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> 4188590253SShijie Hu #include <linux/sched/mm.h> 421da177e4SLinus Torvalds 43ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops; 44f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops; 454b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations; 4692e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations; 4792e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations; 481da177e4SLinus Torvalds 4932021982SDavid Howells enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT }; 5032021982SDavid Howells 5132021982SDavid Howells struct hugetlbfs_fs_context { 524a25220dSDavid Howells struct hstate *hstate; 5332021982SDavid Howells unsigned long long max_size_opt; 5432021982SDavid Howells unsigned long long min_size_opt; 554a25220dSDavid Howells long max_hpages; 564a25220dSDavid Howells long nr_inodes; 574a25220dSDavid Howells long min_hpages; 5832021982SDavid Howells enum hugetlbfs_size_type max_val_type; 5932021982SDavid Howells enum hugetlbfs_size_type min_val_type; 60a0eb3a05SEric W. Biederman kuid_t uid; 61a0eb3a05SEric W. Biederman kgid_t gid; 62a1d776eeSDavid Gibson umode_t mode; 63a1d776eeSDavid Gibson }; 64a1d776eeSDavid Gibson 651da177e4SLinus Torvalds int sysctl_hugetlb_shm_group; 661da177e4SLinus Torvalds 6732021982SDavid Howells enum hugetlb_param { 6832021982SDavid Howells Opt_gid, 6932021982SDavid Howells Opt_min_size, 7032021982SDavid Howells Opt_mode, 7132021982SDavid Howells Opt_nr_inodes, 7232021982SDavid Howells Opt_pagesize, 7332021982SDavid Howells Opt_size, 7432021982SDavid Howells Opt_uid, 75e73a75faSRandy Dunlap }; 76e73a75faSRandy Dunlap 77d7167b14SAl Viro static const struct fs_parameter_spec hugetlb_fs_parameters[] = { 7832021982SDavid Howells fsparam_u32 ("gid", Opt_gid), 7932021982SDavid Howells fsparam_string("min_size", Opt_min_size), 8032021982SDavid Howells fsparam_u32 ("mode", Opt_mode), 8132021982SDavid Howells fsparam_string("nr_inodes", Opt_nr_inodes), 8232021982SDavid Howells fsparam_string("pagesize", Opt_pagesize), 8332021982SDavid Howells fsparam_string("size", Opt_size), 8432021982SDavid Howells fsparam_u32 ("uid", Opt_uid), 8532021982SDavid Howells {} 8632021982SDavid Howells }; 8732021982SDavid Howells 8870c3547eSMike Kravetz #ifdef CONFIG_NUMA 8970c3547eSMike Kravetz static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, 9070c3547eSMike Kravetz struct inode *inode, pgoff_t index) 9170c3547eSMike Kravetz { 9270c3547eSMike Kravetz vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy, 9370c3547eSMike Kravetz index); 9470c3547eSMike Kravetz } 9570c3547eSMike Kravetz 9670c3547eSMike Kravetz static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) 9770c3547eSMike Kravetz { 9870c3547eSMike Kravetz mpol_cond_put(vma->vm_policy); 9970c3547eSMike Kravetz } 10070c3547eSMike Kravetz #else 10170c3547eSMike Kravetz static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, 10270c3547eSMike Kravetz struct inode *inode, pgoff_t index) 10370c3547eSMike Kravetz { 10470c3547eSMike Kravetz } 10570c3547eSMike Kravetz 10670c3547eSMike Kravetz static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) 10770c3547eSMike Kravetz { 10870c3547eSMike Kravetz } 10970c3547eSMike Kravetz #endif 11070c3547eSMike Kravetz 1112e9b367cSAdam Litke static void huge_pagevec_release(struct pagevec *pvec) 1122e9b367cSAdam Litke { 1132e9b367cSAdam Litke int i; 1142e9b367cSAdam Litke 1152e9b367cSAdam Litke for (i = 0; i < pagevec_count(pvec); ++i) 1162e9b367cSAdam Litke put_page(pvec->pages[i]); 1172e9b367cSAdam Litke 1182e9b367cSAdam Litke pagevec_reinit(pvec); 1192e9b367cSAdam Litke } 1202e9b367cSAdam Litke 12163489f8eSMike Kravetz /* 12263489f8eSMike Kravetz * Mask used when checking the page offset value passed in via system 12363489f8eSMike Kravetz * calls. This value will be converted to a loff_t which is signed. 12463489f8eSMike Kravetz * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the 12563489f8eSMike Kravetz * value. The extra bit (- 1 in the shift value) is to take the sign 12663489f8eSMike Kravetz * bit into account. 12763489f8eSMike Kravetz */ 12863489f8eSMike Kravetz #define PGOFF_LOFFT_MAX \ 12963489f8eSMike Kravetz (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1))) 13063489f8eSMike Kravetz 1311da177e4SLinus Torvalds static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) 1321da177e4SLinus Torvalds { 133496ad9aaSAl Viro struct inode *inode = file_inode(file); 1341da177e4SLinus Torvalds loff_t len, vma_len; 1351da177e4SLinus Torvalds int ret; 136a5516438SAndi Kleen struct hstate *h = hstate_file(file); 1371da177e4SLinus Torvalds 13868589bc3SHugh Dickins /* 139dec4ad86SDavid Gibson * vma address alignment (but not the pgoff alignment) has 140dec4ad86SDavid Gibson * already been checked by prepare_hugepage_range. If you add 141dec4ad86SDavid Gibson * any error returns here, do so after setting VM_HUGETLB, so 142dec4ad86SDavid Gibson * is_vm_hugetlb_page tests below unmap_region go the right 14345e55300SPeter Collingbourne * way when do_mmap unwinds (may be important on powerpc 144dec4ad86SDavid Gibson * and ia64). 14568589bc3SHugh Dickins */ 146a2fce914SNaoya Horiguchi vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; 14768589bc3SHugh Dickins vma->vm_ops = &hugetlb_vm_ops; 1481da177e4SLinus Torvalds 149045c7a3fSMike Kravetz /* 15063489f8eSMike Kravetz * page based offset in vm_pgoff could be sufficiently large to 1515df63c2aSMike Kravetz * overflow a loff_t when converted to byte offset. This can 1525df63c2aSMike Kravetz * only happen on architectures where sizeof(loff_t) == 1535df63c2aSMike Kravetz * sizeof(unsigned long). So, only check in those instances. 154045c7a3fSMike Kravetz */ 1555df63c2aSMike Kravetz if (sizeof(unsigned long) == sizeof(loff_t)) { 15663489f8eSMike Kravetz if (vma->vm_pgoff & PGOFF_LOFFT_MAX) 157045c7a3fSMike Kravetz return -EINVAL; 1585df63c2aSMike Kravetz } 159045c7a3fSMike Kravetz 16063489f8eSMike Kravetz /* must be huge page aligned */ 1612b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 162dec4ad86SDavid Gibson return -EINVAL; 163dec4ad86SDavid Gibson 1641da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 165045c7a3fSMike Kravetz len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 166045c7a3fSMike Kravetz /* check for overflow */ 167045c7a3fSMike Kravetz if (len < vma_len) 168045c7a3fSMike Kravetz return -EINVAL; 1691da177e4SLinus Torvalds 1705955102cSAl Viro inode_lock(inode); 1711da177e4SLinus Torvalds file_accessed(file); 1721da177e4SLinus Torvalds 1731da177e4SLinus Torvalds ret = -ENOMEM; 174*33b8f84aSMike Kravetz if (!hugetlb_reserve_pages(inode, 175a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1765a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1775a6fe125SMel Gorman vma->vm_flags)) 178b45b5bd6SDavid Gibson goto out; 179b45b5bd6SDavid Gibson 1804c887265SAdam Litke ret = 0; 181b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 182045c7a3fSMike Kravetz i_size_write(inode, len); 1831da177e4SLinus Torvalds out: 1845955102cSAl Viro inode_unlock(inode); 1851da177e4SLinus Torvalds 1861da177e4SLinus Torvalds return ret; 1871da177e4SLinus Torvalds } 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds /* 1903e4e28c5SMichel Lespinasse * Called under mmap_write_lock(mm). 1911da177e4SLinus Torvalds */ 1921da177e4SLinus Torvalds 193d2ba27e8SAdrian Bunk #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 1941da177e4SLinus Torvalds static unsigned long 19588590253SShijie Hu hugetlb_get_unmapped_area_bottomup(struct file *file, unsigned long addr, 19688590253SShijie Hu unsigned long len, unsigned long pgoff, unsigned long flags) 19788590253SShijie Hu { 19888590253SShijie Hu struct hstate *h = hstate_file(file); 19988590253SShijie Hu struct vm_unmapped_area_info info; 20088590253SShijie Hu 20188590253SShijie Hu info.flags = 0; 20288590253SShijie Hu info.length = len; 20388590253SShijie Hu info.low_limit = current->mm->mmap_base; 20488590253SShijie Hu info.high_limit = TASK_SIZE; 20588590253SShijie Hu info.align_mask = PAGE_MASK & ~huge_page_mask(h); 20688590253SShijie Hu info.align_offset = 0; 20788590253SShijie Hu return vm_unmapped_area(&info); 20888590253SShijie Hu } 20988590253SShijie Hu 21088590253SShijie Hu static unsigned long 21188590253SShijie Hu hugetlb_get_unmapped_area_topdown(struct file *file, unsigned long addr, 21288590253SShijie Hu unsigned long len, unsigned long pgoff, unsigned long flags) 21388590253SShijie Hu { 21488590253SShijie Hu struct hstate *h = hstate_file(file); 21588590253SShijie Hu struct vm_unmapped_area_info info; 21688590253SShijie Hu 21788590253SShijie Hu info.flags = VM_UNMAPPED_AREA_TOPDOWN; 21888590253SShijie Hu info.length = len; 21988590253SShijie Hu info.low_limit = max(PAGE_SIZE, mmap_min_addr); 22088590253SShijie Hu info.high_limit = current->mm->mmap_base; 22188590253SShijie Hu info.align_mask = PAGE_MASK & ~huge_page_mask(h); 22288590253SShijie Hu info.align_offset = 0; 22388590253SShijie Hu addr = vm_unmapped_area(&info); 22488590253SShijie Hu 22588590253SShijie Hu /* 22688590253SShijie Hu * A failed mmap() very likely causes application failure, 22788590253SShijie Hu * so fall back to the bottom-up function here. This scenario 22888590253SShijie Hu * can happen with large stack limits and large mmap() 22988590253SShijie Hu * allocations. 