namespace.c 55 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164
  1. /*
  2. * linux/fs/namespace.c
  3. *
  4. * (C) Copyright Al Viro 2000, 2001
  5. * Released under GPL v2.
  6. *
  7. * Based on code from fs/super.c, copyright Linus Torvalds and others.
  8. * Heavily rewritten.
  9. */
  10. #include <linux/syscalls.h>
  11. #include <linux/slab.h>
  12. #include <linux/sched.h>
  13. #include <linux/smp_lock.h>
  14. #include <linux/init.h>
  15. #include <linux/kernel.h>
  16. #include <linux/quotaops.h>
  17. #include <linux/acct.h>
  18. #include <linux/capability.h>
  19. #include <linux/cpumask.h>
  20. #include <linux/module.h>
  21. #include <linux/sysfs.h>
  22. #include <linux/seq_file.h>
  23. #include <linux/mnt_namespace.h>
  24. #include <linux/namei.h>
  25. #include <linux/security.h>
  26. #include <linux/mount.h>
  27. #include <linux/ramfs.h>
  28. #include <linux/log2.h>
  29. #include <asm/uaccess.h>
  30. #include <asm/unistd.h>
  31. #include "pnode.h"
  32. #include "internal.h"
  33. #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
  34. #define HASH_SIZE (1UL << HASH_SHIFT)
  35. /* spinlock for vfsmount related operations, inplace of dcache_lock */
  36. __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
  37. static int event;
  38. static struct list_head *mount_hashtable __read_mostly;
  39. static struct kmem_cache *mnt_cache __read_mostly;
  40. static struct rw_semaphore namespace_sem;
  41. /* /sys/fs */
  42. struct kobject *fs_kobj;
  43. EXPORT_SYMBOL_GPL(fs_kobj);
  44. static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
  45. {
  46. unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
  47. tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
  48. tmp = tmp + (tmp >> HASH_SHIFT);
  49. return tmp & (HASH_SIZE - 1);
  50. }
  51. #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
  52. struct vfsmount *alloc_vfsmnt(const char *name)
  53. {
  54. struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
  55. if (mnt) {
  56. atomic_set(&mnt->mnt_count, 1);
  57. INIT_LIST_HEAD(&mnt->mnt_hash);
  58. INIT_LIST_HEAD(&mnt->mnt_child);
  59. INIT_LIST_HEAD(&mnt->mnt_mounts);
  60. INIT_LIST_HEAD(&mnt->mnt_list);
  61. INIT_LIST_HEAD(&mnt->mnt_expire);
  62. INIT_LIST_HEAD(&mnt->mnt_share);
  63. INIT_LIST_HEAD(&mnt->mnt_slave_list);
  64. INIT_LIST_HEAD(&mnt->mnt_slave);
  65. atomic_set(&mnt->__mnt_writers, 0);
  66. if (name) {
  67. int size = strlen(name) + 1;
  68. char *newname = kmalloc(size, GFP_KERNEL);
  69. if (newname) {
  70. memcpy(newname, name, size);
  71. mnt->mnt_devname = newname;
  72. }
  73. }
  74. }
  75. return mnt;
  76. }
  77. /*
  78. * Most r/o checks on a fs are for operations that take
  79. * discrete amounts of time, like a write() or unlink().
  80. * We must keep track of when those operations start
  81. * (for permission checks) and when they end, so that
  82. * we can determine when writes are able to occur to
  83. * a filesystem.
  84. */
  85. /*
  86. * __mnt_is_readonly: check whether a mount is read-only
  87. * @mnt: the mount to check for its write status
  88. *
  89. * This shouldn't be used directly ouside of the VFS.
  90. * It does not guarantee that the filesystem will stay
  91. * r/w, just that it is right *now*. This can not and
  92. * should not be used in place of IS_RDONLY(inode).
  93. * mnt_want/drop_write() will _keep_ the filesystem
  94. * r/w.
  95. */
  96. int __mnt_is_readonly(struct vfsmount *mnt)
  97. {
  98. if (mnt->mnt_flags & MNT_READONLY)
  99. return 1;
  100. if (mnt->mnt_sb->s_flags & MS_RDONLY)
  101. return 1;
  102. return 0;
  103. }
  104. EXPORT_SYMBOL_GPL(__mnt_is_readonly);
  105. struct mnt_writer {
  106. /*
  107. * If holding multiple instances of this lock, they
  108. * must be ordered by cpu number.
  109. */
  110. spinlock_t lock;
  111. struct lock_class_key lock_class; /* compiles out with !lockdep */
  112. unsigned long count;
  113. struct vfsmount *mnt;
  114. } ____cacheline_aligned_in_smp;
  115. static DEFINE_PER_CPU(struct mnt_writer, mnt_writers);
  116. static int __init init_mnt_writers(void)
  117. {
  118. int cpu;
  119. for_each_possible_cpu(cpu) {
  120. struct mnt_writer *writer = &per_cpu(mnt_writers, cpu);
  121. spin_lock_init(&writer->lock);
  122. lockdep_set_class(&writer->lock, &writer->lock_class);
  123. writer->count = 0;
  124. }
  125. return 0;
  126. }
  127. fs_initcall(init_mnt_writers);
  128. static void unlock_mnt_writers(void)
  129. {
  130. int cpu;
  131. struct mnt_writer *cpu_writer;
  132. for_each_possible_cpu(cpu) {
  133. cpu_writer = &per_cpu(mnt_writers, cpu);
  134. spin_unlock(&cpu_writer->lock);
  135. }
  136. }
  137. static inline void __clear_mnt_count(struct mnt_writer *cpu_writer)
  138. {
  139. if (!cpu_writer->mnt)
  140. return;
  141. /*
  142. * This is in case anyone ever leaves an invalid,
  143. * old ->mnt and a count of 0.
  144. */
  145. if (!cpu_writer->count)
  146. return;
  147. atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers);
  148. cpu_writer->count = 0;
  149. }
  150. /*
  151. * must hold cpu_writer->lock
  152. */
  153. static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer,
  154. struct vfsmount *mnt)
  155. {
  156. if (cpu_writer->mnt == mnt)
  157. return;
  158. __clear_mnt_count(cpu_writer);
  159. cpu_writer->mnt = mnt;
  160. }
  161. /*
  162. * Most r/o checks on a fs are for operations that take
  163. * discrete amounts of time, like a write() or unlink().
  164. * We must keep track of when those operations start
  165. * (for permission checks) and when they end, so that
  166. * we can determine when writes are able to occur to
  167. * a filesystem.
  168. */
  169. /**
  170. * mnt_want_write - get write access to a mount
  171. * @mnt: the mount on which to take a write
  172. *
  173. * This tells the low-level filesystem that a write is
  174. * about to be performed to it, and makes sure that
  175. * writes are allowed before returning success. When
  176. * the write operation is finished, mnt_drop_write()
  177. * must be called. This is effectively a refcount.
  178. */
  179. int mnt_want_write(struct vfsmount *mnt)
  180. {
  181. int ret = 0;
  182. struct mnt_writer *cpu_writer;
  183. cpu_writer = &get_cpu_var(mnt_writers);
  184. spin_lock(&cpu_writer->lock);
  185. if (__mnt_is_readonly(mnt)) {
  186. ret = -EROFS;
  187. goto out;
  188. }
  189. use_cpu_writer_for_mount(cpu_writer, mnt);
  190. cpu_writer->count++;
  191. out:
  192. spin_unlock(&cpu_writer->lock);
  193. put_cpu_var(mnt_writers);
  194. return ret;
  195. }
  196. EXPORT_SYMBOL_GPL(mnt_want_write);
  197. static void lock_mnt_writers(void)
  198. {
  199. int cpu;
  200. struct mnt_writer *cpu_writer;
  201. for_each_possible_cpu(cpu) {
  202. cpu_writer = &per_cpu(mnt_writers, cpu);
  203. spin_lock(&cpu_writer->lock);
  204. __clear_mnt_count(cpu_writer);
  205. cpu_writer->mnt = NULL;
  206. }
  207. }
  208. /*
  209. * These per-cpu write counts are not guaranteed to have
  210. * matched increments and decrements on any given cpu.
  211. * A file open()ed for write on one cpu and close()d on
  212. * another cpu will imbalance this count. Make sure it
  213. * does not get too far out of whack.
  214. */
  215. static void handle_write_count_underflow(struct vfsmount *mnt)
  216. {
  217. if (atomic_read(&mnt->__mnt_writers) >=
  218. MNT_WRITER_UNDERFLOW_LIMIT)
  219. return;
  220. /*
  221. * It isn't necessary to hold all of the locks
  222. * at the same time, but doing it this way makes
  223. * us share a lot more code.
  224. */
  225. lock_mnt_writers();
  226. /*
  227. * vfsmount_lock is for mnt_flags.
  228. */
  229. spin_lock(&vfsmount_lock);
  230. /*
  231. * If coalescing the per-cpu writer counts did not
  232. * get us back to a positive writer count, we have
  233. * a bug.
