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), 80e0f7e2b2SMike Kravetz fsparam_u32oct("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 11163489f8eSMike Kravetz /* 11263489f8eSMike Kravetz * Mask used when checking the page offset value passed in via system 11363489f8eSMike Kravetz * calls. This value will be converted to a loff_t which is signed. 11463489f8eSMike Kravetz * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the 11563489f8eSMike Kravetz * value. The extra bit (- 1 in the shift value) is to take the sign 11663489f8eSMike Kravetz * bit into account. 11763489f8eSMike Kravetz */ 11863489f8eSMike Kravetz #define PGOFF_LOFFT_MAX \ 11963489f8eSMike Kravetz (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1))) 12063489f8eSMike Kravetz 1211da177e4SLinus Torvalds static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) 1221da177e4SLinus Torvalds { 123496ad9aaSAl Viro struct inode *inode = file_inode(file); 12422247efdSPeter Xu struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 1251da177e4SLinus Torvalds loff_t len, vma_len; 1261da177e4SLinus Torvalds int ret; 127a5516438SAndi Kleen struct hstate *h = hstate_file(file); 1281da177e4SLinus Torvalds 12968589bc3SHugh Dickins /* 130dec4ad86SDavid Gibson * vma address alignment (but not the pgoff alignment) has 131dec4ad86SDavid Gibson * already been checked by prepare_hugepage_range. If you add 132dec4ad86SDavid Gibson * any error returns here, do so after setting VM_HUGETLB, so 133dec4ad86SDavid Gibson * is_vm_hugetlb_page tests below unmap_region go the right 13445e55300SPeter Collingbourne * way when do_mmap unwinds (may be important on powerpc 135dec4ad86SDavid Gibson * and ia64). 13668589bc3SHugh Dickins */ 137a2fce914SNaoya Horiguchi vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; 13868589bc3SHugh Dickins vma->vm_ops = &hugetlb_vm_ops; 1391da177e4SLinus Torvalds 14022247efdSPeter Xu ret = seal_check_future_write(info->seals, vma); 14122247efdSPeter Xu if (ret) 14222247efdSPeter Xu return ret; 14322247efdSPeter Xu 144045c7a3fSMike Kravetz /* 14563489f8eSMike Kravetz * page based offset in vm_pgoff could be sufficiently large to 1465df63c2aSMike Kravetz * overflow a loff_t when converted to byte offset. This can 1475df63c2aSMike Kravetz * only happen on architectures where sizeof(loff_t) == 1485df63c2aSMike Kravetz * sizeof(unsigned long). So, only check in those instances. 149045c7a3fSMike Kravetz */ 1505df63c2aSMike Kravetz if (sizeof(unsigned long) == sizeof(loff_t)) { 15163489f8eSMike Kravetz if (vma->vm_pgoff & PGOFF_LOFFT_MAX) 152045c7a3fSMike Kravetz return -EINVAL; 1535df63c2aSMike Kravetz } 154045c7a3fSMike Kravetz 15563489f8eSMike Kravetz /* must be huge page aligned */ 1562b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 157dec4ad86SDavid Gibson return -EINVAL; 158dec4ad86SDavid Gibson 1591da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 160045c7a3fSMike Kravetz len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 161045c7a3fSMike Kravetz /* check for overflow */ 162045c7a3fSMike Kravetz if (len < vma_len) 163045c7a3fSMike Kravetz return -EINVAL; 1641da177e4SLinus Torvalds 1655955102cSAl Viro inode_lock(inode); 1661da177e4SLinus Torvalds file_accessed(file); 1671da177e4SLinus Torvalds 1681da177e4SLinus Torvalds ret = -ENOMEM; 16933b8f84aSMike Kravetz if (!hugetlb_reserve_pages(inode, 170a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1715a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1725a6fe125SMel Gorman vma->vm_flags)) 173b45b5bd6SDavid Gibson goto out; 174b45b5bd6SDavid Gibson 1754c887265SAdam Litke ret = 0; 176b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 177045c7a3fSMike Kravetz i_size_write(inode, len); 1781da177e4SLinus Torvalds out: 1795955102cSAl Viro inode_unlock(inode); 1801da177e4SLinus Torvalds 1811da177e4SLinus Torvalds return ret; 1821da177e4SLinus Torvalds } 1831da177e4SLinus Torvalds 1841da177e4SLinus Torvalds /* 1853e4e28c5SMichel Lespinasse * Called under mmap_write_lock(mm). 1861da177e4SLinus Torvalds */ 1871da177e4SLinus Torvalds 1881da177e4SLinus Torvalds static unsigned long 18988590253SShijie Hu hugetlb_get_unmapped_area_bottomup(struct file *file, unsigned long addr, 19088590253SShijie Hu unsigned long len, unsigned long pgoff, unsigned long flags) 19188590253SShijie Hu { 19288590253SShijie Hu struct hstate *h = hstate_file(file); 19388590253SShijie Hu struct vm_unmapped_area_info info; 19488590253SShijie Hu 19588590253SShijie Hu info.flags = 0; 19688590253SShijie Hu info.length = len; 19788590253SShijie Hu info.low_limit = current->mm->mmap_base; 1982cb4de08SChristophe Leroy info.high_limit = arch_get_mmap_end(addr, len, flags); 19988590253SShijie Hu info.align_mask = PAGE_MASK & ~huge_page_mask(h); 20088590253SShijie Hu info.align_offset = 0; 20188590253SShijie Hu return vm_unmapped_area(&info); 20288590253SShijie Hu } 20388590253SShijie Hu 20488590253SShijie Hu static unsigned long 20588590253SShijie Hu hugetlb_get_unmapped_area_topdown(struct file *file, unsigned long addr, 20688590253SShijie Hu unsigned long len, unsigned long pgoff, unsigned long flags) 20788590253SShijie Hu { 20888590253SShijie Hu struct hstate *h = hstate_file(file); 20988590253SShijie Hu struct vm_unmapped_area_info info; 21088590253SShijie Hu 21188590253SShijie Hu info.flags = VM_UNMAPPED_AREA_TOPDOWN; 21288590253SShijie Hu info.length = len; 21388590253SShijie Hu info.low_limit = max(PAGE_SIZE, mmap_min_addr); 2145f24d5a5SChristophe Leroy info.high_limit = arch_get_mmap_base(addr, current->mm->mmap_base); 21588590253SShijie Hu info.align_mask = PAGE_MASK & ~huge_page_mask(h); 21688590253SShijie Hu info.align_offset = 0; 21788590253SShijie Hu addr = vm_unmapped_area(&info); 21888590253SShijie Hu 21988590253SShijie Hu /* 22088590253SShijie Hu * A failed mmap() very likely causes application failure, 22188590253SShijie Hu * so fall back to the bottom-up function here. This scenario 22288590253SShijie Hu * can happen with large stack limits and large mmap() 22388590253SShijie Hu * allocations. 22488590253SShijie Hu */ 22588590253SShijie Hu if (unlikely(offset_in_page(addr))) { 22688590253SShijie Hu VM_BUG_ON(addr != -ENOMEM); 22788590253SShijie Hu info.flags = 0; 22888590253SShijie Hu info.low_limit = current->mm->mmap_base; 2292cb4de08SChristophe Leroy info.high_limit = arch_get_mmap_end(addr, len, flags); 23088590253SShijie Hu addr = vm_unmapped_area(&info); 23188590253SShijie Hu } 23288590253SShijie Hu 23388590253SShijie Hu return addr; 23488590253SShijie Hu } 23588590253SShijie Hu 2364b439e25SChristophe Leroy unsigned long 2374b439e25SChristophe Leroy generic_hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 2384b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 2394b439e25SChristophe Leroy unsigned long flags) 2401da177e4SLinus Torvalds { 2411da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2421da177e4SLinus Torvalds struct vm_area_struct *vma; 243a5516438SAndi Kleen struct hstate *h = hstate_file(file); 2442cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 2451da177e4SLinus Torvalds 246a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 2471da177e4SLinus Torvalds return -EINVAL; 2481da177e4SLinus Torvalds if (len > TASK_SIZE) 2491da177e4SLinus Torvalds return -ENOMEM; 2501da177e4SLinus Torvalds 251036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 252a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 253036e0856SBenjamin Herrenschmidt return -EINVAL; 254036e0856SBenjamin Herrenschmidt return addr; 255036e0856SBenjamin Herrenschmidt } 256036e0856SBenjamin Herrenschmidt 2571da177e4SLinus Torvalds if (addr) { 258a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 2591da177e4SLinus Torvalds vma = find_vma(mm, addr); 2605f24d5a5SChristophe Leroy if (mmap_end - len >= addr && 2611be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma))) 2621da177e4SLinus Torvalds return addr; 2631da177e4SLinus Torvalds } 2641da177e4SLinus Torvalds 26588590253SShijie Hu /* 26688590253SShijie Hu * Use mm->get_unmapped_area value as a hint to use topdown routine. 