23088590253SShijie Hu */ 23188590253SShijie Hu if (unlikely(offset_in_page(addr))) { 23288590253SShijie Hu VM_BUG_ON(addr != -ENOMEM); 23388590253SShijie Hu info.flags = 0; 23488590253SShijie Hu info.low_limit = current->mm->mmap_base; 23588590253SShijie Hu info.high_limit = TASK_SIZE; 23688590253SShijie Hu addr = vm_unmapped_area(&info); 23788590253SShijie Hu } 23888590253SShijie Hu 23988590253SShijie Hu return addr; 24088590253SShijie Hu } 24188590253SShijie Hu 24288590253SShijie Hu static unsigned long 2431da177e4SLinus Torvalds hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 2441da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 2451da177e4SLinus Torvalds { 2461da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2471da177e4SLinus Torvalds struct vm_area_struct *vma; 248a5516438SAndi Kleen struct hstate *h = hstate_file(file); 2491da177e4SLinus Torvalds 250a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 2511da177e4SLinus Torvalds return -EINVAL; 2521da177e4SLinus Torvalds if (len > TASK_SIZE) 2531da177e4SLinus Torvalds return -ENOMEM; 2541da177e4SLinus Torvalds 255036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 256a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 257036e0856SBenjamin Herrenschmidt return -EINVAL; 258036e0856SBenjamin Herrenschmidt return addr; 259036e0856SBenjamin Herrenschmidt } 260036e0856SBenjamin Herrenschmidt 2611da177e4SLinus Torvalds if (addr) { 262a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 2631da177e4SLinus Torvalds vma = find_vma(mm, addr); 2641da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 2651be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma))) 2661da177e4SLinus Torvalds return addr; 2671da177e4SLinus Torvalds } 2681da177e4SLinus Torvalds 26988590253SShijie Hu /* 27088590253SShijie Hu * Use mm->get_unmapped_area value as a hint to use topdown routine. 27188590253SShijie Hu * If architectures have special needs, they should define their own 27288590253SShijie Hu * version of hugetlb_get_unmapped_area. 27388590253SShijie Hu */ 27488590253SShijie Hu if (mm->get_unmapped_area == arch_get_unmapped_area_topdown) 27588590253SShijie Hu return hugetlb_get_unmapped_area_topdown(file, addr, len, 27688590253SShijie Hu pgoff, flags); 27788590253SShijie Hu return hugetlb_get_unmapped_area_bottomup(file, addr, len, 27888590253SShijie Hu pgoff, flags); 2791da177e4SLinus Torvalds } 2801da177e4SLinus Torvalds #endif 2811da177e4SLinus Torvalds 28234d0640eSAl Viro static size_t 283e63e1e5aSBadari Pulavarty hugetlbfs_read_actor(struct page *page, unsigned long offset, 28434d0640eSAl Viro struct iov_iter *to, unsigned long size) 285e63e1e5aSBadari Pulavarty { 28634d0640eSAl Viro size_t copied = 0; 287e63e1e5aSBadari Pulavarty int i, chunksize; 288e63e1e5aSBadari Pulavarty 289e63e1e5aSBadari Pulavarty /* Find which 4k chunk and offset with in that chunk */ 29009cbfeafSKirill A. Shutemov i = offset >> PAGE_SHIFT; 29109cbfeafSKirill A. Shutemov offset = offset & ~PAGE_MASK; 292e63e1e5aSBadari Pulavarty 293e63e1e5aSBadari Pulavarty while (size) { 29434d0640eSAl Viro size_t n; 29509cbfeafSKirill A. Shutemov chunksize = PAGE_SIZE; 296e63e1e5aSBadari Pulavarty if (offset) 297e63e1e5aSBadari Pulavarty chunksize -= offset; 298e63e1e5aSBadari Pulavarty if (chunksize > size) 299e63e1e5aSBadari Pulavarty chunksize = size; 30034d0640eSAl Viro n = copy_page_to_iter(&page[i], offset, chunksize, to); 30134d0640eSAl Viro copied += n; 30234d0640eSAl Viro if (n != chunksize) 30334d0640eSAl Viro return copied; 304e63e1e5aSBadari Pulavarty offset = 0; 305e63e1e5aSBadari Pulavarty size -= chunksize; 306e63e1e5aSBadari Pulavarty i++; 307e63e1e5aSBadari Pulavarty } 30834d0640eSAl Viro return copied; 309e63e1e5aSBadari Pulavarty } 310e63e1e5aSBadari Pulavarty 311e63e1e5aSBadari Pulavarty /* 312e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 313e63e1e5aSBadari Pulavarty * data. Its *very* similar to do_generic_mapping_read(), we can't use that 314ea1754a0SKirill A. Shutemov * since it has PAGE_SIZE assumptions. 315e63e1e5aSBadari Pulavarty */ 31634d0640eSAl Viro static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) 317e63e1e5aSBadari Pulavarty { 31834d0640eSAl Viro struct file *file = iocb->ki_filp; 31934d0640eSAl Viro struct hstate *h = hstate_file(file); 32034d0640eSAl Viro struct address_space *mapping = file->f_mapping; 321e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 32234d0640eSAl Viro unsigned long index = iocb->ki_pos >> huge_page_shift(h); 32334d0640eSAl Viro unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); 324e63e1e5aSBadari Pulavarty unsigned long end_index; 325e63e1e5aSBadari Pulavarty loff_t isize; 326e63e1e5aSBadari Pulavarty ssize_t retval = 0; 327e63e1e5aSBadari Pulavarty 32834d0640eSAl Viro while (iov_iter_count(to)) { 329e63e1e5aSBadari Pulavarty struct page *page; 33034d0640eSAl Viro size_t nr, copied; 331e63e1e5aSBadari Pulavarty 332e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 333a5516438SAndi Kleen nr = huge_page_size(h); 334a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 335a05b0855SAneesh Kumar K.V if (!isize) 33634d0640eSAl Viro break; 337a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 338e63e1e5aSBadari Pulavarty if (index > end_index) 33934d0640eSAl Viro break; 34034d0640eSAl Viro if (index == end_index) { 341a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 342a05b0855SAneesh Kumar K.V if (nr <= offset) 34334d0640eSAl Viro break; 344e63e1e5aSBadari Pulavarty } 345e63e1e5aSBadari Pulavarty nr = nr - offset; 346e63e1e5aSBadari Pulavarty 347e63e1e5aSBadari Pulavarty /* Find the page */ 348a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 349e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 350e63e1e5aSBadari Pulavarty /* 351e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 352e63e1e5aSBadari Pulavarty * length of the hole or request. 353e63e1e5aSBadari Pulavarty */ 35434d0640eSAl Viro copied = iov_iter_zero(nr, to); 355e63e1e5aSBadari Pulavarty } else { 356a05b0855SAneesh Kumar K.V unlock_page(page); 357a05b0855SAneesh Kumar K.V 358e63e1e5aSBadari Pulavarty /* 359e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 360e63e1e5aSBadari Pulavarty */ 36134d0640eSAl Viro copied = hugetlbfs_read_actor(page, offset, to, nr); 36209cbfeafSKirill A. Shutemov put_page(page); 363e63e1e5aSBadari Pulavarty } 36434d0640eSAl Viro offset += copied; 36534d0640eSAl Viro retval += copied; 36634d0640eSAl Viro if (copied != nr && iov_iter_count(to)) { 36734d0640eSAl Viro if (!retval) 36834d0640eSAl Viro retval = -EFAULT; 369e63e1e5aSBadari Pulavarty break; 370e63e1e5aSBadari Pulavarty } 37134d0640eSAl Viro index += offset >> huge_page_shift(h); 37234d0640eSAl Viro offset &= ~huge_page_mask(h); 37334d0640eSAl Viro } 37434d0640eSAl Viro iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; 375e63e1e5aSBadari Pulavarty return retval; 376e63e1e5aSBadari Pulavarty } 377e63e1e5aSBadari Pulavarty 378800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 379800d15a5SNick Piggin struct address_space *mapping, 380800d15a5SNick Piggin loff_t pos, unsigned len, unsigned flags, 381800d15a5SNick Piggin struct page **pagep, void **fsdata) 3821da177e4SLinus Torvalds { 3831da177e4SLinus Torvalds return -EINVAL; 3841da177e4SLinus Torvalds } 3851da177e4SLinus Torvalds 386800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 387800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 388800d15a5SNick Piggin struct page *page, void *fsdata) 3891da177e4SLinus Torvalds { 390800d15a5SNick Piggin BUG(); 3911da177e4SLinus Torvalds return -EINVAL; 3921da177e4SLinus Torvalds } 3931da177e4SLinus Torvalds 394b5cec28dSMike Kravetz static void remove_huge_page(struct page *page) 3951da177e4SLinus Torvalds { 396b9ea2515SKonstantin Khlebnikov ClearPageDirty(page); 3971da177e4SLinus Torvalds ClearPageUptodate(page); 398bd65cb86SMinchan Kim delete_from_page_cache(page); 3991da177e4SLinus Torvalds } 4001da177e4SLinus Torvalds 4014aae8d1cSMike Kravetz static void 402f808c13fSDavidlohr Bueso hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end) 4034aae8d1cSMike Kravetz { 4044aae8d1cSMike Kravetz struct vm_area_struct *vma; 4054aae8d1cSMike Kravetz 4064aae8d1cSMike Kravetz /* 4074aae8d1cSMike Kravetz * end == 0 indicates that the entire range after 4084aae8d1cSMike Kravetz * start should be unmapped. 4094aae8d1cSMike Kravetz */ 4104aae8d1cSMike Kravetz vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { 4114aae8d1cSMike Kravetz unsigned long v_offset; 4124aae8d1cSMike Kravetz unsigned long v_end; 4134aae8d1cSMike Kravetz 4144aae8d1cSMike Kravetz /* 4154aae8d1cSMike Kravetz * Can the expression below overflow on 32-bit arches? 4164aae8d1cSMike Kravetz * No, because the interval tree returns us only those vmas 4174aae8d1cSMike Kravetz * which overlap the truncated area starting at pgoff, 4184aae8d1cSMike Kravetz * and no vma on a 32-bit arch can span beyond the 4GB. 4194aae8d1cSMike Kravetz */ 4204aae8d1cSMike Kravetz if (vma->vm_pgoff < start) 4214aae8d1cSMike Kravetz v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; 4224aae8d1cSMike Kravetz else 4234aae8d1cSMike Kravetz v_offset = 0; 4244aae8d1cSMike Kravetz 4254aae8d1cSMike Kravetz if (!end) 4264aae8d1cSMike Kravetz v_end = vma->vm_end; 4274aae8d1cSMike Kravetz else { 4284aae8d1cSMike Kravetz v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) 4294aae8d1cSMike Kravetz + vma->vm_start; 4304aae8d1cSMike Kravetz if (v_end > vma->vm_end) 4314aae8d1cSMike Kravetz v_end = vma->vm_end; 4324aae8d1cSMike Kravetz } 4334aae8d1cSMike Kravetz 4344aae8d1cSMike Kravetz unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, 4354aae8d1cSMike Kravetz NULL); 4364aae8d1cSMike Kravetz } 4374aae8d1cSMike Kravetz } 438b5cec28dSMike Kravetz 439b5cec28dSMike Kravetz /* 440b5cec28dSMike Kravetz * remove_inode_hugepages handles two distinct cases: truncation and hole 441b5cec28dSMike Kravetz * punch. There are subtle differences in operation for each case. 4424aae8d1cSMike Kravetz * 443b5cec28dSMike Kravetz * truncation is indicated by end of range being LLONG_MAX 444b5cec28dSMike Kravetz * In this case, we first scan the range and release found pages. 