namespace.c 65 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742
  1. /*
  2. * linux/fs/namespace.c
  3. *
  4. * (C) Copyright Al Viro 2000, 2001
  5. * Released under GPL v2.
  6. *
  7. * Based on code from fs/super.c, copyright Linus Torvalds and others.
  8. * Heavily rewritten.
  9. */
  10. #include <linux/syscalls.h>
  11. #include <linux/slab.h>
  12. #include <linux/sched.h>
  13. #include <linux/spinlock.h>
  14. #include <linux/percpu.h>
  15. #include <linux/init.h>
  16. #include <linux/kernel.h>
  17. #include <linux/acct.h>
  18. #include <linux/capability.h>
  19. #include <linux/cpumask.h>
  20. #include <linux/module.h>
  21. #include <linux/sysfs.h>
  22. #include <linux/seq_file.h>
  23. #include <linux/mnt_namespace.h>
  24. #include <linux/namei.h>
  25. #include <linux/nsproxy.h>
  26. #include <linux/security.h>
  27. #include <linux/mount.h>
  28. #include <linux/ramfs.h>
  29. #include <linux/log2.h>
  30. #include <linux/idr.h>
  31. #include <linux/fs_struct.h>
  32. #include <linux/fsnotify.h>
  33. #include <asm/uaccess.h>
  34. #include <asm/unistd.h>
  35. #include "pnode.h"
  36. #include "internal.h"
  37. #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
  38. #define HASH_SIZE (1UL << HASH_SHIFT)
  39. static int event;
  40. static DEFINE_IDA(mnt_id_ida);
  41. static DEFINE_IDA(mnt_group_ida);
  42. static DEFINE_SPINLOCK(mnt_id_lock);
  43. static int mnt_id_start = 0;
  44. static int mnt_group_start = 1;
  45. static struct list_head *mount_hashtable __read_mostly;
  46. static struct kmem_cache *mnt_cache __read_mostly;
  47. static struct rw_semaphore namespace_sem;
  48. /* /sys/fs */
  49. struct kobject *fs_kobj;
  50. EXPORT_SYMBOL_GPL(fs_kobj);
  51. /*
  52. * vfsmount lock may be taken for read to prevent changes to the
  53. * vfsmount hash, ie. during mountpoint lookups or walking back
  54. * up the tree.
  55. *
  56. * It should be taken for write in all cases where the vfsmount
  57. * tree or hash is modified or when a vfsmount structure is modified.
  58. */
  59. DEFINE_BRLOCK(vfsmount_lock);
  60. static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
  61. {
  62. unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
  63. tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
  64. tmp = tmp + (tmp >> HASH_SHIFT);
  65. return tmp & (HASH_SIZE - 1);
  66. }
  67. #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
  68. /*
  69. * allocation is serialized by namespace_sem, but we need the spinlock to
  70. * serialize with freeing.
  71. */
  72. static int mnt_alloc_id(struct vfsmount *mnt)
  73. {
  74. int res;
  75. retry:
  76. ida_pre_get(&mnt_id_ida, GFP_KERNEL);
  77. spin_lock(&mnt_id_lock);
  78. res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
  79. if (!res)
  80. mnt_id_start = mnt->mnt_id + 1;
  81. spin_unlock(&mnt_id_lock);
  82. if (res == -EAGAIN)
  83. goto retry;
  84. return res;
  85. }
  86. static void mnt_free_id(struct vfsmount *mnt)
  87. {
  88. int id = mnt->mnt_id;
  89. spin_lock(&mnt_id_lock);
  90. ida_remove(&mnt_id_ida, id);
  91. if (mnt_id_start > id)
  92. mnt_id_start = id;
  93. spin_unlock(&mnt_id_lock);
  94. }
  95. /*
  96. * Allocate a new peer group ID
  97. *
  98. * mnt_group_ida is protected by namespace_sem
  99. */
  100. static int mnt_alloc_group_id(struct vfsmount *mnt)
  101. {
  102. int res;
  103. if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
  104. return -ENOMEM;
  105. res = ida_get_new_above(&mnt_group_ida,
  106. mnt_group_start,
  107. &mnt->mnt_group_id);
  108. if (!res)
  109. mnt_group_start = mnt->mnt_group_id + 1;
  110. return res;
  111. }
  112. /*
  113. * Release a peer group ID
  114. */
  115. void mnt_release_group_id(struct vfsmount *mnt)
  116. {
  117. int id = mnt->mnt_group_id;
  118. ida_remove(&mnt_group_ida, id);
  119. if (mnt_group_start > id)
  120. mnt_group_start = id;
  121. mnt->mnt_group_id = 0;
  122. }
  123. /*
  124. * vfsmount lock must be held for read
  125. */
  126. static inline void mnt_add_count(struct vfsmount *mnt, int n)
  127. {
  128. #ifdef CONFIG_SMP
  129. this_cpu_add(mnt->mnt_pcp->mnt_count, n);
  130. #else
  131. preempt_disable();
  132. mnt->mnt_count += n;
  133. preempt_enable();
  134. #endif
  135. }
  136. static inline void mnt_set_count(struct vfsmount *mnt, int n)
  137. {
  138. #ifdef CONFIG_SMP
  139. this_cpu_write(mnt->mnt_pcp->mnt_count, n);
  140. #else
  141. mnt->mnt_count = n;
  142. #endif
  143. }
  144. /*
  145. * vfsmount lock must be held for read
  146. */
  147. static inline void mnt_inc_count(struct vfsmount *mnt)
  148. {
  149. mnt_add_count(mnt, 1);
  150. }
  151. /*
  152. * vfsmount lock must be held for read
  153. */
  154. static inline void mnt_dec_count(struct vfsmount *mnt)
  155. {
  156. mnt_add_count(mnt, -1);
  157. }
  158. /*
  159. * vfsmount lock must be held for write
  160. */
  161. unsigned int mnt_get_count(struct vfsmount *mnt)
  162. {
  163. #ifdef CONFIG_SMP
  164. unsigned int count = 0;
  165. int cpu;
  166. for_each_possible_cpu(cpu) {
  167. count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
  168. }
  169. return count;
  170. #else
  171. return mnt->mnt_count;
  172. #endif
  173. }
  174. static struct vfsmount *alloc_vfsmnt(const char *name)
  175. {
  176. struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
  177. if (mnt) {
  178. int err;
  179. err = mnt_alloc_id(mnt);
  180. if (err)
  181. goto out_free_cache;
  182. if (name) {
  183. mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
  184. if (!mnt->mnt_devname)
  185. goto out_free_id;
  186. }
  187. #ifdef CONFIG_SMP
  188. mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
  189. if (!mnt->mnt_pcp)
  190. goto out_free_devname;
  191. this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
  192. #else
  193. mnt->mnt_count = 1;
  194. mnt->mnt_writers = 0;
  195. #endif
  196. INIT_LIST_HEAD(&mnt->mnt_hash);
  197. INIT_LIST_HEAD(&mnt->mnt_child);
  198. INIT_LIST_HEAD(&mnt->mnt_mounts);
  199. INIT_LIST_HEAD(&mnt->mnt_list);
  200. INIT_LIST_HEAD(&mnt->mnt_expire);
  201. INIT_LIST_HEAD(&mnt->mnt_share);
  202. INIT_LIST_HEAD(&mnt->mnt_slave_list);
  203. INIT_LIST_HEAD(&mnt->mnt_slave);
  204. #ifdef CONFIG_FSNOTIFY
  205. INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
  206. #endif
  207. }
  208. return mnt;
  209. #ifdef CONFIG_SMP
  210. out_free_devname:
  211. kfree(mnt->mnt_devname);
  212. #endif
  213. out_free_id:
  214. mnt_free_id(mnt);
  215. out_free_cache:
  216. kmem_cache_free(mnt_cache, mnt);
  217. return NULL;
  218. }
  219. /*
  220. * Most r/o checks on a fs are for operations that take
  221. * discrete amounts of time, like a write() or unlink().
  222. * We must keep track of when those operations start
  223. * (for permission checks) and when they end, so that
  224. * we can determine when writes are able to occur to
  225. * a filesystem.
  226. */
  227. /*
  228. * __mnt_is_readonly: check whether a mount is read-only
  229. * @mnt: the mount to check for its write status
  230. *
  231. * This shouldn't be used directly ouside of the VFS.
  232. * It does not guarantee that the filesystem will stay
  233. * r/w, just that it is right *now*. This can not and
  234. * should not be used in place of IS_RDONLY(inode).
  235. * mnt_want/drop_write() will _keep_ the filesystem
  236. * r/w.
  237. */
  238. int __mnt_is_readonly(struct vfsmount *mnt)
  239. {
  240. if (mnt->mnt_flags & MNT_READONLY)
  241. return 1;
  242. if (mnt->mnt_sb->s_flags & MS_RDONLY)
  243. return 1;
  244. return 0;
  245. }
  246. EXPORT_SYMBOL_GPL(__mnt_is_readonly);
  247. static inline void mnt_inc_writers(struct vfsmount *mnt)
  248. {
  249. #ifdef CONFIG_SMP
  250. this_cpu_inc(mnt->mnt_pcp->mnt_writers);
  251. #else
  252. mnt->mnt_writers++;
  253. #endif
  254. }
  255. static inline void mnt_dec_writers(struct vfsmount *mnt)
  256. {
  257. #ifdef CONFIG_SMP
  258. this_cpu_dec(mnt->mnt_pcp->mnt_writers);
  259. #else
  260. mnt->mnt_writers--;
  261. #endif
  262. }
  263. static unsigned int mnt_get_writers(struct vfsmount *mnt)
  264. {
  265. #ifdef CONFIG_SMP
  266. unsigned int count = 0;
  267. int cpu;
  268. for_each_possible_cpu(cpu) {
  269. count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
  270. }
  271. return count;
  272. #else
  273. return mnt->mnt_writers;
  274. #endif
  275. }
  276. /*
  277. * Most r/o checks on a fs are for operations that take
  278. * discrete amounts of time, like a write() or unlink().
  279. * We must keep track of when those operations start
  280. * (for permission checks) and when they end, so that
  281. * we can determine when writes are able to occur to
  282. * a filesystem.
  283. */
  284. /**
  285. * mnt_want_write - get write access to a mount
  286. * @mnt: the mount on which to take a write
  287. *
  288. * This tells the low-level filesystem that a write is
  289. * about to be performed to it, and makes sure that
  290. * writes are allowed before returning success. When
  291. * the write operation is finished, mnt_drop_write()
  292. * must be called. This is effectively a refcount.
  293. */
  294. int mnt_want_write(struct vfsmount *mnt)
  295. {
  296. int ret = 0;
  297. preempt_disable();
  298. mnt_inc_writers(mnt);
  299. /*
  300. * The store to mnt_inc_writers must be visible before we pass
  301. * MNT_WRITE_HOLD loop below, so that the slowpath can see our
  302. * incremented count after it has set MNT_WRITE_HOLD.
  303. */
  304. smp_mb();
  305. while (mnt->mnt_flags & MNT_WRITE_HOLD)
  306. cpu_relax();
  307. /*
  308. * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
  309. * be set to match its requirements. So we must not load that until
  310. * MNT_WRITE_HOLD is cleared.