26788590253SShijie Hu * If architectures have special needs, they should define their own 26888590253SShijie Hu * version of hugetlb_get_unmapped_area. 26988590253SShijie Hu */ 27088590253SShijie Hu if (mm->get_unmapped_area == arch_get_unmapped_area_topdown) 27188590253SShijie Hu return hugetlb_get_unmapped_area_topdown(file, addr, len, 27288590253SShijie Hu pgoff, flags); 27388590253SShijie Hu return hugetlb_get_unmapped_area_bottomup(file, addr, len, 27488590253SShijie Hu pgoff, flags); 2751da177e4SLinus Torvalds } 2764b439e25SChristophe Leroy 2774b439e25SChristophe Leroy #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 2784b439e25SChristophe Leroy static unsigned long 2794b439e25SChristophe Leroy hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 2804b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 2814b439e25SChristophe Leroy unsigned long flags) 2824b439e25SChristophe Leroy { 2834b439e25SChristophe Leroy return generic_hugetlb_get_unmapped_area(file, addr, len, pgoff, flags); 2844b439e25SChristophe Leroy } 2851da177e4SLinus Torvalds #endif 2861da177e4SLinus Torvalds 28734d0640eSAl Viro static size_t 288e63e1e5aSBadari Pulavarty hugetlbfs_read_actor(struct page *page, unsigned long offset, 28934d0640eSAl Viro struct iov_iter *to, unsigned long size) 290e63e1e5aSBadari Pulavarty { 29134d0640eSAl Viro size_t copied = 0; 292e63e1e5aSBadari Pulavarty int i, chunksize; 293e63e1e5aSBadari Pulavarty 294e63e1e5aSBadari Pulavarty /* Find which 4k chunk and offset with in that chunk */ 29509cbfeafSKirill A. Shutemov i = offset >> PAGE_SHIFT; 29609cbfeafSKirill A. Shutemov offset = offset & ~PAGE_MASK; 297e63e1e5aSBadari Pulavarty 298e63e1e5aSBadari Pulavarty while (size) { 29934d0640eSAl Viro size_t n; 30009cbfeafSKirill A. Shutemov chunksize = PAGE_SIZE; 301e63e1e5aSBadari Pulavarty if (offset) 302e63e1e5aSBadari Pulavarty chunksize -= offset; 303e63e1e5aSBadari Pulavarty if (chunksize > size) 304e63e1e5aSBadari Pulavarty chunksize = size; 30534d0640eSAl Viro n = copy_page_to_iter(&page[i], offset, chunksize, to); 30634d0640eSAl Viro copied += n; 30734d0640eSAl Viro if (n != chunksize) 30834d0640eSAl Viro return copied; 309e63e1e5aSBadari Pulavarty offset = 0; 310e63e1e5aSBadari Pulavarty size -= chunksize; 311e63e1e5aSBadari Pulavarty i++; 312e63e1e5aSBadari Pulavarty } 31334d0640eSAl Viro return copied; 314e63e1e5aSBadari Pulavarty } 315e63e1e5aSBadari Pulavarty 316e63e1e5aSBadari Pulavarty /* 317e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 318c7e285e3SMiaohe Lin * data. Its *very* similar to generic_file_buffered_read(), we can't use that 319ea1754a0SKirill A. Shutemov * since it has PAGE_SIZE assumptions. 320e63e1e5aSBadari Pulavarty */ 32134d0640eSAl Viro static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) 322e63e1e5aSBadari Pulavarty { 32334d0640eSAl Viro struct file *file = iocb->ki_filp; 32434d0640eSAl Viro struct hstate *h = hstate_file(file); 32534d0640eSAl Viro struct address_space *mapping = file->f_mapping; 326e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 32734d0640eSAl Viro unsigned long index = iocb->ki_pos >> huge_page_shift(h); 32834d0640eSAl Viro unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); 329e63e1e5aSBadari Pulavarty unsigned long end_index; 330e63e1e5aSBadari Pulavarty loff_t isize; 331e63e1e5aSBadari Pulavarty ssize_t retval = 0; 332e63e1e5aSBadari Pulavarty 33334d0640eSAl Viro while (iov_iter_count(to)) { 334e63e1e5aSBadari Pulavarty struct page *page; 33534d0640eSAl Viro size_t nr, copied; 336e63e1e5aSBadari Pulavarty 337e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 338a5516438SAndi Kleen nr = huge_page_size(h); 339a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 340a05b0855SAneesh Kumar K.V if (!isize) 34134d0640eSAl Viro break; 342a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 343e63e1e5aSBadari Pulavarty if (index > end_index) 34434d0640eSAl Viro break; 34534d0640eSAl Viro if (index == end_index) { 346a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 347a05b0855SAneesh Kumar K.V if (nr <= offset) 34834d0640eSAl Viro break; 349e63e1e5aSBadari Pulavarty } 350e63e1e5aSBadari Pulavarty nr = nr - offset; 351e63e1e5aSBadari Pulavarty 352e63e1e5aSBadari Pulavarty /* Find the page */ 353a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 354e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 355e63e1e5aSBadari Pulavarty /* 356e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 357e63e1e5aSBadari Pulavarty * length of the hole or request. 358e63e1e5aSBadari Pulavarty */ 35934d0640eSAl Viro copied = iov_iter_zero(nr, to); 360e63e1e5aSBadari Pulavarty } else { 361a05b0855SAneesh Kumar K.V unlock_page(page); 362a05b0855SAneesh Kumar K.V 363e63e1e5aSBadari Pulavarty /* 364e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 365e63e1e5aSBadari Pulavarty */ 36634d0640eSAl Viro copied = hugetlbfs_read_actor(page, offset, to, nr); 36709cbfeafSKirill A. Shutemov put_page(page); 368e63e1e5aSBadari Pulavarty } 36934d0640eSAl Viro offset += copied; 37034d0640eSAl Viro retval += copied; 37134d0640eSAl Viro if (copied != nr && iov_iter_count(to)) { 37234d0640eSAl Viro if (!retval) 37334d0640eSAl Viro retval = -EFAULT; 374e63e1e5aSBadari Pulavarty break; 375e63e1e5aSBadari Pulavarty } 37634d0640eSAl Viro index += offset >> huge_page_shift(h); 37734d0640eSAl Viro offset &= ~huge_page_mask(h); 37834d0640eSAl Viro } 37934d0640eSAl Viro iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; 380e63e1e5aSBadari Pulavarty return retval; 381e63e1e5aSBadari Pulavarty } 382e63e1e5aSBadari Pulavarty 383800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 384800d15a5SNick Piggin struct address_space *mapping, 3859d6b0cd7SMatthew Wilcox (Oracle) loff_t pos, unsigned len, 386800d15a5SNick Piggin struct page **pagep, void **fsdata) 3871da177e4SLinus Torvalds { 3881da177e4SLinus Torvalds return -EINVAL; 3891da177e4SLinus Torvalds } 3901da177e4SLinus Torvalds 391800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 392800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 393800d15a5SNick Piggin struct page *page, void *fsdata) 3941da177e4SLinus Torvalds { 395800d15a5SNick Piggin BUG(); 3961da177e4SLinus Torvalds return -EINVAL; 3971da177e4SLinus Torvalds } 3981da177e4SLinus Torvalds 399b5cec28dSMike Kravetz static void remove_huge_page(struct page *page) 4001da177e4SLinus Torvalds { 401b9ea2515SKonstantin Khlebnikov ClearPageDirty(page); 4021da177e4SLinus Torvalds ClearPageUptodate(page); 403bd65cb86SMinchan Kim delete_from_page_cache(page); 4041da177e4SLinus Torvalds } 4051da177e4SLinus Torvalds 4064aae8d1cSMike Kravetz static void 40705e90bd0SPeter Xu hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end, 40805e90bd0SPeter Xu zap_flags_t zap_flags) 4094aae8d1cSMike Kravetz { 4104aae8d1cSMike Kravetz struct vm_area_struct *vma; 4114aae8d1cSMike Kravetz 4124aae8d1cSMike Kravetz /* 413d6aba4c8SSean Christopherson * end == 0 indicates that the entire range after start should be 414d6aba4c8SSean Christopherson * unmapped. Note, end is exclusive, whereas the interval tree takes 415d6aba4c8SSean Christopherson * an inclusive "last". 4164aae8d1cSMike Kravetz */ 417d6aba4c8SSean Christopherson vma_interval_tree_foreach(vma, root, start, end ? end - 1 : ULONG_MAX) { 4184aae8d1cSMike Kravetz unsigned long v_offset; 4194aae8d1cSMike Kravetz unsigned long v_end; 4204aae8d1cSMike Kravetz 4214aae8d1cSMike Kravetz /* 4224aae8d1cSMike Kravetz * Can the expression below overflow on 32-bit arches? 4234aae8d1cSMike Kravetz * No, because the interval tree returns us only those vmas 4244aae8d1cSMike Kravetz * which overlap the truncated area starting at pgoff, 4254aae8d1cSMike Kravetz * and no vma on a 32-bit arch can span beyond the 4GB. 4264aae8d1cSMike Kravetz */ 4274aae8d1cSMike Kravetz if (vma->vm_pgoff < start) 4284aae8d1cSMike Kravetz v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; 4294aae8d1cSMike Kravetz else 4304aae8d1cSMike Kravetz v_offset = 0; 4314aae8d1cSMike Kravetz 4324aae8d1cSMike Kravetz if (!end) 4334aae8d1cSMike Kravetz v_end = vma->vm_end; 4344aae8d1cSMike Kravetz else { 4354aae8d1cSMike Kravetz v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) 4364aae8d1cSMike Kravetz + vma->vm_start; 4374aae8d1cSMike Kravetz if (v_end > vma->vm_end) 4384aae8d1cSMike Kravetz v_end = vma->vm_end; 4394aae8d1cSMike Kravetz } 4404aae8d1cSMike Kravetz 4414aae8d1cSMike Kravetz unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, 44205e90bd0SPeter Xu NULL, zap_flags); 4434aae8d1cSMike Kravetz } 4444aae8d1cSMike Kravetz } 445b5cec28dSMike Kravetz 446b5cec28dSMike Kravetz /* 447b5cec28dSMike Kravetz * remove_inode_hugepages handles two distinct cases: truncation and hole 448b5cec28dSMike Kravetz * punch. There are subtle differences in operation for each case. 4494aae8d1cSMike Kravetz * 450b5cec28dSMike Kravetz * truncation is indicated by end of range being LLONG_MAX 451b5cec28dSMike Kravetz * In this case, we first scan the range and release found pages. 4521935ebd3SMiaohe Lin * After releasing pages, hugetlb_unreserve_pages cleans up region/reserve 453e7c58097SMike Kravetz * maps and global counts. Page faults can not race with truncation 45487bf91d3SMike Kravetz * in this routine. hugetlb_no_page() holds i_mmap_rwsem and prevents 45587bf91d3SMike Kravetz * page faults in the truncated range by checking i_size. i_size is 45687bf91d3SMike Kravetz * modified while holding i_mmap_rwsem. 457b5cec28dSMike Kravetz * hole punch is indicated if end is not LLONG_MAX 458b5cec28dSMike Kravetz * In the hole punch case we scan the range and release found pages. 4591935ebd3SMiaohe Lin * Only when releasing a page is the associated region/reserve map 4601935ebd3SMiaohe Lin * deleted. The region/reserve map for ranges without associated 461e7c58097SMike Kravetz * pages are not modified. Page faults can race with hole punch. 462e7c58097SMike Kravetz * This is indicated if we find a mapped page. 463b5cec28dSMike Kravetz * Note: If the passed end of range value is beyond the end of file, but 464b5cec28dSMike Kravetz * not LLONG_MAX this routine still performs a hole punch operation. 