_MOD_B, &dquot->dq_flags))
  274. return 0;
  275. list_del_init(&dquot->dq_dirty);
  276. return 1;
  277. }
  278. void mark_info_dirty(struct super_block *sb, int type)
  279. {
  280. set_bit(DQF_INFO_DIRTY_B, &sb_dqopt(sb)->info[type].dqi_flags);
  281. }
  282. EXPORT_SYMBOL(mark_info_dirty);
  283. /*
  284. * Read dquot from disk and alloc space for it
  285. */
  286. int dquot_acquire(struct dquot *dquot)
  287. {
  288. int ret = 0, ret2 = 0;
  289. struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
  290. mutex_lock(&dquot->dq_lock);
  291. mutex_lock(&dqopt->dqio_mutex);
  292. if (!test_bit(DQ_READ_B, &dquot->dq_flags))
  293. ret = dqopt->ops[dquot->dq_type]->read_dqblk(dquot);
  294. if (ret < 0)
  295. goto out_iolock;
  296. set_bit(DQ_READ_B, &dquot->dq_flags);
  297. /* Instantiate dquot if needed */
  298. if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && !dquot->dq_off) {
  299. ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
  300. /* Write the info if needed */
  301. if (info_dirty(&dqopt->info[dquot->dq_type]))
  302. ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
  303. if (ret < 0)
  304. goto out_iolock;
  305. if (ret2 < 0) {
  306. ret = ret2;
  307. goto out_iolock;
  308. }
  309. }
  310. set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
  311. out_iolock:
  312. mutex_unlock(&dqopt->dqio_mutex);
  313. mutex_unlock(&dquot->dq_lock);
  314. return ret;
  315. }
  316. /*
  317. * Write dquot to disk
  318. */
  319. int dquot_commit(struct dquot *dquot)
  320. {
  321. int ret = 0, ret2 = 0;
  322. struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
  323. mutex_lock(&dqopt->dqio_mutex);
  324. spin_lock(&dq_list_lock);
  325. if (!clear_dquot_dirty(dquot)) {
  326. spin_unlock(&dq_list_lock);
  327. goto out_sem;
  328. }
  329. spin_unlock(&dq_list_lock);
  330. /* Inactive dquot can be only if there was error during read/init
  331. * => we have better not writing it */
  332. if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
  333. ret = dqopt->ops[dquot->dq_type]->commit_dqblk(dquot);
  334. if (info_dirty(&dqopt->info[dquot->dq_type]))
  335. ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
  336. if (ret >= 0)
  337. ret = ret2;
  338. }
  339. out_sem:
  340. mutex_unlock(&dqopt->dqio_mutex);
  341. return ret;
  342. }
  343. /*
  344. * Release dquot
  345. */
  346. int dquot_release(struct dquot *dquot)
  347. {
  348. int ret = 0, ret2 = 0;
  349. struct quota_info *dqopt = sb_dqopt(dquot->dq_sb);
  350. mutex_lock(&dquot->dq_lock);
  351. /* Check whether we are not racing with some other dqget() */
  352. if (atomic_read(&dquot->dq_count) > 1)
  353. goto out_dqlock;
  354. mutex_lock(&dqopt->dqio_mutex);
  355. if (dqopt->ops[dquot->dq_type]->release_dqblk) {
  356. ret = dqopt->ops[dquot->dq_type]->release_dqblk(dquot);
  357. /* Write the info */
  358. if (info_dirty(&dqopt->info[dquot->dq_type]))
  359. ret2 = dqopt->ops[dquot->dq_type]->write_file_info(dquot->dq_sb, dquot->dq_type);
  360. if (ret >= 0)
  361. ret = ret2;
  362. }
  363. clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
  364. mutex_unlock(&dqopt->dqio_mutex);
  365. out_dqlock:
  366. mutex_unlock(&dquot->dq_lock);
  367. return ret;
  368. }
  369. /* Invalidate all dquots on the list. Note that this function is called after
  370. * quota is disabled and pointers from inodes removed so there cannot be new
  371. * quota users. There can still be some users of quotas due to inodes being
  372. * just deleted or pruned by prune_icache() (those are not attached to any
  373. * list). We have to wait for such users.
  374. */
  375. static void invalidate_dquots(struct super_block *sb, int type)
  376. {
  377. struct dquot *dquot, *tmp;
  378. restart:
  379. spin_lock(&dq_list_lock);
  380. list_for_each_entry_safe(dquot, tmp, &inuse_list, dq_inuse) {
  381. if (dquot->dq_sb != sb)
  382. continue;
  383. if (dquot->dq_type != type)
  384. continue;
  385. /* Wait for dquot users */
  386. if (atomic_read(&dquot->dq_count)) {
  387. DEFINE_WAIT(wait);
  388. atomic_inc(&dquot->dq_count);
  389. prepare_to_wait(&dquot->dq_wait_unused, &wait,
  390. TASK_UNINTERRUPTIBLE);
  391. spin_unlock(&dq_list_lock);
  392. /* Once dqput() wakes us up, we know it's time to free
  393. * the dquot.
  394. * IMPORTANT: we rely on the fact that there is always
  395. * at most one process waiting for dquot to free.
  396. * Otherwise dq_count would be > 1 and we would never
  397. * wake up.
  398. */
  399. if (atomic_read(&dquot->dq_count) > 1)
  400. schedule();
  401. finish_wait(&dquot->dq_wait_unused, &wait);
  402. dqput(dquot);
  403. /* At this moment dquot() need not exist (it could be
  404. * reclaimed by prune_dqcache(). Hence we must
  405. * restart. */
  406. goto restart;
  407. }
  408. /*
  409. * Quota now has no users and it has been written on last
  410. * dqput()
  411. */
  412. remove_dquot_hash(dquot);
  413. remove_free_dquot(dquot);
  414. remove_inuse(dquot);
  415. kmem_cache_free(dquot_cachep, dquot);
  416. }
  417. spin_unlock(&dq_list_lock);
  418. }
  419. int vfs_quota_sync(struct super_block *sb, int type)
  420. {
  421. struct list_head *dirty;
  422. struct dquot *dquot;
  423. struct quota_info *dqopt = sb_dqopt(sb);
  424. int cnt;
  425. mutex_lock(&dqopt->dqonoff_mutex);
  426. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  427. if (type != -1 && cnt != type)
  428. continue;
  429. if (!sb_has_quota_enabled(sb, cnt))
  430. continue;
  431. spin_lock(&dq_list_lock);
  432. dirty = &dqopt->info[cnt].dqi_dirty_list;
  433. while (!list_empty(dirty)) {
  434. dquot = list_first_entry(dirty, struct dquot, dq_dirty);
  435. /* Dirty and inactive can be only bad dquot... */
  436. if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
  437. clear_dquot_dirty(dquot);
  438. continue;
  439. }
  440. /* Now we have active dquot from which someone is
  441. * holding reference so we can safely just increase
  442. * use count */
  443. atomic_inc(&dquot->dq_count);
  444. dqstats.lookups++;
  445. spin_unlock(&dq_list_lock);
  446. sb->dq_op->write_dquot(dquot);
  447. dqput(dquot);
  448. spin_lock(&dq_list_lock);
  449. }
  450. spin_unlock(&dq_list_lock);
  451. }
  452. for (cnt = 0; cnt < MAXQUOTAS; cnt++)
  453. if ((cnt == type || type == -1) && sb_has_quota_enabled(sb, cnt)
  454. && info_dirty(&dqopt->info[cnt]))
  455. sb->dq_op->write_info(sb, cnt);
  456. spin_lock(&dq_list_lock);
  457. dqstats.syncs++;
  458. spin_unlock(&dq_list_lock);
  459. mutex_unlock(&dqopt->dqonoff_mutex);
  460. return 0;
  461. }
  462. /* Free unused dquots from cache */
  463. static void prune_dqcache(int count)
  464. {
  465. struct list_head *head;
  466. struct dquot *dquot;
  467. head = free_dquots.prev;
  468. while (head != &free_dquots && count) {
  469. dquot = list_entry(head, struct dquot, dq_free);
  470. remove_dquot_hash(dquot);
  471. remove_free_dquot(dquot);
  472. remove_inuse(dquot);
  473. kmem_cache_free(dquot_cachep, dquot);
  474. count--;
  475. head = free_dquots.prev;
  476. }
  477. }
  478. /*
  479. * This is called from kswapd when we think we need some
  480. * more memory
  481. */
  482. static int shrink_dqcache_memory(int nr, gfp_t gfp_mask)
  483. {
  484. if (nr) {
  485. spin_lock(&dq_list_lock);
  486. prune_dqcache(nr);
  487. spin_unlock(&dq_list_lock);
  488. }
  489. return (dqstats.free_dquots / 100) * sysctl_vfs_cache_pressure;
  490. }
  491. static struct shrinker dqcache_shrinker = {
  492. .shrink = shrink_dqcache_memory,
  493. .seeks = DEFAULT_SEEKS,
  494. };
  495. /*
  496. * Put reference to dquot
  497. * NOTE: If you change this function please check whether dqput_blocks() works right...
