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; 17433b8f84aSMike 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 7642f221d6fSChristian Brauner error = setattr_prepare(&init_user_ns, dentry, attr); 7651da177e4SLinus Torvalds if (error) 7661025774cSChristoph Hellwig return error; 7671da177e4SLinus Torvalds 7681da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 769ff62a342SMarc-André Lureau loff_t oldsize = inode->i_size; 770ff62a342SMarc-André Lureau loff_t newsize = attr->ia_size; 771ff62a342SMarc-André Lureau 772ff62a342SMarc-André Lureau if (newsize & ~huge_page_mask(h)) 7731025774cSChristoph Hellwig return -EINVAL; 774ff62a342SMarc-André Lureau /* protected by i_mutex */ 775ff62a342SMarc-André Lureau if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || 776ff62a342SMarc-André Lureau (newsize > oldsize && (info->seals & F_SEAL_GROW))) 777ff62a342SMarc-André Lureau return -EPERM; 778ff62a342SMarc-André Lureau error = hugetlb_vmtruncate(inode, newsize); 7791da177e4SLinus Torvalds if (error) 7801da177e4SLinus Torvalds return error; 7811da177e4SLinus Torvalds } 7821da177e4SLinus Torvalds 7832f221d6fSChristian Brauner setattr_copy(&init_user_ns, inode, attr); 7841025774cSChristoph Hellwig mark_inode_dirty(inode); 7851025774cSChristoph Hellwig return 0; 7861025774cSChristoph Hellwig } 7871025774cSChristoph Hellwig 7887d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 78932021982SDavid Howells struct hugetlbfs_fs_context *ctx) 7901da177e4SLinus Torvalds { 7911da177e4SLinus Torvalds struct inode *inode; 7921da177e4SLinus Torvalds 7931da177e4SLinus Torvalds inode = new_inode(sb); 7941da177e4SLinus Torvalds if (inode) { 79585fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 79632021982SDavid Howells inode->i_mode = S_IFDIR | ctx->mode; 79732021982SDavid Howells inode->i_uid = ctx->uid; 79832021982SDavid Howells inode->i_gid = ctx->gid; 799078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 8007d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 8017d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 8027d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 8037d54fa64SAl Viro inc_nlink(inode); 80465ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 8057d54fa64SAl Viro } 8067d54fa64SAl Viro return inode; 8077d54fa64SAl Viro } 8087d54fa64SAl Viro 809b610ded7SMichal Hocko /* 810c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 811b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 81288f306b6SKirill A. Shutemov * annotation because huge_pmd_share() does an allocation under hugetlb's 813c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 814b610ded7SMichal Hocko */ 815c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 816b610ded7SMichal Hocko 8177d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 8187d54fa64SAl Viro struct inode *dir, 81918df2252SAl Viro umode_t mode, dev_t dev) 8207d54fa64SAl Viro { 8217d54fa64SAl Viro struct inode *inode; 82258b6e5e8SMike Kravetz struct resv_map *resv_map = NULL; 8239119a41eSJoonsoo Kim 82458b6e5e8SMike Kravetz /* 82558b6e5e8SMike Kravetz * Reserve maps are only needed for inodes that can have associated 82658b6e5e8SMike Kravetz * page allocations. 82758b6e5e8SMike Kravetz */ 82858b6e5e8SMike Kravetz if (S_ISREG(mode) || S_ISLNK(mode)) { 8299119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 8309119a41eSJoonsoo Kim if (!resv_map) 8319119a41eSJoonsoo Kim return NULL; 83258b6e5e8SMike Kravetz } 8337d54fa64SAl Viro 8347d54fa64SAl Viro inode = new_inode(sb); 8357d54fa64SAl Viro if (inode) { 836ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 837ff62a342SMarc-André Lureau 8387d54fa64SAl Viro inode->i_ino = get_next_ino(); 83921cb47beSChristian Brauner inode_init_owner(&init_user_ns, inode, dir, mode); 840c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 841c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 8421da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 843078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 8449119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 845ff62a342SMarc-André Lureau info->seals = F_SEAL_SEAL; 8461da177e4SLinus Torvalds switch (mode & S_IFMT) { 8471da177e4SLinus Torvalds default: 8481da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 8491da177e4SLinus Torvalds break; 8501da177e4SLinus Torvalds case S_IFREG: 8511da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 8521da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 8531da177e4SLinus Torvalds break; 8541da177e4SLinus Torvalds case S_IFDIR: 8551da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 8561da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 8571da177e4SLinus Torvalds 8581da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 859d8c76e6fSDave Hansen inc_nlink(inode); 8601da177e4SLinus Torvalds break; 8611da177e4SLinus Torvalds case S_IFLNK: 8621da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 