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 1112e9b367cSAdam Litke static void huge_pagevec_release(struct pagevec *pvec) 1122e9b367cSAdam Litke { 1132e9b367cSAdam Litke int i; 1142e9b367cSAdam Litke 1152e9b367cSAdam Litke for (i = 0; i < pagevec_count(pvec); ++i) 1162e9b367cSAdam Litke put_page(pvec->pages[i]); 1172e9b367cSAdam Litke 1182e9b367cSAdam Litke pagevec_reinit(pvec); 1192e9b367cSAdam Litke } 1202e9b367cSAdam Litke 12163489f8eSMike Kravetz /* 12263489f8eSMike Kravetz * Mask used when checking the page offset value passed in via system 12363489f8eSMike Kravetz * calls. This value will be converted to a loff_t which is signed. 12463489f8eSMike Kravetz * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the 12563489f8eSMike Kravetz * value. The extra bit (- 1 in the shift value) is to take the sign 12663489f8eSMike Kravetz * bit into account. 12763489f8eSMike Kravetz */ 12863489f8eSMike Kravetz #define PGOFF_LOFFT_MAX \ 12963489f8eSMike Kravetz (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1))) 13063489f8eSMike Kravetz 1311da177e4SLinus Torvalds static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) 1321da177e4SLinus Torvalds { 133496ad9aaSAl Viro struct inode *inode = file_inode(file); 13422247efdSPeter Xu struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 1351da177e4SLinus Torvalds loff_t len, vma_len; 1361da177e4SLinus Torvalds int ret; 137a5516438SAndi Kleen struct hstate *h = hstate_file(file); 1381da177e4SLinus Torvalds 13968589bc3SHugh Dickins /* 140dec4ad86SDavid Gibson * vma address alignment (but not the pgoff alignment) has 141dec4ad86SDavid Gibson * already been checked by prepare_hugepage_range. If you add 142dec4ad86SDavid Gibson * any error returns here, do so after setting VM_HUGETLB, so 143dec4ad86SDavid Gibson * is_vm_hugetlb_page tests below unmap_region go the right 14445e55300SPeter Collingbourne * way when do_mmap unwinds (may be important on powerpc 145dec4ad86SDavid Gibson * and ia64). 14668589bc3SHugh Dickins */ 147a2fce914SNaoya Horiguchi vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; 14868589bc3SHugh Dickins vma->vm_ops = &hugetlb_vm_ops; 1491da177e4SLinus Torvalds 15022247efdSPeter Xu ret = seal_check_future_write(info->seals, vma); 15122247efdSPeter Xu if (ret) 15222247efdSPeter Xu return ret; 15322247efdSPeter Xu 154045c7a3fSMike Kravetz /* 15563489f8eSMike Kravetz * page based offset in vm_pgoff could be sufficiently large to 1565df63c2aSMike Kravetz * overflow a loff_t when converted to byte offset. This can 1575df63c2aSMike Kravetz * only happen on architectures where sizeof(loff_t) == 1585df63c2aSMike Kravetz * sizeof(unsigned long). So, only check in those instances. 159045c7a3fSMike Kravetz */ 1605df63c2aSMike Kravetz if (sizeof(unsigned long) == sizeof(loff_t)) { 16163489f8eSMike Kravetz if (vma->vm_pgoff & PGOFF_LOFFT_MAX) 162045c7a3fSMike Kravetz return -EINVAL; 1635df63c2aSMike Kravetz } 164045c7a3fSMike Kravetz 16563489f8eSMike Kravetz /* must be huge page aligned */ 1662b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 167dec4ad86SDavid Gibson return -EINVAL; 168dec4ad86SDavid Gibson 1691da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 170045c7a3fSMike Kravetz len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 171045c7a3fSMike Kravetz /* check for overflow */ 172045c7a3fSMike Kravetz if (len < vma_len) 173045c7a3fSMike Kravetz return -EINVAL; 1741da177e4SLinus Torvalds 1755955102cSAl Viro inode_lock(inode); 1761da177e4SLinus Torvalds file_accessed(file); 1771da177e4SLinus Torvalds 1781da177e4SLinus Torvalds ret = -ENOMEM; 17933b8f84aSMike Kravetz if (!hugetlb_reserve_pages(inode, 180a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1815a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1825a6fe125SMel Gorman vma->vm_flags)) 183b45b5bd6SDavid Gibson goto out; 184b45b5bd6SDavid Gibson 1854c887265SAdam Litke ret = 0; 186b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 187045c7a3fSMike Kravetz i_size_write(inode, len); 1881da177e4SLinus Torvalds out: 1895955102cSAl Viro inode_unlock(inode); 1901da177e4SLinus Torvalds 1911da177e4SLinus Torvalds return ret; 1921da177e4SLinus Torvalds } 1931da177e4SLinus Torvalds 1941da177e4SLinus Torvalds /* 1953e4e28c5SMichel Lespinasse * Called under mmap_write_lock(mm). 1961da177e4SLinus Torvalds */ 1971da177e4SLinus Torvalds 1981da177e4SLinus Torvalds static unsigned long 19988590253SShijie Hu hugetlb_get_unmapped_area_bottomup(struct file *file, unsigned long addr, 20088590253SShijie Hu unsigned long len, unsigned long pgoff, unsigned long flags) 20188590253SShijie Hu { 20288590253SShijie Hu struct hstate *h = hstate_file(file); 20388590253SShijie Hu struct vm_unmapped_area_info info; 20488590253SShijie Hu 20588590253SShijie Hu info.flags = 0; 20688590253SShijie Hu info.length = len; 20788590253SShijie Hu info.low_limit = current->mm->mmap_base; 2082cb4de08SChristophe Leroy info.high_limit = arch_get_mmap_end(addr, len, flags); 20988590253SShijie Hu info.align_mask = PAGE_MASK & ~huge_page_mask(h); 21088590253SShijie Hu info.align_offset = 0; 21188590253SShijie Hu return vm_unmapped_area(&info); 21288590253SShijie Hu } 21388590253SShijie Hu 21488590253SShijie Hu static unsigned long 21588590253SShijie Hu hugetlb_get_unmapped_area_topdown(struct file *file, unsigned long addr, 21688590253SShijie Hu unsigned long len, unsigned long pgoff, unsigned long flags) 21788590253SShijie Hu { 21888590253SShijie Hu struct hstate *h = hstate_file(file); 21988590253SShijie Hu struct vm_unmapped_area_info info; 22088590253SShijie Hu 22188590253SShijie Hu info.flags = VM_UNMAPPED_AREA_TOPDOWN; 22288590253SShijie Hu info.length = len; 22388590253SShijie Hu info.low_limit = max(PAGE_SIZE, mmap_min_addr); 2245f24d5a5SChristophe Leroy info.high_limit = arch_get_mmap_base(addr, current->mm->mmap_base); 22588590253SShijie Hu info.align_mask = PAGE_MASK & ~huge_page_mask(h); 22688590253SShijie Hu info.align_offset = 0; 22788590253SShijie Hu addr = vm_unmapped_area(&info); 22888590253SShijie Hu 22988590253SShijie Hu /* 23088590253SShijie Hu * A failed mmap() very likely causes application failure, 23188590253SShijie Hu * so fall back to the bottom-up function here. This scenario 23288590253SShijie Hu * can happen with large stack limits and large mmap() 23388590253SShijie Hu * allocations. 23488590253SShijie Hu */ 23588590253SShijie Hu if (unlikely(offset_in_page(addr))) { 23688590253SShijie Hu VM_BUG_ON(addr != -ENOMEM); 23788590253SShijie Hu info.flags = 0; 23888590253SShijie Hu info.low_limit = current->mm->mmap_base; 2392cb4de08SChristophe Leroy info.high_limit = arch_get_mmap_end(addr, len, flags); 24088590253SShijie Hu addr = vm_unmapped_area(&info); 24188590253SShijie Hu } 24288590253SShijie Hu 24388590253SShijie Hu return addr; 24488590253SShijie Hu } 24588590253SShijie Hu 2464b439e25SChristophe Leroy unsigned long 2474b439e25SChristophe Leroy generic_hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 2484b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 2494b439e25SChristophe Leroy unsigned long flags) 2501da177e4SLinus Torvalds { 2511da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2521da177e4SLinus Torvalds struct vm_area_struct *vma; 253a5516438SAndi Kleen struct hstate *h = hstate_file(file); 2542cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 2551da177e4SLinus Torvalds 256a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 2571da177e4SLinus Torvalds return -EINVAL; 2581da177e4SLinus Torvalds if (len > TASK_SIZE) 2591da177e4SLinus Torvalds return -ENOMEM; 2601da177e4SLinus Torvalds 261036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 262a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 263036e0856SBenjamin Herrenschmidt return -EINVAL; 264036e0856SBenjamin Herrenschmidt return addr; 265036e0856SBenjamin Herrenschmidt } 266036e0856SBenjamin Herrenschmidt 2671da177e4SLinus Torvalds if (addr) { 268a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 2691da177e4SLinus Torvalds vma = find_vma(mm, addr); 2705f24d5a5SChristophe Leroy if (mmap_end - len >= addr && 2711be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma))) 2721da177e4SLinus Torvalds return addr; 2731da177e4SLinus Torvalds } 2741da177e4SLinus Torvalds 27588590253SShijie Hu /* 27688590253SShijie Hu * Use mm->get_unmapped_area value as a hint to use topdown routine. 