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> 30e73a75faSRandy Dunlap #include <linux/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> 411da177e4SLinus Torvalds 42ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops; 43f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops; 444b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations; 4592e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations; 4692e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations; 471da177e4SLinus Torvalds 48a1d776eeSDavid Gibson struct hugetlbfs_config { 494a25220dSDavid Howells struct hstate *hstate; 504a25220dSDavid Howells long max_hpages; 514a25220dSDavid Howells long nr_inodes; 524a25220dSDavid Howells long min_hpages; 53a0eb3a05SEric W. Biederman kuid_t uid; 54a0eb3a05SEric W. Biederman kgid_t gid; 55a1d776eeSDavid Gibson umode_t mode; 56a1d776eeSDavid Gibson }; 57a1d776eeSDavid Gibson 58a1d776eeSDavid Gibson struct hugetlbfs_inode_info { 59a1d776eeSDavid Gibson struct shared_policy policy; 60a1d776eeSDavid Gibson struct inode vfs_inode; 61a1d776eeSDavid Gibson }; 62a1d776eeSDavid Gibson 63a1d776eeSDavid Gibson static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode) 64a1d776eeSDavid Gibson { 65a1d776eeSDavid Gibson return container_of(inode, struct hugetlbfs_inode_info, vfs_inode); 66a1d776eeSDavid Gibson } 67a1d776eeSDavid Gibson 681da177e4SLinus Torvalds int sysctl_hugetlb_shm_group; 691da177e4SLinus Torvalds 70e73a75faSRandy Dunlap enum { 71e73a75faSRandy Dunlap Opt_size, Opt_nr_inodes, 72e73a75faSRandy Dunlap Opt_mode, Opt_uid, Opt_gid, 737ca02d0aSMike Kravetz Opt_pagesize, Opt_min_size, 74e73a75faSRandy Dunlap Opt_err, 75e73a75faSRandy Dunlap }; 76e73a75faSRandy Dunlap 77a447c093SSteven Whitehouse static const match_table_t tokens = { 78e73a75faSRandy Dunlap {Opt_size, "size=%s"}, 79e73a75faSRandy Dunlap {Opt_nr_inodes, "nr_inodes=%s"}, 80e73a75faSRandy Dunlap {Opt_mode, "mode=%o"}, 81e73a75faSRandy Dunlap {Opt_uid, "uid=%u"}, 82e73a75faSRandy Dunlap {Opt_gid, "gid=%u"}, 83a137e1ccSAndi Kleen {Opt_pagesize, "pagesize=%s"}, 847ca02d0aSMike Kravetz {Opt_min_size, "min_size=%s"}, 85e73a75faSRandy Dunlap {Opt_err, NULL}, 86e73a75faSRandy Dunlap }; 87e73a75faSRandy Dunlap 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 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); 1241da177e4SLinus Torvalds loff_t len, vma_len; 1251da177e4SLinus Torvalds int ret; 126a5516438SAndi Kleen struct hstate *h = hstate_file(file); 1271da177e4SLinus Torvalds 12868589bc3SHugh Dickins /* 129dec4ad86SDavid Gibson * vma address alignment (but not the pgoff alignment) has 130dec4ad86SDavid Gibson * already been checked by prepare_hugepage_range. If you add 131dec4ad86SDavid Gibson * any error returns here, do so after setting VM_HUGETLB, so 132dec4ad86SDavid Gibson * is_vm_hugetlb_page tests below unmap_region go the right 133dec4ad86SDavid Gibson * way when do_mmap_pgoff unwinds (may be important on powerpc 134dec4ad86SDavid Gibson * and ia64). 13568589bc3SHugh Dickins */ 136a2fce914SNaoya Horiguchi vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; 13768589bc3SHugh Dickins vma->vm_ops = &hugetlb_vm_ops; 1381da177e4SLinus Torvalds 139045c7a3fSMike Kravetz /* 140045c7a3fSMike Kravetz * Offset passed to mmap (before page shift) could have been 141045c7a3fSMike Kravetz * negative when represented as a (l)off_t. 142045c7a3fSMike Kravetz */ 143045c7a3fSMike Kravetz if (((loff_t)vma->vm_pgoff << PAGE_SHIFT) < 0) 144045c7a3fSMike Kravetz return -EINVAL; 145045c7a3fSMike Kravetz 1462b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 147dec4ad86SDavid Gibson return -EINVAL; 148dec4ad86SDavid Gibson 1491da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 150045c7a3fSMike Kravetz len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 151045c7a3fSMike Kravetz /* check for overflow */ 152045c7a3fSMike Kravetz if (len < vma_len) 153045c7a3fSMike Kravetz return -EINVAL; 1541da177e4SLinus Torvalds 1555955102cSAl Viro inode_lock(inode); 1561da177e4SLinus Torvalds file_accessed(file); 1571da177e4SLinus Torvalds 1581da177e4SLinus Torvalds ret = -ENOMEM; 159a1e78772SMel Gorman if (hugetlb_reserve_pages(inode, 160a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1615a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1625a6fe125SMel Gorman vma->vm_flags)) 163b45b5bd6SDavid Gibson goto out; 164b45b5bd6SDavid Gibson 1654c887265SAdam Litke ret = 0; 166b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 167045c7a3fSMike Kravetz i_size_write(inode, len); 1681da177e4SLinus Torvalds out: 1695955102cSAl Viro inode_unlock(inode); 1701da177e4SLinus Torvalds 1711da177e4SLinus Torvalds return ret; 1721da177e4SLinus Torvalds } 1731da177e4SLinus Torvalds 1741da177e4SLinus Torvalds /* 175508034a3SHugh Dickins * Called under down_write(mmap_sem). 1761da177e4SLinus Torvalds */ 1771da177e4SLinus Torvalds 178d2ba27e8SAdrian Bunk #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 1791da177e4SLinus Torvalds static unsigned long 1801da177e4SLinus Torvalds hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 1811da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 1821da177e4SLinus Torvalds { 1831da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1841da177e4SLinus Torvalds struct vm_area_struct *vma; 185a5516438SAndi Kleen struct hstate *h = hstate_file(file); 18608659355SMichel Lespinasse struct vm_unmapped_area_info info; 1871da177e4SLinus Torvalds 188a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 1891da177e4SLinus Torvalds return -EINVAL; 1901da177e4SLinus Torvalds if (len > TASK_SIZE) 1911da177e4SLinus Torvalds return -ENOMEM; 1921da177e4SLinus Torvalds 193036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 194a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 195036e0856SBenjamin Herrenschmidt return -EINVAL; 196036e0856SBenjamin Herrenschmidt return addr; 197036e0856SBenjamin Herrenschmidt } 198036e0856SBenjamin Herrenschmidt 1991da177e4SLinus Torvalds if (addr) { 200a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 2011da177e4SLinus Torvalds vma = find_vma(mm, addr); 2021da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 2031be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma))) 2041da177e4SLinus Torvalds return addr; 2051da177e4SLinus Torvalds } 2061da177e4SLinus Torvalds 20708659355SMichel Lespinasse info.flags = 0; 20808659355SMichel Lespinasse info.length = len; 20908659355SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 21008659355SMichel Lespinasse info.high_limit = TASK_SIZE; 21108659355SMichel Lespinasse info.align_mask = PAGE_MASK & ~huge_page_mask(h); 21208659355SMichel Lespinasse info.align_offset = 0; 21308659355SMichel Lespinasse return vm_unmapped_area(&info); 2141da177e4SLinus Torvalds } 2151da177e4SLinus Torvalds #endif 2161da177e4SLinus Torvalds 21734d0640eSAl Viro static size_t 218e63e1e5aSBadari Pulavarty hugetlbfs_read_actor(struct page *page, unsigned long offset, 21934d0640eSAl Viro struct iov_iter *to, unsigned long size) 220e63e1e5aSBadari Pulavarty { 22134d0640eSAl Viro size_t copied = 0; 222e63e1e5aSBadari Pulavarty int i, chunksize; 223e63e1e5aSBadari Pulavarty 224e63e1e5aSBadari Pulavarty /* Find which 4k chunk and offset with in that chunk */ 22509cbfeafSKirill A. Shutemov i = offset >> PAGE_SHIFT; 22609cbfeafSKirill A. Shutemov offset = offset & ~PAGE_MASK; 227e63e1e5aSBadari Pulavarty 228e63e1e5aSBadari Pulavarty while (size) { 22934d0640eSAl Viro size_t n; 23009cbfeafSKirill A. Shutemov chunksize = PAGE_SIZE; 231e63e1e5aSBadari Pulavarty if (offset) 232e63e1e5aSBadari Pulavarty chunksize -= offset; 233e63e1e5aSBadari Pulavarty if (chunksize > size) 234e63e1e5aSBadari Pulavarty chunksize = size; 23534d0640eSAl Viro n = copy_page_to_iter(&page[i], offset, chunksize, to); 23634d0640eSAl Viro copied += n; 23734d0640eSAl Viro if (n != chunksize) 23834d0640eSAl Viro return copied; 239e63e1e5aSBadari Pulavarty offset = 0; 240e63e1e5aSBadari Pulavarty size -= chunksize; 241e63e1e5aSBadari Pulavarty i++; 242e63e1e5aSBadari Pulavarty } 24334d0640eSAl Viro return copied; 244e63e1e5aSBadari Pulavarty } 245e63e1e5aSBadari Pulavarty 246e63e1e5aSBadari Pulavarty /* 247e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 248e63e1e5aSBadari Pulavarty * data. Its *very* similar to do_generic_mapping_read(), we can't use that 249ea1754a0SKirill A. Shutemov * since it has PAGE_SIZE assumptions. 