465b5cec28dSMike Kravetz */ 466b5cec28dSMike Kravetz static void remove_inode_hugepages(struct inode *inode, loff_t lstart, 467b5cec28dSMike Kravetz loff_t lend) 4681da177e4SLinus Torvalds { 469a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 470b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 471a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 472b5cec28dSMike Kravetz const pgoff_t end = lend >> huge_page_shift(h); 4731508062eSMatthew Wilcox (Oracle) struct folio_batch fbatch; 474d72dc8a2SJan Kara pgoff_t next, index; 475a43a8c39SChen, Kenneth W int i, freed = 0; 476b5cec28dSMike Kravetz bool truncate_op = (lend == LLONG_MAX); 4771da177e4SLinus Torvalds 4781508062eSMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 4791da177e4SLinus Torvalds next = start; 4801508062eSMatthew Wilcox (Oracle) while (filemap_get_folios(mapping, &next, end - 1, &fbatch)) { 4811508062eSMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); ++i) { 4821508062eSMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 483d4241a04SMiaohe Lin u32 hash = 0; 484b5cec28dSMike Kravetz 4851508062eSMatthew Wilcox (Oracle) index = folio->index; 48687bf91d3SMike Kravetz if (!truncate_op) { 48787bf91d3SMike Kravetz /* 48887bf91d3SMike Kravetz * Only need to hold the fault mutex in the 48987bf91d3SMike Kravetz * hole punch case. This prevents races with 49087bf91d3SMike Kravetz * page faults. Races are not possible in the 49187bf91d3SMike Kravetz * case of truncation. 49287bf91d3SMike Kravetz */ 493d4241a04SMiaohe Lin hash = hugetlb_fault_mutex_hash(mapping, index); 494e7c58097SMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 49587bf91d3SMike Kravetz } 496e7c58097SMike Kravetz 497b5cec28dSMike Kravetz /* 4981508062eSMatthew Wilcox (Oracle) * If folio is mapped, it was faulted in after being 499e7c58097SMike Kravetz * unmapped in caller. Unmap (again) now after taking 500e7c58097SMike Kravetz * the fault mutex. The mutex will prevent faults 5011508062eSMatthew Wilcox (Oracle) * until we finish removing the folio. 502e7c58097SMike Kravetz * 503e7c58097SMike Kravetz * This race can only happen in the hole punch case. 504e7c58097SMike Kravetz * Getting here in a truncate operation is a bug. 505b5cec28dSMike Kravetz */ 5061508062eSMatthew Wilcox (Oracle) if (unlikely(folio_mapped(folio))) { 507e7c58097SMike Kravetz BUG_ON(truncate_op); 508e7c58097SMike Kravetz 509c0d0381aSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 510e7c58097SMike Kravetz i_mmap_lock_write(mapping); 511c0d0381aSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 512e7c58097SMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 513e7c58097SMike Kravetz index * pages_per_huge_page(h), 51405e90bd0SPeter Xu (index + 1) * pages_per_huge_page(h), 51505e90bd0SPeter Xu ZAP_FLAG_DROP_MARKER); 516e7c58097SMike Kravetz i_mmap_unlock_write(mapping); 517e7c58097SMike Kravetz } 5184aae8d1cSMike Kravetz 5191508062eSMatthew Wilcox (Oracle) folio_lock(folio); 5204aae8d1cSMike Kravetz /* 5214aae8d1cSMike Kravetz * We must free the huge page and remove from page 5224aae8d1cSMike Kravetz * cache (remove_huge_page) BEFORE removing the 5234aae8d1cSMike Kravetz * region/reserve map (hugetlb_unreserve_pages). In 5244aae8d1cSMike Kravetz * rare out of memory conditions, removal of the 52572e2936cSzhong jiang * region/reserve map could fail. Correspondingly, 52672e2936cSzhong jiang * the subpool and global reserve usage count can need 52772e2936cSzhong jiang * to be adjusted. 5284aae8d1cSMike Kravetz */ 5291508062eSMatthew Wilcox (Oracle) VM_BUG_ON(HPageRestoreReserve(&folio->page)); 5301508062eSMatthew Wilcox (Oracle) remove_huge_page(&folio->page); 531b5cec28dSMike Kravetz freed++; 532b5cec28dSMike Kravetz if (!truncate_op) { 5334aae8d1cSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, 534d72dc8a2SJan Kara index, index + 1, 1))) 53572e2936cSzhong jiang hugetlb_fix_reserve_counts(inode); 536b5cec28dSMike Kravetz } 537b5cec28dSMike Kravetz 5381508062eSMatthew Wilcox (Oracle) folio_unlock(folio); 53987bf91d3SMike Kravetz if (!truncate_op) 540e7c58097SMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 5411da177e4SLinus Torvalds } 5421508062eSMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 5431817889eSMike Kravetz cond_resched(); 5441da177e4SLinus Torvalds } 545b5cec28dSMike Kravetz 546b5cec28dSMike Kravetz if (truncate_op) 547b5cec28dSMike Kravetz (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); 5481da177e4SLinus Torvalds } 5491da177e4SLinus Torvalds 5502bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 5511da177e4SLinus Torvalds { 5529119a41eSJoonsoo Kim struct resv_map *resv_map; 5539119a41eSJoonsoo Kim 554b5cec28dSMike Kravetz remove_inode_hugepages(inode, 0, LLONG_MAX); 555f27a5136SMike Kravetz 556f27a5136SMike Kravetz /* 557f27a5136SMike Kravetz * Get the resv_map from the address space embedded in the inode. 558f27a5136SMike Kravetz * This is the address space which points to any resv_map allocated 559f27a5136SMike Kravetz * at inode creation time. If this is a device special inode, 560f27a5136SMike Kravetz * i_mapping may not point to the original address space. 561f27a5136SMike Kravetz */ 562f27a5136SMike Kravetz resv_map = (struct resv_map *)(&inode->i_data)->private_data; 563f27a5136SMike Kravetz /* Only regular and link inodes have associated reserve maps */ 5649119a41eSJoonsoo Kim if (resv_map) 5659119a41eSJoonsoo Kim resv_map_release(&resv_map->refs); 566dbd5768fSJan Kara clear_inode(inode); 567149f4211SChristoph Hellwig } 568149f4211SChristoph Hellwig 569e5d319deSMiaohe Lin static void hugetlb_vmtruncate(struct inode *inode, loff_t offset) 5701da177e4SLinus Torvalds { 571856fc295SHugh Dickins pgoff_t pgoff; 5721da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 573a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 5741da177e4SLinus Torvalds 575a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 576856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 5771da177e4SLinus Torvalds 57883cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 57987bf91d3SMike Kravetz i_size_write(inode, offset); 580f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 58105e90bd0SPeter Xu hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0, 58205e90bd0SPeter Xu ZAP_FLAG_DROP_MARKER); 583c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 584e7c58097SMike Kravetz remove_inode_hugepages(inode, offset, LLONG_MAX); 5851da177e4SLinus Torvalds } 5861da177e4SLinus Torvalds 58770c3547eSMike Kravetz static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) 58870c3547eSMike Kravetz { 58970c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 59070c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 59170c3547eSMike Kravetz loff_t hole_start, hole_end; 59270c3547eSMike Kravetz 59370c3547eSMike Kravetz /* 59470c3547eSMike Kravetz * For hole punch round up the beginning offset of the hole and 59570c3547eSMike Kravetz * round down the end. 59670c3547eSMike Kravetz */ 59770c3547eSMike Kravetz hole_start = round_up(offset, hpage_size); 59870c3547eSMike Kravetz hole_end = round_down(offset + len, hpage_size); 59970c3547eSMike Kravetz 60070c3547eSMike Kravetz if (hole_end > hole_start) { 60170c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 602ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 60370c3547eSMike Kravetz 6045955102cSAl Viro inode_lock(inode); 605ff62a342SMarc-André Lureau 606398c0da7SMiaohe Lin /* protected by i_rwsem */ 607ab3948f5SJoel Fernandes (Google) if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { 608ff62a342SMarc-André Lureau inode_unlock(inode); 609ff62a342SMarc-André Lureau return -EPERM; 610ff62a342SMarc-André Lureau } 611ff62a342SMarc-André Lureau 61270c3547eSMike Kravetz i_mmap_lock_write(mapping); 613f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 61470c3547eSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 61570c3547eSMike Kravetz hole_start >> PAGE_SHIFT, 61605e90bd0SPeter Xu hole_end >> PAGE_SHIFT, 0); 617c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 618e7c58097SMike Kravetz remove_inode_hugepages(inode, hole_start, hole_end); 6195955102cSAl Viro inode_unlock(inode); 62070c3547eSMike Kravetz } 62170c3547eSMike Kravetz 62270c3547eSMike Kravetz return 0; 62370c3547eSMike Kravetz } 62470c3547eSMike Kravetz 62570c3547eSMike Kravetz static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, 62670c3547eSMike Kravetz loff_t len) 62770c3547eSMike Kravetz { 62870c3547eSMike Kravetz struct inode *inode = file_inode(file); 629ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 63070c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 63170c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 63270c3547eSMike Kravetz struct vm_area_struct pseudo_vma; 63370c3547eSMike Kravetz struct mm_struct *mm = current->mm; 63470c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 63570c3547eSMike Kravetz unsigned long hpage_shift = huge_page_shift(h); 63670c3547eSMike Kravetz pgoff_t start, index, end; 63770c3547eSMike Kravetz int error; 63870c3547eSMike Kravetz u32 hash; 63970c3547eSMike Kravetz 64070c3547eSMike Kravetz if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 64170c3547eSMike Kravetz return -EOPNOTSUPP; 64270c3547eSMike Kravetz 64370c3547eSMike Kravetz if (mode & FALLOC_FL_PUNCH_HOLE) 64470c3547eSMike Kravetz return hugetlbfs_punch_hole(inode, offset, len); 64570c3547eSMike Kravetz 64670c3547eSMike Kravetz /* 64770c3547eSMike Kravetz * Default preallocate case. 64870c3547eSMike Kravetz * For this range, start is rounded down and end is rounded up 64970c3547eSMike Kravetz * as well as being converted to page offsets. 