27788590253SShijie Hu * If architectures have special needs, they should define their own 27888590253SShijie Hu * version of hugetlb_get_unmapped_area. 27988590253SShijie Hu */ 28088590253SShijie Hu if (mm->get_unmapped_area == arch_get_unmapped_area_topdown) 28188590253SShijie Hu return hugetlb_get_unmapped_area_topdown(file, addr, len, 28288590253SShijie Hu pgoff, flags); 28388590253SShijie Hu return hugetlb_get_unmapped_area_bottomup(file, addr, len, 28488590253SShijie Hu pgoff, flags); 2851da177e4SLinus Torvalds } 2864b439e25SChristophe Leroy 2874b439e25SChristophe Leroy #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 2884b439e25SChristophe Leroy static unsigned long 2894b439e25SChristophe Leroy hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 2904b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 2914b439e25SChristophe Leroy unsigned long flags) 2924b439e25SChristophe Leroy { 2934b439e25SChristophe Leroy return generic_hugetlb_get_unmapped_area(file, addr, len, pgoff, flags); 2944b439e25SChristophe Leroy } 2951da177e4SLinus Torvalds #endif 2961da177e4SLinus Torvalds 29734d0640eSAl Viro static size_t 298e63e1e5aSBadari Pulavarty hugetlbfs_read_actor(struct page *page, unsigned long offset, 29934d0640eSAl Viro struct iov_iter *to, unsigned long size) 300e63e1e5aSBadari Pulavarty { 30134d0640eSAl Viro size_t copied = 0; 302e63e1e5aSBadari Pulavarty int i, chunksize; 303e63e1e5aSBadari Pulavarty 304e63e1e5aSBadari Pulavarty /* Find which 4k chunk and offset with in that chunk */ 30509cbfeafSKirill A. Shutemov i = offset >> PAGE_SHIFT; 30609cbfeafSKirill A. Shutemov offset = offset & ~PAGE_MASK; 307e63e1e5aSBadari Pulavarty 308e63e1e5aSBadari Pulavarty while (size) { 30934d0640eSAl Viro size_t n; 31009cbfeafSKirill A. Shutemov chunksize = PAGE_SIZE; 311e63e1e5aSBadari Pulavarty if (offset) 312e63e1e5aSBadari Pulavarty chunksize -= offset; 313e63e1e5aSBadari Pulavarty if (chunksize > size) 314e63e1e5aSBadari Pulavarty chunksize = size; 31534d0640eSAl Viro n = copy_page_to_iter(&page[i], offset, chunksize, to); 31634d0640eSAl Viro copied += n; 31734d0640eSAl Viro if (n != chunksize) 31834d0640eSAl Viro return copied; 319e63e1e5aSBadari Pulavarty offset = 0; 320e63e1e5aSBadari Pulavarty size -= chunksize; 321e63e1e5aSBadari Pulavarty i++; 322e63e1e5aSBadari Pulavarty } 32334d0640eSAl Viro return copied; 324e63e1e5aSBadari Pulavarty } 325e63e1e5aSBadari Pulavarty 326e63e1e5aSBadari Pulavarty /* 327e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 328c7e285e3SMiaohe Lin * data. Its *very* similar to generic_file_buffered_read(), we can't use that 329ea1754a0SKirill A. Shutemov * since it has PAGE_SIZE assumptions. 330e63e1e5aSBadari Pulavarty */ 33134d0640eSAl Viro static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) 332e63e1e5aSBadari Pulavarty { 33334d0640eSAl Viro struct file *file = iocb->ki_filp; 33434d0640eSAl Viro struct hstate *h = hstate_file(file); 33534d0640eSAl Viro struct address_space *mapping = file->f_mapping; 336e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 33734d0640eSAl Viro unsigned long index = iocb->ki_pos >> huge_page_shift(h); 33834d0640eSAl Viro unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); 339e63e1e5aSBadari Pulavarty unsigned long end_index; 340e63e1e5aSBadari Pulavarty loff_t isize; 341e63e1e5aSBadari Pulavarty ssize_t retval = 0; 342e63e1e5aSBadari Pulavarty 34334d0640eSAl Viro while (iov_iter_count(to)) { 344e63e1e5aSBadari Pulavarty struct page *page; 34534d0640eSAl Viro size_t nr, copied; 346e63e1e5aSBadari Pulavarty 347e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 348a5516438SAndi Kleen nr = huge_page_size(h); 349a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 350a05b0855SAneesh Kumar K.V if (!isize) 35134d0640eSAl Viro break; 352a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 353e63e1e5aSBadari Pulavarty if (index > end_index) 35434d0640eSAl Viro break; 35534d0640eSAl Viro if (index == end_index) { 356a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 357a05b0855SAneesh Kumar K.V if (nr <= offset) 35834d0640eSAl Viro break; 359e63e1e5aSBadari Pulavarty } 360e63e1e5aSBadari Pulavarty nr = nr - offset; 361e63e1e5aSBadari Pulavarty 362e63e1e5aSBadari Pulavarty /* Find the page */ 363a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 364e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 365e63e1e5aSBadari Pulavarty /* 366e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 367e63e1e5aSBadari Pulavarty * length of the hole or request. 368e63e1e5aSBadari Pulavarty */ 36934d0640eSAl Viro copied = iov_iter_zero(nr, to); 370e63e1e5aSBadari Pulavarty } else { 371a05b0855SAneesh Kumar K.V unlock_page(page); 372a05b0855SAneesh Kumar K.V 373e63e1e5aSBadari Pulavarty /* 374e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 375e63e1e5aSBadari Pulavarty */ 37634d0640eSAl Viro copied = hugetlbfs_read_actor(page, offset, to, nr); 37709cbfeafSKirill A. Shutemov put_page(page); 378e63e1e5aSBadari Pulavarty } 37934d0640eSAl Viro offset += copied; 38034d0640eSAl Viro retval += copied; 38134d0640eSAl Viro if (copied != nr && iov_iter_count(to)) { 38234d0640eSAl Viro if (!retval) 38334d0640eSAl Viro retval = -EFAULT; 384e63e1e5aSBadari Pulavarty break; 385e63e1e5aSBadari Pulavarty } 38634d0640eSAl Viro index += offset >> huge_page_shift(h); 38734d0640eSAl Viro offset &= ~huge_page_mask(h); 38834d0640eSAl Viro } 38934d0640eSAl Viro iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; 390e63e1e5aSBadari Pulavarty return retval; 391e63e1e5aSBadari Pulavarty } 392e63e1e5aSBadari Pulavarty 393800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 394800d15a5SNick Piggin struct address_space *mapping, 3959d6b0cd7SMatthew Wilcox (Oracle) loff_t pos, unsigned len, 396800d15a5SNick Piggin struct page **pagep, void **fsdata) 3971da177e4SLinus Torvalds { 3981da177e4SLinus Torvalds return -EINVAL; 3991da177e4SLinus Torvalds } 4001da177e4SLinus Torvalds 401800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 402800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 403800d15a5SNick Piggin struct page *page, void *fsdata) 4041da177e4SLinus Torvalds { 405800d15a5SNick Piggin BUG(); 4061da177e4SLinus Torvalds return -EINVAL; 4071da177e4SLinus Torvalds } 4081da177e4SLinus Torvalds 409b5cec28dSMike Kravetz static void remove_huge_page(struct page *page) 4101da177e4SLinus Torvalds { 411b9ea2515SKonstantin Khlebnikov ClearPageDirty(page); 4121da177e4SLinus Torvalds ClearPageUptodate(page); 413bd65cb86SMinchan Kim delete_from_page_cache(page); 4141da177e4SLinus Torvalds } 4151da177e4SLinus Torvalds 4164aae8d1cSMike Kravetz static void 41705e90bd0SPeter Xu hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end, 41805e90bd0SPeter Xu zap_flags_t zap_flags) 4194aae8d1cSMike Kravetz { 4204aae8d1cSMike Kravetz struct vm_area_struct *vma; 4214aae8d1cSMike Kravetz 4224aae8d1cSMike Kravetz /* 423d6aba4c8SSean Christopherson * end == 0 indicates that the entire range after start should be 424d6aba4c8SSean Christopherson * unmapped. Note, end is exclusive, whereas the interval tree takes 425d6aba4c8SSean Christopherson * an inclusive "last". 4264aae8d1cSMike Kravetz */ 427d6aba4c8SSean Christopherson vma_interval_tree_foreach(vma, root, start, end ? end - 1 : ULONG_MAX) { 4284aae8d1cSMike Kravetz unsigned long v_offset; 4294aae8d1cSMike Kravetz unsigned long v_end; 4304aae8d1cSMike Kravetz 4314aae8d1cSMike Kravetz /* 4324aae8d1cSMike Kravetz * Can the expression below overflow on 32-bit arches? 4334aae8d1cSMike Kravetz * No, because the interval tree returns us only those vmas 4344aae8d1cSMike Kravetz * which overlap the truncated area starting at pgoff, 4354aae8d1cSMike Kravetz * and no vma on a 32-bit arch can span beyond the 4GB. 4364aae8d1cSMike Kravetz */ 4374aae8d1cSMike Kravetz if (vma->vm_pgoff < start) 4384aae8d1cSMike Kravetz v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; 4394aae8d1cSMike Kravetz else 4404aae8d1cSMike Kravetz v_offset = 0; 4414aae8d1cSMike Kravetz 4424aae8d1cSMike Kravetz if (!end) 4434aae8d1cSMike Kravetz v_end = vma->vm_end; 4444aae8d1cSMike Kravetz else { 4454aae8d1cSMike Kravetz v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) 4464aae8d1cSMike Kravetz + vma->vm_start; 4474aae8d1cSMike Kravetz if (v_end > vma->vm_end) 4484aae8d1cSMike Kravetz v_end = vma->vm_end; 4494aae8d1cSMike Kravetz } 4504aae8d1cSMike Kravetz 4514aae8d1cSMike Kravetz unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, 45205e90bd0SPeter Xu NULL, zap_flags); 4534aae8d1cSMike Kravetz } 4544aae8d1cSMike Kravetz } 455b5cec28dSMike Kravetz 456b5cec28dSMike Kravetz /* 457b5cec28dSMike Kravetz * remove_inode_hugepages handles two distinct cases: truncation and hole 458b5cec28dSMike Kravetz * punch. There are subtle differences in operation for each case. 4594aae8d1cSMike Kravetz * 460b5cec28dSMike Kravetz * truncation is indicated by end of range being LLONG_MAX 461b5cec28dSMike Kravetz * In this case, we first scan the range and release found pages. 4621935ebd3SMiaohe Lin * After releasing pages, hugetlb_unreserve_pages cleans up region/reserve 463e7c58097SMike Kravetz * maps and global counts. Page faults can not race with truncation 46487bf91d3SMike Kravetz * in this routine. hugetlb_no_page() holds i_mmap_rwsem and prevents 46587bf91d3SMike Kravetz * page faults in the truncated range by checking i_size. i_size is 46687bf91d3SMike Kravetz * modified while holding i_mmap_rwsem. 467b5cec28dSMike Kravetz * hole punch is indicated if end is not LLONG_MAX 468b5cec28dSMike Kravetz * In the hole punch case we scan the range and release found pages. 4691935ebd3SMiaohe Lin * Only when releasing a page is the associated region/reserve map 4701935ebd3SMiaohe Lin * deleted. The region/reserve map for ranges without associated 471e7c58097SMike Kravetz * pages are not modified. Page faults can race with hole punch. 472e7c58097SMike Kravetz * This is indicated if we find a mapped page. 473b5cec28dSMike Kravetz * Note: If the passed end of range value is beyond the end of file, but 474b5cec28dSMike Kravetz * not LLONG_MAX this routine still performs a hole punch operation. 