namespace.c 71 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968
  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/export.h>
  12. #include <linux/capability.h>
  13. #include <linux/mnt_namespace.h>
  14. #include <linux/user_namespace.h>
  15. #include <linux/namei.h>
  16. #include <linux/security.h>
  17. #include <linux/idr.h>
  18. #include <linux/acct.h> /* acct_auto_close_mnt */
  19. #include <linux/init.h> /* init_rootfs */
  20. #include <linux/fs_struct.h> /* get_fs_root et.al. */
  21. #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
  22. #include <linux/uaccess.h>
  23. #include <linux/proc_ns.h>
  24. #include <linux/magic.h>
  25. #include "pnode.h"
  26. #include "internal.h"
  27. #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
  28. #define HASH_SIZE (1UL << HASH_SHIFT)
  29. static int event;
  30. static DEFINE_IDA(mnt_id_ida);
  31. static DEFINE_IDA(mnt_group_ida);
  32. static DEFINE_SPINLOCK(mnt_id_lock);
  33. static int mnt_id_start = 0;
  34. static int mnt_group_start = 1;
  35. static struct list_head *mount_hashtable __read_mostly;
  36. static struct list_head *mountpoint_hashtable __read_mostly;
  37. static struct kmem_cache *mnt_cache __read_mostly;
  38. static DECLARE_RWSEM(namespace_sem);
  39. /* /sys/fs */
  40. struct kobject *fs_kobj;
  41. EXPORT_SYMBOL_GPL(fs_kobj);
  42. /*
  43. * vfsmount lock may be taken for read to prevent changes to the
  44. * vfsmount hash, ie. during mountpoint lookups or walking back
  45. * up the tree.
  46. *
  47. * It should be taken for write in all cases where the vfsmount
  48. * tree or hash is modified or when a vfsmount structure is modified.
  49. */
  50. __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
  51. static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
  52. {
  53. unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
  54. tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
  55. tmp = tmp + (tmp >> HASH_SHIFT);
  56. return tmp & (HASH_SIZE - 1);
  57. }
  58. /*
  59. * allocation is serialized by namespace_sem, but we need the spinlock to
  60. * serialize with freeing.
  61. */
  62. static int mnt_alloc_id(struct mount *mnt)
  63. {
  64. int res;
  65. retry:
  66. ida_pre_get(&mnt_id_ida, GFP_KERNEL);
  67. spin_lock(&mnt_id_lock);
  68. res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
  69. if (!res)
  70. mnt_id_start = mnt->mnt_id + 1;
  71. spin_unlock(&mnt_id_lock);
  72. if (res == -EAGAIN)
  73. goto retry;
  74. return res;
  75. }
  76. static void mnt_free_id(struct mount *mnt)
  77. {
  78. int id = mnt->mnt_id;
  79. spin_lock(&mnt_id_lock);
  80. ida_remove(&mnt_id_ida, id);
  81. if (mnt_id_start > id)
  82. mnt_id_start = id;
  83. spin_unlock(&mnt_id_lock);
  84. }
  85. /*
  86. * Allocate a new peer group ID
  87. *
  88. * mnt_group_ida is protected by namespace_sem
  89. */
  90. static int mnt_alloc_group_id(struct mount *mnt)
  91. {
  92. int res;
  93. if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
  94. return -ENOMEM;
  95. res = ida_get_new_above(&mnt_group_ida,
  96. mnt_group_start,
  97. &mnt->mnt_group_id);
  98. if (!res)
  99. mnt_group_start = mnt->mnt_group_id + 1;
  100. return res;
  101. }
  102. /*
  103. * Release a peer group ID
  104. */
  105. void mnt_release_group_id(struct mount *mnt)
  106. {
  107. int id = mnt->mnt_group_id;
  108. ida_remove(&mnt_group_ida, id);
  109. if (mnt_group_start > id)
  110. mnt_group_start = id;
  111. mnt->mnt_group_id = 0;
  112. }
  113. /*
  114. * vfsmount lock must be held for read
  115. */
  116. static inline void mnt_add_count(struct mount *mnt, int n)
  117. {
  118. #ifdef CONFIG_SMP
  119. this_cpu_add(mnt->mnt_pcp->mnt_count, n);
  120. #else
  121. preempt_disable();
  122. mnt->mnt_count += n;
  123. preempt_enable();
  124. #endif
  125. }
  126. /*
  127. * vfsmount lock must be held for write
  128. */
  129. unsigned int mnt_get_count(struct mount *mnt)
  130. {
  131. #ifdef CONFIG_SMP
  132. unsigned int count = 0;
  133. int cpu;
  134. for_each_possible_cpu(cpu) {
  135. count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
  136. }
  137. return count;
  138. #else
  139. return mnt->mnt_count;
  140. #endif
  141. }
  142. static struct mount *alloc_vfsmnt(const char *name)
  143. {
  144. struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
  145. if (mnt) {
  146. int err;
  147. err = mnt_alloc_id(mnt);
  148. if (err)
  149. goto out_free_cache;
  150. if (name) {
  151. mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
  152. if (!mnt->mnt_devname)
  153. goto out_free_id;
  154. }
  155. #ifdef CONFIG_SMP
  156. mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
  157. if (!mnt->mnt_pcp)
  158. goto out_free_devname;
  159. this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
  160. #else
  161. mnt->mnt_count = 1;
  162. mnt->mnt_writers = 0;
  163. #endif
  164. INIT_LIST_HEAD(&mnt->mnt_hash);
  165. INIT_LIST_HEAD(&mnt->mnt_child);
  166. INIT_LIST_HEAD(&mnt->mnt_mounts);
  167. INIT_LIST_HEAD(&mnt->mnt_list);
  168. INIT_LIST_HEAD(&mnt->mnt_expire);
  169. INIT_LIST_HEAD(&mnt->mnt_share);
  170. INIT_LIST_HEAD(&mnt->mnt_slave_list);
  171. INIT_LIST_HEAD(&mnt->mnt_slave);
  172. #ifdef CONFIG_FSNOTIFY
  173. INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
  174. #endif
  175. }
  176. return mnt;
  177. #ifdef CONFIG_SMP
  178. out_free_devname:
  179. kfree(mnt->mnt_devname);
  180. #endif
  181. out_free_id:
  182. mnt_free_id(mnt);
  183. out_free_cache:
  184. kmem_cache_free(mnt_cache, mnt);
  185. return NULL;
  186. }
  187. /*
  188. * Most r/o checks on a fs are for operations that take
  189. * discrete amounts of time, like a write() or unlink().
  190. * We must keep track of when those operations start
  191. * (for permission checks) and when they end, so that
  192. * we can determine when writes are able to occur to
  193. * a filesystem.
  194. */
  195. /*
  196. * __mnt_is_readonly: check whether a mount is read-only
  197. * @mnt: the mount to check for its write status
  198. *
  199. * This shouldn't be used directly ouside of the VFS.
  200. * It does not guarantee that the filesystem will stay
  201. * r/w, just that it is right *now*. This can not and
  202. * should not be used in place of IS_RDONLY(inode).
  203. * mnt_want/drop_write() will _keep_ the filesystem
  204. * r/w.
  205. */
  206. int __mnt_is_readonly(struct vfsmount *mnt)
  207. {
  208. if (mnt->mnt_flags & MNT_READONLY)
  209. return 1;
  210. if (mnt->mnt_sb->s_flags & MS_RDONLY)
  211. return 1;
  212. return 0;
  213. }
  214. EXPORT_SYMBOL_GPL(__mnt_is_readonly);
  215. static inline void mnt_inc_writers(struct mount *mnt)
  216. {
  217. #ifdef CONFIG_SMP
  218. this_cpu_inc(mnt->mnt_pcp->mnt_writers);
  219. #else
  220. mnt->mnt_writers++;
  221. #endif
  222. }
  223. static inline void mnt_dec_writers(struct mount *mnt)
  224. {
  225. #ifdef CONFIG_SMP
  226. this_cpu_dec(mnt->mnt_pcp->mnt_writers);
  227. #else
  228. mnt->mnt_writers--;
  229. #endif
  230. }
  231. static unsigned int mnt_get_writers(struct mount *mnt)
  232. {
  233. #ifdef CONFIG_SMP
  234. unsigned int count = 0;
  235. int cpu;
  236. for_each_possible_cpu(cpu) {
  237. count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
  238. }
  239. return count;
  240. #else
  241. return mnt->mnt_writers;
  242. #endif
  243. }
  244. static int mnt_is_readonly(struct vfsmount *mnt)
  245. {
  246. if (mnt->mnt_sb->s_readonly_remount)
  247. return 1;
  248. /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
  249. smp_rmb();
  250. return __mnt_is_readonly(mnt);
  251. }
  252. /*
  253. * Most r/o & frozen checks on a fs are for operations that take discrete
  254. * amounts of time, like a write() or unlink(). We must keep track of when
  255. * those operations start (for permission checks) and when they end, so that we
  256. * can determine when writes are able to occur to a filesystem.
  257. */
  258. /**
  259. * __mnt_want_write - get write access to a mount without freeze protection
  260. * @m: the mount on which to take a write
  261. *
  262. * This tells the low-level filesystem that a write is about to be performed to
  263. * it, and makes sure that writes are allowed (mnt it read-write) before
  264. * returning success. This operation does not protect against filesystem being
  265. * frozen. When the write operation is finished, __mnt_drop_write() must be
  266. * called. This is effectively a refcount.
  267. */
  268. int __mnt_want_write(struct vfsmount *m)
  269. {
  270. struct mount *mnt = real_mount(m);
  271. int ret = 0;
  272. preempt_disable();
  273. mnt_inc_writers(mnt);
  274. /*
  275. * The store to mnt_inc_writers must be visible before we pass
  276. * MNT_WRITE_HOLD loop below, so that the slowpath can see our
  277. * incremented count after it has set MNT_WRITE_HOLD.
  278. */
  279. smp_mb();
  280. while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
  281. cpu_relax();
  282. /*
  283. * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
  284. * be set to match its requirements. So we must not load that until
  285. * MNT_WRITE_HOLD is cleared.
  286. */
  287. smp_rmb();
  288. if (mnt_is_readonly(m)) {
  289. mnt_dec_writers(mnt);
  290. ret = -EROFS;
  291. }
  292. preempt_enable();
  293. return ret;
  294. }
  295. /**
  296. * mnt_want_write - get write access to a mount
  297. * @m: the mount on which to take a write
  298. *
  299. * This tells the low-level filesystem that a write is about to be performed to
  300. * it, and makes sure that writes are allowed (mount is read-write, filesystem
  301. * is not frozen) before returning success. When the write operation is
  302. * finished, mnt_drop_write() must be called. This is effectively a refcount.
  303. */
  304. int mnt_want_write(struct vfsmount *m)
  305. {
  306. int ret;
  307. sb_start_write(m->mnt_sb);
  308. ret = __mnt_want_write(m);
  309. if (ret)
  310. sb_end_write(m->mnt_sb);
  311. return ret;
  312. }
  313. EXPORT_SYMBOL_GPL(mnt_want_write);
  314. /**
  315. * mnt_clone_write - get write access to a mount
  316. * @mnt: the mount on which to take a write
  317. *
  318. * This is effectively like mnt_want_write, except
  319. * it must only be used to take an extra write reference
  320. * on a mountpoint that we already know has a write reference
  321. * on it. This allows some optimisation.
  322. *
  323. * After finished, mnt_drop_write must be called as usual to
  324. * drop the reference.