86321fc61c7SAl Viro inode_nohighmem(inode); 8641da177e4SLinus Torvalds break; 8651da177e4SLinus Torvalds } 866e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 86758b6e5e8SMike Kravetz } else { 86858b6e5e8SMike Kravetz if (resv_map) 8699119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 87058b6e5e8SMike Kravetz } 8719119a41eSJoonsoo Kim 8721da177e4SLinus Torvalds return inode; 8731da177e4SLinus Torvalds } 8741da177e4SLinus Torvalds 8751da177e4SLinus Torvalds /* 8761da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 8771da177e4SLinus Torvalds */ 8781ab5b82fSPiotr Sarna static int do_hugetlbfs_mknod(struct inode *dir, 8791ab5b82fSPiotr Sarna struct dentry *dentry, 8801ab5b82fSPiotr Sarna umode_t mode, 8811ab5b82fSPiotr Sarna dev_t dev, 8821ab5b82fSPiotr Sarna bool tmpfile) 8831da177e4SLinus Torvalds { 8841da177e4SLinus Torvalds struct inode *inode; 8851da177e4SLinus Torvalds int error = -ENOSPC; 8861da177e4SLinus Torvalds 8877d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 8881da177e4SLinus Torvalds if (inode) { 889078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 8901ab5b82fSPiotr Sarna if (tmpfile) { 8911ab5b82fSPiotr Sarna d_tmpfile(dentry, inode); 8921ab5b82fSPiotr Sarna } else { 8931da177e4SLinus Torvalds d_instantiate(dentry, inode); 8941da177e4SLinus Torvalds dget(dentry);/* Extra count - pin the dentry in core */ 8951ab5b82fSPiotr Sarna } 8961da177e4SLinus Torvalds error = 0; 8971da177e4SLinus Torvalds } 8981da177e4SLinus Torvalds return error; 8991da177e4SLinus Torvalds } 9001da177e4SLinus Torvalds 901549c7297SChristian Brauner static int hugetlbfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 9021ab5b82fSPiotr Sarna struct dentry *dentry, umode_t mode, dev_t dev) 9031ab5b82fSPiotr Sarna { 9041ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode, dev, false); 9051ab5b82fSPiotr Sarna } 9061ab5b82fSPiotr Sarna 907549c7297SChristian Brauner static int hugetlbfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 908549c7297SChristian Brauner struct dentry *dentry, umode_t mode) 9091da177e4SLinus Torvalds { 910549c7297SChristian Brauner int retval = hugetlbfs_mknod(&init_user_ns, dir, dentry, 911549c7297SChristian Brauner mode | S_IFDIR, 0); 9121da177e4SLinus Torvalds if (!retval) 913d8c76e6fSDave Hansen inc_nlink(dir); 9141da177e4SLinus Torvalds return retval; 9151da177e4SLinus Torvalds } 9161da177e4SLinus Torvalds 917549c7297SChristian Brauner static int hugetlbfs_create(struct user_namespace *mnt_userns, 918549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 919549c7297SChristian Brauner umode_t mode, bool excl) 9201da177e4SLinus Torvalds { 921549c7297SChristian Brauner return hugetlbfs_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0); 9221da177e4SLinus Torvalds } 9231da177e4SLinus Torvalds 924549c7297SChristian Brauner static int hugetlbfs_tmpfile(struct user_namespace *mnt_userns, 925549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 926549c7297SChristian Brauner umode_t mode) 9271ab5b82fSPiotr Sarna { 9281ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0, true); 9291ab5b82fSPiotr Sarna } 9301ab5b82fSPiotr Sarna 931549c7297SChristian Brauner static int hugetlbfs_symlink(struct user_namespace *mnt_userns, 932549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 933549c7297SChristian Brauner const char *symname) 9341da177e4SLinus Torvalds { 9351da177e4SLinus Torvalds struct inode *inode; 9361da177e4SLinus Torvalds int error = -ENOSPC; 9371da177e4SLinus Torvalds 9387d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 9391da177e4SLinus Torvalds if (inode) { 9401da177e4SLinus Torvalds int l = strlen(symname)+1; 9411da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 9421da177e4SLinus Torvalds if (!error) { 9431da177e4SLinus Torvalds d_instantiate(dentry, inode); 9441da177e4SLinus Torvalds dget(dentry); 9451da177e4SLinus Torvalds } else 9461da177e4SLinus Torvalds iput(inode); 9471da177e4SLinus Torvalds } 948078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 9491da177e4SLinus Torvalds 9501da177e4SLinus Torvalds return error; 9511da177e4SLinus Torvalds } 9521da177e4SLinus Torvalds 953290408d4SNaoya Horiguchi static int hugetlbfs_migrate_page(struct address_space *mapping, 954b969c4abSMel Gorman struct page *newpage, struct page *page, 955a6bc32b8SMel Gorman enum migrate_mode mode) 956290408d4SNaoya Horiguchi { 957290408d4SNaoya Horiguchi int rc; 958290408d4SNaoya Horiguchi 959290408d4SNaoya Horiguchi rc = migrate_huge_page_move_mapping(mapping, newpage, page); 96078bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 961290408d4SNaoya Horiguchi return rc; 962cb6acd01SMike Kravetz 963d6995da3SMike Kravetz if (hugetlb_page_subpool(page)) { 964d6995da3SMike Kravetz hugetlb_set_page_subpool(newpage, hugetlb_page_subpool(page)); 965d6995da3SMike Kravetz hugetlb_set_page_subpool(page, NULL); 966cb6acd01SMike Kravetz } 967cb6acd01SMike Kravetz 9682916ecc0SJérôme