475b5cec28dSMike Kravetz */ 476b5cec28dSMike Kravetz static void remove_inode_hugepages(struct inode *inode, loff_t lstart, 477b5cec28dSMike Kravetz loff_t lend) 4781da177e4SLinus Torvalds { 479a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 480b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 481a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 482b5cec28dSMike Kravetz const pgoff_t end = lend >> huge_page_shift(h); 4831da177e4SLinus Torvalds struct pagevec pvec; 484d72dc8a2SJan Kara pgoff_t next, index; 485a43a8c39SChen, Kenneth W int i, freed = 0; 486b5cec28dSMike Kravetz bool truncate_op = (lend == LLONG_MAX); 4871da177e4SLinus Torvalds 48886679820SMel Gorman pagevec_init(&pvec); 4891da177e4SLinus Torvalds next = start; 490b5cec28dSMike Kravetz while (next < end) { 491b5cec28dSMike Kravetz /* 4921817889eSMike Kravetz * When no more pages are found, we are done. 493b5cec28dSMike Kravetz */ 494397162ffSJan Kara if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1)) 4951da177e4SLinus Torvalds break; 4961da177e4SLinus Torvalds 4971da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); ++i) { 4981da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 499d4241a04SMiaohe Lin u32 hash = 0; 500b5cec28dSMike Kravetz 501d72dc8a2SJan Kara index = page->index; 50287bf91d3SMike Kravetz if (!truncate_op) { 50387bf91d3SMike Kravetz /* 50487bf91d3SMike Kravetz * Only need to hold the fault mutex in the 50587bf91d3SMike Kravetz * hole punch case. This prevents races with 50687bf91d3SMike Kravetz * page faults. Races are not possible in the 50787bf91d3SMike Kravetz * case of truncation. 50887bf91d3SMike Kravetz */ 509d4241a04SMiaohe Lin hash = hugetlb_fault_mutex_hash(mapping, index); 510e7c58097SMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 51187bf91d3SMike Kravetz } 512e7c58097SMike Kravetz 513b5cec28dSMike Kravetz /* 514e7c58097SMike Kravetz * If page is mapped, it was faulted in after being 515e7c58097SMike Kravetz * unmapped in caller. Unmap (again) now after taking 516e7c58097SMike Kravetz * the fault mutex. The mutex will prevent faults 517e7c58097SMike Kravetz * until we finish removing the page. 518e7c58097SMike Kravetz * 519e7c58097SMike Kravetz * This race can only happen in the hole punch case. 520e7c58097SMike Kravetz * Getting here in a truncate operation is a bug. 521b5cec28dSMike Kravetz */ 522e7c58097SMike Kravetz if (unlikely(page_mapped(page))) { 523e7c58097SMike Kravetz BUG_ON(truncate_op); 524e7c58097SMike Kravetz 525c0d0381aSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 526e7c58097SMike Kravetz i_mmap_lock_write(mapping); 527c0d0381aSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 528e7c58097SMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 529e7c58097SMike Kravetz index * pages_per_huge_page(h), 53005e90bd0SPeter Xu (index + 1) * pages_per_huge_page(h), 53105e90bd0SPeter Xu ZAP_FLAG_DROP_MARKER); 532e7c58097SMike Kravetz i_mmap_unlock_write(mapping); 533e7c58097SMike Kravetz } 5344aae8d1cSMike Kravetz 5354aae8d1cSMike Kravetz lock_page(page); 5364aae8d1cSMike Kravetz /* 5374aae8d1cSMike Kravetz * We must free the huge page and remove from page 5384aae8d1cSMike Kravetz * cache (remove_huge_page) BEFORE removing the 5394aae8d1cSMike Kravetz * region/reserve map (hugetlb_unreserve_pages). In 5404aae8d1cSMike Kravetz * rare out of memory conditions, removal of the 54172e2936cSzhong jiang * region/reserve map could fail. Correspondingly, 54272e2936cSzhong jiang * the subpool and global reserve usage count can need 54372e2936cSzhong jiang * to be adjusted. 5444aae8d1cSMike Kravetz */ 545e32905e5SMike Kravetz VM_BUG_ON(HPageRestoreReserve(page)); 546b5cec28dSMike Kravetz remove_huge_page(page); 547b5cec28dSMike Kravetz freed++; 548b5cec28dSMike Kravetz if (!truncate_op) { 5494aae8d1cSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, 550d72dc8a2SJan Kara index, index + 1, 1))) 55172e2936cSzhong jiang hugetlb_fix_reserve_counts(inode); 552b5cec28dSMike Kravetz } 553b5cec28dSMike Kravetz 5541da177e4SLinus Torvalds unlock_page(page); 55587bf91d3SMike Kravetz if (!truncate_op) 556e7c58097SMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 5571da177e4SLinus Torvalds } 5581da177e4SLinus Torvalds huge_pagevec_release(&pvec); 5591817889eSMike Kravetz cond_resched(); 5601da177e4SLinus Torvalds } 561b5cec28dSMike Kravetz 562b5cec28dSMike Kravetz if (truncate_op) 563b5cec28dSMike Kravetz (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); 5641da177e4SLinus Torvalds } 5651da177e4SLinus Torvalds 5662bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 5671da177e4SLinus Torvalds { 5689119a41eSJoonsoo Kim struct resv_map *resv_map; 5699119a41eSJoonsoo Kim 570b5cec28dSMike Kravetz remove_inode_hugepages(inode, 0, LLONG_MAX); 571f27a5136SMike Kravetz 572f27a5136SMike Kravetz /* 573f27a5136SMike Kravetz * Get the resv_map from the address space embedded in the inode. 574f27a5136SMike Kravetz * This is the address space which points to any resv_map allocated 575f27a5136SMike Kravetz * at inode creation time. If this is a device special inode, 576f27a5136SMike Kravetz * i_mapping may not point to the original address space. 577f27a5136SMike Kravetz */ 578f27a5136SMike Kravetz resv_map = (struct resv_map *)(&inode->i_data)->private_data; 579f27a5136SMike Kravetz /* Only regular and link inodes have associated reserve maps */ 5809119a41eSJoonsoo Kim if (resv_map) 5819119a41eSJoonsoo Kim resv_map_release(&resv_map->refs); 582dbd5768fSJan Kara clear_inode(inode); 583149f4211SChristoph Hellwig } 584149f4211SChristoph Hellwig 585e5d319deSMiaohe Lin static void hugetlb_vmtruncate(struct inode *inode, loff_t offset) 5861da177e4SLinus Torvalds { 587856fc295SHugh Dickins pgoff_t pgoff; 5881da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 589a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 5901da177e4SLinus Torvalds 591a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 592856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 5931da177e4SLinus Torvalds 59483cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 59587bf91d3SMike Kravetz i_size_write(inode, offset); 596f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 59705e90bd0SPeter Xu hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0, 59805e90bd0SPeter Xu ZAP_FLAG_DROP_MARKER); 599c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 600e7c58097SMike Kravetz remove_inode_hugepages(inode, offset, LLONG_MAX); 6011da177e4SLinus Torvalds } 6021da177e4SLinus Torvalds 60368d32527SMike Kravetz static void hugetlbfs_zero_partial_page(struct hstate *h, 60468d32527SMike Kravetz struct address_space *mapping, 60568d32527SMike Kravetz loff_t start, 60668d32527SMike Kravetz loff_t end) 60768d32527SMike Kravetz { 60868d32527SMike Kravetz pgoff_t idx = start >> huge_page_shift(h); 60968d32527SMike Kravetz struct folio *folio; 61068d32527SMike Kravetz 61168d32527SMike Kravetz folio = filemap_lock_folio(mapping, idx); 61268d32527SMike Kravetz if (!folio) 61368d32527SMike Kravetz return; 61468d32527SMike Kravetz 61568d32527SMike Kravetz start = start & ~huge_page_mask(h); 61668d32527SMike Kravetz end = end & ~huge_page_mask(h); 61768d32527SMike Kravetz if (!end) 61868d32527SMike Kravetz end = huge_page_size(h); 61968d32527SMike Kravetz 62068d32527SMike Kravetz folio_zero_segment(folio, (size_t)start, (size_t)end); 62168d32527SMike Kravetz 62268d32527SMike Kravetz folio_unlock(folio); 62368d32527SMike Kravetz folio_put(folio); 62468d32527SMike Kravetz } 62568d32527SMike Kravetz 62670c3547eSMike Kravetz static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) 62770c3547eSMike Kravetz { 62868d32527SMike Kravetz struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 62968d32527SMike Kravetz struct address_space *mapping = inode->i_mapping; 63070c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 63170c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 63270c3547eSMike Kravetz loff_t hole_start, hole_end; 63370c3547eSMike Kravetz 63470c3547eSMike Kravetz /* 63568d32527SMike Kravetz * hole_start and hole_end indicate the full pages within the hole. 63670c3547eSMike Kravetz */ 63770c3547eSMike Kravetz hole_start = round_up(offset, hpage_size); 63870c3547eSMike Kravetz hole_end = round_down(offset + len, hpage_size); 63970c3547eSMike Kravetz 6405955102cSAl Viro inode_lock(inode); 641ff62a342SMarc-André Lureau 642398c0da7SMiaohe Lin /* protected by i_rwsem */ 643ab3948f5SJoel Fernandes (Google) if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { 644ff62a342SMarc-André Lureau inode_unlock(inode); 645ff62a342SMarc-André Lureau return -EPERM; 646ff62a342SMarc-André Lureau } 647ff62a342SMarc-André Lureau 64870c3547eSMike Kravetz i_mmap_lock_write(mapping); 64968d32527SMike Kravetz 65068d32527SMike Kravetz /* If range starts before first full page, zero partial page. */ 65168d32527SMike Kravetz if (offset < hole_start) 65268d32527SMike Kravetz hugetlbfs_zero_partial_page(h, mapping, 65368d32527SMike Kravetz offset, min(offset + len, hole_start)); 65468d32527SMike Kravetz 65568d32527SMike Kravetz /* Unmap users of full pages in the hole. */ 65668d32527SMike Kravetz if (hole_end > hole_start) { 657f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 65870c3547eSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 65970c3547eSMike Kravetz hole_start >> PAGE_SHIFT, 66005e90bd0SPeter Xu hole_end >> PAGE_SHIFT, 0); 66170c3547eSMike Kravetz } 66270c3547eSMike Kravetz 66368d32527SMike Kravetz /* If range extends beyond last full page, zero partial page. */ 66468d32527SMike Kravetz if ((offset + len) > hole_end && (offset + len) > hole_start) 66568d32527SMike Kravetz hugetlbfs_zero_partial_page(h, mapping, 66668d32527SMike Kravetz hole_end, offset + len); 66768d32527SMike Kravetz 66868d32527SMike Kravetz i_mmap_unlock_write(mapping); 66968d32527SMike Kravetz 67068d32527SMike Kravetz /* Remove full pages from the file. */ 67168d32527SMike Kravetz if (hole_end > hole_start) 67268d32527SMike Kravetz remove_inode_hugepages(inode, hole_start, hole_end); 67368d32527SMike Kravetz 67468d32527SMike Kravetz inode_unlock(inode); 67568d32527SMike Kravetz 67670c3547eSMike Kravetz return 0; 67770c3547eSMike Kravetz } 67870c3547eSMike Kravetz 67970c3547eSMike Kravetz static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, 68070c3547eSMike Kravetz loff_t len) 68170c3547eSMike Kravetz { 68270c3547eSMike Kravetz struct inode *inode = file_inode(file); 683ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 68470c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 68570c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 68670c3547eSMike Kravetz struct vm_area_struct pseudo_vma; 68770c3547eSMike Kravetz struct mm_struct *mm = current->mm; 68870c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 68970c3547eSMike Kravetz unsigned long hpage_shift = huge_page_shift(h); 69070c3547eSMike Kravetz pgoff_t start, index, end; 69170c3547eSMike Kravetz int error; 69270c3547eSMike Kravetz u32 hash; 69370c3547eSMike Kravetz 69470c3547eSMike Kravetz if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 69570c3547eSMike Kravetz return -EOPNOTSUPP; 69670c3547eSMike Kravetz 69770c3547eSMike Kravetz if (mode & FALLOC_FL_PUNCH_HOLE) 69870c3547eSMike Kravetz return hugetlbfs_punch_hole(inode, offset, len); 69970c3547eSMike Kravetz 70070c3547eSMike Kravetz /* 70170c3547eSMike Kravetz * Default preallocate case. 70270c3547eSMike Kravetz * For this range, start is rounded down and end is rounded up 70370c3547eSMike Kravetz * as well as being converted to page offsets. 70470c3547eSMike Kravetz */ 70570c3547eSMike Kravetz start = offset >> hpage_shift; 70670c3547eSMike Kravetz end = (offset + len + hpage_size - 1) >> hpage_shift; 70770c3547eSMike Kravetz 7085955102cSAl Viro inode_lock(inode); 70970c3547eSMike Kravetz 71070c3547eSMike Kravetz /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ 71170c3547eSMike Kravetz error = inode_newsize_ok(inode, offset + len); 71270c3547eSMike Kravetz if (error) 71370c3547eSMike Kravetz goto out; 71470c3547eSMike Kravetz 715ff62a342SMarc-André Lureau if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { 716ff62a342SMarc-André Lureau error = -EPERM; 717ff62a342SMarc-André Lureau goto out; 718ff62a342SMarc-André Lureau } 719ff62a342SMarc-André Lureau 72070c3547eSMike Kravetz /* 72170c3547eSMike Kravetz * Initialize a pseudo vma as this is required by the huge page 72270c3547eSMike Kravetz * allocation routines. If NUMA is configured, use page index 72370c3547eSMike Kravetz * as input to create an allocation policy. 72470c3547eSMike Kravetz */ 7252c4541e2SKirill A. Shutemov vma_init(&pseudo_vma, mm); 72670c3547eSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 72770c3547eSMike Kravetz pseudo_vma.vm_file = file; 72870c3547eSMike Kravetz 72970c3547eSMike Kravetz for (index = start; index < end; index++) { 73070c3547eSMike Kravetz /* 73170c3547eSMike Kravetz * This is supposed to be the vaddr where the page is being 73270c3547eSMike Kravetz * faulted in, but we have no vaddr here. 73370c3547eSMike Kravetz */ 73470c3547eSMike Kravetz struct page *page; 73570c3547eSMike Kravetz unsigned long addr; 73670c3547eSMike Kravetz 73770c3547eSMike Kravetz cond_resched(); 73870c3547eSMike Kravetz 73970c3547eSMike Kravetz /* 74070c3547eSMike Kravetz * fallocate(2) manpage permits EINTR; we may have been 74170c3547eSMike Kravetz * interrupted because we are using up too much memory. 74270c3547eSMike Kravetz */ 74370c3547eSMike Kravetz if (signal_pending(current)) { 74470c3547eSMike Kravetz error = -EINTR; 74570c3547eSMike Kravetz break; 74670c3547eSMike Kravetz } 74770c3547eSMike Kravetz 74870c3547eSMike Kravetz /* Set numa allocation policy based on index */ 74970c3547eSMike Kravetz hugetlb_set_vma_policy(&pseudo_vma, inode, index); 75070c3547eSMike Kravetz 75170c3547eSMike Kravetz /* addr is the offset within the file (zero based) */ 75270c3547eSMike Kravetz addr = index * hpage_size; 75370c3547eSMike Kravetz 75487bf91d3SMike Kravetz /* 75587bf91d3SMike Kravetz * fault mutex taken here, protects against fault path 75687bf91d3SMike Kravetz * and hole punch. inode_lock previously taken protects 75787bf91d3SMike Kravetz * against truncation. 75887bf91d3SMike Kravetz */ 759188b04a7SWei Yang hash = hugetlb_fault_mutex_hash(mapping, index); 76070c3547eSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 76170c3547eSMike Kravetz 76270c3547eSMike Kravetz /* See if already present in mapping to avoid alloc/free */ 76370c3547eSMike Kravetz page = find_get_page(mapping, index); 76470c3547eSMike Kravetz if (page) { 76570c3547eSMike Kravetz put_page(page); 76670c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 76770c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 76870c3547eSMike Kravetz continue; 76970c3547eSMike Kravetz } 77070c3547eSMike Kravetz 77188ce3fefSMiaohe Lin /* 77288ce3fefSMiaohe Lin * Allocate page without setting the avoid_reserve argument. 77388ce3fefSMiaohe Lin * There certainly are no reserves associated with the 77488ce3fefSMiaohe Lin * pseudo_vma. However, there could be shared mappings with 77588ce3fefSMiaohe Lin * reserves for the file at the inode level. If we fallocate 77688ce3fefSMiaohe Lin * pages in these areas, we need to consume the reserves 77788ce3fefSMiaohe Lin * to keep reservation accounting consistent. 77888ce3fefSMiaohe Lin */ 77988ce3fefSMiaohe Lin page = alloc_huge_page(&pseudo_vma, addr, 0); 78070c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 78170c3547eSMike Kravetz if (IS_ERR(page)) { 78270c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 78370c3547eSMike Kravetz error = PTR_ERR(page); 78470c3547eSMike Kravetz goto out; 78570c3547eSMike Kravetz } 78670c3547eSMike Kravetz clear_huge_page(page, addr, pages_per_huge_page(h)); 78770c3547eSMike Kravetz __SetPageUptodate(page); 78870c3547eSMike Kravetz error = huge_add_to_page_cache(page, mapping, index); 78970c3547eSMike Kravetz if (unlikely(error)) { 790846be085SMike Kravetz restore_reserve_on_error(h, &pseudo_vma, addr, page); 79170c3547eSMike Kravetz put_page(page); 79270c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 79370c3547eSMike Kravetz goto out; 79470c3547eSMike Kravetz } 79570c3547eSMike Kravetz 79670c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 79770c3547eSMike Kravetz 7988f251a3dSMike Kravetz SetHPageMigratable(page); 79970c3547eSMike Kravetz /* 80070c3547eSMike Kravetz * unlock_page because locked by add_to_page_cache() 801585fc0d2SMuchun Song * put_page() due to reference from alloc_huge_page() 80270c3547eSMike Kravetz */ 80370c3547eSMike Kravetz unlock_page(page); 80472639e6dSNadav Amit put_page(page); 80570c3547eSMike Kravetz } 80670c3547eSMike Kravetz 80770c3547eSMike Kravetz if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) 80870c3547eSMike Kravetz i_size_write(inode, offset + len); 809078cd827SDeepa Dinamani inode->i_ctime = current_time(inode); 81070c3547eSMike Kravetz out: 8115955102cSAl Viro inode_unlock(inode); 81270c3547eSMike Kravetz return error; 81370c3547eSMike Kravetz } 81470c3547eSMike Kravetz 815549c7297SChristian Brauner static int hugetlbfs_setattr(struct user_namespace *mnt_userns, 816549c7297SChristian Brauner struct dentry *dentry, struct iattr *attr) 8171da177e4SLinus Torvalds { 8182b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 819a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 8201da177e4SLinus Torvalds int error; 8211da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 822ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 8231da177e4SLinus Torvalds 8242f221d6fSChristian Brauner error = setattr_prepare(&init_user_ns, dentry, attr); 8251da177e4SLinus Torvalds if (error) 8261025774cSChristoph Hellwig return error; 8271da177e4SLinus Torvalds 8281da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 829ff62a342SMarc-André Lureau loff_t oldsize = inode->i_size; 830ff62a342SMarc-André Lureau loff_t newsize = attr->ia_size; 831ff62a342SMarc-André Lureau 832ff62a342SMarc-André Lureau if (newsize & ~huge_page_mask(h)) 8331025774cSChristoph Hellwig return -EINVAL; 834398c0da7SMiaohe Lin /* protected by i_rwsem */ 835ff62a342SMarc-André Lureau if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || 836ff62a342SMarc-André Lureau (newsize > oldsize && (info->seals & F_SEAL_GROW))) 837ff62a342SMarc-André Lureau return -EPERM; 838e5d319deSMiaohe Lin hugetlb_vmtruncate(inode, newsize); 8391da177e4SLinus Torvalds } 8401da177e4SLinus Torvalds 8412f221d6fSChristian Brauner setattr_copy(&init_user_ns, inode, attr); 8421025774cSChristoph Hellwig mark_inode_dirty(inode); 8431025774cSChristoph Hellwig return 0; 8441025774cSChristoph Hellwig } 8451025774cSChristoph Hellwig 8467d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 84732021982SDavid Howells struct hugetlbfs_fs_context *ctx) 8481da177e4SLinus Torvalds { 8491da177e4SLinus Torvalds struct inode *inode; 8501da177e4SLinus Torvalds 8511da177e4SLinus Torvalds