11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * hugetlbpage-backed filesystem. Based on ramfs. 31da177e4SLinus Torvalds * 46d49e352SNadia Yvette Chambers * Nadia Yvette Chambers, 2002 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 2002 Linus Torvalds. 73e89e1c5SPaul Gortmaker * License: GPL 81da177e4SLinus Torvalds */ 91da177e4SLinus Torvalds 109b857d26SAndrew Morton #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 119b857d26SAndrew Morton 121da177e4SLinus Torvalds #include <linux/thread_info.h> 131da177e4SLinus Torvalds #include <asm/current.h> 14174cd4b1SIngo Molnar #include <linux/sched/signal.h> /* remove ASAP */ 1570c3547eSMike Kravetz #include <linux/falloc.h> 161da177e4SLinus Torvalds #include <linux/fs.h> 171da177e4SLinus Torvalds #include <linux/mount.h> 181da177e4SLinus Torvalds #include <linux/file.h> 19e73a75faSRandy Dunlap #include <linux/kernel.h> 201da177e4SLinus Torvalds #include <linux/writeback.h> 211da177e4SLinus Torvalds #include <linux/pagemap.h> 221da177e4SLinus Torvalds #include <linux/highmem.h> 231da177e4SLinus Torvalds #include <linux/init.h> 241da177e4SLinus Torvalds #include <linux/string.h> 2516f7e0feSRandy Dunlap #include <linux/capability.h> 26e73a75faSRandy Dunlap #include <linux/ctype.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 281da177e4SLinus Torvalds #include <linux/hugetlb.h> 291da177e4SLinus Torvalds #include <linux/pagevec.h> 3032021982SDavid Howells #include <linux/fs_parser.h> 31036e0856SBenjamin Herrenschmidt #include <linux/mman.h> 321da177e4SLinus Torvalds #include <linux/slab.h> 331da177e4SLinus Torvalds #include <linux/dnotify.h> 341da177e4SLinus Torvalds #include <linux/statfs.h> 351da177e4SLinus Torvalds #include <linux/security.h> 361fd7317dSNick Black #include <linux/magic.h> 37290408d4SNaoya Horiguchi #include <linux/migrate.h> 3834d0640eSAl Viro #include <linux/uio.h> 391da177e4SLinus Torvalds 407c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 411da177e4SLinus Torvalds 42ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops; 43f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops; 444b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations; 4592e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations; 4692e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations; 471da177e4SLinus Torvalds 4832021982SDavid Howells enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT }; 4932021982SDavid Howells 5032021982SDavid Howells struct hugetlbfs_fs_context { 514a25220dSDavid Howells struct hstate *hstate; 5232021982SDavid Howells unsigned long long max_size_opt; 5332021982SDavid Howells unsigned long long min_size_opt; 544a25220dSDavid Howells long max_hpages; 554a25220dSDavid Howells long nr_inodes; 564a25220dSDavid Howells long min_hpages; 5732021982SDavid Howells enum hugetlbfs_size_type max_val_type; 5832021982SDavid Howells enum hugetlbfs_size_type min_val_type; 59a0eb3a05SEric W. Biederman kuid_t uid; 60a0eb3a05SEric W. Biederman kgid_t gid; 61a1d776eeSDavid Gibson umode_t mode; 62a1d776eeSDavid Gibson }; 63a1d776eeSDavid Gibson 641da177e4SLinus Torvalds int sysctl_hugetlb_shm_group; 651da177e4SLinus Torvalds 6632021982SDavid Howells enum hugetlb_param { 6732021982SDavid Howells Opt_gid, 6832021982SDavid Howells Opt_min_size, 6932021982SDavid Howells Opt_mode, 7032021982SDavid Howells Opt_nr_inodes, 7132021982SDavid Howells Opt_pagesize, 7232021982SDavid Howells Opt_size, 7332021982SDavid Howells Opt_uid, 74e73a75faSRandy Dunlap }; 75e73a75faSRandy Dunlap 76d7167b14SAl Viro static const struct fs_parameter_spec hugetlb_fs_parameters[] = { 7732021982SDavid Howells fsparam_u32 ("gid", Opt_gid), 7832021982SDavid Howells fsparam_string("min_size", Opt_min_size), 7932021982SDavid Howells fsparam_u32 ("mode", Opt_mode), 8032021982SDavid Howells fsparam_string("nr_inodes", Opt_nr_inodes), 8132021982SDavid Howells fsparam_string("pagesize", Opt_pagesize), 8232021982SDavid Howells fsparam_string("size", Opt_size), 8332021982SDavid Howells fsparam_u32 ("uid", Opt_uid), 8432021982SDavid Howells {} 8532021982SDavid Howells }; 8632021982SDavid Howells 8770c3547eSMike Kravetz #ifdef CONFIG_NUMA 8870c3547eSMike Kravetz static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, 8970c3547eSMike Kravetz struct inode *inode, pgoff_t index) 9070c3547eSMike Kravetz { 9170c3547eSMike Kravetz vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy, 9270c3547eSMike Kravetz index); 9370c3547eSMike Kravetz } 9470c3547eSMike Kravetz 9570c3547eSMike Kravetz static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) 9670c3547eSMike Kravetz { 9770c3547eSMike Kravetz mpol_cond_put(vma->vm_policy); 9870c3547eSMike Kravetz } 9970c3547eSMike Kravetz #else 10070c3547eSMike Kravetz static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, 10170c3547eSMike Kravetz struct inode *inode, pgoff_t index) 10270c3547eSMike Kravetz { 10370c3547eSMike Kravetz } 10470c3547eSMike Kravetz 10570c3547eSMike Kravetz static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) 10670c3547eSMike Kravetz { 10770c3547eSMike Kravetz } 10870c3547eSMike Kravetz #endif 10970c3547eSMike Kravetz 1102e9b367cSAdam Litke static void huge_pagevec_release(struct pagevec *pvec) 1112e9b367cSAdam Litke { 1122e9b367cSAdam Litke int i; 1132e9b367cSAdam Litke 1142e9b367cSAdam Litke for (i = 0; i < pagevec_count(pvec); ++i) 1152e9b367cSAdam Litke put_page(pvec->pages[i]); 1162e9b367cSAdam Litke 1172e9b367cSAdam Litke pagevec_reinit(pvec); 1182e9b367cSAdam Litke } 1192e9b367cSAdam Litke 12063489f8eSMike Kravetz /* 12163489f8eSMike Kravetz * Mask used when checking the page offset value passed in via system 12263489f8eSMike Kravetz * calls. This value will be converted to a loff_t which is signed. 12363489f8eSMike Kravetz * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the 12463489f8eSMike Kravetz * value. The extra bit (- 1 in the shift value) is to take the sign 12563489f8eSMike Kravetz * bit into account. 12663489f8eSMike Kravetz */ 12763489f8eSMike Kravetz #define PGOFF_LOFFT_MAX \ 12863489f8eSMike Kravetz (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1))) 12963489f8eSMike Kravetz 1301da177e4SLinus Torvalds static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) 1311da177e4SLinus Torvalds { 132496ad9aaSAl Viro struct inode *inode = file_inode(file); 1331da177e4SLinus Torvalds loff_t len, vma_len; 1341da177e4SLinus Torvalds int ret; 135a5516438SAndi Kleen struct hstate *h = hstate_file(file); 1361da177e4SLinus Torvalds 13768589bc3SHugh Dickins /* 138dec4ad86SDavid Gibson * vma address alignment (but not the pgoff alignment) has 139dec4ad86SDavid Gibson * already been checked by prepare_hugepage_range. If you add 140dec4ad86SDavid Gibson * any error returns here, do so after setting VM_HUGETLB, so 141dec4ad86SDavid Gibson * is_vm_hugetlb_page tests below unmap_region go the right 142dec4ad86SDavid Gibson * way when do_mmap_pgoff unwinds (may be important on powerpc 143dec4ad86SDavid Gibson * and ia64). 14468589bc3SHugh Dickins */ 145a2fce914SNaoya Horiguchi vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; 14668589bc3SHugh Dickins vma->vm_ops = &hugetlb_vm_ops; 1471da177e4SLinus Torvalds 148045c7a3fSMike Kravetz /* 14963489f8eSMike Kravetz * page based offset in vm_pgoff could be sufficiently large to 1505df63c2aSMike Kravetz * overflow a loff_t when converted to byte offset. This can 1515df63c2aSMike Kravetz * only happen on architectures where sizeof(loff_t) == 1525df63c2aSMike Kravetz * sizeof(unsigned long). So, only check in those instances. 153045c7a3fSMike Kravetz */ 1545df63c2aSMike Kravetz if (sizeof(unsigned long) == sizeof(loff_t)) { 15563489f8eSMike Kravetz if (vma->vm_pgoff & PGOFF_LOFFT_MAX) 156045c7a3fSMike Kravetz return -EINVAL; 1575df63c2aSMike Kravetz } 158045c7a3fSMike Kravetz 15963489f8eSMike Kravetz /* must be huge page aligned */ 1602b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 161dec4ad86SDavid Gibson return -EINVAL; 162dec4ad86SDavid Gibson 1631da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 164045c7a3fSMike Kravetz len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 165045c7a3fSMike Kravetz /* check for overflow */ 166045c7a3fSMike Kravetz if (len < vma_len) 167045c7a3fSMike Kravetz return -EINVAL; 1681da177e4SLinus Torvalds 1695955102cSAl Viro inode_lock(inode); 1701da177e4SLinus Torvalds file_accessed(file); 1711da177e4SLinus Torvalds 1721da177e4SLinus Torvalds ret = -ENOMEM; 173a1e78772SMel Gorman if (hugetlb_reserve_pages(inode, 174a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1755a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1765a6fe125SMel Gorman vma->vm_flags)) 177b45b5bd6SDavid Gibson goto out; 178b45b5bd6SDavid Gibson 1794c887265SAdam Litke ret = 0; 180b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 181045c7a3fSMike Kravetz i_size_write(inode, len); 1821da177e4SLinus Torvalds out: 1835955102cSAl Viro inode_unlock(inode); 1841da177e4SLinus Torvalds 1851da177e4SLinus Torvalds return ret; 1861da177e4SLinus Torvalds } 1871da177e4SLinus Torvalds 1881da177e4SLinus Torvalds /* 189508034a3SHugh Dickins * Called under down_write(mmap_sem). 