  498. * MUST be called with either dqptr_sem or dqonoff_mutex held
  499. */
  500. static void dqput(struct dquot *dquot)
  501. {
  502. if (!dquot)
  503. return;
  504. #ifdef __DQUOT_PARANOIA
  505. if (!atomic_read(&dquot->dq_count)) {
  506. printk("VFS: dqput: trying to free free dquot\n");
  507. printk("VFS: device %s, dquot of %s %d\n",
  508. dquot->dq_sb->s_id,
  509. quotatypes[dquot->dq_type],
  510. dquot->dq_id);
  511. BUG();
  512. }
  513. #endif
  514. spin_lock(&dq_list_lock);
  515. dqstats.drops++;
  516. spin_unlock(&dq_list_lock);
  517. we_slept:
  518. spin_lock(&dq_list_lock);
  519. if (atomic_read(&dquot->dq_count) > 1) {
  520. /* We have more than one user... nothing to do */
  521. atomic_dec(&dquot->dq_count);
  522. /* Releasing dquot during quotaoff phase? */
  523. if (!sb_has_quota_enabled(dquot->dq_sb, dquot->dq_type) &&
  524. atomic_read(&dquot->dq_count) == 1)
  525. wake_up(&dquot->dq_wait_unused);
  526. spin_unlock(&dq_list_lock);
  527. return;
  528. }
  529. /* Need to release dquot? */
  530. if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && dquot_dirty(dquot)) {
  531. spin_unlock(&dq_list_lock);
  532. /* Commit dquot before releasing */
  533. dquot->dq_sb->dq_op->write_dquot(dquot);
  534. goto we_slept;
  535. }
  536. /* Clear flag in case dquot was inactive (something bad happened) */
  537. clear_dquot_dirty(dquot);
  538. if (test_bit(DQ_ACTIVE_B, &dquot->dq_flags)) {
  539. spin_unlock(&dq_list_lock);
  540. dquot->dq_sb->dq_op->release_dquot(dquot);
  541. goto we_slept;
  542. }
  543. atomic_dec(&dquot->dq_count);
  544. #ifdef __DQUOT_PARANOIA
  545. /* sanity check */
  546. BUG_ON(!list_empty(&dquot->dq_free));
  547. #endif
  548. put_dquot_last(dquot);
  549. spin_unlock(&dq_list_lock);
  550. }
  551. static struct dquot *get_empty_dquot(struct super_block *sb, int type)
  552. {
  553. struct dquot *dquot;
  554. dquot = kmem_cache_zalloc(dquot_cachep, GFP_NOFS);
  555. if(!dquot)
  556. return NODQUOT;
  557. mutex_init(&dquot->dq_lock);
  558. INIT_LIST_HEAD(&dquot->dq_free);
  559. INIT_LIST_HEAD(&dquot->dq_inuse);
  560. INIT_HLIST_NODE(&dquot->dq_hash);
  561. INIT_LIST_HEAD(&dquot->dq_dirty);
  562. init_waitqueue_head(&dquot->dq_wait_unused);
  563. dquot->dq_sb = sb;
  564. dquot->dq_type = type;
  565. atomic_set(&dquot->dq_count, 1);
  566. return dquot;
  567. }
  568. /*
  569. * Get reference to dquot
  570. * MUST be called with either dqptr_sem or dqonoff_mutex held
  571. */
  572. static struct dquot *dqget(struct super_block *sb, unsigned int id, int type)
  573. {
  574. unsigned int hashent = hashfn(sb, id, type);
  575. struct dquot *dquot, *empty = NODQUOT;
  576. if (!sb_has_quota_enabled(sb, type))
  577. return NODQUOT;
  578. we_slept:
  579. spin_lock(&dq_list_lock);
  580. if ((dquot = find_dquot(hashent, sb, id, type)) == NODQUOT) {
  581. if (empty == NODQUOT) {
  582. spin_unlock(&dq_list_lock);
  583. if ((empty = get_empty_dquot(sb, type)) == NODQUOT)
  584. schedule(); /* Try to wait for a moment... */
  585. goto we_slept;
  586. }
  587. dquot = empty;
  588. dquot->dq_id = id;
  589. /* all dquots go on the inuse_list */
  590. put_inuse(dquot);
  591. /* hash it first so it can be found */
  592. insert_dquot_hash(dquot);
  593. dqstats.lookups++;
  594. spin_unlock(&dq_list_lock);
  595. } else {
  596. if (!atomic_read(&dquot->dq_count))
  597. remove_free_dquot(dquot);
  598. atomic_inc(&dquot->dq_count);
  599. dqstats.cache_hits++;
  600. dqstats.lookups++;
  601. spin_unlock(&dq_list_lock);
  602. if (empty)
  603. kmem_cache_free(dquot_cachep, empty);
  604. }
  605. /* Wait for dq_lock - after this we know that either dquot_release() is already
  606. * finished or it will be canceled due to dq_count > 1 test */
  607. wait_on_dquot(dquot);
  608. /* Read the dquot and instantiate it (everything done only if needed) */
  609. if (!test_bit(DQ_ACTIVE_B, &dquot->dq_flags) && sb->dq_op->acquire_dquot(dquot) < 0) {
  610. dqput(dquot);
  611. return NODQUOT;
  612. }
  613. #ifdef __DQUOT_PARANOIA
  614. BUG_ON(!dquot->dq_sb); /* Has somebody invalidated entry under us? */
  615. #endif
  616. return dquot;
  617. }
  618. static int dqinit_needed(struct inode *inode, int type)
  619. {
  620. int cnt;
  621. if (IS_NOQUOTA(inode))
  622. return 0;
  623. if (type != -1)
  624. return inode->i_dquot[type] == NODQUOT;
  625. for (cnt = 0; cnt < MAXQUOTAS; cnt++)
  626. if (inode->i_dquot[cnt] == NODQUOT)
  627. return 1;
  628. return 0;
  629. }
  630. /* This routine is guarded by dqonoff_mutex mutex */
  631. static void add_dquot_ref(struct super_block *sb, int type)
  632. {
  633. struct inode *inode;
  634. restart:
  635. spin_lock(&inode_lock);
  636. list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
  637. if (!atomic_read(&inode->i_writecount))
  638. continue;
  639. if (!dqinit_needed(inode, type))
  640. continue;
  641. if (inode->i_state & (I_FREEING|I_WILL_FREE))
  642. continue;
  643. __iget(inode);
  644. spin_unlock(&inode_lock);
  645. sb->dq_op->initialize(inode, type);
  646. iput(inode);
  647. /* As we may have blocked we had better restart... */
  648. goto restart;
  649. }
  650. spin_unlock(&inode_lock);
  651. }
  652. /* Return 0 if dqput() won't block (note that 1 doesn't necessarily mean blocking) */
  653. static inline int dqput_blocks(struct dquot *dquot)
  654. {
  655. if (atomic_read(&dquot->dq_count) <= 1)
  656. return 1;
  657. return 0;
  658. }
  659. /* Remove references to dquots from inode - add dquot to list for freeing if needed */
  660. /* We can't race with anybody because we hold dqptr_sem for writing... */
  661. static int remove_inode_dquot_ref(struct inode *inode, int type,
  662. struct list_head *tofree_head)
  663. {
  664. struct dquot *dquot = inode->i_dquot[type];
  665. inode->i_dquot[type] = NODQUOT;
  666. if (dquot != NODQUOT) {
  667. if (dqput_blocks(dquot)) {
  668. #ifdef __DQUOT_PARANOIA
  669. if (atomic_read(&dquot->dq_count) != 1)
  670. printk(KERN_WARNING "VFS: Adding dquot with dq_count %d to dispose list.\n", atomic_read(&dquot->dq_count));
  671. #endif
  672. spin_lock(&dq_list_lock);
  673. list_add(&dquot->dq_free, tofree_head); /* As dquot must have currently users it can't be on the free list... */