inode = new_inode(sb); 8521da177e4SLinus Torvalds if (inode) { 85385fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 85432021982SDavid Howells inode->i_mode = S_IFDIR | ctx->mode; 85532021982SDavid Howells inode->i_uid = ctx->uid; 85632021982SDavid Howells inode->i_gid = ctx->gid; 857078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 8587d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 8597d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 8607d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 8617d54fa64SAl Viro inc_nlink(inode); 86265ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 8637d54fa64SAl Viro } 8647d54fa64SAl Viro return inode; 8657d54fa64SAl Viro } 8667d54fa64SAl Viro 867b610ded7SMichal Hocko /* 868c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 869b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 87088f306b6SKirill A. Shutemov * annotation because huge_pmd_share() does an allocation under hugetlb's 871c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 872b610ded7SMichal Hocko */ 873c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 874b610ded7SMichal Hocko 8757d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 8767d54fa64SAl Viro struct inode *dir, 87718df2252SAl Viro umode_t mode, dev_t dev) 8787d54fa64SAl Viro { 8797d54fa64SAl Viro struct inode *inode; 88058b6e5e8SMike Kravetz struct resv_map *resv_map = NULL; 8819119a41eSJoonsoo Kim 88258b6e5e8SMike Kravetz /* 88358b6e5e8SMike Kravetz * Reserve maps are only needed for inodes that can have associated 88458b6e5e8SMike Kravetz * page allocations. 88558b6e5e8SMike Kravetz */ 88658b6e5e8SMike Kravetz if (S_ISREG(mode) || S_ISLNK(mode)) { 8879119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 8889119a41eSJoonsoo Kim if (!resv_map) 8899119a41eSJoonsoo Kim return NULL; 89058b6e5e8SMike Kravetz } 8917d54fa64SAl Viro 8927d54fa64SAl Viro inode = new_inode(sb); 8937d54fa64SAl Viro if (inode) { 894ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 895ff62a342SMarc-André Lureau 8967d54fa64SAl Viro inode->i_ino = get_next_ino(); 89721cb47beSChristian Brauner inode_init_owner(&init_user_ns, inode, dir, mode); 898c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 899c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 9001da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 901078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 9029119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 903ff62a342SMarc-André Lureau info->seals = F_SEAL_SEAL; 9041da177e4SLinus Torvalds switch (mode & S_IFMT) { 9051da177e4SLinus Torvalds default: 9061da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 9071da177e4SLinus Torvalds break; 9081da177e4SLinus Torvalds case S_IFREG: 9091da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 9101da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 9111da177e4SLinus Torvalds break; 9121da177e4SLinus Torvalds case S_IFDIR: 9131da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 9141da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 9151da177e4SLinus Torvalds 9161da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 917d8c76e6fSDave Hansen inc_nlink(inode); 9181da177e4SLinus Torvalds break; 9191da177e4SLinus Torvalds case S_IFLNK: 9201da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 92121fc61c7SAl Viro inode_nohighmem(inode); 9221da177e4SLinus Torvalds break; 9231da177e4SLinus Torvalds } 924e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 92558b6e5e8SMike Kravetz } else { 92658b6e5e8SMike Kravetz if (resv_map) 9279119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 92858b6e5e8SMike Kravetz } 9299119a41eSJoonsoo Kim 9301da177e4SLinus Torvalds return inode; 9311da177e4SLinus Torvalds } 9321da177e4SLinus Torvalds 9331da177e4SLinus Torvalds /* 9341da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 9351da177e4SLinus Torvalds */ 9361ab5b82fSPiotr Sarna static int do_hugetlbfs_mknod(struct inode *dir, 9371ab5b82fSPiotr Sarna struct dentry *dentry, 9381ab5b82fSPiotr Sarna umode_t mode, 9391ab5b82fSPiotr Sarna dev_t dev, 9401ab5b82fSPiotr Sarna bool tmpfile) 9411da177e4SLinus Torvalds { 9421da177e4SLinus Torvalds struct inode *inode; 9431da177e4SLinus Torvalds int error = -ENOSPC; 9441da177e4SLinus Torvalds 9457d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 9461da177e4SLinus Torvalds if (inode) { 947078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 9481ab5b82fSPiotr Sarna if (tmpfile) { 9491ab5b82fSPiotr Sarna d_tmpfile(dentry, inode); 9501ab5b82fSPiotr Sarna } else { 9511da177e4SLinus Torvalds d_instantiate(dentry, inode); 9521da177e4SLinus Torvalds dget(dentry);/* Extra count - pin the dentry in core */ 9531ab5b82fSPiotr Sarna } 9541da177e4SLinus Torvalds error = 0; 9551da177e4SLinus Torvalds } 9561da177e4SLinus Torvalds return error; 9571da177e4SLinus Torvalds } 9581da177e4SLinus Torvalds 959549c7297SChristian Brauner static int hugetlbfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 9601ab5b82fSPiotr Sarna struct dentry *dentry, umode_t mode, dev_t dev) 9611ab5b82fSPiotr Sarna { 9621ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode, dev, false); 9631ab5b82fSPiotr Sarna } 9641ab5b82fSPiotr Sarna 965549c7297SChristian Brauner static int hugetlbfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 966549c7297SChristian Brauner struct dentry *dentry, umode_t mode) 9671da177e4SLinus Torvalds { 968549c7297SChristian Brauner int retval = hugetlbfs_mknod(&init_user_ns, dir, dentry, 969549c7297SChristian Brauner mode | S_IFDIR, 0); 9701da177e4SLinus Torvalds if (!retval) 971d8c76e6fSDave Hansen inc_nlink(dir); 9721da177e4SLinus Torvalds return retval; 9731da177e4SLinus Torvalds } 9741da177e4SLinus Torvalds 975549c7297SChristian Brauner static int hugetlbfs_create(struct user_namespace *mnt_userns, 976549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 977549c7297SChristian Brauner umode_t mode, bool excl) 9781da177e4SLinus Torvalds { 979549c7297SChristian Brauner return hugetlbfs_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0); 9801da177e4SLinus Torvalds } 9811da177e4SLinus Torvalds 982549c7297SChristian Brauner static int hugetlbfs_tmpfile(struct user_namespace *mnt_userns, 983549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 984549c7297SChristian Brauner umode_t mode) 9851ab5b82fSPiotr Sarna { 9861ab5b82fSPiotr Sarna return do_hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0, true); 9871ab5b82fSPiotr Sarna } 9881ab5b82fSPiotr Sarna 989549c7297SChristian Brauner static int hugetlbfs_symlink(struct user_namespace *mnt_userns, 990549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 991549c7297SChristian Brauner const char *symname) 9921da177e4SLinus Torvalds { 9931da177e4SLinus Torvalds struct inode *inode; 9941da177e4SLinus Torvalds int error = -ENOSPC; 9951da177e4SLinus Torvalds 9967d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 9971da177e4SLinus Torvalds if (inode) { 9981da177e4SLinus Torvalds int l = strlen(symname)+1; 9991da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 10001da177e4SLinus Torvalds if (!error) { 10011da177e4SLinus Torvalds d_instantiate(dentry, inode); 10021da177e4SLinus Torvalds dget(dentry); 10031da177e4SLinus Torvalds } else 10041da177e4SLinus Torvalds iput(inode); 10051da177e4SLinus Torvalds } 1006078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 10071da177e4SLinus Torvalds 10081da177e4SLinus Torvalds return error; 10091da177e4SLinus Torvalds } 10101da177e4SLinus Torvalds 1011290408d4SNaoya Horiguchi static int hugetlbfs_migrate_page(struct address_space *mapping, 1012b969c4abSMel Gorman struct page *newpage, struct page *page, 1013a6bc32b8SMel Gorman enum migrate_mode mode) 1014290408d4SNaoya Horiguchi { 1015290408d4SNaoya Horiguchi int rc; 1016290408d4SNaoya Horiguchi 1017290408d4SNaoya Horiguchi rc = migrate_huge_page_move_mapping(mapping, newpage, page); 101878bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 1019290408d4SNaoya Horiguchi return rc; 1020cb6acd01SMike Kravetz 1021d6995da3SMike Kravetz if (hugetlb_page_subpool(page)) { 1022d6995da3SMike Kravetz hugetlb_set_page_subpool(newpage, hugetlb_page_subpool(page)); 1023d6995da3SMike Kravetz hugetlb_set_page_subpool(page, NULL); 1024cb6acd01SMike Kravetz } 1025cb6acd01SMike Kravetz 10262916ecc0SJérôme Glisse if (mode != MIGRATE_SYNC_NO_COPY) 1027290408d4SNaoya Horiguchi migrate_page_copy(newpage, page); 10282916ecc0SJérôme Glisse else 10292916ecc0SJérôme Glisse migrate_page_states(newpage, page); 1030290408d4SNaoya Horiguchi 103178bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 1032290408d4SNaoya Horiguchi } 1033290408d4SNaoya Horiguchi 103478bb9203SNaoya Horiguchi static int hugetlbfs_error_remove_page(struct address_space *mapping, 103578bb9203SNaoya Horiguchi struct page *page) 103678bb9203SNaoya Horiguchi { 103778bb9203SNaoya Horiguchi struct inode *inode = mapping->host; 1038ab615a5bSMike Kravetz pgoff_t index = page->index; 103978bb9203SNaoya Horiguchi 104078bb9203SNaoya Horiguchi remove_huge_page(page); 1041ab615a5bSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) 104278bb9203SNaoya Horiguchi hugetlb_fix_reserve_counts(inode); 1043ab615a5bSMike Kravetz 104478bb9203SNaoya Horiguchi return 0; 104578bb9203SNaoya Horiguchi } 104678bb9203SNaoya Horiguchi 10474a25220dSDavid Howells /* 10484a25220dSDavid Howells * Display the mount options in /proc/mounts. 