  311. */
  312. smp_rmb();
  313. if (__mnt_is_readonly(mnt)) {
  314. mnt_dec_writers(mnt);
  315. ret = -EROFS;
  316. goto out;
  317. }
  318. out:
  319. preempt_enable();
  320. return ret;
  321. }
  322. EXPORT_SYMBOL_GPL(mnt_want_write);
  323. /**
  324. * mnt_clone_write - get write access to a mount
  325. * @mnt: the mount on which to take a write
  326. *
  327. * This is effectively like mnt_want_write, except
  328. * it must only be used to take an extra write reference
  329. * on a mountpoint that we already know has a write reference
  330. * on it. This allows some optimisation.
  331. *
  332. * After finished, mnt_drop_write must be called as usual to
  333. * drop the reference.
  334. */
  335. int mnt_clone_write(struct vfsmount *mnt)
  336. {
  337. /* superblock may be r/o */
  338. if (__mnt_is_readonly(mnt))
  339. return -EROFS;
  340. preempt_disable();
  341. mnt_inc_writers(mnt);
  342. preempt_enable();
  343. return 0;
  344. }
  345. EXPORT_SYMBOL_GPL(mnt_clone_write);
  346. /**
  347. * mnt_want_write_file - get write access to a file's mount
  348. * @file: the file who's mount on which to take a write
  349. *
  350. * This is like mnt_want_write, but it takes a file and can
  351. * do some optimisations if the file is open for write already
  352. */
  353. int mnt_want_write_file(struct file *file)
  354. {
  355. struct inode *inode = file->f_dentry->d_inode;
  356. if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
  357. return mnt_want_write(file->f_path.mnt);
  358. else
  359. return mnt_clone_write(file->f_path.mnt);
  360. }
  361. EXPORT_SYMBOL_GPL(mnt_want_write_file);
  362. /**
  363. * mnt_drop_write - give up write access to a mount
  364. * @mnt: the mount on which to give up write access
  365. *
  366. * Tells the low-level filesystem that we are done
  367. * performing writes to it. Must be matched with
  368. * mnt_want_write() call above.
  369. */
  370. void mnt_drop_write(struct vfsmount *mnt)
  371. {
  372. preempt_disable();
  373. mnt_dec_writers(mnt);
  374. preempt_enable();
  375. }
  376. EXPORT_SYMBOL_GPL(mnt_drop_write);
  377. static int mnt_make_readonly(struct vfsmount *mnt)
  378. {
  379. int ret = 0;
  380. br_write_lock(vfsmount_lock);
  381. mnt->mnt_flags |= MNT_WRITE_HOLD;
  382. /*
  383. * After storing MNT_WRITE_HOLD, we'll read the counters. This store
  384. * should be visible before we do.
  385. */
  386. smp_mb();
  387. /*
  388. * With writers on hold, if this value is zero, then there are
  389. * definitely no active writers (although held writers may subsequently
  390. * increment the count, they'll have to wait, and decrement it after
  391. * seeing MNT_READONLY).
  392. *
  393. * It is OK to have counter incremented on one CPU and decremented on
  394. * another: the sum will add up correctly. The danger would be when we
  395. * sum up each counter, if we read a counter before it is incremented,
  396. * but then read another CPU's count which it has been subsequently
  397. * decremented from -- we would see more decrements than we should.
  398. * MNT_WRITE_HOLD protects against this scenario, because
  399. * mnt_want_write first increments count, then smp_mb, then spins on
  400. * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
  401. * we're counting up here.
  402. */
  403. if (mnt_get_writers(mnt) > 0)
  404. ret = -EBUSY;
  405. else
  406. mnt->mnt_flags |= MNT_READONLY;
  407. /*
  408. * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
  409. * that become unheld will see MNT_READONLY.
  410. */
  411. smp_wmb();
  412. mnt->mnt_flags &= ~MNT_WRITE_HOLD;
  413. br_write_unlock(vfsmount_lock);
  414. return ret;
  415. }
  416. static void __mnt_unmake_readonly(struct vfsmount *mnt)
  417. {
  418. br_write_lock(vfsmount_lock);
  419. mnt->mnt_flags &= ~MNT_READONLY;
  420. br_write_unlock(vfsmount_lock);
  421. }
  422. static void free_vfsmnt(struct vfsmount *mnt)
  423. {
  424. kfree(mnt->mnt_devname);
  425. mnt_free_id(mnt);
  426. #ifdef CONFIG_SMP
  427. free_percpu(mnt->mnt_pcp);
  428. #endif
  429. kmem_cache_free(mnt_cache, mnt);
  430. }
  431. /*
  432. * find the first or last mount at @dentry on vfsmount @mnt depending on
  433. * @dir. If @dir is set return the first mount else return the last mount.
  434. * vfsmount_lock must be held for read or write.
  435. */
  436. struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
  437. int dir)
  438. {
  439. struct list_head *head = mount_hashtable + hash(mnt, dentry);
  440. struct list_head *tmp = head;
  441. struct vfsmount *p, *found = NULL;
  442. for (;;) {
  443. tmp = dir ? tmp->next : tmp->prev;
  444. p = NULL;
  445. if (tmp == head)
  446. break;
  447. p = list_entry(tmp, struct vfsmount, mnt_hash);
  448. if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
  449. found = p;
  450. break;
  451. }
  452. }
  453. return found;
  454. }
  455. /*
  456. * lookup_mnt increments the ref count before returning
  457. * the vfsmount struct.
  458. */
  459. struct vfsmount *lookup_mnt(struct path *path)
  460. {
  461. struct vfsmount *child_mnt;
  462. br_read_lock(vfsmount_lock);
  463. if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
  464. mntget(child_mnt);
  465. br_read_unlock(vfsmount_lock);
  466. return child_mnt;
  467. }
  468. static inline int check_mnt(struct vfsmount *mnt)
  469. {
  470. return mnt->mnt_ns == current->nsproxy->mnt_ns;
  471. }
  472. /*
  473. * vfsmount lock must be held for write
  474. */
  475. static void touch_mnt_namespace(struct mnt_namespace *ns)
  476. {
  477. if (ns) {
  478. ns->event = ++event;
  479. wake_up_interruptible(&ns->poll);
  480. }
  481. }
  482. /*
  483. * vfsmount lock must be held for write
  484. */
  485. static void __touch_mnt_namespace(struct mnt_namespace *ns)
  486. {
  487. if (ns && ns->event != event) {
  488. ns->event = event;
  489. wake_up_interruptible(&ns->poll);
  490. }
  491. }
  492. /*
  493. * Clear dentry's mounted state if it has no remaining mounts.
  494. * vfsmount_lock must be held for write.
  495. */
  496. static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
  497. {
  498. unsigned u;
  499. for (u = 0; u < HASH_SIZE; u++) {
  500. struct vfsmount *p;
  501. list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
  502. if (p->mnt_mountpoint == dentry)
  503. return;
  504. }
  505. }
  506. spin_lock(&dentry->d_lock);
  507. dentry->d_flags &= ~DCACHE_MOUNTED;
  508. spin_unlock(&dentry->d_lock);
  509. }
  510. /*
  511. * vfsmount lock must be held for write
  512. */
  513. static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
  514. {
  515. old_path->dentry = mnt->mnt_mountpoint;
  516. old_path->mnt = mnt->mnt_parent;
  517. mnt->mnt_parent = mnt;
  518. mnt->mnt_mountpoint = mnt->mnt_root;
  519. list_del_init(&mnt->mnt_child);
  520. list_del_init(&mnt->mnt_hash);
  521. dentry_reset_mounted(old_path->mnt, old_path->dentry);
  522. }
  523. /*
  524. * vfsmount lock must be held for write
  525. */
  526. void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
  527. struct vfsmount *child_mnt)
  528. {
  529. child_mnt->mnt_parent = mntget(mnt);
  530. child_mnt->mnt_mountpoint = dget(dentry);
  531. spin_lock(&dentry->d_lock);
  532. dentry->d_flags |= DCACHE_MOUNTED;
  533. spin_unlock(&dentry->d_lock);
  534. }
  535. /*
  536. * vfsmount lock must be held for write
  537. */
  538. static void attach_mnt(struct vfsmount *mnt, struct path *path)
  539. {
  540. mnt_set_mountpoint(path->mnt, path->dentry, mnt);
  541. list_add_tail(&mnt->mnt_hash, mount_hashtable +
  542. hash(path->mnt, path->dentry));
  543. list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
  544. }
  545. static inline void __mnt_make_longterm(struct vfsmount *mnt)
  546. {
  547. #ifdef CONFIG_SMP
  548. atomic_inc(&mnt->mnt_longterm);
  549. #endif
  550. }
  551. /* needs vfsmount lock for write */
  552. static inline void __mnt_make_shortterm(struct vfsmount *mnt)
  553. {
  554. #ifdef CONFIG_SMP
  555. atomic_dec(&mnt->mnt_longterm);
  556. #endif
  557. }
  558. /*
  559. * vfsmount lock must be held for write
  560. */
  561. static void commit_tree(struct vfsmount *mnt)
  562. {
  563. struct vfsmount *parent = mnt->mnt_parent;
  564. struct vfsmount *m;
  565. LIST_HEAD(head);
  566. struct mnt_namespace *n = parent->mnt_ns;
  567. BUG_ON(parent == mnt);
  568. list_add_tail(&head, &mnt->mnt_list);
  569. list_for_each_entry(m, &head, mnt_list) {
  570. m->mnt_ns = n;
  571. __mnt_make_longterm(m);
  572. }
  573. list_splice(&head, n->list.prev);
  574. list_add_tail(&mnt->mnt_hash, mount_hashtable +
  575. hash(parent, mnt->mnt_mountpoint));
  576. list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
  577. touch_mnt_namespace(n);
  578. }
  579. static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
  580. {
  581. struct list_head *next = p->mnt_mounts.next;
  582. if (next == &p->mnt_mounts) {
  583. while (1) {
  584. if (p == root)
  585. return NULL;
  586. next = p->mnt_child.next;
  587. if (next != &p->mnt_parent->mnt_mounts)
  588. break;
  589. p = p->mnt_parent;
  590. }
  591. }
  592. return list_entry(next, struct vfsmount, mnt_child);
  593. }
  594. static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
  595. {
  596. struct list_head *prev = p->mnt_mounts.prev;
  597. while (prev != &p->mnt_mounts) {
  598. p = list_entry(prev, struct vfsmount, mnt_child);
  599. prev = p->mnt_mounts.prev;
  600. }
  601. return p;
  602. }
  603. struct vfsmount *
  604. vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
  605. {
  606. struct vfsmount *mnt;
  607. struct dentry *root;
  608. if (!type)
  609. return ERR_PTR(-ENODEV);
  610. mnt = alloc_vfsmnt(name);
  611. if (!mnt)
  612. return ERR_PTR(-ENOMEM);
  613. if (flags & MS_KERNMOUNT)
  614. mnt->mnt_flags = MNT_INTERNAL;
  615. root = mount_fs(type, flags, name, data);
  616. if (IS_ERR(root)) {
  617. free_vfsmnt(mnt);
  618. return ERR_CAST(root);
  619. }
  620. mnt->mnt_root = root;
  621. mnt->mnt_sb = root->d_sb;
  622. mnt->mnt_mountpoint = mnt->mnt_root;
  623. mnt->mnt_parent = mnt;
  624. return mnt;
  625. }
  626. EXPORT_SYMBOL_GPL(vfs_kern_mount);
  627. static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
  628. int flag)
  629. {
  630. struct super_block *sb = old->mnt_sb;
  631. struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
  632. if (mnt) {
  633. if (flag & (CL_SLAVE | CL_PRIVATE))
  634. mnt->mnt_group_id = 0; /* not a peer of original */
  635. else
  636. mnt->mnt_group_id = old->mnt_group_id;
  637. if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
  638. int err = mnt_alloc_group_id(mnt);
  639. if (err)
  640. goto out_free;
  641. }
  642. mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
  643. atomic_inc(&sb->s_active);
  644. mnt->mnt_sb = sb;
  645. mnt->mnt_root = dget(root);
  646. mnt->mnt_mountpoint = mnt->mnt_root;
  647. mnt->mnt_parent = mnt;
  648. if (flag & CL_SLAVE) {
  649. list_add(&mnt->mnt_slave, &old->mnt_slave_list);
  650. mnt->mnt_master = old;
  651. CLEAR_MNT_SHARED(mnt);
  652. } else if (!(flag & CL_PRIVATE)) {
  653. if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
  654. list_add(&mnt->mnt_share, &old->mnt_share);
  655. if (IS_MNT_SLAVE(old))
  656. list_add(&mnt->mnt_slave, &old->mnt_slave);
  657. mnt->mnt_master = old->mnt_master;
  658. }
  659. if (flag & CL_MAKE_SHARED)
  660. set_mnt_shared(mnt);
  661. /* stick the duplicate mount on the same expiry list
  662. * as the original if that was on one */
  663. if (flag & CL_EXPIRE) {
  664. if (!list_empty(&old->mnt_expire))
  665. list_add(&mnt->mnt_expire, &old->mnt_expire);
  666. }
  667. }
  668. return mnt;
  669. out_free:
  670. free_vfsmnt(mnt);
  671. return NULL;
  672. }
  673. static inline void mntfree(struct vfsmount *mnt)
  674. {
  675. struct super_block *sb = mnt->mnt_sb;
  676. /*
  677. * This probably indicates that somebody messed
  678. * up a mnt_want/drop_write() pair. If this
  679. * happens, the filesystem was probably unable
  680. * to make r/w->r/o transitions.