  234. */
  235. if ((atomic_read(&mnt->__mnt_writers) < 0) &&
  236. !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) {
  237. printk(KERN_DEBUG "leak detected on mount(%p) writers "
  238. "count: %d\n",
  239. mnt, atomic_read(&mnt->__mnt_writers));
  240. WARN_ON(1);
  241. /* use the flag to keep the dmesg spam down */
  242. mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT;
  243. }
  244. spin_unlock(&vfsmount_lock);
  245. unlock_mnt_writers();
  246. }
  247. /**
  248. * mnt_drop_write - give up write access to a mount
  249. * @mnt: the mount on which to give up write access
  250. *
  251. * Tells the low-level filesystem that we are done
  252. * performing writes to it. Must be matched with
  253. * mnt_want_write() call above.
  254. */
  255. void mnt_drop_write(struct vfsmount *mnt)
  256. {
  257. int must_check_underflow = 0;
  258. struct mnt_writer *cpu_writer;
  259. cpu_writer = &get_cpu_var(mnt_writers);
  260. spin_lock(&cpu_writer->lock);
  261. use_cpu_writer_for_mount(cpu_writer, mnt);
  262. if (cpu_writer->count > 0) {
  263. cpu_writer->count--;
  264. } else {
  265. must_check_underflow = 1;
  266. atomic_dec(&mnt->__mnt_writers);
  267. }
  268. spin_unlock(&cpu_writer->lock);
  269. /*
  270. * Logically, we could call this each time,
  271. * but the __mnt_writers cacheline tends to
  272. * be cold, and makes this expensive.
  273. */
  274. if (must_check_underflow)
  275. handle_write_count_underflow(mnt);
  276. /*
  277. * This could be done right after the spinlock
  278. * is taken because the spinlock keeps us on
  279. * the cpu, and disables preemption. However,
  280. * putting it here bounds the amount that
  281. * __mnt_writers can underflow. Without it,
  282. * we could theoretically wrap __mnt_writers.
  283. */
  284. put_cpu_var(mnt_writers);
  285. }
  286. EXPORT_SYMBOL_GPL(mnt_drop_write);
  287. static int mnt_make_readonly(struct vfsmount *mnt)
  288. {
  289. int ret = 0;
  290. lock_mnt_writers();
  291. /*
  292. * With all the locks held, this value is stable
  293. */
  294. if (atomic_read(&mnt->__mnt_writers) > 0) {
  295. ret = -EBUSY;
  296. goto out;
  297. }
  298. /*
  299. * nobody can do a successful mnt_want_write() with all
  300. * of the counts in MNT_DENIED_WRITE and the locks held.
  301. */
  302. spin_lock(&vfsmount_lock);
  303. if (!ret)
  304. mnt->mnt_flags |= MNT_READONLY;
  305. spin_unlock(&vfsmount_lock);
  306. out:
  307. unlock_mnt_writers();
  308. return ret;
  309. }
  310. static void __mnt_unmake_readonly(struct vfsmount *mnt)
  311. {
  312. spin_lock(&vfsmount_lock);
  313. mnt->mnt_flags &= ~MNT_READONLY;
  314. spin_unlock(&vfsmount_lock);
  315. }
  316. int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
  317. {
  318. mnt->mnt_sb = sb;
  319. mnt->mnt_root = dget(sb->s_root);
  320. return 0;
  321. }
  322. EXPORT_SYMBOL(simple_set_mnt);
  323. void free_vfsmnt(struct vfsmount *mnt)
  324. {
  325. kfree(mnt->mnt_devname);
  326. kmem_cache_free(mnt_cache, mnt);
  327. }
  328. /*
  329. * find the first or last mount at @dentry on vfsmount @mnt depending on
  330. * @dir. If @dir is set return the first mount else return the last mount.
  331. */
  332. struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
  333. int dir)
  334. {
  335. struct list_head *head = mount_hashtable + hash(mnt, dentry);
  336. struct list_head *tmp = head;
  337. struct vfsmount *p, *found = NULL;
  338. for (;;) {
  339. tmp = dir ? tmp->next : tmp->prev;
  340. p = NULL;
  341. if (tmp == head)
  342. break;
  343. p = list_entry(tmp, struct vfsmount, mnt_hash);
  344. if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
  345. found = p;
  346. break;
  347. }
  348. }
  349. return found;
  350. }
  351. /*
  352. * lookup_mnt increments the ref count before returning
  353. * the vfsmount struct.
  354. */
  355. struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
  356. {
  357. struct vfsmount *child_mnt;
  358. spin_lock(&vfsmount_lock);
  359. if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
  360. mntget(child_mnt);
  361. spin_unlock(&vfsmount_lock);
  362. return child_mnt;
  363. }
  364. static inline int check_mnt(struct vfsmount *mnt)
  365. {
  366. return mnt->mnt_ns == current->nsproxy->mnt_ns;
  367. }
  368. static void touch_mnt_namespace(struct mnt_namespace *ns)
  369. {
  370. if (ns) {
  371. ns->event = ++event;
  372. wake_up_interruptible(&ns->poll);
  373. }
  374. }
  375. static void __touch_mnt_namespace(struct mnt_namespace *ns)
  376. {
  377. if (ns && ns->event != event) {
  378. ns->event = event;
  379. wake_up_interruptible(&ns->poll);
  380. }
  381. }
  382. static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
  383. {
  384. old_path->dentry = mnt->mnt_mountpoint;
  385. old_path->mnt = mnt->mnt_parent;
  386. mnt->mnt_parent = mnt;
  387. mnt->mnt_mountpoint = mnt->mnt_root;
  388. list_del_init(&mnt->mnt_child);
  389. list_del_init(&mnt->mnt_hash);
  390. old_path->dentry->d_mounted--;
  391. }
  392. void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
  393. struct vfsmount *child_mnt)
  394. {
  395. child_mnt->mnt_parent = mntget(mnt);
  396. child_mnt->mnt_mountpoint = dget(dentry);
  397. dentry->d_mounted++;
  398. }
  399. static void attach_mnt(struct vfsmount *mnt, struct path *path)
  400. {
  401. mnt_set_mountpoint(path->mnt, path->dentry, mnt);
  402. list_add_tail(&mnt->mnt_hash, mount_hashtable +
  403. hash(path->mnt, path->dentry));
  404. list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
  405. }
  406. /*
  407. * the caller must hold vfsmount_lock
  408. */
  409. static void commit_tree(struct vfsmount *mnt)
  410. {
  411. struct vfsmount *parent = mnt->mnt_parent;
  412. struct vfsmount *m;
  413. LIST_HEAD(head);
  414. struct mnt_namespace *n = parent->mnt_ns;
  415. BUG_ON(parent == mnt);
  416. list_add_tail(&head, &mnt->mnt_list);
  417. list_for_each_entry(m, &head, mnt_list)
  418. m->mnt_ns = n;
  419. list_splice(&head, n->list.prev);
  420. list_add_tail(&mnt->mnt_hash, mount_hashtable +
  421. hash(parent, mnt->mnt_mountpoint));
  422. list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
  423. touch_mnt_namespace(n);
  424. }
  425. static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
  426. {
  427. struct list_head *next = p->mnt_mounts.next;
  428. if (next == &p->mnt_mounts) {
  429. while (1) {
  430. if (p == root)
  431. return NULL;
  432. next = p->mnt_child.next;
  433. if (next != &p->mnt_parent->mnt_mounts)
  434. break;
  435. p = p->mnt_parent;
  436. }
  437. }
  438. return list_entry(next, struct vfsmount, mnt_child);
  439. }
  440. static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
  441. {
  442. struct list_head *prev = p->mnt_mounts.prev;
  443. while (prev != &p->mnt_mounts) {
  444. p = list_entry(prev, struct vfsmount, mnt_child);
  445. prev = p->mnt_mounts.prev;
  446. }
  447. return p;
  448. }
  449. static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
  450. int flag)
  451. {
  452. struct super_block *sb = old->mnt_sb;
  453. struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
  454. if (mnt) {
  455. mnt->mnt_flags = old->mnt_flags;
  456. atomic_inc(&sb->s_active);
  457. mnt->mnt_sb = sb;
  458. mnt->mnt_root = dget(root);
  459. mnt->mnt_mountpoint = mnt->mnt_root;
  460. mnt->mnt_parent = mnt;
  461. if (flag & CL_SLAVE) {
  462. list_add(&mnt->mnt_slave, &old->mnt_slave_list);
  463. mnt->mnt_master = old;
  464. CLEAR_MNT_SHARED(mnt);
  465. } else if (!(flag & CL_PRIVATE)) {
  466. if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
  467. list_add(&mnt->mnt_share, &old->mnt_share);
  468. if (IS_MNT_SLAVE(old))
  469. list_add(&mnt->mnt_slave, &old->mnt_slave);
  470. mnt->mnt_master = old->mnt_master;
  471. }
  472. if (flag & CL_MAKE_SHARED)
  473. set_mnt_shared(mnt);
  474. /* stick the duplicate mount on the same expiry list
  475. * as the original if that was on one */
  476. if (flag & CL_EXPIRE) {
  477. if (!list_empty(&old->mnt_expire))
  478. list_add(&mnt->mnt_expire, &old->mnt_expire);
  479. }
  480. }
  481. return mnt;
  482. }
  483. static inline void __mntput(struct vfsmount *mnt)
  484. {
  485. int cpu;
  486. struct super_block *sb = mnt->mnt_sb;
  487. /*
  488. * We don't have to hold all of the locks at the
  489. * same time here because we know that we're the
  490. * last reference to mnt and that no new writers
  491. * can come in.
