inode.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837
  1. /*
  2. * hugetlbpage-backed filesystem. Based on ramfs.
  3. *
  4. * William Irwin, 2002
  5. *
  6. * Copyright (C) 2002 Linus Torvalds.
  7. */
  8. #include <linux/module.h>
  9. #include <linux/thread_info.h>
  10. #include <asm/current.h>
  11. #include <linux/sched.h> /* remove ASAP */
  12. #include <linux/fs.h>
  13. #include <linux/mount.h>
  14. #include <linux/file.h>
  15. #include <linux/writeback.h>
  16. #include <linux/pagemap.h>
  17. #include <linux/highmem.h>
  18. #include <linux/init.h>
  19. #include <linux/string.h>
  20. #include <linux/capability.h>
  21. #include <linux/backing-dev.h>
  22. #include <linux/hugetlb.h>
  23. #include <linux/pagevec.h>
  24. #include <linux/quotaops.h>
  25. #include <linux/slab.h>
  26. #include <linux/dnotify.h>
  27. #include <linux/statfs.h>
  28. #include <linux/security.h>
  29. #include <asm/uaccess.h>
  30. /* some random number */
  31. #define HUGETLBFS_MAGIC 0x958458f6
  32. static const struct super_operations hugetlbfs_ops;
  33. static const struct address_space_operations hugetlbfs_aops;
  34. const struct file_operations hugetlbfs_file_operations;
  35. static const struct inode_operations hugetlbfs_dir_inode_operations;
  36. static const struct inode_operations hugetlbfs_inode_operations;
  37. static struct backing_dev_info hugetlbfs_backing_dev_info = {
  38. .ra_pages = 0, /* No readahead */
  39. .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
  40. };
  41. int sysctl_hugetlb_shm_group;
  42. static void huge_pagevec_release(struct pagevec *pvec)
  43. {
  44. int i;
  45. for (i = 0; i < pagevec_count(pvec); ++i)
  46. put_page(pvec->pages[i]);
  47. pagevec_reinit(pvec);
  48. }
  49. static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  50. {
  51. struct inode *inode = file->f_path.dentry->d_inode;
  52. loff_t len, vma_len;
  53. int ret;
  54. /*
  55. * vma alignment has already been checked by prepare_hugepage_range.
  56. * If you add any error returns here, do so after setting VM_HUGETLB,
  57. * so is_vm_hugetlb_page tests below unmap_region go the right way
  58. * when do_mmap_pgoff unwinds (may be important on powerpc and ia64).
  59. */
  60. vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
  61. vma->vm_ops = &hugetlb_vm_ops;
  62. vma_len = (loff_t)(vma->vm_end - vma->vm_start);
  63. mutex_lock(&inode->i_mutex);
  64. file_accessed(file);
  65. ret = -ENOMEM;
  66. len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
  67. if (vma->vm_flags & VM_MAYSHARE &&
  68. hugetlb_reserve_pages(inode, vma->vm_pgoff >> (HPAGE_SHIFT-PAGE_SHIFT),
  69. len >> HPAGE_SHIFT))
  70. goto out;
  71. ret = 0;
  72. hugetlb_prefault_arch_hook(vma->vm_mm);
  73. if (vma->vm_flags & VM_WRITE && inode->i_size < len)
  74. inode->i_size = len;
  75. out:
  76. mutex_unlock(&inode->i_mutex);
  77. return ret;
  78. }
  79. /*
  80. * Called under down_write(mmap_sem).
  81. */
  82. #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
  83. unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
  84. unsigned long len, unsigned long pgoff, unsigned long flags);
  85. #else
  86. static unsigned long
  87. hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
  88. unsigned long len, unsigned long pgoff, unsigned long flags)
  89. {
  90. struct mm_struct *mm = current->mm;
  91. struct vm_area_struct *vma;
  92. unsigned long start_addr;
  93. if (len & ~HPAGE_MASK)
  94. return -EINVAL;
  95. if (len > TASK_SIZE)
  96. return -ENOMEM;
  97. if (addr) {
  98. addr = ALIGN(addr, HPAGE_SIZE);
  99. vma = find_vma(mm, addr);
  100. if (TASK_SIZE - len >= addr &&
  101. (!vma || addr + len <= vma->vm_start))
  102. return addr;
  103. }
  104. start_addr = mm->free_area_cache;
  105. if (len <= mm->cached_hole_size)
  106. start_addr = TASK_UNMAPPED_BASE;
  107. full_search:
  108. addr = ALIGN(start_addr, HPAGE_SIZE);
  109. for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
  110. /* At this point: (!vma || addr < vma->vm_end). */
  111. if (TASK_SIZE - len < addr) {
  112. /*
  113. * Start a new search - just in case we missed
  114. * some holes.