445b5cec28dSMike Kravetz * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv 446e7c58097SMike Kravetz * maps and global counts. Page faults can not race with truncation 44787bf91d3SMike Kravetz * in this routine. hugetlb_no_page() holds i_mmap_rwsem and prevents 44887bf91d3SMike Kravetz * page faults in the truncated range by checking i_size. i_size is 44987bf91d3SMike Kravetz * modified while holding i_mmap_rwsem. 450b5cec28dSMike Kravetz * hole punch is indicated if end is not LLONG_MAX 451b5cec28dSMike Kravetz * In the hole punch case we scan the range and release found pages. 452b5cec28dSMike Kravetz * Only when releasing a page is the associated region/reserv map 453b5cec28dSMike Kravetz * deleted. The region/reserv map for ranges without associated 454e7c58097SMike Kravetz * pages are not modified. Page faults can race with hole punch. 455e7c58097SMike Kravetz * This is indicated if we find a mapped page. 456b5cec28dSMike Kravetz * Note: If the passed end of range value is beyond the end of file, but 457b5cec28dSMike Kravetz * not LLONG_MAX this routine still performs a hole punch operation. 458b5cec28dSMike Kravetz */ 459b5cec28dSMike Kravetz static void remove_inode_hugepages(struct inode *inode, loff_t lstart, 460b5cec28dSMike Kravetz loff_t lend) 4611da177e4SLinus Torvalds { 462a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 463b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 464a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 465b5cec28dSMike Kravetz const pgoff_t end = lend >> huge_page_shift(h); 466b5cec28dSMike Kravetz struct vm_area_struct pseudo_vma; 4671da177e4SLinus Torvalds struct pagevec pvec; 468d72dc8a2SJan Kara pgoff_t next, index; 469a43a8c39SChen, Kenneth W int i, freed = 0; 470b5cec28dSMike Kravetz bool truncate_op = (lend == LLONG_MAX); 4711da177e4SLinus Torvalds 4722c4541e2SKirill A. Shutemov vma_init(&pseudo_vma, current->mm); 473b5cec28dSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 47486679820SMel Gorman pagevec_init(&pvec); 4751da177e4SLinus Torvalds next = start; 476b5cec28dSMike Kravetz while (next < end) { 477b5cec28dSMike Kravetz /* 4781817889eSMike Kravetz * When no more pages are found, we are done. 479b5cec28dSMike Kravetz */ 480397162ffSJan Kara if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1)) 4811da177e4SLinus Torvalds break; 4821da177e4SLinus Torvalds 4831da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); ++i) { 4841da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 485e7c58097SMike Kravetz u32 hash; 486b5cec28dSMike Kravetz 487d72dc8a2SJan Kara index = page->index; 488188b04a7SWei Yang hash = hugetlb_fault_mutex_hash(mapping, index); 48987bf91d3SMike Kravetz if (!truncate_op) { 49087bf91d3SMike Kravetz /* 49187bf91d3SMike Kravetz * Only need to hold the fault mutex in the 49287bf91d3SMike Kravetz * hole punch case. This prevents races with 49387bf91d3SMike Kravetz * page faults. Races are not possible in the 49487bf91d3SMike Kravetz * case of truncation. 49587bf91d3SMike Kravetz */ 496e7c58097SMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 49787bf91d3SMike Kravetz } 498e7c58097SMike Kravetz 499b5cec28dSMike Kravetz /* 500e7c58097SMike Kravetz * If page is mapped, it was faulted in after being 501e7c58097SMike Kravetz * unmapped in caller. Unmap (again) now after taking 502e7c58097SMike Kravetz * the fault mutex. The mutex will prevent faults 503e7c58097SMike Kravetz * until we finish removing the page. 504e7c58097SMike Kravetz * 505e7c58097SMike Kravetz * This race can only happen in the hole punch case. 506e7c58097SMike Kravetz * Getting here in a truncate operation is a bug. 507b5cec28dSMike Kravetz */ 508e7c58097SMike Kravetz if (unlikely(page_mapped(page))) { 509e7c58097SMike Kravetz BUG_ON(truncate_op); 510e7c58097SMike Kravetz 511c0d0381aSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 512e7c58097SMike Kravetz i_mmap_lock_write(mapping); 513c0d0381aSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 514e7c58097SMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 515e7c58097SMike Kravetz index * pages_per_huge_page(h), 516e7c58097SMike Kravetz (index + 1) * pages_per_huge_page(h)); 517e7c58097SMike Kravetz i_mmap_unlock_write(mapping); 518e7c58097SMike Kravetz } 5194aae8d1cSMike Kravetz 5204aae8d1cSMike Kravetz lock_page(page); 5214aae8d1cSMike Kravetz /* 5224aae8d1cSMike Kravetz * We must free the huge page and remove from page 5234aae8d1cSMike Kravetz * cache (remove_huge_page) BEFORE removing the 5244aae8d1cSMike Kravetz * region/reserve map (hugetlb_unreserve_pages). In 5254aae8d1cSMike Kravetz * rare out of memory conditions, removal of the 52672e2936cSzhong jiang * region/reserve map could fail. Correspondingly, 52772e2936cSzhong jiang * the subpool and global reserve usage count can need 52872e2936cSzhong jiang * to be adjusted. 5294aae8d1cSMike Kravetz */ 53072e2936cSzhong jiang VM_BUG_ON(PagePrivate(page)); 531b5cec28dSMike Kravetz remove_huge_page(page); 532b5cec28dSMike Kravetz freed++; 533b5cec28dSMike Kravetz if (!truncate_op) { 5344aae8d1cSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, 535d72dc8a2SJan Kara index, index + 1, 1))) 53672e2936cSzhong jiang hugetlb_fix_reserve_counts(inode); 537b5cec28dSMike Kravetz } 538b5cec28dSMike Kravetz 5391da177e4SLinus Torvalds unlock_page(page); 54087bf91d3SMike Kravetz if (!truncate_op) 541e7c58097SMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 5421da177e4SLinus Torvalds } 5431da177e4SLinus Torvalds huge_pagevec_release(&pvec); 5441817889eSMike Kravetz cond_resched(); 5451da177e4SLinus Torvalds } 546b5cec28dSMike Kravetz 547b5cec28dSMike Kravetz if (truncate_op) 548b5cec28dSMike Kravetz (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); 5491da177e4SLinus Torvalds } 5501da177e4SLinus Torvalds 5512bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 5521da177e4SLinus Torvalds { 5539119a41eSJoonsoo Kim struct resv_map *resv_map; 5549119a41eSJoonsoo Kim 555b5cec28dSMike Kravetz remove_inode_hugepages(inode, 0, LLONG_MAX); 556f27a5136SMike Kravetz 557f27a5136SMike Kravetz /* 558f27a5136SMike Kravetz * Get the resv_map from the address space embedded in the inode. 559f27a5136SMike Kravetz * This is the address space which points to any resv_map allocated 560f27a5136SMike Kravetz * at inode creation time. If this is a device special inode, 561f27a5136SMike Kravetz * i_mapping may not point to the original address space. 562f27a5136SMike Kravetz */ 563f27a5136SMike Kravetz resv_map = (struct resv_map *)(&inode->i_data)->private_data; 564f27a5136SMike Kravetz /* Only regular and link inodes have associated reserve maps */ 5659119a41eSJoonsoo Kim if (resv_map) 5669119a41eSJoonsoo Kim resv_map_release(&resv_map->refs); 567dbd5768fSJan Kara clear_inode(inode); 568149f4211SChristoph Hellwig } 569149f4211SChristoph Hellwig 5701da177e4SLinus Torvalds static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) 5711da177e4SLinus Torvalds { 572856fc295SHugh Dickins pgoff_t pgoff; 5731da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 574a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 5751da177e4SLinus Torvalds 576a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 577856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 5781da177e4SLinus Torvalds 57983cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 58087bf91d3SMike Kravetz i_size_write(inode, offset); 581f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 5821bfad99aSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); 583c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 584e7c58097SMike Kravetz remove_inode_hugepages(inode, offset, LLONG_MAX); 5851da177e4SLinus Torvalds return 0; 5861da177e4SLinus Torvalds } 5871da177e4SLinus Torvalds 58870c3547eSMike Kravetz static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) 58970c3547eSMike Kravetz { 59070c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 59170c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 59270c3547eSMike Kravetz loff_t hole_start, hole_end; 59370c3547eSMike Kravetz 59470c3547eSMike Kravetz /* 59570c3547eSMike Kravetz * For hole punch round up the beginning offset of the hole and 59670c3547eSMike Kravetz * round down the end. 59770c3547eSMike Kravetz */ 59870c3547eSMike Kravetz hole_start = round_up(offset, hpage_size); 59970c3547eSMike Kravetz hole_end = round_down(offset + len, hpage_size); 60070c3547eSMike Kravetz 60170c3547eSMike Kravetz if (hole_end > hole_start) { 60270c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 603ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 60470c3547eSMike Kravetz 6055955102cSAl Viro inode_lock(inode); 606ff62a342SMarc-André Lureau 607ff62a342SMarc-André Lureau /* protected by i_mutex */ 608ab3948f5SJoel Fernandes (Google) if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { 609ff62a342SMarc-André Lureau inode_unlock(inode); 610ff62a342SMarc-André Lureau return -EPERM; 611ff62a342SMarc-André Lureau } 612ff62a342SMarc-André Lureau 61370c3547eSMike Kravetz i_mmap_lock_write(mapping); 614f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 61570c3547eSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 61670c3547eSMike Kravetz hole_start >> PAGE_SHIFT, 61770c3547eSMike Kravetz hole_end >> PAGE_SHIFT); 618c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 619e7c58097SMike Kravetz remove_inode_hugepages(inode, hole_start, hole_end); 6205955102cSAl Viro inode_unlock(inode); 62170c3547eSMike Kravetz } 62270c3547eSMike Kravetz 62370c3547eSMike Kravetz return 0; 62470c3547eSMike Kravetz } 62570c3547eSMike Kravetz 62670c3547eSMike Kravetz static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, 62770c3547eSMike Kravetz loff_t len) 62870c3547eSMike Kravetz { 62970c3547eSMike Kravetz struct inode *inode = file_inode(file); 630ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 63170c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 63270c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 63370c3547eSMike Kravetz struct vm_area_struct pseudo_vma; 63470c3547eSMike Kravetz struct mm_struct *mm = current->mm; 63570c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 63670c3547eSMike Kravetz unsigned long hpage_shift = huge_page_shift(h); 63770c3547eSMike Kravetz pgoff_t start, index, end; 63870c3547eSMike Kravetz int error; 63970c3547eSMike Kravetz u32 hash; 64070c3547eSMike Kravetz 64170c3547eSMike Kravetz if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 64270c3547eSMike Kravetz return -EOPNOTSUPP; 64370c3547eSMike Kravetz 64470c3547eSMike Kravetz if (mode & FALLOC_FL_PUNCH_HOLE) 64570c3547eSMike Kravetz return hugetlbfs_punch_hole(inode, offset, len); 64670c3547eSMike Kravetz 64770c3547eSMike Kravetz /* 64870c3547eSMike Kravetz * Default preallocate case. 64970c3547eSMike Kravetz * For this range, start is rounded down and end is rounded up 65070c3547eSMike Kravetz * as well as being converted to page offsets. 65170c3547eSMike Kravetz */ 65270c3547eSMike Kravetz start = offset >> hpage_shift; 65370c3547eSMike Kravetz end = (offset + len + hpage_size - 1) >> hpage_shift; 65470c3547eSMike Kravetz 6555955102cSAl Viro inode_lock(inode); 65670c3547eSMike Kravetz 65770c3547eSMike Kravetz /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ 65870c3547eSMike Kravetz error = inode_newsize_ok(inode, offset + len); 65970c3547eSMike Kravetz if (error) 66070c3547eSMike Kravetz goto out; 66170c3547eSMike Kravetz 662ff62a342SMarc-André Lureau if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { 663ff62a342SMarc-André Lureau error = -EPERM; 664ff62a342SMarc-André Lureau goto out; 665ff62a342SMarc-André Lureau } 666ff62a342SMarc-André Lureau 66770c3547eSMike Kravetz /* 66870c3547eSMike Kravetz * Initialize a pseudo vma as this is required by the huge page 66970c3547eSMike Kravetz * allocation routines. If NUMA is configured, use page index 67070c3547eSMike Kravetz * as input to create an allocation policy. 67170c3547eSMike Kravetz */ 6722c4541e2SKirill A. Shutemov vma_init(&pseudo_vma, mm); 67370c3547eSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 67470c3547eSMike Kravetz pseudo_vma.vm_file = file; 67570c3547eSMike Kravetz 67670c3547eSMike Kravetz for (index = start; index < end; index++) { 67770c3547eSMike Kravetz /* 67870c3547eSMike Kravetz * This is supposed to be the vaddr where the page is being 67970c3547eSMike Kravetz * faulted in, but we have no vaddr here. 68070c3547eSMike Kravetz */ 68170c3547eSMike Kravetz struct page *page; 68270c3547eSMike Kravetz unsigned long addr; 68370c3547eSMike Kravetz int avoid_reserve = 0; 68470c3547eSMike Kravetz 68570c3547eSMike Kravetz cond_resched(); 68670c3547eSMike Kravetz 68770c3547eSMike Kravetz /* 68870c3547eSMike Kravetz * fallocate(2) manpage permits EINTR; we may have been 68970c3547eSMike Kravetz * interrupted because we are using up too much memory. 69070c3547eSMike Kravetz */ 69170c3547eSMike Kravetz if (signal_pending(current)) { 69270c3547eSMike Kravetz error = -EINTR; 69370c3547eSMike Kravetz break; 69470c3547eSMike Kravetz } 69570c3547eSMike Kravetz 69670c3547eSMike Kravetz /* Set numa allocation policy based on index */ 69770c3547eSMike Kravetz hugetlb_set_vma_policy(&pseudo_vma, inode, index); 69870c3547eSMike Kravetz 69970c3547eSMike Kravetz /* addr is the offset within the file (zero based) */ 70070c3547eSMike Kravetz addr = index * hpage_size; 70170c3547eSMike Kravetz 70287bf91d3SMike Kravetz /* 70387bf91d3SMike Kravetz * fault mutex taken here, protects against fault path 70487bf91d3SMike Kravetz * and hole punch. inode_lock previously taken protects 70587bf91d3SMike Kravetz * against truncation. 70687bf91d3SMike Kravetz */ 707188b04a7SWei Yang hash = hugetlb_fault_mutex_hash(mapping, index); 70870c3547eSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 70970c3547eSMike Kravetz 71070c3547eSMike Kravetz /* See if already present in mapping to avoid alloc/free */ 71170c3547eSMike Kravetz page = find_get_page(mapping, index); 71270c3547eSMike Kravetz if (page) { 71370c3547eSMike Kravetz put_page(page); 71470c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 71570c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 71670c3547eSMike Kravetz continue; 71770c3547eSMike Kravetz } 71870c3547eSMike Kravetz 71970c3547eSMike Kravetz /* Allocate page and add to page cache */ 72070c3547eSMike Kravetz page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); 72170c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 72270c3547eSMike Kravetz if (IS_ERR(page)) { 72370c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 72470c3547eSMike Kravetz error = PTR_ERR(page); 72570c3547eSMike Kravetz goto out; 72670c3547eSMike Kravetz } 72770c3547eSMike Kravetz clear_huge_page(page, addr, pages_per_huge_page(h)); 72870c3547eSMike Kravetz __SetPageUptodate(page); 72970c3547eSMike Kravetz error = huge_add_to_page_cache(page, mapping, index); 73070c3547eSMike Kravetz if (unlikely(error)) { 73170c3547eSMike Kravetz put_page(page); 73270c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 73370c3547eSMike Kravetz goto out; 73470c3547eSMike Kravetz } 73570c3547eSMike Kravetz 73670c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 73770c3547eSMike Kravetz 7388f251a3dSMike Kravetz SetHPageMigratable(page); 73970c3547eSMike Kravetz /* 74070c3547eSMike Kravetz * unlock_page because locked by add_to_page_cache() 741585fc0d2SMuchun Song * put_page() due to reference from alloc_huge_page() 74270c3547eSMike Kravetz */ 74370c3547eSMike Kravetz unlock_page(page); 74472639e6dSNadav Amit put_page(page); 74570c3547eSMike Kravetz } 74670c3547eSMike Kravetz 74770c3547eSMike Kravetz if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) 74870c3547eSMike Kravetz i_size_write(inode, offset + len); 749078cd827SDeepa Dinamani inode->i_ctime = current_time(inode); 75070c3547eSMike Kravetz out: 7515955102cSAl Viro inode_unlock(inode); 75270c3547eSMike Kravetz return error; 75370c3547eSMike Kravetz } 75470c3547eSMike Kravetz 755549c7297SChristian Brauner static int hugetlbfs_setattr(struct user_namespace *mnt_userns, 756549c7297SChristian Brauner struct dentry *dentry, struct iattr *attr) 7571da177e4SLinus Torvalds { 7582b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 759a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 7601da177e4SLinus Torvalds int error; 7611da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 762ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 7631da177e4SLinus Torvalds 7641da177e4SLinus Torvalds BUG_ON(!inode); 7651da177e4SLinus Torvalds 7662f221d6fSChristian Brauner error = setattr_prepare(&init_user_ns, dentry, attr); 7671da177e4SLinus Torvalds if (error) 7681025774cSChristoph Hellwig return error; 7691da177e4SLinus Torvalds 7701da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 771ff62a342SMarc-André Lureau loff_t oldsize = inode->i_size; 772ff62a342SMarc-André Lureau loff_t newsize = attr->ia_size; 773ff62a342SMarc-André Lureau 774ff62a342SMarc-André Lureau if (newsize & ~huge_page_mask(h)) 7751025774cSChristoph Hellwig return -EINVAL; 776ff62a342SMarc-André Lureau /* protected by i_mutex */ 777ff62a342SMarc-André Lureau if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || 778ff62a342SMarc-André Lureau (newsize > oldsize && (info->seals & F_SEAL_GROW))) 779ff62a342SMarc-André Lureau return -EPERM; 780ff62a342SMarc-André Lureau error = hugetlb_vmtruncate(inode, newsize); 7811da177e4SLinus Torvalds if (error) 7821da177e4SLinus Torvalds return error; 7831da177e4SLinus Torvalds } 7841da177e4SLinus Torvalds 7852f221d6fSChristian Brauner setattr_copy(&init_user_ns, inode, attr); 7861025774cSChristoph Hellwig mark_inode_dirty(inode); 7871025774cSChristoph Hellwig return 0; 7881025774cSChristoph Hellwig } 7891025774cSChristoph Hellwig 7907d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 79132021982SDavid Howells struct hugetlbfs_fs_context *ctx) 7921da177e4SLinus Torvalds { 7931da177e4SLinus Torvalds struct inode *inode; 7941da177e4SLinus Torvalds 7951da177e4SLinus Torvalds inode = new_inode(sb); 7961da177e4SLinus Torvalds if (inode) { 79785fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 79832021982SDavid Howells inode->i_mode = S_IFDIR | ctx->mode; 79932021982SDavid Howells inode->i_uid = ctx->uid; 80032021982SDavid Howells inode->i_gid = ctx->gid; 801078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 8027d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 8037d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 8047d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 8057d54fa64SAl Viro inc_nlink(inode); 80665ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 8077d54fa64SAl Viro } 8087d54fa64SAl Viro return inode; 8097d54fa64SAl Viro } 8107d54fa64SAl Viro 811b610ded7SMichal Hocko /* 812c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 813b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 81488f306b6SKirill A. Shutemov * annotation because huge_pmd_share() does an allocation under hugetlb's 815c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 816b610ded7SMichal Hocko */ 817c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 818b610ded7SMichal Hocko 8197d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 8207d54fa64SAl Viro struct inode *dir, 82118df2252SAl Viro umode_t mode, dev_t dev) 8227d54fa64SAl Viro { 8237d54fa64SAl Viro struct inode *inode; 82458b6e5e8SMike Kravetz struct resv_map *resv_map = NULL; 8259119a41eSJoonsoo Kim 82658b6e5e8SMike Kravetz /* 82758b6e5e8SMike Kravetz * Reserve maps are only needed for inodes that can have associated 82858b6e5e8SMike Kravetz * page allocations. 