Glisse if (mode != MIGRATE_SYNC_NO_COPY) 969290408d4SNaoya Horiguchi migrate_page_copy(newpage, page); 9702916ecc0SJérôme Glisse else 9712916ecc0SJérôme Glisse migrate_page_states(newpage, page); 972290408d4SNaoya Horiguchi 97378bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 974290408d4SNaoya Horiguchi } 975290408d4SNaoya Horiguchi 97678bb9203SNaoya Horiguchi static int hugetlbfs_error_remove_page(struct address_space *mapping, 97778bb9203SNaoya Horiguchi struct page *page) 97878bb9203SNaoya Horiguchi { 97978bb9203SNaoya Horiguchi struct inode *inode = mapping->host; 980ab615a5bSMike Kravetz pgoff_t index = page->index; 98178bb9203SNaoya Horiguchi 98278bb9203SNaoya Horiguchi remove_huge_page(page); 983ab615a5bSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) 98478bb9203SNaoya Horiguchi hugetlb_fix_reserve_counts(inode); 985ab615a5bSMike Kravetz 98678bb9203SNaoya Horiguchi return 0; 98778bb9203SNaoya Horiguchi } 98878bb9203SNaoya Horiguchi 9894a25220dSDavid Howells /* 9904a25220dSDavid Howells * Display the mount options in /proc/mounts. 9914a25220dSDavid Howells */ 9924a25220dSDavid Howells static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) 9934a25220dSDavid Howells { 9944a25220dSDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); 9954a25220dSDavid Howells struct hugepage_subpool *spool = sbinfo->spool; 9964a25220dSDavid Howells unsigned long hpage_size = huge_page_size(sbinfo->hstate); 9974a25220dSDavid Howells unsigned hpage_shift = huge_page_shift(sbinfo->hstate); 9984a25220dSDavid Howells char mod; 9994a25220dSDavid Howells 10004a25220dSDavid Howells if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) 10014a25220dSDavid Howells seq_printf(m, ",uid=%u", 10024a25220dSDavid Howells from_kuid_munged(&init_user_ns, sbinfo->uid)); 10034a25220dSDavid Howells if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) 10044a25220dSDavid Howells seq_printf(m, ",gid=%u", 10054a25220dSDavid Howells from_kgid_munged(&init_user_ns, sbinfo->gid)); 10064a25220dSDavid Howells if (sbinfo->mode != 0755) 10074a25220dSDavid Howells seq_printf(m, ",mode=%o", sbinfo->mode); 10084a25220dSDavid Howells if (sbinfo->max_inodes != -1) 10094a25220dSDavid Howells seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); 10104a25220dSDavid Howells 10114a25220dSDavid Howells hpage_size /= 1024; 10124a25220dSDavid Howells mod = 'K'; 10134a25220dSDavid Howells if (hpage_size >= 1024) { 10144a25220dSDavid Howells hpage_size /= 1024; 10154a25220dSDavid Howells mod = 'M'; 10164a25220dSDavid Howells } 10174a25220dSDavid Howells seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); 10184a25220dSDavid Howells if (spool) { 10194a25220dSDavid Howells if (spool->max_hpages != -1) 10204a25220dSDavid Howells seq_printf(m, ",size=%llu", 10214a25220dSDavid Howells (unsigned long long)spool->max_hpages << hpage_shift); 10224a25220dSDavid Howells if (spool->min_hpages != -1) 10234a25220dSDavid Howells seq_printf(m, ",min_size=%llu", 10244a25220dSDavid Howells (unsigned long long)spool->min_hpages << hpage_shift); 10254a25220dSDavid Howells } 10264a25220dSDavid Howells return 0; 10274a25220dSDavid Howells } 10284a25220dSDavid Howells 1029726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 10301da177e4SLinus Torvalds { 1031726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 10322b0143b5SDavid Howells struct hstate *h = hstate_inode(d_inode(dentry)); 10331da177e4SLinus Torvalds 10341da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 1035a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 10361da177e4SLinus Torvalds if (sbinfo) { 10371da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 103874a8a65cSDavid Gibson /* If no limits set, just report 0 for max/free/used 103974a8a65cSDavid Gibson * blocks, like simple_statfs() */ 104090481622SDavid Gibson if (sbinfo->spool) { 104190481622SDavid Gibson long free_pages; 104290481622SDavid Gibson 104390481622SDavid Gibson spin_lock(&sbinfo->spool->lock); 104490481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 104590481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 104690481622SDavid Gibson - sbinfo->spool->used_hpages; 104790481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 104890481622SDavid Gibson spin_unlock(&sbinfo->spool->lock); 10491da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 10501da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 105174a8a65cSDavid Gibson } 10521da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 10531da177e4SLinus Torvalds } 10541da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 10551da177e4SLinus Torvalds return 0; 10561da177e4SLinus Torvalds } 10571da177e4SLinus Torvalds 10581da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 10591da177e4SLinus Torvalds { 10601da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 10611da177e4SLinus Torvalds 10621da177e4SLinus Torvalds if (sbi) { 10631da177e4SLinus