  115. */
  116. if (start_addr != TASK_UNMAPPED_BASE) {
  117. start_addr = TASK_UNMAPPED_BASE;
  118. goto full_search;
  119. }
  120. return -ENOMEM;
  121. }
  122. if (!vma || addr + len <= vma->vm_start)
  123. return addr;
  124. addr = ALIGN(vma->vm_end, HPAGE_SIZE);
  125. }
  126. }
  127. #endif
  128. /*
  129. * Read a page. Again trivial. If it didn't already exist
  130. * in the page cache, it is zero-filled.
  131. */
  132. static int hugetlbfs_readpage(struct file *file, struct page * page)
  133. {
  134. unlock_page(page);
  135. return -EINVAL;
  136. }
  137. static int hugetlbfs_prepare_write(struct file *file,
  138. struct page *page, unsigned offset, unsigned to)
  139. {
  140. return -EINVAL;
  141. }
  142. static int hugetlbfs_commit_write(struct file *file,
  143. struct page *page, unsigned offset, unsigned to)
  144. {
  145. return -EINVAL;
  146. }
  147. static void truncate_huge_page(struct page *page)
  148. {
  149. cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
  150. ClearPageUptodate(page);
  151. remove_from_page_cache(page);
  152. put_page(page);
  153. }
  154. static void truncate_hugepages(struct inode *inode, loff_t lstart)
  155. {
  156. struct address_space *mapping = &inode->i_data;
  157. const pgoff_t start = lstart >> HPAGE_SHIFT;
  158. struct pagevec pvec;
  159. pgoff_t next;
  160. int i, freed = 0;
  161. pagevec_init(&pvec, 0);
  162. next = start;
  163. while (1) {
  164. if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
  165. if (next == start)
  166. break;
  167. next = start;
  168. continue;
  169. }
  170. for (i = 0; i < pagevec_count(&pvec); ++i) {
  171. struct page *page = pvec.pages[i];
  172. lock_page(page);
  173. if (page->index > next)
  174. next = page->index;
  175. ++next;
  176. truncate_huge_page(page);
  177. unlock_page(page);
  178. hugetlb_put_quota(mapping);
  179. freed++;
  180. }
  181. huge_pagevec_release(&pvec);
  182. }
  183. BUG_ON(!lstart && mapping->nrpages);
  184. hugetlb_unreserve_pages(inode, start, freed);
  185. }
  186. static void hugetlbfs_delete_inode(struct inode *inode)
  187. {
  188. truncate_hugepages(inode, 0);
  189. clear_inode(inode);
  190. }
  191. static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
  192. {
  193. struct super_block *sb = inode->i_sb;
  194. if (!hlist_unhashed(&inode->i_hash)) {
  195. if (!(inode->i_state & (I_DIRTY|I_LOCK)))
  196. list_move(&inode->i_list, &inode_unused);
  197. inodes_stat.nr_unused++;
  198. if (!sb || (sb->s_flags & MS_ACTIVE)) {
  199. spin_unlock(&inode_lock);
  200. return;
  201. }
  202. inode->i_state |= I_WILL_FREE;
  203. spin_unlock(&inode_lock);
  204. /*
  205. * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
  206. * in our backing_dev_info.
  207. */
  208. write_inode_now(inode, 1);
  209. spin_lock(&inode_lock);
  210. inode->i_state &= ~I_WILL_FREE;
  211. inodes_stat.nr_unused--;
  212. hlist_del_init(&inode->i_hash);
  213. }
  214. list_del_init(&inode->i_list);
  215. list_del_init(&inode->i_sb_list);
  216. inode->i_state |= I_FREEING;
  217. inodes_stat.nr_inodes--;
  218. spin_unlock(&inode_lock);
  219. truncate_hugepages(inode, 0);
  220. clear_inode(inode);
  221. destroy_inode(inode);
  222. }
  223. static void hugetlbfs_drop_inode(struct inode *inode)
  224. {
  225. if (!inode->i_nlink)
  226. generic_delete_inode(inode);
  227. else
  228. hugetlbfs_forget_inode(inode);
  229. }
  230. static inline void
  231. hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
  232. {
  233. struct vm_area_struct *vma;
  234. struct prio_tree_iter iter;
  235. vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
  236. unsigned long v_offset;
  237. /*
  238. * Can the expression below overflow on 32-bit arches?