  674. spin_unlock(&dq_list_lock);
  675. return 1;
  676. }
  677. else
  678. dqput(dquot); /* We have guaranteed we won't block */
  679. }
  680. return 0;
  681. }
  682. /* Free list of dquots - called from inode.c */
  683. /* dquots are removed from inodes, no new references can be got so we are the only ones holding reference */
  684. static void put_dquot_list(struct list_head *tofree_head)
  685. {
  686. struct list_head *act_head;
  687. struct dquot *dquot;
  688. act_head = tofree_head->next;
  689. /* So now we have dquots on the list... Just free them */
  690. while (act_head != tofree_head) {
  691. dquot = list_entry(act_head, struct dquot, dq_free);
  692. act_head = act_head->next;
  693. list_del_init(&dquot->dq_free); /* Remove dquot from the list so we won't have problems... */
  694. dqput(dquot);
  695. }
  696. }
  697. static void remove_dquot_ref(struct super_block *sb, int type,
  698. struct list_head *tofree_head)
  699. {
  700. struct inode *inode;
  701. spin_lock(&inode_lock);
  702. list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
  703. if (!IS_NOQUOTA(inode))
  704. remove_inode_dquot_ref(inode, type, tofree_head);
  705. }
  706. spin_unlock(&inode_lock);
  707. }
  708. /* Gather all references from inodes and drop them */
  709. static void drop_dquot_ref(struct super_block *sb, int type)
  710. {
  711. LIST_HEAD(tofree_head);
  712. if (sb->dq_op) {
  713. down_write(&sb_dqopt(sb)->dqptr_sem);
  714. remove_dquot_ref(sb, type, &tofree_head);
  715. up_write(&sb_dqopt(sb)->dqptr_sem);
  716. put_dquot_list(&tofree_head);
  717. }
  718. }
  719. static inline void dquot_incr_inodes(struct dquot *dquot, unsigned long number)
  720. {
  721. dquot->dq_dqb.dqb_curinodes += number;
  722. }
  723. static inline void dquot_incr_space(struct dquot *dquot, qsize_t number)
  724. {
  725. dquot->dq_dqb.dqb_curspace += number;
  726. }
  727. static inline void dquot_decr_inodes(struct dquot *dquot, unsigned long number)
  728. {
  729. if (dquot->dq_dqb.dqb_curinodes > number)
  730. dquot->dq_dqb.dqb_curinodes -= number;
  731. else
  732. dquot->dq_dqb.dqb_curinodes = 0;
  733. if (dquot->dq_dqb.dqb_curinodes <= dquot->dq_dqb.dqb_isoftlimit)
  734. dquot->dq_dqb.dqb_itime = (time_t) 0;
  735. clear_bit(DQ_INODES_B, &dquot->dq_flags);
  736. }
  737. static inline void dquot_decr_space(struct dquot *dquot, qsize_t number)
  738. {
  739. if (dquot->dq_dqb.dqb_curspace > number)
  740. dquot->dq_dqb.dqb_curspace -= number;
  741. else
  742. dquot->dq_dqb.dqb_curspace = 0;
  743. if (toqb(dquot->dq_dqb.dqb_curspace) <= dquot->dq_dqb.dqb_bsoftlimit)
  744. dquot->dq_dqb.dqb_btime = (time_t) 0;
  745. clear_bit(DQ_BLKS_B, &dquot->dq_flags);
  746. }
  747. static int warning_issued(struct dquot *dquot, const int warntype)
  748. {
  749. int flag = (warntype == QUOTA_NL_BHARDWARN ||
  750. warntype == QUOTA_NL_BSOFTLONGWARN) ? DQ_BLKS_B :
  751. ((warntype == QUOTA_NL_IHARDWARN ||
  752. warntype == QUOTA_NL_ISOFTLONGWARN) ? DQ_INODES_B : 0);
  753. if (!flag)
  754. return 0;
  755. return test_and_set_bit(flag, &dquot->dq_flags);
  756. }
  757. #ifdef CONFIG_PRINT_QUOTA_WARNING
  758. static int flag_print_warnings = 1;
  759. static inline int need_print_warning(struct dquot *dquot)
  760. {
  761. if (!flag_print_warnings)
  762. return 0;
  763. switch (dquot->dq_type) {
  764. case USRQUOTA:
  765. return current->fsuid == dquot->dq_id;
  766. case GRPQUOTA:
  767. return in_group_p(dquot->dq_id);
  768. }
  769. return 0;
  770. }
  771. /* Print warning to user which exceeded quota */
  772. static void print_warning(struct dquot *dquot, const int warntype)
  773. {
  774. char *msg = NULL;
  775. struct tty_struct *tty;
  776. if (!need_print_warning(dquot))
  777. return;
  778. mutex_lock(&tty_mutex);
  779. tty = get_current_tty();
  780. if (!tty)
  781. goto out_lock;
  782. tty_write_message(tty, dquot->dq_sb->s_id);
  783. if (warntype == QUOTA_NL_ISOFTWARN || warntype == QUOTA_NL_BSOFTWARN)
  784. tty_write_message(tty, ": warning, ");
  785. else
  786. tty_write_message(tty, ": write failed, ");
  787. tty_write_message(tty, quotatypes[dquot->dq_type]);
  788. switch (warntype) {
  789. case QUOTA_NL_IHARDWARN:
  790. msg = " file limit reached.\r\n";
  791. break;
  792. case QUOTA_NL_ISOFTLONGWARN:
  793. msg = " file quota exceeded too long.\r\n";
  794. break;
  795. case QUOTA_NL_ISOFTWARN:
  796. msg = " file quota exceeded.\r\n";
  797. break;
  798. case QUOTA_NL_BHARDWARN:
  799. msg = " block limit reached.\r\n";
  800. break;
  801. case QUOTA_NL_BSOFTLONGWARN:
  802. msg = " block quota exceeded too long.\r\n";
  803. break;
  804. case QUOTA_NL_BSOFTWARN:
  805. msg = " block quota exceeded.\r\n";
  806. break;
  807. }
  808. tty_write_message(tty, msg);
  809. out_lock:
  810. mutex_unlock(&tty_mutex);
  811. }
  812. #endif
  813. #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
  814. /* Netlink family structure for quota */
  815. static struct genl_family quota_genl_family = {
  816. .id = GENL_ID_GENERATE,
  817. .hdrsize = 0,
  818. .name = "VFS_DQUOT",
  819. .version = 1,
  820. .maxattr = QUOTA_NL_A_MAX,
  821. };
  822. /* Send warning to userspace about user which exceeded quota */
  823. static void send_warning(const struct dquot *dquot, const char warntype)
  824. {
  825. static atomic_t seq;
  826. struct sk_buff *skb;
  827. void *msg_head;
  828. int ret;
  829. int msg_size = 4 * nla_total_size(sizeof(u32)) +
  830. 2 * nla_total_size(sizeof(u64));
  831. /* We have to allocate using GFP_NOFS as we are called from a
  832. * filesystem performing write and thus further recursion into
  833. * the fs to free some data could cause deadlocks. */
  834. skb = genlmsg_new(msg_size, GFP_NOFS);
  835. if (!skb) {
  836. printk(KERN_ERR
  837. "VFS: Not enough memory to send quota warning.\n");
  838. return;
  839. }
  840. msg_head = genlmsg_put(skb, 0, atomic_add_return(1, &seq),
  841. &quota_genl_family, 0, QUOTA_NL_C_WARNING);
  842. if (!msg_head) {
  843. printk(KERN_ERR
  844. "VFS: Cannot store netlink header in quota warning.\n");
  845. goto err_out;
  846. }
  847. ret = nla_put_u32(skb, QUOTA_NL_A_QTYPE, dquot->dq_type);
  848. if (ret)
  849. goto attr_err_out;