65070c3547eSMike Kravetz */ 65170c3547eSMike Kravetz start = offset >> hpage_shift; 65270c3547eSMike Kravetz end = (offset + len + hpage_size - 1) >> hpage_shift; 65370c3547eSMike Kravetz 6545955102cSAl Viro inode_lock(inode); 65570c3547eSMike Kravetz 65670c3547eSMike Kravetz /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ 65770c3547eSMike Kravetz error = inode_newsize_ok(inode, offset + len); 65870c3547eSMike Kravetz if (error) 65970c3547eSMike Kravetz goto out; 66070c3547eSMike Kravetz 661ff62a342SMarc-André Lureau if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { 662ff62a342SMarc-André Lureau error = -EPERM; 663ff62a342SMarc-André Lureau goto out; 664ff62a342SMarc-André Lureau } 665ff62a342SMarc-André Lureau 66670c3547eSMike Kravetz /* 66770c3547eSMike Kravetz * Initialize a pseudo vma as this is required by the huge page 66870c3547eSMike Kravetz * allocation routines. If NUMA is configured, use page index 66970c3547eSMike Kravetz * as input to create an allocation policy. 67070c3547eSMike Kravetz */ 6712c4541e2SKirill A. Shutemov vma_init(&pseudo_vma, mm); 67270c3547eSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 67370c3547eSMike Kravetz pseudo_vma.vm_file = file; 67470c3547eSMike Kravetz 67570c3547eSMike Kravetz for (index = start; index < end; index++) { 67670c3547eSMike Kravetz /* 67770c3547eSMike Kravetz * This is supposed to be the vaddr where the page is being 67870c3547eSMike Kravetz * faulted in, but we have no vaddr here. 67970c3547eSMike Kravetz */ 68070c3547eSMike Kravetz struct page *page; 68170c3547eSMike Kravetz unsigned long addr; 68270c3547eSMike Kravetz 68370c3547eSMike Kravetz cond_resched(); 68470c3547eSMike Kravetz 68570c3547eSMike Kravetz /* 68670c3547eSMike Kravetz * fallocate(2) manpage permits EINTR; we may have been 68770c3547eSMike Kravetz * interrupted because we are using up too much memory. 68870c3547eSMike Kravetz */ 68970c3547eSMike Kravetz if (signal_pending(current)) { 69070c3547eSMike Kravetz error = -EINTR; 69170c3547eSMike Kravetz break; 69270c3547eSMike Kravetz } 69370c3547eSMike Kravetz 69470c3547eSMike Kravetz /* Set numa allocation policy based on index */ 69570c3547eSMike Kravetz hugetlb_set_vma_policy(&pseudo_vma, inode, index); 69670c3547eSMike Kravetz 69770c3547eSMike Kravetz /* addr is the offset within the file (zero based) */ 69870c3547eSMike Kravetz addr = index * hpage_size; 69970c3547eSMike Kravetz 70087bf91d3SMike Kravetz /* 70187bf91d3SMike Kravetz * fault mutex taken here, protects against fault path 70287bf91d3SMike Kravetz * and hole punch. inode_lock previously taken protects 70387bf91d3SMike Kravetz * against truncation. 70487bf91d3SMike Kravetz */ 705188b04a7SWei Yang hash = hugetlb_fault_mutex_hash(mapping, index); 70670c3547eSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 70770c3547eSMike Kravetz 70870c3547eSMike Kravetz /* See if already present in mapping to avoid alloc/free */ 70970c3547eSMike Kravetz page = find_get_page(mapping, index); 71070c3547eSMike Kravetz if (page) { 71170c3547eSMike Kravetz put_page(page); 71270c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 71370c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 71470c3547eSMike Kravetz continue; 71570c3547eSMike Kravetz } 71670c3547eSMike Kravetz 71788ce3fefSMiaohe Lin /* 71888ce3fefSMiaohe Lin * Allocate page without setting the avoid_reserve argument. 71988ce3fefSMiaohe Lin * There certainly are no reserves associated with the 72088ce3fefSMiaohe Lin * pseudo_vma. However, there could be shared mappings with 72188ce3fefSMiaohe Lin * reserves for the file at the inode level. If we fallocate 72288ce3fefSMiaohe Lin * pages in these areas, we need to consume the reserves 72388ce3fefSMiaohe Lin * to keep reservation accounting consistent. 72488ce3fefSMiaohe Lin */ 72588ce3fefSMiaohe Lin page = alloc_huge_page(&pseudo_vma, addr, 0); 72670c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 72770c3547eSMike Kravetz if (IS_ERR(page)) { 72870c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 72970c3547eSMike Kravetz error = PTR_ERR(page); 73070c3547eSMike Kravetz goto out; 73170c3547eSMike Kravetz } 73270c3547eSMike Kravetz clear_huge_page(page, addr, pages_per_huge_page(h)); 73370c3547eSMike Kravetz __SetPageUptodate(page); 73470c3547eSMike Kravetz error = huge_add_to_page_cache(page, mapping, index); 73570c3547eSMike Kravetz if (unlikely(error)) { 736846be085SMike Kravetz restore_reserve_on_error(h, &pseudo_vma, addr, page); 73770c3547eSMike Kravetz put_page(page); 73870c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 73970c3547eSMike Kravetz goto out; 74070c3547eSMike Kravetz } 74170c3547eSMike Kravetz 74270c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 74370c3547eSMike Kravetz 7448f251a3dSMike Kravetz SetHPageMigratable(page); 74570c3547eSMike Kravetz /* 746d9ef44deSMatthew Wilcox (Oracle) * unlock_page because locked by huge_add_to_page_cache() 747585fc0d2SMuchun Song * put_page() due to reference from alloc_huge_page() 74870c3547eSMike Kravetz */ 74970c3547eSMike Kravetz unlock_page(page); 75072639e6dSNadav Amit put_page(page); 75170c3547eSMike Kravetz } 75270c3547eSMike Kravetz 75370c3547eSMike Kravetz if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) 75470c3547eSMike Kravetz i_size_write(inode, offset + len); 755078cd827SDeepa Dinamani inode->i_ctime = current_time(inode); 75670c3547eSMike Kravetz out: 7575955102cSAl Viro inode_unlock(inode); 75870c3547eSMike Kravetz return error; 75970c3547eSMike Kravetz } 76070c3547eSMike Kravetz 761549c7297SChristian Brauner static int hugetlbfs_setattr(struct user_namespace *mnt_userns, 762549c7297SChristian Brauner struct dentry *dentry, struct iattr *attr) 7631da177e4SLinus Torvalds { 7642b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 765a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 7661da177e4SLinus Torvalds int error; 7671da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 768ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 7691da177e4SLinus Torvalds 7702f221d6fSChristian Brauner error = setattr_prepare(&init_user_ns, dentry, attr); 7711da177e4SLinus Torvalds if (error) 7721025774cSChristoph Hellwig return error; 7731da177e4SLinus Torvalds 7741da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 775ff62a342SMarc-André Lureau loff_t oldsize = inode->i_size; 776ff62a342SMarc-André Lureau loff_t newsize = attr->ia_size; 777ff62a342SMarc-André Lureau 778ff62a342SMarc-André Lureau if (newsize & ~huge_page_mask(h)) 7791025774cSChristoph Hellwig return -EINVAL; 780398c0da7SMiaohe Lin /* protected by i_rwsem */ 781ff62a342SMarc-André Lureau if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || 782ff62a342SMarc-André Lureau (newsize > oldsize && (info->seals & F_SEAL_GROW))) 783ff62a342SMarc-André Lureau return -EPERM; 784e5d319deSMiaohe Lin hugetlb_vmtruncate(inode, newsize); 7851da177e4SLinus Torvalds } 7861da177e4SLinus Torvalds 7872f221d6fSChristian Brauner setattr_copy(&init_user_ns, inode, attr); 7881025774cSChristoph Hellwig mark_inode_dirty(inode); 7891025774cSChristoph Hellwig return 0; 7901025774cSChristoph Hellwig } 7911025774cSChristoph Hellwig 7927d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 79332021982SDavid Howells struct hugetlbfs_fs_context *ctx) 7941da177e4SLinus Torvalds { 7951da177e4SLinus Torvalds struct inode *inode; 7961da177e4SLinus Torvalds 7971da177e4SLinus Torvalds inode = new_inode(sb); 7981da177e4SLinus Torvalds if (inode) { 79985fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 80032021982SDavid Howells inode->i_mode = S_IFDIR | ctx->mode; 80132021982SDavid Howells inode->i_uid = ctx->uid; 80232021982SDavid Howells inode->i_gid = ctx->gid; 803078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 8047d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 8057d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 8067d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 8077d54fa64SAl Viro inc_nlink(inode); 80865ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 8097d54fa64SAl Viro } 8107d54fa64SAl Viro return inode; 8117d54fa64SAl Viro } 8127d54fa64SAl Viro 813b610ded7SMichal Hocko /* 814c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 815b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 81688f306b6SKirill A. Shutemov * annotation because huge_pmd_share() does an allocation under hugetlb's 817c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 818b610ded7SMichal Hocko */ 819c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 820b610ded7SMichal Hocko 8217d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 8227d54fa64SAl Viro struct inode *dir, 82318df2252SAl Viro umode_t mode, dev_t dev) 8247d54fa64SAl Viro { 8257d54fa64SAl Viro struct inode *inode; 82658b6e5e8SMike Kravetz struct resv_map *resv_map = NULL; 8279119a41eSJoonsoo Kim 82858b6e5e8SMike Kravetz /* 82958b6e5e8SMike Kravetz * Reserve maps are only needed for inodes that can have associated 83058b6e5e8SMike Kravetz * page allocations. 