  239. * No, because the prio_tree returns us only those vmas
  240. * which overlap the truncated area starting at pgoff,
  241. * and no vma on a 32-bit arch can span beyond the 4GB.
  242. */
  243. if (vma->vm_pgoff < pgoff)
  244. v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
  245. else
  246. v_offset = 0;
  247. __unmap_hugepage_range(vma,
  248. vma->vm_start + v_offset, vma->vm_end);
  249. }
  250. }
  251. /*
  252. * Expanding truncates are not allowed.
  253. */
  254. static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
  255. {
  256. pgoff_t pgoff;
  257. struct address_space *mapping = inode->i_mapping;
  258. if (offset > inode->i_size)
  259. return -EINVAL;
  260. BUG_ON(offset & ~HPAGE_MASK);
  261. pgoff = offset >> PAGE_SHIFT;
  262. inode->i_size = offset;
  263. spin_lock(&mapping->i_mmap_lock);
  264. if (!prio_tree_empty(&mapping->i_mmap))
  265. hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
  266. spin_unlock(&mapping->i_mmap_lock);
  267. truncate_hugepages(inode, offset);
  268. return 0;
  269. }
  270. static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
  271. {
  272. struct inode *inode = dentry->d_inode;
  273. int error;
  274. unsigned int ia_valid = attr->ia_valid;
  275. BUG_ON(!inode);
  276. error = inode_change_ok(inode, attr);
  277. if (error)
  278. goto out;
  279. if (ia_valid & ATTR_SIZE) {
  280. error = -EINVAL;
  281. if (!(attr->ia_size & ~HPAGE_MASK))
  282. error = hugetlb_vmtruncate(inode, attr->ia_size);
  283. if (error)
  284. goto out;
  285. attr->ia_valid &= ~ATTR_SIZE;
  286. }
  287. error = inode_setattr(inode, attr);
  288. out:
  289. return error;
  290. }
  291. static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
  292. gid_t gid, int mode, dev_t dev)
  293. {
  294. struct inode *inode;
  295. inode = new_inode(sb);
  296. if (inode) {
  297. struct hugetlbfs_inode_info *info;
  298. inode->i_mode = mode;
  299. inode->i_uid = uid;
  300. inode->i_gid = gid;
  301. inode->i_blocks = 0;
  302. inode->i_mapping->a_ops = &hugetlbfs_aops;
  303. inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
  304. inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
  305. INIT_LIST_HEAD(&inode->i_mapping->private_list);
  306. info = HUGETLBFS_I(inode);
  307. mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
  308. switch (mode & S_IFMT) {
  309. default:
  310. init_special_inode(inode, mode, dev);
  311. break;
  312. case S_IFREG:
  313. inode->i_op = &hugetlbfs_inode_operations;
  314. inode->i_fop = &hugetlbfs_file_operations;
  315. break;
  316. case S_IFDIR:
  317. inode->i_op = &hugetlbfs_dir_inode_operations;
  318. inode->i_fop = &simple_dir_operations;
  319. /* directory inodes start off with i_nlink == 2 (for "." entry) */
  320. inc_nlink(inode);
  321. break;
  322. case S_IFLNK:
  323. inode->i_op = &page_symlink_inode_operations;
  324. break;
  325. }
  326. }
  327. return inode;
  328. }
  329. /*
  330. * File creation. Allocate an inode, and we're done..