250e63e1e5aSBadari Pulavarty */ 25134d0640eSAl Viro static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) 252e63e1e5aSBadari Pulavarty { 25334d0640eSAl Viro struct file *file = iocb->ki_filp; 25434d0640eSAl Viro struct hstate *h = hstate_file(file); 25534d0640eSAl Viro struct address_space *mapping = file->f_mapping; 256e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 25734d0640eSAl Viro unsigned long index = iocb->ki_pos >> huge_page_shift(h); 25834d0640eSAl Viro unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); 259e63e1e5aSBadari Pulavarty unsigned long end_index; 260e63e1e5aSBadari Pulavarty loff_t isize; 261e63e1e5aSBadari Pulavarty ssize_t retval = 0; 262e63e1e5aSBadari Pulavarty 26334d0640eSAl Viro while (iov_iter_count(to)) { 264e63e1e5aSBadari Pulavarty struct page *page; 26534d0640eSAl Viro size_t nr, copied; 266e63e1e5aSBadari Pulavarty 267e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 268a5516438SAndi Kleen nr = huge_page_size(h); 269a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 270a05b0855SAneesh Kumar K.V if (!isize) 27134d0640eSAl Viro break; 272a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 273e63e1e5aSBadari Pulavarty if (index > end_index) 27434d0640eSAl Viro break; 27534d0640eSAl Viro if (index == end_index) { 276a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 277a05b0855SAneesh Kumar K.V if (nr <= offset) 27834d0640eSAl Viro break; 279e63e1e5aSBadari Pulavarty } 280e63e1e5aSBadari Pulavarty nr = nr - offset; 281e63e1e5aSBadari Pulavarty 282e63e1e5aSBadari Pulavarty /* Find the page */ 283a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 284e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 285e63e1e5aSBadari Pulavarty /* 286e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 287e63e1e5aSBadari Pulavarty * length of the hole or request. 288e63e1e5aSBadari Pulavarty */ 28934d0640eSAl Viro copied = iov_iter_zero(nr, to); 290e63e1e5aSBadari Pulavarty } else { 291a05b0855SAneesh Kumar K.V unlock_page(page); 292a05b0855SAneesh Kumar K.V 293e63e1e5aSBadari Pulavarty /* 294e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 295e63e1e5aSBadari Pulavarty */ 29634d0640eSAl Viro copied = hugetlbfs_read_actor(page, offset, to, nr); 29709cbfeafSKirill A. Shutemov put_page(page); 298e63e1e5aSBadari Pulavarty } 29934d0640eSAl Viro offset += copied; 30034d0640eSAl Viro retval += copied; 30134d0640eSAl Viro if (copied != nr && iov_iter_count(to)) { 30234d0640eSAl Viro if (!retval) 30334d0640eSAl Viro retval = -EFAULT; 304e63e1e5aSBadari Pulavarty break; 305e63e1e5aSBadari Pulavarty } 30634d0640eSAl Viro index += offset >> huge_page_shift(h); 30734d0640eSAl Viro offset &= ~huge_page_mask(h); 30834d0640eSAl Viro } 30934d0640eSAl Viro iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; 310e63e1e5aSBadari Pulavarty return retval; 311e63e1e5aSBadari Pulavarty } 312e63e1e5aSBadari Pulavarty 313800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 314800d15a5SNick Piggin struct address_space *mapping, 315800d15a5SNick Piggin loff_t pos, unsigned len, unsigned flags, 316800d15a5SNick Piggin struct page **pagep, void **fsdata) 3171da177e4SLinus Torvalds { 3181da177e4SLinus Torvalds return -EINVAL; 3191da177e4SLinus Torvalds } 3201da177e4SLinus Torvalds 321800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 322800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 323800d15a5SNick Piggin struct page *page, void *fsdata) 3241da177e4SLinus Torvalds { 325800d15a5SNick Piggin BUG(); 3261da177e4SLinus Torvalds return -EINVAL; 3271da177e4SLinus Torvalds } 3281da177e4SLinus Torvalds 329b5cec28dSMike Kravetz static void remove_huge_page(struct page *page) 3301da177e4SLinus Torvalds { 331b9ea2515SKonstantin Khlebnikov ClearPageDirty(page); 3321da177e4SLinus Torvalds ClearPageUptodate(page); 333bd65cb86SMinchan Kim delete_from_page_cache(page); 3341da177e4SLinus Torvalds } 3351da177e4SLinus Torvalds 3364aae8d1cSMike Kravetz static void 3374aae8d1cSMike Kravetz hugetlb_vmdelete_list(struct rb_root *root, pgoff_t start, pgoff_t end) 3384aae8d1cSMike Kravetz { 3394aae8d1cSMike Kravetz struct vm_area_struct *vma; 3404aae8d1cSMike Kravetz 3414aae8d1cSMike Kravetz /* 3424aae8d1cSMike Kravetz * end == 0 indicates that the entire range after 3434aae8d1cSMike Kravetz * start should be unmapped. 3444aae8d1cSMike Kravetz */ 3454aae8d1cSMike Kravetz vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { 3464aae8d1cSMike Kravetz unsigned long v_offset; 3474aae8d1cSMike Kravetz unsigned long v_end; 3484aae8d1cSMike Kravetz 3494aae8d1cSMike Kravetz /* 3504aae8d1cSMike Kravetz * Can the expression below overflow on 32-bit arches? 3514aae8d1cSMike Kravetz * No, because the interval tree returns us only those vmas 3524aae8d1cSMike Kravetz * which overlap the truncated area starting at pgoff, 3534aae8d1cSMike Kravetz * and no vma on a 32-bit arch can span beyond the 4GB. 3544aae8d1cSMike Kravetz */ 3554aae8d1cSMike Kravetz if (vma->vm_pgoff < start) 3564aae8d1cSMike Kravetz v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; 3574aae8d1cSMike Kravetz else 3584aae8d1cSMike Kravetz v_offset = 0; 3594aae8d1cSMike Kravetz 3604aae8d1cSMike Kravetz if (!end) 3614aae8d1cSMike Kravetz v_end = vma->vm_end; 3624aae8d1cSMike Kravetz else { 3634aae8d1cSMike Kravetz v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) 3644aae8d1cSMike Kravetz + vma->vm_start; 3654aae8d1cSMike Kravetz if (v_end > vma->vm_end) 3664aae8d1cSMike Kravetz v_end = vma->vm_end; 3674aae8d1cSMike Kravetz } 3684aae8d1cSMike Kravetz 3694aae8d1cSMike Kravetz unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, 3704aae8d1cSMike Kravetz NULL); 3714aae8d1cSMike Kravetz } 3724aae8d1cSMike Kravetz } 373b5cec28dSMike Kravetz 374b5cec28dSMike Kravetz /* 375b5cec28dSMike Kravetz * remove_inode_hugepages handles two distinct cases: truncation and hole 376b5cec28dSMike Kravetz * punch. There are subtle differences in operation for each case. 3774aae8d1cSMike Kravetz * 378b5cec28dSMike Kravetz * truncation is indicated by end of range being LLONG_MAX 379b5cec28dSMike Kravetz * In this case, we first scan the range and release found pages. 380b5cec28dSMike Kravetz * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv 3811817889eSMike Kravetz * maps and global counts. Page faults can not race with truncation 3821817889eSMike Kravetz * in this routine. hugetlb_no_page() prevents page faults in the 3831817889eSMike Kravetz * truncated range. It checks i_size before allocation, and again after 3841817889eSMike Kravetz * with the page table lock for the page held. The same lock must be 3851817889eSMike Kravetz * acquired to unmap a page. 386b5cec28dSMike Kravetz * hole punch is indicated if end is not LLONG_MAX 387b5cec28dSMike Kravetz * In the hole punch case we scan the range and release found pages. 388b5cec28dSMike Kravetz * Only when releasing a page is the associated region/reserv map 389b5cec28dSMike Kravetz * deleted. The region/reserv map for ranges without associated 3901817889eSMike Kravetz * pages are not modified. Page faults can race with hole punch. 3911817889eSMike Kravetz * This is indicated if we find a mapped page. 392b5cec28dSMike Kravetz * Note: If the passed end of range value is beyond the end of file, but 393b5cec28dSMike Kravetz * not LLONG_MAX this routine still performs a hole punch operation. 394b5cec28dSMike Kravetz */ 395b5cec28dSMike Kravetz static void remove_inode_hugepages(struct inode *inode, loff_t lstart, 396b5cec28dSMike Kravetz loff_t lend) 3971da177e4SLinus Torvalds { 398a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 399b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 400a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 401b5cec28dSMike Kravetz const pgoff_t end = lend >> huge_page_shift(h); 402b5cec28dSMike Kravetz struct vm_area_struct pseudo_vma; 4031da177e4SLinus Torvalds struct pagevec pvec; 404d72dc8a2SJan Kara pgoff_t next, index; 405a43a8c39SChen, Kenneth W int i, freed = 0; 406b5cec28dSMike Kravetz long lookup_nr = PAGEVEC_SIZE; 407b5cec28dSMike Kravetz bool truncate_op = (lend == LLONG_MAX); 4081da177e4SLinus Torvalds 409b5cec28dSMike Kravetz memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); 410b5cec28dSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 4111da177e4SLinus Torvalds pagevec_init(&pvec, 0); 4121da177e4SLinus Torvalds next = start; 413b5cec28dSMike Kravetz while (next < end) { 414b5cec28dSMike Kravetz /* 4151817889eSMike Kravetz * Don't grab more pages than the number left in the range. 