10494a25220dSDavid Howells */ 10504a25220dSDavid Howells static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) 10514a25220dSDavid Howells { 10524a25220dSDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); 10534a25220dSDavid Howells struct hugepage_subpool *spool = sbinfo->spool; 10544a25220dSDavid Howells unsigned long hpage_size = huge_page_size(sbinfo->hstate); 10554a25220dSDavid Howells unsigned hpage_shift = huge_page_shift(sbinfo->hstate); 10564a25220dSDavid Howells char mod; 10574a25220dSDavid Howells 10584a25220dSDavid Howells if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) 10594a25220dSDavid Howells seq_printf(m, ",uid=%u", 10604a25220dSDavid Howells from_kuid_munged(&init_user_ns, sbinfo->uid)); 10614a25220dSDavid Howells if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) 10624a25220dSDavid Howells seq_printf(m, ",gid=%u", 10634a25220dSDavid Howells from_kgid_munged(&init_user_ns, sbinfo->gid)); 10644a25220dSDavid Howells if (sbinfo->mode != 0755) 10654a25220dSDavid Howells seq_printf(m, ",mode=%o", sbinfo->mode); 10664a25220dSDavid Howells if (sbinfo->max_inodes != -1) 10674a25220dSDavid Howells seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); 10684a25220dSDavid Howells 10694a25220dSDavid Howells hpage_size /= 1024; 10704a25220dSDavid Howells mod = 'K'; 10714a25220dSDavid Howells if (hpage_size >= 1024) { 10724a25220dSDavid Howells hpage_size /= 1024; 10734a25220dSDavid Howells mod = 'M'; 10744a25220dSDavid Howells } 10754a25220dSDavid Howells seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); 10764a25220dSDavid Howells if (spool) { 10774a25220dSDavid Howells if (spool->max_hpages != -1) 10784a25220dSDavid Howells seq_printf(m, ",size=%llu", 10794a25220dSDavid Howells (unsigned long long)spool->max_hpages << hpage_shift); 10804a25220dSDavid Howells if (spool->min_hpages != -1) 10814a25220dSDavid Howells seq_printf(m, ",min_size=%llu", 10824a25220dSDavid Howells (unsigned long long)spool->min_hpages << hpage_shift); 10834a25220dSDavid Howells } 10844a25220dSDavid Howells return 0; 10854a25220dSDavid Howells } 10864a25220dSDavid Howells 1087726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 10881da177e4SLinus Torvalds { 1089726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 10902b0143b5SDavid Howells struct hstate *h = hstate_inode(d_inode(dentry)); 10911da177e4SLinus Torvalds 10921da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 1093a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 10941da177e4SLinus Torvalds if (sbinfo) { 10951da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 109674a8a65cSDavid Gibson /* If no limits set, just report 0 for max/free/used 109774a8a65cSDavid Gibson * blocks, like simple_statfs() */ 109890481622SDavid Gibson if (sbinfo->spool) { 109990481622SDavid Gibson long free_pages; 110090481622SDavid Gibson 11014b25f030SMina Almasry spin_lock_irq(&sbinfo->spool->lock); 110290481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 110390481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 110490481622SDavid Gibson - sbinfo->spool->used_hpages; 110590481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 11064b25f030SMina Almasry spin_unlock_irq(&sbinfo->spool->lock); 11071da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 11081da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 110974a8a65cSDavid Gibson } 11101da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 11131da177e4SLinus Torvalds return 0; 11141da177e4SLinus Torvalds } 11151da177e4SLinus Torvalds 11161da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 11171da177e4SLinus Torvalds { 11181da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 11191da177e4SLinus Torvalds 11201da177e4SLinus Torvalds if (sbi) { 11211da177e4SLinus Torvalds sb->s_fs_info = NULL; 112290481622SDavid Gibson 112390481622SDavid Gibson if (sbi->spool) 112490481622SDavid Gibson hugepage_put_subpool(sbi->spool); 112590481622SDavid Gibson 11261da177e4SLinus Torvalds kfree(sbi); 11271da177e4SLinus Torvalds } 11281da177e4SLinus Torvalds } 11291da177e4SLinus Torvalds 113096527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 113196527980SChristoph Hellwig { 113296527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 113396527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 113496527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 113596527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 113696527980SChristoph Hellwig return 0; 113796527980SChristoph Hellwig } 113896527980SChristoph Hellwig sbinfo->free_inodes--; 113996527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 114096527980SChristoph Hellwig } 114196527980SChristoph Hellwig 114296527980SChristoph Hellwig return 1; 114396527980SChristoph Hellwig } 114496527980SChristoph Hellwig 114596527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 114696527980SChristoph Hellwig { 114796527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 114896527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 114996527980SChristoph Hellwig sbinfo->free_inodes++; 115096527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 115196527980SChristoph Hellwig } 115296527980SChristoph Hellwig } 115396527980SChristoph Hellwig 115496527980SChristoph Hellwig 1155e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 11561da177e4SLinus Torvalds 11571da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 11581da177e4SLinus Torvalds { 115996527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 11601da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 11611da177e4SLinus Torvalds 116296527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 11631da177e4SLinus Torvalds return NULL; 1164fd60b288SMuchun Song p = alloc_inode_sb(sb, hugetlbfs_inode_cachep, GFP_KERNEL); 116596527980SChristoph Hellwig if (unlikely(!p)) { 116696527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 116796527980SChristoph Hellwig return NULL; 11681da177e4SLinus Torvalds } 11694742a35dSMike Kravetz 11704742a35dSMike Kravetz /* 11714742a35dSMike Kravetz * Any time after allocation, hugetlbfs_destroy_inode can be called 11724742a35dSMike Kravetz * for the inode. mpol_free_shared_policy is unconditionally called 11734742a35dSMike Kravetz * as part of hugetlbfs_destroy_inode. So, initialize policy here 11744742a35dSMike Kravetz * in case of a quick call to destroy. 11754742a35dSMike Kravetz * 11764742a35dSMike Kravetz * Note that the policy is initialized even if we are creating a 11774742a35dSMike Kravetz * private inode. This simplifies hugetlbfs_destroy_inode. 11784742a35dSMike Kravetz */ 11794742a35dSMike Kravetz mpol_shared_policy_init(&p->policy, NULL); 11804742a35dSMike Kravetz 118196527980SChristoph Hellwig return &p->vfs_inode; 11821da177e4SLinus Torvalds } 11831da177e4SLinus Torvalds 1184b62de322SAl Viro static void hugetlbfs_free_inode(struct inode *inode) 1185fa0d7e3dSNick Piggin { 1186fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 1187fa0d7e3dSNick Piggin } 1188fa0d7e3dSNick Piggin 11891da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 11901da177e4SLinus Torvalds { 119196527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 11921da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 11931da177e4SLinus Torvalds } 11941da177e4SLinus Torvalds 1195f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 1196800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 1197800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 119846de8b97SMatthew Wilcox (Oracle) .dirty_folio = noop_dirty_folio, 1199290408d4SNaoya Horiguchi .migratepage = hugetlbfs_migrate_page, 120078bb9203SNaoya Horiguchi .error_remove_page = hugetlbfs_error_remove_page, 12011da177e4SLinus Torvalds }; 12021da177e4SLinus Torvalds 120396527980SChristoph Hellwig 120451cc5068SAlexey Dobriyan static void init_once(void *foo) 120596527980SChristoph Hellwig { 120696527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 120796527980SChristoph Hellwig 120896527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 120996527980SChristoph Hellwig } 121096527980SChristoph Hellwig 12114b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 121234d0640eSAl Viro .read_iter = hugetlbfs_read_iter, 12131da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 12141b061d92SChristoph Hellwig .fsync = noop_fsync, 12151da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 12166038f373SArnd Bergmann .llseek = default_llseek, 121770c3547eSMike Kravetz .fallocate = hugetlbfs_fallocate, 12181da177e4SLinus Torvalds }; 12191da177e4SLinus Torvalds 122092e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 12211da177e4SLinus Torvalds .create = hugetlbfs_create, 12221da177e4SLinus Torvalds .lookup = simple_lookup, 12231da177e4SLinus Torvalds .link = simple_link, 12241da177e4SLinus Torvalds .unlink = simple_unlink, 12251da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 12261da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 12271da177e4SLinus Torvalds .rmdir = simple_rmdir, 12281da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 12291da177e4SLinus Torvalds .rename = simple_rename, 12301da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 12311ab5b82fSPiotr Sarna .tmpfile = hugetlbfs_tmpfile, 12321da177e4SLinus Torvalds }; 12331da177e4SLinus Torvalds 123492e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 