83158b6e5e8SMike Kravetz */ 83258b6e5e8SMike Kravetz if (S_ISREG(mode) || S_ISLNK(mode)) { 8339119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 8349119a41eSJoonsoo Kim if (!resv_map) 8359119a41eSJoonsoo Kim return NULL; 83658b6e5e8SMike Kravetz } 8377d54fa64SAl Viro 8387d54fa64SAl Viro inode = new_inode(sb); 8397d54fa64SAl Viro if (inode) { 840ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 841ff62a342SMarc-André Lureau 8427d54fa64SAl Viro inode->i_ino = get_next_ino(); 84321cb47beSChristian Brauner inode_init_owner(&init_user_ns, inode, dir, mode); 844c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 845c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 8461da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 847078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 8489119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 849ff62a342SMarc-André Lureau info->seals = F_SEAL_SEAL; 8501da177e4SLinus Torvalds switch (mode & S_IFMT) { 8511da177e4SLinus Torvalds default: 8521da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 8531da177e4SLinus Torvalds break; 8541da177e4SLinus Torvalds case S_IFREG: 8551da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 8561da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 8571da177e4SLinus Torvalds break; 8581da177e4SLinus Torvalds case S_IFDIR: 8591da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 8601da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 8611da177e4SLinus Torvalds 8621da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 863d8c76e6fSDave Hansen inc_nlink(inode); 8641da177e4SLinus Torvalds break; 8651da177e4SLinus Torvalds case S_IFLNK: 8661da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 86721fc61c7SAl Viro inode_nohighmem(inode); 8681da177e4SLinus Torvalds break; 8691da177e4SLinus Torvalds } 870e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 87158b6e5e8SMike Kravetz } else { 87258b6e5e8SMike Kravetz if (resv_map) 8739119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 87458b6e5e8SMike Kravetz } 8759119a41eSJoonsoo Kim 8761da177e4SLinus Torvalds return inode; 8771da177e4SLinus Torvalds } 8781da177e4SLinus Torvalds 8791da177e4SLinus Torvalds /* 8801da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 8811da177e4SLinus Torvalds */ 8821ab5b82fSPiotr Sarna static int do_hugetlbfs_mknod(struct inode *dir, 8831ab5b82fSPiotr Sarna struct dentry *dentry, 8841ab5b82fSPiotr Sarna umode_t mode, 8851ab5b82fSPiotr Sarna dev_t dev, 8861ab5b82fSPiotr Sarna bool tmpfile) 8871da177e4SLinus Torvalds { 8881da177e4SLinus Torvalds struct inode *inode; 8891da177e4SLinus Torvalds int error = -ENOSPC; 8901da177e4SLinus Torvalds 8917d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 8921da177e4SLinus Torvalds if (inode) { 893078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 8941ab5b82fSPiotr Sarna if (tmpfile) { 8951ab5b82fSPiotr Sarna d_tmpfile(dentry, inode); 8961ab5b82fSPiotr Sarna } else { 8971da177e4SLinus Torvalds d_instantiate(dentry, inode); 8981da177e4SLinus Torvalds dget(dentry);/* Extra count - pin the dentry in core */ 8991ab5b82fSPiotr Sarna } 9001da177e4SLinus Torvalds error = 0; 9011da177e4SLinus Torvalds } 9021da177e4SLinus Torvalds return error; 9031da177e4SLinus Torvalds } 9041da177e4SLinus Torvalds 905549c7297SChristian Brauner static int hugetlbfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 9061ab5b82fSPiotr Sarna struct dentry *dentry, umode_t mode, dev_t dev) 9071ab5b82fSPiotr Sarna { 9081ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode, dev, false); 9091ab5b82fSPiotr Sarna } 9101ab5b82fSPiotr Sarna 911549c7297SChristian Brauner static int hugetlbfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 912549c7297SChristian Brauner struct dentry *dentry, umode_t mode) 9131da177e4SLinus Torvalds { 914549c7297SChristian Brauner int retval = hugetlbfs_mknod(&init_user_ns, dir, dentry, 915549c7297SChristian Brauner mode | S_IFDIR, 0); 9161da177e4SLinus Torvalds if (!retval) 917d8c76e6fSDave Hansen inc_nlink(dir); 9181da177e4SLinus Torvalds return retval; 9191da177e4SLinus Torvalds } 9201da177e4SLinus Torvalds 921549c7297SChristian Brauner static int hugetlbfs_create(struct user_namespace *mnt_userns, 922549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 923549c7297SChristian Brauner umode_t mode, bool excl) 9241da177e4SLinus Torvalds { 925549c7297SChristian Brauner return hugetlbfs_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0); 9261da177e4SLinus Torvalds } 9271da177e4SLinus Torvalds 928549c7297SChristian Brauner static int hugetlbfs_tmpfile(struct user_namespace *mnt_userns, 929549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 930549c7297SChristian Brauner umode_t mode) 9311ab5b82fSPiotr Sarna { 9321ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0, true); 9331ab5b82fSPiotr Sarna } 9341ab5b82fSPiotr Sarna 935549c7297SChristian Brauner static int hugetlbfs_symlink(struct user_namespace *mnt_userns, 936549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 937549c7297SChristian Brauner const char *symname) 9381da177e4SLinus Torvalds { 9391da177e4SLinus Torvalds struct inode *inode; 9401da177e4SLinus Torvalds int error = -ENOSPC; 9411da177e4SLinus Torvalds 9427d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 9431da177e4SLinus Torvalds if (inode) { 9441da177e4SLinus Torvalds int l = strlen(symname)+1; 9451da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 9461da177e4SLinus Torvalds if (!error) { 9471da177e4SLinus Torvalds d_instantiate(dentry, inode); 9481da177e4SLinus Torvalds dget(dentry); 9491da177e4SLinus Torvalds } else 9501da177e4SLinus Torvalds iput(inode); 9511da177e4SLinus Torvalds } 952078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 9531da177e4SLinus Torvalds 9541da177e4SLinus Torvalds return error; 9551da177e4SLinus Torvalds } 9561da177e4SLinus Torvalds 957*b890ec2aSMatthew Wilcox (Oracle) #ifdef CONFIG_MIGRATION 958*b890ec2aSMatthew Wilcox (Oracle) static int hugetlbfs_migrate_folio(struct address_space *mapping, 959*b890ec2aSMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, 960a6bc32b8SMel Gorman enum migrate_mode mode) 961290408d4SNaoya Horiguchi { 962290408d4SNaoya Horiguchi int rc; 963290408d4SNaoya Horiguchi 964*b890ec2aSMatthew Wilcox (Oracle) rc = migrate_huge_page_move_mapping(mapping, dst, src); 96578bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 966290408d4SNaoya Horiguchi return rc; 967cb6acd01SMike Kravetz 968*b890ec2aSMatthew Wilcox (Oracle) if (hugetlb_page_subpool(&src->page)) { 969*b890ec2aSMatthew Wilcox (Oracle) hugetlb_set_page_subpool(&dst->page, 970*b890ec2aSMatthew Wilcox (Oracle) hugetlb_page_subpool(&src->page)); 971*b890ec2aSMatthew Wilcox (Oracle) hugetlb_set_page_subpool(&src->page, NULL); 972cb6acd01SMike Kravetz } 973cb6acd01SMike Kravetz 9742916ecc0SJérôme Glisse if (mode != MIGRATE_SYNC_NO_COPY) 975*b890ec2aSMatthew Wilcox (Oracle) folio_migrate_copy(dst, src); 9762916ecc0SJérôme Glisse else 977*b890ec2aSMatthew Wilcox (Oracle) folio_migrate_flags(dst, src); 978290408d4SNaoya Horiguchi 97978bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 980290408d4SNaoya Horiguchi } 981*b890ec2aSMatthew Wilcox (Oracle) #else 982*b890ec2aSMatthew Wilcox (Oracle) #define hugetlbfs_migrate_folio NULL 983*b890ec2aSMatthew Wilcox (Oracle) #endif 984290408d4SNaoya Horiguchi 98578bb9203SNaoya Horiguchi static int hugetlbfs_error_remove_page(struct address_space *mapping, 98678bb9203SNaoya Horiguchi struct page *page) 98778bb9203SNaoya Horiguchi { 98878bb9203SNaoya Horiguchi struct inode *inode = mapping->host; 989ab615a5bSMike Kravetz pgoff_t index = page->index; 99078bb9203SNaoya Horiguchi 99178bb9203SNaoya Horiguchi remove_huge_page(page); 992ab615a5bSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) 99378bb9203SNaoya Horiguchi hugetlb_fix_reserve_counts(inode); 994ab615a5bSMike Kravetz 99578bb9203SNaoya Horiguchi return 0; 99678bb9203SNaoya Horiguchi } 99778bb9203SNaoya Horiguchi 9984a25220dSDavid Howells /* 9994a25220dSDavid Howells * Display the mount options in /proc/mounts. 