416b5cec28dSMike Kravetz */ 417b5cec28dSMike Kravetz if (end - next < lookup_nr) 418b5cec28dSMike Kravetz lookup_nr = end - next; 419b5cec28dSMike Kravetz 420b5cec28dSMike Kravetz /* 4211817889eSMike Kravetz * When no more pages are found, we are done. 422b5cec28dSMike Kravetz */ 423d72dc8a2SJan Kara if (!pagevec_lookup(&pvec, mapping, &next, lookup_nr)) 4241da177e4SLinus Torvalds break; 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); ++i) { 4271da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 428b5cec28dSMike Kravetz u32 hash; 429b5cec28dSMike Kravetz 4301817889eSMike Kravetz /* 4311817889eSMike Kravetz * The page (index) could be beyond end. This is 4321817889eSMike Kravetz * only possible in the punch hole case as end is 4331817889eSMike Kravetz * max page offset in the truncate case. 4341817889eSMike Kravetz */ 435d72dc8a2SJan Kara index = page->index; 436d72dc8a2SJan Kara if (index >= end) 4371817889eSMike Kravetz break; 4381817889eSMike Kravetz 439b5cec28dSMike Kravetz hash = hugetlb_fault_mutex_hash(h, current->mm, 440b5cec28dSMike Kravetz &pseudo_vma, 441d72dc8a2SJan Kara mapping, index, 0); 442b5cec28dSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 4431da177e4SLinus Torvalds 444b5cec28dSMike Kravetz /* 4454aae8d1cSMike Kravetz * If page is mapped, it was faulted in after being 4464aae8d1cSMike Kravetz * unmapped in caller. Unmap (again) now after taking 4474aae8d1cSMike Kravetz * the fault mutex. The mutex will prevent faults 4484aae8d1cSMike Kravetz * until we finish removing the page. 4494aae8d1cSMike Kravetz * 4504aae8d1cSMike Kravetz * This race can only happen in the hole punch case. 4514aae8d1cSMike Kravetz * Getting here in a truncate operation is a bug. 452b5cec28dSMike Kravetz */ 4534aae8d1cSMike Kravetz if (unlikely(page_mapped(page))) { 4544aae8d1cSMike Kravetz BUG_ON(truncate_op); 4554aae8d1cSMike Kravetz 4564aae8d1cSMike Kravetz i_mmap_lock_write(mapping); 4574aae8d1cSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 458d72dc8a2SJan Kara index * pages_per_huge_page(h), 459d72dc8a2SJan Kara (index + 1) * pages_per_huge_page(h)); 4604aae8d1cSMike Kravetz i_mmap_unlock_write(mapping); 4614aae8d1cSMike Kravetz } 4624aae8d1cSMike Kravetz 4634aae8d1cSMike Kravetz lock_page(page); 4644aae8d1cSMike Kravetz /* 4654aae8d1cSMike Kravetz * We must free the huge page and remove from page 4664aae8d1cSMike Kravetz * cache (remove_huge_page) BEFORE removing the 4674aae8d1cSMike Kravetz * region/reserve map (hugetlb_unreserve_pages). In 4684aae8d1cSMike Kravetz * rare out of memory conditions, removal of the 46972e2936cSzhong jiang * region/reserve map could fail. Correspondingly, 47072e2936cSzhong jiang * the subpool and global reserve usage count can need 47172e2936cSzhong jiang * to be adjusted. 4724aae8d1cSMike Kravetz */ 47372e2936cSzhong jiang VM_BUG_ON(PagePrivate(page)); 474b5cec28dSMike Kravetz remove_huge_page(page); 475b5cec28dSMike Kravetz freed++; 476b5cec28dSMike Kravetz if (!truncate_op) { 4774aae8d1cSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, 478d72dc8a2SJan Kara index, index + 1, 1))) 47972e2936cSzhong jiang hugetlb_fix_reserve_counts(inode); 480b5cec28dSMike Kravetz } 481b5cec28dSMike Kravetz 4821da177e4SLinus Torvalds unlock_page(page); 483b5cec28dSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 4841da177e4SLinus Torvalds } 4851da177e4SLinus Torvalds huge_pagevec_release(&pvec); 4861817889eSMike Kravetz cond_resched(); 4871da177e4SLinus Torvalds } 488b5cec28dSMike Kravetz 489b5cec28dSMike Kravetz if (truncate_op) 490b5cec28dSMike Kravetz (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); 4911da177e4SLinus Torvalds } 4921da177e4SLinus Torvalds 4932bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 4941da177e4SLinus Torvalds { 4959119a41eSJoonsoo Kim struct resv_map *resv_map; 4969119a41eSJoonsoo Kim 497b5cec28dSMike Kravetz remove_inode_hugepages(inode, 0, LLONG_MAX); 4989119a41eSJoonsoo Kim resv_map = (struct resv_map *)inode->i_mapping->private_data; 4999119a41eSJoonsoo Kim /* root inode doesn't have the resv_map, so we should check it */ 5009119a41eSJoonsoo Kim if (resv_map) 5019119a41eSJoonsoo Kim resv_map_release(&resv_map->refs); 502dbd5768fSJan Kara clear_inode(inode); 503149f4211SChristoph Hellwig } 504149f4211SChristoph Hellwig 5051da177e4SLinus Torvalds static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) 5061da177e4SLinus Torvalds { 507856fc295SHugh Dickins pgoff_t pgoff; 5081da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 509a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 5101da177e4SLinus Torvalds 511a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 512856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 5131da177e4SLinus Torvalds 5147aa91e10SKen Chen i_size_write(inode, offset); 51583cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 5166b2dbba8SMichel Lespinasse if (!RB_EMPTY_ROOT(&mapping->i_mmap)) 5171bfad99aSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); 51883cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 519b5cec28dSMike Kravetz remove_inode_hugepages(inode, offset, LLONG_MAX); 5201da177e4SLinus Torvalds return 0; 5211da177e4SLinus Torvalds } 5221da177e4SLinus Torvalds 52370c3547eSMike Kravetz static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) 52470c3547eSMike Kravetz { 52570c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 52670c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 52770c3547eSMike Kravetz loff_t hole_start, hole_end; 52870c3547eSMike Kravetz 52970c3547eSMike Kravetz /* 53070c3547eSMike Kravetz * For hole punch round up the beginning offset of the hole and 53170c3547eSMike Kravetz * round down the end. 53270c3547eSMike Kravetz */ 53370c3547eSMike Kravetz hole_start = round_up(offset, hpage_size); 53470c3547eSMike Kravetz hole_end = round_down(offset + len, hpage_size); 53570c3547eSMike Kravetz 53670c3547eSMike Kravetz if (hole_end > hole_start) { 53770c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 53870c3547eSMike Kravetz 5395955102cSAl Viro inode_lock(inode); 54070c3547eSMike Kravetz i_mmap_lock_write(mapping); 54170c3547eSMike Kravetz if (!RB_EMPTY_ROOT(&mapping->i_mmap)) 54270c3547eSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 54370c3547eSMike Kravetz hole_start >> PAGE_SHIFT, 54470c3547eSMike Kravetz hole_end >> PAGE_SHIFT); 54570c3547eSMike Kravetz i_mmap_unlock_write(mapping); 54670c3547eSMike Kravetz remove_inode_hugepages(inode, hole_start, hole_end); 5475955102cSAl Viro inode_unlock(inode); 54870c3547eSMike Kravetz } 54970c3547eSMike Kravetz 55070c3547eSMike Kravetz return 0; 55170c3547eSMike Kravetz } 55270c3547eSMike Kravetz 55370c3547eSMike Kravetz static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, 55470c3547eSMike Kravetz loff_t len) 55570c3547eSMike Kravetz { 55670c3547eSMike Kravetz struct inode *inode = file_inode(file); 55770c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 55870c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 55970c3547eSMike Kravetz struct vm_area_struct pseudo_vma; 56070c3547eSMike Kravetz struct mm_struct *mm = current->mm; 56170c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 56270c3547eSMike Kravetz unsigned long hpage_shift = huge_page_shift(h); 56370c3547eSMike Kravetz pgoff_t start, index, end; 56470c3547eSMike Kravetz int error; 56570c3547eSMike Kravetz u32 hash; 56670c3547eSMike Kravetz 56770c3547eSMike Kravetz if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 56870c3547eSMike Kravetz return -EOPNOTSUPP; 56970c3547eSMike Kravetz 57070c3547eSMike Kravetz if (mode & FALLOC_FL_PUNCH_HOLE) 57170c3547eSMike Kravetz return hugetlbfs_punch_hole(inode, offset, len); 57270c3547eSMike Kravetz 57370c3547eSMike Kravetz /* 57470c3547eSMike Kravetz * Default preallocate case. 57570c3547eSMike Kravetz * For this range, start is rounded down and end is rounded up 57670c3547eSMike Kravetz * as well as being converted to page offsets. 