  492. */
  493. for_each_possible_cpu(cpu) {
  494. struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu);
  495. if (cpu_writer->mnt != mnt)
  496. continue;
  497. spin_lock(&cpu_writer->lock);
  498. atomic_add(cpu_writer->count, &mnt->__mnt_writers);
  499. cpu_writer->count = 0;
  500. /*
  501. * Might as well do this so that no one
  502. * ever sees the pointer and expects
  503. * it to be valid.
  504. */
  505. cpu_writer->mnt = NULL;
  506. spin_unlock(&cpu_writer->lock);
  507. }
  508. /*
  509. * This probably indicates that somebody messed
  510. * up a mnt_want/drop_write() pair. If this
  511. * happens, the filesystem was probably unable
  512. * to make r/w->r/o transitions.
  513. */
  514. WARN_ON(atomic_read(&mnt->__mnt_writers));
  515. dput(mnt->mnt_root);
  516. free_vfsmnt(mnt);
  517. deactivate_super(sb);
  518. }
  519. void mntput_no_expire(struct vfsmount *mnt)
  520. {
  521. repeat:
  522. if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
  523. if (likely(!mnt->mnt_pinned)) {
  524. spin_unlock(&vfsmount_lock);
  525. __mntput(mnt);
  526. return;
  527. }
  528. atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
  529. mnt->mnt_pinned = 0;
  530. spin_unlock(&vfsmount_lock);
  531. acct_auto_close_mnt(mnt);
  532. security_sb_umount_close(mnt);
  533. goto repeat;
  534. }
  535. }
  536. EXPORT_SYMBOL(mntput_no_expire);
  537. void mnt_pin(struct vfsmount *mnt)
  538. {
  539. spin_lock(&vfsmount_lock);
  540. mnt->mnt_pinned++;
  541. spin_unlock(&vfsmount_lock);
  542. }
  543. EXPORT_SYMBOL(mnt_pin);
  544. void mnt_unpin(struct vfsmount *mnt)
  545. {
  546. spin_lock(&vfsmount_lock);
  547. if (mnt->mnt_pinned) {
  548. atomic_inc(&mnt->mnt_count);
  549. mnt->mnt_pinned--;
  550. }
  551. spin_unlock(&vfsmount_lock);
  552. }
  553. EXPORT_SYMBOL(mnt_unpin);
  554. static inline void mangle(struct seq_file *m, const char *s)
  555. {
  556. seq_escape(m, s, " \t\n\\");
  557. }
  558. /*
  559. * Simple .show_options callback for filesystems which don't want to
  560. * implement more complex mount option showing.
  561. *
  562. * See also save_mount_options().
  563. */
  564. int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
  565. {
  566. const char *options = mnt->mnt_sb->s_options;
  567. if (options != NULL && options[0]) {
  568. seq_putc(m, ',');
  569. mangle(m, options);
  570. }
  571. return 0;
  572. }
  573. EXPORT_SYMBOL(generic_show_options);
  574. /*
  575. * If filesystem uses generic_show_options(), this function should be
  576. * called from the fill_super() callback.
  577. *
  578. * The .remount_fs callback usually needs to be handled in a special
  579. * way, to make sure, that previous options are not overwritten if the
  580. * remount fails.
  581. *
  582. * Also note, that if the filesystem's .remount_fs function doesn't
  583. * reset all options to their default value, but changes only newly
  584. * given options, then the displayed options will not reflect reality
  585. * any more.
  586. */
  587. void save_mount_options(struct super_block *sb, char *options)
  588. {
  589. kfree(sb->s_options);
  590. sb->s_options = kstrdup(options, GFP_KERNEL);
  591. }
  592. EXPORT_SYMBOL(save_mount_options);
  593. /* iterator */
  594. static void *m_start(struct seq_file *m, loff_t *pos)
  595. {
  596. struct mnt_namespace *n = m->private;
  597. down_read(&namespace_sem);
  598. return seq_list_start(&n->list, *pos);
  599. }
  600. static void *m_next(struct seq_file *m, void *v, loff_t *pos)
  601. {
  602. struct mnt_namespace *n = m->private;
  603. return seq_list_next(v, &n->list, pos);
  604. }
  605. static void m_stop(struct seq_file *m, void *v)
  606. {
  607. up_read(&namespace_sem);
  608. }
  609. static int show_vfsmnt(struct seq_file *m, void *v)
  610. {
  611. struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
  612. int err = 0;
  613. static struct proc_fs_info {
  614. int flag;
  615. char *str;
  616. } fs_info[] = {
  617. { MS_SYNCHRONOUS, ",sync" },
  618. { MS_DIRSYNC, ",dirsync" },
  619. { MS_MANDLOCK, ",mand" },
  620. { 0, NULL }
  621. };
  622. static struct proc_fs_info mnt_info[] = {
  623. { MNT_NOSUID, ",nosuid" },
  624. { MNT_NODEV, ",nodev" },
  625. { MNT_NOEXEC, ",noexec" },
  626. { MNT_NOATIME, ",noatime" },
  627. { MNT_NODIRATIME, ",nodiratime" },
  628. { MNT_RELATIME, ",relatime" },
  629. { 0, NULL }
  630. };
  631. struct proc_fs_info *fs_infop;
  632. struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
  633. mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
  634. seq_putc(m, ' ');
  635. seq_path(m, &mnt_path, " \t\n\\");
  636. seq_putc(m, ' ');
  637. mangle(m, mnt->mnt_sb->s_type->name);
  638. if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) {
  639. seq_putc(m, '.');
  640. mangle(m, mnt->mnt_sb->s_subtype);
  641. }
  642. seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
  643. for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
  644. if (mnt->mnt_sb->s_flags & fs_infop->flag)
  645. seq_puts(m, fs_infop->str);
  646. }
  647. for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
  648. if (mnt->mnt_flags & fs_infop->flag)
  649. seq_puts(m, fs_infop->str);
  650. }
  651. if (mnt->mnt_sb->s_op->show_options)
  652. err = mnt->mnt_sb->s_op->show_options(m, mnt);
  653. seq_puts(m, " 0 0\n");
  654. return err;
  655. }
  656. struct seq_operations mounts_op = {
  657. .start = m_start,
  658. .next = m_next,
  659. .stop = m_stop,
  660. .show = show_vfsmnt
  661. };
  662. static int show_vfsstat(struct seq_file *m, void *v)
  663. {
  664. struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
  665. struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
  666. int err = 0;
  667. /* device */
  668. if (mnt->mnt_devname) {
  669. seq_puts(m, "device ");
  670. mangle(m, mnt->mnt_devname);
  671. } else
  672. seq_puts(m, "no device");
  673. /* mount point */
  674. seq_puts(m, " mounted on ");
  675. seq_path(m, &mnt_path, " \t\n\\");
  676. seq_putc(m, ' ');
  677. /* file system type */
  678. seq_puts(m, "with fstype ");
  679. mangle(m, mnt->mnt_sb->s_type->name);
  680. /* optional statistics */
  681. if (mnt->mnt_sb->s_op->show_stats) {
  682. seq_putc(m, ' ');
  683. err = mnt->mnt_sb->s_op->show_stats(m, mnt);
  684. }
  685. seq_putc(m, '\n');
  686. return err;
  687. }
  688. struct seq_operations mountstats_op = {
  689. .start = m_start,
  690. .next = m_next,
  691. .stop = m_stop,
  692. .show = show_vfsstat,
  693. };
  694. /**
  695. * may_umount_tree - check if a mount tree is busy
  696. * @mnt: root of mount tree
  697. *
  698. * This is called to check if a tree of mounts has any
  699. * open files, pwds, chroots or sub mounts that are
  700. * busy.
  701. */
  702. int may_umount_tree(struct vfsmount *mnt)
  703. {
  704. int actual_refs = 0;
  705. int minimum_refs = 0;
  706. struct vfsmount *p;
  707. spin_lock(&vfsmount_lock);
  708. for (p = mnt; p; p = next_mnt(p, mnt)) {
  709. actual_refs += atomic_read(&p->mnt_count);
  710. minimum_refs += 2;
  711. }
  712. spin_unlock(&vfsmount_lock);
  713. if (actual_refs > minimum_refs)
  714. return 0;
  715. return 1;
  716. }
  717. EXPORT_SYMBOL(may_umount_tree);
  718. /**
  719. * may_umount - check if a mount point is busy
  720. * @mnt: root of mount
  721. *
  722. * This is called to check if a mount point has any
  723. * open files, pwds, chroots or sub mounts. If the
  724. * mount has sub mounts this will return busy
  725. * regardless of whether the sub mounts are busy.
  726. *
  727. * Doesn't take quota and stuff into account. IOW, in some cases it will
  728. * give false negatives. The main reason why it's here is that we need
  729. * a non-destructive way to look for easily umountable filesystems.