1901da177e4SLinus Torvalds */ 1911da177e4SLinus Torvalds 192d2ba27e8SAdrian Bunk #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 1931da177e4SLinus Torvalds static unsigned long 1941da177e4SLinus Torvalds hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 1951da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 1961da177e4SLinus Torvalds { 1971da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1981da177e4SLinus Torvalds struct vm_area_struct *vma; 199a5516438SAndi Kleen struct hstate *h = hstate_file(file); 20008659355SMichel Lespinasse struct vm_unmapped_area_info info; 2011da177e4SLinus Torvalds 202a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 2031da177e4SLinus Torvalds return -EINVAL; 2041da177e4SLinus Torvalds if (len > TASK_SIZE) 2051da177e4SLinus Torvalds return -ENOMEM; 2061da177e4SLinus Torvalds 207036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 208a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 209036e0856SBenjamin Herrenschmidt return -EINVAL; 210036e0856SBenjamin Herrenschmidt return addr; 211036e0856SBenjamin Herrenschmidt } 212036e0856SBenjamin Herrenschmidt 2131da177e4SLinus Torvalds if (addr) { 214a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 2151da177e4SLinus Torvalds vma = find_vma(mm, addr); 2161da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 2171be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma))) 2181da177e4SLinus Torvalds return addr; 2191da177e4SLinus Torvalds } 2201da177e4SLinus Torvalds 22108659355SMichel Lespinasse info.flags = 0; 22208659355SMichel Lespinasse info.length = len; 22308659355SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 22408659355SMichel Lespinasse info.high_limit = TASK_SIZE; 22508659355SMichel Lespinasse info.align_mask = PAGE_MASK & ~huge_page_mask(h); 22608659355SMichel Lespinasse info.align_offset = 0; 22708659355SMichel Lespinasse return vm_unmapped_area(&info); 2281da177e4SLinus Torvalds } 2291da177e4SLinus Torvalds #endif 2301da177e4SLinus Torvalds 23134d0640eSAl Viro static size_t 232e63e1e5aSBadari Pulavarty hugetlbfs_read_actor(struct page *page, unsigned long offset, 23334d0640eSAl Viro struct iov_iter *to, unsigned long size) 234e63e1e5aSBadari Pulavarty { 23534d0640eSAl Viro size_t copied = 0; 236e63e1e5aSBadari Pulavarty int i, chunksize; 237e63e1e5aSBadari Pulavarty 238e63e1e5aSBadari Pulavarty /* Find which 4k chunk and offset with in that chunk */ 23909cbfeafSKirill A. Shutemov i = offset >> PAGE_SHIFT; 24009cbfeafSKirill A. Shutemov offset = offset & ~PAGE_MASK; 241e63e1e5aSBadari Pulavarty 242e63e1e5aSBadari Pulavarty while (size) { 24334d0640eSAl Viro size_t n; 24409cbfeafSKirill A. Shutemov chunksize = PAGE_SIZE; 245e63e1e5aSBadari Pulavarty if (offset) 246e63e1e5aSBadari Pulavarty chunksize -= offset; 247e63e1e5aSBadari Pulavarty if (chunksize > size) 248e63e1e5aSBadari Pulavarty chunksize = size; 24934d0640eSAl Viro n = copy_page_to_iter(&page[i], offset, chunksize, to); 25034d0640eSAl Viro copied += n; 25134d0640eSAl Viro if (n != chunksize) 25234d0640eSAl Viro return copied; 253e63e1e5aSBadari Pulavarty offset = 0; 254e63e1e5aSBadari Pulavarty size -= chunksize; 255e63e1e5aSBadari Pulavarty i++; 256e63e1e5aSBadari Pulavarty } 25734d0640eSAl Viro return copied; 258e63e1e5aSBadari Pulavarty } 259e63e1e5aSBadari Pulavarty 260e63e1e5aSBadari Pulavarty /* 261e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 262e63e1e5aSBadari Pulavarty * data. Its *very* similar to do_generic_mapping_read(), we can't use that 263ea1754a0SKirill A. Shutemov * since it has PAGE_SIZE assumptions. 264e63e1e5aSBadari Pulavarty */ 26534d0640eSAl Viro static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) 266e63e1e5aSBadari Pulavarty { 26734d0640eSAl Viro struct file *file = iocb->ki_filp; 26834d0640eSAl Viro struct hstate *h = hstate_file(file); 26934d0640eSAl Viro struct address_space *mapping = file->f_mapping; 270e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 27134d0640eSAl Viro unsigned long index = iocb->ki_pos >> huge_page_shift(h); 27234d0640eSAl Viro unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); 273e63e1e5aSBadari Pulavarty unsigned long end_index; 274e63e1e5aSBadari Pulavarty loff_t isize; 275e63e1e5aSBadari Pulavarty ssize_t retval = 0; 276e63e1e5aSBadari Pulavarty 27734d0640eSAl Viro while (iov_iter_count(to)) { 278e63e1e5aSBadari Pulavarty struct page *page; 27934d0640eSAl Viro size_t nr, copied; 280e63e1e5aSBadari Pulavarty 281e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 282a5516438SAndi Kleen nr = huge_page_size(h); 283a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 284a05b0855SAneesh Kumar K.V if (!isize) 28534d0640eSAl Viro break; 286a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 287e63e1e5aSBadari Pulavarty if (index > end_index) 28834d0640eSAl Viro break; 28934d0640eSAl Viro if (index == end_index) { 290a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 291a05b0855SAneesh Kumar K.V if (nr <= offset) 29234d0640eSAl Viro break; 293e63e1e5aSBadari Pulavarty } 294e63e1e5aSBadari Pulavarty nr = nr - offset; 295e63e1e5aSBadari Pulavarty 296e63e1e5aSBadari Pulavarty /* Find the page */ 297a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 298e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 299e63e1e5aSBadari Pulavarty /* 300e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 301e63e1e5aSBadari Pulavarty * length of the hole or request. 302e63e1e5aSBadari Pulavarty */ 30334d0640eSAl Viro copied = iov_iter_zero(nr, to); 304e63e1e5aSBadari Pulavarty } else { 305a05b0855SAneesh Kumar K.V unlock_page(page); 306a05b0855SAneesh Kumar K.V 307e63e1e5aSBadari Pulavarty /* 308e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 309e63e1e5aSBadari Pulavarty */ 31034d0640eSAl Viro copied = hugetlbfs_read_actor(page, offset, to, nr); 31109cbfeafSKirill A. Shutemov put_page(page); 312e63e1e5aSBadari Pulavarty } 31334d0640eSAl Viro offset += copied; 31434d0640eSAl Viro retval += copied; 31534d0640eSAl Viro if (copied != nr && iov_iter_count(to)) { 31634d0640eSAl Viro if (!retval) 31734d0640eSAl Viro retval = -EFAULT; 318e63e1e5aSBadari Pulavarty break; 319e63e1e5aSBadari Pulavarty } 32034d0640eSAl Viro index += offset >> huge_page_shift(h); 32134d0640eSAl Viro offset &= ~huge_page_mask(h); 32234d0640eSAl Viro } 32334d0640eSAl Viro iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; 324e63e1e5aSBadari Pulavarty return retval; 325e63e1e5aSBadari Pulavarty } 326e63e1e5aSBadari Pulavarty 327800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 328800d15a5SNick Piggin struct address_space *mapping, 329800d15a5SNick Piggin loff_t pos, unsigned len, unsigned flags, 330800d15a5SNick Piggin struct page **pagep, void **fsdata) 3311da177e4SLinus Torvalds { 3321da177e4SLinus Torvalds return -EINVAL; 3331da177e4SLinus Torvalds } 3341da177e4SLinus Torvalds 335800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 336800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 337800d15a5SNick Piggin struct page *page, void *fsdata) 3381da177e4SLinus Torvalds { 339800d15a5SNick Piggin BUG(); 3401da177e4SLinus Torvalds return -EINVAL; 3411da177e4SLinus Torvalds } 3421da177e4SLinus Torvalds 343b5cec28dSMike Kravetz static void remove_huge_page(struct page *page) 3441da177e4SLinus Torvalds { 345b9ea2515SKonstantin Khlebnikov ClearPageDirty(page); 3461da177e4SLinus Torvalds ClearPageUptodate(page); 347bd65cb86SMinchan Kim delete_from_page_cache(page); 3481da177e4SLinus Torvalds } 3491da177e4SLinus Torvalds 3504aae8d1cSMike Kravetz static void 351f808c13fSDavidlohr Bueso hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end) 3524aae8d1cSMike Kravetz { 3534aae8d1cSMike Kravetz struct vm_area_struct *vma; 3544aae8d1cSMike Kravetz 3554aae8d1cSMike Kravetz /* 3564aae8d1cSMike Kravetz * end == 0 indicates that the entire range after 3574aae8d1cSMike Kravetz * start should be unmapped. 3584aae8d1cSMike Kravetz */ 3594aae8d1cSMike Kravetz vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { 3604aae8d1cSMike Kravetz unsigned long v_offset; 3614aae8d1cSMike Kravetz unsigned long v_end; 3624aae8d1cSMike Kravetz 3634aae8d1cSMike Kravetz /* 3644aae8d1cSMike Kravetz * Can the expression below overflow on 32-bit arches? 3654aae8d1cSMike Kravetz * No, because the interval tree returns us only those vmas 3664aae8d1cSMike Kravetz * which overlap the truncated area starting at pgoff, 3674aae8d1cSMike Kravetz * and no vma on a 32-bit arch can span beyond the 4GB. 3684aae8d1cSMike Kravetz */ 3694aae8d1cSMike Kravetz if (vma->vm_pgoff < start) 3704aae8d1cSMike Kravetz v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; 3714aae8d1cSMike Kravetz else 3724aae8d1cSMike Kravetz v_offset = 0; 3734aae8d1cSMike Kravetz 3744aae8d1cSMike Kravetz if (!end) 3754aae8d1cSMike Kravetz v_end = vma->vm_end; 3764aae8d1cSMike Kravetz else { 3774aae8d1cSMike Kravetz v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) 3784aae8d1cSMike Kravetz + vma->vm_start; 3794aae8d1cSMike Kravetz if (v_end > vma->vm_end) 3804aae8d1cSMike Kravetz v_end = vma->vm_end; 3814aae8d1cSMike Kravetz } 3824aae8d1cSMike Kravetz 3834aae8d1cSMike Kravetz unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, 3844aae8d1cSMike Kravetz NULL); 3854aae8d1cSMike Kravetz } 3864aae8d1cSMike Kravetz } 387b5cec28dSMike Kravetz 388b5cec28dSMike Kravetz /* 389b5cec28dSMike Kravetz * remove_inode_hugepages handles two distinct cases: truncation and hole 390b5cec28dSMike Kravetz * punch. There are subtle differences in operation for each case. 3914aae8d1cSMike Kravetz * 392b5cec28dSMike Kravetz * truncation is indicated by end of range being LLONG_MAX 393b5cec28dSMike Kravetz * In this case, we first scan the range and release found pages. 