57770c3547eSMike Kravetz */ 57870c3547eSMike Kravetz start = offset >> hpage_shift; 57970c3547eSMike Kravetz end = (offset + len + hpage_size - 1) >> hpage_shift; 58070c3547eSMike Kravetz 5815955102cSAl Viro inode_lock(inode); 58270c3547eSMike Kravetz 58370c3547eSMike Kravetz /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ 58470c3547eSMike Kravetz error = inode_newsize_ok(inode, offset + len); 58570c3547eSMike Kravetz if (error) 58670c3547eSMike Kravetz goto out; 58770c3547eSMike Kravetz 58870c3547eSMike Kravetz /* 58970c3547eSMike Kravetz * Initialize a pseudo vma as this is required by the huge page 59070c3547eSMike Kravetz * allocation routines. If NUMA is configured, use page index 59170c3547eSMike Kravetz * as input to create an allocation policy. 59270c3547eSMike Kravetz */ 59370c3547eSMike Kravetz memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); 59470c3547eSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 59570c3547eSMike Kravetz pseudo_vma.vm_file = file; 59670c3547eSMike Kravetz 59770c3547eSMike Kravetz for (index = start; index < end; index++) { 59870c3547eSMike Kravetz /* 59970c3547eSMike Kravetz * This is supposed to be the vaddr where the page is being 60070c3547eSMike Kravetz * faulted in, but we have no vaddr here. 60170c3547eSMike Kravetz */ 60270c3547eSMike Kravetz struct page *page; 60370c3547eSMike Kravetz unsigned long addr; 60470c3547eSMike Kravetz int avoid_reserve = 0; 60570c3547eSMike Kravetz 60670c3547eSMike Kravetz cond_resched(); 60770c3547eSMike Kravetz 60870c3547eSMike Kravetz /* 60970c3547eSMike Kravetz * fallocate(2) manpage permits EINTR; we may have been 61070c3547eSMike Kravetz * interrupted because we are using up too much memory. 61170c3547eSMike Kravetz */ 61270c3547eSMike Kravetz if (signal_pending(current)) { 61370c3547eSMike Kravetz error = -EINTR; 61470c3547eSMike Kravetz break; 61570c3547eSMike Kravetz } 61670c3547eSMike Kravetz 61770c3547eSMike Kravetz /* Set numa allocation policy based on index */ 61870c3547eSMike Kravetz hugetlb_set_vma_policy(&pseudo_vma, inode, index); 61970c3547eSMike Kravetz 62070c3547eSMike Kravetz /* addr is the offset within the file (zero based) */ 62170c3547eSMike Kravetz addr = index * hpage_size; 62270c3547eSMike Kravetz 62370c3547eSMike Kravetz /* mutex taken here, fault path and hole punch */ 62470c3547eSMike Kravetz hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping, 62570c3547eSMike Kravetz index, addr); 62670c3547eSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 62770c3547eSMike Kravetz 62870c3547eSMike Kravetz /* See if already present in mapping to avoid alloc/free */ 62970c3547eSMike Kravetz page = find_get_page(mapping, index); 63070c3547eSMike Kravetz if (page) { 63170c3547eSMike Kravetz put_page(page); 63270c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 63370c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 63470c3547eSMike Kravetz continue; 63570c3547eSMike Kravetz } 63670c3547eSMike Kravetz 63770c3547eSMike Kravetz /* Allocate page and add to page cache */ 63870c3547eSMike Kravetz page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); 63970c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 64070c3547eSMike Kravetz if (IS_ERR(page)) { 64170c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 64270c3547eSMike Kravetz error = PTR_ERR(page); 64370c3547eSMike Kravetz goto out; 64470c3547eSMike Kravetz } 64570c3547eSMike Kravetz clear_huge_page(page, addr, pages_per_huge_page(h)); 64670c3547eSMike Kravetz __SetPageUptodate(page); 64770c3547eSMike Kravetz error = huge_add_to_page_cache(page, mapping, index); 64870c3547eSMike Kravetz if (unlikely(error)) { 64970c3547eSMike Kravetz put_page(page); 65070c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 65170c3547eSMike Kravetz goto out; 65270c3547eSMike Kravetz } 65370c3547eSMike Kravetz 65470c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 65570c3547eSMike Kravetz 65670c3547eSMike Kravetz /* 65770c3547eSMike Kravetz * page_put due to reference from alloc_huge_page() 65870c3547eSMike Kravetz * unlock_page because locked by add_to_page_cache() 65970c3547eSMike Kravetz */ 66070c3547eSMike Kravetz put_page(page); 66170c3547eSMike Kravetz unlock_page(page); 66270c3547eSMike Kravetz } 66370c3547eSMike Kravetz 66470c3547eSMike Kravetz if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) 66570c3547eSMike Kravetz i_size_write(inode, offset + len); 666078cd827SDeepa Dinamani inode->i_ctime = current_time(inode); 66770c3547eSMike Kravetz out: 6685955102cSAl Viro inode_unlock(inode); 66970c3547eSMike Kravetz return error; 67070c3547eSMike Kravetz } 67170c3547eSMike Kravetz 6721da177e4SLinus Torvalds static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) 6731da177e4SLinus Torvalds { 6742b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 675a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 6761da177e4SLinus Torvalds int error; 6771da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 6781da177e4SLinus Torvalds 6791da177e4SLinus Torvalds BUG_ON(!inode); 6801da177e4SLinus Torvalds 68131051c85SJan Kara error = setattr_prepare(dentry, attr); 6821da177e4SLinus Torvalds if (error) 6831025774cSChristoph Hellwig return error; 6841da177e4SLinus Torvalds 6851da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 6861da177e4SLinus Torvalds error = -EINVAL; 6871025774cSChristoph Hellwig if (attr->ia_size & ~huge_page_mask(h)) 6881025774cSChristoph Hellwig return -EINVAL; 6891da177e4SLinus Torvalds error = hugetlb_vmtruncate(inode, attr->ia_size); 6901da177e4SLinus Torvalds if (error) 6911da177e4SLinus Torvalds return error; 6921da177e4SLinus Torvalds } 6931da177e4SLinus Torvalds 6941025774cSChristoph Hellwig setattr_copy(inode, attr); 6951025774cSChristoph Hellwig mark_inode_dirty(inode); 6961025774cSChristoph Hellwig return 0; 6971025774cSChristoph Hellwig } 6981025774cSChristoph Hellwig 6997d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 7007d54fa64SAl Viro struct hugetlbfs_config *config) 7011da177e4SLinus Torvalds { 7021da177e4SLinus Torvalds struct inode *inode; 7031da177e4SLinus Torvalds 7041da177e4SLinus Torvalds inode = new_inode(sb); 7051da177e4SLinus Torvalds if (inode) { 70685fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 7077d54fa64SAl Viro inode->i_mode = S_IFDIR | config->mode; 7087d54fa64SAl Viro inode->i_uid = config->uid; 7097d54fa64SAl Viro inode->i_gid = config->gid; 710078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 7117d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 7127d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 7137d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 7147d54fa64SAl Viro inc_nlink(inode); 71565ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 7167d54fa64SAl Viro } 7177d54fa64SAl Viro return inode; 7187d54fa64SAl Viro } 7197d54fa64SAl Viro 720b610ded7SMichal Hocko /* 721c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 722b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 72388f306b6SKirill A. Shutemov * annotation because huge_pmd_share() does an allocation under hugetlb's 724c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 725b610ded7SMichal Hocko */ 726c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 727b610ded7SMichal Hocko 7287d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 7297d54fa64SAl Viro struct inode *dir, 73018df2252SAl Viro umode_t mode, dev_t dev) 7317d54fa64SAl Viro { 7327d54fa64SAl Viro struct inode *inode; 7339119a41eSJoonsoo Kim struct resv_map *resv_map; 7349119a41eSJoonsoo Kim 7359119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 7369119a41eSJoonsoo Kim if (!resv_map) 7379119a41eSJoonsoo Kim return NULL; 7387d54fa64SAl Viro 7397d54fa64SAl Viro inode = new_inode(sb); 7407d54fa64SAl Viro if (inode) { 7417d54fa64SAl Viro inode->i_ino = get_next_ino(); 7427d54fa64SAl Viro inode_init_owner(inode, dir, mode); 743c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 