  325. */
  326. int mnt_clone_write(struct vfsmount *mnt)
  327. {
  328. /* superblock may be r/o */
  329. if (__mnt_is_readonly(mnt))
  330. return -EROFS;
  331. preempt_disable();
  332. mnt_inc_writers(real_mount(mnt));
  333. preempt_enable();
  334. return 0;
  335. }
  336. EXPORT_SYMBOL_GPL(mnt_clone_write);
  337. /**
  338. * __mnt_want_write_file - get write access to a file's mount
  339. * @file: the file who's mount on which to take a write
  340. *
  341. * This is like __mnt_want_write, but it takes a file and can
  342. * do some optimisations if the file is open for write already
  343. */
  344. int __mnt_want_write_file(struct file *file)
  345. {
  346. struct inode *inode = file_inode(file);
  347. if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
  348. return __mnt_want_write(file->f_path.mnt);
  349. else
  350. return mnt_clone_write(file->f_path.mnt);
  351. }
  352. /**
  353. * mnt_want_write_file - get write access to a file's mount
  354. * @file: the file who's mount on which to take a write
  355. *
  356. * This is like mnt_want_write, but it takes a file and can
  357. * do some optimisations if the file is open for write already
  358. */
  359. int mnt_want_write_file(struct file *file)
  360. {
  361. int ret;
  362. sb_start_write(file->f_path.mnt->mnt_sb);
  363. ret = __mnt_want_write_file(file);
  364. if (ret)
  365. sb_end_write(file->f_path.mnt->mnt_sb);
  366. return ret;
  367. }
  368. EXPORT_SYMBOL_GPL(mnt_want_write_file);
  369. /**
  370. * __mnt_drop_write - give up write access to a mount
  371. * @mnt: the mount on which to give up write access
  372. *
  373. * Tells the low-level filesystem that we are done
  374. * performing writes to it. Must be matched with
  375. * __mnt_want_write() call above.
  376. */
  377. void __mnt_drop_write(struct vfsmount *mnt)
  378. {
  379. preempt_disable();
  380. mnt_dec_writers(real_mount(mnt));
  381. preempt_enable();
  382. }
  383. /**
  384. * mnt_drop_write - give up write access to a mount
  385. * @mnt: the mount on which to give up write access
  386. *
  387. * Tells the low-level filesystem that we are done performing writes to it and
  388. * also allows filesystem to be frozen again. Must be matched with
  389. * mnt_want_write() call above.
  390. */
  391. void mnt_drop_write(struct vfsmount *mnt)
  392. {
  393. __mnt_drop_write(mnt);
  394. sb_end_write(mnt->mnt_sb);
  395. }
  396. EXPORT_SYMBOL_GPL(mnt_drop_write);
  397. void __mnt_drop_write_file(struct file *file)
  398. {
  399. __mnt_drop_write(file->f_path.mnt);
  400. }
  401. void mnt_drop_write_file(struct file *file)
  402. {
  403. mnt_drop_write(file->f_path.mnt);
  404. }
  405. EXPORT_SYMBOL(mnt_drop_write_file);
  406. static int mnt_make_readonly(struct mount *mnt)
  407. {
  408. int ret = 0;
  409. lock_mount_hash();
  410. mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
  411. /*
  412. * After storing MNT_WRITE_HOLD, we'll read the counters. This store
  413. * should be visible before we do.
  414. */
  415. smp_mb();
  416. /*
  417. * With writers on hold, if this value is zero, then there are
  418. * definitely no active writers (although held writers may subsequently
  419. * increment the count, they'll have to wait, and decrement it after
  420. * seeing MNT_READONLY).
  421. *
  422. * It is OK to have counter incremented on one CPU and decremented on
  423. * another: the sum will add up correctly. The danger would be when we
  424. * sum up each counter, if we read a counter before it is incremented,
  425. * but then read another CPU's count which it has been subsequently
  426. * decremented from -- we would see more decrements than we should.
  427. * MNT_WRITE_HOLD protects against this scenario, because
  428. * mnt_want_write first increments count, then smp_mb, then spins on
  429. * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
  430. * we're counting up here.
  431. */
  432. if (mnt_get_writers(mnt) > 0)
  433. ret = -EBUSY;
  434. else
  435. mnt->mnt.mnt_flags |= MNT_READONLY;
  436. /*
  437. * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
  438. * that become unheld will see MNT_READONLY.
  439. */
  440. smp_wmb();
  441. mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
  442. unlock_mount_hash();
  443. return ret;
  444. }
  445. static void __mnt_unmake_readonly(struct mount *mnt)
  446. {
  447. lock_mount_hash();
  448. mnt->mnt.mnt_flags &= ~MNT_READONLY;
  449. unlock_mount_hash();
  450. }
  451. int sb_prepare_remount_readonly(struct super_block *sb)
  452. {
  453. struct mount *mnt;
  454. int err = 0;
  455. /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
  456. if (atomic_long_read(&sb->s_remove_count))
  457. return -EBUSY;
  458. lock_mount_hash();
  459. list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
  460. if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
  461. mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
  462. smp_mb();
  463. if (mnt_get_writers(mnt) > 0) {
  464. err = -EBUSY;
  465. break;
  466. }
  467. }
  468. }
  469. if (!err && atomic_long_read(&sb->s_remove_count))
  470. err = -EBUSY;
  471. if (!err) {
  472. sb->s_readonly_remount = 1;
  473. smp_wmb();
  474. }
  475. list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
  476. if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
  477. mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
  478. }
  479. unlock_mount_hash();
  480. return err;
  481. }
  482. static void free_vfsmnt(struct mount *mnt)
  483. {
  484. kfree(mnt->mnt_devname);
  485. mnt_free_id(mnt);
  486. #ifdef CONFIG_SMP
  487. free_percpu(mnt->mnt_pcp);
  488. #endif
  489. kmem_cache_free(mnt_cache, mnt);
  490. }
  491. /* call under rcu_read_lock */
  492. bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
  493. {
  494. struct mount *mnt;
  495. if (read_seqretry(&mount_lock, seq))
  496. return false;
  497. if (bastard == NULL)
  498. return true;
  499. mnt = real_mount(bastard);
  500. mnt_add_count(mnt, 1);
  501. if (likely(!read_seqretry(&mount_lock, seq)))
  502. return true;
  503. if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
  504. mnt_add_count(mnt, -1);
  505. return false;
  506. }
  507. rcu_read_unlock();
  508. mntput(bastard);
  509. rcu_read_lock();
  510. return false;
  511. }
  512. /*
  513. * find the first mount at @dentry on vfsmount @mnt.
  514. * call under rcu_read_lock()
  515. */
  516. struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
  517. {
  518. struct list_head *head = mount_hashtable + hash(mnt, dentry);
  519. struct mount *p;
  520. list_for_each_entry_rcu(p, head, mnt_hash)
  521. if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
  522. return p;
  523. return NULL;
  524. }
  525. /*
  526. * find the last mount at @dentry on vfsmount @mnt.
  527. * mount_lock must be held.
  528. */
  529. struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
  530. {
  531. struct list_head *head = mount_hashtable + hash(mnt, dentry);
  532. struct mount *p;
  533. list_for_each_entry_reverse(p, head, mnt_hash)
  534. if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
  535. return p;
  536. return NULL;
  537. }
  538. /*
  539. * lookup_mnt - Return the first child mount mounted at path
  540. *
  541. * "First" means first mounted chronologically. If you create the
  542. * following mounts:
  543. *
  544. * mount /dev/sda1 /mnt
  545. * mount /dev/sda2 /mnt
  546. * mount /dev/sda3 /mnt
  547. *
  548. * Then lookup_mnt() on the base /mnt dentry in the root mount will
  549. * return successively the root dentry and vfsmount of /dev/sda1, then
  550. * /dev/sda2, then /dev/sda3, then NULL.
  551. *
  552. * lookup_mnt takes a reference to the found vfsmount.
  553. */
  554. struct vfsmount *lookup_mnt(struct path *path)
  555. {
  556. struct mount *child_mnt;
  557. struct vfsmount *m;
  558. unsigned seq;
  559. rcu_read_lock();
  560. do {
  561. seq = read_seqbegin(&mount_lock);
  562. child_mnt = __lookup_mnt(path->mnt, path->dentry);
  563. m = child_mnt ? &child_mnt->mnt : NULL;
  564. } while (!legitimize_mnt(m, seq));
  565. rcu_read_unlock();
  566. return m;
  567. }
  568. static struct mountpoint *new_mountpoint(struct dentry *dentry)
  569. {
  570. struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry);
  571. struct mountpoint *mp;
  572. int ret;
  573. list_for_each_entry(mp, chain, m_hash) {
  574. if (mp->m_dentry == dentry) {
  575. /* might be worth a WARN_ON() */
  576. if (d_unlinked(dentry))
  577. return ERR_PTR(-ENOENT);
  578. mp->m_count++;
  579. return mp;
  580. }
  581. }
  582. mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
  583. if (!mp)
  584. return ERR_PTR(-ENOMEM);
  585. ret = d_set_mounted(dentry);
  586. if (ret) {
  587. kfree(mp);
  588. return ERR_PTR(ret);
  589. }
  590. mp->m_dentry = dentry;
  591. mp->m_count = 1;
  592. list_add(&mp->m_hash, chain);
  593. return mp;
  594. }
  595. static void put_mountpoint(struct mountpoint *mp)
  596. {
  597. if (!--mp->m_count) {
  598. struct dentry *dentry = mp->m_dentry;
  599. spin_lock(&dentry->d_lock);
  600. dentry->d_flags &= ~DCACHE_MOUNTED;
  601. spin_unlock(&dentry->d_lock);
  602. list_del(&mp->m_hash);
  603. kfree(mp);
  604. }
  605. }
  606. static inline int check_mnt(struct mount *mnt)
  607. {
  608. return mnt->mnt_ns == current->nsproxy->mnt_ns;
  609. }
  610. /*
  611. * vfsmount lock must be held for write
  612. */
  613. static void touch_mnt_namespace(struct mnt_namespace *ns)
  614. {
  615. if (ns) {
  616. ns->event = ++event;
  617. wake_up_interruptible(&ns->poll);
  618. }
  619. }
  620. /*
  621. * vfsmount lock must be held for write
  622. */
  623. static void __touch_mnt_namespace(struct mnt_namespace *ns)
  624. {
  625. if (ns && ns->event != event) {
  626. ns->event = event;
  627. wake_up_interruptible(&ns->poll);
  628. }
  629. }
  630. /*
  631. * vfsmount lock must be held for write
  632. */
  633. static void detach_mnt(struct mount *mnt, struct path *old_path)
  634. {
  635. old_path->dentry = mnt->mnt_mountpoint;
  636. old_path->mnt = &mnt->mnt_parent->mnt;
  637. mnt->mnt_parent = mnt;
  638. mnt->mnt_mountpoint = mnt->mnt.mnt_root;
  639. list_del_init(&mnt->mnt_child);
  640. list_del_init(&mnt->mnt_hash);
  641. put_mountpoint(mnt->mnt_mp);
  642. mnt->mnt_mp = NULL;
  643. }
  644. /*
  645. * vfsmount lock must be held for write
  646. */
  647. void mnt_set_mountpoint(struct mount *mnt,
  648. struct mountpoint *mp,
  649. struct mount *child_mnt)
  650. {
  651. mp->m_count++;
  652. mnt_add_count(mnt, 1); /* essentially, that's mntget */
  653. child_mnt->mnt_mountpoint = dget(mp->m_dentry);
  654. child_mnt->mnt_parent = mnt;
  655. child_mnt->mnt_mp = mp;
  656. }
  657. /*
  658. * vfsmount lock must be held for write
  659. */
  660. static void attach_mnt(struct mount *mnt,
  661. struct mount *parent,
  662. struct mountpoint *mp)
  663. {
  664. mnt_set_mountpoint(parent, mp, mnt);
  665. list_add_tail(&mnt->mnt_hash, mount_hashtable +
  666. hash(&parent->mnt, mp->m_dentry));
  667. list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