  730. */
  731. int may_umount(struct vfsmount *mnt)
  732. {
  733. int ret = 1;
  734. spin_lock(&vfsmount_lock);
  735. if (propagate_mount_busy(mnt, 2))
  736. ret = 0;
  737. spin_unlock(&vfsmount_lock);
  738. return ret;
  739. }
  740. EXPORT_SYMBOL(may_umount);
  741. void release_mounts(struct list_head *head)
  742. {
  743. struct vfsmount *mnt;
  744. while (!list_empty(head)) {
  745. mnt = list_first_entry(head, struct vfsmount, mnt_hash);
  746. list_del_init(&mnt->mnt_hash);
  747. if (mnt->mnt_parent != mnt) {
  748. struct dentry *dentry;
  749. struct vfsmount *m;
  750. spin_lock(&vfsmount_lock);
  751. dentry = mnt->mnt_mountpoint;
  752. m = mnt->mnt_parent;
  753. mnt->mnt_mountpoint = mnt->mnt_root;
  754. mnt->mnt_parent = mnt;
  755. m->mnt_ghosts--;
  756. spin_unlock(&vfsmount_lock);
  757. dput(dentry);
  758. mntput(m);
  759. }
  760. mntput(mnt);
  761. }
  762. }
  763. void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
  764. {
  765. struct vfsmount *p;
  766. for (p = mnt; p; p = next_mnt(p, mnt))
  767. list_move(&p->mnt_hash, kill);
  768. if (propagate)
  769. propagate_umount(kill);
  770. list_for_each_entry(p, kill, mnt_hash) {
  771. list_del_init(&p->mnt_expire);
  772. list_del_init(&p->mnt_list);
  773. __touch_mnt_namespace(p->mnt_ns);
  774. p->mnt_ns = NULL;
  775. list_del_init(&p->mnt_child);
  776. if (p->mnt_parent != p) {
  777. p->mnt_parent->mnt_ghosts++;
  778. p->mnt_mountpoint->d_mounted--;
  779. }
  780. change_mnt_propagation(p, MS_PRIVATE);
  781. }
  782. }
  783. static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
  784. static int do_umount(struct vfsmount *mnt, int flags)
  785. {
  786. struct super_block *sb = mnt->mnt_sb;
  787. int retval;
  788. LIST_HEAD(umount_list);
  789. retval = security_sb_umount(mnt, flags);
  790. if (retval)
  791. return retval;
  792. /*
  793. * Allow userspace to request a mountpoint be expired rather than
  794. * unmounting unconditionally. Unmount only happens if:
  795. * (1) the mark is already set (the mark is cleared by mntput())
  796. * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
  797. */
  798. if (flags & MNT_EXPIRE) {
  799. if (mnt == current->fs->root.mnt ||
  800. flags & (MNT_FORCE | MNT_DETACH))
  801. return -EINVAL;
  802. if (atomic_read(&mnt->mnt_count) != 2)
  803. return -EBUSY;
  804. if (!xchg(&mnt->mnt_expiry_mark, 1))
  805. return -EAGAIN;
  806. }
  807. /*
  808. * If we may have to abort operations to get out of this
  809. * mount, and they will themselves hold resources we must
  810. * allow the fs to do things. In the Unix tradition of
  811. * 'Gee thats tricky lets do it in userspace' the umount_begin
  812. * might fail to complete on the first run through as other tasks
  813. * must return, and the like. Thats for the mount program to worry
  814. * about for the moment.
  815. */
  816. lock_kernel();
  817. if (sb->s_op->umount_begin)
  818. sb->s_op->umount_begin(mnt, flags);
  819. unlock_kernel();
  820. /*
  821. * No sense to grab the lock for this test, but test itself looks
  822. * somewhat bogus. Suggestions for better replacement?
  823. * Ho-hum... In principle, we might treat that as umount + switch
  824. * to rootfs. GC would eventually take care of the old vfsmount.
  825. * Actually it makes sense, especially if rootfs would contain a
  826. * /reboot - static binary that would close all descriptors and
  827. * call reboot(9). Then init(8) could umount root and exec /reboot.
  828. */
  829. if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
  830. /*
  831. * Special case for "unmounting" root ...
  832. * we just try to remount it readonly.
  833. */
  834. down_write(&sb->s_umount);
  835. if (!(sb->s_flags & MS_RDONLY)) {
  836. lock_kernel();
  837. DQUOT_OFF(sb);
  838. retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
  839. unlock_kernel();
  840. }
  841. up_write(&sb->s_umount);
  842. return retval;
  843. }
  844. down_write(&namespace_sem);
  845. spin_lock(&vfsmount_lock);
  846. event++;
  847. if (!(flags & MNT_DETACH))
  848. shrink_submounts(mnt, &umount_list);
  849. retval = -EBUSY;
  850. if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
  851. if (!list_empty(&mnt->mnt_list))
  852. umount_tree(mnt, 1, &umount_list);
  853. retval = 0;
  854. }
  855. spin_unlock(&vfsmount_lock);
  856. if (retval)
  857. security_sb_umount_busy(mnt);
  858. up_write(&namespace_sem);
  859. release_mounts(&umount_list);
  860. return retval;
  861. }
  862. /*
  863. * Now umount can handle mount points as well as block devices.
  864. * This is important for filesystems which use unnamed block devices.
  865. *
  866. * We now support a flag for forced unmount like the other 'big iron'
  867. * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
  868. */
  869. asmlinkage long sys_umount(char __user * name, int flags)
  870. {
  871. struct nameidata nd;
  872. int retval;
  873. retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
  874. if (retval)
  875. goto out;
  876. retval = -EINVAL;
  877. if (nd.path.dentry != nd.path.mnt->mnt_root)
  878. goto dput_and_out;
  879. if (!check_mnt(nd.path.mnt))
  880. goto dput_and_out;
  881. retval = -EPERM;
  882. if (!capable(CAP_SYS_ADMIN))
  883. goto dput_and_out;
  884. retval = do_umount(nd.path.mnt, flags);
  885. dput_and_out:
  886. /* we mustn't call path_put() as that would clear mnt_expiry_mark */
  887. dput(nd.path.dentry);
  888. mntput_no_expire(nd.path.mnt);
  889. out:
  890. return retval;
  891. }
  892. #ifdef __ARCH_WANT_SYS_OLDUMOUNT
  893. /*
  894. * The 2.0 compatible umount. No flags.
  895. */
  896. asmlinkage long sys_oldumount(char __user * name)
  897. {
  898. return sys_umount(name, 0);
  899. }
  900. #endif
  901. static int mount_is_safe(struct nameidata *nd)
  902. {
  903. if (capable(CAP_SYS_ADMIN))
  904. return 0;
  905. return -EPERM;
  906. #ifdef notyet
  907. if (S_ISLNK(nd->path.dentry->d_inode->i_mode))
  908. return -EPERM;
  909. if (nd->path.dentry->d_inode->i_mode & S_ISVTX) {
  910. if (current->uid != nd->path.dentry->d_inode->i_uid)
  911. return -EPERM;
  912. }
  913. if (vfs_permission(nd, MAY_WRITE))
  914. return -EPERM;
  915. return 0;
  916. #endif
  917. }
  918. static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
  919. {
  920. while (1) {
  921. if (d == dentry)
  922. return 1;
  923. if (d == NULL || d == d->d_parent)
  924. return 0;
  925. d = d->d_parent;
  926. }
  927. }
  928. struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
  929. int flag)
  930. {
  931. struct vfsmount *res, *p, *q, *r, *s;
  932. struct path path;
  933. if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
  934. return NULL;
  935. res = q = clone_mnt(mnt, dentry, flag);
  936. if (!q)
  937. goto Enomem;
  938. q->mnt_mountpoint = mnt->mnt_mountpoint;
  939. p = mnt;
  940. list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
  941. if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
  942. continue;
  943. for (s = r; s; s = next_mnt(s, r)) {
  944. if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
  945. s = skip_mnt_tree(s);
  946. continue;
  947. }
  948. while (p != s->mnt_parent) {
  949. p = p->mnt_parent;
  950. q = q->mnt_parent;
  951. }
  952. p = s;
  953. path.mnt = q;
  954. path.dentry = p->mnt_mountpoint;
  955. q = clone_mnt(p, p->mnt_root, flag);
  956. if (!q)
  957. goto Enomem;
  958. spin_lock(&vfsmount_lock);
  959. list_add_tail(&q->mnt_list, &res->mnt_list);
  960. attach_mnt(q, &path);
  961. spin_unlock(&vfsmount_lock);
  962. }
  963. }
  964. return res;
  965. Enomem:
  966. if (res) {
  967. LIST_HEAD(umount_list);
  968. spin_lock(&vfsmount_lock);
  969. umount_tree(res, 0, &umount_list);
  970. spin_unlock(&vfsmount_lock);
  971. release_mounts(&umount_list);
  972. }
  973. return NULL;
  974. }
  975. struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
  976. {
  977. struct vfsmount *tree;
  978. down_write(&namespace_sem);
  979. tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
  980. up_write(&namespace_sem);
  981. return tree;
  982. }
  983. void drop_collected_mounts(struct vfsmount *mnt)
  984. {
  985. LIST_HEAD(umount_list);
  986. down_write(&namespace_sem);
  987. spin_lock(&vfsmount_lock);
  988. umount_tree(mnt, 0, &umount_list);
  989. spin_unlock(&vfsmount_lock);
  990. up_write(&namespace_sem);
  991. release_mounts(&umount_list);
  992. }
  993. /*
  994. * @source_mnt : mount tree to be attached
  995. * @nd : place the mount tree @source_mnt is attached
  996. * @parent_nd : if non-null, detach the source_mnt from its parent and
  997. * store the parent mount and mountpoint dentry.