744c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 7451da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 746078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 7479119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 7481da177e4SLinus Torvalds switch (mode & S_IFMT) { 7491da177e4SLinus Torvalds default: 7501da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 7511da177e4SLinus Torvalds break; 7521da177e4SLinus Torvalds case S_IFREG: 7531da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 7541da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 7551da177e4SLinus Torvalds break; 7561da177e4SLinus Torvalds case S_IFDIR: 7571da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 7581da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 7591da177e4SLinus Torvalds 7601da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 761d8c76e6fSDave Hansen inc_nlink(inode); 7621da177e4SLinus Torvalds break; 7631da177e4SLinus Torvalds case S_IFLNK: 7641da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 76521fc61c7SAl Viro inode_nohighmem(inode); 7661da177e4SLinus Torvalds break; 7671da177e4SLinus Torvalds } 768e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 7699119a41eSJoonsoo Kim } else 7709119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 7719119a41eSJoonsoo Kim 7721da177e4SLinus Torvalds return inode; 7731da177e4SLinus Torvalds } 7741da177e4SLinus Torvalds 7751da177e4SLinus Torvalds /* 7761da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 7771da177e4SLinus Torvalds */ 7781da177e4SLinus Torvalds static int hugetlbfs_mknod(struct inode *dir, 7791a67aafbSAl Viro struct dentry *dentry, umode_t mode, dev_t dev) 7801da177e4SLinus Torvalds { 7811da177e4SLinus Torvalds struct inode *inode; 7821da177e4SLinus Torvalds int error = -ENOSPC; 7831da177e4SLinus Torvalds 7847d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 7851da177e4SLinus Torvalds if (inode) { 786078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 7871da177e4SLinus Torvalds d_instantiate(dentry, inode); 7881da177e4SLinus Torvalds dget(dentry); /* Extra count - pin the dentry in core */ 7891da177e4SLinus Torvalds error = 0; 7901da177e4SLinus Torvalds } 7911da177e4SLinus Torvalds return error; 7921da177e4SLinus Torvalds } 7931da177e4SLinus Torvalds 79418bb1db3SAl Viro static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 7951da177e4SLinus Torvalds { 7961da177e4SLinus Torvalds int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); 7971da177e4SLinus Torvalds if (!retval) 798d8c76e6fSDave Hansen inc_nlink(dir); 7991da177e4SLinus Torvalds return retval; 8001da177e4SLinus Torvalds } 8011da177e4SLinus Torvalds 802ebfc3b49SAl Viro static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) 8031da177e4SLinus Torvalds { 8041da177e4SLinus Torvalds return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); 8051da177e4SLinus Torvalds } 8061da177e4SLinus Torvalds 8071da177e4SLinus Torvalds static int hugetlbfs_symlink(struct inode *dir, 8081da177e4SLinus Torvalds struct dentry *dentry, const char *symname) 8091da177e4SLinus Torvalds { 8101da177e4SLinus Torvalds struct inode *inode; 8111da177e4SLinus Torvalds int error = -ENOSPC; 8121da177e4SLinus Torvalds 8137d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 8141da177e4SLinus Torvalds if (inode) { 8151da177e4SLinus Torvalds int l = strlen(symname)+1; 8161da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 8171da177e4SLinus Torvalds if (!error) { 8181da177e4SLinus Torvalds d_instantiate(dentry, inode); 8191da177e4SLinus Torvalds dget(dentry); 8201da177e4SLinus Torvalds } else 8211da177e4SLinus Torvalds iput(inode); 8221da177e4SLinus Torvalds } 823078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 8241da177e4SLinus Torvalds 8251da177e4SLinus Torvalds return error; 8261da177e4SLinus Torvalds } 8271da177e4SLinus Torvalds 8281da177e4SLinus Torvalds /* 8296649a386SKen Chen * mark the head page dirty 8301da177e4SLinus Torvalds */ 8311da177e4SLinus Torvalds static int hugetlbfs_set_page_dirty(struct page *page) 8321da177e4SLinus Torvalds { 833d85f3385SChristoph Lameter struct page *head = compound_head(page); 8346649a386SKen Chen 8356649a386SKen Chen SetPageDirty(head); 8361da177e4SLinus Torvalds return 0; 8371da177e4SLinus Torvalds } 8381da177e4SLinus Torvalds 839290408d4SNaoya Horiguchi static int hugetlbfs_migrate_page(struct address_space *mapping, 840b969c4abSMel Gorman struct page *newpage, struct page *page, 841a6bc32b8SMel Gorman enum migrate_mode mode) 842290408d4SNaoya Horiguchi { 843290408d4SNaoya Horiguchi int rc; 844290408d4SNaoya Horiguchi 845290408d4SNaoya Horiguchi rc = migrate_huge_page_move_mapping(mapping, newpage, page); 84678bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 847290408d4SNaoya Horiguchi return rc; 848290408d4SNaoya Horiguchi migrate_page_copy(newpage, page); 849290408d4SNaoya Horiguchi 85078bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 851290408d4SNaoya Horiguchi } 852290408d4SNaoya Horiguchi 85378bb9203SNaoya Horiguchi static int hugetlbfs_error_remove_page(struct address_space *mapping, 85478bb9203SNaoya Horiguchi struct page *page) 85578bb9203SNaoya Horiguchi { 85678bb9203SNaoya Horiguchi struct inode *inode = mapping->host; 85778bb9203SNaoya Horiguchi 85878bb9203SNaoya Horiguchi remove_huge_page(page); 85978bb9203SNaoya Horiguchi hugetlb_fix_reserve_counts(inode); 86078bb9203SNaoya Horiguchi return 0; 86178bb9203SNaoya Horiguchi } 86278bb9203SNaoya Horiguchi 8634a25220dSDavid Howells /* 8644a25220dSDavid Howells * Display the mount options in /proc/mounts. 8654a25220dSDavid Howells */ 8664a25220dSDavid Howells static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) 8674a25220dSDavid Howells { 8684a25220dSDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); 8694a25220dSDavid Howells struct hugepage_subpool *spool = sbinfo->spool; 8704a25220dSDavid Howells unsigned long hpage_size = huge_page_size(sbinfo->hstate); 8714a25220dSDavid Howells unsigned hpage_shift = huge_page_shift(sbinfo->hstate); 8724a25220dSDavid Howells char mod; 8734a25220dSDavid Howells 8744a25220dSDavid Howells if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) 8754a25220dSDavid Howells seq_printf(m, ",uid=%u", 8764a25220dSDavid Howells from_kuid_munged(&init_user_ns, sbinfo->uid)); 8774a25220dSDavid Howells if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) 8784a25220dSDavid Howells seq_printf(m, ",gid=%u", 8794a25220dSDavid Howells from_kgid_munged(&init_user_ns, sbinfo->gid)); 8804a25220dSDavid Howells if (sbinfo->mode != 0755) 8814a25220dSDavid Howells seq_printf(m, ",mode=%o", sbinfo->mode); 8824a25220dSDavid Howells if (sbinfo->max_inodes != -1) 8834a25220dSDavid Howells seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); 8844a25220dSDavid Howells 8854a25220dSDavid Howells hpage_size /= 1024; 8864a25220dSDavid Howells mod = 'K'; 8874a25220dSDavid Howells if (hpage_size >= 1024) { 8884a25220dSDavid Howells hpage_size /= 1024; 8894a25220dSDavid Howells mod = 'M'; 8904a25220dSDavid Howells } 8914a25220dSDavid Howells seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); 8924a25220dSDavid Howells if (spool) { 8934a25220dSDavid Howells if (spool->max_hpages != -1) 8944a25220dSDavid Howells seq_printf(m, ",size=%llu", 8954a25220dSDavid Howells (unsigned long long)spool->max_hpages << hpage_shift); 8964a25220dSDavid Howells if (spool->min_hpages != -1) 8974a25220dSDavid Howells seq_printf(m, ",min_size=%llu", 8984a25220dSDavid Howells (unsigned long long)spool->min_hpages << hpage_shift); 8994a25220dSDavid Howells } 9004a25220dSDavid Howells return 0; 9014a25220dSDavid Howells } 9024a25220dSDavid Howells 903726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 9041da177e4SLinus Torvalds { 905726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 9062b0143b5SDavid Howells struct hstate *h = hstate_inode(d_inode(dentry)); 9071da177e4SLinus Torvalds 9081da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 909a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 9101da177e4SLinus Torvalds if (sbinfo) { 9111da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 91274a8a65cSDavid Gibson /* If no limits set, just report 0 for max/free/used 