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> 141da177e4SLinus Torvalds #include <linux/sched.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 401da177e4SLinus Torvalds #include <asm/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 { 49a0eb3a05SEric W. Biederman kuid_t uid; 50a0eb3a05SEric W. Biederman kgid_t gid; 51a1d776eeSDavid Gibson umode_t mode; 527ca02d0aSMike Kravetz long max_hpages; 53a1d776eeSDavid Gibson long nr_inodes; 54a1d776eeSDavid Gibson struct hstate *hstate; 557ca02d0aSMike Kravetz long min_hpages; 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 1392b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 140dec4ad86SDavid Gibson return -EINVAL; 141dec4ad86SDavid Gibson 1421da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 1431da177e4SLinus Torvalds 1441b1dcc1bSJes Sorensen mutex_lock(&inode->i_mutex); 1451da177e4SLinus Torvalds file_accessed(file); 1461da177e4SLinus Torvalds 1471da177e4SLinus Torvalds ret = -ENOMEM; 1481da177e4SLinus Torvalds len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 1491da177e4SLinus Torvalds 150a1e78772SMel Gorman if (hugetlb_reserve_pages(inode, 151a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1525a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1535a6fe125SMel Gorman vma->vm_flags)) 154b45b5bd6SDavid Gibson goto out; 155b45b5bd6SDavid Gibson 1564c887265SAdam Litke ret = 0; 157b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 1581da177e4SLinus Torvalds inode->i_size = len; 1591da177e4SLinus Torvalds out: 1601b1dcc1bSJes Sorensen mutex_unlock(&inode->i_mutex); 1611da177e4SLinus Torvalds 1621da177e4SLinus Torvalds return ret; 1631da177e4SLinus Torvalds } 1641da177e4SLinus Torvalds 1651da177e4SLinus Torvalds /* 166508034a3SHugh Dickins * Called under down_write(mmap_sem). 1671da177e4SLinus Torvalds */ 1681da177e4SLinus Torvalds 169d2ba27e8SAdrian Bunk #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 1701da177e4SLinus Torvalds static unsigned long 1711da177e4SLinus Torvalds hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 1721da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 1731da177e4SLinus Torvalds { 1741da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1751da177e4SLinus Torvalds struct vm_area_struct *vma; 176a5516438SAndi Kleen struct hstate *h = hstate_file(file); 17708659355SMichel Lespinasse struct vm_unmapped_area_info info; 1781da177e4SLinus Torvalds 179a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 1801da177e4SLinus Torvalds return -EINVAL; 1811da177e4SLinus Torvalds if (len > TASK_SIZE) 1821da177e4SLinus Torvalds return -ENOMEM; 1831da177e4SLinus Torvalds 184036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 185a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 186036e0856SBenjamin Herrenschmidt return -EINVAL; 187036e0856SBenjamin Herrenschmidt return addr; 188036e0856SBenjamin Herrenschmidt } 189036e0856SBenjamin Herrenschmidt 1901da177e4SLinus Torvalds if (addr) { 191a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 1921da177e4SLinus Torvalds vma = find_vma(mm, addr); 1931da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 1941da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 1951da177e4SLinus Torvalds return addr; 1961da177e4SLinus Torvalds } 1971da177e4SLinus Torvalds 19808659355SMichel Lespinasse info.flags = 0; 19908659355SMichel Lespinasse info.length = len; 20008659355SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 20108659355SMichel Lespinasse info.high_limit = TASK_SIZE; 20208659355SMichel Lespinasse info.align_mask = PAGE_MASK & ~huge_page_mask(h); 20308659355SMichel Lespinasse info.align_offset = 0; 20408659355SMichel Lespinasse return vm_unmapped_area(&info); 2051da177e4SLinus Torvalds } 2061da177e4SLinus Torvalds #endif 2071da177e4SLinus Torvalds 20834d0640eSAl Viro static size_t 209e63e1e5aSBadari Pulavarty hugetlbfs_read_actor(struct page *page, unsigned long offset, 21034d0640eSAl Viro struct iov_iter *to, unsigned long size) 211e63e1e5aSBadari Pulavarty { 21234d0640eSAl Viro size_t copied = 0; 213e63e1e5aSBadari Pulavarty int i, chunksize; 214e63e1e5aSBadari Pulavarty 215e63e1e5aSBadari Pulavarty /* Find which 4k chunk and offset with in that chunk */ 216e63e1e5aSBadari Pulavarty i = offset >> PAGE_CACHE_SHIFT; 217e63e1e5aSBadari Pulavarty offset = offset & ~PAGE_CACHE_MASK; 218e63e1e5aSBadari Pulavarty 219e63e1e5aSBadari Pulavarty while (size) { 22034d0640eSAl Viro size_t n; 221e63e1e5aSBadari Pulavarty chunksize = PAGE_CACHE_SIZE; 222e63e1e5aSBadari Pulavarty if (offset) 223e63e1e5aSBadari Pulavarty chunksize -= offset; 224e63e1e5aSBadari Pulavarty if (chunksize > size) 225e63e1e5aSBadari Pulavarty chunksize = size; 22634d0640eSAl Viro n = copy_page_to_iter(&page[i], offset, chunksize, to); 22734d0640eSAl Viro copied += n; 22834d0640eSAl Viro if (n != chunksize) 22934d0640eSAl Viro return copied; 230e63e1e5aSBadari Pulavarty offset = 0; 231e63e1e5aSBadari Pulavarty size -= chunksize; 232e63e1e5aSBadari Pulavarty i++; 233e63e1e5aSBadari Pulavarty } 23434d0640eSAl Viro return copied; 235e63e1e5aSBadari Pulavarty } 236e63e1e5aSBadari Pulavarty 237e63e1e5aSBadari Pulavarty /* 238e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 239e63e1e5aSBadari Pulavarty * data. Its *very* similar to do_generic_mapping_read(), we can't use that 240e63e1e5aSBadari Pulavarty * since it has PAGE_CACHE_SIZE assumptions. 241e63e1e5aSBadari Pulavarty */ 24234d0640eSAl Viro static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) 243e63e1e5aSBadari Pulavarty { 24434d0640eSAl Viro struct file *file = iocb->ki_filp; 24534d0640eSAl Viro struct hstate *h = hstate_file(file); 24634d0640eSAl Viro struct address_space *mapping = file->f_mapping; 247e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 24834d0640eSAl Viro unsigned long index = iocb->ki_pos >> huge_page_shift(h); 24934d0640eSAl Viro unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); 250e63e1e5aSBadari Pulavarty unsigned long end_index; 251e63e1e5aSBadari Pulavarty loff_t isize; 252e63e1e5aSBadari Pulavarty ssize_t retval = 0; 253e63e1e5aSBadari Pulavarty 25434d0640eSAl Viro while (iov_iter_count(to)) { 255e63e1e5aSBadari Pulavarty struct page *page; 25634d0640eSAl Viro size_t nr, copied; 257e63e1e5aSBadari Pulavarty 258e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 259a5516438SAndi Kleen nr = huge_page_size(h); 260a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 261a05b0855SAneesh Kumar K.V if (!isize) 26234d0640eSAl Viro break; 263a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 264e63e1e5aSBadari Pulavarty if (index > end_index) 26534d0640eSAl Viro break; 26634d0640eSAl Viro if (index == end_index) { 267a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 268a05b0855SAneesh Kumar K.V if (nr <= offset) 26934d0640eSAl Viro break; 270e63e1e5aSBadari Pulavarty } 271e63e1e5aSBadari Pulavarty nr = nr - offset; 272e63e1e5aSBadari Pulavarty 273e63e1e5aSBadari Pulavarty /* Find the page */ 274a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 275e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 276e63e1e5aSBadari Pulavarty /* 277e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 278e63e1e5aSBadari Pulavarty * length of the hole or request. 279e63e1e5aSBadari Pulavarty */ 28034d0640eSAl Viro copied = iov_iter_zero(nr, to); 281e63e1e5aSBadari Pulavarty } else { 282a05b0855SAneesh Kumar K.V unlock_page(page); 283a05b0855SAneesh Kumar K.V 284e63e1e5aSBadari Pulavarty /* 285e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 286e63e1e5aSBadari Pulavarty */ 28734d0640eSAl Viro copied = hugetlbfs_read_actor(page, offset, to, nr); 288a05b0855SAneesh Kumar K.V page_cache_release(page); 289e63e1e5aSBadari Pulavarty } 29034d0640eSAl Viro offset += copied; 29134d0640eSAl Viro retval += copied; 29234d0640eSAl Viro if (copied != nr && iov_iter_count(to)) { 29334d0640eSAl Viro if (!retval) 29434d0640eSAl Viro retval = -EFAULT; 295e63e1e5aSBadari Pulavarty break; 296e63e1e5aSBadari Pulavarty } 29734d0640eSAl Viro index += offset >> huge_page_shift(h); 29834d0640eSAl Viro offset &= ~huge_page_mask(h); 29934d0640eSAl Viro } 30034d0640eSAl Viro iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; 301e63e1e5aSBadari Pulavarty return retval; 302e63e1e5aSBadari Pulavarty } 303e63e1e5aSBadari Pulavarty 304800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 305800d15a5SNick Piggin struct address_space *mapping, 306800d15a5SNick Piggin loff_t pos, unsigned len, unsigned flags, 307800d15a5SNick Piggin struct page **pagep, void **fsdata) 3081da177e4SLinus Torvalds { 3091da177e4SLinus Torvalds return -EINVAL; 3101da177e4SLinus Torvalds } 3111da177e4SLinus Torvalds 312800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 313800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 314800d15a5SNick