  998. * (done when source_mnt is moved)
  999. *
  1000. * NOTE: in the table below explains the semantics when a source mount
  1001. * of a given type is attached to a destination mount of a given type.
  1002. * ---------------------------------------------------------------------------
  1003. * | BIND MOUNT OPERATION |
  1004. * |**************************************************************************
  1005. * | source-->| shared | private | slave | unbindable |
  1006. * | dest | | | | |
  1007. * | | | | | | |
  1008. * | v | | | | |
  1009. * |**************************************************************************
  1010. * | shared | shared (++) | shared (+) | shared(+++)| invalid |
  1011. * | | | | | |
  1012. * |non-shared| shared (+) | private | slave (*) | invalid |
  1013. * ***************************************************************************
  1014. * A bind operation clones the source mount and mounts the clone on the
  1015. * destination mount.
  1016. *
  1017. * (++) the cloned mount is propagated to all the mounts in the propagation
  1018. * tree of the destination mount and the cloned mount is added to
  1019. * the peer group of the source mount.
  1020. * (+) the cloned mount is created under the destination mount and is marked
  1021. * as shared. The cloned mount is added to the peer group of the source
  1022. * mount.
  1023. * (+++) the mount is propagated to all the mounts in the propagation tree
  1024. * of the destination mount and the cloned mount is made slave
  1025. * of the same master as that of the source mount. The cloned mount
  1026. * is marked as 'shared and slave'.
  1027. * (*) the cloned mount is made a slave of the same master as that of the
  1028. * source mount.
  1029. *
  1030. * ---------------------------------------------------------------------------
  1031. * | MOVE MOUNT OPERATION |
  1032. * |**************************************************************************
  1033. * | source-->| shared | private | slave | unbindable |
  1034. * | dest | | | | |
  1035. * | | | | | | |
  1036. * | v | | | | |
  1037. * |**************************************************************************
  1038. * | shared | shared (+) | shared (+) | shared(+++) | invalid |
  1039. * | | | | | |
  1040. * |non-shared| shared (+*) | private | slave (*) | unbindable |
  1041. * ***************************************************************************
  1042. *
  1043. * (+) the mount is moved to the destination. And is then propagated to
  1044. * all the mounts in the propagation tree of the destination mount.
  1045. * (+*) the mount is moved to the destination.
  1046. * (+++) the mount is moved to the destination and is then propagated to
  1047. * all the mounts belonging to the destination mount's propagation tree.
  1048. * the mount is marked as 'shared and slave'.
  1049. * (*) the mount continues to be a slave at the new location.
  1050. *
  1051. * if the source mount is a tree, the operations explained above is
  1052. * applied to each mount in the tree.
  1053. * Must be called without spinlocks held, since this function can sleep
  1054. * in allocations.
  1055. */
  1056. static int attach_recursive_mnt(struct vfsmount *source_mnt,
  1057. struct path *path, struct path *parent_path)
  1058. {
  1059. LIST_HEAD(tree_list);
  1060. struct vfsmount *dest_mnt = path->mnt;
  1061. struct dentry *dest_dentry = path->dentry;
  1062. struct vfsmount *child, *p;
  1063. if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
  1064. return -EINVAL;
  1065. if (IS_MNT_SHARED(dest_mnt)) {
  1066. for (p = source_mnt; p; p = next_mnt(p, source_mnt))
  1067. set_mnt_shared(p);
  1068. }
  1069. spin_lock(&vfsmount_lock);
  1070. if (parent_path) {
  1071. detach_mnt(source_mnt, parent_path);
  1072. attach_mnt(source_mnt, path);
  1073. touch_mnt_namespace(current->nsproxy->mnt_ns);
  1074. } else {
  1075. mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
  1076. commit_tree(source_mnt);
  1077. }
  1078. list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
  1079. list_del_init(&child->mnt_hash);
  1080. commit_tree(child);
  1081. }
  1082. spin_unlock(&vfsmount_lock);
  1083. return 0;
  1084. }
  1085. static int graft_tree(struct vfsmount *mnt, struct path *path)
  1086. {
  1087. int err;
  1088. if (mnt->mnt_sb->s_flags & MS_NOUSER)
  1089. return -EINVAL;
  1090. if (S_ISDIR(path->dentry->d_inode->i_mode) !=
  1091. S_ISDIR(mnt->mnt_root->d_inode->i_mode))
  1092. return -ENOTDIR;
  1093. err = -ENOENT;
  1094. mutex_lock(&path->dentry->d_inode->i_mutex);
  1095. if (IS_DEADDIR(path->dentry->d_inode))
  1096. goto out_unlock;
  1097. err = security_sb_check_sb(mnt, path);
  1098. if (err)
  1099. goto out_unlock;
  1100. err = -ENOENT;
  1101. if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry))
  1102. err = attach_recursive_mnt(mnt, path, NULL);
  1103. out_unlock:
  1104. mutex_unlock(&path->dentry->d_inode->i_mutex);
  1105. if (!err)
  1106. security_sb_post_addmount(mnt, path);
  1107. return err;
  1108. }
  1109. /*
  1110. * recursively change the type of the mountpoint.
  1111. * noinline this do_mount helper to save do_mount stack space.
  1112. */
  1113. static noinline int do_change_type(struct nameidata *nd, int flag)
  1114. {
  1115. struct vfsmount *m, *mnt = nd->path.mnt;
  1116. int recurse = flag & MS_REC;
  1117. int type = flag & ~MS_REC;
  1118. if (!capable(CAP_SYS_ADMIN))
  1119. return -EPERM;
  1120. if (nd->path.dentry != nd->path.mnt->mnt_root)
  1121. return -EINVAL;
  1122. down_write(&namespace_sem);
  1123. spin_lock(&vfsmount_lock);
  1124. for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
  1125. change_mnt_propagation(m, type);
  1126. spin_unlock(&vfsmount_lock);
  1127. up_write(&namespace_sem);
  1128. return 0;
  1129. }
  1130. /*
  1131. * do loopback mount.
  1132. * noinline this do_mount helper to save do_mount stack space.
  1133. */
  1134. static noinline int do_loopback(struct nameidata *nd, char *old_name,
  1135. int recurse)
  1136. {
  1137. struct nameidata old_nd;
  1138. struct vfsmount *mnt = NULL;
  1139. int err = mount_is_safe(nd);
  1140. if (err)
  1141. return err;
  1142. if (!old_name || !*old_name)
  1143. return -EINVAL;
  1144. err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
  1145. if (err)
  1146. return err;
  1147. down_write(&namespace_sem);
  1148. err = -EINVAL;
  1149. if (IS_MNT_UNBINDABLE(old_nd.path.mnt))
  1150. goto out;
  1151. if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
  1152. goto out;
  1153. err = -ENOMEM;
  1154. if (recurse)
  1155. mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0);
  1156. else
  1157. mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0);
  1158. if (!mnt)
  1159. goto out;
  1160. err = graft_tree(mnt, &nd->path);
  1161. if (err) {
  1162. LIST_HEAD(umount_list);
  1163. spin_lock(&vfsmount_lock);
  1164. umount_tree(mnt, 0, &umount_list);
  1165. spin_unlock(&vfsmount_lock);
  1166. release_mounts(&umount_list);
  1167. }
  1168. out:
  1169. up_write(&namespace_sem);
  1170. path_put(&old_nd.path);
  1171. return err;
  1172. }
  1173. static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
  1174. {
  1175. int error = 0;
  1176. int readonly_request = 0;
  1177. if (ms_flags & MS_RDONLY)
  1178. readonly_request = 1;
  1179. if (readonly_request == __mnt_is_readonly(mnt))
  1180. return 0;
  1181. if (readonly_request)
  1182. error = mnt_make_readonly(mnt);
  1183. else
  1184. __mnt_unmake_readonly(mnt);
  1185. return error;
  1186. }
  1187. /*
  1188. * change filesystem flags. dir should be a physical root of filesystem.
  1189. * If you've mounted a non-root directory somewhere and want to do remount
  1190. * on it - tough luck.
  1191. * noinline this do_mount helper to save do_mount stack space.