91374a8a65cSDavid Gibson * blocks, like simple_statfs() */ 91490481622SDavid Gibson if (sbinfo->spool) { 91590481622SDavid Gibson long free_pages; 91690481622SDavid Gibson 91790481622SDavid Gibson spin_lock(&sbinfo->spool->lock); 91890481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 91990481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 92090481622SDavid Gibson - sbinfo->spool->used_hpages; 92190481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 92290481622SDavid Gibson spin_unlock(&sbinfo->spool->lock); 9231da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 9241da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 92574a8a65cSDavid Gibson } 9261da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 9271da177e4SLinus Torvalds } 9281da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 9291da177e4SLinus Torvalds return 0; 9301da177e4SLinus Torvalds } 9311da177e4SLinus Torvalds 9321da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 9331da177e4SLinus Torvalds { 9341da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 9351da177e4SLinus Torvalds 9361da177e4SLinus Torvalds if (sbi) { 9371da177e4SLinus Torvalds sb->s_fs_info = NULL; 93890481622SDavid Gibson 93990481622SDavid Gibson if (sbi->spool) 94090481622SDavid Gibson hugepage_put_subpool(sbi->spool); 94190481622SDavid Gibson 9421da177e4SLinus Torvalds kfree(sbi); 9431da177e4SLinus Torvalds } 9441da177e4SLinus Torvalds } 9451da177e4SLinus Torvalds 94696527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 94796527980SChristoph Hellwig { 94896527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 94996527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 95096527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 95196527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 95296527980SChristoph Hellwig return 0; 95396527980SChristoph Hellwig } 95496527980SChristoph Hellwig sbinfo->free_inodes--; 95596527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 95696527980SChristoph Hellwig } 95796527980SChristoph Hellwig 95896527980SChristoph Hellwig return 1; 95996527980SChristoph Hellwig } 96096527980SChristoph Hellwig 96196527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 96296527980SChristoph Hellwig { 96396527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 96496527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 96596527980SChristoph Hellwig sbinfo->free_inodes++; 96696527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 96796527980SChristoph Hellwig } 96896527980SChristoph Hellwig } 96996527980SChristoph Hellwig 97096527980SChristoph Hellwig 971e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 9721da177e4SLinus Torvalds 9731da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 9741da177e4SLinus Torvalds { 97596527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 9761da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 9771da177e4SLinus Torvalds 97896527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 9791da177e4SLinus Torvalds return NULL; 980e94b1766SChristoph Lameter p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); 98196527980SChristoph Hellwig if (unlikely(!p)) { 98296527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 98396527980SChristoph Hellwig return NULL; 9841da177e4SLinus Torvalds } 9854742a35dSMike Kravetz 9864742a35dSMike Kravetz /* 9874742a35dSMike Kravetz * Any time after allocation, hugetlbfs_destroy_inode can be called 9884742a35dSMike Kravetz * for the inode. mpol_free_shared_policy is unconditionally called 9894742a35dSMike Kravetz * as part of hugetlbfs_destroy_inode. So, initialize policy here 9904742a35dSMike Kravetz * in case of a quick call to destroy. 9914742a35dSMike Kravetz * 9924742a35dSMike Kravetz * Note that the policy is initialized even if we are creating a 9934742a35dSMike Kravetz * private inode. This simplifies hugetlbfs_destroy_inode. 9944742a35dSMike Kravetz */ 9954742a35dSMike Kravetz mpol_shared_policy_init(&p->policy, NULL); 9964742a35dSMike Kravetz 99796527980SChristoph Hellwig return &p->vfs_inode; 9981da177e4SLinus Torvalds } 9991da177e4SLinus Torvalds 1000fa0d7e3dSNick Piggin static void hugetlbfs_i_callback(struct rcu_head *head) 1001fa0d7e3dSNick Piggin { 1002fa0d7e3dSNick Piggin struct inode *inode = container_of(head, struct inode, i_rcu); 1003fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 1004fa0d7e3dSNick Piggin } 1005fa0d7e3dSNick Piggin 10061da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 10071da177e4SLinus Torvalds { 100896527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 10091da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 1010fa0d7e3dSNick Piggin call_rcu(&inode->i_rcu, hugetlbfs_i_callback); 10111da177e4SLinus Torvalds } 10121da177e4SLinus Torvalds 1013f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 1014800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 1015800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 10161da177e4SLinus Torvalds .set_page_dirty = hugetlbfs_set_page_dirty, 1017290408d4SNaoya Horiguchi .migratepage = hugetlbfs_migrate_page, 101878bb9203SNaoya Horiguchi .error_remove_page = hugetlbfs_error_remove_page, 10191da177e4SLinus Torvalds }; 10201da177e4SLinus Torvalds 102196527980SChristoph Hellwig 102251cc5068SAlexey Dobriyan static void init_once(void *foo) 102396527980SChristoph Hellwig { 102496527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 102596527980SChristoph Hellwig 102696527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 102796527980SChristoph Hellwig } 102896527980SChristoph Hellwig 10294b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 103034d0640eSAl Viro .read_iter = hugetlbfs_read_iter, 10311da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 10321b061d92SChristoph Hellwig .fsync = noop_fsync, 10331da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 10346038f373SArnd Bergmann .llseek = default_llseek, 103570c3547eSMike Kravetz .fallocate = hugetlbfs_fallocate, 10361da177e4SLinus Torvalds }; 10371da177e4SLinus Torvalds 103892e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 10391da177e4SLinus Torvalds .create = hugetlbfs_create, 10401da177e4SLinus Torvalds .lookup = simple_lookup, 10411da177e4SLinus Torvalds .link = simple_link, 10421da177e4SLinus Torvalds .unlink = simple_unlink, 10431da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 10441da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 10451da177e4SLinus Torvalds .rmdir = simple_rmdir, 10461da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 10471da177e4SLinus Torvalds .rename = simple_rename, 10481da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 10491da177e4SLinus Torvalds }; 10501da177e4SLinus Torvalds 105192e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 10521da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 10531da177e4SLinus Torvalds }; 10541da177e4SLinus Torvalds 1055ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 10561da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 10571da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 10582bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 10591da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 10601da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 10614a25220dSDavid Howells .show_options = hugetlbfs_show_options, 10621da177e4SLinus Torvalds }; 10631da177e4SLinus Torvalds 10644a25220dSDavid Howells enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT }; 10657ca02d0aSMike Kravetz 10667ca02d0aSMike Kravetz /* 10677ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 10687ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 10697ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 10707ca02d0aSMike Kravetz */ 10714a25220dSDavid Howells static long 10727ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 10734a25220dSDavid Howells enum hugetlbfs_size_type val_type) 10747ca02d0aSMike Kravetz { 10757ca02d0aSMike Kravetz if (val_type == NO_SIZE) 10767ca02d0aSMike Kravetz return -1; 10777ca02d0aSMike Kravetz 10787ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 10797ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 10807ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 10817ca02d0aSMike Kravetz do_div(size_opt, 100); 10827ca02d0aSMike Kravetz } 10837ca02d0aSMike Kravetz 10847ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 10857ca02d0aSMike Kravetz return size_opt; 10867ca02d0aSMike Kravetz } 10877ca02d0aSMike Kravetz 10881da177e4SLinus Torvalds static int 10891da177e4SLinus Torvalds hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) 10901da177e4SLinus Torvalds { 1091e73a75faSRandy Dunlap char *p, *rest; 1092e73a75faSRandy Dunlap substring_t args[MAX_OPT_ARGS]; 1093e73a75faSRandy Dunlap int option; 10947ca02d0aSMike Kravetz unsigned long long max_size_opt = 0, min_size_opt = 0; 10954a25220dSDavid Howells enum hugetlbfs_size_type max_val_type = NO_SIZE, min_val_type = NO_SIZE; 10961da177e4SLinus Torvalds 10971da177e4SLinus Torvalds if (!options) 10981da177e4SLinus Torvalds return 0; 10991da177e4SLinus Torvalds 1100e73a75faSRandy Dunlap while ((p = strsep(&options, ",")) != NULL) { 1101e73a75faSRandy Dunlap int token; 1102b4c07bceSLee Schermerhorn if (!