  331. */
  332. static int hugetlbfs_mknod(struct inode *dir,
  333. struct dentry *dentry, int mode, dev_t dev)
  334. {
  335. struct inode *inode;
  336. int error = -ENOSPC;
  337. gid_t gid;
  338. if (dir->i_mode & S_ISGID) {
  339. gid = dir->i_gid;
  340. if (S_ISDIR(mode))
  341. mode |= S_ISGID;
  342. } else {
  343. gid = current->fsgid;
  344. }
  345. inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
  346. if (inode) {
  347. dir->i_ctime = dir->i_mtime = CURRENT_TIME;
  348. d_instantiate(dentry, inode);
  349. dget(dentry); /* Extra count - pin the dentry in core */
  350. error = 0;
  351. }
  352. return error;
  353. }
  354. static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
  355. {
  356. int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
  357. if (!retval)
  358. inc_nlink(dir);
  359. return retval;
  360. }
  361. static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
  362. {
  363. return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
  364. }
  365. static int hugetlbfs_symlink(struct inode *dir,
  366. struct dentry *dentry, const char *symname)
  367. {
  368. struct inode *inode;
  369. int error = -ENOSPC;
  370. gid_t gid;
  371. if (dir->i_mode & S_ISGID)
  372. gid = dir->i_gid;
  373. else
  374. gid = current->fsgid;
  375. inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
  376. gid, S_IFLNK|S_IRWXUGO, 0);
  377. if (inode) {
  378. int l = strlen(symname)+1;
  379. error = page_symlink(inode, symname, l);
  380. if (!error) {
  381. d_instantiate(dentry, inode);
  382. dget(dentry);
  383. } else
  384. iput(inode);
  385. }
  386. dir->i_ctime = dir->i_mtime = CURRENT_TIME;
  387. return error;
  388. }
  389. /*
  390. * mark the head page dirty
  391. */
  392. static int hugetlbfs_set_page_dirty(struct page *page)
  393. {
  394. struct page *head = (struct page *)page_private(page);
  395. SetPageDirty(head);
  396. return 0;
  397. }
  398. static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  399. {
  400. struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
  401. buf->f_type = HUGETLBFS_MAGIC;
  402. buf->f_bsize = HPAGE_SIZE;
  403. if (sbinfo) {
  404. spin_lock(&sbinfo->stat_lock);
  405. /* If no limits set, just report 0 for max/free/used
  406. * blocks, like simple_statfs() */
  407. if (sbinfo->max_blocks >= 0) {
  408. buf->f_blocks = sbinfo->max_blocks;
  409. buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
  410. buf->f_files = sbinfo->max_inodes;
  411. buf->f_ffree = sbinfo->free_inodes;
  412. }
  413. spin_unlock(&sbinfo->stat_lock);
  414. }
  415. buf->f_namelen = NAME_MAX;
  416. return 0;
  417. }
  418. static void hugetlbfs_put_super(struct super_block *sb)
  419. {
  420. struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
  421. if (sbi) {
  422. sb->s_fs_info = NULL;
  423. kfree(sbi);
  424. }
  425. }
  426. static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
  427. {
  428. if (sbinfo->free_inodes >= 0) {
  429. spin_lock(&sbinfo->stat_lock);
  430. if (unlikely(!sbinfo->free_inodes)) {
  431. spin_unlock(&sbinfo->stat_lock);
  432. return 0;
  433. }
  434. sbinfo->free_inodes--;
  435. spin_unlock(&sbinfo->stat_lock);
  436. }
  437. return 1;
  438. }
  439. static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
  440. {
  441. if (sbinfo->free_inodes >= 0) {
  442. spin_lock(&sbinfo->stat_lock);
  443. sbinfo->free_inodes++;
  444. spin_unlock(&sbinfo->stat_lock);
  445. }
  446. }
  447. static struct kmem_cache *hugetlbfs_inode_cachep;
  448. static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
  449. {
  450. struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
  451. struct hugetlbfs_inode_info *p;
  452. if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
  453. return NULL;
  454. p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
  455. if (unlikely(!p)) {
  456. hugetlbfs_inc_free_inodes(sbinfo);
  457. return NULL;
  458. }
  459. return &p->vfs_inode;
  460. }
  461. static void hugetlbfs_destroy_inode(struct inode *inode)
  462. {
  463. hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
  464. mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
