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> 1470c3547eSMike Kravetz #include <linux/falloc.h> 151da177e4SLinus Torvalds #include <linux/fs.h> 161da177e4SLinus Torvalds #include <linux/mount.h> 171da177e4SLinus Torvalds #include <linux/file.h> 18e73a75faSRandy Dunlap #include <linux/kernel.h> 191da177e4SLinus Torvalds #include <linux/writeback.h> 201da177e4SLinus Torvalds #include <linux/pagemap.h> 211da177e4SLinus Torvalds #include <linux/highmem.h> 221da177e4SLinus Torvalds #include <linux/init.h> 231da177e4SLinus Torvalds #include <linux/string.h> 2416f7e0feSRandy Dunlap #include <linux/capability.h> 25e73a75faSRandy Dunlap #include <linux/ctype.h> 261da177e4SLinus Torvalds #include <linux/backing-dev.h> 271da177e4SLinus Torvalds #include <linux/hugetlb.h> 281da177e4SLinus Torvalds #include <linux/pagevec.h> 2932021982SDavid Howells #include <linux/fs_parser.h> 30036e0856SBenjamin Herrenschmidt #include <linux/mman.h> 311da177e4SLinus Torvalds #include <linux/slab.h> 321da177e4SLinus Torvalds #include <linux/dnotify.h> 331da177e4SLinus Torvalds #include <linux/statfs.h> 341da177e4SLinus Torvalds #include <linux/security.h> 351fd7317dSNick Black #include <linux/magic.h> 36290408d4SNaoya Horiguchi #include <linux/migrate.h> 3734d0640eSAl Viro #include <linux/uio.h> 381da177e4SLinus Torvalds 397c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 4088590253SShijie Hu #include <linux/sched/mm.h> 411da177e4SLinus Torvalds 42f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops; 434b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations; 4492e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations; 4592e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations; 461da177e4SLinus Torvalds 4732021982SDavid Howells enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT }; 4832021982SDavid Howells 4932021982SDavid Howells struct hugetlbfs_fs_context { 504a25220dSDavid Howells struct hstate *hstate; 5132021982SDavid Howells unsigned long long max_size_opt; 5232021982SDavid Howells unsigned long long min_size_opt; 534a25220dSDavid Howells long max_hpages; 544a25220dSDavid Howells long nr_inodes; 554a25220dSDavid Howells long min_hpages; 5632021982SDavid Howells enum hugetlbfs_size_type max_val_type; 5732021982SDavid Howells enum hugetlbfs_size_type min_val_type; 58a0eb3a05SEric W. Biederman kuid_t uid; 59a0eb3a05SEric W. Biederman kgid_t gid; 60a1d776eeSDavid Gibson umode_t mode; 61a1d776eeSDavid Gibson }; 62a1d776eeSDavid Gibson 631da177e4SLinus Torvalds int sysctl_hugetlb_shm_group; 641da177e4SLinus Torvalds 6532021982SDavid Howells enum hugetlb_param { 6632021982SDavid Howells Opt_gid, 6732021982SDavid Howells Opt_min_size, 6832021982SDavid Howells Opt_mode, 6932021982SDavid Howells Opt_nr_inodes, 7032021982SDavid Howells Opt_pagesize, 7132021982SDavid Howells Opt_size, 7232021982SDavid Howells Opt_uid, 73e73a75faSRandy Dunlap }; 74e73a75faSRandy Dunlap 75d7167b14SAl Viro static const struct fs_parameter_spec hugetlb_fs_parameters[] = { 7632021982SDavid Howells fsparam_u32 ("gid", Opt_gid), 7732021982SDavid Howells fsparam_string("min_size", Opt_min_size), 78e0f7e2b2SMike Kravetz fsparam_u32oct("mode", Opt_mode), 7932021982SDavid Howells fsparam_string("nr_inodes", Opt_nr_inodes), 8032021982SDavid Howells fsparam_string("pagesize", Opt_pagesize), 8132021982SDavid Howells fsparam_string("size", Opt_size), 8232021982SDavid Howells fsparam_u32 ("uid", Opt_uid), 8332021982SDavid Howells {} 8432021982SDavid Howells }; 8532021982SDavid Howells 8670c3547eSMike Kravetz #ifdef CONFIG_NUMA 8770c3547eSMike Kravetz static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, 8870c3547eSMike Kravetz struct inode *inode, pgoff_t index) 8970c3547eSMike Kravetz { 9070c3547eSMike Kravetz vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy, 9170c3547eSMike Kravetz index); 9270c3547eSMike Kravetz } 9370c3547eSMike Kravetz 9470c3547eSMike Kravetz static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) 9570c3547eSMike Kravetz { 9670c3547eSMike Kravetz mpol_cond_put(vma->vm_policy); 9770c3547eSMike Kravetz } 9870c3547eSMike Kravetz #else 9970c3547eSMike Kravetz static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, 10070c3547eSMike Kravetz struct inode *inode, pgoff_t index) 10170c3547eSMike Kravetz { 10270c3547eSMike Kravetz } 10370c3547eSMike Kravetz 10470c3547eSMike Kravetz static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) 10570c3547eSMike Kravetz { 10670c3547eSMike Kravetz } 10770c3547eSMike Kravetz #endif 10870c3547eSMike Kravetz 10963489f8eSMike Kravetz /* 11063489f8eSMike Kravetz * Mask used when checking the page offset value passed in via system 11163489f8eSMike Kravetz * calls. This value will be converted to a loff_t which is signed. 11263489f8eSMike Kravetz * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the 11363489f8eSMike Kravetz * value. The extra bit (- 1 in the shift value) is to take the sign 11463489f8eSMike Kravetz * bit into account. 11563489f8eSMike Kravetz */ 11663489f8eSMike Kravetz #define PGOFF_LOFFT_MAX \ 11763489f8eSMike Kravetz (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1))) 11863489f8eSMike Kravetz 1191da177e4SLinus Torvalds static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) 1201da177e4SLinus Torvalds { 121496ad9aaSAl Viro struct inode *inode = file_inode(file); 12222247efdSPeter Xu struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 1231da177e4SLinus Torvalds loff_t len, vma_len; 1241da177e4SLinus Torvalds int ret; 125a5516438SAndi Kleen struct hstate *h = hstate_file(file); 1261da177e4SLinus Torvalds 12768589bc3SHugh Dickins /* 128dec4ad86SDavid Gibson * vma address alignment (but not the pgoff alignment) has 129dec4ad86SDavid Gibson * already been checked by prepare_hugepage_range. If you add 130dec4ad86SDavid Gibson * any error returns here, do so after setting VM_HUGETLB, so 131dec4ad86SDavid Gibson * is_vm_hugetlb_page tests below unmap_region go the right 13245e55300SPeter Collingbourne * way when do_mmap unwinds (may be important on powerpc 133dec4ad86SDavid Gibson * and ia64). 13468589bc3SHugh Dickins */ 135a2fce914SNaoya Horiguchi vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; 13668589bc3SHugh Dickins vma->vm_ops = &hugetlb_vm_ops; 1371da177e4SLinus Torvalds 13822247efdSPeter Xu ret = seal_check_future_write(info->seals, vma); 13922247efdSPeter Xu if (ret) 14022247efdSPeter Xu return ret; 14122247efdSPeter Xu 142045c7a3fSMike Kravetz /* 14363489f8eSMike Kravetz * page based offset in vm_pgoff could be sufficiently large to 1445df63c2aSMike Kravetz * overflow a loff_t when converted to byte offset. This can 1455df63c2aSMike Kravetz * only happen on architectures where sizeof(loff_t) == 1465df63c2aSMike Kravetz * sizeof(unsigned long). So, only check in those instances. 147045c7a3fSMike Kravetz */ 1485df63c2aSMike Kravetz if (sizeof(unsigned long) == sizeof(loff_t)) { 14963489f8eSMike Kravetz if (vma->vm_pgoff & PGOFF_LOFFT_MAX) 150045c7a3fSMike Kravetz return -EINVAL; 1515df63c2aSMike Kravetz } 152045c7a3fSMike Kravetz 15363489f8eSMike Kravetz /* must be huge page aligned */ 1542b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 155dec4ad86SDavid Gibson return -EINVAL; 156dec4ad86SDavid Gibson 1571da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 158045c7a3fSMike Kravetz len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 159045c7a3fSMike Kravetz /* check for overflow */ 160045c7a3fSMike Kravetz if (len < vma_len) 161045c7a3fSMike Kravetz return -EINVAL; 1621da177e4SLinus Torvalds 1635955102cSAl Viro inode_lock(inode); 1641da177e4SLinus Torvalds file_accessed(file); 1651da177e4SLinus Torvalds 1661da177e4SLinus Torvalds ret = -ENOMEM; 16733b8f84aSMike Kravetz if (!hugetlb_reserve_pages(inode, 168a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1695a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1705a6fe125SMel Gorman vma->vm_flags)) 171b45b5bd6SDavid Gibson goto out; 172b45b5bd6SDavid Gibson 1734c887265SAdam Litke ret = 0; 174b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 175045c7a3fSMike Kravetz i_size_write(inode, len); 1761da177e4SLinus Torvalds out: 1775955102cSAl Viro inode_unlock(inode); 1781da177e4SLinus Torvalds 1791da177e4SLinus Torvalds return ret; 1801da177e4SLinus Torvalds } 1811da177e4SLinus Torvalds 1821da177e4SLinus Torvalds /* 1833e4e28c5SMichel Lespinasse * Called under mmap_write_lock(mm). 1841da177e4SLinus Torvalds */ 1851da177e4SLinus Torvalds 1861da177e4SLinus Torvalds static unsigned long 18788590253SShijie Hu hugetlb_get_unmapped_area_bottomup(struct file *file, unsigned long addr, 18888590253SShijie Hu unsigned long len, unsigned long pgoff, unsigned long flags) 18988590253SShijie Hu { 19088590253SShijie Hu struct hstate *h = hstate_file(file); 19188590253SShijie Hu struct vm_unmapped_area_info info; 19288590253SShijie Hu 19388590253SShijie Hu info.flags = 0; 19488590253SShijie Hu info.length = len; 19588590253SShijie Hu info.low_limit = current->mm->mmap_base; 1962cb4de08SChristophe Leroy info.high_limit = arch_get_mmap_end(addr, len, flags); 19788590253SShijie Hu info.align_mask = PAGE_MASK & ~huge_page_mask(h); 19888590253SShijie Hu info.align_offset = 0; 19988590253SShijie Hu return vm_unmapped_area(&info); 20088590253SShijie Hu } 20188590253SShijie Hu 20288590253SShijie Hu static unsigned long 20388590253SShijie Hu hugetlb_get_unmapped_area_topdown(struct file *file, unsigned long addr, 20488590253SShijie Hu unsigned long len, unsigned long pgoff, unsigned long flags) 20588590253SShijie Hu { 20688590253SShijie Hu struct hstate *h = hstate_file(file); 20788590253SShijie Hu struct vm_unmapped_area_info info; 20888590253SShijie Hu 20988590253SShijie Hu info.flags = VM_UNMAPPED_AREA_TOPDOWN; 21088590253SShijie Hu info.length = len; 21188590253SShijie Hu info.low_limit = max(PAGE_SIZE, mmap_min_addr); 2125f24d5a5SChristophe Leroy info.high_limit = arch_get_mmap_base(addr, current->mm->mmap_base); 21388590253SShijie Hu info.align_mask = PAGE_MASK & ~huge_page_mask(h); 21488590253SShijie Hu info.align_offset = 0; 21588590253SShijie Hu addr = vm_unmapped_area(&info); 21688590253SShijie Hu 21788590253SShijie Hu /* 21888590253SShijie Hu * A failed mmap() very likely causes application failure, 21988590253SShijie Hu * so fall back to the bottom-up function here. This scenario 22088590253SShijie Hu * can happen with large stack limits and large mmap() 22188590253SShijie Hu * allocations. 