  668. }
  669. /*
  670. * vfsmount lock must be held for write
  671. */
  672. static void commit_tree(struct mount *mnt)
  673. {
  674. struct mount *parent = mnt->mnt_parent;
  675. struct mount *m;
  676. LIST_HEAD(head);
  677. struct mnt_namespace *n = parent->mnt_ns;
  678. BUG_ON(parent == mnt);
  679. list_add_tail(&head, &mnt->mnt_list);
  680. list_for_each_entry(m, &head, mnt_list)
  681. m->mnt_ns = n;
  682. list_splice(&head, n->list.prev);
  683. list_add_tail(&mnt->mnt_hash, mount_hashtable +
  684. hash(&parent->mnt, mnt->mnt_mountpoint));
  685. list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
  686. touch_mnt_namespace(n);
  687. }
  688. static struct mount *next_mnt(struct mount *p, struct mount *root)
  689. {
  690. struct list_head *next = p->mnt_mounts.next;
  691. if (next == &p->mnt_mounts) {
  692. while (1) {
  693. if (p == root)
  694. return NULL;
  695. next = p->mnt_child.next;
  696. if (next != &p->mnt_parent->mnt_mounts)
  697. break;
  698. p = p->mnt_parent;
  699. }
  700. }
  701. return list_entry(next, struct mount, mnt_child);
  702. }
  703. static struct mount *skip_mnt_tree(struct mount *p)
  704. {
  705. struct list_head *prev = p->mnt_mounts.prev;
  706. while (prev != &p->mnt_mounts) {
  707. p = list_entry(prev, struct mount, mnt_child);
  708. prev = p->mnt_mounts.prev;
  709. }
  710. return p;
  711. }
  712. struct vfsmount *
  713. vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
  714. {
  715. struct mount *mnt;
  716. struct dentry *root;
  717. if (!type)
  718. return ERR_PTR(-ENODEV);
  719. mnt = alloc_vfsmnt(name);
  720. if (!mnt)
  721. return ERR_PTR(-ENOMEM);
  722. if (flags & MS_KERNMOUNT)
  723. mnt->mnt.mnt_flags = MNT_INTERNAL;
  724. root = mount_fs(type, flags, name, data);
  725. if (IS_ERR(root)) {
  726. free_vfsmnt(mnt);
  727. return ERR_CAST(root);
  728. }
  729. mnt->mnt.mnt_root = root;
  730. mnt->mnt.mnt_sb = root->d_sb;
  731. mnt->mnt_mountpoint = mnt->mnt.mnt_root;
  732. mnt->mnt_parent = mnt;
  733. lock_mount_hash();
  734. list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
  735. unlock_mount_hash();
  736. return &mnt->mnt;
  737. }
  738. EXPORT_SYMBOL_GPL(vfs_kern_mount);
  739. static struct mount *clone_mnt(struct mount *old, struct dentry *root,
  740. int flag)
  741. {
  742. struct super_block *sb = old->mnt.mnt_sb;
  743. struct mount *mnt;
  744. int err;
  745. mnt = alloc_vfsmnt(old->mnt_devname);
  746. if (!mnt)
  747. return ERR_PTR(-ENOMEM);
  748. if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
  749. mnt->mnt_group_id = 0; /* not a peer of original */
  750. else
  751. mnt->mnt_group_id = old->mnt_group_id;
  752. if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
  753. err = mnt_alloc_group_id(mnt);
  754. if (err)
  755. goto out_free;
  756. }
  757. mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
  758. /* Don't allow unprivileged users to change mount flags */
  759. if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
  760. mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
  761. /* Don't allow unprivileged users to reveal what is under a mount */
  762. if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire))
  763. mnt->mnt.mnt_flags |= MNT_LOCKED;
  764. atomic_inc(&sb->s_active);
  765. mnt->mnt.mnt_sb = sb;
  766. mnt->mnt.mnt_root = dget(root);
  767. mnt->mnt_mountpoint = mnt->mnt.mnt_root;
  768. mnt->mnt_parent = mnt;
  769. lock_mount_hash();
  770. list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
  771. unlock_mount_hash();
  772. if ((flag & CL_SLAVE) ||
  773. ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
  774. list_add(&mnt->mnt_slave, &old->mnt_slave_list);
  775. mnt->mnt_master = old;
  776. CLEAR_MNT_SHARED(mnt);
  777. } else if (!(flag & CL_PRIVATE)) {
  778. if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
  779. list_add(&mnt->mnt_share, &old->mnt_share);
  780. if (IS_MNT_SLAVE(old))
  781. list_add(&mnt->mnt_slave, &old->mnt_slave);
  782. mnt->mnt_master = old->mnt_master;
  783. }
  784. if (flag & CL_MAKE_SHARED)
  785. set_mnt_shared(mnt);
  786. /* stick the duplicate mount on the same expiry list
  787. * as the original if that was on one */
  788. if (flag & CL_EXPIRE) {
  789. if (!list_empty(&old->mnt_expire))
  790. list_add(&mnt->mnt_expire, &old->mnt_expire);
  791. }
  792. return mnt;
  793. out_free:
  794. free_vfsmnt(mnt);
  795. return ERR_PTR(err);
  796. }
  797. static void delayed_free(struct rcu_head *head)
  798. {
  799. struct mount *mnt = container_of(head, struct mount, mnt_rcu);
  800. kfree(mnt->mnt_devname);
  801. #ifdef CONFIG_SMP
  802. free_percpu(mnt->mnt_pcp);
  803. #endif
  804. kmem_cache_free(mnt_cache, mnt);
  805. }
  806. static void mntput_no_expire(struct mount *mnt)
  807. {
  808. put_again:
  809. rcu_read_lock();
  810. mnt_add_count(mnt, -1);
  811. if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
  812. rcu_read_unlock();
  813. return;
  814. }
  815. lock_mount_hash();
  816. if (mnt_get_count(mnt)) {
  817. rcu_read_unlock();
  818. unlock_mount_hash();
  819. return;
  820. }
  821. if (unlikely(mnt->mnt_pinned)) {
  822. mnt_add_count(mnt, mnt->mnt_pinned + 1);
  823. mnt->mnt_pinned = 0;
  824. rcu_read_unlock();
  825. unlock_mount_hash();
  826. acct_auto_close_mnt(&mnt->mnt);
  827. goto put_again;
  828. }
  829. if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
  830. rcu_read_unlock();
  831. unlock_mount_hash();
  832. return;
  833. }
  834. mnt->mnt.mnt_flags |= MNT_DOOMED;
  835. rcu_read_unlock();
  836. list_del(&mnt->mnt_instance);
  837. unlock_mount_hash();
  838. /*
  839. * This probably indicates that somebody messed
  840. * up a mnt_want/drop_write() pair. If this
  841. * happens, the filesystem was probably unable
  842. * to make r/w->r/o transitions.
  843. */
  844. /*
  845. * The locking used to deal with mnt_count decrement provides barriers,
  846. * so mnt_get_writers() below is safe.
  847. */
  848. WARN_ON(mnt_get_writers(mnt));
  849. fsnotify_vfsmount_delete(&mnt->mnt);
  850. dput(mnt->mnt.mnt_root);
  851. deactivate_super(mnt->mnt.mnt_sb);
  852. mnt_free_id(mnt);
  853. call_rcu(&mnt->mnt_rcu, delayed_free);
  854. }
  855. void mntput(struct vfsmount *mnt)
  856. {
  857. if (mnt) {
  858. struct mount *m = real_mount(mnt);
  859. /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
  860. if (unlikely(m->mnt_expiry_mark))
  861. m->mnt_expiry_mark = 0;
  862. mntput_no_expire(m);
  863. }
  864. }
  865. EXPORT_SYMBOL(mntput);
  866. struct vfsmount *mntget(struct vfsmount *mnt)
  867. {
  868. if (mnt)
  869. mnt_add_count(real_mount(mnt), 1);
  870. return mnt;
  871. }
  872. EXPORT_SYMBOL(mntget);
  873. void mnt_pin(struct vfsmount *mnt)
  874. {
  875. lock_mount_hash();
  876. real_mount(mnt)->mnt_pinned++;
  877. unlock_mount_hash();
  878. }
  879. EXPORT_SYMBOL(mnt_pin);
  880. void mnt_unpin(struct vfsmount *m)
  881. {
  882. struct mount *mnt = real_mount(m);
  883. lock_mount_hash();
  884. if (mnt->mnt_pinned) {
  885. mnt_add_count(mnt, 1);
  886. mnt->mnt_pinned--;
  887. }
  888. unlock_mount_hash();
  889. }
  890. EXPORT_SYMBOL(mnt_unpin);
  891. static inline void mangle(struct seq_file *m, const char *s)
  892. {
  893. seq_escape(m, s, " \t\n\\");
  894. }
  895. /*
  896. * Simple .show_options callback for filesystems which don't want to
  897. * implement more complex mount option showing.
  898. *
  899. * See also save_mount_options().
  900. */
  901. int generic_show_options(struct seq_file *m, struct dentry *root)
  902. {
  903. const char *options;
  904. rcu_read_lock();
  905. options = rcu_dereference(root->d_sb->s_options);
  906. if (options != NULL && options[0]) {
  907. seq_putc(m, ',');
  908. mangle(m, options);
  909. }
  910. rcu_read_unlock();
  911. return 0;
  912. }
  913. EXPORT_SYMBOL(generic_show_options);
  914. /*
  915. * If filesystem uses generic_show_options(), this function should be
  916. * called from the fill_super() callback.
  917. *
  918. * The .remount_fs callback usually needs to be handled in a special
  919. * way, to make sure, that previous options are not overwritten if the
  920. * remount fails.
  921. *
  922. * Also note, that if the filesystem's .remount_fs function doesn't
  923. * reset all options to their default value, but changes only newly
  924. * given options, then the displayed options will not reflect reality
  925. * any more.
  926. */
  927. void save_mount_options(struct super_block *sb, char *options)
  928. {
  929. BUG_ON(sb->s_options);
  930. rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
  931. }
  932. EXPORT_SYMBOL(save_mount_options);
  933. void replace_mount_options(struct super_block *sb, char *options)
  934. {
  935. char *old = sb->s_options;
  936. rcu_assign_pointer(sb->s_options, options);
  937. if (old) {
  938. synchronize_rcu();
  939. kfree(old);
  940. }
  941. }
  942. EXPORT_SYMBOL(replace_mount_options);
  943. #ifdef CONFIG_PROC_FS
  944. /* iterator; we want it to have access to namespace_sem, thus here... */
  945. static void *m_start(struct seq_file *m, loff_t *pos)
  946. {
  947. struct proc_mounts *p = proc_mounts(m);
  948. down_read(&namespace_sem);
  949. return seq_list_start(&p->ns->list, *pos);
  950. }
  951. static void *m_next(struct seq_file *m, void *v, loff_t *pos)
  952. {
  953. struct proc_mounts *p = proc_mounts(m);
  954. return seq_list_next(v, &p->ns->list, pos);
  955. }
  956. static void m_stop(struct seq_file *m, void *v)
  957. {
  958. up_read(&namespace_sem);
  959. }
  960. static int m_show(struct seq_file *m, void *v)
  961. {
  962. struct proc_mounts *p = proc_mounts(m);
  963. struct mount *r = list_entry(v, struct mount, mnt_list);
  964. return p->show(m, &r->mnt);
  965. }
  966. const struct seq_operations mounts_op = {
  967. .start = m_start,
  968. .next = m_next,
  969. .stop = m_stop,
  970. .show = m_show,
  971. };
  972. #endif /* CONFIG_PROC_FS */
  973. /**
  974. * may_umount_tree - check if a mount tree is busy
  975. * @mnt: root of mount tree
  976. *
  977. * This is called to check if a tree of mounts has any
  978. * open files, pwds, chroots or sub mounts that are
  979. * busy.
  980. */
  981. int may_umount_tree(struct vfsmount *m)
  982. {
  983. struct mount *mnt = real_mount(m);
  984. int actual_refs = 0;
  985. int minimum_refs = 0;
  986. struct mount *p;
  987. BUG_ON(!m);
  988. /* write lock needed for mnt_get_count */
  989. lock_mount_hash();
  990. for (p = mnt; p; p = next_mnt(p, mnt)) {
  991. actual_refs += mnt_get_count(p);
  992. minimum_refs += 2;
  993. }
  994. unlock_mount_hash();
  995. if (actual_refs > minimum_refs)
  996. return 0;
  997. return 1;
  998. }
  999. EXPORT_SYMBOL(may_umount_tree);
  1000. /**
  1001. * may_umount - check if a mount point is busy
  1002. * @mnt: root of mount
  1003. *
  1004. * This is called to check if a mount point has any
  1005. * open files, pwds, chroots or sub mounts. If the
  1006. * mount has sub mounts this will return busy
  1007. * regardless of whether the sub mounts are busy.