  681. */
  682. /*
  683. * The locking used to deal with mnt_count decrement provides barriers,
  684. * so mnt_get_writers() below is safe.
  685. */
  686. WARN_ON(mnt_get_writers(mnt));
  687. fsnotify_vfsmount_delete(mnt);
  688. dput(mnt->mnt_root);
  689. free_vfsmnt(mnt);
  690. deactivate_super(sb);
  691. }
  692. static void mntput_no_expire(struct vfsmount *mnt)
  693. {
  694. put_again:
  695. #ifdef CONFIG_SMP
  696. br_read_lock(vfsmount_lock);
  697. if (likely(atomic_read(&mnt->mnt_longterm))) {
  698. mnt_dec_count(mnt);
  699. br_read_unlock(vfsmount_lock);
  700. return;
  701. }
  702. br_read_unlock(vfsmount_lock);
  703. br_write_lock(vfsmount_lock);
  704. mnt_dec_count(mnt);
  705. if (mnt_get_count(mnt)) {
  706. br_write_unlock(vfsmount_lock);
  707. return;
  708. }
  709. #else
  710. mnt_dec_count(mnt);
  711. if (likely(mnt_get_count(mnt)))
  712. return;
  713. br_write_lock(vfsmount_lock);
  714. #endif
  715. if (unlikely(mnt->mnt_pinned)) {
  716. mnt_add_count(mnt, mnt->mnt_pinned + 1);
  717. mnt->mnt_pinned = 0;
  718. br_write_unlock(vfsmount_lock);
  719. acct_auto_close_mnt(mnt);
  720. goto put_again;
  721. }
  722. br_write_unlock(vfsmount_lock);
  723. mntfree(mnt);
  724. }
  725. void mntput(struct vfsmount *mnt)
  726. {
  727. if (mnt) {
  728. /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
  729. if (unlikely(mnt->mnt_expiry_mark))
  730. mnt->mnt_expiry_mark = 0;
  731. mntput_no_expire(mnt);
  732. }
  733. }
  734. EXPORT_SYMBOL(mntput);
  735. struct vfsmount *mntget(struct vfsmount *mnt)
  736. {
  737. if (mnt)
  738. mnt_inc_count(mnt);
  739. return mnt;
  740. }
  741. EXPORT_SYMBOL(mntget);
  742. void mnt_pin(struct vfsmount *mnt)
  743. {
  744. br_write_lock(vfsmount_lock);
  745. mnt->mnt_pinned++;
  746. br_write_unlock(vfsmount_lock);
  747. }
  748. EXPORT_SYMBOL(mnt_pin);
  749. void mnt_unpin(struct vfsmount *mnt)
  750. {
  751. br_write_lock(vfsmount_lock);
  752. if (mnt->mnt_pinned) {
  753. mnt_inc_count(mnt);
  754. mnt->mnt_pinned--;
  755. }
  756. br_write_unlock(vfsmount_lock);
  757. }
  758. EXPORT_SYMBOL(mnt_unpin);
  759. static inline void mangle(struct seq_file *m, const char *s)
  760. {
  761. seq_escape(m, s, " \t\n\\");
  762. }
  763. /*
  764. * Simple .show_options callback for filesystems which don't want to
  765. * implement more complex mount option showing.
  766. *
  767. * See also save_mount_options().
  768. */
  769. int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
  770. {
  771. const char *options;
  772. rcu_read_lock();
  773. options = rcu_dereference(mnt->mnt_sb->s_options);
  774. if (options != NULL && options[0]) {
  775. seq_putc(m, ',');
  776. mangle(m, options);
  777. }
  778. rcu_read_unlock();
  779. return 0;
  780. }
  781. EXPORT_SYMBOL(generic_show_options);
  782. /*
  783. * If filesystem uses generic_show_options(), this function should be
  784. * called from the fill_super() callback.
  785. *
  786. * The .remount_fs callback usually needs to be handled in a special
  787. * way, to make sure, that previous options are not overwritten if the
  788. * remount fails.
  789. *
  790. * Also note, that if the filesystem's .remount_fs function doesn't
  791. * reset all options to their default value, but changes only newly
  792. * given options, then the displayed options will not reflect reality
  793. * any more.
  794. */
  795. void save_mount_options(struct super_block *sb, char *options)
  796. {
  797. BUG_ON(sb->s_options);
  798. rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
  799. }
  800. EXPORT_SYMBOL(save_mount_options);
  801. void replace_mount_options(struct super_block *sb, char *options)
  802. {
  803. char *old = sb->s_options;
  804. rcu_assign_pointer(sb->s_options, options);
  805. if (old) {
  806. synchronize_rcu();
  807. kfree(old);
  808. }
  809. }
  810. EXPORT_SYMBOL(replace_mount_options);
  811. #ifdef CONFIG_PROC_FS
  812. /* iterator */
  813. static void *m_start(struct seq_file *m, loff_t *pos)
  814. {
  815. struct proc_mounts *p = m->private;
  816. down_read(&namespace_sem);
  817. return seq_list_start(&p->ns->list, *pos);
  818. }
  819. static void *m_next(struct seq_file *m, void *v, loff_t *pos)
  820. {
  821. struct proc_mounts *p = m->private;
  822. return seq_list_next(v, &p->ns->list, pos);
  823. }
  824. static void m_stop(struct seq_file *m, void *v)
  825. {
  826. up_read(&namespace_sem);
  827. }
  828. int mnt_had_events(struct proc_mounts *p)
  829. {
  830. struct mnt_namespace *ns = p->ns;
  831. int res = 0;
  832. br_read_lock(vfsmount_lock);
  833. if (p->event != ns->event) {
  834. p->event = ns->event;
  835. res = 1;
  836. }
  837. br_read_unlock(vfsmount_lock);
  838. return res;
  839. }
  840. struct proc_fs_info {
  841. int flag;
  842. const char *str;
  843. };
  844. static int show_sb_opts(struct seq_file *m, struct super_block *sb)
  845. {
  846. static const struct proc_fs_info fs_info[] = {
  847. { MS_SYNCHRONOUS, ",sync" },
  848. { MS_DIRSYNC, ",dirsync" },
  849. { MS_MANDLOCK, ",mand" },
  850. { 0, NULL }
  851. };
  852. const struct proc_fs_info *fs_infop;
  853. for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
  854. if (sb->s_flags & fs_infop->flag)
  855. seq_puts(m, fs_infop->str);
  856. }
  857. return security_sb_show_options(m, sb);
  858. }
  859. static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
  860. {
  861. static const struct proc_fs_info mnt_info[] = {
  862. { MNT_NOSUID, ",nosuid" },
  863. { MNT_NODEV, ",nodev" },
  864. { MNT_NOEXEC, ",noexec" },
  865. { MNT_NOATIME, ",noatime" },
  866. { MNT_NODIRATIME, ",nodiratime" },
  867. { MNT_RELATIME, ",relatime" },
  868. { 0, NULL }
  869. };
  870. const struct proc_fs_info *fs_infop;
  871. for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
  872. if (mnt->mnt_flags & fs_infop->flag)
  873. seq_puts(m, fs_infop->str);
  874. }
  875. }
  876. static void show_type(struct seq_file *m, struct super_block *sb)
  877. {
  878. mangle(m, sb->s_type->name);
  879. if (sb->s_subtype && sb->s_subtype[0]) {
  880. seq_putc(m, '.');
  881. mangle(m, sb->s_subtype);
  882. }
  883. }
  884. static int show_vfsmnt(struct seq_file *m, void *v)
  885. {
  886. struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
  887. int err = 0;
  888. struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
  889. if (mnt->mnt_sb->s_op->show_devname) {
  890. err = mnt->mnt_sb->s_op->show_devname(m, mnt);
  891. if (err)
  892. goto out;
  893. } else {
  894. mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
  895. }
  896. seq_putc(m, ' ');
  897. seq_path(m, &mnt_path, " \t\n\\");
  898. seq_putc(m, ' ');
  899. show_type(m, mnt->mnt_sb);
  900. seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
  901. err = show_sb_opts(m, mnt->mnt_sb);
  902. if (err)
  903. goto out;
  904. show_mnt_opts(m, mnt);
  905. if (mnt->mnt_sb->s_op->show_options)
  906. err = mnt->mnt_sb->s_op->show_options(m, mnt);
  907. seq_puts(m, " 0 0\n");
  908. out:
  909. return err;
  910. }
  911. const struct seq_operations mounts_op = {
  912. .start = m_start,
  913. .next = m_next,
  914. .stop = m_stop,
  915. .show = show_vfsmnt
  916. };
  917. static int uuid_is_nil(u8 *uuid)
  918. {
  919. int i;
  920. u8 *cp = (u8 *)uuid;
  921. for (i = 0; i < 16; i++) {
  922. if (*cp++)
  923. return 0;
  924. }
  925. return 1;
  926. }
  927. static int show_mountinfo(struct seq_file *m, void *v)
  928. {
  929. struct proc_mounts *p = m->private;
  930. struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
  931. struct super_block *sb = mnt->mnt_sb;
  932. struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
  933. struct path root = p->root;
  934. int err = 0;
  935. seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
  936. MAJOR(sb->s_dev), MINOR(sb->s_dev));
  937. if (sb->s_op->show_path)
  938. err = sb->s_op->show_path(m, mnt);
  939. else
  940. seq_dentry(m, mnt->mnt_root, " \t\n\\");
  941. if (err)
  942. goto out;
  943. seq_putc(m, ' ');
  944. seq_path_root(m, &mnt_path, &root, " \t\n\\");
  945. if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
  946. /*
  947. * Mountpoint is outside root, discard that one. Ugly,
  948. * but less so than trying to do that in iterator in a
  949. * race-free way (due to renames).