  850. ret = nla_put_u64(skb, QUOTA_NL_A_EXCESS_ID, dquot->dq_id);
  851. if (ret)
  852. goto attr_err_out;
  853. ret = nla_put_u32(skb, QUOTA_NL_A_WARNING, warntype);
  854. if (ret)
  855. goto attr_err_out;
  856. ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MAJOR,
  857. MAJOR(dquot->dq_sb->s_dev));
  858. if (ret)
  859. goto attr_err_out;
  860. ret = nla_put_u32(skb, QUOTA_NL_A_DEV_MINOR,
  861. MINOR(dquot->dq_sb->s_dev));
  862. if (ret)
  863. goto attr_err_out;
  864. ret = nla_put_u64(skb, QUOTA_NL_A_CAUSED_ID, current->user->uid);
  865. if (ret)
  866. goto attr_err_out;
  867. genlmsg_end(skb, msg_head);
  868. ret = genlmsg_multicast(skb, 0, quota_genl_family.id, GFP_NOFS);
  869. if (ret < 0 && ret != -ESRCH)
  870. printk(KERN_ERR
  871. "VFS: Failed to send notification message: %d\n", ret);
  872. return;
  873. attr_err_out:
  874. printk(KERN_ERR "VFS: Not enough space to compose quota message!\n");
  875. err_out:
  876. kfree_skb(skb);
  877. }
  878. #endif
  879. static inline void flush_warnings(struct dquot * const *dquots, char *warntype)
  880. {
  881. int i;
  882. for (i = 0; i < MAXQUOTAS; i++)
  883. if (dquots[i] != NODQUOT && warntype[i] != QUOTA_NL_NOWARN &&
  884. !warning_issued(dquots[i], warntype[i])) {
  885. #ifdef CONFIG_PRINT_QUOTA_WARNING
  886. print_warning(dquots[i], warntype[i]);
  887. #endif
  888. #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
  889. send_warning(dquots[i], warntype[i]);
  890. #endif
  891. }
  892. }
  893. static inline char ignore_hardlimit(struct dquot *dquot)
  894. {
  895. struct mem_dqinfo *info = &sb_dqopt(dquot->dq_sb)->info[dquot->dq_type];
  896. return capable(CAP_SYS_RESOURCE) &&
  897. (info->dqi_format->qf_fmt_id != QFMT_VFS_OLD || !(info->dqi_flags & V1_DQF_RSQUASH));
  898. }
  899. /* needs dq_data_lock */
  900. static int check_idq(struct dquot *dquot, ulong inodes, char *warntype)
  901. {
  902. *warntype = QUOTA_NL_NOWARN;
  903. if (inodes <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
  904. return QUOTA_OK;
  905. if (dquot->dq_dqb.dqb_ihardlimit &&
  906. (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_ihardlimit &&
  907. !ignore_hardlimit(dquot)) {
  908. *warntype = QUOTA_NL_IHARDWARN;
  909. return NO_QUOTA;
  910. }
  911. if (dquot->dq_dqb.dqb_isoftlimit &&
  912. (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
  913. dquot->dq_dqb.dqb_itime && get_seconds() >= dquot->dq_dqb.dqb_itime &&
  914. !ignore_hardlimit(dquot)) {
  915. *warntype = QUOTA_NL_ISOFTLONGWARN;
  916. return NO_QUOTA;
  917. }
  918. if (dquot->dq_dqb.dqb_isoftlimit &&
  919. (dquot->dq_dqb.dqb_curinodes + inodes) > dquot->dq_dqb.dqb_isoftlimit &&
  920. dquot->dq_dqb.dqb_itime == 0) {
  921. *warntype = QUOTA_NL_ISOFTWARN;
  922. dquot->dq_dqb.dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
  923. }
  924. return QUOTA_OK;
  925. }
  926. /* needs dq_data_lock */
  927. static int check_bdq(struct dquot *dquot, qsize_t space, int prealloc, char *warntype)
  928. {
  929. *warntype = QUOTA_NL_NOWARN;
  930. if (space <= 0 || test_bit(DQ_FAKE_B, &dquot->dq_flags))
  931. return QUOTA_OK;
  932. if (dquot->dq_dqb.dqb_bhardlimit &&
  933. toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bhardlimit &&
  934. !ignore_hardlimit(dquot)) {
  935. if (!prealloc)
  936. *warntype = QUOTA_NL_BHARDWARN;
  937. return NO_QUOTA;
  938. }
  939. if (dquot->dq_dqb.dqb_bsoftlimit &&
  940. toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
  941. dquot->dq_dqb.dqb_btime && get_seconds() >= dquot->dq_dqb.dqb_btime &&
  942. !ignore_hardlimit(dquot)) {
  943. if (!prealloc)
  944. *warntype = QUOTA_NL_BSOFTLONGWARN;
  945. return NO_QUOTA;
  946. }
  947. if (dquot->dq_dqb.dqb_bsoftlimit &&
  948. toqb(dquot->dq_dqb.dqb_curspace + space) > dquot->dq_dqb.dqb_bsoftlimit &&
  949. dquot->dq_dqb.dqb_btime == 0) {
  950. if (!prealloc) {
  951. *warntype = QUOTA_NL_BSOFTWARN;
  952. dquot->dq_dqb.dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
  953. }
  954. else
  955. /*
  956. * We don't allow preallocation to exceed softlimit so exceeding will
  957. * be always printed
  958. */
  959. return NO_QUOTA;
  960. }
  961. return QUOTA_OK;
  962. }
  963. /*
  964. * Initialize quota pointers in inode
  965. * Transaction must be started at entry
  966. */
  967. int dquot_initialize(struct inode *inode, int type)
  968. {
  969. unsigned int id = 0;
  970. int cnt, ret = 0;
  971. /* First test before acquiring mutex - solves deadlocks when we
  972. * re-enter the quota code and are already holding the mutex */
  973. if (IS_NOQUOTA(inode))
  974. return 0;
  975. down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
  976. /* Having dqptr_sem we know NOQUOTA flags can't be altered... */
  977. if (IS_NOQUOTA(inode))
  978. goto out_err;
  979. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  980. if (type != -1 && cnt != type)
  981. continue;
  982. if (inode->i_dquot[cnt] == NODQUOT) {
  983. switch (cnt) {
  984. case USRQUOTA:
  985. id = inode->i_uid;
  986. break;
  987. case GRPQUOTA:
  988. id = inode->i_gid;
  989. break;
  990. }
  991. inode->i_dquot[cnt] = dqget(inode->i_sb, id, cnt);
  992. }
  993. }
  994. out_err:
  995. up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
  996. return ret;
  997. }
  998. /*
  999. * Release all quotas referenced by inode
  1000. * Transaction must be started at an entry
  1001. */
  1002. int dquot_drop(struct inode *inode)
  1003. {
  1004. int cnt;
  1005. down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1006. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1007. if (inode->i_dquot[cnt] != NODQUOT) {
  1008. dqput(inode->i_dquot[cnt]);
  1009. inode->i_dquot[cnt] = NODQUOT;
  1010. }
  1011. }
  1012. up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1013. return 0;
  1014. }
  1015. /*
  1016. * Following four functions update i_blocks+i_bytes fields and
  1017. * quota information (together with appropriate checks)
  1018. * NOTE: We absolutely rely on the fact that caller dirties
  1019. * the inode (usually macros in quotaops.h care about this) and
  1020. * holds a handle for the current transaction so that dquot write and
  1021. * inode write go into the same transaction.