*p) 1103b4c07bceSLee Schermerhorn continue; 11041da177e4SLinus Torvalds 1105e73a75faSRandy Dunlap token = match_token(p, tokens, args); 1106e73a75faSRandy Dunlap switch (token) { 1107e73a75faSRandy Dunlap case Opt_uid: 1108e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 1109e73a75faSRandy Dunlap goto bad_val; 1110a0eb3a05SEric W. Biederman pconfig->uid = make_kuid(current_user_ns(), option); 1111a0eb3a05SEric W. Biederman if (!uid_valid(pconfig->uid)) 1112a0eb3a05SEric W. Biederman goto bad_val; 1113e73a75faSRandy Dunlap break; 1114e73a75faSRandy Dunlap 1115e73a75faSRandy Dunlap case Opt_gid: 1116e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 1117e73a75faSRandy Dunlap goto bad_val; 1118a0eb3a05SEric W. Biederman pconfig->gid = make_kgid(current_user_ns(), option); 1119a0eb3a05SEric W. Biederman if (!gid_valid(pconfig->gid)) 1120a0eb3a05SEric W. Biederman goto bad_val; 1121e73a75faSRandy Dunlap break; 1122e73a75faSRandy Dunlap 1123e73a75faSRandy Dunlap case Opt_mode: 1124e73a75faSRandy Dunlap if (match_octal(&args[0], &option)) 1125e73a75faSRandy Dunlap goto bad_val; 112675897d60SKen Chen pconfig->mode = option & 01777U; 1127e73a75faSRandy Dunlap break; 1128e73a75faSRandy Dunlap 1129e73a75faSRandy Dunlap case Opt_size: { 1130e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 1131e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 1132e73a75faSRandy Dunlap goto bad_val; 11337ca02d0aSMike Kravetz max_size_opt = memparse(args[0].from, &rest); 11347ca02d0aSMike Kravetz max_val_type = SIZE_STD; 1135a137e1ccSAndi Kleen if (*rest == '%') 11367ca02d0aSMike Kravetz max_val_type = SIZE_PERCENT; 1137e73a75faSRandy Dunlap break; 1138e73a75faSRandy Dunlap } 11391da177e4SLinus Torvalds 1140e73a75faSRandy Dunlap case Opt_nr_inodes: 1141e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 1142e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 1143e73a75faSRandy Dunlap goto bad_val; 1144e73a75faSRandy Dunlap pconfig->nr_inodes = memparse(args[0].from, &rest); 1145e73a75faSRandy Dunlap break; 1146e73a75faSRandy Dunlap 1147a137e1ccSAndi Kleen case Opt_pagesize: { 1148a137e1ccSAndi Kleen unsigned long ps; 1149a137e1ccSAndi Kleen ps = memparse(args[0].from, &rest); 1150a137e1ccSAndi Kleen pconfig->hstate = size_to_hstate(ps); 1151a137e1ccSAndi Kleen if (!pconfig->hstate) { 11529b857d26SAndrew Morton pr_err("Unsupported page size %lu MB\n", 1153a137e1ccSAndi Kleen ps >> 20); 1154a137e1ccSAndi Kleen return -EINVAL; 1155a137e1ccSAndi Kleen } 1156a137e1ccSAndi Kleen break; 1157a137e1ccSAndi Kleen } 1158a137e1ccSAndi Kleen 11597ca02d0aSMike Kravetz case Opt_min_size: { 11607ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 11617ca02d0aSMike Kravetz if (!isdigit(*args[0].from)) 11627ca02d0aSMike Kravetz goto bad_val; 11637ca02d0aSMike Kravetz min_size_opt = memparse(args[0].from, &rest); 11647ca02d0aSMike Kravetz min_val_type = SIZE_STD; 11657ca02d0aSMike Kravetz if (*rest == '%') 11667ca02d0aSMike Kravetz min_val_type = SIZE_PERCENT; 11677ca02d0aSMike Kravetz break; 11687ca02d0aSMike Kravetz } 11697ca02d0aSMike Kravetz 1170e73a75faSRandy Dunlap default: 11719b857d26SAndrew Morton pr_err("Bad mount option: \"%s\"\n", p); 1172b4c07bceSLee Schermerhorn return -EINVAL; 1173e73a75faSRandy Dunlap break; 1174e73a75faSRandy Dunlap } 11751da177e4SLinus Torvalds } 1176a137e1ccSAndi Kleen 11777ca02d0aSMike Kravetz /* 11787ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 11797ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 11807ca02d0aSMike Kravetz */ 11817ca02d0aSMike Kravetz pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, 11827ca02d0aSMike Kravetz max_size_opt, max_val_type); 11837ca02d0aSMike Kravetz pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, 11847ca02d0aSMike Kravetz min_size_opt, min_val_type); 11857ca02d0aSMike Kravetz 11867ca02d0aSMike Kravetz /* 11877ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 11887ca02d0aSMike Kravetz */ 11897ca02d0aSMike Kravetz if (max_val_type > NO_SIZE && 11907ca02d0aSMike Kravetz pconfig->min_hpages > pconfig->max_hpages) { 11917ca02d0aSMike Kravetz pr_err("minimum size can not be greater than maximum size\n"); 11927ca02d0aSMike Kravetz return -EINVAL; 1193a137e1ccSAndi Kleen } 1194a137e1ccSAndi Kleen 11951da177e4SLinus Torvalds return 0; 1196e73a75faSRandy Dunlap 1197e73a75faSRandy Dunlap bad_val: 11989b857d26SAndrew Morton pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); 1199c12ddba0SAkinobu Mita return -EINVAL; 12001da177e4SLinus Torvalds } 12011da177e4SLinus Torvalds 12021da177e4SLinus Torvalds static int 12031da177e4SLinus Torvalds hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) 12041da177e4SLinus Torvalds { 12051da177e4SLinus Torvalds int ret; 12061da177e4SLinus Torvalds struct hugetlbfs_config config; 12071da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 12081da177e4SLinus Torvalds 12097ca02d0aSMike Kravetz config.max_hpages = -1; /* No limit on size by default */ 12101da177e4SLinus Torvalds config.nr_inodes = -1; /* No limit on number of inodes by default */ 121177c70de1SDavid Howells config.uid = current_fsuid(); 121277c70de1SDavid Howells config.gid = current_fsgid(); 12131da177e4SLinus Torvalds config.mode = 0755; 1214a137e1ccSAndi Kleen config.hstate = &default_hstate; 12157ca02d0aSMike Kravetz config.min_hpages = -1; /* No default minimum size */ 12161da177e4SLinus Torvalds ret = hugetlbfs_parse_options(data, &config); 12171da177e4SLinus Torvalds if (ret) 12181da177e4SLinus Torvalds return ret; 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 12211da177e4SLinus Torvalds if (!sbinfo) 12221da177e4SLinus Torvalds return -ENOMEM; 12231da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 1224a137e1ccSAndi Kleen sbinfo->hstate = config.hstate; 12251da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 12261da177e4SLinus Torvalds sbinfo->max_inodes = config.nr_inodes; 12271da177e4SLinus Torvalds sbinfo->free_inodes = config.nr_inodes; 122890481622SDavid Gibson sbinfo->spool = NULL; 12294a25220dSDavid Howells sbinfo->uid = config.uid; 12304a25220dSDavid Howells sbinfo->gid = config.gid; 12314a25220dSDavid Howells sbinfo->mode = config.mode; 12324a25220dSDavid Howells 12337ca02d0aSMike Kravetz /* 12347ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 12357ca02d0aSMike Kravetz * specified. Any needed reservations (for minimim size) are taken 12367ca02d0aSMike Kravetz * taken when the subpool is created. 