  950. */
  951. return SEQ_SKIP;
  952. }
  953. seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
  954. show_mnt_opts(m, mnt);
  955. /* Tagged fields ("foo:X" or "bar") */
  956. if (IS_MNT_SHARED(mnt))
  957. seq_printf(m, " shared:%i", mnt->mnt_group_id);
  958. if (IS_MNT_SLAVE(mnt)) {
  959. int master = mnt->mnt_master->mnt_group_id;
  960. int dom = get_dominating_id(mnt, &p->root);
  961. seq_printf(m, " master:%i", master);
  962. if (dom && dom != master)
  963. seq_printf(m, " propagate_from:%i", dom);
  964. }
  965. if (IS_MNT_UNBINDABLE(mnt))
  966. seq_puts(m, " unbindable");
  967. if (!uuid_is_nil(mnt->mnt_sb->s_uuid))
  968. /* print the uuid */
  969. seq_printf(m, " uuid:%pU", mnt->mnt_sb->s_uuid);
  970. /* Filesystem specific data */
  971. seq_puts(m, " - ");
  972. show_type(m, sb);
  973. seq_putc(m, ' ');
  974. if (sb->s_op->show_devname)
  975. err = sb->s_op->show_devname(m, mnt);
  976. else
  977. mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
  978. if (err)
  979. goto out;
  980. seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
  981. err = show_sb_opts(m, sb);
  982. if (err)
  983. goto out;
  984. if (sb->s_op->show_options)
  985. err = sb->s_op->show_options(m, mnt);
  986. seq_putc(m, '\n');
  987. out:
  988. return err;
  989. }
  990. const struct seq_operations mountinfo_op = {
  991. .start = m_start,
  992. .next = m_next,
  993. .stop = m_stop,
  994. .show = show_mountinfo,
  995. };
  996. static int show_vfsstat(struct seq_file *m, void *v)
  997. {
  998. struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
  999. struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
  1000. int err = 0;
  1001. /* device */
  1002. if (mnt->mnt_sb->s_op->show_devname) {
  1003. err = mnt->mnt_sb->s_op->show_devname(m, mnt);
  1004. } else {
  1005. if (mnt->mnt_devname) {
  1006. seq_puts(m, "device ");
  1007. mangle(m, mnt->mnt_devname);
  1008. } else
  1009. seq_puts(m, "no device");
  1010. }
  1011. /* mount point */
  1012. seq_puts(m, " mounted on ");
  1013. seq_path(m, &mnt_path, " \t\n\\");
  1014. seq_putc(m, ' ');
  1015. /* file system type */
  1016. seq_puts(m, "with fstype ");
  1017. show_type(m, mnt->mnt_sb);
  1018. /* optional statistics */
  1019. if (mnt->mnt_sb->s_op->show_stats) {
  1020. seq_putc(m, ' ');
  1021. if (!err)
  1022. err = mnt->mnt_sb->s_op->show_stats(m, mnt);
  1023. }
  1024. seq_putc(m, '\n');
  1025. return err;
  1026. }
  1027. const struct seq_operations mountstats_op = {
  1028. .start = m_start,
  1029. .next = m_next,
  1030. .stop = m_stop,
  1031. .show = show_vfsstat,
  1032. };
  1033. #endif /* CONFIG_PROC_FS */
  1034. /**
  1035. * may_umount_tree - check if a mount tree is busy
  1036. * @mnt: root of mount tree
  1037. *
  1038. * This is called to check if a tree of mounts has any
  1039. * open files, pwds, chroots or sub mounts that are
  1040. * busy.
  1041. */
  1042. int may_umount_tree(struct vfsmount *mnt)
  1043. {
  1044. int actual_refs = 0;
  1045. int minimum_refs = 0;
  1046. struct vfsmount *p;
  1047. /* write lock needed for mnt_get_count */
  1048. br_write_lock(vfsmount_lock);
  1049. for (p = mnt; p; p = next_mnt(p, mnt)) {
  1050. actual_refs += mnt_get_count(p);
  1051. minimum_refs += 2;
  1052. }
  1053. br_write_unlock(vfsmount_lock);
  1054. if (actual_refs > minimum_refs)
  1055. return 0;
  1056. return 1;
  1057. }
  1058. EXPORT_SYMBOL(may_umount_tree);
  1059. /**
  1060. * may_umount - check if a mount point is busy
  1061. * @mnt: root of mount
  1062. *
  1063. * This is called to check if a mount point has any
  1064. * open files, pwds, chroots or sub mounts. If the
  1065. * mount has sub mounts this will return busy
  1066. * regardless of whether the sub mounts are busy.
  1067. *
  1068. * Doesn't take quota and stuff into account. IOW, in some cases it will
  1069. * give false negatives. The main reason why it's here is that we need
  1070. * a non-destructive way to look for easily umountable filesystems.
  1071. */
  1072. int may_umount(struct vfsmount *mnt)
  1073. {
  1074. int ret = 1;
  1075. down_read(&namespace_sem);
  1076. br_write_lock(vfsmount_lock);
  1077. if (propagate_mount_busy(mnt, 2))
  1078. ret = 0;
  1079. br_write_unlock(vfsmount_lock);
  1080. up_read(&namespace_sem);
  1081. return ret;
  1082. }
  1083. EXPORT_SYMBOL(may_umount);
  1084. void release_mounts(struct list_head *head)
  1085. {
  1086. struct vfsmount *mnt;
  1087. while (!list_empty(head)) {
  1088. mnt = list_first_entry(head, struct vfsmount, mnt_hash);
  1089. list_del_init(&mnt->mnt_hash);
  1090. if (mnt->mnt_parent != mnt) {
  1091. struct dentry *dentry;
  1092. struct vfsmount *m;
  1093. br_write_lock(vfsmount_lock);
  1094. dentry = mnt->mnt_mountpoint;
  1095. m = mnt->mnt_parent;
  1096. mnt->mnt_mountpoint = mnt->mnt_root;
  1097. mnt->mnt_parent = mnt;
  1098. m->mnt_ghosts--;
  1099. br_write_unlock(vfsmount_lock);
  1100. dput(dentry);
  1101. mntput(m);
  1102. }
  1103. mntput(mnt);
  1104. }
  1105. }
  1106. /*
  1107. * vfsmount lock must be held for write
  1108. * namespace_sem must be held for write
  1109. */
  1110. void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
  1111. {
  1112. LIST_HEAD(tmp_list);
  1113. struct vfsmount *p;
  1114. for (p = mnt; p; p = next_mnt(p, mnt))
  1115. list_move(&p->mnt_hash, &tmp_list);
  1116. if (propagate)
  1117. propagate_umount(&tmp_list);
  1118. list_for_each_entry(p, &tmp_list, mnt_hash) {
  1119. list_del_init(&p->mnt_expire);
  1120. list_del_init(&p->mnt_list);
  1121. __touch_mnt_namespace(p->mnt_ns);
  1122. p->mnt_ns = NULL;
  1123. __mnt_make_shortterm(p);
  1124. list_del_init(&p->mnt_child);
  1125. if (p->mnt_parent != p) {
  1126. p->mnt_parent->mnt_ghosts++;
  1127. dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
  1128. }
  1129. change_mnt_propagation(p, MS_PRIVATE);
  1130. }
  1131. list_splice(&tmp_list, kill);
  1132. }
  1133. static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
  1134. static int do_umount(struct vfsmount *mnt, int flags)
  1135. {
  1136. struct super_block *sb = mnt->mnt_sb;
  1137. int retval;
  1138. LIST_HEAD(umount_list);
  1139. retval = security_sb_umount(mnt, flags);
  1140. if (retval)
  1141. return retval;
  1142. /*
  1143. * Allow userspace to request a mountpoint be expired rather than
  1144. * unmounting unconditionally. Unmount only happens if:
  1145. * (1) the mark is already set (the mark is cleared by mntput())
  1146. * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
  1147. */
  1148. if (flags & MNT_EXPIRE) {
  1149. if (mnt == current->fs->root.mnt ||
  1150. flags & (MNT_FORCE | MNT_DETACH))
  1151. return -EINVAL;
  1152. /*
  1153. * probably don't strictly need the lock here if we examined
  1154. * all race cases, but it's a slowpath.
  1155. */
  1156. br_write_lock(vfsmount_lock);
  1157. if (mnt_get_count(mnt) != 2) {
  1158. br_write_unlock(vfsmount_lock);
  1159. return -EBUSY;
  1160. }
  1161. br_write_unlock(vfsmount_lock);
  1162. if (!xchg(&mnt->mnt_expiry_mark, 1))
  1163. return -EAGAIN;
  1164. }
  1165. /*
  1166. * If we may have to abort operations to get out of this
  1167. * mount, and they will themselves hold resources we must
  1168. * allow the fs to do things. In the Unix tradition of
  1169. * 'Gee thats tricky lets do it in userspace' the umount_begin
  1170. * might fail to complete on the first run through as other tasks
  1171. * must return, and the like. Thats for the mount program to worry
  1172. * about for the moment.
  1173. */
  1174. if (flags & MNT_FORCE && sb->s_op->umount_begin) {
  1175. sb->s_op->umount_begin(sb);
  1176. }
  1177. /*
  1178. * No sense to grab the lock for this test, but test itself looks
  1179. * somewhat bogus. Suggestions for better replacement?
  1180. * Ho-hum... In principle, we might treat that as umount + switch
  1181. * to rootfs. GC would eventually take care of the old vfsmount.
  1182. * Actually it makes sense, especially if rootfs would contain a
  1183. * /reboot - static binary that would close all descriptors and
  1184. * call reboot(9). Then init(8) could umount root and exec /reboot.
  1185. */
  1186. if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
  1187. /*
  1188. * Special case for "unmounting" root ...
  1189. * we just try to remount it readonly.
  1190. */
  1191. down_write(&sb->s_umount);
  1192. if (!(sb->s_flags & MS_RDONLY))
  1193. retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
  1194. up_write(&sb->s_umount);
  1195. return retval;
  1196. }
  1197. down_write(&namespace_sem);
  1198. br_write_lock(vfsmount_lock);
  1199. event++;
  1200. if (!(flags & MNT_DETACH))
  1201. shrink_submounts(mnt, &umount_list);
  1202. retval = -EBUSY;
  1203. if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
  1204. if (!list_empty(&mnt->mnt_list))
  1205. umount_tree(mnt, 1, &umount_list);
  1206. retval = 0;
  1207. }
  1208. br_write_unlock(vfsmount_lock);
  1209. up_write(&namespace_sem);
  1210. release_mounts(&umount_list);
  1211. return retval;
  1212. }
  1213. /*
  1214. * Now umount can handle mount points as well as block devices.