22288590253SShijie Hu */ 22388590253SShijie Hu if (unlikely(offset_in_page(addr))) { 22488590253SShijie Hu VM_BUG_ON(addr != -ENOMEM); 22588590253SShijie Hu info.flags = 0; 22688590253SShijie Hu info.low_limit = current->mm->mmap_base; 2272cb4de08SChristophe Leroy info.high_limit = arch_get_mmap_end(addr, len, flags); 22888590253SShijie Hu addr = vm_unmapped_area(&info); 22988590253SShijie Hu } 23088590253SShijie Hu 23188590253SShijie Hu return addr; 23288590253SShijie Hu } 23388590253SShijie Hu 2344b439e25SChristophe Leroy unsigned long 2354b439e25SChristophe Leroy generic_hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 2364b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 2374b439e25SChristophe Leroy unsigned long flags) 2381da177e4SLinus Torvalds { 2391da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2401da177e4SLinus Torvalds struct vm_area_struct *vma; 241a5516438SAndi Kleen struct hstate *h = hstate_file(file); 2422cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 2431da177e4SLinus Torvalds 244a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 2451da177e4SLinus Torvalds return -EINVAL; 2461da177e4SLinus Torvalds if (len > TASK_SIZE) 2471da177e4SLinus Torvalds return -ENOMEM; 2481da177e4SLinus Torvalds 249036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 250a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 251036e0856SBenjamin Herrenschmidt return -EINVAL; 252036e0856SBenjamin Herrenschmidt return addr; 253036e0856SBenjamin Herrenschmidt } 254036e0856SBenjamin Herrenschmidt 2551da177e4SLinus Torvalds if (addr) { 256a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 2571da177e4SLinus Torvalds vma = find_vma(mm, addr); 2585f24d5a5SChristophe Leroy if (mmap_end - len >= addr && 2591be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma))) 2601da177e4SLinus Torvalds return addr; 2611da177e4SLinus Torvalds } 2621da177e4SLinus Torvalds 26388590253SShijie Hu /* 26488590253SShijie Hu * Use mm->get_unmapped_area value as a hint to use topdown routine. 26588590253SShijie Hu * If architectures have special needs, they should define their own 26688590253SShijie Hu * version of hugetlb_get_unmapped_area. 26788590253SShijie Hu */ 26888590253SShijie Hu if (mm->get_unmapped_area == arch_get_unmapped_area_topdown) 26988590253SShijie Hu return hugetlb_get_unmapped_area_topdown(file, addr, len, 27088590253SShijie Hu pgoff, flags); 27188590253SShijie Hu return hugetlb_get_unmapped_area_bottomup(file, addr, len, 27288590253SShijie Hu pgoff, flags); 2731da177e4SLinus Torvalds } 2744b439e25SChristophe Leroy 2754b439e25SChristophe Leroy #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 2764b439e25SChristophe Leroy static unsigned long 2774b439e25SChristophe Leroy hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 2784b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 2794b439e25SChristophe Leroy unsigned long flags) 2804b439e25SChristophe Leroy { 2814b439e25SChristophe Leroy return generic_hugetlb_get_unmapped_area(file, addr, len, pgoff, flags); 2824b439e25SChristophe Leroy } 2831da177e4SLinus Torvalds #endif 2841da177e4SLinus Torvalds 285e63e1e5aSBadari Pulavarty /* 286e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 287445c8098SMiaohe Lin * data. This provides functionality similar to filemap_read(). 288e63e1e5aSBadari Pulavarty */ 28934d0640eSAl Viro static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) 290e63e1e5aSBadari Pulavarty { 29134d0640eSAl Viro struct file *file = iocb->ki_filp; 29234d0640eSAl Viro struct hstate *h = hstate_file(file); 29334d0640eSAl Viro struct address_space *mapping = file->f_mapping; 294e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 29534d0640eSAl Viro unsigned long index = iocb->ki_pos >> huge_page_shift(h); 29634d0640eSAl Viro unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); 297e63e1e5aSBadari Pulavarty unsigned long end_index; 298e63e1e5aSBadari Pulavarty loff_t isize; 299e63e1e5aSBadari Pulavarty ssize_t retval = 0; 300e63e1e5aSBadari Pulavarty 30134d0640eSAl Viro while (iov_iter_count(to)) { 302e63e1e5aSBadari Pulavarty struct page *page; 30334d0640eSAl Viro size_t nr, copied; 304e63e1e5aSBadari Pulavarty 305e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 306a5516438SAndi Kleen nr = huge_page_size(h); 307a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 308a05b0855SAneesh Kumar K.V if (!isize) 30934d0640eSAl Viro break; 310a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 311e63e1e5aSBadari Pulavarty if (index > end_index) 31234d0640eSAl Viro break; 31334d0640eSAl Viro if (index == end_index) { 314a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 315a05b0855SAneesh Kumar K.V if (nr <= offset) 31634d0640eSAl Viro break; 317e63e1e5aSBadari Pulavarty } 318e63e1e5aSBadari Pulavarty nr = nr - offset; 319e63e1e5aSBadari Pulavarty 320e63e1e5aSBadari Pulavarty /* Find the page */ 321a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 322e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 323e63e1e5aSBadari Pulavarty /* 324e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 325e63e1e5aSBadari Pulavarty * length of the hole or request. 326e63e1e5aSBadari Pulavarty */ 32734d0640eSAl Viro copied = iov_iter_zero(nr, to); 328e63e1e5aSBadari Pulavarty } else { 329a05b0855SAneesh Kumar K.V unlock_page(page); 330a05b0855SAneesh Kumar K.V 331e63e1e5aSBadari Pulavarty /* 332e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 333e63e1e5aSBadari Pulavarty */ 334c7d57ab1SAl Viro copied = copy_page_to_iter(page, offset, nr, to); 33509cbfeafSKirill A. Shutemov put_page(page); 336e63e1e5aSBadari Pulavarty } 33734d0640eSAl Viro offset += copied; 33834d0640eSAl Viro retval += copied; 33934d0640eSAl Viro if (copied != nr && iov_iter_count(to)) { 34034d0640eSAl Viro if (!retval) 34134d0640eSAl Viro retval = -EFAULT; 342e63e1e5aSBadari Pulavarty break; 343e63e1e5aSBadari Pulavarty } 34434d0640eSAl Viro index += offset >> huge_page_shift(h); 34534d0640eSAl Viro offset &= ~huge_page_mask(h); 34634d0640eSAl Viro } 34734d0640eSAl Viro iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; 348e63e1e5aSBadari Pulavarty return retval; 349e63e1e5aSBadari Pulavarty } 350e63e1e5aSBadari Pulavarty 351800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 352800d15a5SNick Piggin struct address_space *mapping, 3539d6b0cd7SMatthew Wilcox (Oracle) loff_t pos, unsigned len, 354800d15a5SNick Piggin struct page **pagep, void **fsdata) 3551da177e4SLinus Torvalds { 3561da177e4SLinus Torvalds return -EINVAL; 3571da177e4SLinus Torvalds } 3581da177e4SLinus Torvalds 359800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 360800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 361800d15a5SNick Piggin struct page *page, void *fsdata) 3621da177e4SLinus Torvalds { 363800d15a5SNick Piggin BUG(); 3641da177e4SLinus Torvalds return -EINVAL; 3651da177e4SLinus Torvalds } 3661da177e4SLinus Torvalds 3677e1813d4SMike Kravetz static void hugetlb_delete_from_page_cache(struct page *page) 3681da177e4SLinus Torvalds { 369b9ea2515SKonstantin Khlebnikov ClearPageDirty(page); 3701da177e4SLinus Torvalds ClearPageUptodate(page); 371bd65cb86SMinchan Kim delete_from_page_cache(page); 3721da177e4SLinus Torvalds } 3731da177e4SLinus Torvalds 3744aae8d1cSMike Kravetz static void 37505e90bd0SPeter Xu hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end, 37605e90bd0SPeter Xu zap_flags_t zap_flags) 3774aae8d1cSMike Kravetz { 3784aae8d1cSMike Kravetz struct vm_area_struct *vma; 3794aae8d1cSMike Kravetz 3804aae8d1cSMike Kravetz /* 381d6aba4c8SSean Christopherson * end == 0 indicates that the entire range after start should be 382d6aba4c8SSean Christopherson * unmapped. Note, end is exclusive, whereas the interval tree takes 383d6aba4c8SSean Christopherson * an inclusive "last". 3844aae8d1cSMike Kravetz */ 385d6aba4c8SSean Christopherson vma_interval_tree_foreach(vma, root, start, end ? end - 1 : ULONG_MAX) { 3864aae8d1cSMike Kravetz unsigned long v_offset; 3874aae8d1cSMike Kravetz unsigned long v_end; 3884aae8d1cSMike Kravetz 3894aae8d1cSMike Kravetz /* 3904aae8d1cSMike Kravetz * Can the expression below overflow on 32-bit arches? 3914aae8d1cSMike Kravetz * No, because the interval tree returns us only those vmas 3924aae8d1cSMike Kravetz * which overlap the truncated area starting at pgoff, 3934aae8d1cSMike Kravetz * and no vma on a 32-bit arch can span beyond the 4GB. 3944aae8d1cSMike Kravetz */ 3954aae8d1cSMike Kravetz if (vma->vm_pgoff < start) 3964aae8d1cSMike Kravetz v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; 3974aae8d1cSMike Kravetz else 3984aae8d1cSMike Kravetz v_offset = 0; 3994aae8d1cSMike Kravetz 4004aae8d1cSMike Kravetz if (!end) 4014aae8d1cSMike Kravetz v_end = vma->vm_end; 4024aae8d1cSMike Kravetz else { 4034aae8d1cSMike Kravetz v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) 4044aae8d1cSMike Kravetz + vma->vm_start; 4054aae8d1cSMike Kravetz if (v_end > vma->vm_end) 4064aae8d1cSMike Kravetz v_end = vma->vm_end; 4074aae8d1cSMike Kravetz } 4084aae8d1cSMike Kravetz 4094aae8d1cSMike Kravetz unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, 41005e90bd0SPeter Xu NULL, zap_flags); 4114aae8d1cSMike Kravetz } 4124aae8d1cSMike Kravetz } 413b5cec28dSMike Kravetz 414b5cec28dSMike Kravetz /* 415*c8627228SMike Kravetz * Called with hugetlb fault mutex held. 416*c8627228SMike Kravetz * Returns true if page was actually removed, false otherwise. 