10004a25220dSDavid Howells */ 10014a25220dSDavid Howells static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) 10024a25220dSDavid Howells { 10034a25220dSDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); 10044a25220dSDavid Howells struct hugepage_subpool *spool = sbinfo->spool; 10054a25220dSDavid Howells unsigned long hpage_size = huge_page_size(sbinfo->hstate); 10064a25220dSDavid Howells unsigned hpage_shift = huge_page_shift(sbinfo->hstate); 10074a25220dSDavid Howells char mod; 10084a25220dSDavid Howells 10094a25220dSDavid Howells if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) 10104a25220dSDavid Howells seq_printf(m, ",uid=%u", 10114a25220dSDavid Howells from_kuid_munged(&init_user_ns, sbinfo->uid)); 10124a25220dSDavid Howells if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) 10134a25220dSDavid Howells seq_printf(m, ",gid=%u", 10144a25220dSDavid Howells from_kgid_munged(&init_user_ns, sbinfo->gid)); 10154a25220dSDavid Howells if (sbinfo->mode != 0755) 10164a25220dSDavid Howells seq_printf(m, ",mode=%o", sbinfo->mode); 10174a25220dSDavid Howells if (sbinfo->max_inodes != -1) 10184a25220dSDavid Howells seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); 10194a25220dSDavid Howells 10204a25220dSDavid Howells hpage_size /= 1024; 10214a25220dSDavid Howells mod = 'K'; 10224a25220dSDavid Howells if (hpage_size >= 1024) { 10234a25220dSDavid Howells hpage_size /= 1024; 10244a25220dSDavid Howells mod = 'M'; 10254a25220dSDavid Howells } 10264a25220dSDavid Howells seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); 10274a25220dSDavid Howells if (spool) { 10284a25220dSDavid Howells if (spool->max_hpages != -1) 10294a25220dSDavid Howells seq_printf(m, ",size=%llu", 10304a25220dSDavid Howells (unsigned long long)spool->max_hpages << hpage_shift); 10314a25220dSDavid Howells if (spool->min_hpages != -1) 10324a25220dSDavid Howells seq_printf(m, ",min_size=%llu", 10334a25220dSDavid Howells (unsigned long long)spool->min_hpages << hpage_shift); 10344a25220dSDavid Howells } 10354a25220dSDavid Howells return 0; 10364a25220dSDavid Howells } 10374a25220dSDavid Howells 1038726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 10391da177e4SLinus Torvalds { 1040726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 10412b0143b5SDavid Howells struct hstate *h = hstate_inode(d_inode(dentry)); 10421da177e4SLinus Torvalds 10431da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 1044a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 10451da177e4SLinus Torvalds if (sbinfo) { 10461da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 104774a8a65cSDavid Gibson /* If no limits set, just report 0 for max/free/used 104874a8a65cSDavid Gibson * blocks, like simple_statfs() */ 104990481622SDavid Gibson if (sbinfo->spool) { 105090481622SDavid Gibson long free_pages; 105190481622SDavid Gibson 10524b25f030SMina Almasry spin_lock_irq(&sbinfo->spool->lock); 105390481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 105490481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 105590481622SDavid Gibson - sbinfo->spool->used_hpages; 105690481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 10574b25f030SMina Almasry spin_unlock_irq(&sbinfo->spool->lock); 10581da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 10591da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 106074a8a65cSDavid Gibson } 10611da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 10621da177e4SLinus Torvalds } 10631da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 10641da177e4SLinus Torvalds return 0; 10651da177e4SLinus Torvalds } 10661da177e4SLinus Torvalds 10671da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 10681da177e4SLinus Torvalds { 10691da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 10701da177e4SLinus Torvalds 10711da177e4SLinus Torvalds if (sbi) { 10721da177e4SLinus Torvalds sb->s_fs_info = NULL; 107390481622SDavid Gibson 107490481622SDavid Gibson if (sbi->spool) 107590481622SDavid Gibson hugepage_put_subpool(sbi->spool); 107690481622SDavid Gibson 10771da177e4SLinus Torvalds kfree(sbi); 10781da177e4SLinus Torvalds } 10791da177e4SLinus Torvalds } 10801da177e4SLinus Torvalds 108196527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 108296527980SChristoph Hellwig { 108396527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 108496527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 108596527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 108696527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 108796527980SChristoph Hellwig return 0; 108896527980SChristoph Hellwig } 108996527980SChristoph Hellwig sbinfo->free_inodes--; 109096527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 109196527980SChristoph Hellwig } 109296527980SChristoph Hellwig 109396527980SChristoph Hellwig return 1; 109496527980SChristoph Hellwig } 109596527980SChristoph Hellwig 109696527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 109796527980SChristoph Hellwig { 109896527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 109996527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 110096527980SChristoph Hellwig sbinfo->free_inodes++; 110196527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 110296527980SChristoph Hellwig } 110396527980SChristoph Hellwig } 110496527980SChristoph Hellwig 110596527980SChristoph Hellwig 1106e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 11071da177e4SLinus Torvalds 11081da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 11091da177e4SLinus Torvalds { 111096527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 11111da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 11121da177e4SLinus Torvalds 111396527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 11141da177e4SLinus Torvalds return NULL; 1115fd60b288SMuchun Song p = alloc_inode_sb(sb, hugetlbfs_inode_cachep, GFP_KERNEL); 111696527980SChristoph Hellwig if (unlikely(!p)) { 111796527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 111896527980SChristoph Hellwig return NULL; 11191da177e4SLinus Torvalds } 11204742a35dSMike Kravetz 11214742a35dSMike Kravetz /* 11224742a35dSMike Kravetz * Any time after allocation, hugetlbfs_destroy_inode can be called 11234742a35dSMike Kravetz * for the inode. mpol_free_shared_policy is unconditionally called 11244742a35dSMike Kravetz * as part of hugetlbfs_destroy_inode. So, initialize policy here 11254742a35dSMike Kravetz * in case of a quick call to destroy. 11264742a35dSMike Kravetz * 11274742a35dSMike Kravetz * Note that the policy is initialized even if we are creating a 11284742a35dSMike Kravetz * private inode. This simplifies hugetlbfs_destroy_inode. 11294742a35dSMike Kravetz */ 11304742a35dSMike Kravetz mpol_shared_policy_init(&p->policy, NULL); 11314742a35dSMike Kravetz 113296527980SChristoph Hellwig return &p->vfs_inode; 11331da177e4SLinus Torvalds } 11341da177e4SLinus Torvalds 1135b62de322SAl Viro static void hugetlbfs_free_inode(struct inode *inode) 1136fa0d7e3dSNick Piggin { 1137fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 1138fa0d7e3dSNick Piggin } 1139fa0d7e3dSNick Piggin 11401da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 11411da177e4SLinus Torvalds { 114296527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 11431da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 11441da177e4SLinus Torvalds } 11451da177e4SLinus Torvalds 1146f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 1147800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 1148800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 114946de8b97SMatthew Wilcox (Oracle) .dirty_folio = noop_dirty_folio, 1150*b890ec2aSMatthew Wilcox (Oracle) .migrate_folio = hugetlbfs_migrate_folio, 115178bb9203SNaoya Horiguchi .error_remove_page = hugetlbfs_error_remove_page, 11521da177e4SLinus Torvalds }; 11531da177e4SLinus Torvalds 115496527980SChristoph Hellwig 115551cc5068SAlexey Dobriyan static void init_once(void *foo) 115696527980SChristoph Hellwig { 115796527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 115896527980SChristoph Hellwig 115996527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 116096527980SChristoph Hellwig } 116196527980SChristoph Hellwig 11624b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 116334d0640eSAl Viro .read_iter = hugetlbfs_read_iter, 11641da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 11651b061d92SChristoph Hellwig .fsync = noop_fsync, 11661da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 11676038f373SArnd Bergmann .llseek = default_llseek, 116870c3547eSMike Kravetz .fallocate = hugetlbfs_fallocate, 11691da177e4SLinus Torvalds }; 11701da177e4SLinus Torvalds 117192e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 11721da177e4SLinus Torvalds .create = hugetlbfs_create, 11731da177e4SLinus Torvalds .lookup = simple_lookup, 11741da177e4SLinus Torvalds .link = simple_link, 11751da177e4SLinus Torvalds .unlink = simple_unlink, 11761da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 11771da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 11781da177e4SLinus Torvalds .rmdir = simple_rmdir, 11791da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 11801da177e4SLinus Torvalds .rename = simple_rename, 11811da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 11821ab5b82fSPiotr Sarna .tmpfile = hugetlbfs_tmpfile, 11831da177e4SLinus Torvalds }; 11841da177e4SLinus Torvalds 118592e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 11861da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 11871da177e4SLinus Torvalds }; 11881da177e4SLinus Torvalds 1189ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 11901da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 1191b62de322SAl Viro .free_inode = hugetlbfs_free_inode, 11921da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 11932bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 11941da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 11951da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 11964a25220dSDavid Howells .show_options = hugetlbfs_show_options, 11971da177e4SLinus Torvalds }; 11981da177e4SLinus Torvalds 11997ca02d0aSMike Kravetz /* 12007ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 12017ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 12027ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 12037ca02d0aSMike Kravetz */ 12044a25220dSDavid Howells static long 12057ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 12064a25220dSDavid Howells enum hugetlbfs_size_type val_type) 12077ca02d0aSMike Kravetz { 12087ca02d0aSMike Kravetz if (val_type == NO_SIZE) 12097ca02d0aSMike Kravetz return -1; 12107ca02d0aSMike Kravetz 12117ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 12127ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 12137ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 12147ca02d0aSMike Kravetz do_div(size_opt, 100); 12157ca02d0aSMike Kravetz } 12167ca02d0aSMike Kravetz 12177ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 12187ca02d0aSMike Kravetz return size_opt; 12197ca02d0aSMike Kravetz } 12207ca02d0aSMike Kravetz 122132021982SDavid Howells /* 122232021982SDavid Howells * Parse one mount parameter. 