  1215. * This is important for filesystems which use unnamed block devices.
  1216. *
  1217. * We now support a flag for forced unmount like the other 'big iron'
  1218. * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
  1219. */
  1220. SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
  1221. {
  1222. struct path path;
  1223. int retval;
  1224. int lookup_flags = 0;
  1225. if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
  1226. return -EINVAL;
  1227. if (!(flags & UMOUNT_NOFOLLOW))
  1228. lookup_flags |= LOOKUP_FOLLOW;
  1229. retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
  1230. if (retval)
  1231. goto out;
  1232. retval = -EINVAL;
  1233. if (path.dentry != path.mnt->mnt_root)
  1234. goto dput_and_out;
  1235. if (!check_mnt(path.mnt))
  1236. goto dput_and_out;
  1237. retval = -EPERM;
  1238. if (!capable(CAP_SYS_ADMIN))
  1239. goto dput_and_out;
  1240. retval = do_umount(path.mnt, flags);
  1241. dput_and_out:
  1242. /* we mustn't call path_put() as that would clear mnt_expiry_mark */
  1243. dput(path.dentry);
  1244. mntput_no_expire(path.mnt);
  1245. out:
  1246. return retval;
  1247. }
  1248. #ifdef __ARCH_WANT_SYS_OLDUMOUNT
  1249. /*
  1250. * The 2.0 compatible umount. No flags.
  1251. */
  1252. SYSCALL_DEFINE1(oldumount, char __user *, name)
  1253. {
  1254. return sys_umount(name, 0);
  1255. }
  1256. #endif
  1257. static int mount_is_safe(struct path *path)
  1258. {
  1259. if (capable(CAP_SYS_ADMIN))
  1260. return 0;
  1261. return -EPERM;
  1262. #ifdef notyet
  1263. if (S_ISLNK(path->dentry->d_inode->i_mode))
  1264. return -EPERM;
  1265. if (path->dentry->d_inode->i_mode & S_ISVTX) {
  1266. if (current_uid() != path->dentry->d_inode->i_uid)
  1267. return -EPERM;
  1268. }
  1269. if (inode_permission(path->dentry->d_inode, MAY_WRITE))
  1270. return -EPERM;
  1271. return 0;
  1272. #endif
  1273. }
  1274. struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
  1275. int flag)
  1276. {
  1277. struct vfsmount *res, *p, *q, *r, *s;
  1278. struct path path;
  1279. if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
  1280. return NULL;
  1281. res = q = clone_mnt(mnt, dentry, flag);
  1282. if (!q)
  1283. goto Enomem;
  1284. q->mnt_mountpoint = mnt->mnt_mountpoint;
  1285. p = mnt;
  1286. list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
  1287. if (!is_subdir(r->mnt_mountpoint, dentry))
  1288. continue;
  1289. for (s = r; s; s = next_mnt(s, r)) {
  1290. if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
  1291. s = skip_mnt_tree(s);
  1292. continue;
  1293. }
  1294. while (p != s->mnt_parent) {
  1295. p = p->mnt_parent;
  1296. q = q->mnt_parent;
  1297. }
  1298. p = s;
  1299. path.mnt = q;
  1300. path.dentry = p->mnt_mountpoint;
  1301. q = clone_mnt(p, p->mnt_root, flag);
  1302. if (!q)
  1303. goto Enomem;
  1304. br_write_lock(vfsmount_lock);
  1305. list_add_tail(&q->mnt_list, &res->mnt_list);
  1306. attach_mnt(q, &path);
  1307. br_write_unlock(vfsmount_lock);
  1308. }
  1309. }
  1310. return res;
  1311. Enomem:
  1312. if (res) {
  1313. LIST_HEAD(umount_list);
  1314. br_write_lock(vfsmount_lock);
  1315. umount_tree(res, 0, &umount_list);
  1316. br_write_unlock(vfsmount_lock);
  1317. release_mounts(&umount_list);
  1318. }
  1319. return NULL;
  1320. }
  1321. struct vfsmount *collect_mounts(struct path *path)
  1322. {
  1323. struct vfsmount *tree;
  1324. down_write(&namespace_sem);
  1325. tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
  1326. up_write(&namespace_sem);
  1327. return tree;
  1328. }
  1329. void drop_collected_mounts(struct vfsmount *mnt)
  1330. {
  1331. LIST_HEAD(umount_list);
  1332. down_write(&namespace_sem);
  1333. br_write_lock(vfsmount_lock);
  1334. umount_tree(mnt, 0, &umount_list);
  1335. br_write_unlock(vfsmount_lock);
  1336. up_write(&namespace_sem);
  1337. release_mounts(&umount_list);
  1338. }
  1339. int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
  1340. struct vfsmount *root)
  1341. {
  1342. struct vfsmount *mnt;
  1343. int res = f(root, arg);
  1344. if (res)
  1345. return res;
  1346. list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
  1347. res = f(mnt, arg);
  1348. if (res)
  1349. return res;
  1350. }
  1351. return 0;
  1352. }
  1353. static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
  1354. {
  1355. struct vfsmount *p;
  1356. for (p = mnt; p != end; p = next_mnt(p, mnt)) {
  1357. if (p->mnt_group_id && !IS_MNT_SHARED(p))
  1358. mnt_release_group_id(p);
  1359. }
  1360. }
  1361. static int invent_group_ids(struct vfsmount *mnt, bool recurse)
  1362. {
  1363. struct vfsmount *p;
  1364. for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
  1365. if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
  1366. int err = mnt_alloc_group_id(p);
  1367. if (err) {
  1368. cleanup_group_ids(mnt, p);
  1369. return err;
  1370. }
  1371. }
  1372. }
  1373. return 0;
  1374. }
  1375. /*
  1376. * @source_mnt : mount tree to be attached
  1377. * @nd : place the mount tree @source_mnt is attached
  1378. * @parent_nd : if non-null, detach the source_mnt from its parent and
  1379. * store the parent mount and mountpoint dentry.
  1380. * (done when source_mnt is moved)
  1381. *
  1382. * NOTE: in the table below explains the semantics when a source mount
  1383. * of a given type is attached to a destination mount of a given type.
  1384. * ---------------------------------------------------------------------------
  1385. * | BIND MOUNT OPERATION |
  1386. * |**************************************************************************
  1387. * | source-->| shared | private | slave | unbindable |
  1388. * | dest | | | | |
  1389. * | | | | | | |
  1390. * | v | | | | |
  1391. * |**************************************************************************
  1392. * | shared | shared (++) | shared (+) | shared(+++)| invalid |
  1393. * | | | | | |
  1394. * |non-shared| shared (+) | private | slave (*) | invalid |
  1395. * ***************************************************************************
  1396. * A bind operation clones the source mount and mounts the clone on the
  1397. * destination mount.
  1398. *
  1399. * (++) the cloned mount is propagated to all the mounts in the propagation
  1400. * tree of the destination mount and the cloned mount is added to
  1401. * the peer group of the source mount.
  1402. * (+) the cloned mount is created under the destination mount and is marked
  1403. * as shared. The cloned mount is added to the peer group of the source
  1404. * mount.
  1405. * (+++) the mount is propagated to all the mounts in the propagation tree
  1406. * of the destination mount and the cloned mount is made slave
  1407. * of the same master as that of the source mount. The cloned mount
  1408. * is marked as 'shared and slave'.
  1409. * (*) the cloned mount is made a slave of the same master as that of the
  1410. * source mount.
  1411. *
  1412. * ---------------------------------------------------------------------------
  1413. * | MOVE MOUNT OPERATION |
  1414. * |**************************************************************************
  1415. * | source-->| shared | private | slave | unbindable |
  1416. * | dest | | | | |
  1417. * | | | | | | |
  1418. * | v | | | | |
  1419. * |**************************************************************************
  1420. * | shared | shared (+) | shared (+) | shared(+++) | invalid |
  1421. * | | | | | |
  1422. * |non-shared| shared (+*) | private | slave (*) | unbindable |
  1423. * ***************************************************************************
  1424. *
  1425. * (+) the mount is moved to the destination. And is then propagated to
  1426. * all the mounts in the propagation tree of the destination mount.
  1427. * (+*) the mount is moved to the destination.
  1428. * (+++) the mount is moved to the destination and is then propagated to
  1429. * all the mounts belonging to the destination mount's propagation tree.
  1430. * the mount is marked as 'shared and slave'.
  1431. * (*) the mount continues to be a slave at the new location.
  1432. *
  1433. * if the source mount is a tree, the operations explained above is
  1434. * applied to each mount in the tree.
  1435. * Must be called without spinlocks held, since this function can sleep
  1436. * in allocations.
  1437. */
  1438. static int attach_recursive_mnt(struct vfsmount *source_mnt,
  1439. struct path *path, struct path *parent_path)
  1440. {
  1441. LIST_HEAD(tree_list);
  1442. struct vfsmount *dest_mnt = path->mnt;
  1443. struct dentry *dest_dentry = path->dentry;
  1444. struct vfsmount *child, *p;
  1445. int err;
  1446. if (IS_MNT_SHARED(dest_mnt)) {
  1447. err = invent_group_ids(source_mnt, true);
  1448. if (err)
  1449. goto out;
  1450. }
  1451. err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
  1452. if (err)
  1453. goto out_cleanup_ids;
  1454. br_write_lock(vfsmount_lock);
  1455. if (IS_MNT_SHARED(dest_mnt)) {
  1456. for (p = source_mnt; p; p = next_mnt(p, source_mnt))
  1457. set_mnt_shared(p);
  1458. }
  1459. if (parent_path) {
  1460. detach_mnt(source_mnt, parent_path);
  1461. attach_mnt(source_mnt, path);
  1462. touch_mnt_namespace(parent_path->mnt->mnt_ns);
  1463. } else {
  1464. mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
  1465. commit_tree(source_mnt);
  1466. }
  1467. list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
  1468. list_del_init(&child->mnt_hash);
  1469. commit_tree(child);
  1470. }
  1471. br_write_unlock(vfsmount_lock);
  1472. return 0;
  1473. out_cleanup_ids:
  1474. if (IS_MNT_SHARED(dest_mnt))
  1475. cleanup_group_ids(source_mnt, NULL);
  1476. out:
  1477. return err;
  1478. }
  1479. static int lock_mount(struct path *path)
  1480. {
  1481. struct vfsmount *mnt;
  1482. retry:
  1483. mutex_lock(&path->dentry->d_inode->i_mutex);
  1484. if (unlikely(cant_mount(path->dentry))) {
  1485. mutex_unlock(&path->dentry->d_inode->i_mutex);
  1486. return -ENOENT;
  1487. }
  1488. down_write(&namespace_sem);
  1489. mnt = lookup_mnt(path);
  1490. if (likely(!mnt))
  1491. return 0;
  1492. up_write(&namespace_sem);
  1493. mutex_unlock(&path->dentry->d_inode->i_mutex);
  1494. path_put(path);
  1495. path->mnt = mnt;
  1496. path->dentry = dget(mnt->mnt_root);
  1497. goto retry;
  1498. }
  1499. static void unlock_mount(struct path *path)
  1500. {
  1501. up_write(&namespace_sem);
  1502. mutex_unlock(&path->dentry->d_inode->i_mutex);
  1503. }
  1504. static int graft_tree(struct vfsmount *mnt, struct path *path)
  1505. {
  1506. if (mnt->mnt_sb->s_flags & MS_NOUSER)
  1507. return -EINVAL;
  1508. if (S_ISDIR(path->dentry->d_inode->i_mode) !=
  1509. S_ISDIR(mnt->mnt_root->d_inode->i_mode))
  1510. return -ENOTDIR;
  1511. if (d_unlinked(path->dentry))
  1512. return -ENOENT;
  1513. return attach_recursive_mnt(mnt, path, NULL);
  1514. }
  1515. /*
  1516. * Sanity check the flags to change_mnt_propagation.