417*c8627228SMike Kravetz */ 418*c8627228SMike Kravetz static bool remove_inode_single_folio(struct hstate *h, struct inode *inode, 419*c8627228SMike Kravetz struct address_space *mapping, 420*c8627228SMike Kravetz struct folio *folio, pgoff_t index, 421*c8627228SMike Kravetz bool truncate_op) 422*c8627228SMike Kravetz { 423*c8627228SMike Kravetz bool ret = false; 424*c8627228SMike Kravetz 425*c8627228SMike Kravetz /* 426*c8627228SMike Kravetz * If folio is mapped, it was faulted in after being 427*c8627228SMike Kravetz * unmapped in caller. Unmap (again) while holding 428*c8627228SMike Kravetz * the fault mutex. The mutex will prevent faults 429*c8627228SMike Kravetz * until we finish removing the folio. 430*c8627228SMike Kravetz */ 431*c8627228SMike Kravetz if (unlikely(folio_mapped(folio))) { 432*c8627228SMike Kravetz i_mmap_lock_write(mapping); 433*c8627228SMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 434*c8627228SMike Kravetz index * pages_per_huge_page(h), 435*c8627228SMike Kravetz (index + 1) * pages_per_huge_page(h), 436*c8627228SMike Kravetz ZAP_FLAG_DROP_MARKER); 437*c8627228SMike Kravetz i_mmap_unlock_write(mapping); 438*c8627228SMike Kravetz } 439*c8627228SMike Kravetz 440*c8627228SMike Kravetz folio_lock(folio); 441*c8627228SMike Kravetz /* 442*c8627228SMike Kravetz * After locking page, make sure mapping is the same. 443*c8627228SMike Kravetz * We could have raced with page fault populate and 444*c8627228SMike Kravetz * backout code. 445*c8627228SMike Kravetz */ 446*c8627228SMike Kravetz if (folio_mapping(folio) == mapping) { 447*c8627228SMike Kravetz /* 448*c8627228SMike Kravetz * We must remove the folio from page cache before removing 449*c8627228SMike Kravetz * the region/ reserve map (hugetlb_unreserve_pages). In 450*c8627228SMike Kravetz * rare out of memory conditions, removal of the region/reserve 451*c8627228SMike Kravetz * map could fail. Correspondingly, the subpool and global 452*c8627228SMike Kravetz * reserve usage count can need to be adjusted. 453*c8627228SMike Kravetz */ 454*c8627228SMike Kravetz VM_BUG_ON(HPageRestoreReserve(&folio->page)); 455*c8627228SMike Kravetz hugetlb_delete_from_page_cache(&folio->page); 456*c8627228SMike Kravetz ret = true; 457*c8627228SMike Kravetz if (!truncate_op) { 458*c8627228SMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, index, 459*c8627228SMike Kravetz index + 1, 1))) 460*c8627228SMike Kravetz hugetlb_fix_reserve_counts(inode); 461*c8627228SMike Kravetz } 462*c8627228SMike Kravetz } 463*c8627228SMike Kravetz 464*c8627228SMike Kravetz folio_unlock(folio); 465*c8627228SMike Kravetz return ret; 466*c8627228SMike Kravetz } 467*c8627228SMike Kravetz 468*c8627228SMike Kravetz /* 469b5cec28dSMike Kravetz * remove_inode_hugepages handles two distinct cases: truncation and hole 470b5cec28dSMike Kravetz * punch. There are subtle differences in operation for each case. 4714aae8d1cSMike Kravetz * 472b5cec28dSMike Kravetz * truncation is indicated by end of range being LLONG_MAX 473b5cec28dSMike Kravetz * In this case, we first scan the range and release found pages. 4741935ebd3SMiaohe Lin * After releasing pages, hugetlb_unreserve_pages cleans up region/reserve 475*c8627228SMike Kravetz * maps and global counts. Page faults can race with truncation. 476*c8627228SMike Kravetz * During faults, hugetlb_no_page() checks i_size before page allocation, 477*c8627228SMike Kravetz * and again after obtaining page table lock. It will 'back out' 478*c8627228SMike Kravetz * allocations in the truncated range. 479b5cec28dSMike Kravetz * hole punch is indicated if end is not LLONG_MAX 480b5cec28dSMike Kravetz * In the hole punch case we scan the range and release found pages. 4811935ebd3SMiaohe Lin * Only when releasing a page is the associated region/reserve map 4821935ebd3SMiaohe Lin * deleted. The region/reserve map for ranges without associated 483e7c58097SMike Kravetz * pages are not modified. Page faults can race with hole punch. 484e7c58097SMike Kravetz * This is indicated if we find a mapped page. 485b5cec28dSMike Kravetz * Note: If the passed end of range value is beyond the end of file, but 486b5cec28dSMike Kravetz * not LLONG_MAX this routine still performs a hole punch operation. 487b5cec28dSMike Kravetz */ 488b5cec28dSMike Kravetz static void remove_inode_hugepages(struct inode *inode, loff_t lstart, 489b5cec28dSMike Kravetz loff_t lend) 4901da177e4SLinus Torvalds { 491a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 492b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 493a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 494b5cec28dSMike Kravetz const pgoff_t end = lend >> huge_page_shift(h); 4951508062eSMatthew Wilcox (Oracle) struct folio_batch fbatch; 496d72dc8a2SJan Kara pgoff_t next, index; 497a43a8c39SChen, Kenneth W int i, freed = 0; 498b5cec28dSMike Kravetz bool truncate_op = (lend == LLONG_MAX); 4991da177e4SLinus Torvalds 5001508062eSMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 5011da177e4SLinus Torvalds next = start; 5021508062eSMatthew Wilcox (Oracle) while (filemap_get_folios(mapping, &next, end - 1, &fbatch)) { 5031508062eSMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); ++i) { 5041508062eSMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 505d4241a04SMiaohe Lin u32 hash = 0; 506b5cec28dSMike Kravetz 5071508062eSMatthew Wilcox (Oracle) index = folio->index; 508d4241a04SMiaohe Lin hash = hugetlb_fault_mutex_hash(mapping, index); 509e7c58097SMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 510e7c58097SMike Kravetz 511b5cec28dSMike Kravetz /* 512*c8627228SMike Kravetz * Remove folio that was part of folio_batch. 513b5cec28dSMike Kravetz */ 514*c8627228SMike Kravetz if (remove_inode_single_folio(h, inode, mapping, folio, 515*c8627228SMike Kravetz index, truncate_op)) 516b5cec28dSMike Kravetz freed++; 517b5cec28dSMike Kravetz 518e7c58097SMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 5191da177e4SLinus Torvalds } 5201508062eSMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 5211817889eSMike Kravetz cond_resched(); 5221da177e4SLinus Torvalds } 523b5cec28dSMike Kravetz 524b5cec28dSMike Kravetz if (truncate_op) 525b5cec28dSMike Kravetz (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); 5261da177e4SLinus Torvalds } 5271da177e4SLinus Torvalds 5282bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 5291da177e4SLinus Torvalds { 5309119a41eSJoonsoo Kim struct resv_map *resv_map; 5319119a41eSJoonsoo Kim 532b5cec28dSMike Kravetz remove_inode_hugepages(inode, 0, LLONG_MAX); 533f27a5136SMike Kravetz 534f27a5136SMike Kravetz /* 535f27a5136SMike Kravetz * Get the resv_map from the address space embedded in the inode. 536f27a5136SMike Kravetz * This is the address space which points to any resv_map allocated 537f27a5136SMike Kravetz * at inode creation time. If this is a device special inode, 538f27a5136SMike Kravetz * i_mapping may not point to the original address space. 539f27a5136SMike Kravetz */ 540f27a5136SMike Kravetz resv_map = (struct resv_map *)(&inode->i_data)->private_data; 541f27a5136SMike Kravetz /* Only regular and link inodes have associated reserve maps */ 5429119a41eSJoonsoo Kim if (resv_map) 5439119a41eSJoonsoo Kim resv_map_release(&resv_map->refs); 544dbd5768fSJan Kara clear_inode(inode); 545149f4211SChristoph Hellwig } 546149f4211SChristoph Hellwig 547e5d319deSMiaohe Lin static void hugetlb_vmtruncate(struct inode *inode, loff_t offset) 5481da177e4SLinus Torvalds { 549856fc295SHugh Dickins pgoff_t pgoff; 5501da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 551a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 5521da177e4SLinus Torvalds 553a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 554856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 5551da177e4SLinus Torvalds 55687bf91d3SMike Kravetz i_size_write(inode, offset); 557188a3972SMike Kravetz i_mmap_lock_write(mapping); 558f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 55905e90bd0SPeter Xu hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0, 56005e90bd0SPeter Xu ZAP_FLAG_DROP_MARKER); 561c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 562e7c58097SMike Kravetz remove_inode_hugepages(inode, offset, LLONG_MAX); 5631da177e4SLinus Torvalds } 5641da177e4SLinus Torvalds 56568d32527SMike Kravetz static void hugetlbfs_zero_partial_page(struct hstate *h, 56668d32527SMike Kravetz struct address_space *mapping, 56768d32527SMike Kravetz loff_t start, 56868d32527SMike Kravetz loff_t end) 56968d32527SMike Kravetz { 57068d32527SMike Kravetz pgoff_t idx = start >> huge_page_shift(h); 57168d32527SMike Kravetz struct folio *folio; 57268d32527SMike Kravetz 57368d32527SMike Kravetz folio = filemap_lock_folio(mapping, idx); 57468d32527SMike Kravetz if (!folio) 57568d32527SMike Kravetz return; 57668d32527SMike Kravetz 57768d32527SMike Kravetz start = start & ~huge_page_mask(h); 57868d32527SMike Kravetz end = end & ~huge_page_mask(h); 57968d32527SMike Kravetz if (!end) 58068d32527SMike Kravetz end = huge_page_size(h); 58168d32527SMike Kravetz 58268d32527SMike Kravetz folio_zero_segment(folio, (size_t)start, (size_t)end); 58368d32527SMike Kravetz 58468d32527SMike Kravetz folio_unlock(folio); 58568d32527SMike Kravetz folio_put(folio); 58668d32527SMike Kravetz } 58768d32527SMike Kravetz 58870c3547eSMike Kravetz static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) 58970c3547eSMike Kravetz { 59068d32527SMike Kravetz struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 59168d32527SMike Kravetz struct address_space *mapping = inode->i_mapping; 59270c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 59370c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 59470c3547eSMike Kravetz loff_t hole_start, hole_end; 59570c3547eSMike Kravetz 59670c3547eSMike Kravetz /* 59768d32527SMike Kravetz * hole_start and hole_end indicate the full pages within the hole. 59870c3547eSMike Kravetz */ 59970c3547eSMike Kravetz hole_start = round_up(offset, hpage_size); 60070c3547eSMike Kravetz hole_end = round_down(offset + len, hpage_size); 60170c3547eSMike Kravetz 6025955102cSAl Viro inode_lock(inode); 603ff62a342SMarc-André Lureau 604398c0da7SMiaohe Lin /* protected by i_rwsem */ 605ab3948f5SJoel Fernandes (Google) if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { 606ff62a342SMarc-André Lureau inode_unlock(inode); 607ff62a342SMarc-André Lureau return -EPERM; 608ff62a342SMarc-André Lureau } 609ff62a342SMarc-André Lureau 61070c3547eSMike Kravetz i_mmap_lock_write(mapping); 61168d32527SMike Kravetz 61268d32527SMike Kravetz /* If range starts before first full page, zero partial page. */ 61368d32527SMike Kravetz if (offset < hole_start) 61468d32527SMike Kravetz hugetlbfs_zero_partial_page(h, mapping, 61568d32527SMike Kravetz offset, min(offset + len, hole_start)); 61668d32527SMike Kravetz 61768d32527SMike Kravetz /* Unmap users of full pages in the hole. */ 61868d32527SMike Kravetz if (hole_end > hole_start) { 619f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 62070c3547eSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 62170c3547eSMike Kravetz hole_start >> PAGE_SHIFT, 62205e90bd0SPeter Xu hole_end >> PAGE_SHIFT, 0); 62370c3547eSMike Kravetz } 62470c3547eSMike Kravetz 62568d32527SMike Kravetz /* If range extends beyond last full page, zero partial page. */ 62668d32527SMike Kravetz if ((offset + len) > hole_end && (offset + len) > hole_start) 62768d32527SMike Kravetz hugetlbfs_zero_partial_page(h, mapping, 62868d32527SMike Kravetz hole_end, offset + len); 62968d32527SMike Kravetz 63068d32527SMike Kravetz i_mmap_unlock_write(mapping); 63168d32527SMike Kravetz 63268d32527SMike Kravetz /* Remove full pages from the file. */ 63368d32527SMike Kravetz if (hole_end > hole_start) 63468d32527SMike Kravetz remove_inode_hugepages(inode, hole_start, hole_end); 63568d32527SMike Kravetz 63668d32527SMike Kravetz inode_unlock(inode); 63768d32527SMike Kravetz 63870c3547eSMike Kravetz return 0; 63970c3547eSMike Kravetz } 64070c3547eSMike Kravetz 64170c3547eSMike Kravetz static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, 64270c3547eSMike Kravetz loff_t len) 64370c3547eSMike Kravetz { 64470c3547eSMike Kravetz struct inode *inode = file_inode(file); 645ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 64670c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 64770c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 64870c3547eSMike Kravetz struct vm_area_struct pseudo_vma; 64970c3547eSMike Kravetz struct mm_struct *mm = current->mm; 65070c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 65170c3547eSMike Kravetz unsigned long hpage_shift = huge_page_shift(h); 65270c3547eSMike Kravetz pgoff_t start, index, end; 65370c3547eSMike Kravetz int error; 65470c3547eSMike Kravetz u32 hash; 65570c3547eSMike Kravetz 65670c3547eSMike Kravetz if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 65770c3547eSMike Kravetz return -EOPNOTSUPP; 65870c3547eSMike Kravetz 65970c3547eSMike Kravetz if (mode & FALLOC_FL_PUNCH_HOLE) 66070c3547eSMike Kravetz return hugetlbfs_punch_hole(inode, offset, len); 66170c3547eSMike Kravetz 66270c3547eSMike Kravetz /* 66370c3547eSMike Kravetz * Default preallocate case. 66470c3547eSMike Kravetz * For this range, start is rounded down and end is rounded up 66570c3547eSMike Kravetz * as well as being converted to page offsets. 66670c3547eSMike Kravetz */ 66770c3547eSMike Kravetz start = offset >> hpage_shift; 66870c3547eSMike Kravetz end = (offset + len + hpage_size - 1) >> hpage_shift; 66970c3547eSMike Kravetz 6705955102cSAl Viro inode_lock(inode); 67170c3547eSMike Kravetz 67270c3547eSMike Kravetz /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ 67370c3547eSMike Kravetz error = inode_newsize_ok(inode, offset + len); 67470c3547eSMike Kravetz if (error) 67570c3547eSMike Kravetz goto out; 67670c3547eSMike Kravetz 677ff62a342SMarc-André Lureau if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { 678ff62a342SMarc-André Lureau error = -EPERM; 679ff62a342SMarc-André Lureau goto out; 680ff62a342SMarc-André Lureau } 681ff62a342SMarc-André Lureau 68270c3547eSMike Kravetz /* 68370c3547eSMike Kravetz * Initialize a pseudo vma as this is required by the huge page 68470c3547eSMike Kravetz * allocation routines. If NUMA is configured, use page index 68570c3547eSMike Kravetz * as input to create an allocation policy. 68670c3547eSMike Kravetz */ 6872c4541e2SKirill A. Shutemov vma_init(&pseudo_vma, mm); 68870c3547eSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 68970c3547eSMike Kravetz pseudo_vma.vm_file = file; 69070c3547eSMike Kravetz 69170c3547eSMike Kravetz for (index = start; index < end; index++) { 69270c3547eSMike Kravetz /* 69370c3547eSMike Kravetz * This is supposed to be the vaddr where the page is being 69470c3547eSMike Kravetz * faulted in, but we have no vaddr here. 69570c3547eSMike Kravetz */ 69670c3547eSMike Kravetz struct page *page; 69770c3547eSMike Kravetz unsigned long addr; 69870c3547eSMike Kravetz 69970c3547eSMike Kravetz cond_resched(); 70070c3547eSMike Kravetz 70170c3547eSMike Kravetz /* 70270c3547eSMike Kravetz * fallocate(2) manpage permits EINTR; we may have been 70370c3547eSMike Kravetz * interrupted because we are using up too much memory. 70470c3547eSMike Kravetz */ 70570c3547eSMike Kravetz if (signal_pending(current)) { 70670c3547eSMike Kravetz error = -EINTR; 70770c3547eSMike Kravetz break; 70870c3547eSMike Kravetz } 70970c3547eSMike Kravetz 71070c3547eSMike Kravetz /* Set numa allocation policy based on index */ 71170c3547eSMike Kravetz hugetlb_set_vma_policy(&pseudo_vma, inode, index); 71270c3547eSMike Kravetz 71370c3547eSMike Kravetz /* addr is the offset within the file (zero based) */ 71470c3547eSMike Kravetz addr = index * hpage_size; 71570c3547eSMike Kravetz 716188a3972SMike Kravetz /* mutex taken here, fault path and hole punch */ 717188b04a7SWei Yang hash = hugetlb_fault_mutex_hash(mapping, index); 71870c3547eSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 71970c3547eSMike Kravetz 72070c3547eSMike Kravetz /* See if already present in mapping to avoid alloc/free */ 72170c3547eSMike Kravetz page = find_get_page(mapping, index); 72270c3547eSMike Kravetz if (page) { 72370c3547eSMike Kravetz put_page(page); 72470c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 72570c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 72670c3547eSMike Kravetz continue; 72770c3547eSMike Kravetz } 72870c3547eSMike Kravetz 72988ce3fefSMiaohe Lin /* 73088ce3fefSMiaohe Lin * Allocate page without setting the avoid_reserve argument. 73188ce3fefSMiaohe Lin * There certainly are no reserves associated with the 73288ce3fefSMiaohe Lin * pseudo_vma. However, there could be shared mappings with 73388ce3fefSMiaohe Lin * reserves for the file at the inode level. If we fallocate 73488ce3fefSMiaohe Lin * pages in these areas, we need to consume the reserves 73588ce3fefSMiaohe Lin * to keep reservation accounting consistent. 73688ce3fefSMiaohe Lin */ 73788ce3fefSMiaohe Lin page = alloc_huge_page(&pseudo_vma, addr, 0); 73870c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 73970c3547eSMike Kravetz if (IS_ERR(page)) { 74070c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 74170c3547eSMike Kravetz error = PTR_ERR(page); 74270c3547eSMike Kravetz goto out; 74370c3547eSMike Kravetz } 74470c3547eSMike Kravetz clear_huge_page(page, addr, pages_per_huge_page(h)); 74570c3547eSMike Kravetz __SetPageUptodate(page); 7467e1813d4SMike Kravetz error = hugetlb_add_to_page_cache(page, mapping, index); 74770c3547eSMike Kravetz if (unlikely(error)) { 748846be085SMike Kravetz restore_reserve_on_error(h, &pseudo_vma, addr, page); 74970c3547eSMike Kravetz put_page(page); 75070c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 75170c3547eSMike Kravetz goto out; 75270c3547eSMike Kravetz } 75370c3547eSMike Kravetz 75470c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 75570c3547eSMike Kravetz 7568f251a3dSMike Kravetz SetHPageMigratable(page); 75770c3547eSMike Kravetz /* 7587e1813d4SMike Kravetz * unlock_page because locked by hugetlb_add_to_page_cache() 759585fc0d2SMuchun Song * put_page() due to reference from alloc_huge_page() 76070c3547eSMike Kravetz */ 76170c3547eSMike Kravetz unlock_page(page); 76272639e6dSNadav Amit put_page(page); 76370c3547eSMike Kravetz } 76470c3547eSMike Kravetz 76570c3547eSMike Kravetz if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) 76670c3547eSMike Kravetz i_size_write(inode, offset + len); 767078cd827SDeepa Dinamani inode->i_ctime = current_time(inode); 76870c3547eSMike Kravetz out: 7695955102cSAl Viro inode_unlock(inode); 77070c3547eSMike Kravetz return error; 77170c3547eSMike Kravetz } 77270c3547eSMike Kravetz 773549c7297SChristian Brauner static int hugetlbfs_setattr(struct user_namespace *mnt_userns, 774549c7297SChristian Brauner struct dentry *dentry, struct iattr *attr) 7751da177e4SLinus Torvalds { 7762b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 777a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 7781da177e4SLinus Torvalds int error; 7791da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 780ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 7811da177e4SLinus Torvalds 7822f221d6fSChristian Brauner