82958b6e5e8SMike Kravetz */ 83058b6e5e8SMike Kravetz if (S_ISREG(mode) || S_ISLNK(mode)) { 8319119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 8329119a41eSJoonsoo Kim if (!resv_map) 8339119a41eSJoonsoo Kim return NULL; 83458b6e5e8SMike Kravetz } 8357d54fa64SAl Viro 8367d54fa64SAl Viro inode = new_inode(sb); 8377d54fa64SAl Viro if (inode) { 838ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 839ff62a342SMarc-André Lureau 8407d54fa64SAl Viro inode->i_ino = get_next_ino(); 84121cb47beSChristian Brauner inode_init_owner(&init_user_ns, inode, dir, mode); 842c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 843c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 8441da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 845078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 8469119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 847ff62a342SMarc-André Lureau info->seals = F_SEAL_SEAL; 8481da177e4SLinus Torvalds switch (mode & S_IFMT) { 8491da177e4SLinus Torvalds default: 8501da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 8511da177e4SLinus Torvalds break; 8521da177e4SLinus Torvalds case S_IFREG: 8531da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 8541da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 8551da177e4SLinus Torvalds break; 8561da177e4SLinus Torvalds case S_IFDIR: 8571da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 8581da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 8591da177e4SLinus Torvalds 8601da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 861d8c76e6fSDave Hansen inc_nlink(inode); 8621da177e4SLinus Torvalds break; 8631da177e4SLinus Torvalds case S_IFLNK: 8641da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 86521fc61c7SAl Viro inode_nohighmem(inode); 8661da177e4SLinus Torvalds break; 8671da177e4SLinus Torvalds } 868e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 86958b6e5e8SMike Kravetz } else { 87058b6e5e8SMike Kravetz if (resv_map) 8719119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 87258b6e5e8SMike Kravetz } 8739119a41eSJoonsoo Kim 8741da177e4SLinus Torvalds return inode; 8751da177e4SLinus Torvalds } 8761da177e4SLinus Torvalds 8771da177e4SLinus Torvalds /* 8781da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 8791da177e4SLinus Torvalds */ 8801ab5b82fSPiotr Sarna static int do_hugetlbfs_mknod(struct inode *dir, 8811ab5b82fSPiotr Sarna struct dentry *dentry, 8821ab5b82fSPiotr Sarna umode_t mode, 8831ab5b82fSPiotr Sarna dev_t dev, 8841ab5b82fSPiotr Sarna bool tmpfile) 8851da177e4SLinus Torvalds { 8861da177e4SLinus Torvalds struct inode *inode; 8871da177e4SLinus Torvalds int error = -ENOSPC; 8881da177e4SLinus Torvalds 8897d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 8901da177e4SLinus Torvalds if (inode) { 891078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 8921ab5b82fSPiotr Sarna if (tmpfile) { 8931ab5b82fSPiotr Sarna d_tmpfile(dentry, inode); 8941ab5b82fSPiotr Sarna } else { 8951da177e4SLinus Torvalds d_instantiate(dentry, inode); 8961da177e4SLinus Torvalds dget(dentry);/* Extra count - pin the dentry in core */ 8971ab5b82fSPiotr Sarna } 8981da177e4SLinus Torvalds error = 0; 8991da177e4SLinus Torvalds } 9001da177e4SLinus Torvalds return error; 9011da177e4SLinus Torvalds } 9021da177e4SLinus Torvalds 903549c7297SChristian Brauner static int hugetlbfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 9041ab5b82fSPiotr Sarna struct dentry *dentry, umode_t mode, dev_t dev) 9051ab5b82fSPiotr Sarna { 9061ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode, dev, false); 9071ab5b82fSPiotr Sarna } 9081ab5b82fSPiotr Sarna 909549c7297SChristian Brauner static int hugetlbfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 910549c7297SChristian Brauner struct dentry *dentry, umode_t mode) 9111da177e4SLinus Torvalds { 912549c7297SChristian Brauner int retval = hugetlbfs_mknod(&init_user_ns, dir, dentry, 913549c7297SChristian Brauner mode | S_IFDIR, 0); 9141da177e4SLinus Torvalds if (!retval) 915d8c76e6fSDave Hansen inc_nlink(dir); 9161da177e4SLinus Torvalds return retval; 9171da177e4SLinus Torvalds } 9181da177e4SLinus Torvalds 919549c7297SChristian Brauner static int hugetlbfs_create(struct user_namespace *mnt_userns, 920549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 921549c7297SChristian Brauner umode_t mode, bool excl) 9221da177e4SLinus Torvalds { 923549c7297SChristian Brauner return hugetlbfs_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0); 9241da177e4SLinus Torvalds } 9251da177e4SLinus Torvalds 926549c7297SChristian Brauner static int hugetlbfs_tmpfile(struct user_namespace *mnt_userns, 927549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 928549c7297SChristian Brauner umode_t mode) 9291ab5b82fSPiotr Sarna { 9301ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0, true); 9311ab5b82fSPiotr Sarna } 9321ab5b82fSPiotr Sarna 933549c7297SChristian Brauner static int hugetlbfs_symlink(struct user_namespace *mnt_userns, 934549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 935549c7297SChristian Brauner const char *symname) 9361da177e4SLinus Torvalds { 9371da177e4SLinus Torvalds struct inode *inode; 9381da177e4SLinus Torvalds int error = -ENOSPC; 9391da177e4SLinus Torvalds 9407d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 9411da177e4SLinus Torvalds if (inode) { 9421da177e4SLinus Torvalds int l = strlen(symname)+1; 9431da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 9441da177e4SLinus Torvalds if (!error) { 9451da177e4SLinus Torvalds d_instantiate(dentry, inode); 9461da177e4SLinus Torvalds dget(dentry); 9471da177e4SLinus Torvalds } else 9481da177e4SLinus Torvalds iput(inode); 9491da177e4SLinus Torvalds } 950078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 9511da177e4SLinus Torvalds 9521da177e4SLinus Torvalds return error; 9531da177e4SLinus Torvalds } 9541da177e4SLinus Torvalds 9551da177e4SLinus Torvalds /* 9566649a386SKen Chen * mark the head page dirty 9571da177e4SLinus Torvalds */ 9581da177e4SLinus Torvalds static int hugetlbfs_set_page_dirty(struct page *page) 9591da177e4SLinus Torvalds { 960d85f3385SChristoph Lameter struct page *head = compound_head(page); 9616649a386SKen Chen 9626649a386SKen Chen SetPageDirty(head); 9631da177e4SLinus Torvalds return 0; 9641da177e4SLinus Torvalds } 9651da177e4SLinus Torvalds 966290408d4SNaoya Horiguchi static int hugetlbfs_migrate_page(struct address_space *mapping, 967b969c4abSMel Gorman struct page *newpage, struct page *page, 968a6bc32b8SMel Gorman enum migrate_mode mode) 969290408d4SNaoya Horiguchi { 970290408d4SNaoya Horiguchi int rc; 971290408d4SNaoya Horiguchi 972290408d4SNaoya Horiguchi rc = migrate_huge_page_move_mapping(mapping, newpage, page); 97378bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 974290408d4SNaoya Horiguchi return rc; 975cb6acd01SMike Kravetz 976d6995da3SMike Kravetz if (hugetlb_page_subpool(page)) { 977d6995da3SMike Kravetz hugetlb_set_page_subpool(newpage, hugetlb_page_subpool(page)); 978d6995da3SMike Kravetz hugetlb_set_page_subpool(page, NULL); 979cb6acd01SMike Kravetz } 980cb6acd01SMike Kravetz 9812916ecc0SJérôme Glisse if (mode != MIGRATE_SYNC_NO_COPY) 982290408d4SNaoya Horiguchi migrate_page_copy(newpage, page); 9832916ecc0SJérôme Glisse else 9842916ecc0SJérôme Glisse migrate_page_states(newpage, page); 985290408d4SNaoya Horiguchi 98678bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 987290408d4SNaoya Horiguchi } 988290408d4SNaoya Horiguchi 98978bb9203SNaoya Horiguchi static int hugetlbfs_error_remove_page(struct address_space *mapping, 99078bb9203SNaoya Horiguchi struct page *page) 99178bb9203SNaoya Horiguchi { 99278bb9203SNaoya Horiguchi struct inode *inode = mapping->host; 993ab615a5bSMike Kravetz pgoff_t index = page->index; 99478bb9203SNaoya Horiguchi 99578bb9203SNaoya Horiguchi remove_huge_page(page); 996ab615a5bSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) 99778bb9203SNaoya Horiguchi hugetlb_fix_reserve_counts(inode); 998ab615a5bSMike Kravetz 99978bb9203SNaoya Horiguchi return 0; 100078bb9203SNaoya Horiguchi } 100178bb9203SNaoya Horiguchi 10024a25220dSDavid Howells /* 10034a25220dSDavid Howells * Display the mount options in /proc/mounts. 