394b5cec28dSMike Kravetz * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv 395e7c58097SMike Kravetz * maps and global counts. Page faults can not race with truncation 396e7c58097SMike Kravetz * in this routine. hugetlb_no_page() prevents page faults in the 397e7c58097SMike Kravetz * truncated range. It checks i_size before allocation, and again after 398e7c58097SMike Kravetz * with the page table lock for the page held. The same lock must be 399e7c58097SMike Kravetz * acquired to unmap a page. 400b5cec28dSMike Kravetz * hole punch is indicated if end is not LLONG_MAX 401b5cec28dSMike Kravetz * In the hole punch case we scan the range and release found pages. 402b5cec28dSMike Kravetz * Only when releasing a page is the associated region/reserv map 403b5cec28dSMike Kravetz * deleted. The region/reserv map for ranges without associated 404e7c58097SMike Kravetz * pages are not modified. Page faults can race with hole punch. 405e7c58097SMike Kravetz * This is indicated if we find a mapped page. 406b5cec28dSMike Kravetz * Note: If the passed end of range value is beyond the end of file, but 407b5cec28dSMike Kravetz * not LLONG_MAX this routine still performs a hole punch operation. 408b5cec28dSMike Kravetz */ 409b5cec28dSMike Kravetz static void remove_inode_hugepages(struct inode *inode, loff_t lstart, 410b5cec28dSMike Kravetz loff_t lend) 4111da177e4SLinus Torvalds { 412a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 413b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 414a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 415b5cec28dSMike Kravetz const pgoff_t end = lend >> huge_page_shift(h); 416b5cec28dSMike Kravetz struct vm_area_struct pseudo_vma; 4171da177e4SLinus Torvalds struct pagevec pvec; 418d72dc8a2SJan Kara pgoff_t next, index; 419a43a8c39SChen, Kenneth W int i, freed = 0; 420b5cec28dSMike Kravetz bool truncate_op = (lend == LLONG_MAX); 4211da177e4SLinus Torvalds 4222c4541e2SKirill A. Shutemov vma_init(&pseudo_vma, current->mm); 423b5cec28dSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 42486679820SMel Gorman pagevec_init(&pvec); 4251da177e4SLinus Torvalds next = start; 426b5cec28dSMike Kravetz while (next < end) { 427b5cec28dSMike Kravetz /* 4281817889eSMike Kravetz * When no more pages are found, we are done. 429b5cec28dSMike Kravetz */ 430397162ffSJan Kara if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1)) 4311da177e4SLinus Torvalds break; 4321da177e4SLinus Torvalds 4331da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); ++i) { 4341da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 435e7c58097SMike Kravetz u32 hash; 436b5cec28dSMike Kravetz 437d72dc8a2SJan Kara index = page->index; 438188b04a7SWei Yang hash = hugetlb_fault_mutex_hash(mapping, index); 439e7c58097SMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 440e7c58097SMike Kravetz 441b5cec28dSMike Kravetz /* 442e7c58097SMike Kravetz * If page is mapped, it was faulted in after being 443e7c58097SMike Kravetz * unmapped in caller. Unmap (again) now after taking 444e7c58097SMike Kravetz * the fault mutex. The mutex will prevent faults 445e7c58097SMike Kravetz * until we finish removing the page. 446e7c58097SMike Kravetz * 447e7c58097SMike Kravetz * This race can only happen in the hole punch case. 448e7c58097SMike Kravetz * Getting here in a truncate operation is a bug. 449b5cec28dSMike Kravetz */ 450e7c58097SMike Kravetz if (unlikely(page_mapped(page))) { 451e7c58097SMike Kravetz BUG_ON(truncate_op); 452e7c58097SMike Kravetz 453e7c58097SMike Kravetz i_mmap_lock_write(mapping); 454e7c58097SMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 455e7c58097SMike Kravetz index * pages_per_huge_page(h), 456e7c58097SMike Kravetz (index + 1) * pages_per_huge_page(h)); 457e7c58097SMike Kravetz i_mmap_unlock_write(mapping); 458e7c58097SMike Kravetz } 4594aae8d1cSMike Kravetz 4604aae8d1cSMike Kravetz lock_page(page); 4614aae8d1cSMike Kravetz /* 4624aae8d1cSMike Kravetz * We must free the huge page and remove from page 4634aae8d1cSMike Kravetz * cache (remove_huge_page) BEFORE removing the 4644aae8d1cSMike Kravetz * region/reserve map (hugetlb_unreserve_pages). In 4654aae8d1cSMike Kravetz * rare out of memory conditions, removal of the 46672e2936cSzhong jiang * region/reserve map could fail. Correspondingly, 46772e2936cSzhong jiang * the subpool and global reserve usage count can need 46872e2936cSzhong jiang * to be adjusted. 4694aae8d1cSMike Kravetz */ 47072e2936cSzhong jiang VM_BUG_ON(PagePrivate(page)); 471b5cec28dSMike Kravetz remove_huge_page(page); 472b5cec28dSMike Kravetz freed++; 473b5cec28dSMike Kravetz if (!truncate_op) { 4744aae8d1cSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, 475d72dc8a2SJan Kara index, index + 1, 1))) 47672e2936cSzhong jiang hugetlb_fix_reserve_counts(inode); 477b5cec28dSMike Kravetz } 478b5cec28dSMike Kravetz 4791da177e4SLinus Torvalds unlock_page(page); 480e7c58097SMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 4811da177e4SLinus Torvalds } 4821da177e4SLinus Torvalds huge_pagevec_release(&pvec); 4831817889eSMike Kravetz cond_resched(); 4841da177e4SLinus Torvalds } 485b5cec28dSMike Kravetz 486b5cec28dSMike Kravetz if (truncate_op) 487b5cec28dSMike Kravetz (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); 4881da177e4SLinus Torvalds } 4891da177e4SLinus Torvalds 4902bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 4911da177e4SLinus Torvalds { 4929119a41eSJoonsoo Kim struct resv_map *resv_map; 4939119a41eSJoonsoo Kim 494b5cec28dSMike Kravetz remove_inode_hugepages(inode, 0, LLONG_MAX); 495f27a5136SMike Kravetz 496f27a5136SMike Kravetz /* 497f27a5136SMike Kravetz * Get the resv_map from the address space embedded in the inode. 498f27a5136SMike Kravetz * This is the address space which points to any resv_map allocated 499f27a5136SMike Kravetz * at inode creation time. If this is a device special inode, 500f27a5136SMike Kravetz * i_mapping may not point to the original address space. 501f27a5136SMike Kravetz */ 502f27a5136SMike Kravetz resv_map = (struct resv_map *)(&inode->i_data)->private_data; 503f27a5136SMike Kravetz /* Only regular and link inodes have associated reserve maps */ 5049119a41eSJoonsoo Kim if (resv_map) 5059119a41eSJoonsoo Kim resv_map_release(&resv_map->refs); 506dbd5768fSJan Kara clear_inode(inode); 507149f4211SChristoph Hellwig } 508149f4211SChristoph Hellwig 5091da177e4SLinus Torvalds static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) 5101da177e4SLinus Torvalds { 511856fc295SHugh Dickins pgoff_t pgoff; 5121da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 513a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 5141da177e4SLinus Torvalds 515a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 516856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 5171da177e4SLinus Torvalds 5187aa91e10SKen Chen i_size_write(inode, offset); 51983cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 520f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 5211bfad99aSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); 522c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 523e7c58097SMike Kravetz remove_inode_hugepages(inode, offset, LLONG_MAX); 5241da177e4SLinus Torvalds return 0; 5251da177e4SLinus Torvalds } 5261da177e4SLinus Torvalds 52770c3547eSMike Kravetz static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) 52870c3547eSMike Kravetz { 52970c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 53070c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 53170c3547eSMike Kravetz loff_t hole_start, hole_end; 53270c3547eSMike Kravetz 53370c3547eSMike Kravetz /* 53470c3547eSMike Kravetz * For hole punch round up the beginning offset of the hole and 53570c3547eSMike Kravetz * round down the end. 53670c3547eSMike Kravetz */ 53770c3547eSMike Kravetz hole_start = round_up(offset, hpage_size); 53870c3547eSMike Kravetz hole_end = round_down(offset + len, hpage_size); 53970c3547eSMike Kravetz 54070c3547eSMike Kravetz if (hole_end > hole_start) { 54170c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 542ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 54370c3547eSMike Kravetz 5445955102cSAl Viro inode_lock(inode); 545ff62a342SMarc-André Lureau 546ff62a342SMarc-André Lureau /* protected by i_mutex */ 547ab3948f5SJoel Fernandes (Google) if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { 548ff62a342SMarc-André Lureau inode_unlock(inode); 549ff62a342SMarc-André Lureau return -EPERM; 550ff62a342SMarc-André Lureau } 551ff62a342SMarc-André Lureau 55270c3547eSMike Kravetz i_mmap_lock_write(mapping); 553f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 55470c3547eSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 55570c3547eSMike Kravetz hole_start >> PAGE_SHIFT, 55670c3547eSMike Kravetz hole_end >> PAGE_SHIFT); 557c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 558e7c58097SMike Kravetz remove_inode_hugepages(inode, hole_start, hole_end); 5595955102cSAl Viro inode_unlock(inode); 56070c3547eSMike Kravetz } 56170c3547eSMike Kravetz 56270c3547eSMike Kravetz return 0; 56370c3547eSMike Kravetz } 56470c3547eSMike Kravetz 56570c3547eSMike Kravetz static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, 56670c3547eSMike Kravetz loff_t len) 56770c3547eSMike Kravetz { 56870c3547eSMike Kravetz struct inode *inode = file_inode(file); 569ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 57070c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 57170c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 57270c3547eSMike Kravetz struct vm_area_struct pseudo_vma; 57370c3547eSMike Kravetz struct mm_struct *mm = current->mm; 57470c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 57570c3547eSMike Kravetz unsigned long hpage_shift = huge_page_shift(h); 57670c3547eSMike Kravetz pgoff_t start, index, end; 57770c3547eSMike Kravetz int error; 57870c3547eSMike Kravetz u32 hash; 57970c3547eSMike Kravetz 58070c3547eSMike Kravetz if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 58170c3547eSMike Kravetz return -EOPNOTSUPP; 58270c3547eSMike Kravetz 58370c3547eSMike Kravetz if (mode & FALLOC_FL_PUNCH_HOLE) 58470c3547eSMike Kravetz return hugetlbfs_punch_hole(inode, offset, len); 58570c3547eSMike Kravetz 58670c3547eSMike Kravetz /* 58770c3547eSMike Kravetz * Default preallocate case. 58870c3547eSMike Kravetz * For this range, start is rounded down and end is rounded up 58970c3547eSMike Kravetz * as well as being converted to page offsets. 59070c3547eSMike Kravetz */ 59170c3547eSMike Kravetz start = offset >> hpage_shift; 59270c3547eSMike Kravetz end = (offset + len + hpage_size - 1) >> hpage_shift; 59370c3547eSMike Kravetz 5945955102cSAl Viro inode_lock(inode); 59570c3547eSMike Kravetz 59670c3547eSMike Kravetz /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ 59770c3547eSMike Kravetz error = inode_newsize_ok(inode, offset + len); 59870c3547eSMike Kravetz if (error) 59970c3547eSMike Kravetz goto out; 60070c3547eSMike Kravetz 601ff62a342SMarc-André Lureau if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { 602ff62a342SMarc-André Lureau error = -EPERM; 603ff62a342SMarc-André Lureau goto out; 604ff62a342SMarc-André Lureau } 605ff62a342SMarc-André Lureau 60670c3547eSMike Kravetz /* 60770c3547eSMike Kravetz * Initialize a pseudo vma as this is required by the huge page 60870c3547eSMike Kravetz * allocation routines. If NUMA is configured, use page index 60970c3547eSMike Kravetz * as input to create an allocation policy. 61070c3547eSMike Kravetz */ 6112c4541e2SKirill A. Shutemov vma_init(&pseudo_vma, mm); 61270c3547eSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 61370c3547eSMike Kravetz pseudo_vma.vm_file = file; 61470c3547eSMike Kravetz 61570c3547eSMike Kravetz for (index = start; index < end; index++) { 61670c3547eSMike Kravetz /* 61770c3547eSMike Kravetz * This is supposed to be the vaddr where the page is being 61870c3547eSMike Kravetz * faulted in, but we have no vaddr here. 61970c3547eSMike Kravetz */ 62070c3547eSMike Kravetz struct page *page; 62170c3547eSMike Kravetz unsigned long addr; 62270c3547eSMike Kravetz int avoid_reserve = 0; 62370c3547eSMike Kravetz 62470c3547eSMike Kravetz cond_resched(); 62570c3547eSMike Kravetz 62670c3547eSMike Kravetz /* 62770c3547eSMike Kravetz * fallocate(2) manpage permits EINTR; we may have been 62870c3547eSMike Kravetz * interrupted because we are using up too much memory. 62970c3547eSMike Kravetz */ 63070c3547eSMike Kravetz if (signal_pending(current)) { 63170c3547eSMike Kravetz error = -EINTR; 63270c3547eSMike Kravetz break; 63370c3547eSMike Kravetz } 63470c3547eSMike Kravetz 63570c3547eSMike Kravetz /* Set numa allocation policy based on index */ 63670c3547eSMike Kravetz hugetlb_set_vma_policy(&pseudo_vma, inode, index); 63770c3547eSMike Kravetz 63870c3547eSMike Kravetz /* addr is the offset within the file (zero based) */ 63970c3547eSMike Kravetz addr = index * hpage_size; 64070c3547eSMike Kravetz 641e7c58097SMike Kravetz /* mutex taken here, fault path and hole punch */ 642188b04a7SWei Yang hash = hugetlb_fault_mutex_hash(mapping, index); 64370c3547eSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 64470c3547eSMike Kravetz 64570c3547eSMike Kravetz /* See if already present in mapping to avoid alloc/free */ 64670c3547eSMike Kravetz page = find_get_page(mapping, index); 64770c3547eSMike Kravetz if (page) { 64870c3547eSMike Kravetz put_page(page); 64970c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 65070c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 65170c3547eSMike Kravetz continue; 65270c3547eSMike Kravetz } 65370c3547eSMike Kravetz 65470c3547eSMike Kravetz /* Allocate page and add to page cache */ 65570c3547eSMike Kravetz page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); 65670c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 65770c3547eSMike Kravetz if (IS_ERR(page)) { 65870c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 65970c3547eSMike Kravetz error = PTR_ERR(page); 66070c3547eSMike Kravetz goto out; 66170c3547eSMike Kravetz } 66270c3547eSMike Kravetz clear_huge_page(page, addr, pages_per_huge_page(h)); 66370c3547eSMike Kravetz __SetPageUptodate(page); 66470c3547eSMike Kravetz error = huge_add_to_page_cache(page, mapping, index); 66570c3547eSMike Kravetz if (unlikely(error)) { 66670c3547eSMike Kravetz put_page(page); 66770c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 66870c3547eSMike Kravetz goto out; 66970c3547eSMike Kravetz } 67070c3547eSMike Kravetz 67170c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 67270c3547eSMike Kravetz 67370c3547eSMike Kravetz /* 67470c3547eSMike Kravetz * unlock_page because locked by add_to_page_cache() 67572639e6dSNadav Amit * page_put due to reference from alloc_huge_page() 67670c3547eSMike Kravetz */ 67770c3547eSMike Kravetz unlock_page(page); 67872639e6dSNadav Amit put_page(page); 67970c3547eSMike Kravetz } 68070c3547eSMike Kravetz 68170c3547eSMike Kravetz if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) 68270c3547eSMike Kravetz i_size_write(inode, offset + len); 683078cd827SDeepa Dinamani inode->i_ctime = current_time(inode); 68470c3547eSMike Kravetz out: 6855955102cSAl Viro inode_unlock(inode); 68670c3547eSMike Kravetz return error; 68770c3547eSMike Kravetz } 68870c3547eSMike Kravetz 6891da177e4SLinus Torvalds static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) 6901da177e4SLinus Torvalds { 6912b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 692a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 6931da177e4SLinus Torvalds int error; 6941da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 695ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 6961da177e4SLinus Torvalds 6971da177e4SLinus Torvalds BUG_ON(!inode); 6981da177e4SLinus Torvalds 69931051c85SJan Kara error = setattr_prepare(dentry, attr); 7001da177e4SLinus Torvalds if (error) 7011025774cSChristoph Hellwig return error; 7021da177e4SLinus Torvalds 7031da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 704ff62a342SMarc-André Lureau loff_t oldsize = inode->i_size; 705ff62a342SMarc-André Lureau loff_t newsize = attr->ia_size; 706ff62a342SMarc-André Lureau 707ff62a342SMarc-André Lureau if (newsize & ~huge_page_mask(h)) 7081025774cSChristoph Hellwig return -EINVAL; 709ff62a342SMarc-André Lureau /* protected by i_mutex */ 710ff62a342SMarc-André Lureau if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || 711ff62a342SMarc-André Lureau (newsize > oldsize && (info->seals & F_SEAL_GROW))) 712ff62a342SMarc-André Lureau return -EPERM; 713ff62a342SMarc-André Lureau error = hugetlb_vmtruncate(inode, newsize); 7141da177e4SLinus Torvalds if (error) 7151da177e4SLinus Torvalds return error; 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 7181025774cSChristoph Hellwig setattr_copy(inode, attr); 7191025774cSChristoph Hellwig mark_inode_dirty(inode); 7201025774cSChristoph Hellwig return 0; 7211025774cSChristoph Hellwig } 7221025774cSChristoph Hellwig 7237d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 72432021982SDavid Howells struct hugetlbfs_fs_context *ctx) 7251da177e4SLinus Torvalds { 7261da177e4SLinus Torvalds struct inode *inode; 7271da177e4SLinus Torvalds 7281da177e4SLinus Torvalds inode = new_inode(sb); 7291da177e4SLinus Torvalds if (inode) { 73085fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 73132021982SDavid Howells inode->i_mode = S_IFDIR | ctx->mode; 73232021982SDavid Howells inode->i_uid = ctx->uid; 73332021982SDavid Howells inode->i_gid = ctx->gid; 734078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 7357d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 7367d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 7377d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 7387d54fa64SAl Viro inc_nlink(inode); 73965ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 7407d54fa64SAl Viro } 7417d54fa64SAl Viro return inode; 7427d54fa64SAl Viro } 7437d54fa64SAl Viro 744b610ded7SMichal Hocko /* 745c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 746b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 74788f306b6SKirill A. Shutemov * annotation because huge_pmd_share() does an allocation under hugetlb's 748c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 749b610ded7SMichal Hocko */ 750c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 751b610ded7SMichal Hocko 7527d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 7537d54fa64SAl Viro struct inode *dir, 75418df2252SAl Viro umode_t mode, dev_t dev) 7557d54fa64SAl Viro { 7567d54fa64SAl Viro struct inode *inode; 75758b6e5e8SMike Kravetz struct resv_map *resv_map = NULL; 7589119a41eSJoonsoo Kim 75958b6e5e8SMike Kravetz /* 76058b6e5e8SMike Kravetz * Reserve maps are only needed for inodes that can have associated 76158b6e5e8SMike Kravetz * page allocations. 