error = setattr_prepare(&init_user_ns, dentry, attr); 7831da177e4SLinus Torvalds if (error) 7841025774cSChristoph Hellwig return error; 7851da177e4SLinus Torvalds 7861da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 787ff62a342SMarc-André Lureau loff_t oldsize = inode->i_size; 788ff62a342SMarc-André Lureau loff_t newsize = attr->ia_size; 789ff62a342SMarc-André Lureau 790ff62a342SMarc-André Lureau if (newsize & ~huge_page_mask(h)) 7911025774cSChristoph Hellwig return -EINVAL; 792398c0da7SMiaohe Lin /* protected by i_rwsem */ 793ff62a342SMarc-André Lureau if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || 794ff62a342SMarc-André Lureau (newsize > oldsize && (info->seals & F_SEAL_GROW))) 795ff62a342SMarc-André Lureau return -EPERM; 796e5d319deSMiaohe Lin hugetlb_vmtruncate(inode, newsize); 7971da177e4SLinus Torvalds } 7981da177e4SLinus Torvalds 7992f221d6fSChristian Brauner setattr_copy(&init_user_ns, inode, attr); 8001025774cSChristoph Hellwig mark_inode_dirty(inode); 8011025774cSChristoph Hellwig return 0; 8021025774cSChristoph Hellwig } 8031025774cSChristoph Hellwig 8047d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 80532021982SDavid Howells struct hugetlbfs_fs_context *ctx) 8061da177e4SLinus Torvalds { 8071da177e4SLinus Torvalds struct inode *inode; 8081da177e4SLinus Torvalds 8091da177e4SLinus Torvalds inode = new_inode(sb); 8101da177e4SLinus Torvalds if (inode) { 81185fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 81232021982SDavid Howells inode->i_mode = S_IFDIR | ctx->mode; 81332021982SDavid Howells inode->i_uid = ctx->uid; 81432021982SDavid Howells inode->i_gid = ctx->gid; 815078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 8167d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 8177d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 8187d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 8197d54fa64SAl Viro inc_nlink(inode); 82065ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 8217d54fa64SAl Viro } 8227d54fa64SAl Viro return inode; 8237d54fa64SAl Viro } 8247d54fa64SAl Viro 825b610ded7SMichal Hocko /* 826c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 827b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 82888f306b6SKirill A. Shutemov * annotation because huge_pmd_share() does an allocation under hugetlb's 829c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 830b610ded7SMichal Hocko */ 831c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 832b610ded7SMichal Hocko 8337d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 8347d54fa64SAl Viro struct inode *dir, 83518df2252SAl Viro umode_t mode, dev_t dev) 8367d54fa64SAl Viro { 8377d54fa64SAl Viro struct inode *inode; 83858b6e5e8SMike Kravetz struct resv_map *resv_map = NULL; 8399119a41eSJoonsoo Kim 84058b6e5e8SMike Kravetz /* 84158b6e5e8SMike Kravetz * Reserve maps are only needed for inodes that can have associated 84258b6e5e8SMike Kravetz * page allocations. 84358b6e5e8SMike Kravetz */ 84458b6e5e8SMike Kravetz if (S_ISREG(mode) || S_ISLNK(mode)) { 8459119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 8469119a41eSJoonsoo Kim if (!resv_map) 8479119a41eSJoonsoo Kim return NULL; 84858b6e5e8SMike Kravetz } 8497d54fa64SAl Viro 8507d54fa64SAl Viro inode = new_inode(sb); 8517d54fa64SAl Viro if (inode) { 852ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 853ff62a342SMarc-André Lureau 8547d54fa64SAl Viro inode->i_ino = get_next_ino(); 85521cb47beSChristian Brauner inode_init_owner(&init_user_ns, inode, dir, mode); 856c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 857c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 8581da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 859078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 8609119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 861ff62a342SMarc-André Lureau info->seals = F_SEAL_SEAL; 8621da177e4SLinus Torvalds switch (mode & S_IFMT) { 8631da177e4SLinus Torvalds default: 8641da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 8651da177e4SLinus Torvalds break; 8661da177e4SLinus Torvalds case S_IFREG: 8671da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 8681da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 8691da177e4SLinus Torvalds break; 8701da177e4SLinus Torvalds case S_IFDIR: 8711da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 8721da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 8731da177e4SLinus Torvalds 8741da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 875d8c76e6fSDave Hansen inc_nlink(inode); 8761da177e4SLinus Torvalds break; 8771da177e4SLinus Torvalds case S_IFLNK: 8781da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 87921fc61c7SAl Viro inode_nohighmem(inode); 8801da177e4SLinus Torvalds break; 8811da177e4SLinus Torvalds } 882e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 88358b6e5e8SMike Kravetz } else { 88458b6e5e8SMike Kravetz if (resv_map) 8859119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 88658b6e5e8SMike Kravetz } 8879119a41eSJoonsoo Kim 8881da177e4SLinus Torvalds return inode; 8891da177e4SLinus Torvalds } 8901da177e4SLinus Torvalds 8911da177e4SLinus Torvalds /* 8921da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 8931da177e4SLinus Torvalds */ 8941ab5b82fSPiotr Sarna static int do_hugetlbfs_mknod(struct inode *dir, 8951ab5b82fSPiotr Sarna struct dentry *dentry, 8961ab5b82fSPiotr Sarna umode_t mode, 8971ab5b82fSPiotr Sarna dev_t dev, 8981ab5b82fSPiotr Sarna bool tmpfile) 8991da177e4SLinus Torvalds { 9001da177e4SLinus Torvalds struct inode *inode; 9011da177e4SLinus Torvalds int error = -ENOSPC; 9021da177e4SLinus Torvalds 9037d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 9041da177e4SLinus Torvalds if (inode) { 905078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 9061ab5b82fSPiotr Sarna if (tmpfile) { 9071ab5b82fSPiotr Sarna d_tmpfile(dentry, inode); 9081ab5b82fSPiotr Sarna } else { 9091da177e4SLinus Torvalds d_instantiate(dentry, inode); 9101da177e4SLinus Torvalds dget(dentry);/* Extra count - pin the dentry in core */ 9111ab5b82fSPiotr Sarna } 9121da177e4SLinus Torvalds error = 0; 9131da177e4SLinus Torvalds } 9141da177e4SLinus Torvalds return error; 9151da177e4SLinus Torvalds } 9161da177e4SLinus Torvalds 917549c7297SChristian Brauner static int hugetlbfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 9181ab5b82fSPiotr Sarna struct dentry *dentry, umode_t mode, dev_t dev) 9191ab5b82fSPiotr Sarna { 9201ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode, dev, false); 9211ab5b82fSPiotr Sarna } 9221ab5b82fSPiotr Sarna 923549c7297SChristian Brauner static int hugetlbfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 924549c7297SChristian Brauner struct dentry *dentry, umode_t mode) 9251da177e4SLinus Torvalds { 926549c7297SChristian Brauner int retval = hugetlbfs_mknod(&init_user_ns, dir, dentry, 927549c7297SChristian Brauner mode | S_IFDIR, 0); 9281da177e4SLinus Torvalds if (!retval) 929d8c76e6fSDave Hansen inc_nlink(dir); 9301da177e4SLinus Torvalds return retval; 9311da177e4SLinus Torvalds } 9321da177e4SLinus Torvalds 933549c7297SChristian Brauner static int hugetlbfs_create(struct user_namespace *mnt_userns, 934549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 935549c7297SChristian Brauner umode_t mode, bool excl) 9361da177e4SLinus Torvalds { 937549c7297SChristian Brauner return hugetlbfs_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0); 9381da177e4SLinus Torvalds } 9391da177e4SLinus Torvalds 940549c7297SChristian Brauner static int hugetlbfs_tmpfile(struct user_namespace *mnt_userns, 941549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 942549c7297SChristian Brauner umode_t mode) 9431ab5b82fSPiotr Sarna { 9441ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0, true); 9451ab5b82fSPiotr Sarna } 9461ab5b82fSPiotr Sarna 947549c7297SChristian Brauner static int hugetlbfs_symlink(struct user_namespace *mnt_userns, 948549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 949549c7297SChristian Brauner const char *symname) 9501da177e4SLinus Torvalds { 9511da177e4SLinus Torvalds struct inode *inode; 9521da177e4SLinus Torvalds int error = -ENOSPC; 9531da177e4SLinus Torvalds 9547d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 9551da177e4SLinus Torvalds if (inode) { 9561da177e4SLinus Torvalds int l = strlen(symname)+1; 9571da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 9581da177e4SLinus Torvalds if (!error) { 9591da177e4SLinus Torvalds d_instantiate(dentry, inode); 9601da177e4SLinus Torvalds dget(dentry); 9611da177e4SLinus Torvalds } else 9621da177e4SLinus Torvalds iput(inode); 9631da177e4SLinus Torvalds } 964078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 9651da177e4SLinus Torvalds 9661da177e4SLinus Torvalds return error; 9671da177e4SLinus Torvalds } 9681da177e4SLinus Torvalds 969b890ec2aSMatthew Wilcox (Oracle) #ifdef CONFIG_MIGRATION 970b890ec2aSMatthew Wilcox (Oracle) static int hugetlbfs_migrate_folio(struct address_space *mapping, 971b890ec2aSMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, 972a6bc32b8SMel Gorman enum migrate_mode mode) 973290408d4SNaoya Horiguchi { 974290408d4SNaoya Horiguchi int rc; 975290408d4SNaoya Horiguchi 976b890ec2aSMatthew Wilcox (Oracle) rc = migrate_huge_page_move_mapping(mapping, dst, src); 97778bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 978290408d4SNaoya Horiguchi return rc; 979cb6acd01SMike Kravetz 980b890ec2aSMatthew Wilcox (Oracle) if (hugetlb_page_subpool(&src->page)) { 981b890ec2aSMatthew Wilcox (Oracle) hugetlb_set_page_subpool(&dst->page, 982b890ec2aSMatthew Wilcox (Oracle) hugetlb_page_subpool(&src->page)); 983b890ec2aSMatthew Wilcox (Oracle) hugetlb_set_page_subpool(&src->page, NULL); 984cb6acd01SMike Kravetz } 985cb6acd01SMike Kravetz 9862916ecc0SJérôme Glisse if (mode != MIGRATE_SYNC_NO_COPY) 987b890ec2aSMatthew Wilcox (Oracle) folio_migrate_copy(dst, src); 9882916ecc0SJérôme Glisse else 989b890ec2aSMatthew Wilcox (Oracle) folio_migrate_flags(dst, src); 990290408d4SNaoya Horiguchi 99178bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 992290408d4SNaoya Horiguchi } 993b890ec2aSMatthew Wilcox (Oracle) #else 994b890ec2aSMatthew Wilcox (Oracle) #define hugetlbfs_migrate_folio NULL 995b890ec2aSMatthew Wilcox (Oracle) #endif 996290408d4SNaoya Horiguchi 99778bb9203SNaoya Horiguchi static int hugetlbfs_error_remove_page(struct address_space *mapping, 99878bb9203SNaoya Horiguchi struct page *page) 99978bb9203SNaoya Horiguchi { 100078bb9203SNaoya Horiguchi struct inode *inode = mapping->host; 1001ab615a5bSMike Kravetz pgoff_t index = page->index; 100278bb9203SNaoya Horiguchi 10037e1813d4SMike Kravetz hugetlb_delete_from_page_cache(page); 1004ab615a5bSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) 100578bb9203SNaoya Horiguchi hugetlb_fix_reserve_counts(inode); 1006ab615a5bSMike Kravetz 100778bb9203SNaoya Horiguchi return 0; 100878bb9203SNaoya Horiguchi } 100978bb9203SNaoya Horiguchi 10104a25220dSDavid Howells /* 10114a25220dSDavid Howells * Display the mount options in /proc/mounts. 