  1022. */
  1023. /*
  1024. * This operation can block, but only after everything is updated
  1025. */
  1026. int dquot_alloc_space(struct inode *inode, qsize_t number, int warn)
  1027. {
  1028. int cnt, ret = NO_QUOTA;
  1029. char warntype[MAXQUOTAS];
  1030. /* First test before acquiring mutex - solves deadlocks when we
  1031. * re-enter the quota code and are already holding the mutex */
  1032. if (IS_NOQUOTA(inode)) {
  1033. out_add:
  1034. inode_add_bytes(inode, number);
  1035. return QUOTA_OK;
  1036. }
  1037. for (cnt = 0; cnt < MAXQUOTAS; cnt++)
  1038. warntype[cnt] = QUOTA_NL_NOWARN;
  1039. down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1040. if (IS_NOQUOTA(inode)) { /* Now we can do reliable test... */
  1041. up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1042. goto out_add;
  1043. }
  1044. spin_lock(&dq_data_lock);
  1045. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1046. if (inode->i_dquot[cnt] == NODQUOT)
  1047. continue;
  1048. if (check_bdq(inode->i_dquot[cnt], number, warn, warntype+cnt) == NO_QUOTA)
  1049. goto warn_put_all;
  1050. }
  1051. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1052. if (inode->i_dquot[cnt] == NODQUOT)
  1053. continue;
  1054. dquot_incr_space(inode->i_dquot[cnt], number);
  1055. }
  1056. inode_add_bytes(inode, number);
  1057. ret = QUOTA_OK;
  1058. warn_put_all:
  1059. spin_unlock(&dq_data_lock);
  1060. if (ret == QUOTA_OK)
  1061. /* Dirtify all the dquots - this can block when journalling */
  1062. for (cnt = 0; cnt < MAXQUOTAS; cnt++)
  1063. if (inode->i_dquot[cnt])
  1064. mark_dquot_dirty(inode->i_dquot[cnt]);
  1065. flush_warnings(inode->i_dquot, warntype);
  1066. up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1067. return ret;
  1068. }
  1069. /*
  1070. * This operation can block, but only after everything is updated
  1071. */
  1072. int dquot_alloc_inode(const struct inode *inode, unsigned long number)
  1073. {
  1074. int cnt, ret = NO_QUOTA;
  1075. char warntype[MAXQUOTAS];
  1076. /* First test before acquiring mutex - solves deadlocks when we
  1077. * re-enter the quota code and are already holding the mutex */
  1078. if (IS_NOQUOTA(inode))
  1079. return QUOTA_OK;
  1080. for (cnt = 0; cnt < MAXQUOTAS; cnt++)
  1081. warntype[cnt] = QUOTA_NL_NOWARN;
  1082. down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1083. if (IS_NOQUOTA(inode)) {
  1084. up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1085. return QUOTA_OK;
  1086. }
  1087. spin_lock(&dq_data_lock);
  1088. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1089. if (inode->i_dquot[cnt] == NODQUOT)
  1090. continue;
  1091. if (check_idq(inode->i_dquot[cnt], number, warntype+cnt) == NO_QUOTA)
  1092. goto warn_put_all;
  1093. }
  1094. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1095. if (inode->i_dquot[cnt] == NODQUOT)
  1096. continue;
  1097. dquot_incr_inodes(inode->i_dquot[cnt], number);
  1098. }
  1099. ret = QUOTA_OK;
  1100. warn_put_all:
  1101. spin_unlock(&dq_data_lock);
  1102. if (ret == QUOTA_OK)
  1103. /* Dirtify all the dquots - this can block when journalling */
  1104. for (cnt = 0; cnt < MAXQUOTAS; cnt++)
  1105. if (inode->i_dquot[cnt])
  1106. mark_dquot_dirty(inode->i_dquot[cnt]);
  1107. flush_warnings(inode->i_dquot, warntype);
  1108. up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1109. return ret;
  1110. }
  1111. /*
  1112. * This operation can block, but only after everything is updated
  1113. */
  1114. int dquot_free_space(struct inode *inode, qsize_t number)
  1115. {
  1116. unsigned int cnt;
  1117. /* First test before acquiring mutex - solves deadlocks when we
  1118. * re-enter the quota code and are already holding the mutex */
  1119. if (IS_NOQUOTA(inode)) {
  1120. out_sub:
  1121. inode_sub_bytes(inode, number);
  1122. return QUOTA_OK;
  1123. }
  1124. down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1125. /* Now recheck reliably when holding dqptr_sem */
  1126. if (IS_NOQUOTA(inode)) {
  1127. up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1128. goto out_sub;
  1129. }
  1130. spin_lock(&dq_data_lock);
  1131. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1132. if (inode->i_dquot[cnt] == NODQUOT)
  1133. continue;
  1134. dquot_decr_space(inode->i_dquot[cnt], number);
  1135. }
  1136. inode_sub_bytes(inode, number);
  1137. spin_unlock(&dq_data_lock);
  1138. /* Dirtify all the dquots - this can block when journalling */
  1139. for (cnt = 0; cnt < MAXQUOTAS; cnt++)
  1140. if (inode->i_dquot[cnt])
  1141. mark_dquot_dirty(inode->i_dquot[cnt]);
  1142. up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1143. return QUOTA_OK;
  1144. }
  1145. /*
  1146. * This operation can block, but only after everything is updated
  1147. */
  1148. int dquot_free_inode(const struct inode *inode, unsigned long number)
  1149. {
  1150. unsigned int cnt;
  1151. /* First test before acquiring mutex - solves deadlocks when we
  1152. * re-enter the quota code and are already holding the mutex */
  1153. if (IS_NOQUOTA(inode))
  1154. return QUOTA_OK;
  1155. down_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1156. /* Now recheck reliably when holding dqptr_sem */
  1157. if (IS_NOQUOTA(inode)) {
  1158. up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1159. return QUOTA_OK;
  1160. }
  1161. spin_lock(&dq_data_lock);
  1162. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1163. if (inode->i_dquot[cnt] == NODQUOT)
  1164. continue;
  1165. dquot_decr_inodes(inode->i_dquot[cnt], number);
  1166. }
  1167. spin_unlock(&dq_data_lock);
  1168. /* Dirtify all the dquots - this can block when journalling */
  1169. for (cnt = 0; cnt < MAXQUOTAS; cnt++)
  1170. if (inode->i_dquot[cnt])
  1171. mark_dquot_dirty(inode->i_dquot[cnt]);
  1172. up_read(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1173. return QUOTA_OK;
  1174. }
  1175. /*
  1176. * Transfer the number of inode and blocks from one diskquota to an other.
  1177. *
  1178. * This operation can block, but only after everything is updated
  1179. * A transaction must be started when entering this function.
  1180. */
  1181. int dquot_transfer(struct inode *inode, struct iattr *iattr)
  1182. {
  1183. qsize_t space;
  1184. struct dquot *transfer_from[MAXQUOTAS];
  1185. struct dquot *transfer_to[MAXQUOTAS];
  1186. int cnt, ret = NO_QUOTA, chuid = (iattr->ia_valid & ATTR_UID) && inode->i_uid != iattr->ia_uid,
  1187. chgid = (iattr->ia_valid & ATTR_GID) && inode->i_gid != iattr->ia_gid;
  1188. char warntype[MAXQUOTAS];
  1189. /* First test before acquiring mutex - solves deadlocks when we
  1190. * re-enter the quota code and are already holding the mutex */
  1191. if (IS_NOQUOTA(inode))
  1192. return QUOTA_OK;
  1193. /* Clear the arrays */
  1194. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1195. transfer_to[cnt] = transfer_from[cnt] = NODQUOT;
  1196. warntype[cnt] = QUOTA_NL_NOWARN;
  1197. }
  1198. down_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1199. /* Now recheck reliably when holding dqptr_sem */
  1200. if (IS_NOQUOTA(inode)) { /* File without quota accounting? */
  1201. up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1202. return QUOTA_OK;
  1203. }
  1204. /* First build the transfer_to list - here we can block on
  1205. * reading/instantiating of dquots. We know that the transaction for
  1206. * us was already started so we don't violate lock ranking here */
  1207. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1208. switch (cnt) {
  1209. case USRQUOTA:
  1210. if (!chuid)
  1211. continue;
  1212. transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_uid, cnt);
  1213. break;
  1214. case GRPQUOTA:
  1215. if (!chgid)
  1216. continue;
  1217. transfer_to[cnt] = dqget(inode->i_sb, iattr->ia_gid, cnt);
  1218. break;
  1219. }
  1220. }
  1221. spin_lock(&dq_data_lock);
  1222. space = inode_get_bytes(inode);
  1223. /* Build the transfer_from list and check the limits */
  1224. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1225. if (transfer_to[cnt] == NODQUOT)
  1226. continue;
  1227. transfer_from[cnt] = inode->i_dquot[cnt];
  1228. if (check_idq(transfer_to[cnt], 1, warntype+cnt) == NO_QUOTA ||
  1229. check_bdq(transfer_to[cnt], space, 0, warntype+cnt) == NO_QUOTA)
  1230. goto warn_put_all;
  1231. }
  1232. /*
  1233. * Finally perform the needed transfer from transfer_from to transfer_to
  1234. */
  1235. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1236. /*
  1237. * Skip changes for same uid or gid or for turned off quota-type.
  1238. */
  1239. if (transfer_to[cnt] == NODQUOT)
  1240. continue;
  1241. /* Due to IO error we might not have transfer_from[] structure */
  1242. if (transfer_from[cnt]) {
  1243. dquot_decr_inodes(transfer_from[cnt], 1);
  1244. dquot_decr_space(transfer_from[cnt], space);
  1245. }
  1246. dquot_incr_inodes(transfer_to[cnt], 1);
  1247. dquot_incr_space(transfer_to[cnt], space);
  1248. inode->i_dquot[cnt] = transfer_to[cnt];
  1249. }
  1250. ret = QUOTA_OK;
  1251. warn_put_all:
  1252. spin_unlock(&dq_data_lock);
  1253. /* Dirtify all the dquots - this can block when journalling */
  1254. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1255. if (transfer_from[cnt])
  1256. mark_dquot_dirty(transfer_from[cnt]);
  1257. if (transfer_to[cnt])
  1258. mark_dquot_dirty(transfer_to[cnt]);
  1259. }
  1260. flush_warnings(transfer_to, warntype);
  1261. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1262. if (ret == QUOTA_OK && transfer_from[cnt] != NODQUOT)
  1263. dqput(transfer_from[cnt]);
  1264. if (ret == NO_QUOTA && transfer_to[cnt] != NODQUOT)
  1265. dqput(transfer_to[cnt]);
  1266. }
  1267. up_write(&sb_dqopt(inode->i_sb)->dqptr_sem);
  1268. return ret;
  1269. }
  1270. /*
  1271. * Write info of quota file to disk
  1272. */
  1273. int dquot_commit_info(struct super_block *sb, int type)
  1274. {
  1275. int ret;
  1276. struct quota_info *dqopt = sb_dqopt(sb);
  1277. mutex_lock(&dqopt->dqio_mutex);
  1278. ret = dqopt->ops[type]->write_file_info(sb, type);
  1279. mutex_unlock(&dqopt->dqio_mutex);
  1280. return ret;
  1281. }
  1282. /*
  1283. * Definitions of diskquota operations.