Torvalds sb->s_fs_info = NULL; 106490481622SDavid Gibson 106590481622SDavid Gibson if (sbi->spool) 106690481622SDavid Gibson hugepage_put_subpool(sbi->spool); 106790481622SDavid Gibson 10681da177e4SLinus Torvalds kfree(sbi); 10691da177e4SLinus Torvalds } 10701da177e4SLinus Torvalds } 10711da177e4SLinus Torvalds 107296527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 107396527980SChristoph Hellwig { 107496527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 107596527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 107696527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 107796527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 107896527980SChristoph Hellwig return 0; 107996527980SChristoph Hellwig } 108096527980SChristoph Hellwig sbinfo->free_inodes--; 108196527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 108296527980SChristoph Hellwig } 108396527980SChristoph Hellwig 108496527980SChristoph Hellwig return 1; 108596527980SChristoph Hellwig } 108696527980SChristoph Hellwig 108796527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 108896527980SChristoph Hellwig { 108996527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 109096527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 109196527980SChristoph Hellwig sbinfo->free_inodes++; 109296527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 109396527980SChristoph Hellwig } 109496527980SChristoph Hellwig } 109596527980SChristoph Hellwig 109696527980SChristoph Hellwig 1097e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 10981da177e4SLinus Torvalds 10991da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 11001da177e4SLinus Torvalds { 110196527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 11021da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 11031da177e4SLinus Torvalds 110496527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 11051da177e4SLinus Torvalds return NULL; 1106e94b1766SChristoph Lameter p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); 110796527980SChristoph Hellwig if (unlikely(!p)) { 110896527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 110996527980SChristoph Hellwig return NULL; 11101da177e4SLinus Torvalds } 11114742a35dSMike Kravetz 11124742a35dSMike Kravetz /* 11134742a35dSMike Kravetz * Any time after allocation, hugetlbfs_destroy_inode can be called 11144742a35dSMike Kravetz * for the inode. mpol_free_shared_policy is unconditionally called 11154742a35dSMike Kravetz * as part of hugetlbfs_destroy_inode. So, initialize policy here 11164742a35dSMike Kravetz * in case of a quick call to destroy. 11174742a35dSMike Kravetz * 11184742a35dSMike Kravetz * Note that the policy is initialized even if we are creating a 11194742a35dSMike Kravetz * private inode. This simplifies hugetlbfs_destroy_inode. 11204742a35dSMike Kravetz */ 11214742a35dSMike Kravetz mpol_shared_policy_init(&p->policy, NULL); 11224742a35dSMike Kravetz 112396527980SChristoph Hellwig return &p->vfs_inode; 11241da177e4SLinus Torvalds } 11251da177e4SLinus Torvalds 1126b62de322SAl Viro static void hugetlbfs_free_inode(struct inode *inode) 1127fa0d7e3dSNick Piggin { 1128fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 1129fa0d7e3dSNick Piggin } 1130fa0d7e3dSNick Piggin 11311da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 11321da177e4SLinus Torvalds { 113396527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 11341da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 11351da177e4SLinus Torvalds } 11361da177e4SLinus Torvalds 1137f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 1138800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 1139800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 1140a4fa34cdSMike Kravetz .set_page_dirty = __set_page_dirty_no_writeback, 1141290408d4SNaoya Horiguchi .migratepage = hugetlbfs_migrate_page, 114278bb9203SNaoya Horiguchi .error_remove_page = hugetlbfs_error_remove_page, 11431da177e4SLinus Torvalds }; 11441da177e4SLinus Torvalds 114596527980SChristoph Hellwig 114651cc5068SAlexey Dobriyan static void init_once(void *foo) 114796527980SChristoph Hellwig { 114896527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 114996527980SChristoph Hellwig 115096527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 115196527980SChristoph Hellwig } 115296527980SChristoph Hellwig 11534b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 115434d0640eSAl Viro .read_iter = hugetlbfs_read_iter, 11551da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 11561b061d92SChristoph Hellwig .fsync = noop_fsync, 11571da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 11586038f373SArnd Bergmann .llseek = default_llseek, 115970c3547eSMike Kravetz .fallocate = hugetlbfs_fallocate, 11601da177e4SLinus Torvalds }; 11611da177e4SLinus Torvalds 116292e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 11631da177e4SLinus Torvalds .create = hugetlbfs_create, 11641da177e4SLinus