122332021982SDavid Howells */ 122432021982SDavid Howells static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param) 12251da177e4SLinus Torvalds { 122632021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 122732021982SDavid Howells struct fs_parse_result result; 122832021982SDavid Howells char *rest; 122932021982SDavid Howells unsigned long ps; 123032021982SDavid Howells int opt; 12311da177e4SLinus Torvalds 1232d7167b14SAl Viro opt = fs_parse(fc, hugetlb_fs_parameters, param, &result); 123332021982SDavid Howells if (opt < 0) 123432021982SDavid Howells return opt; 123532021982SDavid Howells 123632021982SDavid Howells switch (opt) { 123732021982SDavid Howells case Opt_uid: 123832021982SDavid Howells ctx->uid = make_kuid(current_user_ns(), result.uint_32); 123932021982SDavid Howells if (!uid_valid(ctx->uid)) 124032021982SDavid Howells goto bad_val; 12411da177e4SLinus Torvalds return 0; 12421da177e4SLinus Torvalds 1243e73a75faSRandy Dunlap case Opt_gid: 124432021982SDavid Howells ctx->gid = make_kgid(current_user_ns(), result.uint_32); 124532021982SDavid Howells if (!gid_valid(ctx->gid)) 1246e73a75faSRandy Dunlap goto bad_val; 124732021982SDavid Howells return 0; 1248e73a75faSRandy Dunlap 1249e73a75faSRandy Dunlap case Opt_mode: 125032021982SDavid Howells ctx->mode = result.uint_32 & 01777U; 125132021982SDavid Howells return 0; 1252e73a75faSRandy Dunlap 125332021982SDavid Howells case Opt_size: 1254e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 125532021982SDavid Howells if (!isdigit(param->string[0])) 1256e73a75faSRandy Dunlap goto bad_val; 125732021982SDavid Howells ctx->max_size_opt = memparse(param->string, &rest); 125832021982SDavid Howells ctx->max_val_type = SIZE_STD; 1259a137e1ccSAndi Kleen if (*rest == '%') 126032021982SDavid Howells ctx->max_val_type = SIZE_PERCENT; 126132021982SDavid Howells return 0; 12621da177e4SLinus Torvalds 1263e73a75faSRandy Dunlap case Opt_nr_inodes: 1264e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 126532021982SDavid Howells if (!isdigit(param->string[0])) 1266e73a75faSRandy Dunlap goto bad_val; 126732021982SDavid Howells ctx->nr_inodes = memparse(param->string, &rest); 126832021982SDavid Howells return 0; 1269e73a75faSRandy Dunlap 127032021982SDavid Howells case Opt_pagesize: 127132021982SDavid Howells ps = memparse(param->string, &rest); 127232021982SDavid Howells ctx->hstate = size_to_hstate(ps); 127332021982SDavid Howells if (!ctx->hstate) { 127432021982SDavid Howells pr_err("Unsupported page size %lu MB\n", ps >> 20); 1275a137e1ccSAndi Kleen return -EINVAL; 1276a137e1ccSAndi Kleen } 127732021982SDavid Howells return 0; 1278a137e1ccSAndi Kleen 127932021982SDavid Howells case Opt_min_size: 12807ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 128132021982SDavid Howells if (!isdigit(param->string[0])) 12827ca02d0aSMike Kravetz goto bad_val; 128332021982SDavid Howells ctx->min_size_opt = memparse(param->string, &rest); 128432021982SDavid Howells ctx->min_val_type = SIZE_STD; 12857ca02d0aSMike Kravetz if (*rest == '%') 128632021982SDavid Howells ctx->min_val_type = SIZE_PERCENT; 128732021982SDavid Howells return 0; 12887ca02d0aSMike Kravetz 1289e73a75faSRandy Dunlap default: 1290b4c07bceSLee Schermerhorn return -EINVAL; 1291e73a75faSRandy Dunlap } 129232021982SDavid Howells 129332021982SDavid Howells bad_val: 1294b5db30cfSAl Viro return invalfc(fc, "Bad value '%s' for mount option '%s'\n", 129532021982SDavid Howells param->string, param->key); 12961da177e4SLinus Torvalds } 1297a137e1ccSAndi Kleen 12987ca02d0aSMike Kravetz /* 129932021982SDavid Howells * Validate the parsed options. 130032021982SDavid Howells */ 130132021982SDavid Howells static int hugetlbfs_validate(struct fs_context *fc) 130232021982SDavid Howells { 130332021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 130432021982SDavid Howells 130532021982SDavid Howells /* 13067ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 13077ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 13087ca02d0aSMike Kravetz */ 130932021982SDavid Howells ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 131032021982SDavid Howells ctx->max_size_opt, 131132021982SDavid Howells ctx->max_val_type); 131232021982SDavid Howells ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 131332021982SDavid Howells ctx->min_size_opt, 131432021982SDavid Howells ctx->min_val_type); 13157ca02d0aSMike Kravetz 13167ca02d0aSMike Kravetz /* 13177ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 13187ca02d0aSMike Kravetz */ 131932021982SDavid Howells if (ctx->max_val_type > NO_SIZE && 132032021982SDavid Howells ctx->min_hpages > ctx->max_hpages) { 132132021982SDavid Howells pr_err("Minimum size can not be greater than maximum size\n"); 13227ca02d0aSMike Kravetz return -EINVAL; 1323a137e1ccSAndi Kleen } 1324a137e1ccSAndi Kleen 13251da177e4SLinus Torvalds return 0; 13261da177e4SLinus Torvalds } 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds static int 132932021982SDavid Howells hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc) 13301da177e4SLinus Torvalds { 133132021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 13321da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 13331da177e4SLinus Torvalds 13341da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 13351da177e4SLinus Torvalds if (!sbinfo) 13361da177e4SLinus Torvalds return -ENOMEM; 13371da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 13381da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 133932021982SDavid Howells sbinfo->hstate = ctx->hstate; 134032021982SDavid Howells sbinfo->max_inodes = ctx->nr_inodes; 134132021982SDavid Howells sbinfo->free_inodes = ctx->nr_inodes; 134290481622SDavid Gibson sbinfo->spool = NULL; 134332021982SDavid Howells sbinfo->uid = ctx->uid; 134432021982SDavid Howells sbinfo->gid = ctx->gid; 134532021982SDavid Howells sbinfo->mode = ctx->mode; 13464a25220dSDavid Howells 13477ca02d0aSMike Kravetz /* 13487ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 13491935ebd3SMiaohe Lin * specified. Any needed reservations (for minimum size) are taken 13507ca02d0aSMike Kravetz * taken when the subpool is created. 13517ca02d0aSMike Kravetz */ 135232021982SDavid Howells if (ctx->max_hpages != -1 || ctx->min_hpages != -1) { 135332021982SDavid Howells sbinfo->spool = hugepage_new_subpool(ctx->hstate, 135432021982SDavid Howells ctx->max_hpages, 135532021982SDavid Howells ctx->min_hpages); 135690481622SDavid Gibson if (!sbinfo->spool) 135790481622SDavid Gibson goto out_free; 135890481622SDavid Gibson } 13591da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 136032021982SDavid Howells sb->s_blocksize = huge_page_size(ctx->hstate); 136132021982SDavid Howells sb->s_blocksize_bits = huge_page_shift(ctx->hstate); 13621da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 13631da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 13641da177e4SLinus Torvalds sb->s_time_gran = 1; 136515568299SMike Kravetz 136615568299SMike Kravetz /* 136715568299SMike Kravetz * Due to the special and limited functionality of hugetlbfs, it does 136815568299SMike Kravetz * not work well as a stacking filesystem. 