10044a25220dSDavid Howells */ 10054a25220dSDavid Howells static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) 10064a25220dSDavid Howells { 10074a25220dSDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); 10084a25220dSDavid Howells struct hugepage_subpool *spool = sbinfo->spool; 10094a25220dSDavid Howells unsigned long hpage_size = huge_page_size(sbinfo->hstate); 10104a25220dSDavid Howells unsigned hpage_shift = huge_page_shift(sbinfo->hstate); 10114a25220dSDavid Howells char mod; 10124a25220dSDavid Howells 10134a25220dSDavid Howells if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) 10144a25220dSDavid Howells seq_printf(m, ",uid=%u", 10154a25220dSDavid Howells from_kuid_munged(&init_user_ns, sbinfo->uid)); 10164a25220dSDavid Howells if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) 10174a25220dSDavid Howells seq_printf(m, ",gid=%u", 10184a25220dSDavid Howells from_kgid_munged(&init_user_ns, sbinfo->gid)); 10194a25220dSDavid Howells if (sbinfo->mode != 0755) 10204a25220dSDavid Howells seq_printf(m, ",mode=%o", sbinfo->mode); 10214a25220dSDavid Howells if (sbinfo->max_inodes != -1) 10224a25220dSDavid Howells seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); 10234a25220dSDavid Howells 10244a25220dSDavid Howells hpage_size /= 1024; 10254a25220dSDavid Howells mod = 'K'; 10264a25220dSDavid Howells if (hpage_size >= 1024) { 10274a25220dSDavid Howells hpage_size /= 1024; 10284a25220dSDavid Howells mod = 'M'; 10294a25220dSDavid Howells } 10304a25220dSDavid Howells seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); 10314a25220dSDavid Howells if (spool) { 10324a25220dSDavid Howells if (spool->max_hpages != -1) 10334a25220dSDavid Howells seq_printf(m, ",size=%llu", 10344a25220dSDavid Howells (unsigned long long)spool->max_hpages << hpage_shift); 10354a25220dSDavid Howells if (spool->min_hpages != -1) 10364a25220dSDavid Howells seq_printf(m, ",min_size=%llu", 10374a25220dSDavid Howells (unsigned long long)spool->min_hpages << hpage_shift); 10384a25220dSDavid Howells } 10394a25220dSDavid Howells return 0; 10404a25220dSDavid Howells } 10414a25220dSDavid Howells 1042726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 10431da177e4SLinus Torvalds { 1044726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 10452b0143b5SDavid Howells struct hstate *h = hstate_inode(d_inode(dentry)); 10461da177e4SLinus Torvalds 10471da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 1048a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 10491da177e4SLinus Torvalds if (sbinfo) { 10501da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 105174a8a65cSDavid Gibson /* If no limits set, just report 0 for max/free/used 105274a8a65cSDavid Gibson * blocks, like simple_statfs() */ 105390481622SDavid Gibson if (sbinfo->spool) { 105490481622SDavid Gibson long free_pages; 105590481622SDavid Gibson 105690481622SDavid Gibson spin_lock(&sbinfo->spool->lock); 105790481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 105890481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 105990481622SDavid Gibson - sbinfo->spool->used_hpages; 106090481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 106190481622SDavid Gibson spin_unlock(&sbinfo->spool->lock); 10621da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 10631da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 106474a8a65cSDavid Gibson } 10651da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 10661da177e4SLinus Torvalds } 10671da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 10681da177e4SLinus Torvalds return 0; 10691da177e4SLinus Torvalds } 10701da177e4SLinus Torvalds 10711da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 10721da177e4SLinus Torvalds { 10731da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 10741da177e4SLinus Torvalds 10751da177e4SLinus Torvalds if (sbi) { 10761da177e4SLinus Torvalds sb->s_fs_info = NULL; 107790481622SDavid Gibson 107890481622SDavid Gibson if (sbi->spool) 107990481622SDavid Gibson hugepage_put_subpool(sbi->spool); 108090481622SDavid Gibson 10811da177e4SLinus Torvalds kfree(sbi); 10821da177e4SLinus Torvalds } 10831da177e4SLinus Torvalds } 10841da177e4SLinus Torvalds 108596527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 108696527980SChristoph Hellwig { 108796527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 108896527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 108996527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 109096527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 109196527980SChristoph Hellwig return 0; 109296527980SChristoph Hellwig } 109396527980SChristoph Hellwig sbinfo->free_inodes--; 109496527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 109596527980SChristoph Hellwig } 109696527980SChristoph Hellwig 109796527980SChristoph Hellwig return 1; 109896527980SChristoph Hellwig } 109996527980SChristoph Hellwig 110096527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 110196527980SChristoph Hellwig { 110296527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 110396527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 110496527980SChristoph Hellwig sbinfo->free_inodes++; 110596527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 110696527980SChristoph Hellwig } 110796527980SChristoph Hellwig } 110896527980SChristoph Hellwig 110996527980SChristoph Hellwig 1110e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 11111da177e4SLinus Torvalds 11121da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 11131da177e4SLinus Torvalds { 111496527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 11151da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 11161da177e4SLinus Torvalds 111796527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 11181da177e4SLinus Torvalds return NULL; 1119e94b1766SChristoph Lameter p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); 112096527980SChristoph Hellwig if (unlikely(!p)) { 112196527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 112296527980SChristoph Hellwig return NULL; 11231da177e4SLinus Torvalds } 11244742a35dSMike Kravetz 11254742a35dSMike Kravetz /* 11264742a35dSMike Kravetz * Any time after allocation, hugetlbfs_destroy_inode can be called 11274742a35dSMike Kravetz * for the inode. mpol_free_shared_policy is unconditionally called 11284742a35dSMike Kravetz * as part of hugetlbfs_destroy_inode. So, initialize policy here 11294742a35dSMike Kravetz * in case of a quick call to destroy. 11304742a35dSMike Kravetz * 11314742a35dSMike Kravetz * Note that the policy is initialized even if we are creating a 11324742a35dSMike Kravetz * private inode. This simplifies hugetlbfs_destroy_inode. 11334742a35dSMike Kravetz */ 11344742a35dSMike Kravetz mpol_shared_policy_init(&p->policy, NULL); 11354742a35dSMike Kravetz 113696527980SChristoph Hellwig return &p->vfs_inode; 11371da177e4SLinus Torvalds } 11381da177e4SLinus Torvalds 1139b62de322SAl Viro static void hugetlbfs_free_inode(struct inode *inode) 1140fa0d7e3dSNick Piggin { 1141fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 1142fa0d7e3dSNick Piggin } 1143fa0d7e3dSNick Piggin 11441da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 11451da177e4SLinus Torvalds { 114696527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 11471da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 11481da177e4SLinus Torvalds } 11491da177e4SLinus Torvalds 1150f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 1151800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 1152800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 11531da177e4SLinus Torvalds .set_page_dirty = hugetlbfs_set_page_dirty, 1154290408d4SNaoya Horiguchi .migratepage = hugetlbfs_migrate_page, 115578bb9203SNaoya Horiguchi .error_remove_page = hugetlbfs_error_remove_page, 11561da177e4SLinus Torvalds }; 11571da177e4SLinus Torvalds 115896527980SChristoph Hellwig 115951cc5068SAlexey Dobriyan static void init_once(void *foo) 116096527980SChristoph Hellwig { 116196527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 116296527980SChristoph Hellwig 116396527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 116496527980SChristoph Hellwig } 116596527980SChristoph Hellwig 11664b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 116734d0640eSAl Viro .read_iter = hugetlbfs_read_iter, 11681da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 11691b061d92SChristoph Hellwig .fsync = noop_fsync, 11701da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 11716038f373SArnd Bergmann .llseek = default_llseek, 117270c3547eSMike Kravetz .fallocate = hugetlbfs_fallocate, 11731da177e4SLinus Torvalds }; 11741da177e4SLinus Torvalds 117592e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 11761da177e4SLinus Torvalds .create = hugetlbfs_create, 11771da177e4SLinus Torvalds .lookup = simple_lookup, 11781da177e4SLinus Torvalds .link = simple_link, 11791da177e4SLinus Torvalds .unlink = simple_unlink, 11801da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 11811da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 11821da177e4SLinus Torvalds .rmdir = simple_rmdir, 11831da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 11841da177e4SLinus Torvalds .rename = simple_rename, 11851da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 11861ab5b82fSPiotr Sarna .tmpfile = hugetlbfs_tmpfile, 11871da177e4SLinus Torvalds }; 11881da177e4SLinus Torvalds 118992e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 