Torvalds .lookup = simple_lookup, 11651da177e4SLinus Torvalds .link = simple_link, 11661da177e4SLinus Torvalds .unlink = simple_unlink, 11671da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 11681da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 11691da177e4SLinus Torvalds .rmdir = simple_rmdir, 11701da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 11711da177e4SLinus Torvalds .rename = simple_rename, 11721da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 11731ab5b82fSPiotr Sarna .tmpfile = hugetlbfs_tmpfile, 11741da177e4SLinus Torvalds }; 11751da177e4SLinus Torvalds 117692e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 11771da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 11781da177e4SLinus Torvalds }; 11791da177e4SLinus Torvalds 1180ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 11811da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 1182b62de322SAl Viro .free_inode = hugetlbfs_free_inode, 11831da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 11842bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 11851da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 11861da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 11874a25220dSDavid Howells .show_options = hugetlbfs_show_options, 11881da177e4SLinus Torvalds }; 11891da177e4SLinus Torvalds 11907ca02d0aSMike Kravetz /* 11917ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 11927ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 11937ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 11947ca02d0aSMike Kravetz */ 11954a25220dSDavid Howells static long 11967ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 11974a25220dSDavid Howells enum hugetlbfs_size_type val_type) 11987ca02d0aSMike Kravetz { 11997ca02d0aSMike Kravetz if (val_type == NO_SIZE) 12007ca02d0aSMike Kravetz return -1; 12017ca02d0aSMike Kravetz 12027ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 12037ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 12047ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 12057ca02d0aSMike Kravetz do_div(size_opt, 100); 12067ca02d0aSMike Kravetz } 12077ca02d0aSMike Kravetz 12087ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 12097ca02d0aSMike Kravetz return size_opt; 12107ca02d0aSMike Kravetz } 12117ca02d0aSMike Kravetz 121232021982SDavid Howells /* 121332021982SDavid Howells * Parse one mount parameter. 121432021982SDavid Howells */ 121532021982SDavid Howells static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param) 12161da177e4SLinus Torvalds { 121732021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 121832021982SDavid Howells struct fs_parse_result result; 121932021982SDavid Howells char *rest; 122032021982SDavid Howells unsigned long ps; 122132021982SDavid Howells int opt; 12221da177e4SLinus Torvalds 1223d7167b14SAl Viro opt = fs_parse(fc, hugetlb_fs_parameters, param, &result); 122432021982SDavid Howells if (opt < 0) 122532021982SDavid Howells return opt; 122632021982SDavid Howells 122732021982SDavid Howells switch (opt) { 122832021982SDavid Howells case Opt_uid: 122932021982SDavid Howells ctx->uid = make_kuid(current_user_ns(), result.uint_32); 123032021982SDavid Howells if (!uid_valid(ctx->uid)) 123132021982SDavid Howells goto bad_val; 12321da177e4SLinus Torvalds return 0; 12331da177e4SLinus Torvalds 1234e73a75faSRandy Dunlap case Opt_gid: 123532021982SDavid Howells ctx->gid = make_kgid(current_user_ns(), result.uint_32); 123632021982SDavid Howells if (!gid_valid(ctx->gid)) 1237e73a75faSRandy Dunlap goto bad_val; 123832021982SDavid Howells return 0; 1239e73a75faSRandy Dunlap 1240e73a75faSRandy Dunlap case Opt_mode: 124132021982SDavid Howells ctx->mode = result.uint_32 & 01777U; 124232021982SDavid Howells return 0; 1243e73a75faSRandy Dunlap 124432021982SDavid Howells case Opt_size: 1245e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 124632021982SDavid Howells if (!isdigit(param->string[0])) 1247e73a75faSRandy Dunlap goto bad_val; 124832021982SDavid Howells ctx->max_size_opt = memparse(param->string, &rest); 124932021982SDavid Howells ctx->max_val_type = SIZE_STD; 1250a137e1ccSAndi Kleen if (*rest == '%') 125132021982SDavid Howells ctx->max_val_type = SIZE_PERCENT; 125232021982SDavid Howells return 0; 12531da177e4SLinus Torvalds 1254e73a75faSRandy Dunlap case Opt_nr_inodes: 1255e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 125632021982SDavid Howells if (!isdigit(param->string[0])) 1257e73a75faSRandy Dunlap goto bad_val; 125832021982SDavid Howells ctx->nr_inodes = memparse(param->string, &rest); 125932021982SDavid Howells return 0; 1260e73a75faSRandy Dunlap 126132021982SDavid Howells case Opt_pagesize: 126232021982SDavid Howells ps = memparse(param->string, &rest); 126332021982SDavid Howells ctx->hstate = size_to_hstate(ps); 126432021982SDavid Howells if (!ctx->hstate) { 126532021982SDavid Howells pr_err("Unsupported page size %lu MB\n", ps >> 20); 1266a137e1ccSAndi