76258b6e5e8SMike Kravetz */ 76358b6e5e8SMike Kravetz if (S_ISREG(mode) || S_ISLNK(mode)) { 7649119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 7659119a41eSJoonsoo Kim if (!resv_map) 7669119a41eSJoonsoo Kim return NULL; 76758b6e5e8SMike Kravetz } 7687d54fa64SAl Viro 7697d54fa64SAl Viro inode = new_inode(sb); 7707d54fa64SAl Viro if (inode) { 771ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 772ff62a342SMarc-André Lureau 7737d54fa64SAl Viro inode->i_ino = get_next_ino(); 7747d54fa64SAl Viro inode_init_owner(inode, dir, mode); 775c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 776c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 7771da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 778078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 7799119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 780ff62a342SMarc-André Lureau info->seals = F_SEAL_SEAL; 7811da177e4SLinus Torvalds switch (mode & S_IFMT) { 7821da177e4SLinus Torvalds default: 7831da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 7841da177e4SLinus Torvalds break; 7851da177e4SLinus Torvalds case S_IFREG: 7861da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 7871da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 7881da177e4SLinus Torvalds break; 7891da177e4SLinus Torvalds case S_IFDIR: 7901da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 7911da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 7921da177e4SLinus Torvalds 7931da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 794d8c76e6fSDave Hansen inc_nlink(inode); 7951da177e4SLinus Torvalds break; 7961da177e4SLinus Torvalds case S_IFLNK: 7971da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 79821fc61c7SAl Viro inode_nohighmem(inode); 7991da177e4SLinus Torvalds break; 8001da177e4SLinus Torvalds } 801e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 80258b6e5e8SMike Kravetz } else { 80358b6e5e8SMike Kravetz if (resv_map) 8049119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 80558b6e5e8SMike Kravetz } 8069119a41eSJoonsoo Kim 8071da177e4SLinus Torvalds return inode; 8081da177e4SLinus Torvalds } 8091da177e4SLinus Torvalds 8101da177e4SLinus Torvalds /* 8111da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 8121da177e4SLinus Torvalds */ 8131ab5b82fSPiotr Sarna static int do_hugetlbfs_mknod(struct inode *dir, 8141ab5b82fSPiotr Sarna struct dentry *dentry, 8151ab5b82fSPiotr Sarna umode_t mode, 8161ab5b82fSPiotr Sarna dev_t dev, 8171ab5b82fSPiotr Sarna bool tmpfile) 8181da177e4SLinus Torvalds { 8191da177e4SLinus Torvalds struct inode *inode; 8201da177e4SLinus Torvalds int error = -ENOSPC; 8211da177e4SLinus Torvalds 8227d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 8231da177e4SLinus Torvalds if (inode) { 824078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 8251ab5b82fSPiotr Sarna if (tmpfile) { 8261ab5b82fSPiotr Sarna d_tmpfile(dentry, inode); 8271ab5b82fSPiotr Sarna } else { 8281da177e4SLinus Torvalds d_instantiate(dentry, inode); 8291da177e4SLinus Torvalds dget(dentry);/* Extra count - pin the dentry in core */ 8301ab5b82fSPiotr Sarna } 8311da177e4SLinus Torvalds error = 0; 8321da177e4SLinus Torvalds } 8331da177e4SLinus Torvalds return error; 8341da177e4SLinus Torvalds } 8351da177e4SLinus Torvalds 8361ab5b82fSPiotr Sarna static int hugetlbfs_mknod(struct inode *dir, 8371ab5b82fSPiotr Sarna struct dentry *dentry, umode_t mode, dev_t dev) 8381ab5b82fSPiotr Sarna { 8391ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode, dev, false); 8401ab5b82fSPiotr Sarna } 8411ab5b82fSPiotr Sarna 84218bb1db3SAl Viro static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 8431da177e4SLinus Torvalds { 8441da177e4SLinus Torvalds int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); 8451da177e4SLinus Torvalds if (!retval) 846d8c76e6fSDave Hansen inc_nlink(dir); 8471da177e4SLinus Torvalds return retval; 8481da177e4SLinus Torvalds } 8491da177e4SLinus Torvalds 850ebfc3b49SAl Viro static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) 8511da177e4SLinus Torvalds { 8521da177e4SLinus Torvalds return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); 8531da177e4SLinus Torvalds } 8541da177e4SLinus Torvalds 8551ab5b82fSPiotr Sarna static int hugetlbfs_tmpfile(struct inode *dir, 8561ab5b82fSPiotr Sarna struct dentry *dentry, umode_t mode) 8571ab5b82fSPiotr Sarna { 8581ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0, true); 8591ab5b82fSPiotr Sarna } 8601ab5b82fSPiotr Sarna 8611da177e4SLinus Torvalds static int hugetlbfs_symlink(struct inode *dir, 8621da177e4SLinus Torvalds struct dentry *dentry, const char *symname) 8631da177e4SLinus Torvalds { 8641da177e4SLinus Torvalds struct inode *inode; 8651da177e4SLinus Torvalds int error = -ENOSPC; 8661da177e4SLinus Torvalds 8677d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 8681da177e4SLinus Torvalds if (inode) { 8691da177e4SLinus Torvalds int l = strlen(symname)+1; 8701da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 8711da177e4SLinus Torvalds if (!error) { 8721da177e4SLinus Torvalds d_instantiate(dentry, inode); 8731da177e4SLinus Torvalds dget(dentry); 8741da177e4SLinus Torvalds } else 8751da177e4SLinus Torvalds iput(inode); 8761da177e4SLinus Torvalds } 877078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 8781da177e4SLinus Torvalds 8791da177e4SLinus Torvalds return error; 8801da177e4SLinus Torvalds } 8811da177e4SLinus Torvalds 8821da177e4SLinus Torvalds /* 8836649a386SKen Chen * mark the head page dirty 8841da177e4SLinus Torvalds */ 8851da177e4SLinus Torvalds static int hugetlbfs_set_page_dirty(struct page *page) 8861da177e4SLinus Torvalds { 887d85f3385SChristoph Lameter struct page *head = compound_head(page); 8886649a386SKen Chen 8896649a386SKen Chen SetPageDirty(head); 8901da177e4SLinus Torvalds return 0; 8911da177e4SLinus Torvalds } 8921da177e4SLinus Torvalds 893290408d4SNaoya Horiguchi static int hugetlbfs_migrate_page(struct address_space *mapping, 894b969c4abSMel Gorman struct page *newpage, struct page *page, 895a6bc32b8SMel Gorman enum migrate_mode mode) 896290408d4SNaoya Horiguchi { 897290408d4SNaoya Horiguchi int rc; 898290408d4SNaoya Horiguchi 899290408d4SNaoya Horiguchi rc = migrate_huge_page_move_mapping(mapping, newpage, page); 90078bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 901290408d4SNaoya Horiguchi return rc; 902cb6acd01SMike Kravetz 903cb6acd01SMike Kravetz /* 904cb6acd01SMike Kravetz * page_private is subpool pointer in hugetlb pages. Transfer to 905cb6acd01SMike Kravetz * new page. PagePrivate is not associated with page_private for 906cb6acd01SMike Kravetz * hugetlb pages and can not be set here as only page_huge_active 907cb6acd01SMike Kravetz * pages can be migrated. 908cb6acd01SMike Kravetz */ 909cb6acd01SMike Kravetz if (page_private(page)) { 910cb6acd01SMike Kravetz set_page_private(newpage, page_private(page)); 911cb6acd01SMike Kravetz set_page_private(page, 0); 912cb6acd01SMike Kravetz } 913cb6acd01SMike Kravetz 9142916ecc0SJérôme Glisse if (mode != MIGRATE_SYNC_NO_COPY) 915290408d4SNaoya Horiguchi migrate_page_copy(newpage, page); 9162916ecc0SJérôme Glisse else 9172916ecc0SJérôme Glisse migrate_page_states(newpage, page); 918290408d4SNaoya Horiguchi 91978bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 920290408d4SNaoya Horiguchi } 921290408d4SNaoya Horiguchi 92278bb9203SNaoya Horiguchi static int hugetlbfs_error_remove_page(struct address_space *mapping, 92378bb9203SNaoya Horiguchi struct page *page) 92478bb9203SNaoya Horiguchi { 92578bb9203SNaoya Horiguchi struct inode *inode = mapping->host; 926ab615a5bSMike Kravetz pgoff_t index = page->index; 92778bb9203SNaoya Horiguchi 92878bb9203SNaoya Horiguchi remove_huge_page(page); 929ab615a5bSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) 93078bb9203SNaoya Horiguchi hugetlb_fix_reserve_counts(inode); 931ab615a5bSMike Kravetz 93278bb9203SNaoya Horiguchi return 0; 93378bb9203SNaoya Horiguchi } 93478bb9203SNaoya Horiguchi 9354a25220dSDavid Howells /* 9364a25220dSDavid Howells * Display the mount options in /proc/mounts. 