10124a25220dSDavid Howells */ 10134a25220dSDavid Howells static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) 10144a25220dSDavid Howells { 10154a25220dSDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); 10164a25220dSDavid Howells struct hugepage_subpool *spool = sbinfo->spool; 10174a25220dSDavid Howells unsigned long hpage_size = huge_page_size(sbinfo->hstate); 10184a25220dSDavid Howells unsigned hpage_shift = huge_page_shift(sbinfo->hstate); 10194a25220dSDavid Howells char mod; 10204a25220dSDavid Howells 10214a25220dSDavid Howells if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) 10224a25220dSDavid Howells seq_printf(m, ",uid=%u", 10234a25220dSDavid Howells from_kuid_munged(&init_user_ns, sbinfo->uid)); 10244a25220dSDavid Howells if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) 10254a25220dSDavid Howells seq_printf(m, ",gid=%u", 10264a25220dSDavid Howells from_kgid_munged(&init_user_ns, sbinfo->gid)); 10274a25220dSDavid Howells if (sbinfo->mode != 0755) 10284a25220dSDavid Howells seq_printf(m, ",mode=%o", sbinfo->mode); 10294a25220dSDavid Howells if (sbinfo->max_inodes != -1) 10304a25220dSDavid Howells seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); 10314a25220dSDavid Howells 10324a25220dSDavid Howells hpage_size /= 1024; 10334a25220dSDavid Howells mod = 'K'; 10344a25220dSDavid Howells if (hpage_size >= 1024) { 10354a25220dSDavid Howells hpage_size /= 1024; 10364a25220dSDavid Howells mod = 'M'; 10374a25220dSDavid Howells } 10384a25220dSDavid Howells seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); 10394a25220dSDavid Howells if (spool) { 10404a25220dSDavid Howells if (spool->max_hpages != -1) 10414a25220dSDavid Howells seq_printf(m, ",size=%llu", 10424a25220dSDavid Howells (unsigned long long)spool->max_hpages << hpage_shift); 10434a25220dSDavid Howells if (spool->min_hpages != -1) 10444a25220dSDavid Howells seq_printf(m, ",min_size=%llu", 10454a25220dSDavid Howells (unsigned long long)spool->min_hpages << hpage_shift); 10464a25220dSDavid Howells } 10474a25220dSDavid Howells return 0; 10484a25220dSDavid Howells } 10494a25220dSDavid Howells 1050726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 10511da177e4SLinus Torvalds { 1052726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 10532b0143b5SDavid Howells struct hstate *h = hstate_inode(d_inode(dentry)); 10541da177e4SLinus Torvalds 10551da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 1056a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 10571da177e4SLinus Torvalds if (sbinfo) { 10581da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 105911680763SMiaohe Lin /* If no limits set, just report 0 or -1 for max/free/used 106074a8a65cSDavid Gibson * blocks, like simple_statfs() */ 106190481622SDavid Gibson if (sbinfo->spool) { 106290481622SDavid Gibson long free_pages; 106390481622SDavid Gibson 10644b25f030SMina Almasry spin_lock_irq(&sbinfo->spool->lock); 106590481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 106690481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 106790481622SDavid Gibson - sbinfo->spool->used_hpages; 106890481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 10694b25f030SMina Almasry spin_unlock_irq(&sbinfo->spool->lock); 10701da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 10711da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 107274a8a65cSDavid Gibson } 10731da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 10741da177e4SLinus Torvalds } 10751da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 10761da177e4SLinus Torvalds return 0; 10771da177e4SLinus Torvalds } 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 10801da177e4SLinus Torvalds { 10811da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 10821da177e4SLinus Torvalds 10831da177e4SLinus Torvalds if (sbi) { 10841da177e4SLinus Torvalds sb->s_fs_info = NULL; 108590481622SDavid Gibson 108690481622SDavid Gibson if (sbi->spool) 108790481622SDavid Gibson hugepage_put_subpool(sbi->spool); 108890481622SDavid Gibson 10891da177e4SLinus Torvalds kfree(sbi); 10901da177e4SLinus Torvalds } 10911da177e4SLinus Torvalds } 10921da177e4SLinus Torvalds 109396527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 109496527980SChristoph Hellwig { 109596527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 109696527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 109796527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 109896527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 109996527980SChristoph Hellwig return 0; 110096527980SChristoph Hellwig } 110196527980SChristoph Hellwig sbinfo->free_inodes--; 110296527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 110396527980SChristoph Hellwig } 110496527980SChristoph Hellwig 110596527980SChristoph Hellwig return 1; 110696527980SChristoph Hellwig } 110796527980SChristoph Hellwig 110896527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 110996527980SChristoph Hellwig { 111096527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 111196527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 111296527980SChristoph Hellwig sbinfo->free_inodes++; 111396527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 111496527980SChristoph Hellwig } 111596527980SChristoph Hellwig } 111696527980SChristoph Hellwig 111796527980SChristoph Hellwig 1118e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 11191da177e4SLinus Torvalds 11201da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 11211da177e4SLinus Torvalds { 112296527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 11231da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 11241da177e4SLinus Torvalds 112596527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 11261da177e4SLinus Torvalds return NULL; 1127fd60b288SMuchun Song p = alloc_inode_sb(sb, hugetlbfs_inode_cachep, GFP_KERNEL); 112896527980SChristoph Hellwig if (unlikely(!p)) { 112996527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 113096527980SChristoph Hellwig return NULL; 11311da177e4SLinus Torvalds } 11324742a35dSMike Kravetz 11334742a35dSMike Kravetz /* 11344742a35dSMike Kravetz * Any time after allocation, hugetlbfs_destroy_inode can be called 11354742a35dSMike Kravetz * for the inode. mpol_free_shared_policy is unconditionally called 11364742a35dSMike Kravetz * as part of hugetlbfs_destroy_inode. So, initialize policy here 11374742a35dSMike Kravetz * in case of a quick call to destroy. 11384742a35dSMike Kravetz * 11394742a35dSMike Kravetz * Note that the policy is initialized even if we are creating a 11404742a35dSMike Kravetz * private inode. This simplifies hugetlbfs_destroy_inode. 11414742a35dSMike Kravetz */ 11424742a35dSMike Kravetz mpol_shared_policy_init(&p->policy, NULL); 11434742a35dSMike Kravetz 114496527980SChristoph Hellwig return &p->vfs_inode; 11451da177e4SLinus Torvalds } 11461da177e4SLinus Torvalds 1147b62de322SAl Viro static void hugetlbfs_free_inode(struct inode *inode) 1148fa0d7e3dSNick Piggin { 1149fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 1150fa0d7e3dSNick Piggin } 1151fa0d7e3dSNick Piggin 11521da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 11531da177e4SLinus Torvalds { 115496527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 11551da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 11561da177e4SLinus Torvalds } 11571da177e4SLinus Torvalds 1158f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 1159800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 1160800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 116146de8b97SMatthew Wilcox (Oracle) .dirty_folio = noop_dirty_folio, 1162b890ec2aSMatthew Wilcox (Oracle) .migrate_folio = hugetlbfs_migrate_folio, 116378bb9203SNaoya Horiguchi .error_remove_page = hugetlbfs_error_remove_page, 11641da177e4SLinus Torvalds }; 11651da177e4SLinus Torvalds 116696527980SChristoph Hellwig 116751cc5068SAlexey Dobriyan static void init_once(void *foo) 116896527980SChristoph Hellwig { 116996527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 117096527980SChristoph Hellwig 117196527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 117296527980SChristoph Hellwig } 117396527980SChristoph Hellwig 11744b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 117534d0640eSAl Viro .read_iter = hugetlbfs_read_iter, 11761da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 11771b061d92SChristoph Hellwig .fsync = noop_fsync, 11781da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 11796038f373SArnd Bergmann .llseek = default_llseek, 