  1284. */
  1285. struct dquot_operations dquot_operations = {
  1286. .initialize = dquot_initialize,
  1287. .drop = dquot_drop,
  1288. .alloc_space = dquot_alloc_space,
  1289. .alloc_inode = dquot_alloc_inode,
  1290. .free_space = dquot_free_space,
  1291. .free_inode = dquot_free_inode,
  1292. .transfer = dquot_transfer,
  1293. .write_dquot = dquot_commit,
  1294. .acquire_dquot = dquot_acquire,
  1295. .release_dquot = dquot_release,
  1296. .mark_dirty = dquot_mark_dquot_dirty,
  1297. .write_info = dquot_commit_info
  1298. };
  1299. static inline void set_enable_flags(struct quota_info *dqopt, int type)
  1300. {
  1301. switch (type) {
  1302. case USRQUOTA:
  1303. dqopt->flags |= DQUOT_USR_ENABLED;
  1304. break;
  1305. case GRPQUOTA:
  1306. dqopt->flags |= DQUOT_GRP_ENABLED;
  1307. break;
  1308. }
  1309. }
  1310. static inline void reset_enable_flags(struct quota_info *dqopt, int type)
  1311. {
  1312. switch (type) {
  1313. case USRQUOTA:
  1314. dqopt->flags &= ~DQUOT_USR_ENABLED;
  1315. break;
  1316. case GRPQUOTA:
  1317. dqopt->flags &= ~DQUOT_GRP_ENABLED;
  1318. break;
  1319. }
  1320. }
  1321. /*
  1322. * Turn quota off on a device. type == -1 ==> quotaoff for all types (umount)
  1323. */
  1324. int vfs_quota_off(struct super_block *sb, int type)
  1325. {
  1326. int cnt;
  1327. struct quota_info *dqopt = sb_dqopt(sb);
  1328. struct inode *toputinode[MAXQUOTAS];
  1329. /* We need to serialize quota_off() for device */
  1330. mutex_lock(&dqopt->dqonoff_mutex);
  1331. for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
  1332. toputinode[cnt] = NULL;
  1333. if (type != -1 && cnt != type)
  1334. continue;
  1335. if (!sb_has_quota_enabled(sb, cnt))
  1336. continue;
  1337. reset_enable_flags(dqopt, cnt);
  1338. /* Note: these are blocking operations */
  1339. drop_dquot_ref(sb, cnt);
  1340. invalidate_dquots(sb, cnt);
  1341. /*
  1342. * Now all dquots should be invalidated, all writes done so we should be only
  1343. * users of the info. No locks needed.
  1344. */
  1345. if (info_dirty(&dqopt->info[cnt]))
  1346. sb->dq_op->write_info(sb, cnt);
  1347. if (dqopt->ops[cnt]->free_file_info)
  1348. dqopt->ops[cnt]->free_file_info(sb, cnt);
  1349. put_quota_format(dqopt->info[cnt].dqi_format);
  1350. toputinode[cnt] = dqopt->files[cnt];
  1351. dqopt->files[cnt] = NULL;
  1352. dqopt->info[cnt].dqi_flags = 0;
  1353. dqopt->info[cnt].dqi_igrace = 0;
  1354. dqopt->info[cnt].dqi_bgrace = 0;
  1355. dqopt->ops[cnt] = NULL;
  1356. }
  1357. mutex_unlock(&dqopt->dqonoff_mutex);
  1358. /* Sync the superblock so that buffers with quota data are written to
  1359. * disk (and so userspace sees correct data afterwards). */
  1360. if (sb->s_op->sync_fs)
  1361. sb->s_op->sync_fs(sb, 1);
  1362. sync_blockdev(sb->s_bdev);
  1363. /* Now the quota files are just ordinary files and we can set the
  1364. * inode flags back. Moreover we discard the pagecache so that
  1365. * userspace sees the writes we did bypassing the pagecache. We
  1366. * must also discard the blockdev buffers so that we see the
  1367. * changes done by userspace on the next quotaon() */
  1368. for (cnt = 0; cnt < MAXQUOTAS; cnt++)
  1369. if (toputinode[cnt]) {
  1370. mutex_lock(&dqopt->dqonoff_mutex);
  1371. /* If quota was reenabled in the meantime, we have
  1372. * nothing to do */
  1373. if (!sb_has_quota_enabled(sb, cnt)) {
  1374. mutex_lock_nested(&toputinode[cnt]->i_mutex, I_MUTEX_QUOTA);
  1375. toputinode[cnt]->i_flags &= ~(S_IMMUTABLE |
  1376. S_NOATIME | S_NOQUOTA);
  1377. truncate_inode_pages(&toputinode[cnt]->i_data, 0);
  1378. mutex_unlock(&toputinode[cnt]->i_mutex);
  1379. mark_inode_dirty(toputinode[cnt]);
  1380. iput(toputinode[cnt]);
  1381. }
  1382. mutex_unlock(&dqopt->dqonoff_mutex);
  1383. }
  1384. if (sb->s_bdev)
  1385. invalidate_bdev(sb->s_bdev);
  1386. return 0;
  1387. }
  1388. /*
  1389. * Turn quotas on on a device
  1390. */
  1391. /* Helper function when we already have the inode */
  1392. static int vfs_quota_on_inode(struct inode *inode, int type, int format_id)
  1393. {
  1394. struct quota_format_type *fmt = find_quota_format(format_id);
  1395. struct super_block *sb = inode->i_sb;
  1396. struct quota_info *dqopt = sb_dqopt(sb);
  1397. int error;
  1398. int oldflags = -1;
  1399. if (!fmt)
  1400. return -ESRCH;
  1401. if (!S_ISREG(inode->i_mode)) {
  1402. error = -EACCES;
  1403. goto out_fmt;
  1404. }
  1405. if (IS_RDONLY(inode)) {
  1406. error = -EROFS;
  1407. goto out_fmt;
  1408. }
  1409. if (!sb->s_op->quota_write || !sb->s_op->quota_read) {
  1410. error = -EINVAL;
  1411. goto out_fmt;
  1412. }
  1413. /* As we bypass the pagecache we must now flush the inode so that
  1414. * we see all the changes from userspace... */
  1415. write_inode_now(inode, 1);
  1416. /* And now flush the block cache so that kernel sees the changes */
  1417. invalidate_bdev(sb->s_bdev);
  1418. mutex_lock(&inode->i_mutex);
  1419. mutex_lock(&dqopt->dqonoff_mutex);
  1420. if (sb_has_quota_enabled(sb, type)) {
  1421. error = -EBUSY;
  1422. goto out_lock;
  1423. }
  1424. /* We don't want quota and atime on quota files (deadlocks possible)
  1425. * Also nobody should write to the file - we use special IO operations
  1426. * which ignore the immutable bit. */
  1427. down_write(&dqopt->dqptr_sem);
  1428. oldflags = inode->i_flags & (S_NOATIME | S_IMMUTABLE | S_NOQUOTA);
  1429. inode->i_flags |= S_NOQUOTA | S_NOATIME | S_IMMUTABLE;
  1430. up_write(&dqopt->dqptr_sem);
  1431. sb->dq_op->drop(inode);
  1432. error = -EIO;
  1433. dqopt->files[type] = igrab(inode);
  1434. if (!dqopt->files[type])
  1435. goto out_lock;
  1436. error = -EINVAL;
  1437. if (!fmt->qf_ops->check_quota_file(sb, type))
  1438. goto out_file_init;
  1439. dqopt->ops[type] = fmt->qf_ops;
  1440. dqopt->info[type].dqi_format = fmt;
  1441. INIT_LIST_HEAD(&dqopt->info[type].dqi_dirty_list);
  1442. mutex_lock(&dqopt->dqio_mutex);
  1443. if ((error = dqopt->ops[type]->read_file_info(sb, type)) < 0) {
  1444. mutex_unlock(&dqopt->dqio_mutex);
  1445. goto out_file_init;
  1446. }
  1447. mutex_unlock(&dqopt->dqio_mutex);
  1448. mutex_unlock(&inode->i_mutex);
  1449. set_enable_flags(dqopt, type);
  1450. add_dquot_ref(sb, type);
  1451. mutex_unlock(&dqopt->dqonoff_mutex);
  1452. return 0;
  1453. out_file_init:
  1454. dqopt->files[type] = NULL;
  1455. iput(inode);
  1456. out_lock:
  1457. mutex_unlock(&dqopt->dqonoff_mutex);
  1458. if (oldflags != -1) {
  1459. down_write(&dqopt->dqptr_sem);
  1460. /* Set the flags back (in the case of accidental quotaon()
  1461. * on a wrong file we don't want to mess up the flags) */
  1462. inode->i_flags &= ~(S_NOATIME | S_NOQUOTA | S_IMMUTABLE);
  1463. inode->i_flags |= oldflags;
  1464. up_write(&dqopt->dqptr_sem);
  1465. }
  1466. mutex_unlock(&inode->i_mutex);
  1467. out_fmt:
  1468. put_quota_format(fmt);
  1469. return error;
  1470. }
  1471. /* Actual function called from quotactl() */
  1472. int vfs_quota_on(struct super_block *sb, int type, int format_id, char *path)
  1473. {
  1474. struct nameidata nd;
  1475. int error;
  1476. error = path_lookup(path, LOOKUP_FOLLOW, &nd);
  1477. if (error < 0)
  1478. return error;
  1479. error = security_quota_on(nd.dentry);
  1480. if (error)
  1481. goto out_path;
  1482. /* Quota file not on the same filesystem? */
  1483. if (nd.mnt->mnt_sb != sb)
  1484. error = -EXDEV;
  1485. else
  1486. error = vfs_quota_on_inode(nd.dentry->d_inode, type, format_id);
  1487. out_path:
  1488. path_release(&nd);
  1489. return error;
  1490. }
  1491. /*
  1492. * This function is used when filesystem needs to initialize quotas
  1493. * during mount time.