12351da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 12361da177e4SLinus Torvalds }; 12371da177e4SLinus Torvalds 1238ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 12391da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 1240b62de322SAl Viro .free_inode = hugetlbfs_free_inode, 12411da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 12422bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 12431da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 12441da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 12454a25220dSDavid Howells .show_options = hugetlbfs_show_options, 12461da177e4SLinus Torvalds }; 12471da177e4SLinus Torvalds 12487ca02d0aSMike Kravetz /* 12497ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 12507ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 12517ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 12527ca02d0aSMike Kravetz */ 12534a25220dSDavid Howells static long 12547ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 12554a25220dSDavid Howells enum hugetlbfs_size_type val_type) 12567ca02d0aSMike Kravetz { 12577ca02d0aSMike Kravetz if (val_type == NO_SIZE) 12587ca02d0aSMike Kravetz return -1; 12597ca02d0aSMike Kravetz 12607ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 12617ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 12627ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 12637ca02d0aSMike Kravetz do_div(size_opt, 100); 12647ca02d0aSMike Kravetz } 12657ca02d0aSMike Kravetz 12667ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 12677ca02d0aSMike Kravetz return size_opt; 12687ca02d0aSMike Kravetz } 12697ca02d0aSMike Kravetz 127032021982SDavid Howells /* 127132021982SDavid Howells * Parse one mount parameter. 127232021982SDavid Howells */ 127332021982SDavid Howells static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param) 12741da177e4SLinus Torvalds { 127532021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 127632021982SDavid Howells struct fs_parse_result result; 127732021982SDavid Howells char *rest; 127832021982SDavid Howells unsigned long ps; 127932021982SDavid Howells int opt; 12801da177e4SLinus Torvalds 1281d7167b14SAl Viro opt = fs_parse(fc, hugetlb_fs_parameters, param, &result); 128232021982SDavid Howells if (opt < 0) 128332021982SDavid Howells return opt; 128432021982SDavid Howells 128532021982SDavid Howells switch (opt) { 128632021982SDavid Howells case Opt_uid: 128732021982SDavid Howells ctx->uid = make_kuid(current_user_ns(), result.uint_32); 128832021982SDavid Howells if (!uid_valid(ctx->uid)) 128932021982SDavid Howells goto bad_val; 12901da177e4SLinus Torvalds return 0; 12911da177e4SLinus Torvalds 1292e73a75faSRandy Dunlap case Opt_gid: 129332021982SDavid Howells ctx->gid = make_kgid(current_user_ns(), result.uint_32); 129432021982SDavid Howells if (!gid_valid(ctx->gid)) 1295e73a75faSRandy Dunlap goto bad_val; 129632021982SDavid Howells return 0; 1297e73a75faSRandy Dunlap 1298e73a75faSRandy Dunlap case Opt_mode: 129932021982SDavid Howells ctx->mode = result.uint_32 & 01777U; 130032021982SDavid Howells return 0; 1301e73a75faSRandy Dunlap 130232021982SDavid Howells case Opt_size: 1303e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 130432021982SDavid Howells if (!isdigit(param->string[0])) 1305e73a75faSRandy Dunlap goto bad_val; 130632021982SDavid Howells ctx->max_size_opt = memparse(param->string, &rest); 130732021982SDavid Howells ctx->max_val_type = SIZE_STD; 1308a137e1ccSAndi Kleen if (*rest == '%') 130932021982SDavid Howells ctx->max_val_type = SIZE_PERCENT; 131032021982SDavid Howells return 0; 13111da177e4SLinus Torvalds 1312e73a75faSRandy Dunlap case Opt_nr_inodes: 1313e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 131432021982SDavid Howells if (!isdigit(param->string[0])) 1315e73a75faSRandy Dunlap goto bad_val; 131632021982SDavid Howells ctx->nr_inodes = memparse(param->string, &rest); 131732021982SDavid Howells return 0; 1318e73a75faSRandy Dunlap 131932021982SDavid Howells case Opt_pagesize: 132032021982SDavid Howells ps = memparse(param->string, &rest); 132132021982SDavid Howells ctx->hstate = size_to_hstate(ps); 132232021982SDavid Howells if (!ctx->hstate) { 1323*d0036517SMiaohe Lin pr_err("Unsupported page size %lu MB\n", ps / SZ_1M); 1324a137e1ccSAndi Kleen return -EINVAL; 1325a137e1ccSAndi Kleen } 132632021982SDavid Howells return 0; 1327a137e1ccSAndi Kleen 132832021982SDavid Howells case Opt_min_size: 13297ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 133032021982SDavid Howells if (!isdigit(param->string[0])) 13317ca02d0aSMike Kravetz goto bad_val; 133232021982SDavid Howells ctx->min_size_opt = memparse(param->string, &rest); 133332021982SDavid Howells ctx->min_val_type = SIZE_STD; 13347ca02d0aSMike Kravetz if (*rest == '%') 133532021982SDavid Howells ctx->min_val_type = SIZE_PERCENT; 133632021982SDavid Howells return 0; 13377ca02d0aSMike Kravetz 1338e73a75faSRandy Dunlap default: 1339b4c07bceSLee Schermerhorn return -EINVAL; 1340e73a75faSRandy Dunlap } 134132021982SDavid Howells 134232021982SDavid Howells bad_val: 1343b5db30cfSAl Viro return invalfc(fc, "Bad value '%s' for mount option '%s'\n", 134432021982SDavid Howells param->string, param->key); 13451da177e4SLinus Torvalds } 1346a137e1ccSAndi Kleen 13477ca02d0aSMike Kravetz /* 134832021982SDavid Howells * Validate the parsed options. 134932021982SDavid Howells */ 135032021982SDavid Howells static int hugetlbfs_validate(struct fs_context *fc) 135132021982SDavid Howells { 135232021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 135332021982SDavid Howells 135432021982SDavid Howells /* 13557ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 13567ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 13577ca02d0aSMike Kravetz */ 135832021982SDavid Howells ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 135932021982SDavid Howells ctx->max_size_opt, 136032021982SDavid Howells ctx->max_val_type); 136132021982SDavid Howells ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 136232021982SDavid Howells ctx->min_size_opt, 136332021982SDavid Howells ctx->min_val_type); 13647ca02d0aSMike Kravetz 13657ca02d0aSMike Kravetz /* 13667ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 13677ca02d0aSMike Kravetz */ 136832021982SDavid Howells if (ctx->max_val_type > NO_SIZE && 136932021982SDavid Howells ctx->min_hpages > ctx->max_hpages) { 137032021982SDavid Howells pr_err("Minimum size can not be greater than maximum size\n"); 13717ca02d0aSMike Kravetz return -EINVAL; 1372a137e1ccSAndi Kleen } 1373a137e1ccSAndi Kleen 13741da177e4SLinus Torvalds return 0; 13751da177e4SLinus Torvalds } 13761da177e4SLinus Torvalds 13771da177e4SLinus Torvalds static int 137832021982SDavid Howells hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc) 13791da177e4SLinus Torvalds { 138032021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 13811da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 13821da177e4SLinus Torvalds 13831da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 13841da177e4SLinus Torvalds if (!sbinfo) 13851da177e4SLinus Torvalds return -ENOMEM; 13861da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 13871da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 138832021982SDavid Howells sbinfo->hstate = ctx->hstate; 138932021982SDavid Howells sbinfo->max_inodes = ctx->nr_inodes; 139032021982SDavid Howells sbinfo->free_inodes = ctx->nr_inodes; 139190481622SDavid Gibson sbinfo->spool = NULL; 139232021982SDavid Howells sbinfo->uid = ctx->uid; 139332021982SDavid Howells sbinfo->gid = ctx->gid; 139432021982SDavid Howells sbinfo->mode = ctx->mode; 13954a25220dSDavid Howells 13967ca02d0aSMike Kravetz /* 13977ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 13981935ebd3SMiaohe Lin * specified. Any needed reservations (for minimum size) are taken 13997ca02d0aSMike Kravetz * taken when the subpool is created. 14007ca02d0aSMike Kravetz */ 140132021982SDavid Howells if (ctx->max_hpages != -1 || ctx->min_hpages != -1) { 140232021982SDavid Howells sbinfo->spool = hugepage_new_subpool(ctx->hstate, 140332021982SDavid Howells ctx->max_hpages, 140432021982SDavid Howells ctx->min_hpages); 140590481622SDavid Gibson if (!sbinfo->spool) 140690481622SDavid Gibson goto out_free; 140790481622SDavid Gibson } 14081da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 140932021982SDavid Howells sb->s_blocksize = huge_page_size(ctx->hstate); 141032021982SDavid Howells sb->s_blocksize_bits = huge_page_shift(ctx->hstate); 14111da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 14121da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 14131da177e4SLinus Torvalds sb->s_time_gran = 1; 141415568299SMike Kravetz 141515568299SMike Kravetz /* 141615568299SMike Kravetz * Due to the special and limited functionality of hugetlbfs, it does 141715568299SMike Kravetz * not work well as a stacking filesystem. 