12377ca02d0aSMike Kravetz */ 12387ca02d0aSMike Kravetz if (config.max_hpages != -1 || config.min_hpages != -1) { 12397ca02d0aSMike Kravetz sbinfo->spool = hugepage_new_subpool(config.hstate, 12407ca02d0aSMike Kravetz config.max_hpages, 12417ca02d0aSMike Kravetz config.min_hpages); 124290481622SDavid Gibson if (!sbinfo->spool) 124390481622SDavid Gibson goto out_free; 124490481622SDavid Gibson } 12451da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 1246a137e1ccSAndi Kleen sb->s_blocksize = huge_page_size(config.hstate); 1247a137e1ccSAndi Kleen sb->s_blocksize_bits = huge_page_shift(config.hstate); 12481da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 12491da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 12501da177e4SLinus Torvalds sb->s_time_gran = 1; 125148fde701SAl Viro sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); 125248fde701SAl Viro if (!sb->s_root) 12531da177e4SLinus Torvalds goto out_free; 12541da177e4SLinus Torvalds return 0; 12551da177e4SLinus Torvalds out_free: 125690481622SDavid Gibson kfree(sbinfo->spool); 12571da177e4SLinus Torvalds kfree(sbinfo); 12581da177e4SLinus Torvalds return -ENOMEM; 12591da177e4SLinus Torvalds } 12601da177e4SLinus Torvalds 12613c26ff6eSAl Viro static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, 12623c26ff6eSAl Viro int flags, const char *dev_name, void *data) 12631da177e4SLinus Torvalds { 12643c26ff6eSAl Viro return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); 12651da177e4SLinus Torvalds } 12661da177e4SLinus Torvalds 12671da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 12681da177e4SLinus Torvalds .name = "hugetlbfs", 12693c26ff6eSAl Viro .mount = hugetlbfs_mount, 12701da177e4SLinus Torvalds .kill_sb = kill_litter_super, 12711da177e4SLinus Torvalds }; 12721da177e4SLinus Torvalds 127342d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 12741da177e4SLinus Torvalds 1275ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 12761da177e4SLinus Torvalds { 1277a0eb3a05SEric W. Biederman kgid_t shm_group; 1278a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 1279a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 12801da177e4SLinus Torvalds } 12811da177e4SLinus Torvalds 128242d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 128342d7395fSAndi Kleen { 1284af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 128542d7395fSAndi Kleen 128642d7395fSAndi Kleen if (!h) 128742d7395fSAndi Kleen return -1; 128842d7395fSAndi Kleen return h - hstates; 128942d7395fSAndi Kleen } 129042d7395fSAndi Kleen 1291be1d2cf5SFabian Frederick static const struct dentry_operations anon_ops = { 1292118b2302SAl Viro .d_dname = simple_dname 12930df4d6e5SAl Viro }; 12940df4d6e5SAl Viro 1295af73e4d9SNaoya Horiguchi /* 1296af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 1297af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 1298af73e4d9SNaoya Horiguchi */ 1299af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 1300af73e4d9SNaoya Horiguchi vm_flags_t acctflag, struct user_struct **user, 130142d7395fSAndi Kleen int creat_flags, int page_size_log) 13021da177e4SLinus Torvalds { 130339b65252SAnatol Pomozov struct file *file = ERR_PTR(-ENOMEM); 13041da177e4SLinus Torvalds struct inode *inode; 13052c48b9c4SAl Viro struct path path; 13060df4d6e5SAl Viro struct super_block *sb; 13071da177e4SLinus Torvalds struct qstr quick_string; 130842d7395fSAndi Kleen int hstate_idx; 130942d7395fSAndi Kleen 131042d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 131142d7395fSAndi Kleen if (hstate_idx < 0) 131242d7395fSAndi Kleen return ERR_PTR(-ENODEV); 13131da177e4SLinus Torvalds 1314353d5c30SHugh Dickins *user = NULL; 131542d7395fSAndi Kleen if (!hugetlbfs_vfsmount[hstate_idx]) 13165bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 13175bc98594SAkinobu Mita 1318ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 1319353d5c30SHugh Dickins *user = current_user(); 1320353d5c30SHugh Dickins if (user_shm_lock(size, *user)) { 132121a3c273SDavid Rientjes task_lock(current); 13229b857d26SAndrew Morton pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", 132321a3c273SDavid Rientjes current->comm, current->pid); 132421a3c273SDavid Rientjes task_unlock(current); 1325353d5c30SHugh Dickins } else { 1326353d5c30SHugh Dickins *user = NULL; 13271da177e4SLinus Torvalds return ERR_PTR(-EPERM); 13282584e517SRavikiran G Thirumalai } 1329353d5c30SHugh Dickins } 13301da177e4SLinus Torvalds 13310df4d6e5SAl Viro sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb; 13329d66586fSEric W. Biederman quick_string.name = name; 13331da177e4SLinus Torvalds quick_string.len = strlen(quick_string.name); 13341da177e4SLinus Torvalds quick_string.hash = 0; 13350df4d6e5SAl Viro path.dentry = d_alloc_pseudo(sb, &quick_string); 13362c48b9c4SAl Viro if (!path.dentry) 13371da177e4SLinus Torvalds goto out_shm_unlock; 13381da177e4SLinus Torvalds 13390df4d6e5SAl Viro d_set_d_op(path.dentry, &anon_ops); 134042d7395fSAndi Kleen path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); 134139b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 13420df4d6e5SAl Viro inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0); 13431da177e4SLinus Torvalds if (!inode) 1344ce8d2cdfSDave Hansen goto out_dentry; 1345e1832f29SStephen Smalley if (creat_flags == HUGETLB_SHMFS_INODE) 1346e1832f29SStephen Smalley inode->i_flags |= S_PRIVATE; 13471da177e4SLinus Torvalds 134839b65252SAnatol Pomozov file = ERR_PTR(-ENOMEM); 1349af73e4d9SNaoya Horiguchi if (hugetlb_reserve_pages(inode, 0, 1350af73e4d9SNaoya Horiguchi size >> huge_page_shift(hstate_inode(inode)), NULL, 1351af73e4d9SNaoya Horiguchi acctflag)) 1352b45b5bd6SDavid Gibson goto out_inode; 1353b45b5bd6SDavid Gibson 13542c48b9c4SAl Viro d_instantiate(path.dentry, inode); 13551da177e4SLinus Torvalds inode->i_size = size; 13566d6b77f1SMiklos Szeredi clear_nlink(inode); 1357ce8d2cdfSDave Hansen 13582c48b9c4SAl Viro file = alloc_file(&path, FMODE_WRITE | FMODE_READ, 1359ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 136039b65252SAnatol Pomozov if (IS_ERR(file)) 1361b4d232e6SAl Viro goto out_dentry; /* inode is already attached */ 1362ce8d2cdfSDave Hansen 13631da177e4SLinus Torvalds return file; 13641da177e4SLinus Torvalds 1365b45b5bd6SDavid Gibson out_inode: 1366b45b5bd6SDavid Gibson iput(inode); 13671da177e4SLinus Torvalds out_dentry: 13682c48b9c4SAl Viro path_put(&path); 13691da177e4SLinus Torvalds out_shm_unlock: 1370353d5c30SHugh Dickins if (*user) { 1371353d5c30SHugh Dickins user_shm_unlock(size, *user); 1372353d5c30SHugh Dickins *user = NULL; 1373353d5c30SHugh Dickins } 137439b65252SAnatol Pomozov return file; 13751da177e4SLinus Torvalds } 13761da177e4SLinus Torvalds 13771da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 13781da177e4SLinus Torvalds { 137942d7395fSAndi Kleen struct hstate *h; 13801da177e4SLinus Torvalds int error; 138142d7395fSAndi Kleen int i; 13821da177e4SLinus Torvalds 1383457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 13849b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1385457c1b27SNishanth Aravamudan return -ENOTSUPP; 1386457c1b27SNishanth Aravamudan } 1387457c1b27SNishanth Aravamudan 1388d1d5e05fSHillf Danton error = -ENOMEM; 13891da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 13901da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 13915d097056SVladimir Davydov 0, SLAB_ACCOUNT, init_once); 13921da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 1393e0bf68ddSPeter Zijlstra goto out2; 13941da177e4SLinus Torvalds 13951da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 13961da177e4SLinus Torvalds if (error) 13971da177e4SLinus Torvalds goto out; 13981da177e4SLinus Torvalds 139942d7395fSAndi Kleen i = 0; 140042d7395fSAndi Kleen for_each_hstate(h) { 140142d7395fSAndi Kleen char buf[50]; 140242d7395fSAndi Kleen unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); 14031da177e4SLinus Torvalds 140442d7395fSAndi Kleen snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); 140542d7395fSAndi Kleen hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, 140642d7395fSAndi Kleen buf); 140742d7395fSAndi Kleen 140842d7395fSAndi Kleen if (IS_ERR(hugetlbfs_vfsmount[i])) { 14099b857d26SAndrew Morton pr_err("Cannot mount internal hugetlbfs for " 141042d7395fSAndi Kleen "page size %uK", ps_kb); 141142d7395fSAndi Kleen error = PTR_ERR(hugetlbfs_vfsmount[i]); 141242d7395fSAndi Kleen hugetlbfs_vfsmount[i] = NULL; 14131da177e4SLinus Torvalds } 141442d7395fSAndi Kleen i++; 141542d7395fSAndi Kleen } 141642d7395fSAndi Kleen /* Non default hstates are optional */ 141742d7395fSAndi Kleen if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) 141842d7395fSAndi Kleen return 0; 14191da177e4SLinus Torvalds 14201da177e4SLinus Torvalds out: 14211da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 1422e0bf68ddSPeter Zijlstra out2: 14231da177e4SLinus Torvalds return error; 14241da177e4SLinus Torvalds } 14253e89e1c5SPaul Gortmaker fs_initcall(init_hugetlbfs_fs) 1426