9374a25220dSDavid Howells */ 9384a25220dSDavid Howells static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) 9394a25220dSDavid Howells { 9404a25220dSDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); 9414a25220dSDavid Howells struct hugepage_subpool *spool = sbinfo->spool; 9424a25220dSDavid Howells unsigned long hpage_size = huge_page_size(sbinfo->hstate); 9434a25220dSDavid Howells unsigned hpage_shift = huge_page_shift(sbinfo->hstate); 9444a25220dSDavid Howells char mod; 9454a25220dSDavid Howells 9464a25220dSDavid Howells if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) 9474a25220dSDavid Howells seq_printf(m, ",uid=%u", 9484a25220dSDavid Howells from_kuid_munged(&init_user_ns, sbinfo->uid)); 9494a25220dSDavid Howells if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) 9504a25220dSDavid Howells seq_printf(m, ",gid=%u", 9514a25220dSDavid Howells from_kgid_munged(&init_user_ns, sbinfo->gid)); 9524a25220dSDavid Howells if (sbinfo->mode != 0755) 9534a25220dSDavid Howells seq_printf(m, ",mode=%o", sbinfo->mode); 9544a25220dSDavid Howells if (sbinfo->max_inodes != -1) 9554a25220dSDavid Howells seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); 9564a25220dSDavid Howells 9574a25220dSDavid Howells hpage_size /= 1024; 9584a25220dSDavid Howells mod = 'K'; 9594a25220dSDavid Howells if (hpage_size >= 1024) { 9604a25220dSDavid Howells hpage_size /= 1024; 9614a25220dSDavid Howells mod = 'M'; 9624a25220dSDavid Howells } 9634a25220dSDavid Howells seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); 9644a25220dSDavid Howells if (spool) { 9654a25220dSDavid Howells if (spool->max_hpages != -1) 9664a25220dSDavid Howells seq_printf(m, ",size=%llu", 9674a25220dSDavid Howells (unsigned long long)spool->max_hpages << hpage_shift); 9684a25220dSDavid Howells if (spool->min_hpages != -1) 9694a25220dSDavid Howells seq_printf(m, ",min_size=%llu", 9704a25220dSDavid Howells (unsigned long long)spool->min_hpages << hpage_shift); 9714a25220dSDavid Howells } 9724a25220dSDavid Howells return 0; 9734a25220dSDavid Howells } 9744a25220dSDavid Howells 975726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 9761da177e4SLinus Torvalds { 977726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 9782b0143b5SDavid Howells struct hstate *h = hstate_inode(d_inode(dentry)); 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 981a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 9821da177e4SLinus Torvalds if (sbinfo) { 9831da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 98474a8a65cSDavid Gibson /* If no limits set, just report 0 for max/free/used 98574a8a65cSDavid Gibson * blocks, like simple_statfs() */ 98690481622SDavid Gibson if (sbinfo->spool) { 98790481622SDavid Gibson long free_pages; 98890481622SDavid Gibson 98990481622SDavid Gibson spin_lock(&sbinfo->spool->lock); 99090481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 99190481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 99290481622SDavid Gibson - sbinfo->spool->used_hpages; 99390481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 99490481622SDavid Gibson spin_unlock(&sbinfo->spool->lock); 9951da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 9961da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 99774a8a65cSDavid Gibson } 9981da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 9991da177e4SLinus Torvalds } 10001da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 10011da177e4SLinus Torvalds return 0; 10021da177e4SLinus Torvalds } 10031da177e4SLinus Torvalds 10041da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 10051da177e4SLinus Torvalds { 10061da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 10071da177e4SLinus Torvalds 10081da177e4SLinus Torvalds if (sbi) { 10091da177e4SLinus Torvalds sb->s_fs_info = NULL; 101090481622SDavid Gibson 101190481622SDavid Gibson if (sbi->spool) 101290481622SDavid Gibson hugepage_put_subpool(sbi->spool); 101390481622SDavid Gibson 10141da177e4SLinus Torvalds kfree(sbi); 10151da177e4SLinus Torvalds } 10161da177e4SLinus Torvalds } 10171da177e4SLinus Torvalds 101896527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 101996527980SChristoph Hellwig { 102096527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 102196527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 102296527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 102396527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 102496527980SChristoph Hellwig return 0; 102596527980SChristoph Hellwig } 102696527980SChristoph Hellwig sbinfo->free_inodes--; 102796527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 102896527980SChristoph Hellwig } 102996527980SChristoph Hellwig 103096527980SChristoph Hellwig return 1; 103196527980SChristoph Hellwig } 103296527980SChristoph Hellwig 103396527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 103496527980SChristoph Hellwig { 103596527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 103696527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 103796527980SChristoph Hellwig sbinfo->free_inodes++; 103896527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 103996527980SChristoph Hellwig } 104096527980SChristoph Hellwig } 104196527980SChristoph Hellwig 104296527980SChristoph Hellwig 1043e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 10441da177e4SLinus Torvalds 10451da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 10461da177e4SLinus Torvalds { 104796527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 10481da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 10491da177e4SLinus Torvalds 105096527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 10511da177e4SLinus Torvalds return NULL; 1052e94b1766SChristoph Lameter p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); 105396527980SChristoph Hellwig if (unlikely(!p)) { 105496527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 105596527980SChristoph Hellwig return NULL; 10561da177e4SLinus Torvalds } 10574742a35dSMike Kravetz 10584742a35dSMike Kravetz /* 10594742a35dSMike Kravetz * Any time after allocation, hugetlbfs_destroy_inode can be called 10604742a35dSMike Kravetz * for the inode. mpol_free_shared_policy is unconditionally called 10614742a35dSMike Kravetz * as part of hugetlbfs_destroy_inode. So, initialize policy here 10624742a35dSMike Kravetz * in case of a quick call to destroy. 10634742a35dSMike Kravetz * 10644742a35dSMike Kravetz * Note that the policy is initialized even if we are creating a 10654742a35dSMike Kravetz * private inode. This simplifies hugetlbfs_destroy_inode. 10664742a35dSMike Kravetz */ 10674742a35dSMike Kravetz mpol_shared_policy_init(&p->policy, NULL); 10684742a35dSMike Kravetz 106996527980SChristoph Hellwig return &p->vfs_inode; 10701da177e4SLinus Torvalds } 10711da177e4SLinus Torvalds 1072b62de322SAl Viro static void hugetlbfs_free_inode(struct inode *inode) 1073fa0d7e3dSNick Piggin { 1074fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 1075fa0d7e3dSNick Piggin } 1076fa0d7e3dSNick Piggin 10771da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 10781da177e4SLinus Torvalds { 107996527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 10801da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 10811da177e4SLinus Torvalds } 10821da177e4SLinus Torvalds 1083f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 1084800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 1085800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 10861da177e4SLinus Torvalds .set_page_dirty = hugetlbfs_set_page_dirty, 1087290408d4SNaoya Horiguchi .migratepage = hugetlbfs_migrate_page, 108878bb9203SNaoya Horiguchi .error_remove_page = hugetlbfs_error_remove_page, 10891da177e4SLinus Torvalds }; 10901da177e4SLinus Torvalds 109196527980SChristoph Hellwig 109251cc5068SAlexey Dobriyan static void init_once(void *foo) 109396527980SChristoph Hellwig { 109496527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 109596527980SChristoph Hellwig 109696527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 109796527980SChristoph Hellwig } 109896527980SChristoph Hellwig 10994b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 110034d0640eSAl Viro .read_iter = hugetlbfs_read_iter, 11011da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 11021b061d92SChristoph Hellwig .fsync = noop_fsync, 11031da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 11046038f373SArnd Bergmann .llseek = default_llseek, 110570c3547eSMike Kravetz .fallocate = hugetlbfs_fallocate, 11061da177e4SLinus Torvalds }; 11071da177e4SLinus Torvalds 110892e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 11091da177e4SLinus Torvalds .create = hugetlbfs_create, 11101da177e4SLinus Torvalds .lookup = simple_lookup, 11111da177e4SLinus Torvalds .link = simple_link, 11121da177e4SLinus Torvalds .unlink = simple_unlink, 11131da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 11141da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 11151da177e4SLinus Torvalds .rmdir = simple_rmdir, 11161da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 11171da177e4SLinus Torvalds .rename = simple_rename, 11181da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 11191ab5b82fSPiotr Sarna .tmpfile = hugetlbfs_tmpfile, 11201da177e4SLinus Torvalds }; 11211da177e4SLinus Torvalds 112292e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 11231da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 11241da177e4SLinus Torvalds }; 11251da177e4SLinus Torvalds 1126ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 11271da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 1128b62de322SAl Viro .free_inode = hugetlbfs_free_inode, 