  1517. */
  1518. static int flags_to_propagation_type(int flags)
  1519. {
  1520. int type = flags & ~MS_REC;
  1521. /* Fail if any non-propagation flags are set */
  1522. if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
  1523. return 0;
  1524. /* Only one propagation flag should be set */
  1525. if (!is_power_of_2(type))
  1526. return 0;
  1527. return type;
  1528. }
  1529. /*
  1530. * recursively change the type of the mountpoint.
  1531. */
  1532. static int do_change_type(struct path *path, int flag)
  1533. {
  1534. struct vfsmount *m, *mnt = path->mnt;
  1535. int recurse = flag & MS_REC;
  1536. int type;
  1537. int err = 0;
  1538. if (!capable(CAP_SYS_ADMIN))
  1539. return -EPERM;
  1540. if (path->dentry != path->mnt->mnt_root)
  1541. return -EINVAL;
  1542. type = flags_to_propagation_type(flag);
  1543. if (!type)
  1544. return -EINVAL;
  1545. down_write(&namespace_sem);
  1546. if (type == MS_SHARED) {
  1547. err = invent_group_ids(mnt, recurse);
  1548. if (err)
  1549. goto out_unlock;
  1550. }
  1551. br_write_lock(vfsmount_lock);
  1552. for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
  1553. change_mnt_propagation(m, type);
  1554. br_write_unlock(vfsmount_lock);
  1555. out_unlock:
  1556. up_write(&namespace_sem);
  1557. return err;
  1558. }
  1559. /*
  1560. * do loopback mount.
  1561. */
  1562. static int do_loopback(struct path *path, char *old_name,
  1563. int recurse)
  1564. {
  1565. LIST_HEAD(umount_list);
  1566. struct path old_path;
  1567. struct vfsmount *mnt = NULL;
  1568. int err = mount_is_safe(path);
  1569. if (err)
  1570. return err;
  1571. if (!old_name || !*old_name)
  1572. return -EINVAL;
  1573. err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
  1574. if (err)
  1575. return err;
  1576. err = lock_mount(path);
  1577. if (err)
  1578. goto out;
  1579. err = -EINVAL;
  1580. if (IS_MNT_UNBINDABLE(old_path.mnt))
  1581. goto out2;
  1582. if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
  1583. goto out2;
  1584. err = -ENOMEM;
  1585. if (recurse)
  1586. mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
  1587. else
  1588. mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
  1589. if (!mnt)
  1590. goto out2;
  1591. err = graft_tree(mnt, path);
  1592. if (err) {
  1593. br_write_lock(vfsmount_lock);
  1594. umount_tree(mnt, 0, &umount_list);
  1595. br_write_unlock(vfsmount_lock);
  1596. }
  1597. out2:
  1598. unlock_mount(path);
  1599. release_mounts(&umount_list);
  1600. out:
  1601. path_put(&old_path);
  1602. return err;
  1603. }
  1604. static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
  1605. {
  1606. int error = 0;
  1607. int readonly_request = 0;
  1608. if (ms_flags & MS_RDONLY)
  1609. readonly_request = 1;
  1610. if (readonly_request == __mnt_is_readonly(mnt))
  1611. return 0;
  1612. if (readonly_request)
  1613. error = mnt_make_readonly(mnt);
  1614. else
  1615. __mnt_unmake_readonly(mnt);
  1616. return error;
  1617. }
  1618. /*
  1619. * change filesystem flags. dir should be a physical root of filesystem.
  1620. * If you've mounted a non-root directory somewhere and want to do remount
  1621. * on it - tough luck.
  1622. */
  1623. static int do_remount(struct path *path, int flags, int mnt_flags,
  1624. void *data)
  1625. {
  1626. int err;
  1627. struct super_block *sb = path->mnt->mnt_sb;
  1628. if (!capable(CAP_SYS_ADMIN))
  1629. return -EPERM;
  1630. if (!check_mnt(path->mnt))
  1631. return -EINVAL;
  1632. if (path->dentry != path->mnt->mnt_root)
  1633. return -EINVAL;
  1634. err = security_sb_remount(sb, data);
  1635. if (err)
  1636. return err;
  1637. down_write(&sb->s_umount);
  1638. if (flags & MS_BIND)
  1639. err = change_mount_flags(path->mnt, flags);
  1640. else
  1641. err = do_remount_sb(sb, flags, data, 0);
  1642. if (!err) {
  1643. br_write_lock(vfsmount_lock);
  1644. mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
  1645. path->mnt->mnt_flags = mnt_flags;
  1646. br_write_unlock(vfsmount_lock);
  1647. }
  1648. up_write(&sb->s_umount);
  1649. if (!err) {
  1650. br_write_lock(vfsmount_lock);
  1651. touch_mnt_namespace(path->mnt->mnt_ns);
  1652. br_write_unlock(vfsmount_lock);
  1653. }
  1654. return err;
  1655. }
  1656. static inline int tree_contains_unbindable(struct vfsmount *mnt)
  1657. {
  1658. struct vfsmount *p;
  1659. for (p = mnt; p; p = next_mnt(p, mnt)) {
  1660. if (IS_MNT_UNBINDABLE(p))
  1661. return 1;
  1662. }
  1663. return 0;
  1664. }
  1665. static int do_move_mount(struct path *path, char *old_name)
  1666. {
  1667. struct path old_path, parent_path;
  1668. struct vfsmount *p;
  1669. int err = 0;
  1670. if (!capable(CAP_SYS_ADMIN))
  1671. return -EPERM;
  1672. if (!old_name || !*old_name)
  1673. return -EINVAL;
  1674. err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
  1675. if (err)
  1676. return err;
  1677. err = lock_mount(path);
  1678. if (err < 0)
  1679. goto out;
  1680. err = -EINVAL;
  1681. if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
  1682. goto out1;
  1683. if (d_unlinked(path->dentry))
  1684. goto out1;
  1685. err = -EINVAL;
  1686. if (old_path.dentry != old_path.mnt->mnt_root)
  1687. goto out1;
  1688. if (old_path.mnt == old_path.mnt->mnt_parent)
  1689. goto out1;
  1690. if (S_ISDIR(path->dentry->d_inode->i_mode) !=
  1691. S_ISDIR(old_path.dentry->d_inode->i_mode))
  1692. goto out1;
  1693. /*
  1694. * Don't move a mount residing in a shared parent.
  1695. */
  1696. if (old_path.mnt->mnt_parent &&
  1697. IS_MNT_SHARED(old_path.mnt->mnt_parent))
  1698. goto out1;
  1699. /*
  1700. * Don't move a mount tree containing unbindable mounts to a destination
  1701. * mount which is shared.
  1702. */
  1703. if (IS_MNT_SHARED(path->mnt) &&
  1704. tree_contains_unbindable(old_path.mnt))
  1705. goto out1;
  1706. err = -ELOOP;
  1707. for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
  1708. if (p == old_path.mnt)
  1709. goto out1;
  1710. err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
  1711. if (err)
  1712. goto out1;
  1713. /* if the mount is moved, it should no longer be expire
  1714. * automatically */
  1715. list_del_init(&old_path.mnt->mnt_expire);
  1716. out1:
  1717. unlock_mount(path);
  1718. out:
  1719. if (!err)
  1720. path_put(&parent_path);
  1721. path_put(&old_path);
  1722. return err;
  1723. }
  1724. static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
  1725. {
  1726. int err;
  1727. const char *subtype = strchr(fstype, '.');
  1728. if (subtype) {
  1729. subtype++;
  1730. err = -EINVAL;
  1731. if (!subtype[0])
  1732. goto err;
  1733. } else
  1734. subtype = "";
  1735. mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
  1736. err = -ENOMEM;
  1737. if (!mnt->mnt_sb->s_subtype)
  1738. goto err;
  1739. return mnt;
  1740. err:
  1741. mntput(mnt);
  1742. return ERR_PTR(err);
  1743. }
  1744. struct vfsmount *
  1745. do_kern_mount(const char *fstype, int flags, const char *name, void *data)
  1746. {
  1747. struct file_system_type *type = get_fs_type(fstype);
  1748. struct vfsmount *mnt;
  1749. if (!type)
  1750. return ERR_PTR(-ENODEV);
  1751. mnt = vfs_kern_mount(type, flags, name, data);
  1752. if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
  1753. !mnt->mnt_sb->s_subtype)
  1754. mnt = fs_set_subtype(mnt, fstype);
  1755. put_filesystem(type);
  1756. return mnt;
  1757. }
  1758. EXPORT_SYMBOL_GPL(do_kern_mount);
  1759. /*
  1760. * add a mount into a namespace's mount tree
  1761. */
  1762. static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
  1763. {
  1764. int err;
  1765. mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
  1766. err = lock_mount(path);
  1767. if (err)
  1768. return err;
  1769. err = -EINVAL;
  1770. if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
  1771. goto unlock;
  1772. /* Refuse the same filesystem on the same mount point */
  1773. err = -EBUSY;
  1774. if (path->mnt->mnt_sb == newmnt->mnt_sb &&
  1775. path->mnt->mnt_root == path->dentry)
  1776. goto unlock;
  1777. err = -EINVAL;
  1778. if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
  1779. goto unlock;
  1780. newmnt->mnt_flags = mnt_flags;
  1781. err = graft_tree(newmnt, path);
  1782. unlock:
  1783. unlock_mount(path);
  1784. return err;
  1785. }
  1786. /*
  1787. * create a new mount for userspace and request it to be added into the
  1788. * namespace's tree
  1789. */
  1790. static int do_new_mount(struct path *path, char *type, int flags,
  1791. int mnt_flags, char *name, void *data)
  1792. {
  1793. struct vfsmount *mnt;
  1794. int err;
  1795. if (!type)
  1796. return -EINVAL;
  1797. /* we need capabilities... */
  1798. if (!capable(CAP_SYS_ADMIN))
  1799. return -EPERM;
  1800. mnt = do_kern_mount(type, flags, name, data);
  1801. if (IS_ERR(mnt))
  1802. return PTR_ERR(mnt);
  1803. err = do_add_mount(mnt, path, mnt_flags);
  1804. if (err)
  1805. mntput(mnt);
  1806. return err;
  1807. }
  1808. int finish_automount(struct vfsmount *m, struct path *path)
  1809. {
  1810. int err;
  1811. /* The new mount record should have at least 2 refs to prevent it being
  1812. * expired before we get a chance to add it
  1813. */
  1814. BUG_ON(mnt_get_count(m) < 2);
  1815. if (m->mnt_sb == path->mnt->mnt_sb &&
  1816. m->mnt_root == path->dentry) {
  1817. err = -ELOOP;
  1818. goto fail;
  1819. }
  1820. err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
  1821. if (!err)
  1822. return 0;
  1823. fail:
  1824. /* remove m from any expiration list it may be on */
  1825. if (!list_empty(&m->mnt_expire)) {
  1826. down_write(&namespace_sem);
  1827. br_write_lock(vfsmount_lock);
  1828. list_del_init(&m->mnt_expire);
  1829. br_write_unlock(vfsmount_lock);
  1830. up_write(&namespace_sem);
  1831. }
  1832. mntput(m);
  1833. mntput(m);
  1834. return err;
  1835. }
  1836. /**
  1837. * mnt_set_expiry - Put a mount on an expiration list
  1838. * @mnt: The mount to list.