  1008. *
  1009. * Doesn't take quota and stuff into account. IOW, in some cases it will
  1010. * give false negatives. The main reason why it's here is that we need
  1011. * a non-destructive way to look for easily umountable filesystems.
  1012. */
  1013. int may_umount(struct vfsmount *mnt)
  1014. {
  1015. int ret = 1;
  1016. down_read(&namespace_sem);
  1017. lock_mount_hash();
  1018. if (propagate_mount_busy(real_mount(mnt), 2))
  1019. ret = 0;
  1020. unlock_mount_hash();
  1021. up_read(&namespace_sem);
  1022. return ret;
  1023. }
  1024. EXPORT_SYMBOL(may_umount);
  1025. static LIST_HEAD(unmounted); /* protected by namespace_sem */
  1026. static void namespace_unlock(void)
  1027. {
  1028. struct mount *mnt;
  1029. LIST_HEAD(head);
  1030. if (likely(list_empty(&unmounted))) {
  1031. up_write(&namespace_sem);
  1032. return;
  1033. }
  1034. list_splice_init(&unmounted, &head);
  1035. up_write(&namespace_sem);
  1036. synchronize_rcu();
  1037. while (!list_empty(&head)) {
  1038. mnt = list_first_entry(&head, struct mount, mnt_hash);
  1039. list_del_init(&mnt->mnt_hash);
  1040. if (mnt->mnt_ex_mountpoint.mnt)
  1041. path_put(&mnt->mnt_ex_mountpoint);
  1042. mntput(&mnt->mnt);
  1043. }
  1044. }
  1045. static inline void namespace_lock(void)
  1046. {
  1047. down_write(&namespace_sem);
  1048. }
  1049. /*
  1050. * mount_lock must be held
  1051. * namespace_sem must be held for write
  1052. * how = 0 => just this tree, don't propagate
  1053. * how = 1 => propagate; we know that nobody else has reference to any victims
  1054. * how = 2 => lazy umount
  1055. */
  1056. void umount_tree(struct mount *mnt, int how)
  1057. {
  1058. LIST_HEAD(tmp_list);
  1059. struct mount *p;
  1060. for (p = mnt; p; p = next_mnt(p, mnt))
  1061. list_move(&p->mnt_hash, &tmp_list);
  1062. if (how)
  1063. propagate_umount(&tmp_list);
  1064. list_for_each_entry(p, &tmp_list, mnt_hash) {
  1065. list_del_init(&p->mnt_expire);
  1066. list_del_init(&p->mnt_list);
  1067. __touch_mnt_namespace(p->mnt_ns);
  1068. p->mnt_ns = NULL;
  1069. if (how < 2)
  1070. p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
  1071. list_del_init(&p->mnt_child);
  1072. if (mnt_has_parent(p)) {
  1073. put_mountpoint(p->mnt_mp);
  1074. /* move the reference to mountpoint into ->mnt_ex_mountpoint */
  1075. p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint;
  1076. p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt;
  1077. p->mnt_mountpoint = p->mnt.mnt_root;
  1078. p->mnt_parent = p;
  1079. p->mnt_mp = NULL;
  1080. }
  1081. change_mnt_propagation(p, MS_PRIVATE);
  1082. }
  1083. list_splice(&tmp_list, &unmounted);
  1084. }
  1085. static void shrink_submounts(struct mount *mnt);
  1086. static int do_umount(struct mount *mnt, int flags)
  1087. {
  1088. struct super_block *sb = mnt->mnt.mnt_sb;
  1089. int retval;
  1090. retval = security_sb_umount(&mnt->mnt, flags);
  1091. if (retval)
  1092. return retval;
  1093. /*
  1094. * Allow userspace to request a mountpoint be expired rather than
  1095. * unmounting unconditionally. Unmount only happens if:
  1096. * (1) the mark is already set (the mark is cleared by mntput())
  1097. * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
  1098. */
  1099. if (flags & MNT_EXPIRE) {
  1100. if (&mnt->mnt == current->fs->root.mnt ||
  1101. flags & (MNT_FORCE | MNT_DETACH))
  1102. return -EINVAL;
  1103. /*
  1104. * probably don't strictly need the lock here if we examined
  1105. * all race cases, but it's a slowpath.
  1106. */
  1107. lock_mount_hash();
  1108. if (mnt_get_count(mnt) != 2) {
  1109. unlock_mount_hash();
  1110. return -EBUSY;
  1111. }
  1112. unlock_mount_hash();
  1113. if (!xchg(&mnt->mnt_expiry_mark, 1))
  1114. return -EAGAIN;
  1115. }
  1116. /*
  1117. * If we may have to abort operations to get out of this
  1118. * mount, and they will themselves hold resources we must
  1119. * allow the fs to do things. In the Unix tradition of
  1120. * 'Gee thats tricky lets do it in userspace' the umount_begin
  1121. * might fail to complete on the first run through as other tasks
  1122. * must return, and the like. Thats for the mount program to worry
  1123. * about for the moment.
  1124. */
  1125. if (flags & MNT_FORCE && sb->s_op->umount_begin) {
  1126. sb->s_op->umount_begin(sb);
  1127. }
  1128. /*
  1129. * No sense to grab the lock for this test, but test itself looks
  1130. * somewhat bogus. Suggestions for better replacement?
  1131. * Ho-hum... In principle, we might treat that as umount + switch
  1132. * to rootfs. GC would eventually take care of the old vfsmount.
  1133. * Actually it makes sense, especially if rootfs would contain a
  1134. * /reboot - static binary that would close all descriptors and
  1135. * call reboot(9). Then init(8) could umount root and exec /reboot.
  1136. */
  1137. if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
  1138. /*
  1139. * Special case for "unmounting" root ...
  1140. * we just try to remount it readonly.
  1141. */
  1142. down_write(&sb->s_umount);
  1143. if (!(sb->s_flags & MS_RDONLY))
  1144. retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
  1145. up_write(&sb->s_umount);
  1146. return retval;
  1147. }
  1148. namespace_lock();
  1149. lock_mount_hash();
  1150. event++;
  1151. if (flags & MNT_DETACH) {
  1152. if (!list_empty(&mnt->mnt_list))
  1153. umount_tree(mnt, 2);
  1154. retval = 0;
  1155. } else {
  1156. shrink_submounts(mnt);
  1157. retval = -EBUSY;
  1158. if (!propagate_mount_busy(mnt, 2)) {
  1159. if (!list_empty(&mnt->mnt_list))
  1160. umount_tree(mnt, 1);
  1161. retval = 0;
  1162. }
  1163. }
  1164. unlock_mount_hash();
  1165. namespace_unlock();
  1166. return retval;
  1167. }
  1168. /*
  1169. * Is the caller allowed to modify his namespace?
  1170. */
  1171. static inline bool may_mount(void)
  1172. {
  1173. return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
  1174. }
  1175. /*
  1176. * Now umount can handle mount points as well as block devices.
  1177. * This is important for filesystems which use unnamed block devices.
  1178. *
  1179. * We now support a flag for forced unmount like the other 'big iron'
  1180. * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
  1181. */
  1182. SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
  1183. {
  1184. struct path path;
  1185. struct mount *mnt;
  1186. int retval;
  1187. int lookup_flags = 0;
  1188. if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
  1189. return -EINVAL;
  1190. if (!may_mount())
  1191. return -EPERM;
  1192. if (!(flags & UMOUNT_NOFOLLOW))
  1193. lookup_flags |= LOOKUP_FOLLOW;
  1194. retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
  1195. if (retval)
  1196. goto out;
  1197. mnt = real_mount(path.mnt);
  1198. retval = -EINVAL;
  1199. if (path.dentry != path.mnt->mnt_root)
  1200. goto dput_and_out;
  1201. if (!check_mnt(mnt))
  1202. goto dput_and_out;
  1203. if (mnt->mnt.mnt_flags & MNT_LOCKED)
  1204. goto dput_and_out;
  1205. retval = do_umount(mnt, flags);
  1206. dput_and_out:
  1207. /* we mustn't call path_put() as that would clear mnt_expiry_mark */
  1208. dput(path.dentry);
  1209. mntput_no_expire(mnt);
  1210. out:
  1211. return retval;
  1212. }
  1213. #ifdef __ARCH_WANT_SYS_OLDUMOUNT
  1214. /*
  1215. * The 2.0 compatible umount. No flags.
  1216. */
  1217. SYSCALL_DEFINE1(oldumount, char __user *, name)
  1218. {
  1219. return sys_umount(name, 0);
  1220. }
  1221. #endif
  1222. static bool is_mnt_ns_file(struct dentry *dentry)
  1223. {
  1224. /* Is this a proxy for a mount namespace? */
  1225. struct inode *inode = dentry->d_inode;
  1226. struct proc_ns *ei;
  1227. if (!proc_ns_inode(inode))
  1228. return false;
  1229. ei = get_proc_ns(inode);
  1230. if (ei->ns_ops != &mntns_operations)
  1231. return false;
  1232. return true;
  1233. }
  1234. static bool mnt_ns_loop(struct dentry *dentry)
  1235. {
  1236. /* Could bind mounting the mount namespace inode cause a
  1237. * mount namespace loop?
  1238. */
  1239. struct mnt_namespace *mnt_ns;
  1240. if (!is_mnt_ns_file(dentry))
  1241. return false;
  1242. mnt_ns = get_proc_ns(dentry->d_inode)->ns;
  1243. return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
  1244. }
  1245. struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
  1246. int flag)
  1247. {
  1248. struct mount *res, *p, *q, *r, *parent;
  1249. if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
  1250. return ERR_PTR(-EINVAL);
  1251. if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
  1252. return ERR_PTR(-EINVAL);
  1253. res = q = clone_mnt(mnt, dentry, flag);
  1254. if (IS_ERR(q))
  1255. return q;
  1256. q->mnt.mnt_flags &= ~MNT_LOCKED;
  1257. q->mnt_mountpoint = mnt->mnt_mountpoint;
  1258. p = mnt;
  1259. list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
  1260. struct mount *s;
  1261. if (!is_subdir(r->mnt_mountpoint, dentry))
  1262. continue;
  1263. for (s = r; s; s = next_mnt(s, r)) {
  1264. if (!(flag & CL_COPY_UNBINDABLE) &&
  1265. IS_MNT_UNBINDABLE(s)) {
  1266. s = skip_mnt_tree(s);
  1267. continue;
  1268. }
  1269. if (!(flag & CL_COPY_MNT_NS_FILE) &&
  1270. is_mnt_ns_file(s->mnt.mnt_root)) {
  1271. s = skip_mnt_tree(s);
  1272. continue;
  1273. }
  1274. while (p != s->mnt_parent) {
  1275. p = p->mnt_parent;
  1276. q = q->mnt_parent;
  1277. }
  1278. p = s;
  1279. parent = q;
  1280. q = clone_mnt(p, p->mnt.mnt_root, flag);
  1281. if (IS_ERR(q))
  1282. goto out;
  1283. lock_mount_hash();
  1284. list_add_tail(&q->mnt_list, &res->mnt_list);
  1285. attach_mnt(q, parent, p->mnt_mp);
  1286. unlock_mount_hash();
  1287. }
  1288. }
  1289. return res;
  1290. out:
  1291. if (res) {
  1292. lock_mount_hash();
  1293. umount_tree(res, 0);
  1294. unlock_mount_hash();
  1295. }
  1296. return q;
  1297. }
  1298. /* Caller should check returned pointer for errors */
  1299. struct vfsmount *collect_mounts(struct path *path)
  1300. {
  1301. struct mount *tree;
  1302. namespace_lock();
  1303. tree = copy_tree(real_mount(path->mnt), path->dentry,
  1304. CL_COPY_ALL | CL_PRIVATE);
  1305. namespace_unlock();
  1306. if (IS_ERR(tree))
  1307. return ERR_CAST(tree);
  1308. return &tree->mnt;
  1309. }
  1310. void drop_collected_mounts(struct vfsmount *mnt)
  1311. {
  1312. namespace_lock();
  1313. lock_mount_hash();
  1314. umount_tree(real_mount(mnt), 0);
  1315. unlock_mount_hash();
  1316. namespace_unlock();
  1317. }
  1318. int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
  1319. struct vfsmount *root)
  1320. {
  1321. struct mount *mnt;
  1322. int res = f(root, arg);
  1323. if (res)
  1324. return res;
  1325. list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
  1326. res = f(&mnt->mnt, arg);
  1327. if (res)
  1328. return res;
  1329. }
  1330. return 0;
  1331. }
  1332. static void cleanup_group_ids(struct mount *mnt, struct mount *end)
  1333. {
  1334. struct mount *p;
  1335. for (p = mnt; p != end; p = next_mnt(p, mnt)) {
  1336. if (p->mnt_group_id && !IS_MNT_SHARED(p))
  1337. mnt_release_group_id(p);
  1338. }
  1339. }
  1340. static int invent_group_ids(struct mount *mnt, bool recurse)
  1341. {
  1342. struct mount *p;
  1343. for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
  1344. if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
  1345. int err = mnt_alloc_group_id(p);
  1346. if (err) {
  1347. cleanup_group_ids(mnt, p);
  1348. return err;
  1349. }
  1350. }
  1351. }
  1352. return 0;
  1353. }
  1354. /*
  1355. * @source_mnt : mount tree to be attached
  1356. * @nd : place the mount tree @source_mnt is attached
  1357. * @parent_nd : if non-null, detach the source_mnt from its parent and
  1358. * store the parent mount and mountpoint dentry.