11901da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 11911da177e4SLinus Torvalds }; 11921da177e4SLinus Torvalds 1193ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 11941da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 1195b62de322SAl Viro .free_inode = hugetlbfs_free_inode, 11961da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 11972bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 11981da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 11991da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 12004a25220dSDavid Howells .show_options = hugetlbfs_show_options, 12011da177e4SLinus Torvalds }; 12021da177e4SLinus Torvalds 12037ca02d0aSMike Kravetz /* 12047ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 12057ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 12067ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 12077ca02d0aSMike Kravetz */ 12084a25220dSDavid Howells static long 12097ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 12104a25220dSDavid Howells enum hugetlbfs_size_type val_type) 12117ca02d0aSMike Kravetz { 12127ca02d0aSMike Kravetz if (val_type == NO_SIZE) 12137ca02d0aSMike Kravetz return -1; 12147ca02d0aSMike Kravetz 12157ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 12167ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 12177ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 12187ca02d0aSMike Kravetz do_div(size_opt, 100); 12197ca02d0aSMike Kravetz } 12207ca02d0aSMike Kravetz 12217ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 12227ca02d0aSMike Kravetz return size_opt; 12237ca02d0aSMike Kravetz } 12247ca02d0aSMike Kravetz 122532021982SDavid Howells /* 122632021982SDavid Howells * Parse one mount parameter. 122732021982SDavid Howells */ 122832021982SDavid Howells static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param) 12291da177e4SLinus Torvalds { 123032021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 123132021982SDavid Howells struct fs_parse_result result; 123232021982SDavid Howells char *rest; 123332021982SDavid Howells unsigned long ps; 123432021982SDavid Howells int opt; 12351da177e4SLinus Torvalds 1236d7167b14SAl Viro opt = fs_parse(fc, hugetlb_fs_parameters, param, &result); 123732021982SDavid Howells if (opt < 0) 123832021982SDavid Howells return opt; 123932021982SDavid Howells 124032021982SDavid Howells switch (opt) { 124132021982SDavid Howells case Opt_uid: 124232021982SDavid Howells ctx->uid = make_kuid(current_user_ns(), result.uint_32); 124332021982SDavid Howells if (!uid_valid(ctx->uid)) 124432021982SDavid Howells goto bad_val; 12451da177e4SLinus Torvalds return 0; 12461da177e4SLinus Torvalds 1247e73a75faSRandy Dunlap case Opt_gid: 124832021982SDavid Howells ctx->gid = make_kgid(current_user_ns(), result.uint_32); 124932021982SDavid Howells if (!gid_valid(ctx->gid)) 1250e73a75faSRandy Dunlap goto bad_val; 125132021982SDavid Howells return 0; 1252e73a75faSRandy Dunlap 1253e73a75faSRandy Dunlap case Opt_mode: 125432021982SDavid Howells ctx->mode = result.uint_32 & 01777U; 125532021982SDavid Howells return 0; 1256e73a75faSRandy Dunlap 125732021982SDavid Howells case Opt_size: 1258e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 125932021982SDavid Howells if (!isdigit(param->string[0])) 1260e73a75faSRandy Dunlap goto bad_val; 126132021982SDavid Howells ctx->max_size_opt = memparse(param->string, &rest); 126232021982SDavid Howells ctx->max_val_type = SIZE_STD; 1263a137e1ccSAndi Kleen if (*rest == '%') 126432021982SDavid Howells ctx->max_val_type = SIZE_PERCENT; 126532021982SDavid Howells return 0; 12661da177e4SLinus Torvalds 1267e73a75faSRandy Dunlap case Opt_nr_inodes: 1268e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 126932021982SDavid Howells if (!isdigit(param->string[0])) 1270e73a75faSRandy Dunlap goto bad_val; 127132021982SDavid Howells ctx->nr_inodes = memparse(param->string, &rest); 127232021982SDavid Howells return 0; 1273e73a75faSRandy Dunlap 127432021982SDavid Howells case Opt_pagesize: 127532021982SDavid Howells ps = memparse(param->string, &rest); 127632021982SDavid Howells ctx->hstate = size_to_hstate(ps); 127732021982SDavid Howells if (!ctx->hstate) { 127832021982SDavid Howells pr_err("Unsupported page size %lu MB\n", ps >> 20); 1279a137e1ccSAndi Kleen return -EINVAL; 1280a137e1ccSAndi Kleen } 128132021982SDavid Howells return 0; 1282a137e1ccSAndi Kleen 128332021982SDavid Howells case Opt_min_size: 12847ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 128532021982SDavid Howells if (!isdigit(param->string[0])) 12867ca02d0aSMike Kravetz goto bad_val; 128732021982SDavid Howells ctx->min_size_opt = memparse(param->string, &rest); 128832021982SDavid Howells ctx->min_val_type = SIZE_STD; 12897ca02d0aSMike Kravetz if (*rest == '%') 129032021982SDavid Howells ctx->min_val_type = SIZE_PERCENT; 129132021982SDavid Howells return 0; 12927ca02d0aSMike Kravetz 1293e73a75faSRandy Dunlap default: 1294b4c07bceSLee Schermerhorn return -EINVAL; 1295e73a75faSRandy Dunlap } 129632021982SDavid Howells 129732021982SDavid Howells bad_val: 1298b5db30cfSAl Viro return invalfc(fc, "Bad value '%s' for mount option '%s'\n", 129932021982SDavid Howells param->string, param->key); 13001da177e4SLinus Torvalds } 1301a137e1ccSAndi Kleen 13027ca02d0aSMike Kravetz /* 130332021982SDavid Howells * Validate the parsed options. 130432021982SDavid Howells */ 130532021982SDavid Howells static int hugetlbfs_validate(struct fs_context *fc) 130632021982SDavid Howells { 130732021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 130832021982SDavid Howells 130932021982SDavid Howells /* 13107ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 13117ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 13127ca02d0aSMike Kravetz */ 131332021982SDavid Howells ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 131432021982SDavid Howells ctx->max_size_opt, 131532021982SDavid Howells ctx->max_val_type); 131632021982SDavid Howells ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 131732021982SDavid Howells ctx->min_size_opt, 131832021982SDavid Howells ctx->min_val_type); 13197ca02d0aSMike Kravetz 13207ca02d0aSMike Kravetz /* 13217ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 13227ca02d0aSMike Kravetz */ 132332021982SDavid Howells if (ctx->max_val_type > NO_SIZE && 132432021982SDavid Howells ctx->min_hpages > ctx->max_hpages) { 132532021982SDavid Howells pr_err("Minimum size can not be greater than maximum size\n"); 13267ca02d0aSMike Kravetz return -EINVAL; 1327a137e1ccSAndi Kleen } 1328a137e1ccSAndi Kleen 13291da177e4SLinus Torvalds return 0; 13301da177e4SLinus Torvalds } 13311da177e4SLinus Torvalds 13321da177e4SLinus Torvalds static int 133332021982SDavid Howells hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc) 13341da177e4SLinus Torvalds { 133532021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 13361da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 13391da177e4SLinus Torvalds if (!sbinfo) 13401da177e4SLinus Torvalds return -ENOMEM; 13411da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 13421da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 134332021982SDavid Howells sbinfo->hstate = ctx->hstate; 134432021982SDavid Howells sbinfo->max_inodes = ctx->nr_inodes; 134532021982SDavid Howells sbinfo->free_inodes = ctx->nr_inodes; 134690481622SDavid Gibson sbinfo->spool = NULL; 134732021982SDavid Howells sbinfo->uid = ctx->uid; 134832021982SDavid Howells sbinfo->gid = ctx->gid; 134932021982SDavid Howells sbinfo->mode = ctx->mode; 13504a25220dSDavid Howells 13517ca02d0aSMike Kravetz /* 13527ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 13537ca02d0aSMike Kravetz * specified. Any needed reservations (for minimim size) are taken 13547ca02d0aSMike Kravetz * taken when the subpool is created. 13557ca02d0aSMike Kravetz */ 135632021982SDavid Howells if (ctx->max_hpages != -1 || ctx->min_hpages != -1) { 135732021982SDavid Howells sbinfo->spool = hugepage_new_subpool(ctx->hstate, 135832021982SDavid Howells ctx->max_hpages, 135932021982SDavid Howells ctx->min_hpages); 136090481622SDavid Gibson if (!sbinfo->spool) 136190481622SDavid Gibson goto out_free; 136290481622SDavid Gibson } 13631da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 136432021982SDavid Howells sb->s_blocksize = huge_page_size(ctx->hstate); 136532021982SDavid Howells sb->s_blocksize_bits = huge_page_shift(ctx->hstate); 13661da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 13671da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 13681da177e4SLinus Torvalds sb->s_time_gran = 1; 136915568299SMike Kravetz 137015568299SMike Kravetz /* 137115568299SMike Kravetz * Due to the special and limited functionality of hugetlbfs, it does 137215568299SMike Kravetz * not work well as a stacking filesystem. 