11291da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 11302bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 11311da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 11321da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 11334a25220dSDavid Howells .show_options = hugetlbfs_show_options, 11341da177e4SLinus Torvalds }; 11351da177e4SLinus Torvalds 11367ca02d0aSMike Kravetz /* 11377ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 11387ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 11397ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 11407ca02d0aSMike Kravetz */ 11414a25220dSDavid Howells static long 11427ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 11434a25220dSDavid Howells enum hugetlbfs_size_type val_type) 11447ca02d0aSMike Kravetz { 11457ca02d0aSMike Kravetz if (val_type == NO_SIZE) 11467ca02d0aSMike Kravetz return -1; 11477ca02d0aSMike Kravetz 11487ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 11497ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 11507ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 11517ca02d0aSMike Kravetz do_div(size_opt, 100); 11527ca02d0aSMike Kravetz } 11537ca02d0aSMike Kravetz 11547ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 11557ca02d0aSMike Kravetz return size_opt; 11567ca02d0aSMike Kravetz } 11577ca02d0aSMike Kravetz 115832021982SDavid Howells /* 115932021982SDavid Howells * Parse one mount parameter. 116032021982SDavid Howells */ 116132021982SDavid Howells static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param) 11621da177e4SLinus Torvalds { 116332021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 116432021982SDavid Howells struct fs_parse_result result; 116532021982SDavid Howells char *rest; 116632021982SDavid Howells unsigned long ps; 116732021982SDavid Howells int opt; 11681da177e4SLinus Torvalds 1169d7167b14SAl Viro opt = fs_parse(fc, hugetlb_fs_parameters, param, &result); 117032021982SDavid Howells if (opt < 0) 117132021982SDavid Howells return opt; 117232021982SDavid Howells 117332021982SDavid Howells switch (opt) { 117432021982SDavid Howells case Opt_uid: 117532021982SDavid Howells ctx->uid = make_kuid(current_user_ns(), result.uint_32); 117632021982SDavid Howells if (!uid_valid(ctx->uid)) 117732021982SDavid Howells goto bad_val; 11781da177e4SLinus Torvalds return 0; 11791da177e4SLinus Torvalds 1180e73a75faSRandy Dunlap case Opt_gid: 118132021982SDavid Howells ctx->gid = make_kgid(current_user_ns(), result.uint_32); 118232021982SDavid Howells if (!gid_valid(ctx->gid)) 1183e73a75faSRandy Dunlap goto bad_val; 118432021982SDavid Howells return 0; 1185e73a75faSRandy Dunlap 1186e73a75faSRandy Dunlap case Opt_mode: 118732021982SDavid Howells ctx->mode = result.uint_32 & 01777U; 118832021982SDavid Howells return 0; 1189e73a75faSRandy Dunlap 119032021982SDavid Howells case Opt_size: 1191e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 119232021982SDavid Howells if (!isdigit(param->string[0])) 1193e73a75faSRandy Dunlap goto bad_val; 119432021982SDavid Howells ctx->max_size_opt = memparse(param->string, &rest); 119532021982SDavid Howells ctx->max_val_type = SIZE_STD; 1196a137e1ccSAndi Kleen if (*rest == '%') 119732021982SDavid Howells ctx->max_val_type = SIZE_PERCENT; 119832021982SDavid Howells return 0; 11991da177e4SLinus Torvalds 1200e73a75faSRandy Dunlap case Opt_nr_inodes: 1201e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 120232021982SDavid Howells if (!isdigit(param->string[0])) 1203e73a75faSRandy Dunlap goto bad_val; 120432021982SDavid Howells ctx->nr_inodes = memparse(param->string, &rest); 120532021982SDavid Howells return 0; 1206e73a75faSRandy Dunlap 120732021982SDavid Howells case Opt_pagesize: 120832021982SDavid Howells ps = memparse(param->string, &rest); 120932021982SDavid Howells ctx->hstate = size_to_hstate(ps); 121032021982SDavid Howells if (!ctx->hstate) { 121132021982SDavid Howells pr_err("Unsupported page size %lu MB\n", ps >> 20); 1212a137e1ccSAndi Kleen return -EINVAL; 1213a137e1ccSAndi Kleen } 121432021982SDavid Howells return 0; 1215a137e1ccSAndi Kleen 121632021982SDavid Howells case Opt_min_size: 12177ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 121832021982SDavid Howells if (!isdigit(param->string[0])) 12197ca02d0aSMike Kravetz goto bad_val; 122032021982SDavid Howells ctx->min_size_opt = memparse(param->string, &rest); 122132021982SDavid Howells ctx->min_val_type = SIZE_STD; 12227ca02d0aSMike Kravetz if (*rest == '%') 122332021982SDavid Howells ctx->min_val_type = SIZE_PERCENT; 122432021982SDavid Howells return 0; 12257ca02d0aSMike Kravetz 1226e73a75faSRandy Dunlap default: 1227b4c07bceSLee Schermerhorn return -EINVAL; 1228e73a75faSRandy Dunlap } 122932021982SDavid Howells 123032021982SDavid Howells bad_val: 123132021982SDavid Howells return invalf(fc, "hugetlbfs: Bad value '%s' for mount option '%s'\n", 123232021982SDavid Howells param->string, param->key); 12331da177e4SLinus Torvalds } 1234a137e1ccSAndi Kleen 12357ca02d0aSMike Kravetz /* 123632021982SDavid Howells * Validate the parsed options. 123732021982SDavid Howells */ 123832021982SDavid Howells static int hugetlbfs_validate(struct fs_context *fc) 123932021982SDavid Howells { 124032021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 124132021982SDavid Howells 124232021982SDavid Howells /* 12437ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 12447ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 12457ca02d0aSMike Kravetz */ 124632021982SDavid Howells ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 124732021982SDavid Howells ctx->max_size_opt, 124832021982SDavid Howells ctx->max_val_type); 124932021982SDavid Howells ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 125032021982SDavid Howells ctx->min_size_opt, 125132021982SDavid Howells ctx->min_val_type); 12527ca02d0aSMike Kravetz 12537ca02d0aSMike Kravetz /* 12547ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 12557ca02d0aSMike Kravetz */ 125632021982SDavid Howells if (ctx->max_val_type > NO_SIZE && 125732021982SDavid Howells ctx->min_hpages > ctx->max_hpages) { 125832021982SDavid Howells pr_err("Minimum size can not be greater than maximum size\n"); 12597ca02d0aSMike Kravetz return -EINVAL; 1260a137e1ccSAndi Kleen } 1261a137e1ccSAndi Kleen 12621da177e4SLinus Torvalds return 0; 12631da177e4SLinus Torvalds } 12641da177e4SLinus Torvalds 12651da177e4SLinus Torvalds static int 126632021982SDavid Howells hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc) 12671da177e4SLinus Torvalds { 126832021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 12691da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 12701da177e4SLinus Torvalds 12711da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 12721da177e4SLinus Torvalds if (!sbinfo) 12731da177e4SLinus Torvalds return -ENOMEM; 12741da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 12751da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 127632021982SDavid Howells sbinfo->hstate = ctx->hstate; 127732021982SDavid Howells sbinfo->max_inodes = ctx->nr_inodes; 127832021982SDavid Howells sbinfo->free_inodes = ctx->nr_inodes; 127990481622SDavid Gibson sbinfo->spool = NULL; 128032021982SDavid Howells sbinfo->uid = ctx->uid; 128132021982SDavid Howells sbinfo->gid = ctx->gid; 128232021982SDavid Howells sbinfo->mode = ctx->mode; 12834a25220dSDavid Howells 12847ca02d0aSMike Kravetz /* 12857ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 12867ca02d0aSMike Kravetz * specified. Any needed reservations (for minimim size) are taken 12877ca02d0aSMike Kravetz * taken when the subpool is created. 12887ca02d0aSMike Kravetz */ 128932021982SDavid Howells if (ctx->max_hpages != -1 || ctx->min_hpages != -1) { 129032021982SDavid Howells sbinfo->spool = hugepage_new_subpool(ctx->hstate, 129132021982SDavid Howells ctx->max_hpages, 129232021982SDavid Howells ctx->min_hpages); 129390481622SDavid Gibson if (!sbinfo->spool) 129490481622SDavid Gibson goto out_free; 129590481622SDavid Gibson } 12961da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 129732021982SDavid Howells sb->s_blocksize = huge_page_size(ctx->hstate); 129832021982SDavid Howells sb->s_blocksize_bits = huge_page_shift(ctx->hstate); 12991da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 13001da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 13011da177e4SLinus Torvalds sb->s_time_gran = 1; 130232021982SDavid Howells sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx)); 130348fde701SAl Viro if (!sb->s_root) 13041da177e4SLinus Torvalds goto out_free; 13051da177e4SLinus Torvalds return 0; 13061da177e4SLinus Torvalds out_free: 130790481622SDavid Gibson kfree(sbinfo->spool); 13081da177e4SLinus Torvalds kfree(sbinfo); 13091da177e4SLinus Torvalds return -ENOMEM; 13101da177e4SLinus Torvalds } 13111da177e4SLinus Torvalds 131232021982SDavid Howells static int hugetlbfs_get_tree(struct fs_context *fc) 13131da177e4SLinus Torvalds { 131432021982SDavid Howells int err = hugetlbfs_validate(fc); 131532021982SDavid Howells if (err) 131632021982SDavid Howells return err; 13172ac295d4SAl Viro return get_tree_nodev(fc, hugetlbfs_fill_super); 131832021982SDavid Howells } 131932021982SDavid Howells 132032021982SDavid Howells static void