  1359. * (done when source_mnt is moved)
  1360. *
  1361. * NOTE: in the table below explains the semantics when a source mount
  1362. * of a given type is attached to a destination mount of a given type.
  1363. * ---------------------------------------------------------------------------
  1364. * | BIND MOUNT OPERATION |
  1365. * |**************************************************************************
  1366. * | source-->| shared | private | slave | unbindable |
  1367. * | dest | | | | |
  1368. * | | | | | | |
  1369. * | v | | | | |
  1370. * |**************************************************************************
  1371. * | shared | shared (++) | shared (+) | shared(+++)| invalid |
  1372. * | | | | | |
  1373. * |non-shared| shared (+) | private | slave (*) | invalid |
  1374. * ***************************************************************************
  1375. * A bind operation clones the source mount and mounts the clone on the
  1376. * destination mount.
  1377. *
  1378. * (++) the cloned mount is propagated to all the mounts in the propagation
  1379. * tree of the destination mount and the cloned mount is added to
  1380. * the peer group of the source mount.
  1381. * (+) the cloned mount is created under the destination mount and is marked
  1382. * as shared. The cloned mount is added to the peer group of the source
  1383. * mount.
  1384. * (+++) the mount is propagated to all the mounts in the propagation tree
  1385. * of the destination mount and the cloned mount is made slave
  1386. * of the same master as that of the source mount. The cloned mount
  1387. * is marked as 'shared and slave'.
  1388. * (*) the cloned mount is made a slave of the same master as that of the
  1389. * source mount.
  1390. *
  1391. * ---------------------------------------------------------------------------
  1392. * | MOVE MOUNT OPERATION |
  1393. * |**************************************************************************
  1394. * | source-->| shared | private | slave | unbindable |
  1395. * | dest | | | | |
  1396. * | | | | | | |
  1397. * | v | | | | |
  1398. * |**************************************************************************
  1399. * | shared | shared (+) | shared (+) | shared(+++) | invalid |
  1400. * | | | | | |
  1401. * |non-shared| shared (+*) | private | slave (*) | unbindable |
  1402. * ***************************************************************************
  1403. *
  1404. * (+) the mount is moved to the destination. And is then propagated to
  1405. * all the mounts in the propagation tree of the destination mount.
  1406. * (+*) the mount is moved to the destination.
  1407. * (+++) the mount is moved to the destination and is then propagated to
  1408. * all the mounts belonging to the destination mount's propagation tree.
  1409. * the mount is marked as 'shared and slave'.
  1410. * (*) the mount continues to be a slave at the new location.
  1411. *
  1412. * if the source mount is a tree, the operations explained above is
  1413. * applied to each mount in the tree.
  1414. * Must be called without spinlocks held, since this function can sleep
  1415. * in allocations.
  1416. */
  1417. static int attach_recursive_mnt(struct mount *source_mnt,
  1418. struct mount *dest_mnt,
  1419. struct mountpoint *dest_mp,
  1420. struct path *parent_path)
  1421. {
  1422. LIST_HEAD(tree_list);
  1423. struct mount *child, *p;
  1424. int err;
  1425. if (IS_MNT_SHARED(dest_mnt)) {
  1426. err = invent_group_ids(source_mnt, true);
  1427. if (err)
  1428. goto out;
  1429. }
  1430. err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
  1431. if (err)
  1432. goto out_cleanup_ids;
  1433. lock_mount_hash();
  1434. if (IS_MNT_SHARED(dest_mnt)) {
  1435. for (p = source_mnt; p; p = next_mnt(p, source_mnt))
  1436. set_mnt_shared(p);
  1437. }
  1438. if (parent_path) {
  1439. detach_mnt(source_mnt, parent_path);
  1440. attach_mnt(source_mnt, dest_mnt, dest_mp);
  1441. touch_mnt_namespace(source_mnt->mnt_ns);
  1442. } else {
  1443. mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
  1444. commit_tree(source_mnt);
  1445. }
  1446. list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
  1447. list_del_init(&child->mnt_hash);
  1448. commit_tree(child);
  1449. }
  1450. unlock_mount_hash();
  1451. return 0;
  1452. out_cleanup_ids:
  1453. if (IS_MNT_SHARED(dest_mnt))
  1454. cleanup_group_ids(source_mnt, NULL);
  1455. out:
  1456. return err;
  1457. }
  1458. static struct mountpoint *lock_mount(struct path *path)
  1459. {
  1460. struct vfsmount *mnt;
  1461. struct dentry *dentry = path->dentry;
  1462. retry:
  1463. mutex_lock(&dentry->d_inode->i_mutex);
  1464. if (unlikely(cant_mount(dentry))) {
  1465. mutex_unlock(&dentry->d_inode->i_mutex);
  1466. return ERR_PTR(-ENOENT);
  1467. }
  1468. namespace_lock();
  1469. mnt = lookup_mnt(path);
  1470. if (likely(!mnt)) {
  1471. struct mountpoint *mp = new_mountpoint(dentry);
  1472. if (IS_ERR(mp)) {
  1473. namespace_unlock();
  1474. mutex_unlock(&dentry->d_inode->i_mutex);
  1475. return mp;
  1476. }
  1477. return mp;
  1478. }
  1479. namespace_unlock();
  1480. mutex_unlock(&path->dentry->d_inode->i_mutex);
  1481. path_put(path);
  1482. path->mnt = mnt;
  1483. dentry = path->dentry = dget(mnt->mnt_root);
  1484. goto retry;
  1485. }
  1486. static void unlock_mount(struct mountpoint *where)
  1487. {
  1488. struct dentry *dentry = where->m_dentry;
  1489. put_mountpoint(where);
  1490. namespace_unlock();
  1491. mutex_unlock(&dentry->d_inode->i_mutex);
  1492. }
  1493. static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
  1494. {
  1495. if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
  1496. return -EINVAL;
  1497. if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
  1498. S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
  1499. return -ENOTDIR;
  1500. return attach_recursive_mnt(mnt, p, mp, NULL);
  1501. }
  1502. /*
  1503. * Sanity check the flags to change_mnt_propagation.
  1504. */
  1505. static int flags_to_propagation_type(int flags)
  1506. {
  1507. int type = flags & ~(MS_REC | MS_SILENT);
  1508. /* Fail if any non-propagation flags are set */
  1509. if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
  1510. return 0;
  1511. /* Only one propagation flag should be set */
  1512. if (!is_power_of_2(type))
  1513. return 0;
  1514. return type;
  1515. }
  1516. /*
  1517. * recursively change the type of the mountpoint.
  1518. */
  1519. static int do_change_type(struct path *path, int flag)
  1520. {
  1521. struct mount *m;
  1522. struct mount *mnt = real_mount(path->mnt);
  1523. int recurse = flag & MS_REC;
  1524. int type;
  1525. int err = 0;
  1526. if (path->dentry != path->mnt->mnt_root)
  1527. return -EINVAL;
  1528. type = flags_to_propagation_type(flag);
  1529. if (!type)
  1530. return -EINVAL;
  1531. namespace_lock();
  1532. if (type == MS_SHARED) {
  1533. err = invent_group_ids(mnt, recurse);
  1534. if (err)
  1535. goto out_unlock;
  1536. }
  1537. lock_mount_hash();
  1538. for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
  1539. change_mnt_propagation(m, type);
  1540. unlock_mount_hash();
  1541. out_unlock:
  1542. namespace_unlock();
  1543. return err;
  1544. }
  1545. static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
  1546. {
  1547. struct mount *child;
  1548. list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
  1549. if (!is_subdir(child->mnt_mountpoint, dentry))
  1550. continue;
  1551. if (child->mnt.mnt_flags & MNT_LOCKED)
  1552. return true;
  1553. }
  1554. return false;
  1555. }
  1556. /*
  1557. * do loopback mount.
  1558. */
  1559. static int do_loopback(struct path *path, const char *old_name,
  1560. int recurse)
  1561. {
  1562. struct path old_path;
  1563. struct mount *mnt = NULL, *old, *parent;
  1564. struct mountpoint *mp;
  1565. int err;
  1566. if (!old_name || !*old_name)
  1567. return -EINVAL;
  1568. err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
  1569. if (err)
  1570. return err;
  1571. err = -EINVAL;
  1572. if (mnt_ns_loop(old_path.dentry))
  1573. goto out;
  1574. mp = lock_mount(path);
  1575. err = PTR_ERR(mp);
  1576. if (IS_ERR(mp))
  1577. goto out;
  1578. old = real_mount(old_path.mnt);
  1579. parent = real_mount(path->mnt);
  1580. err = -EINVAL;
  1581. if (IS_MNT_UNBINDABLE(old))
  1582. goto out2;
  1583. if (!check_mnt(parent) || !check_mnt(old))
  1584. goto out2;
  1585. if (!recurse && has_locked_children(old, old_path.dentry))
  1586. goto out2;
  1587. if (recurse)
  1588. mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
  1589. else
  1590. mnt = clone_mnt(old, old_path.dentry, 0);
  1591. if (IS_ERR(mnt)) {
  1592. err = PTR_ERR(mnt);
  1593. goto out2;
  1594. }
  1595. mnt->mnt.mnt_flags &= ~MNT_LOCKED;
  1596. err = graft_tree(mnt, parent, mp);
  1597. if (err) {
  1598. lock_mount_hash();
  1599. umount_tree(mnt, 0);
  1600. unlock_mount_hash();
  1601. }
  1602. out2:
  1603. unlock_mount(mp);
  1604. out:
  1605. path_put(&old_path);
  1606. return err;
  1607. }
  1608. static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
  1609. {
  1610. int error = 0;
  1611. int readonly_request = 0;
  1612. if (ms_flags & MS_RDONLY)
  1613. readonly_request = 1;
  1614. if (readonly_request == __mnt_is_readonly(mnt))
  1615. return 0;
  1616. if (mnt->mnt_flags & MNT_LOCK_READONLY)
  1617. return -EPERM;
  1618. if (readonly_request)
  1619. error = mnt_make_readonly(real_mount(mnt));
  1620. else
  1621. __mnt_unmake_readonly(real_mount(mnt));
  1622. return error;
  1623. }
  1624. /*
  1625. * change filesystem flags. dir should be a physical root of filesystem.