  1192. */
  1193. static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags,
  1194. void *data)
  1195. {
  1196. int err;
  1197. struct super_block *sb = nd->path.mnt->mnt_sb;
  1198. if (!capable(CAP_SYS_ADMIN))
  1199. return -EPERM;
  1200. if (!check_mnt(nd->path.mnt))
  1201. return -EINVAL;
  1202. if (nd->path.dentry != nd->path.mnt->mnt_root)
  1203. return -EINVAL;
  1204. down_write(&sb->s_umount);
  1205. if (flags & MS_BIND)
  1206. err = change_mount_flags(nd->path.mnt, flags);
  1207. else
  1208. err = do_remount_sb(sb, flags, data, 0);
  1209. if (!err)
  1210. nd->path.mnt->mnt_flags = mnt_flags;
  1211. up_write(&sb->s_umount);
  1212. if (!err)
  1213. security_sb_post_remount(nd->path.mnt, flags, data);
  1214. return err;
  1215. }
  1216. static inline int tree_contains_unbindable(struct vfsmount *mnt)
  1217. {
  1218. struct vfsmount *p;
  1219. for (p = mnt; p; p = next_mnt(p, mnt)) {
  1220. if (IS_MNT_UNBINDABLE(p))
  1221. return 1;
  1222. }
  1223. return 0;
  1224. }
  1225. /*
  1226. * noinline this do_mount helper to save do_mount stack space.
  1227. */
  1228. static noinline int do_move_mount(struct nameidata *nd, char *old_name)
  1229. {
  1230. struct nameidata old_nd;
  1231. struct path parent_path;
  1232. struct vfsmount *p;
  1233. int err = 0;
  1234. if (!capable(CAP_SYS_ADMIN))
  1235. return -EPERM;
  1236. if (!old_name || !*old_name)
  1237. return -EINVAL;
  1238. err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
  1239. if (err)
  1240. return err;
  1241. down_write(&namespace_sem);
  1242. while (d_mountpoint(nd->path.dentry) &&
  1243. follow_down(&nd->path.mnt, &nd->path.dentry))
  1244. ;
  1245. err = -EINVAL;
  1246. if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
  1247. goto out;
  1248. err = -ENOENT;
  1249. mutex_lock(&nd->path.dentry->d_inode->i_mutex);
  1250. if (IS_DEADDIR(nd->path.dentry->d_inode))
  1251. goto out1;
  1252. if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry))
  1253. goto out1;
  1254. err = -EINVAL;
  1255. if (old_nd.path.dentry != old_nd.path.mnt->mnt_root)
  1256. goto out1;
  1257. if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent)
  1258. goto out1;
  1259. if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
  1260. S_ISDIR(old_nd.path.dentry->d_inode->i_mode))
  1261. goto out1;
  1262. /*
  1263. * Don't move a mount residing in a shared parent.
  1264. */
  1265. if (old_nd.path.mnt->mnt_parent &&
  1266. IS_MNT_SHARED(old_nd.path.mnt->mnt_parent))
  1267. goto out1;
  1268. /*
  1269. * Don't move a mount tree containing unbindable mounts to a destination
  1270. * mount which is shared.
  1271. */
  1272. if (IS_MNT_SHARED(nd->path.mnt) &&
  1273. tree_contains_unbindable(old_nd.path.mnt))
  1274. goto out1;
  1275. err = -ELOOP;
  1276. for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent)
  1277. if (p == old_nd.path.mnt)
  1278. goto out1;
  1279. err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path);
  1280. if (err)
  1281. goto out1;
  1282. /* if the mount is moved, it should no longer be expire
  1283. * automatically */
  1284. list_del_init(&old_nd.path.mnt->mnt_expire);
  1285. out1:
  1286. mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
  1287. out:
  1288. up_write(&namespace_sem);
  1289. if (!err)
  1290. path_put(&parent_path);
  1291. path_put(&old_nd.path);
  1292. return err;
  1293. }
  1294. /*
  1295. * create a new mount for userspace and request it to be added into the
  1296. * namespace's tree
  1297. * noinline this do_mount helper to save do_mount stack space.
  1298. */
  1299. static noinline int do_new_mount(struct nameidata *nd, char *type, int flags,
  1300. int mnt_flags, char *name, void *data)
  1301. {
  1302. struct vfsmount *mnt;
  1303. if (!type || !memchr(type, 0, PAGE_SIZE))
  1304. return -EINVAL;
  1305. /* we need capabilities... */
  1306. if (!capable(CAP_SYS_ADMIN))
  1307. return -EPERM;
  1308. mnt = do_kern_mount(type, flags, name, data);
  1309. if (IS_ERR(mnt))
  1310. return PTR_ERR(mnt);
  1311. return do_add_mount(mnt, nd, mnt_flags, NULL);
  1312. }
  1313. /*
  1314. * add a mount into a namespace's mount tree
  1315. * - provide the option of adding the new mount to an expiration list
  1316. */
  1317. int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
  1318. int mnt_flags, struct list_head *fslist)
  1319. {
  1320. int err;
  1321. down_write(&namespace_sem);
  1322. /* Something was mounted here while we slept */
  1323. while (d_mountpoint(nd->path.dentry) &&
  1324. follow_down(&nd->path.mnt, &nd->path.dentry))
  1325. ;
  1326. err = -EINVAL;
  1327. if (!check_mnt(nd->path.mnt))
  1328. goto unlock;
  1329. /* Refuse the same filesystem on the same mount point */
  1330. err = -EBUSY;
  1331. if (nd->path.mnt->mnt_sb == newmnt->mnt_sb &&
  1332. nd->path.mnt->mnt_root == nd->path.dentry)
  1333. goto unlock;
  1334. err = -EINVAL;
  1335. if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
  1336. goto unlock;
  1337. newmnt->mnt_flags = mnt_flags;
  1338. if ((err = graft_tree(newmnt, &nd->path)))
  1339. goto unlock;
  1340. if (fslist) /* add to the specified expiration list */
  1341. list_add_tail(&newmnt->mnt_expire, fslist);
  1342. up_write(&namespace_sem);
  1343. return 0;
  1344. unlock:
  1345. up_write(&namespace_sem);
  1346. mntput(newmnt);
  1347. return err;
  1348. }
  1349. EXPORT_SYMBOL_GPL(do_add_mount);
  1350. /*
  1351. * process a list of expirable mountpoints with the intent of discarding any
  1352. * mountpoints that aren't in use and haven't been touched since last we came
  1353. * here
  1354. */
  1355. void mark_mounts_for_expiry(struct list_head *mounts)
  1356. {
  1357. struct vfsmount *mnt, *next;
  1358. LIST_HEAD(graveyard);
  1359. LIST_HEAD(umounts);
  1360. if (list_empty(mounts))
  1361. return;
  1362. down_write(&namespace_sem);
  1363. spin_lock(&vfsmount_lock);
  1364. /* extract from the expiration list every vfsmount that matches the
  1365. * following criteria:
  1366. * - only referenced by its parent vfsmount
  1367. * - still marked for expiry (marked on the last call here; marks are
  1368. * cleared by mntput())
  1369. */
  1370. list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
  1371. if (!xchg(&mnt->mnt_expiry_mark, 1) ||
  1372. propagate_mount_busy(mnt, 1))
  1373. continue;
  1374. list_move(&mnt->mnt_expire, &graveyard);
  1375. }
  1376. while (!list_empty(&graveyard)) {
  1377. mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
  1378. touch_mnt_namespace(mnt->mnt_ns);
  1379. umount_tree(mnt, 1, &umounts);
  1380. }
  1381. spin_unlock(&vfsmount_lock);
  1382. up_write(&namespace_sem);
  1383. release_mounts(&umounts);
  1384. }
  1385. EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
  1386. /*
  1387. * Ripoff of 'select_parent()'
  1388. *
  1389. * search the list of submounts for a given mountpoint, and move any
  1390. * shrinkable submounts to the 'graveyard' list.
  1391. */
  1392. static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
  1393. {
  1394. struct vfsmount *this_parent = parent;
  1395. struct list_head *next;
  1396. int found = 0;
  1397. repeat:
  1398. next = this_parent->mnt_mounts.next;
  1399. resume:
  1400. while (next != &this_parent->mnt_mounts) {
  1401. struct list_head *tmp = next;
  1402. struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
  1403. next = tmp->next;
  1404. if (!(mnt->mnt_flags & MNT_SHRINKABLE))
  1405. continue;
  1406. /*
  1407. * Descend a level if the d_mounts list is non-empty.
  1408. */
  1409. if (!list_empty(&mnt->mnt_mounts)) {
  1410. this_parent = mnt;
  1411. goto repeat;
  1412. }
  1413. if (!propagate_mount_busy(mnt, 1)) {
  1414. list_move_tail(&mnt->mnt_expire, graveyard);
  1415. found++;
  1416. }
  1417. }
  1418. /*
  1419. * All done at this level ... ascend and resume the search
  1420. */
  1421. if (this_parent != parent) {
  1422. next = this_parent->mnt_child.next;
  1423. this_parent = this_parent->mnt_parent;
  1424. goto resume;
  1425. }
  1426. return found;
  1427. }
  1428. /*
  1429. * process a list of expirable mountpoints with the intent of discarding any
  1430. * submounts of a specific parent mountpoint
  1431. */
  1432. static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
  1433. {
  1434. LIST_HEAD(graveyard);
  1435. struct vfsmount *m;
  1436. /* extract submounts of 'mountpoint' from the expiration list */
  1437. while (select_submounts(mnt, &graveyard)) {
  1438. while (!list_empty(&graveyard)) {
  1439. m = list_first_entry(&graveyard, struct vfsmount,
  1440. mnt_expire);
  1441. touch_mnt_namespace(mnt->mnt_ns);
  1442. umount_tree(mnt, 1, umounts);
  1443. }
  1444. }
  1445. }
  1446. /*
  1447. * Some copy_from_user() implementations do not return the exact number of
  1448. * bytes remaining to copy on a fault. But copy_mount_options() requires that.