  1494. */
  1495. int vfs_quota_on_mount(struct super_block *sb, char *qf_name,
  1496. int format_id, int type)
  1497. {
  1498. struct dentry *dentry;
  1499. int error;
  1500. dentry = lookup_one_len(qf_name, sb->s_root, strlen(qf_name));
  1501. if (IS_ERR(dentry))
  1502. return PTR_ERR(dentry);
  1503. if (!dentry->d_inode) {
  1504. error = -ENOENT;
  1505. goto out;
  1506. }
  1507. error = security_quota_on(dentry);
  1508. if (!error)
  1509. error = vfs_quota_on_inode(dentry->d_inode, type, format_id);
  1510. out:
  1511. dput(dentry);
  1512. return error;
  1513. }
  1514. /* Generic routine for getting common part of quota structure */
  1515. static void do_get_dqblk(struct dquot *dquot, struct if_dqblk *di)
  1516. {
  1517. struct mem_dqblk *dm = &dquot->dq_dqb;
  1518. spin_lock(&dq_data_lock);
  1519. di->dqb_bhardlimit = dm->dqb_bhardlimit;
  1520. di->dqb_bsoftlimit = dm->dqb_bsoftlimit;
  1521. di->dqb_curspace = dm->dqb_curspace;
  1522. di->dqb_ihardlimit = dm->dqb_ihardlimit;
  1523. di->dqb_isoftlimit = dm->dqb_isoftlimit;
  1524. di->dqb_curinodes = dm->dqb_curinodes;
  1525. di->dqb_btime = dm->dqb_btime;
  1526. di->dqb_itime = dm->dqb_itime;
  1527. di->dqb_valid = QIF_ALL;
  1528. spin_unlock(&dq_data_lock);
  1529. }
  1530. int vfs_get_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
  1531. {
  1532. struct dquot *dquot;
  1533. mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
  1534. if (!(dquot = dqget(sb, id, type))) {
  1535. mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
  1536. return -ESRCH;
  1537. }
  1538. do_get_dqblk(dquot, di);
  1539. dqput(dquot);
  1540. mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
  1541. return 0;
  1542. }
  1543. /* Generic routine for setting common part of quota structure */
  1544. static void do_set_dqblk(struct dquot *dquot, struct if_dqblk *di)
  1545. {
  1546. struct mem_dqblk *dm = &dquot->dq_dqb;
  1547. int check_blim = 0, check_ilim = 0;
  1548. spin_lock(&dq_data_lock);
  1549. if (di->dqb_valid & QIF_SPACE) {
  1550. dm->dqb_curspace = di->dqb_curspace;
  1551. check_blim = 1;
  1552. }
  1553. if (di->dqb_valid & QIF_BLIMITS) {
  1554. dm->dqb_bsoftlimit = di->dqb_bsoftlimit;
  1555. dm->dqb_bhardlimit = di->dqb_bhardlimit;
  1556. check_blim = 1;
  1557. }
  1558. if (di->dqb_valid & QIF_INODES) {
  1559. dm->dqb_curinodes = di->dqb_curinodes;
  1560. check_ilim = 1;
  1561. }
  1562. if (di->dqb_valid & QIF_ILIMITS) {
  1563. dm->dqb_isoftlimit = di->dqb_isoftlimit;
  1564. dm->dqb_ihardlimit = di->dqb_ihardlimit;
  1565. check_ilim = 1;
  1566. }
  1567. if (di->dqb_valid & QIF_BTIME)
  1568. dm->dqb_btime = di->dqb_btime;
  1569. if (di->dqb_valid & QIF_ITIME)
  1570. dm->dqb_itime = di->dqb_itime;
  1571. if (check_blim) {
  1572. if (!dm->dqb_bsoftlimit || toqb(dm->dqb_curspace) < dm->dqb_bsoftlimit) {
  1573. dm->dqb_btime = 0;
  1574. clear_bit(DQ_BLKS_B, &dquot->dq_flags);
  1575. }
  1576. else if (!(di->dqb_valid & QIF_BTIME)) /* Set grace only if user hasn't provided his own... */
  1577. dm->dqb_btime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_bgrace;
  1578. }
  1579. if (check_ilim) {
  1580. if (!dm->dqb_isoftlimit || dm->dqb_curinodes < dm->dqb_isoftlimit) {
  1581. dm->dqb_itime = 0;
  1582. clear_bit(DQ_INODES_B, &dquot->dq_flags);
  1583. }
  1584. else if (!(di->dqb_valid & QIF_ITIME)) /* Set grace only if user hasn't provided his own... */
  1585. dm->dqb_itime = get_seconds() + sb_dqopt(dquot->dq_sb)->info[dquot->dq_type].dqi_igrace;
  1586. }
  1587. if (dm->dqb_bhardlimit || dm->dqb_bsoftlimit || dm->dqb_ihardlimit || dm->dqb_isoftlimit)
  1588. clear_bit(DQ_FAKE_B, &dquot->dq_flags);
  1589. else
  1590. set_bit(DQ_FAKE_B, &dquot->dq_flags);
  1591. spin_unlock(&dq_data_lock);
  1592. mark_dquot_dirty(dquot);
  1593. }
  1594. int vfs_set_dqblk(struct super_block *sb, int type, qid_t id, struct if_dqblk *di)
  1595. {
  1596. struct dquot *dquot;
  1597. mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
  1598. if (!(dquot = dqget(sb, id, type))) {
  1599. mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
  1600. return -ESRCH;
  1601. }
  1602. do_set_dqblk(dquot, di);
  1603. dqput(dquot);
  1604. mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
  1605. return 0;
  1606. }
  1607. /* Generic routine for getting common part of quota file information */
  1608. int vfs_get_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
  1609. {
  1610. struct mem_dqinfo *mi;
  1611. mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
  1612. if (!sb_has_quota_enabled(sb, type)) {
  1613. mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
  1614. return -ESRCH;
  1615. }
  1616. mi = sb_dqopt(sb)->info + type;
  1617. spin_lock(&dq_data_lock);
  1618. ii->dqi_bgrace = mi->dqi_bgrace;
  1619. ii->dqi_igrace = mi->dqi_igrace;
  1620. ii->dqi_flags = mi->dqi_flags & DQF_MASK;
  1621. ii->dqi_valid = IIF_ALL;
  1622. spin_unlock(&dq_data_lock);
  1623. mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
  1624. return 0;
  1625. }
  1626. /* Generic routine for setting common part of quota file information */
  1627. int vfs_set_dqinfo(struct super_block *sb, int type, struct if_dqinfo *ii)
  1628. {
  1629. struct mem_dqinfo *mi;
  1630. mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
  1631. if (!sb_has_quota_enabled(sb, type)) {
  1632. mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
  1633. return -ESRCH;
  1634. }
  1635. mi = sb_dqopt(sb)->info + type;
  1636. spin_lock(&dq_data_lock);
  1637. if (ii->dqi_valid & IIF_BGRACE)
  1638. mi->dqi_bgrace = ii->dqi_bgrace;