hugetlbfs_fs_context_free(struct fs_context *fc) 132132021982SDavid Howells { 132232021982SDavid Howells kfree(fc->fs_private); 132332021982SDavid Howells } 132432021982SDavid Howells 132532021982SDavid Howells static const struct fs_context_operations hugetlbfs_fs_context_ops = { 132632021982SDavid Howells .free = hugetlbfs_fs_context_free, 132732021982SDavid Howells .parse_param = hugetlbfs_parse_param, 132832021982SDavid Howells .get_tree = hugetlbfs_get_tree, 132932021982SDavid Howells }; 133032021982SDavid Howells 133132021982SDavid Howells static int hugetlbfs_init_fs_context(struct fs_context *fc) 133232021982SDavid Howells { 133332021982SDavid Howells struct hugetlbfs_fs_context *ctx; 133432021982SDavid Howells 133532021982SDavid Howells ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL); 133632021982SDavid Howells if (!ctx) 133732021982SDavid Howells return -ENOMEM; 133832021982SDavid Howells 133932021982SDavid Howells ctx->max_hpages = -1; /* No limit on size by default */ 134032021982SDavid Howells ctx->nr_inodes = -1; /* No limit on number of inodes by default */ 134132021982SDavid Howells ctx->uid = current_fsuid(); 134232021982SDavid Howells ctx->gid = current_fsgid(); 134332021982SDavid Howells ctx->mode = 0755; 134432021982SDavid Howells ctx->hstate = &default_hstate; 134532021982SDavid Howells ctx->min_hpages = -1; /* No default minimum size */ 134632021982SDavid Howells ctx->max_val_type = NO_SIZE; 134732021982SDavid Howells ctx->min_val_type = NO_SIZE; 134832021982SDavid Howells fc->fs_private = ctx; 134932021982SDavid Howells fc->ops = &hugetlbfs_fs_context_ops; 135032021982SDavid Howells return 0; 13511da177e4SLinus Torvalds } 13521da177e4SLinus Torvalds 13531da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 13541da177e4SLinus Torvalds .name = "hugetlbfs", 135532021982SDavid Howells .init_fs_context = hugetlbfs_init_fs_context, 1356d7167b14SAl Viro .parameters = hugetlb_fs_parameters, 13571da177e4SLinus Torvalds .kill_sb = kill_litter_super, 13581da177e4SLinus Torvalds }; 13591da177e4SLinus Torvalds 136042d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 13611da177e4SLinus Torvalds 1362ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 13631da177e4SLinus Torvalds { 1364a0eb3a05SEric W. Biederman kgid_t shm_group; 1365a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 1366a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 13671da177e4SLinus Torvalds } 13681da177e4SLinus Torvalds 136942d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 137042d7395fSAndi Kleen { 1371af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 137242d7395fSAndi Kleen 137342d7395fSAndi Kleen if (!h) 137442d7395fSAndi Kleen return -1; 137542d7395fSAndi Kleen return h - hstates; 137642d7395fSAndi Kleen } 137742d7395fSAndi Kleen 1378af73e4d9SNaoya Horiguchi /* 1379af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 1380af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 1381af73e4d9SNaoya Horiguchi */ 1382af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 1383af73e4d9SNaoya Horiguchi vm_flags_t acctflag, struct user_struct **user, 138442d7395fSAndi Kleen int creat_flags, int page_size_log) 13851da177e4SLinus Torvalds { 13861da177e4SLinus Torvalds struct inode *inode; 1387e68375c8SAl Viro struct vfsmount *mnt; 138842d7395fSAndi Kleen int hstate_idx; 1389e68375c8SAl Viro struct file *file; 139042d7395fSAndi Kleen 139142d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 139242d7395fSAndi Kleen if (hstate_idx < 0) 139342d7395fSAndi Kleen return ERR_PTR(-ENODEV); 13941da177e4SLinus Torvalds 1395353d5c30SHugh Dickins *user = NULL; 1396e68375c8SAl Viro mnt = hugetlbfs_vfsmount[hstate_idx]; 1397e68375c8SAl Viro if (!mnt) 13985bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 13995bc98594SAkinobu Mita 1400ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 1401353d5c30SHugh Dickins *user = current_user(); 1402353d5c30SHugh Dickins if (user_shm_lock(size, *user)) { 140321a3c273SDavid Rientjes task_lock(current); 14049b857d26SAndrew Morton pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", 140521a3c273SDavid Rientjes current->comm, current->pid); 140621a3c273SDavid Rientjes task_unlock(current); 1407353d5c30SHugh Dickins } else { 1408353d5c30SHugh Dickins *user = NULL; 14091da177e4SLinus Torvalds return ERR_PTR(-EPERM); 14102584e517SRavikiran G Thirumalai } 1411353d5c30SHugh Dickins } 14121da177e4SLinus Torvalds 141339b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 1414e68375c8SAl Viro inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0); 14151da177e4SLinus Torvalds if (!inode) 1416e68375c8SAl Viro goto out; 1417e1832f29SStephen Smalley if (creat_flags == HUGETLB_SHMFS_INODE) 1418e1832f29SStephen Smalley inode->i_flags |= S_PRIVATE; 14191da177e4SLinus Torvalds 14201da177e4SLinus Torvalds inode->i_size = size; 14216d6b77f1SMiklos Szeredi clear_nlink(inode); 1422ce8d2cdfSDave Hansen 1423e68375c8SAl Viro if (hugetlb_reserve_pages(inode, 0, 1424e68375c8SAl Viro size >> huge_page_shift(hstate_inode(inode)), NULL, 1425e68375c8SAl Viro acctflag)) 1426e68375c8SAl Viro file = ERR_PTR(-ENOMEM); 1427e68375c8SAl Viro else 1428e68375c8SAl Viro file = alloc_file_pseudo(inode, mnt, name, O_RDWR, 1429ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 1430e68375c8SAl Viro if (!IS_ERR(file)) 14311da177e4SLinus Torvalds return file; 14321da177e4SLinus Torvalds 1433b45b5bd6SDavid Gibson iput(inode); 1434e68375c8SAl Viro out: 1435353d5c30SHugh Dickins if (*user) { 1436353d5c30SHugh Dickins user_shm_unlock(size, *user); 1437353d5c30SHugh Dickins *user = NULL; 1438353d5c30SHugh Dickins } 143939b65252SAnatol Pomozov return file; 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds 144232021982SDavid Howells static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h) 144332021982SDavid Howells { 144432021982SDavid Howells struct fs_context *fc; 144532021982SDavid Howells struct vfsmount *mnt; 144632021982SDavid Howells 144732021982SDavid Howells fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT); 144832021982SDavid Howells if (IS_ERR(fc)) { 144932021982SDavid Howells mnt = ERR_CAST(fc); 145032021982SDavid Howells } else { 145132021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 145232021982SDavid Howells ctx->hstate = h; 145332021982SDavid Howells mnt = fc_mount(fc); 145432021982SDavid Howells put_fs_context(fc); 145532021982SDavid Howells } 145632021982SDavid Howells if (IS_ERR(mnt)) 145732021982SDavid Howells pr_err("Cannot mount internal hugetlbfs for page size %uK", 145832021982SDavid Howells 1U << (h->order + PAGE_SHIFT - 10)); 145932021982SDavid Howells return mnt; 146032021982SDavid Howells } 146132021982SDavid Howells 14621da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 14631da177e4SLinus Torvalds { 146432021982SDavid Howells struct vfsmount *mnt; 146542d7395fSAndi Kleen struct hstate *h; 14661da177e4SLinus Torvalds int error; 146742d7395fSAndi Kleen int i; 14681da177e4SLinus Torvalds 1469457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 14709b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1471457c1b27SNishanth Aravamudan return -ENOTSUPP; 1472457c1b27SNishanth Aravamudan } 1473457c1b27SNishanth Aravamudan 1474d1d5e05fSHillf Danton error = -ENOMEM; 14751da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 14761da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 14775d097056SVladimir Davydov 0, SLAB_ACCOUNT, init_once); 14781da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 14798fc312b3SMike Kravetz goto out; 14801da177e4SLinus Torvalds 14811da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 14821da177e4SLinus Torvalds if (error) 14838fc312b3SMike Kravetz goto out_free; 14841da177e4SLinus Torvalds 14858fc312b3SMike Kravetz /* default hstate mount is required */ 14868fc312b3SMike Kravetz mnt = mount_one_hugetlbfs(&hstates[default_hstate_idx]); 14878fc312b3SMike Kravetz if (IS_ERR(mnt)) { 14888fc312b3SMike Kravetz error = PTR_ERR(mnt); 14898fc312b3SMike Kravetz goto out_unreg; 14908fc312b3SMike Kravetz } 14918fc312b3SMike Kravetz hugetlbfs_vfsmount[default_hstate_idx] = mnt; 14928fc312b3SMike Kravetz 14938fc312b3SMike Kravetz /* other hstates are optional */ 149442d7395fSAndi Kleen i = 0; 149542d7395fSAndi Kleen for_each_hstate(h) { 149615f0ec94SJan Stancek if (i == default_hstate_idx) { 149715f0ec94SJan Stancek i++; 14988fc312b3SMike Kravetz continue; 149915f0ec94SJan Stancek } 15008fc312b3SMike Kravetz 150132021982SDavid Howells mnt = mount_one_hugetlbfs(h); 15028fc312b3SMike Kravetz if (IS_ERR(mnt)) 15038fc312b3SMike Kravetz hugetlbfs_vfsmount[i] = NULL; 15048fc312b3SMike Kravetz else 150532021982SDavid Howells hugetlbfs_vfsmount[i] = mnt; 150642d7395fSAndi Kleen i++; 150742d7395fSAndi Kleen } 150832021982SDavid Howells 150942d7395fSAndi Kleen return 0; 15101da177e4SLinus Torvalds 15118fc312b3SMike Kravetz out_unreg: 15128fc312b3SMike Kravetz (void)unregister_filesystem(&hugetlbfs_fs_type); 15138fc312b3SMike Kravetz out_free: 15141da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 15158fc312b3SMike Kravetz out: 15161da177e4SLinus Torvalds return error; 15171da177e4SLinus Torvalds } 15183e89e1c5SPaul Gortmaker fs_initcall(init_hugetlbfs_fs) 1519