  1626. * If you've mounted a non-root directory somewhere and want to do remount
  1627. * on it - tough luck.
  1628. */
  1629. static int do_remount(struct path *path, int flags, int mnt_flags,
  1630. void *data)
  1631. {
  1632. int err;
  1633. struct super_block *sb = path->mnt->mnt_sb;
  1634. struct mount *mnt = real_mount(path->mnt);
  1635. if (!check_mnt(mnt))
  1636. return -EINVAL;
  1637. if (path->dentry != path->mnt->mnt_root)
  1638. return -EINVAL;
  1639. err = security_sb_remount(sb, data);
  1640. if (err)
  1641. return err;
  1642. down_write(&sb->s_umount);
  1643. if (flags & MS_BIND)
  1644. err = change_mount_flags(path->mnt, flags);
  1645. else if (!capable(CAP_SYS_ADMIN))
  1646. err = -EPERM;
  1647. else
  1648. err = do_remount_sb(sb, flags, data, 0);
  1649. if (!err) {
  1650. lock_mount_hash();
  1651. mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
  1652. mnt->mnt.mnt_flags = mnt_flags;
  1653. touch_mnt_namespace(mnt->mnt_ns);
  1654. unlock_mount_hash();
  1655. }
  1656. up_write(&sb->s_umount);
  1657. return err;
  1658. }
  1659. static inline int tree_contains_unbindable(struct mount *mnt)
  1660. {
  1661. struct mount *p;
  1662. for (p = mnt; p; p = next_mnt(p, mnt)) {
  1663. if (IS_MNT_UNBINDABLE(p))
  1664. return 1;
  1665. }
  1666. return 0;
  1667. }
  1668. static int do_move_mount(struct path *path, const char *old_name)
  1669. {
  1670. struct path old_path, parent_path;
  1671. struct mount *p;
  1672. struct mount *old;
  1673. struct mountpoint *mp;
  1674. int err;
  1675. if (!old_name || !*old_name)
  1676. return -EINVAL;
  1677. err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
  1678. if (err)
  1679. return err;
  1680. mp = lock_mount(path);
  1681. err = PTR_ERR(mp);
  1682. if (IS_ERR(mp))
  1683. goto out;
  1684. old = real_mount(old_path.mnt);
  1685. p = real_mount(path->mnt);
  1686. err = -EINVAL;
  1687. if (!check_mnt(p) || !check_mnt(old))
  1688. goto out1;
  1689. if (old->mnt.mnt_flags & MNT_LOCKED)
  1690. goto out1;
  1691. err = -EINVAL;
  1692. if (old_path.dentry != old_path.mnt->mnt_root)
  1693. goto out1;
  1694. if (!mnt_has_parent(old))
  1695. goto out1;
  1696. if (S_ISDIR(path->dentry->d_inode->i_mode) !=
  1697. S_ISDIR(old_path.dentry->d_inode->i_mode))
  1698. goto out1;
  1699. /*
  1700. * Don't move a mount residing in a shared parent.
  1701. */
  1702. if (IS_MNT_SHARED(old->mnt_parent))
  1703. goto out1;
  1704. /*
  1705. * Don't move a mount tree containing unbindable mounts to a destination
  1706. * mount which is shared.
  1707. */
  1708. if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
  1709. goto out1;
  1710. err = -ELOOP;
  1711. for (; mnt_has_parent(p); p = p->mnt_parent)
  1712. if (p == old)
  1713. goto out1;
  1714. err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
  1715. if (err)
  1716. goto out1;
  1717. /* if the mount is moved, it should no longer be expire
  1718. * automatically */
  1719. list_del_init(&old->mnt_expire);
  1720. out1:
  1721. unlock_mount(mp);
  1722. out:
  1723. if (!err)
  1724. path_put(&parent_path);
  1725. path_put(&old_path);
  1726. return err;
  1727. }
  1728. static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
  1729. {
  1730. int err;
  1731. const char *subtype = strchr(fstype, '.');
  1732. if (subtype) {
  1733. subtype++;
  1734. err = -EINVAL;
  1735. if (!subtype[0])
  1736. goto err;
  1737. } else
  1738. subtype = "";
  1739. mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
  1740. err = -ENOMEM;
  1741. if (!mnt->mnt_sb->s_subtype)
  1742. goto err;
  1743. return mnt;
  1744. err:
  1745. mntput(mnt);
  1746. return ERR_PTR(err);
  1747. }
  1748. /*
  1749. * add a mount into a namespace's mount tree
  1750. */
  1751. static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
  1752. {
  1753. struct mountpoint *mp;
  1754. struct mount *parent;
  1755. int err;
  1756. mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL | MNT_DOOMED | MNT_SYNC_UMOUNT);
  1757. mp = lock_mount(path);
  1758. if (IS_ERR(mp))
  1759. return PTR_ERR(mp);
  1760. parent = real_mount(path->mnt);
  1761. err = -EINVAL;
  1762. if (unlikely(!check_mnt(parent))) {
  1763. /* that's acceptable only for automounts done in private ns */
  1764. if (!(mnt_flags & MNT_SHRINKABLE))
  1765. goto unlock;
  1766. /* ... and for those we'd better have mountpoint still alive */
  1767. if (!parent->mnt_ns)
  1768. goto unlock;
  1769. }
  1770. /* Refuse the same filesystem on the same mount point */
  1771. err = -EBUSY;
  1772. if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
  1773. path->mnt->mnt_root == path->dentry)
  1774. goto unlock;
  1775. err = -EINVAL;
  1776. if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
  1777. goto unlock;
  1778. newmnt->mnt.mnt_flags = mnt_flags;
  1779. err = graft_tree(newmnt, parent, mp);
  1780. unlock:
  1781. unlock_mount(mp);
  1782. return err;
  1783. }
  1784. /*
  1785. * create a new mount for userspace and request it to be added into the
  1786. * namespace's tree
  1787. */
  1788. static int do_new_mount(struct path *path, const char *fstype, int flags,
  1789. int mnt_flags, const char *name, void *data)
  1790. {
  1791. struct file_system_type *type;
  1792. struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
  1793. struct vfsmount *mnt;
  1794. int err;
  1795. if (!fstype)
  1796. return -EINVAL;
  1797. type = get_fs_type(fstype);
  1798. if (!type)
  1799. return -ENODEV;
  1800. if (user_ns != &init_user_ns) {
  1801. if (!(type->fs_flags & FS_USERNS_MOUNT)) {
  1802. put_filesystem(type);
  1803. return -EPERM;
  1804. }
  1805. /* Only in special cases allow devices from mounts
  1806. * created outside the initial user namespace.
  1807. */
  1808. if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
  1809. flags |= MS_NODEV;
  1810. mnt_flags |= MNT_NODEV;
  1811. }
  1812. }
  1813. mnt = vfs_kern_mount(type, flags, name, data);
  1814. if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
  1815. !mnt->mnt_sb->s_subtype)
  1816. mnt = fs_set_subtype(mnt, fstype);
  1817. put_filesystem(type);
  1818. if (IS_ERR(mnt))
  1819. return PTR_ERR(mnt);
  1820. err = do_add_mount(real_mount(mnt), path, mnt_flags);
  1821. if (err)
  1822. mntput(mnt);
  1823. return err;
  1824. }
  1825. int finish_automount(struct vfsmount *m, struct path *path)
  1826. {
  1827. struct mount *mnt = real_mount(m);
  1828. int err;
  1829. /* The new mount record should have at least 2 refs to prevent it being
  1830. * expired before we get a chance to add it
  1831. */
  1832. BUG_ON(mnt_get_count(mnt) < 2);
  1833. if (m->mnt_sb == path->mnt->mnt_sb &&
  1834. m->mnt_root == path->dentry) {
  1835. err = -ELOOP;
  1836. goto fail;
  1837. }
  1838. err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
  1839. if (!err)
  1840. return 0;
  1841. fail:
  1842. /* remove m from any expiration list it may be on */
  1843. if (!list_empty(&mnt->mnt_expire)) {
  1844. namespace_lock();
  1845. list_del_init(&mnt->mnt_expire);
  1846. namespace_unlock();
  1847. }
  1848. mntput(m);
  1849. mntput(m);
  1850. return err;
  1851. }
  1852. /**
  1853. * mnt_set_expiry - Put a mount on an expiration list
  1854. * @mnt: The mount to list.
  1855. * @expiry_list: The list to add the mount to.
  1856. */
  1857. void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
  1858. {
  1859. namespace_lock();
  1860. list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
  1861. namespace_unlock();
  1862. }
  1863. EXPORT_SYMBOL(mnt_set_expiry);
  1864. /*
  1865. * process a list of expirable mountpoints with the intent of discarding any
  1866. * mountpoints that aren't in use and haven't been touched since last we came
  1867. * here
  1868. */
  1869. void mark_mounts_for_expiry(struct list_head *mounts)
  1870. {
  1871. struct mount *mnt, *next;
  1872. LIST_HEAD(graveyard);
  1873. if (list_empty(mounts))
  1874. return;
  1875. namespace_lock();
  1876. lock_mount_hash();
  1877. /* extract from the expiration list every vfsmount that matches the
  1878. * following criteria:
  1879. * - only referenced by its parent vfsmount
  1880. * - still marked for expiry (marked on the last call here; marks are
  1881. * cleared by mntput())
  1882. */
  1883. list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
  1884. if (!xchg(&mnt->mnt_expiry_mark, 1) ||
  1885. propagate_mount_busy(mnt, 1))
  1886. continue;
  1887. list_move(&mnt->mnt_expire, &graveyard);
  1888. }
  1889. while (!list_empty(&graveyard)) {
  1890. mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
  1891. touch_mnt_namespace(mnt->mnt_ns);
  1892. umount_tree(mnt, 1);
  1893. }
  1894. unlock_mount_hash();
  1895. namespace_unlock();
  1896. }
  1897. EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
  1898. /*
  1899. * Ripoff of 'select_parent()'
  1900. *
  1901. * search the list of submounts for a given mountpoint, and move any
  1902. * shrinkable submounts to the 'graveyard' list.
  1903. */
  1904. static int select_submounts(struct mount *parent, struct list_head *graveyard)
  1905. {
  1906. struct mount *this_parent = parent;
  1907. struct list_head *next;
  1908. int found = 0;
  1909. repeat:
  1910. next = this_parent->mnt_mounts.next;
  1911. resume:
  1912. while (next != &this_parent->mnt_mounts) {
  1913. struct list_head *tmp = next;
  1914. struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
  1915. next = tmp->next;
  1916. if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
  1917. continue;
  1918. /*
  1919. * Descend a level if the d_mounts list is non-empty.
  1920. */
  1921. if (!list_empty(&mnt->mnt_mounts)) {
  1922. this_parent = mnt;
  1923. goto repeat;
  1924. }
  1925. if (!propagate_mount_busy(mnt, 1)) {
  1926. list_move_tail(&mnt->mnt_expire, graveyard);
  1927. found++;
  1928. }
  1929. }
  1930. /*
  1931. * All done at this level ... ascend and resume the search
  1932. */
  1933. if (this_parent != parent) {
  1934. next = this_parent->mnt_child.next;
  1935. this_parent = this_parent->mnt_parent;
  1936. goto resume;
  1937. }
  1938. return found;
  1939. }
  1940. /*
  1941. * process a list of expirable mountpoints with the intent of discarding any
  1942. * submounts of a specific parent mountpoint
  1943. *
  1944. * mount_lock must be held for write
  1945. */
  1946. static void shrink_submounts(struct mount *mnt)
  1947. {
  1948. LIST_HEAD(graveyard);
  1949. struct mount *m;
  1950. /* extract submounts of 'mountpoint' from the expiration list */
  1951. while (select_submounts(mnt, &graveyard)) {
  1952. while (!list_empty(&graveyard)) {
  1953. m = list_first_entry(&graveyard, struct mount,
  1954. mnt_expire);
  1955. touch_mnt_namespace(m->mnt_ns);
  1956. umount_tree(m, 1);
  1957. }
  1958. }
  1959. }
  1960. /*
  1961. * Some copy_from_user() implementations do not return the exact number of
  1962. * bytes remaining to copy on a fault. But copy_mount_options() requires that.