  465. kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
  466. }
  467. static const struct address_space_operations hugetlbfs_aops = {
  468. .readpage = hugetlbfs_readpage,
  469. .prepare_write = hugetlbfs_prepare_write,
  470. .commit_write = hugetlbfs_commit_write,
  471. .set_page_dirty = hugetlbfs_set_page_dirty,
  472. };
  473. static void init_once(void *foo, struct kmem_cache *cachep, unsigned long flags)
  474. {
  475. struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
  476. if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
  477. SLAB_CTOR_CONSTRUCTOR)
  478. inode_init_once(&ei->vfs_inode);
  479. }
  480. const struct file_operations hugetlbfs_file_operations = {
  481. .mmap = hugetlbfs_file_mmap,
  482. .fsync = simple_sync_file,
  483. .get_unmapped_area = hugetlb_get_unmapped_area,
  484. };
  485. static const struct inode_operations hugetlbfs_dir_inode_operations = {
  486. .create = hugetlbfs_create,
  487. .lookup = simple_lookup,
  488. .link = simple_link,
  489. .unlink = simple_unlink,
  490. .symlink = hugetlbfs_symlink,
  491. .mkdir = hugetlbfs_mkdir,
  492. .rmdir = simple_rmdir,
  493. .mknod = hugetlbfs_mknod,
  494. .rename = simple_rename,
  495. .setattr = hugetlbfs_setattr,
  496. };
  497. static const struct inode_operations hugetlbfs_inode_operations = {
  498. .setattr = hugetlbfs_setattr,
  499. };
  500. static const struct super_operations hugetlbfs_ops = {
  501. .alloc_inode = hugetlbfs_alloc_inode,
  502. .destroy_inode = hugetlbfs_destroy_inode,
  503. .statfs = hugetlbfs_statfs,
  504. .delete_inode = hugetlbfs_delete_inode,
  505. .drop_inode = hugetlbfs_drop_inode,
  506. .put_super = hugetlbfs_put_super,
  507. };
  508. static int
  509. hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
  510. {
  511. char *opt, *value, *rest;
  512. if (!options)
  513. return 0;
  514. while ((opt = strsep(&options, ",")) != NULL) {
  515. if (!*opt)
  516. continue;
  517. value = strchr(opt, '=');
  518. if (!value || !*value)
  519. return -EINVAL;
  520. else
  521. *value++ = '\0';
  522. if (!strcmp(opt, "uid"))
  523. pconfig->uid = simple_strtoul(value, &value, 0);
  524. else if (!strcmp(opt, "gid"))
  525. pconfig->gid = simple_strtoul(value, &value, 0);
  526. else if (!strcmp(opt, "mode"))
  527. pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
  528. else if (!strcmp(opt, "size")) {
  529. unsigned long long size = memparse(value, &rest);
  530. if (*rest == '%') {
  531. size <<= HPAGE_SHIFT;
  532. size *= max_huge_pages;
  533. do_div(size, 100);
  534. rest++;
  535. }
  536. pconfig->nr_blocks = (size >> HPAGE_SHIFT);
  537. value = rest;
  538. } else if (!strcmp(opt,"nr_inodes")) {
  539. pconfig->nr_inodes = memparse(value, &rest);
  540. value = rest;
  541. } else
  542. return -EINVAL;
  543. if (*value)
  544. return -EINVAL;
  545. }
  546. return 0;
  547. }
  548. static int
  549. hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
  550. {
  551. struct inode * inode;
  552. struct dentry * root;
  553. int ret;
  554. struct hugetlbfs_config config;
  555. struct hugetlbfs_sb_info *sbinfo;
  556. config.nr_blocks = -1; /* No limit on size by default */
  557. config.nr_inodes = -1; /* No limit on number of inodes by default */
  558. config.uid = current->fsuid;
  559. config.gid = current->fsgid;
  560. config.mode = 0755;
  561. ret = hugetlbfs_parse_options(data, &config);
  562. if (ret)
  563. return ret;
  564. sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
  565. if (!sbinfo)
  566. return -ENOMEM;
  567. sb->s_fs_info = sbinfo;
  568. spin_lock_init(&sbinfo->stat_lock);
  569. sbinfo->max_blocks = config.nr_blocks;
  570. sbinfo->free_blocks = config.nr_blocks;
  571. sbinfo->max_inodes = config.nr_inodes;
  572. sbinfo->free_inodes = config.nr_inodes;
  573. sb->s_maxbytes = MAX_LFS_FILESIZE;
  574. sb->s_blocksize = HPAGE_SIZE;
  575. sb->s_blocksize_bits = HPAGE_SHIFT;
  576. sb->s_magic = HUGETLBFS_MAGIC;
  577. sb->s_op = &hugetlbfs_ops;
  578. sb->s_time_gran = 1;
  579. inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
  580. S_IFDIR | config.mode, 0);
  581. if (!inode)
  582. goto out_free;