  1839. * @expiry_list: The list to add the mount to.
  1840. */
  1841. void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
  1842. {
  1843. down_write(&namespace_sem);
  1844. br_write_lock(vfsmount_lock);
  1845. list_add_tail(&mnt->mnt_expire, expiry_list);
  1846. br_write_unlock(vfsmount_lock);
  1847. up_write(&namespace_sem);
  1848. }
  1849. EXPORT_SYMBOL(mnt_set_expiry);
  1850. /*
  1851. * process a list of expirable mountpoints with the intent of discarding any
  1852. * mountpoints that aren't in use and haven't been touched since last we came
  1853. * here
  1854. */
  1855. void mark_mounts_for_expiry(struct list_head *mounts)
  1856. {
  1857. struct vfsmount *mnt, *next;
  1858. LIST_HEAD(graveyard);
  1859. LIST_HEAD(umounts);
  1860. if (list_empty(mounts))
  1861. return;
  1862. down_write(&namespace_sem);
  1863. br_write_lock(vfsmount_lock);
  1864. /* extract from the expiration list every vfsmount that matches the
  1865. * following criteria:
  1866. * - only referenced by its parent vfsmount
  1867. * - still marked for expiry (marked on the last call here; marks are
  1868. * cleared by mntput())
  1869. */
  1870. list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
  1871. if (!xchg(&mnt->mnt_expiry_mark, 1) ||
  1872. propagate_mount_busy(mnt, 1))
  1873. continue;
  1874. list_move(&mnt->mnt_expire, &graveyard);
  1875. }
  1876. while (!list_empty(&graveyard)) {
  1877. mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
  1878. touch_mnt_namespace(mnt->mnt_ns);
  1879. umount_tree(mnt, 1, &umounts);
  1880. }
  1881. br_write_unlock(vfsmount_lock);
  1882. up_write(&namespace_sem);
  1883. release_mounts(&umounts);
  1884. }
  1885. EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
  1886. /*
  1887. * Ripoff of 'select_parent()'
  1888. *
  1889. * search the list of submounts for a given mountpoint, and move any
  1890. * shrinkable submounts to the 'graveyard' list.
  1891. */
  1892. static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
  1893. {
  1894. struct vfsmount *this_parent = parent;
  1895. struct list_head *next;
  1896. int found = 0;
  1897. repeat:
  1898. next = this_parent->mnt_mounts.next;
  1899. resume:
  1900. while (next != &this_parent->mnt_mounts) {
  1901. struct list_head *tmp = next;
  1902. struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
  1903. next = tmp->next;
  1904. if (!(mnt->mnt_flags & MNT_SHRINKABLE))
  1905. continue;
  1906. /*
  1907. * Descend a level if the d_mounts list is non-empty.
  1908. */
  1909. if (!list_empty(&mnt->mnt_mounts)) {
  1910. this_parent = mnt;
  1911. goto repeat;
  1912. }
  1913. if (!propagate_mount_busy(mnt, 1)) {
  1914. list_move_tail(&mnt->mnt_expire, graveyard);
  1915. found++;
  1916. }
  1917. }
  1918. /*
  1919. * All done at this level ... ascend and resume the search
  1920. */
  1921. if (this_parent != parent) {
  1922. next = this_parent->mnt_child.next;
  1923. this_parent = this_parent->mnt_parent;
  1924. goto resume;
  1925. }
  1926. return found;
  1927. }
  1928. /*
  1929. * process a list of expirable mountpoints with the intent of discarding any
  1930. * submounts of a specific parent mountpoint
  1931. *
  1932. * vfsmount_lock must be held for write
  1933. */
  1934. static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
  1935. {
  1936. LIST_HEAD(graveyard);
  1937. struct vfsmount *m;
  1938. /* extract submounts of 'mountpoint' from the expiration list */
  1939. while (select_submounts(mnt, &graveyard)) {
  1940. while (!list_empty(&graveyard)) {
  1941. m = list_first_entry(&graveyard, struct vfsmount,
  1942. mnt_expire);
  1943. touch_mnt_namespace(m->mnt_ns);
  1944. umount_tree(m, 1, umounts);
  1945. }
  1946. }
  1947. }
  1948. /*
  1949. * Some copy_from_user() implementations do not return the exact number of
  1950. * bytes remaining to copy on a fault. But copy_mount_options() requires that.
  1951. * Note that this function differs from copy_from_user() in that it will oops
  1952. * on bad values of `to', rather than returning a short copy.
  1953. */
  1954. static long exact_copy_from_user(void *to, const void __user * from,
  1955. unsigned long n)
  1956. {
  1957. char *t = to;
  1958. const char __user *f = from;
  1959. char c;
  1960. if (!access_ok(VERIFY_READ, from, n))
  1961. return n;
  1962. while (n) {
  1963. if (__get_user(c, f)) {
  1964. memset(t, 0, n);
  1965. break;
  1966. }
  1967. *t++ = c;
  1968. f++;
  1969. n--;
  1970. }
  1971. return n;
  1972. }
  1973. int copy_mount_options(const void __user * data, unsigned long *where)
  1974. {
  1975. int i;
  1976. unsigned long page;
  1977. unsigned long size;
  1978. *where = 0;
  1979. if (!data)
  1980. return 0;
  1981. if (!(page = __get_free_page(GFP_KERNEL)))
  1982. return -ENOMEM;
  1983. /* We only care that *some* data at the address the user
  1984. * gave us is valid. Just in case, we'll zero
  1985. * the remainder of the page.
  1986. */
  1987. /* copy_from_user cannot cross TASK_SIZE ! */
  1988. size = TASK_SIZE - (unsigned long)data;
  1989. if (size > PAGE_SIZE)
  1990. size = PAGE_SIZE;
  1991. i = size - exact_copy_from_user((void *)page, data, size);
  1992. if (!i) {
  1993. free_page(page);
  1994. return -EFAULT;
  1995. }
  1996. if (i != PAGE_SIZE)
  1997. memset((char *)page + i, 0, PAGE_SIZE - i);
  1998. *where = page;
  1999. return 0;
  2000. }
  2001. int copy_mount_string(const void __user *data, char **where)
  2002. {
  2003. char *tmp;
  2004. if (!data) {
  2005. *where = NULL;
  2006. return 0;
  2007. }
  2008. tmp = strndup_user(data, PAGE_SIZE);
  2009. if (IS_ERR(tmp))
  2010. return PTR_ERR(tmp);
  2011. *where = tmp;
  2012. return 0;
  2013. }
  2014. /*
  2015. * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
  2016. * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
  2017. *
  2018. * data is a (void *) that can point to any structure up to
  2019. * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
  2020. * information (or be NULL).
  2021. *
  2022. * Pre-0.97 versions of mount() didn't have a flags word.
  2023. * When the flags word was introduced its top half was required
  2024. * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
  2025. * Therefore, if this magic number is present, it carries no information
  2026. * and must be discarded.
  2027. */
  2028. long do_mount(char *dev_name, char *dir_name, char *type_page,
  2029. unsigned long flags, void *data_page)
  2030. {
  2031. struct path path;
  2032. int retval = 0;
  2033. int mnt_flags = 0;
  2034. /* Discard magic */
  2035. if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
  2036. flags &= ~MS_MGC_MSK;
  2037. /* Basic sanity checks */
  2038. if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
  2039. return -EINVAL;
  2040. if (data_page)
  2041. ((char *)data_page)[PAGE_SIZE - 1] = 0;
  2042. /* ... and get the mountpoint */
  2043. retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
  2044. if (retval)
  2045. return retval;
  2046. retval = security_sb_mount(dev_name, &path,
  2047. type_page, flags, data_page);
  2048. if (retval)
  2049. goto dput_out;
  2050. /* Default to relatime unless overriden */
  2051. if (!(flags & MS_NOATIME))
  2052. mnt_flags |= MNT_RELATIME;
  2053. /* Separate the per-mountpoint flags */
  2054. if (flags & MS_NOSUID)
  2055. mnt_flags |= MNT_NOSUID;
  2056. if (flags & MS_NODEV)
  2057. mnt_flags |= MNT_NODEV;
  2058. if (flags & MS_NOEXEC)
  2059. mnt_flags |= MNT_NOEXEC;
  2060. if (flags & MS_NOATIME)
  2061. mnt_flags |= MNT_NOATIME;
  2062. if (flags & MS_NODIRATIME)
  2063. mnt_flags |= MNT_NODIRATIME;
  2064. if (flags & MS_STRICTATIME)
  2065. mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
  2066. if (flags & MS_RDONLY)
  2067. mnt_flags |= MNT_READONLY;
  2068. flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
  2069. MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
  2070. MS_STRICTATIME);
  2071. if (flags & MS_REMOUNT)
  2072. retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
  2073. data_page);
  2074. else if (flags & MS_BIND)
  2075. retval = do_loopback(&path, dev_name, flags & MS_REC);
  2076. else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
  2077. retval = do_change_type(&path, flags);
  2078. else if (flags & MS_MOVE)
  2079. retval = do_move_mount(&path, dev_name);
  2080. else
  2081. retval = do_new_mount(&path, type_page, flags, mnt_flags,
  2082. dev_name, data_page);
  2083. dput_out:
  2084. path_put(&path);
  2085. return retval;
  2086. }
  2087. static struct mnt_namespace *alloc_mnt_ns(void)
  2088. {
  2089. struct mnt_namespace *new_ns;
  2090. new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
  2091. if (!new_ns)
  2092. return ERR_PTR(-ENOMEM);
  2093. atomic_set(&new_ns->count, 1);
  2094. new_ns->root = NULL;
  2095. INIT_LIST_HEAD(&new_ns->list);
  2096. init_waitqueue_head(&new_ns->poll);
  2097. new_ns->event = 0;
  2098. return new_ns;
  2099. }
  2100. void mnt_make_longterm(struct vfsmount *mnt)
  2101. {
  2102. __mnt_make_longterm(mnt);
  2103. }
  2104. void mnt_make_shortterm(struct vfsmount *mnt)
  2105. {
  2106. #ifdef CONFIG_SMP
  2107. if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
  2108. return;
  2109. br_write_lock(vfsmount_lock);
  2110. atomic_dec(&mnt->mnt_longterm);
  2111. br_write_unlock(vfsmount_lock);
  2112. #endif
  2113. }
  2114. /*
  2115. * Allocate a new namespace structure and populate it with contents
  2116. * copied from the namespace of the passed in task structure.