141815568299SMike Kravetz */ 141915568299SMike Kravetz sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH; 142032021982SDavid Howells sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx)); 142148fde701SAl Viro if (!sb->s_root) 14221da177e4SLinus Torvalds goto out_free; 14231da177e4SLinus Torvalds return 0; 14241da177e4SLinus Torvalds out_free: 142590481622SDavid Gibson kfree(sbinfo->spool); 14261da177e4SLinus Torvalds kfree(sbinfo); 14271da177e4SLinus Torvalds return -ENOMEM; 14281da177e4SLinus Torvalds } 14291da177e4SLinus Torvalds 143032021982SDavid Howells static int hugetlbfs_get_tree(struct fs_context *fc) 14311da177e4SLinus Torvalds { 143232021982SDavid Howells int err = hugetlbfs_validate(fc); 143332021982SDavid Howells if (err) 143432021982SDavid Howells return err; 14352ac295d4SAl Viro return get_tree_nodev(fc, hugetlbfs_fill_super); 143632021982SDavid Howells } 143732021982SDavid Howells 143832021982SDavid Howells static void hugetlbfs_fs_context_free(struct fs_context *fc) 143932021982SDavid Howells { 144032021982SDavid Howells kfree(fc->fs_private); 144132021982SDavid Howells } 144232021982SDavid Howells 144332021982SDavid Howells static const struct fs_context_operations hugetlbfs_fs_context_ops = { 144432021982SDavid Howells .free = hugetlbfs_fs_context_free, 144532021982SDavid Howells .parse_param = hugetlbfs_parse_param, 144632021982SDavid Howells .get_tree = hugetlbfs_get_tree, 144732021982SDavid Howells }; 144832021982SDavid Howells 144932021982SDavid Howells static int hugetlbfs_init_fs_context(struct fs_context *fc) 145032021982SDavid Howells { 145132021982SDavid Howells struct hugetlbfs_fs_context *ctx; 145232021982SDavid Howells 145332021982SDavid Howells ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL); 145432021982SDavid Howells if (!ctx) 145532021982SDavid Howells return -ENOMEM; 145632021982SDavid Howells 145732021982SDavid Howells ctx->max_hpages = -1; /* No limit on size by default */ 145832021982SDavid Howells ctx->nr_inodes = -1; /* No limit on number of inodes by default */ 145932021982SDavid Howells ctx->uid = current_fsuid(); 146032021982SDavid Howells ctx->gid = current_fsgid(); 146132021982SDavid Howells ctx->mode = 0755; 146232021982SDavid Howells ctx->hstate = &default_hstate; 146332021982SDavid Howells ctx->min_hpages = -1; /* No default minimum size */ 146432021982SDavid Howells ctx->max_val_type = NO_SIZE; 146532021982SDavid Howells ctx->min_val_type = NO_SIZE; 146632021982SDavid Howells fc->fs_private = ctx; 146732021982SDavid Howells fc->ops = &hugetlbfs_fs_context_ops; 146832021982SDavid Howells return 0; 14691da177e4SLinus Torvalds } 14701da177e4SLinus Torvalds 14711da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 14721da177e4SLinus Torvalds .name = "hugetlbfs", 147332021982SDavid Howells .init_fs_context = hugetlbfs_init_fs_context, 1474d7167b14SAl Viro .parameters = hugetlb_fs_parameters, 14751da177e4SLinus Torvalds .kill_sb = kill_litter_super, 14761da177e4SLinus Torvalds }; 14771da177e4SLinus Torvalds 147842d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 14791da177e4SLinus Torvalds 1480ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 14811da177e4SLinus Torvalds { 1482a0eb3a05SEric W. Biederman kgid_t shm_group; 1483a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 1484a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 14851da177e4SLinus Torvalds } 14861da177e4SLinus Torvalds 148742d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 148842d7395fSAndi Kleen { 1489af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 149042d7395fSAndi Kleen 149142d7395fSAndi Kleen if (!h) 149242d7395fSAndi Kleen return -1; 149304adbc3fSMiaohe Lin return hstate_index(h); 149442d7395fSAndi Kleen } 149542d7395fSAndi Kleen 1496af73e4d9SNaoya Horiguchi /* 1497af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 1498af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 1499af73e4d9SNaoya Horiguchi */ 1500af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 150183c1fd76Szhangyiru vm_flags_t acctflag, int creat_flags, 150283c1fd76Szhangyiru int page_size_log) 15031da177e4SLinus Torvalds { 15041da177e4SLinus Torvalds struct inode *inode; 1505e68375c8SAl Viro struct vfsmount *mnt; 150642d7395fSAndi Kleen int hstate_idx; 1507e68375c8SAl Viro struct file *file; 150842d7395fSAndi Kleen 150942d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 151042d7395fSAndi Kleen if (hstate_idx < 0) 151142d7395fSAndi Kleen return ERR_PTR(-ENODEV); 15121da177e4SLinus Torvalds 1513e68375c8SAl Viro mnt = hugetlbfs_vfsmount[hstate_idx]; 1514e68375c8SAl Viro if (!mnt) 15155bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 15165bc98594SAkinobu Mita 1517ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 151883c1fd76Szhangyiru struct ucounts *ucounts = current_ucounts(); 151983c1fd76Szhangyiru 152083c1fd76Szhangyiru if (user_shm_lock(size, ucounts)) { 152183c1fd76Szhangyiru pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is obsolete\n", 152221a3c273SDavid Rientjes current->comm, current->pid); 152383c1fd76Szhangyiru user_shm_unlock(size, ucounts); 15242584e517SRavikiran G Thirumalai } 152583c1fd76Szhangyiru return ERR_PTR(-EPERM); 1526353d5c30SHugh Dickins } 15271da177e4SLinus Torvalds 152839b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 1529e68375c8SAl Viro inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0); 15301da177e4SLinus Torvalds if (!inode) 1531e68375c8SAl Viro goto out; 1532e1832f29SStephen Smalley if (creat_flags == HUGETLB_SHMFS_INODE) 1533e1832f29SStephen Smalley inode->i_flags |= S_PRIVATE; 15341da177e4SLinus Torvalds 15351da177e4SLinus Torvalds inode->i_size = size; 15366d6b77f1SMiklos Szeredi clear_nlink(inode); 1537ce8d2cdfSDave Hansen 153833b8f84aSMike Kravetz if (!hugetlb_reserve_pages(inode, 0, 1539e68375c8SAl Viro size >> huge_page_shift(hstate_inode(inode)), NULL, 1540e68375c8SAl Viro acctflag)) 1541e68375c8SAl Viro file = ERR_PTR(-ENOMEM); 1542e68375c8SAl Viro else 1543e68375c8SAl Viro file = alloc_file_pseudo(inode, mnt, name, O_RDWR, 1544ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 1545e68375c8SAl Viro if (!IS_ERR(file)) 15461da177e4SLinus Torvalds return file; 15471da177e4SLinus Torvalds 1548b45b5bd6SDavid Gibson iput(inode); 1549e68375c8SAl Viro out: 155039b65252SAnatol Pomozov return file; 15511da177e4SLinus Torvalds } 15521da177e4SLinus Torvalds 155332021982SDavid Howells static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h) 155432021982SDavid Howells { 155532021982SDavid Howells struct fs_context *fc; 155632021982SDavid Howells struct vfsmount *mnt; 155732021982SDavid Howells 155832021982SDavid Howells fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT); 155932021982SDavid Howells if (IS_ERR(fc)) { 156032021982SDavid Howells mnt = ERR_CAST(fc); 156132021982SDavid Howells } else { 156232021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 156332021982SDavid Howells ctx->hstate = h; 156432021982SDavid Howells mnt = fc_mount(fc); 156532021982SDavid Howells put_fs_context(fc); 156632021982SDavid Howells } 156732021982SDavid Howells if (IS_ERR(mnt)) 1568a25fddceSMiaohe Lin pr_err("Cannot mount internal hugetlbfs for page size %luK", 1569*d0036517SMiaohe Lin huge_page_size(h) / SZ_1K); 157032021982SDavid Howells return mnt; 157132021982SDavid Howells } 157232021982SDavid Howells 15731da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 15741da177e4SLinus Torvalds { 157532021982SDavid Howells struct vfsmount *mnt; 157642d7395fSAndi Kleen struct hstate *h; 15771da177e4SLinus Torvalds int error; 157842d7395fSAndi Kleen int i; 15791da177e4SLinus Torvalds 1580457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 15819b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1582457c1b27SNishanth Aravamudan return -ENOTSUPP; 1583457c1b27SNishanth Aravamudan } 1584457c1b27SNishanth Aravamudan 1585d1d5e05fSHillf Danton error = -ENOMEM; 15861da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 15871da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 15885d097056SVladimir Davydov 0, SLAB_ACCOUNT, init_once); 15891da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 15908fc312b3SMike Kravetz goto out; 15911da177e4SLinus Torvalds 15921da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 15931da177e4SLinus Torvalds if (error) 15948fc312b3SMike Kravetz goto out_free; 15951da177e4SLinus Torvalds 15968fc312b3SMike Kravetz /* default hstate mount is required */ 15973b2275a8SMiaohe Lin mnt = mount_one_hugetlbfs(&default_hstate); 15988fc312b3SMike Kravetz if (IS_ERR(mnt)) { 15998fc312b3SMike Kravetz error = PTR_ERR(mnt); 16008fc312b3SMike Kravetz goto out_unreg; 16018fc312b3SMike Kravetz } 16028fc312b3SMike Kravetz hugetlbfs_vfsmount[default_hstate_idx] = mnt; 16038fc312b3SMike Kravetz 16048fc312b3SMike Kravetz /* other hstates are optional */ 160542d7395fSAndi Kleen i = 0; 160642d7395fSAndi Kleen for_each_hstate(h) { 160715f0ec94SJan Stancek if (i == default_hstate_idx) { 160815f0ec94SJan Stancek i++; 16098fc312b3SMike Kravetz continue; 161015f0ec94SJan Stancek } 16118fc312b3SMike Kravetz 161232021982SDavid Howells mnt = mount_one_hugetlbfs(h); 16138fc312b3SMike Kravetz if (IS_ERR(mnt)) 16148fc312b3SMike Kravetz hugetlbfs_vfsmount[i] = NULL; 16158fc312b3SMike Kravetz else 161632021982SDavid Howells hugetlbfs_vfsmount[i] = mnt; 161742d7395fSAndi Kleen i++; 161842d7395fSAndi Kleen } 161932021982SDavid Howells 162042d7395fSAndi Kleen return 0; 16211da177e4SLinus Torvalds 16228fc312b3SMike Kravetz out_unreg: 16238fc312b3SMike Kravetz (void)unregister_filesystem(&hugetlbfs_fs_type); 16248fc312b3SMike Kravetz out_free: 16251da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 16268fc312b3SMike Kravetz out: 16271da177e4SLinus Torvalds return error; 16281da177e4SLinus Torvalds } 16293e89e1c5SPaul Gortmaker fs_initcall(init_hugetlbfs_fs) 1630