137315568299SMike Kravetz */ 137415568299SMike Kravetz sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH; 137532021982SDavid Howells sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx)); 137648fde701SAl Viro if (!sb->s_root) 13771da177e4SLinus Torvalds goto out_free; 13781da177e4SLinus Torvalds return 0; 13791da177e4SLinus Torvalds out_free: 138090481622SDavid Gibson kfree(sbinfo->spool); 13811da177e4SLinus Torvalds kfree(sbinfo); 13821da177e4SLinus Torvalds return -ENOMEM; 13831da177e4SLinus Torvalds } 13841da177e4SLinus Torvalds 138532021982SDavid Howells static int hugetlbfs_get_tree(struct fs_context *fc) 13861da177e4SLinus Torvalds { 138732021982SDavid Howells int err = hugetlbfs_validate(fc); 138832021982SDavid Howells if (err) 138932021982SDavid Howells return err; 13902ac295d4SAl Viro return get_tree_nodev(fc, hugetlbfs_fill_super); 139132021982SDavid Howells } 139232021982SDavid Howells 139332021982SDavid Howells static void hugetlbfs_fs_context_free(struct fs_context *fc) 139432021982SDavid Howells { 139532021982SDavid Howells kfree(fc->fs_private); 139632021982SDavid Howells } 139732021982SDavid Howells 139832021982SDavid Howells static const struct fs_context_operations hugetlbfs_fs_context_ops = { 139932021982SDavid Howells .free = hugetlbfs_fs_context_free, 140032021982SDavid Howells .parse_param = hugetlbfs_parse_param, 140132021982SDavid Howells .get_tree = hugetlbfs_get_tree, 140232021982SDavid Howells }; 140332021982SDavid Howells 140432021982SDavid Howells static int hugetlbfs_init_fs_context(struct fs_context *fc) 140532021982SDavid Howells { 140632021982SDavid Howells struct hugetlbfs_fs_context *ctx; 140732021982SDavid Howells 140832021982SDavid Howells ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL); 140932021982SDavid Howells if (!ctx) 141032021982SDavid Howells return -ENOMEM; 141132021982SDavid Howells 141232021982SDavid Howells ctx->max_hpages = -1; /* No limit on size by default */ 141332021982SDavid Howells ctx->nr_inodes = -1; /* No limit on number of inodes by default */ 141432021982SDavid Howells ctx->uid = current_fsuid(); 141532021982SDavid Howells ctx->gid = current_fsgid(); 141632021982SDavid Howells ctx->mode = 0755; 141732021982SDavid Howells ctx->hstate = &default_hstate; 141832021982SDavid Howells ctx->min_hpages = -1; /* No default minimum size */ 141932021982SDavid Howells ctx->max_val_type = NO_SIZE; 142032021982SDavid Howells ctx->min_val_type = NO_SIZE; 142132021982SDavid Howells fc->fs_private = ctx; 142232021982SDavid Howells fc->ops = &hugetlbfs_fs_context_ops; 142332021982SDavid Howells return 0; 14241da177e4SLinus Torvalds } 14251da177e4SLinus Torvalds 14261da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 14271da177e4SLinus Torvalds .name = "hugetlbfs", 142832021982SDavid Howells .init_fs_context = hugetlbfs_init_fs_context, 1429d7167b14SAl Viro .parameters = hugetlb_fs_parameters, 14301da177e4SLinus Torvalds .kill_sb = kill_litter_super, 14311da177e4SLinus Torvalds }; 14321da177e4SLinus Torvalds 143342d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 14341da177e4SLinus Torvalds 1435ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 14361da177e4SLinus Torvalds { 1437a0eb3a05SEric W. Biederman kgid_t shm_group; 1438a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 1439a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds 144242d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 144342d7395fSAndi Kleen { 1444af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 144542d7395fSAndi Kleen 144642d7395fSAndi Kleen if (!h) 144742d7395fSAndi Kleen return -1; 144842d7395fSAndi Kleen return h - hstates; 144942d7395fSAndi Kleen } 145042d7395fSAndi Kleen 1451af73e4d9SNaoya Horiguchi /* 1452af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 1453af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 1454af73e4d9SNaoya Horiguchi */ 1455af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 1456af73e4d9SNaoya Horiguchi vm_flags_t acctflag, struct user_struct **user, 145742d7395fSAndi Kleen int creat_flags, int page_size_log) 14581da177e4SLinus Torvalds { 14591da177e4SLinus Torvalds struct inode *inode; 1460e68375c8SAl Viro struct vfsmount *mnt; 146142d7395fSAndi Kleen int hstate_idx; 1462e68375c8SAl Viro struct file *file; 146342d7395fSAndi Kleen 146442d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 146542d7395fSAndi Kleen if (hstate_idx < 0) 146642d7395fSAndi Kleen return ERR_PTR(-ENODEV); 14671da177e4SLinus Torvalds 1468353d5c30SHugh Dickins *user = NULL; 1469e68375c8SAl Viro mnt = hugetlbfs_vfsmount[hstate_idx]; 1470e68375c8SAl Viro if (!mnt) 14715bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 14725bc98594SAkinobu Mita 1473ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 1474353d5c30SHugh Dickins *user = current_user(); 1475353d5c30SHugh Dickins if (user_shm_lock(size, *user)) { 147621a3c273SDavid Rientjes task_lock(current); 14779b857d26SAndrew Morton pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", 147821a3c273SDavid Rientjes current->comm, current->pid); 147921a3c273SDavid Rientjes task_unlock(current); 1480353d5c30SHugh Dickins } else { 1481353d5c30SHugh Dickins *user = NULL; 14821da177e4SLinus Torvalds return ERR_PTR(-EPERM); 14832584e517SRavikiran G Thirumalai } 1484353d5c30SHugh Dickins } 14851da177e4SLinus Torvalds 148639b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 1487e68375c8SAl Viro inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0); 14881da177e4SLinus Torvalds if (!inode) 1489e68375c8SAl Viro goto out; 1490e1832f29SStephen Smalley if (creat_flags == HUGETLB_SHMFS_INODE) 1491e1832f29SStephen Smalley inode->i_flags |= S_PRIVATE; 14921da177e4SLinus Torvalds 14931da177e4SLinus Torvalds inode->i_size = size; 14946d6b77f1SMiklos Szeredi clear_nlink(inode); 1495ce8d2cdfSDave Hansen 1496*33b8f84aSMike Kravetz if (!hugetlb_reserve_pages(inode, 0, 1497e68375c8SAl Viro size >> huge_page_shift(hstate_inode(inode)), NULL, 1498e68375c8SAl Viro acctflag)) 1499e68375c8SAl Viro file = ERR_PTR(-ENOMEM); 1500e68375c8SAl Viro else 1501e68375c8SAl Viro file = alloc_file_pseudo(inode, mnt, name, O_RDWR, 1502ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 1503e68375c8SAl Viro if (!IS_ERR(file)) 15041da177e4SLinus Torvalds return file; 15051da177e4SLinus Torvalds 1506b45b5bd6SDavid Gibson iput(inode); 1507e68375c8SAl Viro out: 1508353d5c30SHugh Dickins if (*user) { 1509353d5c30SHugh Dickins user_shm_unlock(size, *user); 1510353d5c30SHugh Dickins *user = NULL; 1511353d5c30SHugh Dickins } 151239b65252SAnatol Pomozov return file; 15131da177e4SLinus Torvalds } 15141da177e4SLinus Torvalds 151532021982SDavid Howells static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h) 151632021982SDavid Howells { 151732021982SDavid Howells struct fs_context *fc; 151832021982SDavid Howells struct vfsmount *mnt; 151932021982SDavid Howells 152032021982SDavid Howells fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT); 152132021982SDavid Howells if (IS_ERR(fc)) { 152232021982SDavid Howells mnt = ERR_CAST(fc); 152332021982SDavid Howells } else { 152432021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 152532021982SDavid Howells ctx->hstate = h; 152632021982SDavid Howells mnt = fc_mount(fc); 152732021982SDavid Howells put_fs_context(fc); 152832021982SDavid Howells } 152932021982SDavid Howells if (IS_ERR(mnt)) 153032021982SDavid Howells pr_err("Cannot mount internal hugetlbfs for page size %uK", 153132021982SDavid Howells 1U << (h->order + PAGE_SHIFT - 10)); 153232021982SDavid Howells return mnt; 153332021982SDavid Howells } 153432021982SDavid Howells 15351da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 15361da177e4SLinus Torvalds { 153732021982SDavid Howells struct vfsmount *mnt; 153842d7395fSAndi Kleen struct hstate *h; 15391da177e4SLinus Torvalds int error; 154042d7395fSAndi Kleen int i; 15411da177e4SLinus Torvalds 1542457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 15439b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1544457c1b27SNishanth Aravamudan return -ENOTSUPP; 1545457c1b27SNishanth Aravamudan } 1546457c1b27SNishanth Aravamudan 1547d1d5e05fSHillf Danton error = -ENOMEM; 15481da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 15491da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 15505d097056SVladimir Davydov 0, SLAB_ACCOUNT, init_once); 15511da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 15528fc312b3SMike Kravetz goto out; 15531da177e4SLinus Torvalds 15541da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 15551da177e4SLinus Torvalds if (error) 15568fc312b3SMike Kravetz goto out_free; 15571da177e4SLinus Torvalds 15588fc312b3SMike Kravetz /* default hstate mount is required */ 15598fc312b3SMike Kravetz mnt = mount_one_hugetlbfs(&hstates[default_hstate_idx]); 15608fc312b3SMike Kravetz if (IS_ERR(mnt)) { 15618fc312b3SMike Kravetz error = PTR_ERR(mnt); 15628fc312b3SMike Kravetz goto out_unreg; 15638fc312b3SMike Kravetz } 15648fc312b3SMike Kravetz hugetlbfs_vfsmount[default_hstate_idx] = mnt; 15658fc312b3SMike Kravetz 15668fc312b3SMike Kravetz /* other hstates are optional */ 156742d7395fSAndi Kleen i = 0; 156842d7395fSAndi Kleen for_each_hstate(h) { 156915f0ec94SJan Stancek if (i == default_hstate_idx) { 157015f0ec94SJan Stancek i++; 15718fc312b3SMike Kravetz continue; 157215f0ec94SJan Stancek } 15738fc312b3SMike Kravetz 157432021982SDavid Howells mnt = mount_one_hugetlbfs(h); 15758fc312b3SMike Kravetz if (IS_ERR(mnt)) 15768fc312b3SMike Kravetz hugetlbfs_vfsmount[i] = NULL; 15778fc312b3SMike Kravetz else 157832021982SDavid Howells hugetlbfs_vfsmount[i] = mnt; 157942d7395fSAndi Kleen i++; 158042d7395fSAndi Kleen } 158132021982SDavid Howells 158242d7395fSAndi Kleen return 0; 15831da177e4SLinus Torvalds 15848fc312b3SMike Kravetz out_unreg: 15858fc312b3SMike Kravetz (void)unregister_filesystem(&hugetlbfs_fs_type); 15868fc312b3SMike Kravetz out_free: 15871da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 15888fc312b3SMike Kravetz out: 15891da177e4SLinus Torvalds return error; 15901da177e4SLinus Torvalds } 15913e89e1c5SPaul Gortmaker fs_initcall(init_hugetlbfs_fs) 1592