  1963. * Note that this function differs from copy_from_user() in that it will oops
  1964. * on bad values of `to', rather than returning a short copy.
  1965. */
  1966. static long exact_copy_from_user(void *to, const void __user * from,
  1967. unsigned long n)
  1968. {
  1969. char *t = to;
  1970. const char __user *f = from;
  1971. char c;
  1972. if (!access_ok(VERIFY_READ, from, n))
  1973. return n;
  1974. while (n) {
  1975. if (__get_user(c, f)) {
  1976. memset(t, 0, n);
  1977. break;
  1978. }
  1979. *t++ = c;
  1980. f++;
  1981. n--;
  1982. }
  1983. return n;
  1984. }
  1985. int copy_mount_options(const void __user * data, unsigned long *where)
  1986. {
  1987. int i;
  1988. unsigned long page;
  1989. unsigned long size;
  1990. *where = 0;
  1991. if (!data)
  1992. return 0;
  1993. if (!(page = __get_free_page(GFP_KERNEL)))
  1994. return -ENOMEM;
  1995. /* We only care that *some* data at the address the user
  1996. * gave us is valid. Just in case, we'll zero
  1997. * the remainder of the page.
  1998. */
  1999. /* copy_from_user cannot cross TASK_SIZE ! */
  2000. size = TASK_SIZE - (unsigned long)data;
  2001. if (size > PAGE_SIZE)
  2002. size = PAGE_SIZE;
  2003. i = size - exact_copy_from_user((void *)page, data, size);
  2004. if (!i) {
  2005. free_page(page);
  2006. return -EFAULT;
  2007. }
  2008. if (i != PAGE_SIZE)
  2009. memset((char *)page + i, 0, PAGE_SIZE - i);
  2010. *where = page;
  2011. return 0;
  2012. }
  2013. int copy_mount_string(const void __user *data, char **where)
  2014. {
  2015. char *tmp;
  2016. if (!data) {
  2017. *where = NULL;
  2018. return 0;
  2019. }
  2020. tmp = strndup_user(data, PAGE_SIZE);
  2021. if (IS_ERR(tmp))
  2022. return PTR_ERR(tmp);
  2023. *where = tmp;
  2024. return 0;
  2025. }
  2026. /*
  2027. * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
  2028. * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
  2029. *
  2030. * data is a (void *) that can point to any structure up to
  2031. * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
  2032. * information (or be NULL).
  2033. *
  2034. * Pre-0.97 versions of mount() didn't have a flags word.
  2035. * When the flags word was introduced its top half was required
  2036. * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
  2037. * Therefore, if this magic number is present, it carries no information
  2038. * and must be discarded.
  2039. */
  2040. long do_mount(const char *dev_name, const char *dir_name,
  2041. const char *type_page, unsigned long flags, void *data_page)
  2042. {
  2043. struct path path;
  2044. int retval = 0;
  2045. int mnt_flags = 0;
  2046. /* Discard magic */
  2047. if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
  2048. flags &= ~MS_MGC_MSK;
  2049. /* Basic sanity checks */
  2050. if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
  2051. return -EINVAL;
  2052. if (data_page)
  2053. ((char *)data_page)[PAGE_SIZE - 1] = 0;
  2054. /* ... and get the mountpoint */
  2055. retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
  2056. if (retval)
  2057. return retval;
  2058. retval = security_sb_mount(dev_name, &path,
  2059. type_page, flags, data_page);
  2060. if (!retval && !may_mount())
  2061. retval = -EPERM;
  2062. if (retval)
  2063. goto dput_out;
  2064. /* Default to relatime unless overriden */
  2065. if (!(flags & MS_NOATIME))
  2066. mnt_flags |= MNT_RELATIME;
  2067. /* Separate the per-mountpoint flags */
  2068. if (flags & MS_NOSUID)
  2069. mnt_flags |= MNT_NOSUID;
  2070. if (flags & MS_NODEV)
  2071. mnt_flags |= MNT_NODEV;
  2072. if (flags & MS_NOEXEC)
  2073. mnt_flags |= MNT_NOEXEC;
  2074. if (flags & MS_NOATIME)
  2075. mnt_flags |= MNT_NOATIME;
  2076. if (flags & MS_NODIRATIME)
  2077. mnt_flags |= MNT_NODIRATIME;
  2078. if (flags & MS_STRICTATIME)
  2079. mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
  2080. if (flags & MS_RDONLY)
  2081. mnt_flags |= MNT_READONLY;
  2082. flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
  2083. MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
  2084. MS_STRICTATIME);
  2085. if (flags & MS_REMOUNT)
  2086. retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
  2087. data_page);
  2088. else if (flags & MS_BIND)
  2089. retval = do_loopback(&path, dev_name, flags & MS_REC);
  2090. else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
  2091. retval = do_change_type(&path, flags);
  2092. else if (flags & MS_MOVE)
  2093. retval = do_move_mount(&path, dev_name);
  2094. else
  2095. retval = do_new_mount(&path, type_page, flags, mnt_flags,
  2096. dev_name, data_page);
  2097. dput_out:
  2098. path_put(&path);
  2099. return retval;
  2100. }
  2101. static void free_mnt_ns(struct mnt_namespace *ns)
  2102. {
  2103. proc_free_inum(ns->proc_inum);
  2104. put_user_ns(ns->user_ns);
  2105. kfree(ns);
  2106. }
  2107. /*
  2108. * Assign a sequence number so we can detect when we attempt to bind
  2109. * mount a reference to an older mount namespace into the current
  2110. * mount namespace, preventing reference counting loops. A 64bit
  2111. * number incrementing at 10Ghz will take 12,427 years to wrap which
  2112. * is effectively never, so we can ignore the possibility.
  2113. */
  2114. static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
  2115. static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
  2116. {
  2117. struct mnt_namespace *new_ns;
  2118. int ret;
  2119. new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
  2120. if (!new_ns)
  2121. return ERR_PTR(-ENOMEM);
  2122. ret = proc_alloc_inum(&new_ns->proc_inum);
  2123. if (ret) {
  2124. kfree(new_ns);
  2125. return ERR_PTR(ret);
  2126. }
  2127. new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
  2128. atomic_set(&new_ns->count, 1);
  2129. new_ns->root = NULL;
  2130. INIT_LIST_HEAD(&new_ns->list);
  2131. init_waitqueue_head(&new_ns->poll);
  2132. new_ns->event = 0;
  2133. new_ns->user_ns = get_user_ns(user_ns);
  2134. return new_ns;
  2135. }
  2136. struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
  2137. struct user_namespace *user_ns, struct fs_struct *new_fs)
  2138. {
  2139. struct mnt_namespace *new_ns;
  2140. struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
  2141. struct mount *p, *q;
  2142. struct mount *old;
  2143. struct mount *new;
  2144. int copy_flags;
  2145. BUG_ON(!ns);
  2146. if (likely(!(flags & CLONE_NEWNS))) {
  2147. get_mnt_ns(ns);
  2148. return ns;
  2149. }
  2150. old = ns->root;
  2151. new_ns = alloc_mnt_ns(user_ns);
  2152. if (IS_ERR(new_ns))
  2153. return new_ns;
  2154. namespace_lock();
  2155. /* First pass: copy the tree topology */
  2156. copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
  2157. if (user_ns != ns->user_ns)
  2158. copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
  2159. new = copy_tree(old, old->mnt.mnt_root, copy_flags);
  2160. if (IS_ERR(new)) {
  2161. namespace_unlock();
  2162. free_mnt_ns(new_ns);
  2163. return ERR_CAST(new);
  2164. }
  2165. new_ns->root = new;
  2166. list_add_tail(&new_ns->list, &new->mnt_list);
  2167. /*
  2168. * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
  2169. * as belonging to new namespace. We have already acquired a private
  2170. * fs_struct, so tsk->fs->lock is not needed.
  2171. */
  2172. p = old;
  2173. q = new;
  2174. while (p) {
  2175. q->mnt_ns = new_ns;
  2176. if (new_fs) {
  2177. if (&p->mnt == new_fs->root.mnt) {
  2178. new_fs->root.mnt = mntget(&q->mnt);
  2179. rootmnt = &p->mnt;
  2180. }
  2181. if (&p->mnt == new_fs->pwd.mnt) {
  2182. new_fs->pwd.mnt = mntget(&q->mnt);
  2183. pwdmnt = &p->mnt;
  2184. }
  2185. }
  2186. p = next_mnt(p, old);
  2187. q = next_mnt(q, new);
  2188. if (!q)
  2189. break;
  2190. while (p->mnt.mnt_root != q->mnt.mnt_root)
  2191. p = next_mnt(p, old);
  2192. }
  2193. namespace_unlock();
  2194. if (rootmnt)
  2195. mntput(rootmnt);
  2196. if (pwdmnt)
  2197. mntput(pwdmnt);
  2198. return new_ns;
  2199. }
  2200. /**
  2201. * create_mnt_ns - creates a private namespace and adds a root filesystem
  2202. * @mnt: pointer to the new root filesystem mountpoint
  2203. */
  2204. static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
  2205. {
  2206. struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
  2207. if (!IS_ERR(new_ns)) {
  2208. struct mount *mnt = real_mount(m);
  2209. mnt->mnt_ns = new_ns;
  2210. new_ns->root = mnt;
  2211. list_add(&mnt->mnt_list, &new_ns->list);
  2212. } else {
  2213. mntput(m);
  2214. }
  2215. return new_ns;
  2216. }
  2217. struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
  2218. {
  2219. struct mnt_namespace *ns;
  2220. struct super_block *s;
  2221. struct path path;
  2222. int err;
  2223. ns = create_mnt_ns(mnt);
  2224. if (IS_ERR(ns))
  2225. return ERR_CAST(ns);
  2226. err = vfs_path_lookup(mnt->mnt_root, mnt,
  2227. name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
  2228. put_mnt_ns(ns);
  2229. if (err)
  2230. return ERR_PTR(err);
  2231. /* trade a vfsmount reference for active sb one */
  2232. s = path.mnt->mnt_sb;
  2233. atomic_inc(&s->s_active);
  2234. mntput(path.mnt);
  2235. /* lock the sucker */
  2236. down_write(&s->s_umount);
  2237. /* ... and return the root of (sub)tree on it */
  2238. return path.dentry;
  2239. }
  2240. EXPORT_SYMBOL(mount_subtree);
  2241. SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
  2242. char __user *, type, unsigned long, flags, void __user *, data)
  2243. {
  2244. int ret;
  2245. char *kernel_type;
  2246. struct filename *kernel_dir;
  2247. char *kernel_dev;
  2248. unsigned long data_page;
  2249. ret = copy_mount_string(type, &kernel_type);
  2250. if (ret < 0)
  2251. goto out_type;
  2252. kernel_dir = getname(dir_name);
  2253. if (IS_ERR(kernel_dir)) {
  2254. ret = PTR_ERR(kernel_dir);
  2255. goto out_dir;
  2256. }
  2257. ret = copy_mount_string(dev_name, &kernel_dev);
  2258. if (ret < 0)
  2259. goto out_dev;
  2260. ret = copy_mount_options(data, &data_page);
  2261. if (ret < 0)
  2262. goto out_data;
  2263. ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
  2264. (void *) data_page);
  2265. free_page(data_page);
  2266. out_data:
  2267. kfree(kernel_dev);
  2268. out_dev:
  2269. putname(kernel_dir);
  2270. out_dir:
  2271. kfree(kernel_type);
  2272. out_type:
  2273. return ret;
  2274. }
  2275. /*
  2276. * Return true if path is reachable from root
  2277. *
  2278. * namespace_sem or mount_lock is held
  2279. */
  2280. bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
  2281. const struct path *root)
  2282. {
  2283. while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
  2284. dentry = mnt->mnt_mountpoint;
  2285. mnt = mnt->mnt_parent;
  2286. }
  2287. return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
  2288. }
  2289. int path_is_under(struct path *path1, struct path *path2)
  2290. {
  2291. int res;
  2292. read_seqlock_excl(&mount_lock);
  2293. res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
  2294. read_sequnlock_excl(&mount_lock);
  2295. return res;
  2296. }
  2297. EXPORT_SYMBOL(path_is_under);
  2298. /*
  2299. * pivot_root Semantics:
  2300. * Moves the root file system of the current process to the directory put_old,
  2301. * makes new_root as the new root file system of the current process, and sets
  2302. * root/cwd of all processes which had them on the current root to new_root.