  2117. */
  2118. static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
  2119. struct fs_struct *fs)
  2120. {
  2121. struct mnt_namespace *new_ns;
  2122. struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
  2123. struct vfsmount *p, *q;
  2124. new_ns = alloc_mnt_ns();
  2125. if (IS_ERR(new_ns))
  2126. return new_ns;
  2127. down_write(&namespace_sem);
  2128. /* First pass: copy the tree topology */
  2129. new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
  2130. CL_COPY_ALL | CL_EXPIRE);
  2131. if (!new_ns->root) {
  2132. up_write(&namespace_sem);
  2133. kfree(new_ns);
  2134. return ERR_PTR(-ENOMEM);
  2135. }
  2136. br_write_lock(vfsmount_lock);
  2137. list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
  2138. br_write_unlock(vfsmount_lock);
  2139. /*
  2140. * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
  2141. * as belonging to new namespace. We have already acquired a private
  2142. * fs_struct, so tsk->fs->lock is not needed.
  2143. */
  2144. p = mnt_ns->root;
  2145. q = new_ns->root;
  2146. while (p) {
  2147. q->mnt_ns = new_ns;
  2148. __mnt_make_longterm(q);
  2149. if (fs) {
  2150. if (p == fs->root.mnt) {
  2151. fs->root.mnt = mntget(q);
  2152. __mnt_make_longterm(q);
  2153. mnt_make_shortterm(p);
  2154. rootmnt = p;
  2155. }
  2156. if (p == fs->pwd.mnt) {
  2157. fs->pwd.mnt = mntget(q);
  2158. __mnt_make_longterm(q);
  2159. mnt_make_shortterm(p);
  2160. pwdmnt = p;
  2161. }
  2162. }
  2163. p = next_mnt(p, mnt_ns->root);
  2164. q = next_mnt(q, new_ns->root);
  2165. }
  2166. up_write(&namespace_sem);
  2167. if (rootmnt)
  2168. mntput(rootmnt);
  2169. if (pwdmnt)
  2170. mntput(pwdmnt);
  2171. return new_ns;
  2172. }
  2173. struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
  2174. struct fs_struct *new_fs)
  2175. {
  2176. struct mnt_namespace *new_ns;
  2177. BUG_ON(!ns);
  2178. get_mnt_ns(ns);
  2179. if (!(flags & CLONE_NEWNS))
  2180. return ns;
  2181. new_ns = dup_mnt_ns(ns, new_fs);
  2182. put_mnt_ns(ns);
  2183. return new_ns;
  2184. }
  2185. /**
  2186. * create_mnt_ns - creates a private namespace and adds a root filesystem
  2187. * @mnt: pointer to the new root filesystem mountpoint
  2188. */
  2189. struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
  2190. {
  2191. struct mnt_namespace *new_ns;
  2192. new_ns = alloc_mnt_ns();
  2193. if (!IS_ERR(new_ns)) {
  2194. mnt->mnt_ns = new_ns;
  2195. __mnt_make_longterm(mnt);
  2196. new_ns->root = mnt;
  2197. list_add(&new_ns->list, &new_ns->root->mnt_list);
  2198. }
  2199. return new_ns;
  2200. }
  2201. EXPORT_SYMBOL(create_mnt_ns);
  2202. SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
  2203. char __user *, type, unsigned long, flags, void __user *, data)
  2204. {
  2205. int ret;
  2206. char *kernel_type;
  2207. char *kernel_dir;
  2208. char *kernel_dev;
  2209. unsigned long data_page;
  2210. ret = copy_mount_string(type, &kernel_type);
  2211. if (ret < 0)
  2212. goto out_type;
  2213. kernel_dir = getname(dir_name);
  2214. if (IS_ERR(kernel_dir)) {
  2215. ret = PTR_ERR(kernel_dir);
  2216. goto out_dir;
  2217. }
  2218. ret = copy_mount_string(dev_name, &kernel_dev);
  2219. if (ret < 0)
  2220. goto out_dev;
  2221. ret = copy_mount_options(data, &data_page);
  2222. if (ret < 0)
  2223. goto out_data;
  2224. ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
  2225. (void *) data_page);
  2226. free_page(data_page);
  2227. out_data:
  2228. kfree(kernel_dev);
  2229. out_dev:
  2230. putname(kernel_dir);
  2231. out_dir:
  2232. kfree(kernel_type);
  2233. out_type:
  2234. return ret;
  2235. }
  2236. /*
  2237. * pivot_root Semantics:
  2238. * Moves the root file system of the current process to the directory put_old,
  2239. * makes new_root as the new root file system of the current process, and sets
  2240. * root/cwd of all processes which had them on the current root to new_root.
  2241. *
  2242. * Restrictions:
  2243. * The new_root and put_old must be directories, and must not be on the
  2244. * same file system as the current process root. The put_old must be
  2245. * underneath new_root, i.e. adding a non-zero number of /.. to the string
  2246. * pointed to by put_old must yield the same directory as new_root. No other
  2247. * file system may be mounted on put_old. After all, new_root is a mountpoint.
  2248. *
  2249. * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
  2250. * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
  2251. * in this situation.
  2252. *
  2253. * Notes:
  2254. * - we don't move root/cwd if they are not at the root (reason: if something
  2255. * cared enough to change them, it's probably wrong to force them elsewhere)
  2256. * - it's okay to pick a root that isn't the root of a file system, e.g.
  2257. * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
  2258. * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
  2259. * first.
  2260. */
  2261. SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
  2262. const char __user *, put_old)
  2263. {
  2264. struct vfsmount *tmp;
  2265. struct path new, old, parent_path, root_parent, root;
  2266. int error;
  2267. if (!capable(CAP_SYS_ADMIN))
  2268. return -EPERM;
  2269. error = user_path_dir(new_root, &new);
  2270. if (error)
  2271. goto out0;
  2272. error = user_path_dir(put_old, &old);
  2273. if (error)
  2274. goto out1;
  2275. error = security_sb_pivotroot(&old, &new);
  2276. if (error)
  2277. goto out2;
  2278. get_fs_root(current->fs, &root);
  2279. error = lock_mount(&old);
  2280. if (error)
  2281. goto out3;
  2282. error = -EINVAL;
  2283. if (IS_MNT_SHARED(old.mnt) ||
  2284. IS_MNT_SHARED(new.mnt->mnt_parent) ||
  2285. IS_MNT_SHARED(root.mnt->mnt_parent))
  2286. goto out4;
  2287. if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
  2288. goto out4;
  2289. error = -ENOENT;
  2290. if (d_unlinked(new.dentry))
  2291. goto out4;
  2292. if (d_unlinked(old.dentry))
  2293. goto out4;
  2294. error = -EBUSY;
  2295. if (new.mnt == root.mnt ||
  2296. old.mnt == root.mnt)
  2297. goto out4; /* loop, on the same file system */
  2298. error = -EINVAL;
  2299. if (root.mnt->mnt_root != root.dentry)
  2300. goto out4; /* not a mountpoint */
  2301. if (root.mnt->mnt_parent == root.mnt)
  2302. goto out4; /* not attached */
  2303. if (new.mnt->mnt_root != new.dentry)
  2304. goto out4; /* not a mountpoint */
  2305. if (new.mnt->mnt_parent == new.mnt)
  2306. goto out4; /* not attached */
  2307. /* make sure we can reach put_old from new_root */
  2308. tmp = old.mnt;
  2309. if (tmp != new.mnt) {
  2310. for (;;) {
  2311. if (tmp->mnt_parent == tmp)
  2312. goto out4; /* already mounted on put_old */
  2313. if (tmp->mnt_parent == new.mnt)
  2314. break;
  2315. tmp = tmp->mnt_parent;
  2316. }
  2317. if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
  2318. goto out4;
  2319. } else if (!is_subdir(old.dentry, new.dentry))
  2320. goto out4;
  2321. br_write_lock(vfsmount_lock);
  2322. detach_mnt(new.mnt, &parent_path);
  2323. detach_mnt(root.mnt, &root_parent);
  2324. /* mount old root on put_old */
  2325. attach_mnt(root.mnt, &old);
  2326. /* mount new_root on / */
  2327. attach_mnt(new.mnt, &root_parent);
  2328. touch_mnt_namespace(current->nsproxy->mnt_ns);
  2329. br_write_unlock(vfsmount_lock);
  2330. chroot_fs_refs(&root, &new);
  2331. error = 0;
  2332. out4:
  2333. unlock_mount(&old);
  2334. if (!error) {
  2335. path_put(&root_parent);
  2336. path_put(&parent_path);
  2337. }
  2338. out3:
  2339. path_put(&root);
  2340. out2:
  2341. path_put(&old);
  2342. out1:
  2343. path_put(&new);
  2344. out0:
  2345. return error;
  2346. }
  2347. static void __init init_mount_tree(void)
  2348. {
  2349. struct vfsmount *mnt;
  2350. struct mnt_namespace *ns;
  2351. struct path root;
  2352. mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
  2353. if (IS_ERR(mnt))
  2354. panic("Can't create rootfs");
  2355. ns = create_mnt_ns(mnt);
  2356. if (IS_ERR(ns))
  2357. panic("Can't allocate initial namespace");
  2358. init_task.nsproxy->mnt_ns = ns;
  2359. get_mnt_ns(ns);
  2360. root.mnt = ns->root;
  2361. root.dentry = ns->root->mnt_root;
  2362. set_fs_pwd(current->fs, &root);
  2363. set_fs_root(current->fs, &root);
  2364. }
  2365. void __init mnt_init(void)
  2366. {
  2367. unsigned u;
  2368. int err;
  2369. init_rwsem(&namespace_sem);
  2370. mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
  2371. 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
  2372. mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
  2373. if (!mount_hashtable)
  2374. panic("Failed to allocate mount hash table\n");
  2375. printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
  2376. for (u = 0; u < HASH_SIZE; u++)
  2377. INIT_LIST_HEAD(&mount_hashtable[u]);
  2378. br_lock_init(vfsmount_lock);
  2379. err = sysfs_init();
  2380. if (err)
  2381. printk(KERN_WARNING "%s: sysfs_init error: %d\n",
  2382. __func__, err);
  2383. fs_kobj = kobject_create_and_add("fs", NULL);
  2384. if (!fs_kobj)
  2385. printk(KERN_WARNING "%s: kobj create error\n", __func__);
  2386. init_rootfs();
  2387. init_mount_tree();
  2388. }
  2389. void put_mnt_ns(struct mnt_namespace *ns)
  2390. {
  2391. LIST_HEAD(umount_list);
  2392. if (!atomic_dec_and_test(&ns->count))
  2393. return;
  2394. down_write(&namespace_sem);
  2395. br_write_lock(vfsmount_lock);
  2396. umount_tree(ns->root, 0, &umount_list);
  2397. br_write_unlock(vfsmount_lock);
  2398. up_write(&namespace_sem);
  2399. release_mounts(&umount_list);
  2400. kfree(ns);
  2401. }
  2402. EXPORT_SYMBOL(put_mnt_ns);
  2403. struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
  2404. {
  2405. return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
  2406. }
  2407. EXPORT_SYMBOL_GPL(kern_mount_data);