  1449. * Note that this function differs from copy_from_user() in that it will oops
  1450. * on bad values of `to', rather than returning a short copy.
  1451. */
  1452. static long exact_copy_from_user(void *to, const void __user * from,
  1453. unsigned long n)
  1454. {
  1455. char *t = to;
  1456. const char __user *f = from;
  1457. char c;
  1458. if (!access_ok(VERIFY_READ, from, n))
  1459. return n;
  1460. while (n) {
  1461. if (__get_user(c, f)) {
  1462. memset(t, 0, n);
  1463. break;
  1464. }
  1465. *t++ = c;
  1466. f++;
  1467. n--;
  1468. }
  1469. return n;
  1470. }
  1471. int copy_mount_options(const void __user * data, unsigned long *where)
  1472. {
  1473. int i;
  1474. unsigned long page;
  1475. unsigned long size;
  1476. *where = 0;
  1477. if (!data)
  1478. return 0;
  1479. if (!(page = __get_free_page(GFP_KERNEL)))
  1480. return -ENOMEM;
  1481. /* We only care that *some* data at the address the user
  1482. * gave us is valid. Just in case, we'll zero
  1483. * the remainder of the page.
  1484. */
  1485. /* copy_from_user cannot cross TASK_SIZE ! */
  1486. size = TASK_SIZE - (unsigned long)data;
  1487. if (size > PAGE_SIZE)
  1488. size = PAGE_SIZE;
  1489. i = size - exact_copy_from_user((void *)page, data, size);
  1490. if (!i) {
  1491. free_page(page);
  1492. return -EFAULT;
  1493. }
  1494. if (i != PAGE_SIZE)
  1495. memset((char *)page + i, 0, PAGE_SIZE - i);
  1496. *where = page;
  1497. return 0;
  1498. }
  1499. /*
  1500. * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
  1501. * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
  1502. *
  1503. * data is a (void *) that can point to any structure up to
  1504. * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
  1505. * information (or be NULL).
  1506. *
  1507. * Pre-0.97 versions of mount() didn't have a flags word.
  1508. * When the flags word was introduced its top half was required
  1509. * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
  1510. * Therefore, if this magic number is present, it carries no information
  1511. * and must be discarded.
  1512. */
  1513. long do_mount(char *dev_name, char *dir_name, char *type_page,
  1514. unsigned long flags, void *data_page)
  1515. {
  1516. struct nameidata nd;
  1517. int retval = 0;
  1518. int mnt_flags = 0;
  1519. /* Discard magic */
  1520. if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
  1521. flags &= ~MS_MGC_MSK;
  1522. /* Basic sanity checks */
  1523. if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
  1524. return -EINVAL;
  1525. if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
  1526. return -EINVAL;
  1527. if (data_page)
  1528. ((char *)data_page)[PAGE_SIZE - 1] = 0;
  1529. /* Separate the per-mountpoint flags */
  1530. if (flags & MS_NOSUID)
  1531. mnt_flags |= MNT_NOSUID;
  1532. if (flags & MS_NODEV)
  1533. mnt_flags |= MNT_NODEV;
  1534. if (flags & MS_NOEXEC)
  1535. mnt_flags |= MNT_NOEXEC;
  1536. if (flags & MS_NOATIME)
  1537. mnt_flags |= MNT_NOATIME;
  1538. if (flags & MS_NODIRATIME)
  1539. mnt_flags |= MNT_NODIRATIME;
  1540. if (flags & MS_RELATIME)
  1541. mnt_flags |= MNT_RELATIME;
  1542. if (flags & MS_RDONLY)
  1543. mnt_flags |= MNT_READONLY;
  1544. flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
  1545. MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
  1546. /* ... and get the mountpoint */
  1547. retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
  1548. if (retval)
  1549. return retval;
  1550. retval = security_sb_mount(dev_name, &nd.path,
  1551. type_page, flags, data_page);
  1552. if (retval)
  1553. goto dput_out;
  1554. if (flags & MS_REMOUNT)
  1555. retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
  1556. data_page);
  1557. else if (flags & MS_BIND)
  1558. retval = do_loopback(&nd, dev_name, flags & MS_REC);
  1559. else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
  1560. retval = do_change_type(&nd, flags);
  1561. else if (flags & MS_MOVE)
  1562. retval = do_move_mount(&nd, dev_name);
  1563. else
  1564. retval = do_new_mount(&nd, type_page, flags, mnt_flags,
  1565. dev_name, data_page);
  1566. dput_out:
  1567. path_put(&nd.path);
  1568. return retval;
  1569. }
  1570. /*
  1571. * Allocate a new namespace structure and populate it with contents
  1572. * copied from the namespace of the passed in task structure.
  1573. */
  1574. static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
  1575. struct fs_struct *fs)
  1576. {
  1577. struct mnt_namespace *new_ns;
  1578. struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
  1579. struct vfsmount *p, *q;
  1580. new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
  1581. if (!new_ns)
  1582. return ERR_PTR(-ENOMEM);
  1583. atomic_set(&new_ns->count, 1);
  1584. INIT_LIST_HEAD(&new_ns->list);
  1585. init_waitqueue_head(&new_ns->poll);
  1586. new_ns->event = 0;
  1587. down_write(&namespace_sem);
  1588. /* First pass: copy the tree topology */
  1589. new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
  1590. CL_COPY_ALL | CL_EXPIRE);
  1591. if (!new_ns->root) {
  1592. up_write(&namespace_sem);
  1593. kfree(new_ns);
  1594. return ERR_PTR(-ENOMEM);;
  1595. }
  1596. spin_lock(&vfsmount_lock);
  1597. list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
  1598. spin_unlock(&vfsmount_lock);
  1599. /*
  1600. * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
  1601. * as belonging to new namespace. We have already acquired a private
  1602. * fs_struct, so tsk->fs->lock is not needed.
  1603. */
  1604. p = mnt_ns->root;
  1605. q = new_ns->root;
  1606. while (p) {
  1607. q->mnt_ns = new_ns;
  1608. if (fs) {
  1609. if (p == fs->root.mnt) {
  1610. rootmnt = p;
  1611. fs->root.mnt = mntget(q);
  1612. }
  1613. if (p == fs->pwd.mnt) {
  1614. pwdmnt = p;
  1615. fs->pwd.mnt = mntget(q);
  1616. }
  1617. if (p == fs->altroot.mnt) {
  1618. altrootmnt = p;
  1619. fs->altroot.mnt = mntget(q);
  1620. }
  1621. }
  1622. p = next_mnt(p, mnt_ns->root);
  1623. q = next_mnt(q, new_ns->root);
  1624. }
  1625. up_write(&namespace_sem);
  1626. if (rootmnt)
  1627. mntput(rootmnt);
  1628. if (pwdmnt)
  1629. mntput(pwdmnt);
  1630. if (altrootmnt)
  1631. mntput(altrootmnt);
  1632. return new_ns;
  1633. }
  1634. struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
  1635. struct fs_struct *new_fs)
  1636. {
  1637. struct mnt_namespace *new_ns;
  1638. BUG_ON(!ns);
  1639. get_mnt_ns(ns);
  1640. if (!(flags & CLONE_NEWNS))
  1641. return ns;
  1642. new_ns = dup_mnt_ns(ns, new_fs);
  1643. put_mnt_ns(ns);
  1644. return new_ns;
  1645. }
  1646. asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
  1647. char __user * type, unsigned long flags,
  1648. void __user * data)
  1649. {
  1650. int retval;
  1651. unsigned long data_page;
  1652. unsigned long type_page;
  1653. unsigned long dev_page;
  1654. char *dir_page;
  1655. retval = copy_mount_options(type, &type_page);
  1656. if (retval < 0)
  1657. return retval;
  1658. dir_page = getname(dir_name);
  1659. retval = PTR_ERR(dir_page);
  1660. if (IS_ERR(dir_page))
  1661. goto out1;
  1662. retval = copy_mount_options(dev_name, &dev_page);
  1663. if (retval < 0)
  1664. goto out2;
  1665. retval = copy_mount_options(data, &data_page);
  1666. if (retval < 0)
  1667. goto out3;
  1668. lock_kernel();
  1669. retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
  1670. flags, (void *)data_page);
  1671. unlock_kernel();
  1672. free_page(data_page);
  1673. out3:
  1674. free_page(dev_page);
  1675. out2:
  1676. putname(dir_page);
  1677. out1:
  1678. free_page(type_page);
  1679. return retval;
  1680. }
  1681. /*
  1682. * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
  1683. * It can block. Requires the big lock held.
  1684. */
  1685. void set_fs_root(struct fs_struct *fs, struct path *path)
  1686. {
  1687. struct path old_root;
  1688. write_lock(&fs->lock);
  1689. old_root = fs->root;
  1690. fs->root = *path;
  1691. path_get(path);
  1692. write_unlock(&fs->lock);
  1693. if (old_root.dentry)
  1694. path_put(&old_root);
  1695. }
  1696. /*
  1697. * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
  1698. * It can block. Requires the big lock held.