118070c3547eSMike Kravetz .fallocate = hugetlbfs_fallocate, 11811da177e4SLinus Torvalds }; 11821da177e4SLinus Torvalds 118392e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 11841da177e4SLinus Torvalds .create = hugetlbfs_create, 11851da177e4SLinus Torvalds .lookup = simple_lookup, 11861da177e4SLinus Torvalds .link = simple_link, 11871da177e4SLinus Torvalds .unlink = simple_unlink, 11881da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 11891da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 11901da177e4SLinus Torvalds .rmdir = simple_rmdir, 11911da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 11921da177e4SLinus Torvalds .rename = simple_rename, 11931da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 11941ab5b82fSPiotr Sarna .tmpfile = hugetlbfs_tmpfile, 11951da177e4SLinus Torvalds }; 11961da177e4SLinus Torvalds 119792e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 11981da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 11991da177e4SLinus Torvalds }; 12001da177e4SLinus Torvalds 1201ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 12021da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 1203b62de322SAl Viro .free_inode = hugetlbfs_free_inode, 12041da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 12052bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 12061da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 12071da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 12084a25220dSDavid Howells .show_options = hugetlbfs_show_options, 12091da177e4SLinus Torvalds }; 12101da177e4SLinus Torvalds 12117ca02d0aSMike Kravetz /* 12127ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 12137ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 12147ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 12157ca02d0aSMike Kravetz */ 12164a25220dSDavid Howells static long 12177ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 12184a25220dSDavid Howells enum hugetlbfs_size_type val_type) 12197ca02d0aSMike Kravetz { 12207ca02d0aSMike Kravetz if (val_type == NO_SIZE) 12217ca02d0aSMike Kravetz return -1; 12227ca02d0aSMike Kravetz 12237ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 12247ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 12257ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 12267ca02d0aSMike Kravetz do_div(size_opt, 100); 12277ca02d0aSMike Kravetz } 12287ca02d0aSMike Kravetz 12297ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 12307ca02d0aSMike Kravetz return size_opt; 12317ca02d0aSMike Kravetz } 12327ca02d0aSMike Kravetz 123332021982SDavid Howells /* 123432021982SDavid Howells * Parse one mount parameter. 123532021982SDavid Howells */ 123632021982SDavid Howells static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param) 12371da177e4SLinus Torvalds { 123832021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 123932021982SDavid Howells struct fs_parse_result result; 124032021982SDavid Howells char *rest; 124132021982SDavid Howells unsigned long ps; 124232021982SDavid Howells int opt; 12431da177e4SLinus Torvalds 1244d7167b14SAl Viro opt = fs_parse(fc, hugetlb_fs_parameters, param, &result); 124532021982SDavid Howells if (opt < 0) 124632021982SDavid Howells return opt; 124732021982SDavid Howells 124832021982SDavid Howells switch (opt) { 124932021982SDavid Howells case Opt_uid: 125032021982SDavid Howells ctx->uid = make_kuid(current_user_ns(), result.uint_32); 125132021982SDavid Howells if (!uid_valid(ctx->uid)) 125232021982SDavid Howells goto bad_val; 12531da177e4SLinus Torvalds return 0; 12541da177e4SLinus Torvalds 1255e73a75faSRandy Dunlap case Opt_gid: 125632021982SDavid Howells ctx->gid = make_kgid(current_user_ns(), result.uint_32); 125732021982SDavid Howells if (!gid_valid(ctx->gid)) 1258e73a75faSRandy Dunlap goto bad_val; 125932021982SDavid Howells return 0; 1260e73a75faSRandy Dunlap 1261e73a75faSRandy Dunlap case Opt_mode: 126232021982SDavid Howells ctx->mode = result.uint_32 & 01777U; 126332021982SDavid Howells return 0; 1264e73a75faSRandy Dunlap 126532021982SDavid Howells case Opt_size: 1266e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 126732021982SDavid Howells if (!isdigit(param->string[0])) 1268e73a75faSRandy Dunlap goto bad_val; 126932021982SDavid Howells ctx->max_size_opt = memparse(param->string, &rest); 127032021982SDavid Howells ctx->max_val_type = SIZE_STD; 1271a137e1ccSAndi Kleen if (*rest == '%') 127232021982SDavid Howells ctx->max_val_type = SIZE_PERCENT; 127332021982SDavid Howells return 0; 12741da177e4SLinus Torvalds 1275e73a75faSRandy Dunlap case Opt_nr_inodes: 1276e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 127732021982SDavid Howells if (!isdigit(param->string[0])) 1278e73a75faSRandy Dunlap goto bad_val; 127932021982SDavid Howells ctx->nr_inodes = memparse(param->string, &rest); 128032021982SDavid Howells return 0; 1281e73a75faSRandy Dunlap 128232021982SDavid Howells case Opt_pagesize: 128332021982SDavid Howells ps = memparse(param->string, &rest); 128432021982SDavid Howells ctx->hstate = size_to_hstate(ps); 128532021982SDavid Howells if (!ctx->hstate) { 1286d0036517SMiaohe Lin pr_err("Unsupported page size %lu MB\n", ps / SZ_1M); 1287a137e1ccSAndi Kleen return -EINVAL; 1288a137e1ccSAndi Kleen } 128932021982SDavid Howells return 0; 1290a137e1ccSAndi Kleen 129132021982SDavid Howells case Opt_min_size: 12927ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 129332021982SDavid Howells if (!isdigit(param->string[0])) 12947ca02d0aSMike Kravetz goto bad_val; 129532021982SDavid Howells ctx->min_size_opt = memparse(param->string, &rest); 129632021982SDavid Howells ctx->min_val_type = SIZE_STD; 12977ca02d0aSMike Kravetz if (*rest == '%') 129832021982SDavid Howells ctx->min_val_type = SIZE_PERCENT; 129932021982SDavid Howells return 0; 13007ca02d0aSMike Kravetz 1301e73a75faSRandy Dunlap default: 1302b4c07bceSLee Schermerhorn return -EINVAL; 1303e73a75faSRandy Dunlap } 130432021982SDavid Howells 130532021982SDavid Howells bad_val: 1306b5db30cfSAl Viro return invalfc(fc, "Bad value '%s' for mount option '%s'\n", 130732021982SDavid Howells param->string, param->key); 13081da177e4SLinus Torvalds } 1309a137e1ccSAndi Kleen 13107ca02d0aSMike Kravetz /* 131132021982SDavid Howells * Validate the parsed options. 131232021982SDavid Howells */ 131332021982SDavid Howells static int hugetlbfs_validate(struct fs_context *fc) 131432021982SDavid Howells { 131532021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 131632021982SDavid Howells 131732021982SDavid Howells /* 13187ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 13197ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 13207ca02d0aSMike Kravetz */ 132132021982SDavid Howells ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 132232021982SDavid Howells ctx->max_size_opt, 132332021982SDavid Howells ctx->max_val_type); 132432021982SDavid Howells ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 132532021982SDavid Howells ctx->min_size_opt, 132632021982SDavid Howells ctx->min_val_type); 13277ca02d0aSMike Kravetz 13287ca02d0aSMike Kravetz /* 13297ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 13307ca02d0aSMike Kravetz */ 133132021982SDavid Howells if (ctx->max_val_type > NO_SIZE && 133232021982SDavid Howells ctx->min_hpages > ctx->max_hpages) { 133332021982SDavid Howells pr_err("Minimum size can not be greater than maximum size\n"); 13347ca02d0aSMike Kravetz return -EINVAL; 1335a137e1ccSAndi Kleen } 1336a137e1ccSAndi Kleen 13371da177e4SLinus Torvalds return 0; 13381da177e4SLinus Torvalds } 13391da177e4SLinus Torvalds 13401da177e4SLinus Torvalds static int 134132021982SDavid Howells hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc) 13421da177e4SLinus Torvalds { 134332021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 13441da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 13471da177e4SLinus Torvalds if (!sbinfo) 13481da177e4SLinus Torvalds return -ENOMEM; 13491da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 13501da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 135132021982SDavid Howells sbinfo->hstate = ctx->hstate; 135232021982SDavid Howells sbinfo->max_inodes = ctx->nr_inodes; 135332021982SDavid Howells sbinfo->free_inodes = ctx->nr_inodes; 135490481622SDavid Gibson sbinfo->spool = NULL; 135532021982SDavid Howells sbinfo->uid = ctx->uid; 135632021982SDavid Howells sbinfo->gid = ctx->gid; 135732021982SDavid Howells sbinfo->mode = ctx->mode; 13584a25220dSDavid Howells 13597ca02d0aSMike Kravetz /* 13607ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 13611935ebd3SMiaohe Lin * specified. Any needed reservations (for minimum size) are taken 1362445c8098SMiaohe Lin * when the subpool is created. 13637ca02d0aSMike Kravetz */ 136432021982SDavid Howells if (ctx->max_hpages != -1 || ctx->min_hpages != -1) { 136532021982SDavid Howells sbinfo->spool = hugepage_new_subpool(ctx->hstate, 136632021982SDavid Howells ctx->max_hpages, 136732021982SDavid Howells ctx->min_hpages); 136890481622SDavid Gibson if (!sbinfo->spool) 136990481622SDavid Gibson goto out_free; 137090481622SDavid Gibson } 13711da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 137232021982SDavid Howells sb->s_blocksize = huge_page_size(ctx->hstate); 137332021982SDavid Howells sb->s_blocksize_bits = huge_page_shift(ctx->hstate); 13741da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 13751da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 13761da177e4SLinus Torvalds sb->s_time_gran = 1; 137715568299SMike Kravetz 137815568299SMike Kravetz /* 137915568299SMike Kravetz * Due to the special and limited functionality of hugetlbfs, it does 138015568299SMike Kravetz * not work well as a stacking filesystem. 