  1639. if (ii->dqi_valid & IIF_IGRACE)
  1640. mi->dqi_igrace = ii->dqi_igrace;
  1641. if (ii->dqi_valid & IIF_FLAGS)
  1642. mi->dqi_flags = (mi->dqi_flags & ~DQF_MASK) | (ii->dqi_flags & DQF_MASK);
  1643. spin_unlock(&dq_data_lock);
  1644. mark_info_dirty(sb, type);
  1645. /* Force write to disk */
  1646. sb->dq_op->write_info(sb, type);
  1647. mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
  1648. return 0;
  1649. }
  1650. struct quotactl_ops vfs_quotactl_ops = {
  1651. .quota_on = vfs_quota_on,
  1652. .quota_off = vfs_quota_off,
  1653. .quota_sync = vfs_quota_sync,
  1654. .get_info = vfs_get_dqinfo,
  1655. .set_info = vfs_set_dqinfo,
  1656. .get_dqblk = vfs_get_dqblk,
  1657. .set_dqblk = vfs_set_dqblk
  1658. };
  1659. static ctl_table fs_dqstats_table[] = {
  1660. {
  1661. .ctl_name = FS_DQ_LOOKUPS,
  1662. .procname = "lookups",
  1663. .data = &dqstats.lookups,
  1664. .maxlen = sizeof(int),
  1665. .mode = 0444,
  1666. .proc_handler = &proc_dointvec,
  1667. },
  1668. {
  1669. .ctl_name = FS_DQ_DROPS,
  1670. .procname = "drops",
  1671. .data = &dqstats.drops,
  1672. .maxlen = sizeof(int),
  1673. .mode = 0444,
  1674. .proc_handler = &proc_dointvec,
  1675. },
  1676. {
  1677. .ctl_name = FS_DQ_READS,
  1678. .procname = "reads",
  1679. .data = &dqstats.reads,
  1680. .maxlen = sizeof(int),
  1681. .mode = 0444,
  1682. .proc_handler = &proc_dointvec,
  1683. },
  1684. {
  1685. .ctl_name = FS_DQ_WRITES,
  1686. .procname = "writes",
  1687. .data = &dqstats.writes,
  1688. .maxlen = sizeof(int),
  1689. .mode = 0444,
  1690. .proc_handler = &proc_dointvec,
  1691. },
  1692. {
  1693. .ctl_name = FS_DQ_CACHE_HITS,
  1694. .procname = "cache_hits",
  1695. .data = &dqstats.cache_hits,
  1696. .maxlen = sizeof(int),
  1697. .mode = 0444,
  1698. .proc_handler = &proc_dointvec,
  1699. },
  1700. {
  1701. .ctl_name = FS_DQ_ALLOCATED,
  1702. .procname = "allocated_dquots",
  1703. .data = &dqstats.allocated_dquots,
  1704. .maxlen = sizeof(int),
  1705. .mode = 0444,
  1706. .proc_handler = &proc_dointvec,
  1707. },
  1708. {
  1709. .ctl_name = FS_DQ_FREE,
  1710. .procname = "free_dquots",
  1711. .data = &dqstats.free_dquots,
  1712. .maxlen = sizeof(int),
  1713. .mode = 0444,
  1714. .proc_handler = &proc_dointvec,
  1715. },
  1716. {
  1717. .ctl_name = FS_DQ_SYNCS,
  1718. .procname = "syncs",
  1719. .data = &dqstats.syncs,
  1720. .maxlen = sizeof(int),
  1721. .mode = 0444,
  1722. .proc_handler = &proc_dointvec,
  1723. },
  1724. #ifdef CONFIG_PRINT_QUOTA_WARNING
  1725. {
  1726. .ctl_name = FS_DQ_WARNINGS,
  1727. .procname = "warnings",
  1728. .data = &flag_print_warnings,
  1729. .maxlen = sizeof(int),
  1730. .mode = 0644,
  1731. .proc_handler = &proc_dointvec,
  1732. },
  1733. #endif
  1734. { .ctl_name = 0 },
  1735. };
  1736. static ctl_table fs_table[] = {
  1737. {
  1738. .ctl_name = FS_DQSTATS,
  1739. .procname = "quota",
  1740. .mode = 0555,
  1741. .child = fs_dqstats_table,
  1742. },
  1743. { .ctl_name = 0 },
  1744. };
  1745. static ctl_table sys_table[] = {
  1746. {
  1747. .ctl_name = CTL_FS,
  1748. .procname = "fs",
  1749. .mode = 0555,
  1750. .child = fs_table,
  1751. },
  1752. { .ctl_name = 0 },
  1753. };
  1754. static int __init dquot_init(void)
  1755. {
  1756. int i;
  1757. unsigned long nr_hash, order;
  1758. printk(KERN_NOTICE "VFS: Disk quotas %s\n", __DQUOT_VERSION__);
  1759. register_sysctl_table(sys_table);
  1760. dquot_cachep = kmem_cache_create("dquot",
  1761. sizeof(struct dquot), sizeof(unsigned long) * 4,
  1762. (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
  1763. SLAB_MEM_SPREAD|SLAB_PANIC),
  1764. NULL);
  1765. order = 0;
  1766. dquot_hash = (struct hlist_head *)__get_free_pages(GFP_ATOMIC, order);
  1767. if (!dquot_hash)
  1768. panic("Cannot create dquot hash table");
  1769. /* Find power-of-two hlist_heads which can fit into allocation */
  1770. nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct hlist_head);
  1771. dq_hash_bits = 0;
  1772. do {
  1773. dq_hash_bits++;
  1774. } while (nr_hash >> dq_hash_bits);
  1775. dq_hash_bits--;
  1776. nr_hash = 1UL << dq_hash_bits;
  1777. dq_hash_mask = nr_hash - 1;
  1778. for (i = 0; i < nr_hash; i++)
  1779. INIT_HLIST_HEAD(dquot_hash + i);
  1780. printk("Dquot-cache hash table entries: %ld (order %ld, %ld bytes)\n",
  1781. nr_hash, order, (PAGE_SIZE << order));
  1782. register_shrinker(&dqcache_shrinker);
  1783. #ifdef CONFIG_QUOTA_NETLINK_INTERFACE
  1784. if (genl_register_family(&quota_genl_family) != 0)
  1785. printk(KERN_ERR "VFS: Failed to create quota netlink interface.\n");
  1786. #endif
  1787. return 0;
  1788. }
  1789. module_init(dquot_init);
  1790. EXPORT_SYMBOL(register_quota_format);
  1791. EXPORT_SYMBOL(unregister_quota_format);
  1792. EXPORT_SYMBOL(dqstats);
  1793. EXPORT_SYMBOL(dq_data_lock);
  1794. EXPORT_SYMBOL(vfs_quota_on);
  1795. EXPORT_SYMBOL(vfs_quota_on_mount);
  1796. EXPORT_SYMBOL(vfs_quota_off);
  1797. EXPORT_SYMBOL(vfs_quota_sync);
  1798. EXPORT_SYMBOL(vfs_get_dqinfo);
  1799. EXPORT_SYMBOL(vfs_set_dqinfo);
  1800. EXPORT_SYMBOL(vfs_get_dqblk);
  1801. EXPORT_SYMBOL(vfs_set_dqblk);
  1802. EXPORT_SYMBOL(dquot_commit);
  1803. EXPORT_SYMBOL(dquot_commit_info);
  1804. EXPORT_SYMBOL(dquot_acquire);
  1805. EXPORT_SYMBOL(dquot_release);
  1806. EXPORT_SYMBOL(dquot_mark_dquot_dirty);
  1807. EXPORT_SYMBOL(dquot_initialize);
  1808. EXPORT_SYMBOL(dquot_drop);
  1809. EXPORT_SYMBOL(dquot_alloc_space);
  1810. EXPORT_SYMBOL(dquot_alloc_inode);
  1811. EXPORT_SYMBOL(dquot_free_space);
  1812. EXPORT_SYMBOL(dquot_free_inode);
  1813. EXPORT_SYMBOL(dquot_transfer);