  1699. */
  1700. void set_fs_pwd(struct fs_struct *fs, struct path *path)
  1701. {
  1702. struct path old_pwd;
  1703. write_lock(&fs->lock);
  1704. old_pwd = fs->pwd;
  1705. fs->pwd = *path;
  1706. path_get(path);
  1707. write_unlock(&fs->lock);
  1708. if (old_pwd.dentry)
  1709. path_put(&old_pwd);
  1710. }
  1711. static void chroot_fs_refs(struct path *old_root, struct path *new_root)
  1712. {
  1713. struct task_struct *g, *p;
  1714. struct fs_struct *fs;
  1715. read_lock(&tasklist_lock);
  1716. do_each_thread(g, p) {
  1717. task_lock(p);
  1718. fs = p->fs;
  1719. if (fs) {
  1720. atomic_inc(&fs->count);
  1721. task_unlock(p);
  1722. if (fs->root.dentry == old_root->dentry
  1723. && fs->root.mnt == old_root->mnt)
  1724. set_fs_root(fs, new_root);
  1725. if (fs->pwd.dentry == old_root->dentry
  1726. && fs->pwd.mnt == old_root->mnt)
  1727. set_fs_pwd(fs, new_root);
  1728. put_fs_struct(fs);
  1729. } else
  1730. task_unlock(p);
  1731. } while_each_thread(g, p);
  1732. read_unlock(&tasklist_lock);
  1733. }
  1734. /*
  1735. * pivot_root Semantics:
  1736. * Moves the root file system of the current process to the directory put_old,
  1737. * makes new_root as the new root file system of the current process, and sets
  1738. * root/cwd of all processes which had them on the current root to new_root.
  1739. *
  1740. * Restrictions:
  1741. * The new_root and put_old must be directories, and must not be on the
  1742. * same file system as the current process root. The put_old must be
  1743. * underneath new_root, i.e. adding a non-zero number of /.. to the string
  1744. * pointed to by put_old must yield the same directory as new_root. No other
  1745. * file system may be mounted on put_old. After all, new_root is a mountpoint.
  1746. *
  1747. * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
  1748. * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
  1749. * in this situation.
  1750. *
  1751. * Notes:
  1752. * - we don't move root/cwd if they are not at the root (reason: if something
  1753. * cared enough to change them, it's probably wrong to force them elsewhere)
  1754. * - it's okay to pick a root that isn't the root of a file system, e.g.
  1755. * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
  1756. * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
  1757. * first.
  1758. */
  1759. asmlinkage long sys_pivot_root(const char __user * new_root,
  1760. const char __user * put_old)
  1761. {
  1762. struct vfsmount *tmp;
  1763. struct nameidata new_nd, old_nd;
  1764. struct path parent_path, root_parent, root;
  1765. int error;
  1766. if (!capable(CAP_SYS_ADMIN))
  1767. return -EPERM;
  1768. error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
  1769. &new_nd);
  1770. if (error)
  1771. goto out0;
  1772. error = -EINVAL;
  1773. if (!check_mnt(new_nd.path.mnt))
  1774. goto out1;
  1775. error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
  1776. if (error)
  1777. goto out1;
  1778. error = security_sb_pivotroot(&old_nd.path, &new_nd.path);
  1779. if (error) {
  1780. path_put(&old_nd.path);
  1781. goto out1;
  1782. }
  1783. read_lock(&current->fs->lock);
  1784. root = current->fs->root;
  1785. path_get(&current->fs->root);
  1786. read_unlock(&current->fs->lock);
  1787. down_write(&namespace_sem);
  1788. mutex_lock(&old_nd.path.dentry->d_inode->i_mutex);
  1789. error = -EINVAL;
  1790. if (IS_MNT_SHARED(old_nd.path.mnt) ||
  1791. IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) ||
  1792. IS_MNT_SHARED(root.mnt->mnt_parent))
  1793. goto out2;
  1794. if (!check_mnt(root.mnt))
  1795. goto out2;
  1796. error = -ENOENT;
  1797. if (IS_DEADDIR(new_nd.path.dentry->d_inode))
  1798. goto out2;
  1799. if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry))
  1800. goto out2;
  1801. if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry))
  1802. goto out2;
  1803. error = -EBUSY;
  1804. if (new_nd.path.mnt == root.mnt ||
  1805. old_nd.path.mnt == root.mnt)
  1806. goto out2; /* loop, on the same file system */
  1807. error = -EINVAL;
  1808. if (root.mnt->mnt_root != root.dentry)
  1809. goto out2; /* not a mountpoint */
  1810. if (root.mnt->mnt_parent == root.mnt)
  1811. goto out2; /* not attached */
  1812. if (new_nd.path.mnt->mnt_root != new_nd.path.dentry)
  1813. goto out2; /* not a mountpoint */
  1814. if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt)
  1815. goto out2; /* not attached */
  1816. /* make sure we can reach put_old from new_root */
  1817. tmp = old_nd.path.mnt;
  1818. spin_lock(&vfsmount_lock);
  1819. if (tmp != new_nd.path.mnt) {
  1820. for (;;) {
  1821. if (tmp->mnt_parent == tmp)
  1822. goto out3; /* already mounted on put_old */
  1823. if (tmp->mnt_parent == new_nd.path.mnt)
  1824. break;
  1825. tmp = tmp->mnt_parent;
  1826. }
  1827. if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry))
  1828. goto out3;
  1829. } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry))
  1830. goto out3;
  1831. detach_mnt(new_nd.path.mnt, &parent_path);
  1832. detach_mnt(root.mnt, &root_parent);
  1833. /* mount old root on put_old */
  1834. attach_mnt(root.mnt, &old_nd.path);
  1835. /* mount new_root on / */
  1836. attach_mnt(new_nd.path.mnt, &root_parent);
  1837. touch_mnt_namespace(current->nsproxy->mnt_ns);
  1838. spin_unlock(&vfsmount_lock);
  1839. chroot_fs_refs(&root, &new_nd.path);
  1840. security_sb_post_pivotroot(&root, &new_nd.path);
  1841. error = 0;
  1842. path_put(&root_parent);
  1843. path_put(&parent_path);
  1844. out2:
  1845. mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex);
  1846. up_write(&namespace_sem);
  1847. path_put(&root);
  1848. path_put(&old_nd.path);
  1849. out1:
  1850. path_put(&new_nd.path);
  1851. out0:
  1852. return error;
  1853. out3:
  1854. spin_unlock(&vfsmount_lock);
  1855. goto out2;
  1856. }
  1857. static void __init init_mount_tree(void)
  1858. {
  1859. struct vfsmount *mnt;
  1860. struct mnt_namespace *ns;
  1861. struct path root;
  1862. mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
  1863. if (IS_ERR(mnt))
  1864. panic("Can't create rootfs");
  1865. ns = kmalloc(sizeof(*ns), GFP_KERNEL);
  1866. if (!ns)
  1867. panic("Can't allocate initial namespace");
  1868. atomic_set(&ns->count, 1);
  1869. INIT_LIST_HEAD(&ns->list);
  1870. init_waitqueue_head(&ns->poll);
  1871. ns->event = 0;
  1872. list_add(&mnt->mnt_list, &ns->list);
  1873. ns->root = mnt;
  1874. mnt->mnt_ns = ns;
  1875. init_task.nsproxy->mnt_ns = ns;
  1876. get_mnt_ns(ns);
  1877. root.mnt = ns->root;
  1878. root.dentry = ns->root->mnt_root;
  1879. set_fs_pwd(current->fs, &root);
  1880. set_fs_root(current->fs, &root);
  1881. }
  1882. void __init mnt_init(void)
  1883. {
  1884. unsigned u;
  1885. int err;
  1886. init_rwsem(&namespace_sem);
  1887. mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
  1888. 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
  1889. mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
  1890. if (!mount_hashtable)
  1891. panic("Failed to allocate mount hash table\n");
  1892. printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
  1893. for (u = 0; u < HASH_SIZE; u++)
  1894. INIT_LIST_HEAD(&mount_hashtable[u]);
  1895. err = sysfs_init();
  1896. if (err)
  1897. printk(KERN_WARNING "%s: sysfs_init error: %d\n",
  1898. __FUNCTION__, err);
  1899. fs_kobj = kobject_create_and_add("fs", NULL);
  1900. if (!fs_kobj)
  1901. printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__);
  1902. init_rootfs();
  1903. init_mount_tree();
  1904. }
  1905. void __put_mnt_ns(struct mnt_namespace *ns)
  1906. {
  1907. struct vfsmount *root = ns->root;
  1908. LIST_HEAD(umount_list);
  1909. ns->root = NULL;
  1910. spin_unlock(&vfsmount_lock);
  1911. down_write(&namespace_sem);
  1912. spin_lock(&vfsmount_lock);
  1913. umount_tree(root, 0, &umount_list);
  1914. spin_unlock(&vfsmount_lock);
  1915. up_write(&namespace_sem);
  1916. release_mounts(&umount_list);
  1917. kfree(ns);
  1918. }