Kleen return -EINVAL; 1267a137e1ccSAndi Kleen } 126832021982SDavid Howells return 0; 1269a137e1ccSAndi Kleen 127032021982SDavid Howells case Opt_min_size: 12717ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 127232021982SDavid Howells if (!isdigit(param->string[0])) 12737ca02d0aSMike Kravetz goto bad_val; 127432021982SDavid Howells ctx->min_size_opt = memparse(param->string, &rest); 127532021982SDavid Howells ctx->min_val_type = SIZE_STD; 12767ca02d0aSMike Kravetz if (*rest == '%') 127732021982SDavid Howells ctx->min_val_type = SIZE_PERCENT; 127832021982SDavid Howells return 0; 12797ca02d0aSMike Kravetz 1280e73a75faSRandy Dunlap default: 1281b4c07bceSLee Schermerhorn return -EINVAL; 1282e73a75faSRandy Dunlap } 128332021982SDavid Howells 128432021982SDavid Howells bad_val: 1285b5db30cfSAl Viro return invalfc(fc, "Bad value '%s' for mount option '%s'\n", 128632021982SDavid Howells param->string, param->key); 12871da177e4SLinus Torvalds } 1288a137e1ccSAndi Kleen 12897ca02d0aSMike Kravetz /* 129032021982SDavid Howells * Validate the parsed options. 129132021982SDavid Howells */ 129232021982SDavid Howells static int hugetlbfs_validate(struct fs_context *fc) 129332021982SDavid Howells { 129432021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 129532021982SDavid Howells 129632021982SDavid Howells /* 12977ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 12987ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 12997ca02d0aSMike Kravetz */ 130032021982SDavid Howells ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 130132021982SDavid Howells ctx->max_size_opt, 130232021982SDavid Howells ctx->max_val_type); 130332021982SDavid Howells ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 130432021982SDavid Howells ctx->min_size_opt, 130532021982SDavid Howells ctx->min_val_type); 13067ca02d0aSMike Kravetz 13077ca02d0aSMike Kravetz /* 13087ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 13097ca02d0aSMike Kravetz */ 131032021982SDavid Howells if (ctx->max_val_type > NO_SIZE && 131132021982SDavid Howells ctx->min_hpages > ctx->max_hpages) { 131232021982SDavid Howells pr_err("Minimum size can not be greater than maximum size\n"); 13137ca02d0aSMike Kravetz return -EINVAL; 1314a137e1ccSAndi Kleen } 1315a137e1ccSAndi Kleen 13161da177e4SLinus Torvalds return 0; 13171da177e4SLinus Torvalds } 13181da177e4SLinus Torvalds 13191da177e4SLinus Torvalds static int 132032021982SDavid Howells hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc) 13211da177e4SLinus Torvalds { 132232021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 13231da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 13241da177e4SLinus Torvalds 13251da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 13261da177e4SLinus Torvalds if (!sbinfo) 13271da177e4SLinus Torvalds return -ENOMEM; 13281da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 13291da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 133032021982SDavid Howells sbinfo->hstate = ctx->hstate; 133132021982SDavid Howells sbinfo->max_inodes = ctx->nr_inodes; 133232021982SDavid Howells sbinfo->free_inodes = ctx->nr_inodes; 133390481622SDavid Gibson sbinfo->spool = NULL; 133432021982SDavid Howells sbinfo->uid = ctx->uid; 133532021982SDavid Howells sbinfo->gid = ctx->gid; 133632021982SDavid Howells sbinfo->mode = ctx->mode; 13374a25220dSDavid Howells 13387ca02d0aSMike Kravetz /* 13397ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 13407ca02d0aSMike Kravetz * specified. Any needed reservations (for minimim size) are taken 13417ca02d0aSMike Kravetz * taken when the subpool is created. 13427ca02d0aSMike Kravetz */ 134332021982SDavid Howells if (ctx->max_hpages != -1 || ctx->min_hpages != -1) { 134432021982SDavid Howells sbinfo->spool = hugepage_new_subpool(ctx->hstate, 134532021982SDavid Howells ctx->max_hpages, 134632021982SDavid Howells ctx->min_hpages); 134790481622SDavid Gibson if (!sbinfo->spool) 134890481622SDavid Gibson goto out_free; 134990481622SDavid Gibson } 13501da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 135132021982SDavid Howells sb->s_blocksize = huge_page_size(ctx->hstate); 135232021982SDavid Howells sb->s_blocksize_bits = huge_page_shift(ctx->hstate); 13531da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 13541da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 13551da177e4SLinus Torvalds sb->s_time_gran = 1; 135615568299SMike Kravetz 135715568299SMike Kravetz /* 135815568299SMike Kravetz * Due to the special and limited functionality of hugetlbfs, it does 135915568299SMike Kravetz * not work well as a stacking filesystem. 