136915568299SMike Kravetz */ 137015568299SMike Kravetz sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH; 137132021982SDavid Howells sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx)); 137248fde701SAl Viro if (!sb->s_root) 13731da177e4SLinus Torvalds goto out_free; 13741da177e4SLinus Torvalds return 0; 13751da177e4SLinus Torvalds out_free: 137690481622SDavid Gibson kfree(sbinfo->spool); 13771da177e4SLinus Torvalds kfree(sbinfo); 13781da177e4SLinus Torvalds return -ENOMEM; 13791da177e4SLinus Torvalds } 13801da177e4SLinus Torvalds 138132021982SDavid Howells static int hugetlbfs_get_tree(struct fs_context *fc) 13821da177e4SLinus Torvalds { 138332021982SDavid Howells int err = hugetlbfs_validate(fc); 138432021982SDavid Howells if (err) 138532021982SDavid Howells return err; 13862ac295d4SAl Viro return get_tree_nodev(fc, hugetlbfs_fill_super); 138732021982SDavid Howells } 138832021982SDavid Howells 138932021982SDavid Howells static void hugetlbfs_fs_context_free(struct fs_context *fc) 139032021982SDavid Howells { 139132021982SDavid Howells kfree(fc->fs_private); 139232021982SDavid Howells } 139332021982SDavid Howells 139432021982SDavid Howells static const struct fs_context_operations hugetlbfs_fs_context_ops = { 139532021982SDavid Howells .free = hugetlbfs_fs_context_free, 139632021982SDavid Howells .parse_param = hugetlbfs_parse_param, 139732021982SDavid Howells .get_tree = hugetlbfs_get_tree, 139832021982SDavid Howells }; 139932021982SDavid Howells 140032021982SDavid Howells static int hugetlbfs_init_fs_context(struct fs_context *fc) 140132021982SDavid Howells { 140232021982SDavid Howells struct hugetlbfs_fs_context *ctx; 140332021982SDavid Howells 140432021982SDavid Howells ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL); 140532021982SDavid Howells if (!ctx) 140632021982SDavid Howells return -ENOMEM; 140732021982SDavid Howells 140832021982SDavid Howells ctx->max_hpages = -1; /* No limit on size by default */ 140932021982SDavid Howells ctx->nr_inodes = -1; /* No limit on number of inodes by default */ 141032021982SDavid Howells ctx->uid = current_fsuid(); 141132021982SDavid Howells ctx->gid = current_fsgid(); 141232021982SDavid Howells ctx->mode = 0755; 141332021982SDavid Howells ctx->hstate = &default_hstate; 141432021982SDavid Howells ctx->min_hpages = -1; /* No default minimum size */ 141532021982SDavid Howells ctx->max_val_type = NO_SIZE; 141632021982SDavid Howells ctx->min_val_type = NO_SIZE; 141732021982SDavid Howells fc->fs_private = ctx; 141832021982SDavid Howells fc->ops = &hugetlbfs_fs_context_ops; 141932021982SDavid Howells return 0; 14201da177e4SLinus Torvalds } 14211da177e4SLinus Torvalds 14221da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 14231da177e4SLinus Torvalds .name = "hugetlbfs", 142432021982SDavid Howells .init_fs_context = hugetlbfs_init_fs_context, 1425d7167b14SAl Viro .parameters = hugetlb_fs_parameters, 14261da177e4SLinus Torvalds .kill_sb = kill_litter_super, 14271da177e4SLinus Torvalds }; 14281da177e4SLinus Torvalds 142942d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 14301da177e4SLinus Torvalds 1431ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 14321da177e4SLinus Torvalds { 1433a0eb3a05SEric W. Biederman kgid_t shm_group; 1434a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 1435a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 14361da177e4SLinus Torvalds } 14371da177e4SLinus Torvalds 143842d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 143942d7395fSAndi Kleen { 1440af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 144142d7395fSAndi Kleen 144242d7395fSAndi Kleen if (!h) 144342d7395fSAndi Kleen return -1; 144404adbc3fSMiaohe Lin return hstate_index(h); 144542d7395fSAndi Kleen } 144642d7395fSAndi Kleen 1447af73e4d9SNaoya Horiguchi /* 1448af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 1449af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 1450af73e4d9SNaoya Horiguchi */ 1451af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 145283c1fd76Szhangyiru vm_flags_t acctflag, int creat_flags, 145383c1fd76Szhangyiru int page_size_log) 14541da177e4SLinus Torvalds { 14551da177e4SLinus Torvalds struct inode *inode; 1456e68375c8SAl Viro struct vfsmount *mnt; 145742d7395fSAndi Kleen int hstate_idx; 1458e68375c8SAl Viro struct file *file; 145942d7395fSAndi Kleen 146042d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 146142d7395fSAndi Kleen if (hstate_idx < 0) 146242d7395fSAndi Kleen return ERR_PTR(-ENODEV); 14631da177e4SLinus Torvalds 1464e68375c8SAl Viro mnt = hugetlbfs_vfsmount[hstate_idx]; 1465e68375c8SAl Viro if (!mnt) 14665bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 14675bc98594SAkinobu Mita 1468ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 146983c1fd76Szhangyiru struct ucounts *ucounts = current_ucounts(); 147083c1fd76Szhangyiru 147183c1fd76Szhangyiru if (user_shm_lock(size, ucounts)) { 147283c1fd76Szhangyiru pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is obsolete\n", 147321a3c273SDavid Rientjes current->comm, current->pid); 147483c1fd76Szhangyiru user_shm_unlock(size, ucounts); 14752584e517SRavikiran G Thirumalai } 147683c1fd76Szhangyiru return ERR_PTR(-EPERM); 1477353d5c30SHugh Dickins } 14781da177e4SLinus Torvalds 147939b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 1480e68375c8SAl Viro inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0); 14811da177e4SLinus Torvalds if (!inode) 1482e68375c8SAl Viro goto out; 1483e1832f29SStephen Smalley if (creat_flags == HUGETLB_SHMFS_INODE) 1484e1832f29SStephen Smalley inode->i_flags |= S_PRIVATE; 14851da177e4SLinus Torvalds 14861da177e4SLinus Torvalds inode->i_size = size; 14876d6b77f1SMiklos Szeredi clear_nlink(inode); 1488ce8d2cdfSDave Hansen 148933b8f84aSMike Kravetz if (!hugetlb_reserve_pages(inode, 0, 1490e68375c8SAl Viro size >> huge_page_shift(hstate_inode(inode)), NULL, 1491e68375c8SAl Viro acctflag)) 1492e68375c8SAl Viro file = ERR_PTR(-ENOMEM); 1493e68375c8SAl Viro else 1494e68375c8SAl Viro file = alloc_file_pseudo(inode, mnt, name, O_RDWR, 1495ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 1496e68375c8SAl Viro if (!IS_ERR(file)) 14971da177e4SLinus Torvalds return file; 14981da177e4SLinus Torvalds 1499b45b5bd6SDavid Gibson iput(inode); 1500e68375c8SAl Viro out: 150139b65252SAnatol Pomozov return file; 15021da177e4SLinus Torvalds } 15031da177e4SLinus Torvalds 150432021982SDavid Howells static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h) 150532021982SDavid Howells { 150632021982SDavid Howells struct fs_context *fc; 150732021982SDavid Howells struct vfsmount *mnt; 150832021982SDavid Howells 150932021982SDavid Howells fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT); 151032021982SDavid Howells if (IS_ERR(fc)) { 151132021982SDavid Howells mnt = ERR_CAST(fc); 151232021982SDavid Howells } else { 151332021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 151432021982SDavid Howells ctx->hstate = h; 151532021982SDavid Howells mnt = fc_mount(fc); 151632021982SDavid Howells put_fs_context(fc); 151732021982SDavid Howells } 151832021982SDavid Howells if (IS_ERR(mnt)) 1519a25fddceSMiaohe Lin pr_err("Cannot mount internal hugetlbfs for page size %luK", 1520a25fddceSMiaohe Lin huge_page_size(h) >> 10); 152132021982SDavid Howells return mnt; 152232021982SDavid Howells } 152332021982SDavid Howells 15241da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 15251da177e4SLinus Torvalds { 152632021982SDavid Howells struct vfsmount *mnt; 152742d7395fSAndi Kleen struct hstate *h; 15281da177e4SLinus Torvalds int error; 152942d7395fSAndi Kleen int i; 15301da177e4SLinus Torvalds 1531457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 15329b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1533457c1b27SNishanth Aravamudan return -ENOTSUPP; 1534457c1b27SNishanth Aravamudan } 1535457c1b27SNishanth Aravamudan 1536d1d5e05fSHillf Danton error = -ENOMEM; 15371da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 15381da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 15395d097056SVladimir Davydov 0, SLAB_ACCOUNT, init_once); 15401da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 15418fc312b3SMike Kravetz goto out; 15421da177e4SLinus Torvalds 15431da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 15441da177e4SLinus Torvalds if (error) 15458fc312b3SMike Kravetz goto out_free; 15461da177e4SLinus Torvalds 15478fc312b3SMike Kravetz /* default hstate mount is required */ 15483b2275a8SMiaohe Lin mnt = mount_one_hugetlbfs(&default_hstate); 15498fc312b3SMike Kravetz if (IS_ERR(mnt)) { 15508fc312b3SMike Kravetz error = PTR_ERR(mnt); 15518fc312b3SMike Kravetz goto out_unreg; 15528fc312b3SMike Kravetz } 15538fc312b3SMike Kravetz hugetlbfs_vfsmount[default_hstate_idx] = mnt; 15548fc312b3SMike Kravetz 15558fc312b3SMike Kravetz /* other hstates are optional */ 155642d7395fSAndi Kleen i = 0; 155742d7395fSAndi Kleen for_each_hstate(h) { 155815f0ec94SJan Stancek if (i == default_hstate_idx) { 155915f0ec94SJan Stancek i++; 15608fc312b3SMike Kravetz continue; 156115f0ec94SJan Stancek } 15628fc312b3SMike Kravetz 156332021982SDavid Howells mnt = mount_one_hugetlbfs(h); 15648fc312b3SMike Kravetz if (IS_ERR(mnt)) 15658fc312b3SMike Kravetz hugetlbfs_vfsmount[i] = NULL; 15668fc312b3SMike Kravetz else 156732021982SDavid Howells hugetlbfs_vfsmount[i] = mnt; 156842d7395fSAndi Kleen i++; 156942d7395fSAndi Kleen } 157032021982SDavid Howells 157142d7395fSAndi Kleen return 0; 15721da177e4SLinus Torvalds 15738fc312b3SMike Kravetz out_unreg: 15748fc312b3SMike Kravetz (void)unregister_filesystem(&hugetlbfs_fs_type); 15758fc312b3SMike Kravetz out_free: 15761da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 15778fc312b3SMike Kravetz out: 15781da177e4SLinus Torvalds return error; 15791da177e4SLinus Torvalds } 15803e89e1c5SPaul Gortmaker fs_initcall(init_hugetlbfs_fs) 1581