138115568299SMike Kravetz */ 138215568299SMike Kravetz sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH; 138332021982SDavid Howells sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx)); 138448fde701SAl Viro if (!sb->s_root) 13851da177e4SLinus Torvalds goto out_free; 13861da177e4SLinus Torvalds return 0; 13871da177e4SLinus Torvalds out_free: 138890481622SDavid Gibson kfree(sbinfo->spool); 13891da177e4SLinus Torvalds kfree(sbinfo); 13901da177e4SLinus Torvalds return -ENOMEM; 13911da177e4SLinus Torvalds } 13921da177e4SLinus Torvalds 139332021982SDavid Howells static int hugetlbfs_get_tree(struct fs_context *fc) 13941da177e4SLinus Torvalds { 139532021982SDavid Howells int err = hugetlbfs_validate(fc); 139632021982SDavid Howells if (err) 139732021982SDavid Howells return err; 13982ac295d4SAl Viro return get_tree_nodev(fc, hugetlbfs_fill_super); 139932021982SDavid Howells } 140032021982SDavid Howells 140132021982SDavid Howells static void hugetlbfs_fs_context_free(struct fs_context *fc) 140232021982SDavid Howells { 140332021982SDavid Howells kfree(fc->fs_private); 140432021982SDavid Howells } 140532021982SDavid Howells 140632021982SDavid Howells static const struct fs_context_operations hugetlbfs_fs_context_ops = { 140732021982SDavid Howells .free = hugetlbfs_fs_context_free, 140832021982SDavid Howells .parse_param = hugetlbfs_parse_param, 140932021982SDavid Howells .get_tree = hugetlbfs_get_tree, 141032021982SDavid Howells }; 141132021982SDavid Howells 141232021982SDavid Howells static int hugetlbfs_init_fs_context(struct fs_context *fc) 141332021982SDavid Howells { 141432021982SDavid Howells struct hugetlbfs_fs_context *ctx; 141532021982SDavid Howells 141632021982SDavid Howells ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL); 141732021982SDavid Howells if (!ctx) 141832021982SDavid Howells return -ENOMEM; 141932021982SDavid Howells 142032021982SDavid Howells ctx->max_hpages = -1; /* No limit on size by default */ 142132021982SDavid Howells ctx->nr_inodes = -1; /* No limit on number of inodes by default */ 142232021982SDavid Howells ctx->uid = current_fsuid(); 142332021982SDavid Howells ctx->gid = current_fsgid(); 142432021982SDavid Howells ctx->mode = 0755; 142532021982SDavid Howells ctx->hstate = &default_hstate; 142632021982SDavid Howells ctx->min_hpages = -1; /* No default minimum size */ 142732021982SDavid Howells ctx->max_val_type = NO_SIZE; 142832021982SDavid Howells ctx->min_val_type = NO_SIZE; 142932021982SDavid Howells fc->fs_private = ctx; 143032021982SDavid Howells fc->ops = &hugetlbfs_fs_context_ops; 143132021982SDavid Howells return 0; 14321da177e4SLinus Torvalds } 14331da177e4SLinus Torvalds 14341da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 14351da177e4SLinus Torvalds .name = "hugetlbfs", 143632021982SDavid Howells .init_fs_context = hugetlbfs_init_fs_context, 1437d7167b14SAl Viro .parameters = hugetlb_fs_parameters, 14381da177e4SLinus Torvalds .kill_sb = kill_litter_super, 14391da177e4SLinus Torvalds }; 14401da177e4SLinus Torvalds 144142d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 14421da177e4SLinus Torvalds 1443ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 14441da177e4SLinus Torvalds { 1445a0eb3a05SEric W. Biederman kgid_t shm_group; 1446a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 1447a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 14481da177e4SLinus Torvalds } 14491da177e4SLinus Torvalds 145042d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 145142d7395fSAndi Kleen { 1452af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 145342d7395fSAndi Kleen 145442d7395fSAndi Kleen if (!h) 145542d7395fSAndi Kleen return -1; 145604adbc3fSMiaohe Lin return hstate_index(h); 145742d7395fSAndi Kleen } 145842d7395fSAndi Kleen 1459af73e4d9SNaoya Horiguchi /* 1460af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 1461af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 1462af73e4d9SNaoya Horiguchi */ 1463af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 146483c1fd76Szhangyiru vm_flags_t acctflag, int creat_flags, 146583c1fd76Szhangyiru int page_size_log) 14661da177e4SLinus Torvalds { 14671da177e4SLinus Torvalds struct inode *inode; 1468e68375c8SAl Viro struct vfsmount *mnt; 146942d7395fSAndi Kleen int hstate_idx; 1470e68375c8SAl Viro struct file *file; 147142d7395fSAndi Kleen 147242d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 147342d7395fSAndi Kleen if (hstate_idx < 0) 147442d7395fSAndi Kleen return ERR_PTR(-ENODEV); 14751da177e4SLinus Torvalds 1476e68375c8SAl Viro mnt = hugetlbfs_vfsmount[hstate_idx]; 1477e68375c8SAl Viro if (!mnt) 14785bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 14795bc98594SAkinobu Mita 1480ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 148183c1fd76Szhangyiru struct ucounts *ucounts = current_ucounts(); 148283c1fd76Szhangyiru 148383c1fd76Szhangyiru if (user_shm_lock(size, ucounts)) { 148483c1fd76Szhangyiru pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is obsolete\n", 148521a3c273SDavid Rientjes current->comm, current->pid); 148683c1fd76Szhangyiru user_shm_unlock(size, ucounts); 14872584e517SRavikiran G Thirumalai } 148883c1fd76Szhangyiru return ERR_PTR(-EPERM); 1489353d5c30SHugh Dickins } 14901da177e4SLinus Torvalds 149139b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 1492e68375c8SAl Viro inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0); 14931da177e4SLinus Torvalds if (!inode) 1494e68375c8SAl Viro goto out; 1495e1832f29SStephen Smalley if (creat_flags == HUGETLB_SHMFS_INODE) 1496e1832f29SStephen Smalley inode->i_flags |= S_PRIVATE; 14971da177e4SLinus Torvalds 14981da177e4SLinus Torvalds inode->i_size = size; 14996d6b77f1SMiklos Szeredi clear_nlink(inode); 1500ce8d2cdfSDave Hansen 150133b8f84aSMike Kravetz if (!hugetlb_reserve_pages(inode, 0, 1502e68375c8SAl Viro size >> huge_page_shift(hstate_inode(inode)), NULL, 1503e68375c8SAl Viro acctflag)) 1504e68375c8SAl Viro file = ERR_PTR(-ENOMEM); 1505e68375c8SAl Viro else 1506e68375c8SAl Viro file = alloc_file_pseudo(inode, mnt, name, O_RDWR, 1507ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 1508e68375c8SAl Viro if (!IS_ERR(file)) 15091da177e4SLinus Torvalds return file; 15101da177e4SLinus Torvalds 1511b45b5bd6SDavid Gibson iput(inode); 1512e68375c8SAl Viro out: 151339b65252SAnatol Pomozov return file; 15141da177e4SLinus Torvalds } 15151da177e4SLinus Torvalds 151632021982SDavid Howells static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h) 151732021982SDavid Howells { 151832021982SDavid Howells struct fs_context *fc; 151932021982SDavid Howells struct vfsmount *mnt; 152032021982SDavid Howells 152132021982SDavid Howells fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT); 152232021982SDavid Howells if (IS_ERR(fc)) { 152332021982SDavid Howells mnt = ERR_CAST(fc); 152432021982SDavid Howells } else { 152532021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 152632021982SDavid Howells ctx->hstate = h; 152732021982SDavid Howells mnt = fc_mount(fc); 152832021982SDavid Howells put_fs_context(fc); 152932021982SDavid Howells } 153032021982SDavid Howells if (IS_ERR(mnt)) 1531a25fddceSMiaohe Lin pr_err("Cannot mount internal hugetlbfs for page size %luK", 1532d0036517SMiaohe Lin huge_page_size(h) / SZ_1K); 153332021982SDavid Howells return mnt; 153432021982SDavid Howells } 153532021982SDavid Howells 15361da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 15371da177e4SLinus Torvalds { 153832021982SDavid Howells struct vfsmount *mnt; 153942d7395fSAndi Kleen struct hstate *h; 15401da177e4SLinus Torvalds int error; 154142d7395fSAndi Kleen int i; 15421da177e4SLinus Torvalds 1543457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 15449b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1545457c1b27SNishanth Aravamudan return -ENOTSUPP; 1546457c1b27SNishanth Aravamudan } 1547457c1b27SNishanth Aravamudan 1548d1d5e05fSHillf Danton error = -ENOMEM; 15491da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 15501da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 15515d097056SVladimir Davydov 0, SLAB_ACCOUNT, init_once); 15521da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 15538fc312b3SMike Kravetz goto out; 15541da177e4SLinus Torvalds 15551da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 15561da177e4SLinus Torvalds if (error) 15578fc312b3SMike Kravetz goto out_free; 15581da177e4SLinus Torvalds 15598fc312b3SMike Kravetz /* default hstate mount is required */ 15603b2275a8SMiaohe Lin mnt = mount_one_hugetlbfs(&default_hstate); 15618fc312b3SMike Kravetz if (IS_ERR(mnt)) { 15628fc312b3SMike Kravetz error = PTR_ERR(mnt); 15638fc312b3SMike Kravetz goto out_unreg; 15648fc312b3SMike Kravetz } 15658fc312b3SMike Kravetz hugetlbfs_vfsmount[default_hstate_idx] = mnt; 15668fc312b3SMike Kravetz 15678fc312b3SMike Kravetz /* other hstates are optional */ 156842d7395fSAndi Kleen i = 0; 156942d7395fSAndi Kleen for_each_hstate(h) { 157015f0ec94SJan Stancek if (i == default_hstate_idx) { 157115f0ec94SJan Stancek i++; 15728fc312b3SMike Kravetz continue; 157315f0ec94SJan Stancek } 15748fc312b3SMike Kravetz 157532021982SDavid Howells mnt = mount_one_hugetlbfs(h); 15768fc312b3SMike Kravetz if (IS_ERR(mnt)) 15778fc312b3SMike Kravetz hugetlbfs_vfsmount[i] = NULL; 15788fc312b3SMike Kravetz else 157932021982SDavid Howells hugetlbfs_vfsmount[i] = mnt; 158042d7395fSAndi Kleen i++; 158142d7395fSAndi Kleen } 158232021982SDavid Howells 158342d7395fSAndi Kleen return 0; 15841da177e4SLinus Torvalds 15858fc312b3SMike Kravetz out_unreg: 15868fc312b3SMike Kravetz (void)unregister_filesystem(&hugetlbfs_fs_type); 15878fc312b3SMike Kravetz out_free: 15881da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 15898fc312b3SMike Kravetz out: 15901da177e4SLinus Torvalds return error; 15911da177e4SLinus Torvalds } 15923e89e1c5SPaul Gortmaker fs_initcall(init_hugetlbfs_fs) 1593