136015568299SMike Kravetz */ 136115568299SMike Kravetz sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH; 136232021982SDavid Howells sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx)); 136348fde701SAl Viro if (!sb->s_root) 13641da177e4SLinus Torvalds goto out_free; 13651da177e4SLinus Torvalds return 0; 13661da177e4SLinus Torvalds out_free: 136790481622SDavid Gibson kfree(sbinfo->spool); 13681da177e4SLinus Torvalds kfree(sbinfo); 13691da177e4SLinus Torvalds return -ENOMEM; 13701da177e4SLinus Torvalds } 13711da177e4SLinus Torvalds 137232021982SDavid Howells static int hugetlbfs_get_tree(struct fs_context *fc) 13731da177e4SLinus Torvalds { 137432021982SDavid Howells int err = hugetlbfs_validate(fc); 137532021982SDavid Howells if (err) 137632021982SDavid Howells return err; 13772ac295d4SAl Viro return get_tree_nodev(fc, hugetlbfs_fill_super); 137832021982SDavid Howells } 137932021982SDavid Howells 138032021982SDavid Howells static void hugetlbfs_fs_context_free(struct fs_context *fc) 138132021982SDavid Howells { 138232021982SDavid Howells kfree(fc->fs_private); 138332021982SDavid Howells } 138432021982SDavid Howells 138532021982SDavid Howells static const struct fs_context_operations hugetlbfs_fs_context_ops = { 138632021982SDavid Howells .free = hugetlbfs_fs_context_free, 138732021982SDavid Howells .parse_param = hugetlbfs_parse_param, 138832021982SDavid Howells .get_tree = hugetlbfs_get_tree, 138932021982SDavid Howells }; 139032021982SDavid Howells 139132021982SDavid Howells static int hugetlbfs_init_fs_context(struct fs_context *fc) 139232021982SDavid Howells { 139332021982SDavid Howells struct hugetlbfs_fs_context *ctx; 139432021982SDavid Howells 139532021982SDavid Howells ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL); 139632021982SDavid Howells if (!ctx) 139732021982SDavid Howells return -ENOMEM; 139832021982SDavid Howells 139932021982SDavid Howells ctx->max_hpages = -1; /* No limit on size by default */ 140032021982SDavid Howells ctx->nr_inodes = -1; /* No limit on number of inodes by default */ 140132021982SDavid Howells ctx->uid = current_fsuid(); 140232021982SDavid Howells ctx->gid = current_fsgid(); 140332021982SDavid Howells ctx->mode = 0755; 140432021982SDavid Howells ctx->hstate = &default_hstate; 140532021982SDavid Howells ctx->min_hpages = -1; /* No default minimum size */ 140632021982SDavid Howells ctx->max_val_type = NO_SIZE; 140732021982SDavid Howells ctx->min_val_type = NO_SIZE; 140832021982SDavid Howells fc->fs_private = ctx; 140932021982SDavid Howells fc->ops = &hugetlbfs_fs_context_ops; 141032021982SDavid Howells return 0; 14111da177e4SLinus Torvalds } 14121da177e4SLinus Torvalds 14131da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 14141da177e4SLinus Torvalds .name = "hugetlbfs", 141532021982SDavid Howells .init_fs_context = hugetlbfs_init_fs_context, 1416d7167b14SAl Viro .parameters = hugetlb_fs_parameters, 14171da177e4SLinus Torvalds .kill_sb = kill_litter_super, 14181da177e4SLinus Torvalds }; 14191da177e4SLinus Torvalds 142042d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 14211da177e4SLinus Torvalds 1422ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 14231da177e4SLinus Torvalds { 1424a0eb3a05SEric W. Biederman kgid_t shm_group; 1425a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 1426a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 14271da177e4SLinus Torvalds } 14281da177e4SLinus Torvalds 142942d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 143042d7395fSAndi Kleen { 1431af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 143242d7395fSAndi Kleen 143342d7395fSAndi Kleen if (!h) 143442d7395fSAndi Kleen return -1; 143542d7395fSAndi Kleen return h - hstates; 143642d7395fSAndi Kleen } 143742d7395fSAndi Kleen 1438af73e4d9SNaoya Horiguchi /* 1439af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 1440af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 1441af73e4d9SNaoya Horiguchi */ 1442af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 1443af73e4d9SNaoya Horiguchi vm_flags_t acctflag, struct user_struct **user, 144442d7395fSAndi Kleen int creat_flags, int page_size_log) 14451da177e4SLinus Torvalds { 14461da177e4SLinus Torvalds struct inode *inode; 1447e68375c8SAl Viro struct vfsmount *mnt; 144842d7395fSAndi Kleen int hstate_idx; 1449e68375c8SAl Viro struct file *file; 145042d7395fSAndi Kleen 145142d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 145242d7395fSAndi Kleen if (hstate_idx < 0) 145342d7395fSAndi Kleen return ERR_PTR(-ENODEV); 14541da177e4SLinus Torvalds 1455353d5c30SHugh Dickins *user = NULL; 1456e68375c8SAl Viro mnt = hugetlbfs_vfsmount[hstate_idx]; 1457e68375c8SAl Viro if (!mnt) 14585bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 14595bc98594SAkinobu Mita 1460ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 1461353d5c30SHugh Dickins *user = current_user(); 1462353d5c30SHugh Dickins if (user_shm_lock(size, *user)) { 146321a3c273SDavid Rientjes task_lock(current); 14649b857d26SAndrew Morton pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", 146521a3c273SDavid Rientjes current->comm, current->pid); 146621a3c273SDavid Rientjes task_unlock(current); 1467353d5c30SHugh Dickins } else { 1468353d5c30SHugh Dickins *user = NULL; 