  2303. *
  2304. * Restrictions:
  2305. * The new_root and put_old must be directories, and must not be on the
  2306. * same file system as the current process root. The put_old must be
  2307. * underneath new_root, i.e. adding a non-zero number of /.. to the string
  2308. * pointed to by put_old must yield the same directory as new_root. No other
  2309. * file system may be mounted on put_old. After all, new_root is a mountpoint.
  2310. *
  2311. * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
  2312. * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
  2313. * in this situation.
  2314. *
  2315. * Notes:
  2316. * - we don't move root/cwd if they are not at the root (reason: if something
  2317. * cared enough to change them, it's probably wrong to force them elsewhere)
  2318. * - it's okay to pick a root that isn't the root of a file system, e.g.
  2319. * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
  2320. * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
  2321. * first.
  2322. */
  2323. SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
  2324. const char __user *, put_old)
  2325. {
  2326. struct path new, old, parent_path, root_parent, root;
  2327. struct mount *new_mnt, *root_mnt, *old_mnt;
  2328. struct mountpoint *old_mp, *root_mp;
  2329. int error;
  2330. if (!may_mount())
  2331. return -EPERM;
  2332. error = user_path_dir(new_root, &new);
  2333. if (error)
  2334. goto out0;
  2335. error = user_path_dir(put_old, &old);
  2336. if (error)
  2337. goto out1;
  2338. error = security_sb_pivotroot(&old, &new);
  2339. if (error)
  2340. goto out2;
  2341. get_fs_root(current->fs, &root);
  2342. old_mp = lock_mount(&old);
  2343. error = PTR_ERR(old_mp);
  2344. if (IS_ERR(old_mp))
  2345. goto out3;
  2346. error = -EINVAL;
  2347. new_mnt = real_mount(new.mnt);
  2348. root_mnt = real_mount(root.mnt);
  2349. old_mnt = real_mount(old.mnt);
  2350. if (IS_MNT_SHARED(old_mnt) ||
  2351. IS_MNT_SHARED(new_mnt->mnt_parent) ||
  2352. IS_MNT_SHARED(root_mnt->mnt_parent))
  2353. goto out4;
  2354. if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
  2355. goto out4;
  2356. if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
  2357. goto out4;
  2358. error = -ENOENT;
  2359. if (d_unlinked(new.dentry))
  2360. goto out4;
  2361. error = -EBUSY;
  2362. if (new_mnt == root_mnt || old_mnt == root_mnt)
  2363. goto out4; /* loop, on the same file system */
  2364. error = -EINVAL;
  2365. if (root.mnt->mnt_root != root.dentry)
  2366. goto out4; /* not a mountpoint */
  2367. if (!mnt_has_parent(root_mnt))
  2368. goto out4; /* not attached */
  2369. root_mp = root_mnt->mnt_mp;
  2370. if (new.mnt->mnt_root != new.dentry)
  2371. goto out4; /* not a mountpoint */
  2372. if (!mnt_has_parent(new_mnt))
  2373. goto out4; /* not attached */
  2374. /* make sure we can reach put_old from new_root */
  2375. if (!is_path_reachable(old_mnt, old.dentry, &new))
  2376. goto out4;
  2377. root_mp->m_count++; /* pin it so it won't go away */
  2378. lock_mount_hash();
  2379. detach_mnt(new_mnt, &parent_path);
  2380. detach_mnt(root_mnt, &root_parent);
  2381. if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
  2382. new_mnt->mnt.mnt_flags |= MNT_LOCKED;
  2383. root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
  2384. }
  2385. /* mount old root on put_old */
  2386. attach_mnt(root_mnt, old_mnt, old_mp);
  2387. /* mount new_root on / */
  2388. attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
  2389. touch_mnt_namespace(current->nsproxy->mnt_ns);
  2390. unlock_mount_hash();
  2391. chroot_fs_refs(&root, &new);
  2392. put_mountpoint(root_mp);
  2393. error = 0;
  2394. out4:
  2395. unlock_mount(old_mp);
  2396. if (!error) {
  2397. path_put(&root_parent);
  2398. path_put(&parent_path);
  2399. }
  2400. out3:
  2401. path_put(&root);
  2402. out2:
  2403. path_put(&old);
  2404. out1:
  2405. path_put(&new);
  2406. out0:
  2407. return error;
  2408. }
  2409. static void __init init_mount_tree(void)
  2410. {
  2411. struct vfsmount *mnt;
  2412. struct mnt_namespace *ns;
  2413. struct path root;
  2414. struct file_system_type *type;
  2415. type = get_fs_type("rootfs");
  2416. if (!type)
  2417. panic("Can't find rootfs type");
  2418. mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
  2419. put_filesystem(type);
  2420. if (IS_ERR(mnt))
  2421. panic("Can't create rootfs");
  2422. ns = create_mnt_ns(mnt);
  2423. if (IS_ERR(ns))
  2424. panic("Can't allocate initial namespace");
  2425. init_task.nsproxy->mnt_ns = ns;
  2426. get_mnt_ns(ns);
  2427. root.mnt = mnt;
  2428. root.dentry = mnt->mnt_root;
  2429. set_fs_pwd(current->fs, &root);
  2430. set_fs_root(current->fs, &root);
  2431. }
  2432. void __init mnt_init(void)
  2433. {
  2434. unsigned u;
  2435. int err;
  2436. mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
  2437. 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
  2438. mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
  2439. mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
  2440. if (!mount_hashtable || !mountpoint_hashtable)
  2441. panic("Failed to allocate mount hash table\n");
  2442. printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
  2443. for (u = 0; u < HASH_SIZE; u++)
  2444. INIT_LIST_HEAD(&mount_hashtable[u]);
  2445. for (u = 0; u < HASH_SIZE; u++)
  2446. INIT_LIST_HEAD(&mountpoint_hashtable[u]);
  2447. err = sysfs_init();
  2448. if (err)
  2449. printk(KERN_WARNING "%s: sysfs_init error: %d\n",
  2450. __func__, err);
  2451. fs_kobj = kobject_create_and_add("fs", NULL);
  2452. if (!fs_kobj)
  2453. printk(KERN_WARNING "%s: kobj create error\n", __func__);
  2454. init_rootfs();
  2455. init_mount_tree();
  2456. }
  2457. void put_mnt_ns(struct mnt_namespace *ns)
  2458. {
  2459. if (!atomic_dec_and_test(&ns->count))
  2460. return;
  2461. drop_collected_mounts(&ns->root->mnt);
  2462. free_mnt_ns(ns);
  2463. }
  2464. struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
  2465. {
  2466. struct vfsmount *mnt;
  2467. mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
  2468. if (!IS_ERR(mnt)) {
  2469. /*
  2470. * it is a longterm mount, don't release mnt until
  2471. * we unmount before file sys is unregistered
  2472. */
  2473. real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
  2474. }
  2475. return mnt;
  2476. }
  2477. EXPORT_SYMBOL_GPL(kern_mount_data);
  2478. void kern_unmount(struct vfsmount *mnt)
  2479. {
  2480. /* release long term mount so mount point can be released */
  2481. if (!IS_ERR_OR_NULL(mnt)) {
  2482. real_mount(mnt)->mnt_ns = NULL;
  2483. synchronize_rcu(); /* yecchhh... */
  2484. mntput(mnt);
  2485. }
  2486. }
  2487. EXPORT_SYMBOL(kern_unmount);
  2488. bool our_mnt(struct vfsmount *mnt)
  2489. {
  2490. return check_mnt(real_mount(mnt));
  2491. }
  2492. bool current_chrooted(void)
  2493. {
  2494. /* Does the current process have a non-standard root */
  2495. struct path ns_root;
  2496. struct path fs_root;
  2497. bool chrooted;
  2498. /* Find the namespace root */
  2499. ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
  2500. ns_root.dentry = ns_root.mnt->mnt_root;
  2501. path_get(&ns_root);
  2502. while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
  2503. ;
  2504. get_fs_root(current->fs, &fs_root);
  2505. chrooted = !path_equal(&fs_root, &ns_root);
  2506. path_put(&fs_root);
  2507. path_put(&ns_root);
  2508. return chrooted;
  2509. }
  2510. bool fs_fully_visible(struct file_system_type *type)
  2511. {
  2512. struct mnt_namespace *ns = current->nsproxy->mnt_ns;
  2513. struct mount *mnt;
  2514. bool visible = false;
  2515. if (unlikely(!ns))
  2516. return false;
  2517. down_read(&namespace_sem);
  2518. list_for_each_entry(mnt, &ns->list, mnt_list) {
  2519. struct mount *child;
  2520. if (mnt->mnt.mnt_sb->s_type != type)
  2521. continue;
  2522. /* This mount is not fully visible if there are any child mounts
  2523. * that cover anything except for empty directories.
  2524. */
  2525. list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
  2526. struct inode *inode = child->mnt_mountpoint->d_inode;
  2527. if (!S_ISDIR(inode->i_mode))
  2528. goto next;
  2529. if (inode->i_nlink != 2)
  2530. goto next;
  2531. }
  2532. visible = true;
  2533. goto found;
  2534. next: ;
  2535. }
  2536. found:
  2537. up_read(&namespace_sem);
  2538. return visible;
  2539. }
  2540. static void *mntns_get(struct task_struct *task)
  2541. {
  2542. struct mnt_namespace *ns = NULL;
  2543. struct nsproxy *nsproxy;
  2544. rcu_read_lock();
  2545. nsproxy = task_nsproxy(task);
  2546. if (nsproxy) {
  2547. ns = nsproxy->mnt_ns;
  2548. get_mnt_ns(ns);
  2549. }
  2550. rcu_read_unlock();
  2551. return ns;
  2552. }
  2553. static void mntns_put(void *ns)
  2554. {
  2555. put_mnt_ns(ns);
  2556. }
  2557. static int mntns_install(struct nsproxy *nsproxy, void *ns)
  2558. {
  2559. struct fs_struct *fs = current->fs;
  2560. struct mnt_namespace *mnt_ns = ns;
  2561. struct path root;
  2562. if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
  2563. !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
  2564. !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
  2565. return -EPERM;
  2566. if (fs->users != 1)
  2567. return -EINVAL;
  2568. get_mnt_ns(mnt_ns);
  2569. put_mnt_ns(nsproxy->mnt_ns);
  2570. nsproxy->mnt_ns = mnt_ns;
  2571. /* Find the root */
  2572. root.mnt = &mnt_ns->root->mnt;
  2573. root.dentry = mnt_ns->root->mnt.mnt_root;
  2574. path_get(&root);
  2575. while(d_mountpoint(root.dentry) && follow_down_one(&root))
  2576. ;
  2577. /* Update the pwd and root */
  2578. set_fs_pwd(fs, &root);
  2579. set_fs_root(fs, &root);
  2580. path_put(&root);
  2581. return 0;
  2582. }
  2583. static unsigned int mntns_inum(void *ns)
  2584. {
  2585. struct mnt_namespace *mnt_ns = ns;
  2586. return mnt_ns->proc_inum;
  2587. }
  2588. const struct proc_ns_operations mntns_operations = {
  2589. .name = "mnt",
  2590. .type = CLONE_NEWNS,
  2591. .get = mntns_get,
  2592. .put = mntns_put,
  2593. .install = mntns_install,
  2594. .inum = mntns_inum,
  2595. };