  583. root = d_alloc_root(inode);
  584. if (!root) {
  585. iput(inode);
  586. goto out_free;
  587. }
  588. sb->s_root = root;
  589. return 0;
  590. out_free:
  591. kfree(sbinfo);
  592. return -ENOMEM;
  593. }
  594. int hugetlb_get_quota(struct address_space *mapping)
  595. {
  596. int ret = 0;
  597. struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
  598. if (sbinfo->free_blocks > -1) {
  599. spin_lock(&sbinfo->stat_lock);
  600. if (sbinfo->free_blocks > 0)
  601. sbinfo->free_blocks--;
  602. else
  603. ret = -ENOMEM;
  604. spin_unlock(&sbinfo->stat_lock);
  605. }
  606. return ret;
  607. }
  608. void hugetlb_put_quota(struct address_space *mapping)
  609. {
  610. struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
  611. if (sbinfo->free_blocks > -1) {
  612. spin_lock(&sbinfo->stat_lock);
  613. sbinfo->free_blocks++;
  614. spin_unlock(&sbinfo->stat_lock);
  615. }
  616. }
  617. static int hugetlbfs_get_sb(struct file_system_type *fs_type,
  618. int flags, const char *dev_name, void *data, struct vfsmount *mnt)
  619. {
  620. return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
  621. }
  622. static struct file_system_type hugetlbfs_fs_type = {
  623. .name = "hugetlbfs",
  624. .get_sb = hugetlbfs_get_sb,
  625. .kill_sb = kill_litter_super,
  626. };
  627. static struct vfsmount *hugetlbfs_vfsmount;
  628. static int can_do_hugetlb_shm(void)
  629. {
  630. return likely(capable(CAP_IPC_LOCK) ||
  631. in_group_p(sysctl_hugetlb_shm_group) ||
  632. can_do_mlock());
  633. }
  634. struct file *hugetlb_zero_setup(size_t size)
  635. {
  636. int error = -ENOMEM;
  637. struct file *file;
  638. struct inode *inode;
  639. struct dentry *dentry, *root;
  640. struct qstr quick_string;
  641. char buf[16];
  642. static atomic_t counter;
  643. if (!can_do_hugetlb_shm())
  644. return ERR_PTR(-EPERM);
  645. if (!user_shm_lock(size, current->user))
  646. return ERR_PTR(-ENOMEM);
  647. root = hugetlbfs_vfsmount->mnt_root;
  648. snprintf(buf, 16, "%u", atomic_inc_return(&counter));
  649. quick_string.name = buf;
  650. quick_string.len = strlen(quick_string.name);
  651. quick_string.hash = 0;
  652. dentry = d_alloc(root, &quick_string);
  653. if (!dentry)
  654. goto out_shm_unlock;
  655. error = -ENFILE;
  656. file = get_empty_filp();
  657. if (!file)
  658. goto out_dentry;
  659. error = -ENOSPC;
  660. inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
  661. current->fsgid, S_IFREG | S_IRWXUGO, 0);
  662. if (!inode)
  663. goto out_file;
  664. error = -ENOMEM;
  665. if (hugetlb_reserve_pages(inode, 0, size >> HPAGE_SHIFT))
  666. goto out_inode;
  667. d_instantiate(dentry, inode);
  668. inode->i_size = size;
  669. inode->i_nlink = 0;
  670. file->f_path.mnt = mntget(hugetlbfs_vfsmount);
  671. file->f_path.dentry = dentry;
  672. file->f_mapping = inode->i_mapping;
  673. file->f_op = &hugetlbfs_file_operations;
  674. file->f_mode = FMODE_WRITE | FMODE_READ;
  675. return file;
  676. out_inode:
  677. iput(inode);
  678. out_file:
  679. put_filp(file);
  680. out_dentry:
  681. dput(dentry);
  682. out_shm_unlock:
  683. user_shm_unlock(size, current->user);
  684. return ERR_PTR(error);
  685. }
  686. static int __init init_hugetlbfs_fs(void)
  687. {
  688. int error;
  689. struct vfsmount *vfsmount;
  690. hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
  691. sizeof(struct hugetlbfs_inode_info),
  692. 0, 0, init_once, NULL);
  693. if (hugetlbfs_inode_cachep == NULL)
  694. return -ENOMEM;
  695. error = register_filesystem(&hugetlbfs_fs_type);
  696. if (error)
  697. goto out;
  698. vfsmount = kern_mount(&hugetlbfs_fs_type);
  699. if (!IS_ERR(vfsmount)) {
  700. hugetlbfs_vfsmount = vfsmount;
  701. return 0;
  702. }
  703. error = PTR_ERR(vfsmount);
  704. out:
  705. if (error)
  706. kmem_cache_destroy(hugetlbfs_inode_cachep);
  707. return error;
  708. }
  709. static void __exit exit_hugetlbfs_fs(void)
  710. {
  711. kmem_cache_destroy(hugetlbfs_inode_cachep);
  712. unregister_filesystem(&hugetlbfs_fs_type);
  713. }
  714. module_init(init_hugetlbfs_fs)
  715. module_exit(exit_hugetlbfs_fs)
  716. MODULE_LICENSE("GPL");