14691da177e4SLinus Torvalds return ERR_PTR(-EPERM); 14702584e517SRavikiran G Thirumalai } 1471353d5c30SHugh Dickins } 14721da177e4SLinus Torvalds 147339b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 1474e68375c8SAl Viro inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0); 14751da177e4SLinus Torvalds if (!inode) 1476e68375c8SAl Viro goto out; 1477e1832f29SStephen Smalley if (creat_flags == HUGETLB_SHMFS_INODE) 1478e1832f29SStephen Smalley inode->i_flags |= S_PRIVATE; 14791da177e4SLinus Torvalds 14801da177e4SLinus Torvalds inode->i_size = size; 14816d6b77f1SMiklos Szeredi clear_nlink(inode); 1482ce8d2cdfSDave Hansen 148333b8f84aSMike Kravetz if (!hugetlb_reserve_pages(inode, 0, 1484e68375c8SAl Viro size >> huge_page_shift(hstate_inode(inode)), NULL, 1485e68375c8SAl Viro acctflag)) 1486e68375c8SAl Viro file = ERR_PTR(-ENOMEM); 1487e68375c8SAl Viro else 1488e68375c8SAl Viro file = alloc_file_pseudo(inode, mnt, name, O_RDWR, 1489ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 1490e68375c8SAl Viro if (!IS_ERR(file)) 14911da177e4SLinus Torvalds return file; 14921da177e4SLinus Torvalds 1493b45b5bd6SDavid Gibson iput(inode); 1494e68375c8SAl Viro out: 1495353d5c30SHugh Dickins if (*user) { 1496353d5c30SHugh Dickins user_shm_unlock(size, *user); 1497353d5c30SHugh Dickins *user = NULL; 1498353d5c30SHugh Dickins } 149939b65252SAnatol Pomozov return file; 15001da177e4SLinus Torvalds } 15011da177e4SLinus Torvalds 150232021982SDavid Howells static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h) 150332021982SDavid Howells { 150432021982SDavid Howells struct fs_context *fc; 150532021982SDavid Howells struct vfsmount *mnt; 150632021982SDavid Howells 150732021982SDavid Howells fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT); 150832021982SDavid Howells if (IS_ERR(fc)) { 150932021982SDavid Howells mnt = ERR_CAST(fc); 151032021982SDavid Howells } else { 151132021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 151232021982SDavid Howells ctx->hstate = h; 151332021982SDavid Howells mnt = fc_mount(fc); 151432021982SDavid Howells put_fs_context(fc); 151532021982SDavid Howells } 151632021982SDavid Howells if (IS_ERR(mnt)) 151732021982SDavid Howells pr_err("Cannot mount internal hugetlbfs for page size %uK", 151832021982SDavid Howells 1U << (h->order + PAGE_SHIFT - 10)); 151932021982SDavid Howells return mnt; 152032021982SDavid Howells } 152132021982SDavid Howells 15221da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 15231da177e4SLinus Torvalds { 152432021982SDavid Howells struct vfsmount *mnt; 152542d7395fSAndi Kleen struct hstate *h; 15261da177e4SLinus Torvalds int error; 152742d7395fSAndi Kleen int i; 15281da177e4SLinus Torvalds 1529457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 15309b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1531457c1b27SNishanth Aravamudan return -ENOTSUPP; 1532457c1b27SNishanth Aravamudan } 1533457c1b27SNishanth Aravamudan 1534d1d5e05fSHillf Danton error = -ENOMEM; 15351da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 15361da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 15375d097056SVladimir Davydov 0, SLAB_ACCOUNT, init_once); 15381da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 15398fc312b3SMike Kravetz goto out; 15401da177e4SLinus Torvalds 15411da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 15421da177e4SLinus Torvalds if (error) 15438fc312b3SMike Kravetz goto out_free; 15441da177e4SLinus Torvalds 15458fc312b3SMike Kravetz /* default hstate mount is required */ 1546*3b2275a8SMiaohe Lin mnt = mount_one_hugetlbfs(&default_hstate); 15478fc312b3SMike Kravetz if (IS_ERR(mnt)) { 15488fc312b3SMike Kravetz error = PTR_ERR(mnt); 15498fc312b3SMike Kravetz goto out_unreg; 15508fc312b3SMike Kravetz } 15518fc312b3SMike Kravetz hugetlbfs_vfsmount[default_hstate_idx] = mnt; 15528fc312b3SMike Kravetz 15538fc312b3SMike Kravetz /* other hstates are optional */ 155442d7395fSAndi Kleen i = 0; 155542d7395fSAndi Kleen for_each_hstate(h) { 155615f0ec94SJan Stancek if (i == default_hstate_idx) { 155715f0ec94SJan Stancek i++; 15588fc312b3SMike Kravetz continue; 155915f0ec94SJan Stancek } 15608fc312b3SMike Kravetz 156132021982SDavid Howells mnt = mount_one_hugetlbfs(h); 15628fc312b3SMike Kravetz if (IS_ERR(mnt)) 15638fc312b3SMike Kravetz hugetlbfs_vfsmount[i] = NULL; 15648fc312b3SMike Kravetz else 156532021982SDavid Howells hugetlbfs_vfsmount[i] = mnt; 156642d7395fSAndi Kleen i++; 156742d7395fSAndi Kleen } 156832021982SDavid Howells 156942d7395fSAndi Kleen return 0; 15701da177e4SLinus Torvalds 15718fc312b3SMike Kravetz out_unreg: 15728fc312b3SMike Kravetz (void)unregister_filesystem(&hugetlbfs_fs_type); 15738fc312b3SMike Kravetz out_free: 15741da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 15758fc312b3SMike Kravetz out: 15761da177e4SLinus Torvalds return error; 15771da177e4SLinus Torvalds } 15783e89e1c5SPaul Gortmaker fs_initcall(init_hugetlbfs_fs) 1579