Piggin struct page *page, void *fsdata) 3151da177e4SLinus Torvalds { 316800d15a5SNick Piggin BUG(); 3171da177e4SLinus Torvalds return -EINVAL; 3181da177e4SLinus Torvalds } 3191da177e4SLinus Torvalds 320b5cec28dSMike Kravetz static void remove_huge_page(struct page *page) 3211da177e4SLinus Torvalds { 322b9ea2515SKonstantin Khlebnikov ClearPageDirty(page); 3231da177e4SLinus Torvalds ClearPageUptodate(page); 324bd65cb86SMinchan Kim delete_from_page_cache(page); 3251da177e4SLinus Torvalds } 3261da177e4SLinus Torvalds 327b5cec28dSMike Kravetz 328b5cec28dSMike Kravetz /* 329b5cec28dSMike Kravetz * remove_inode_hugepages handles two distinct cases: truncation and hole 330b5cec28dSMike Kravetz * punch. There are subtle differences in operation for each case. 331b5cec28dSMike Kravetz 332b5cec28dSMike Kravetz * truncation is indicated by end of range being LLONG_MAX 333b5cec28dSMike Kravetz * In this case, we first scan the range and release found pages. 334b5cec28dSMike Kravetz * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv 3351817889eSMike Kravetz * maps and global counts. Page faults can not race with truncation 3361817889eSMike Kravetz * in this routine. hugetlb_no_page() prevents page faults in the 3371817889eSMike Kravetz * truncated range. It checks i_size before allocation, and again after 3381817889eSMike Kravetz * with the page table lock for the page held. The same lock must be 3391817889eSMike Kravetz * acquired to unmap a page. 340b5cec28dSMike Kravetz * hole punch is indicated if end is not LLONG_MAX 341b5cec28dSMike Kravetz * In the hole punch case we scan the range and release found pages. 342b5cec28dSMike Kravetz * Only when releasing a page is the associated region/reserv map 343b5cec28dSMike Kravetz * deleted. The region/reserv map for ranges without associated 3441817889eSMike Kravetz * pages are not modified. Page faults can race with hole punch. 3451817889eSMike Kravetz * This is indicated if we find a mapped page. 346b5cec28dSMike Kravetz * Note: If the passed end of range value is beyond the end of file, but 347b5cec28dSMike Kravetz * not LLONG_MAX this routine still performs a hole punch operation. 348b5cec28dSMike Kravetz */ 349b5cec28dSMike Kravetz static void remove_inode_hugepages(struct inode *inode, loff_t lstart, 350b5cec28dSMike Kravetz loff_t lend) 3511da177e4SLinus Torvalds { 352a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 353b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 354a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 355b5cec28dSMike Kravetz const pgoff_t end = lend >> huge_page_shift(h); 356b5cec28dSMike Kravetz struct vm_area_struct pseudo_vma; 3571da177e4SLinus Torvalds struct pagevec pvec; 3581da177e4SLinus Torvalds pgoff_t next; 359a43a8c39SChen, Kenneth W int i, freed = 0; 360b5cec28dSMike Kravetz long lookup_nr = PAGEVEC_SIZE; 361b5cec28dSMike Kravetz bool truncate_op = (lend == LLONG_MAX); 3621da177e4SLinus Torvalds 363b5cec28dSMike Kravetz memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); 364b5cec28dSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 3651da177e4SLinus Torvalds pagevec_init(&pvec, 0); 3661da177e4SLinus Torvalds next = start; 367b5cec28dSMike Kravetz while (next < end) { 368b5cec28dSMike Kravetz /* 3691817889eSMike Kravetz * Don't grab more pages than the number left in the range. 370b5cec28dSMike Kravetz */ 371b5cec28dSMike Kravetz if (end - next < lookup_nr) 372b5cec28dSMike Kravetz lookup_nr = end - next; 373b5cec28dSMike Kravetz 374b5cec28dSMike Kravetz /* 3751817889eSMike Kravetz * When no more pages are found, we are done. 376b5cec28dSMike Kravetz */ 3771817889eSMike Kravetz if (!pagevec_lookup(&pvec, mapping, next, lookup_nr)) 3781da177e4SLinus Torvalds break; 3791da177e4SLinus Torvalds 3801da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); ++i) { 3811da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 382b5cec28dSMike Kravetz u32 hash; 383b5cec28dSMike Kravetz 3841817889eSMike Kravetz /* 3851817889eSMike Kravetz * The page (index) could be beyond end. This is 3861817889eSMike Kravetz * only possible in the punch hole case as end is 3871817889eSMike Kravetz * max page offset in the truncate case. 3881817889eSMike Kravetz */ 3891817889eSMike Kravetz next = page->index; 3901817889eSMike Kravetz if (next >= end) 3911817889eSMike Kravetz break; 3921817889eSMike Kravetz 393b5cec28dSMike Kravetz hash = hugetlb_fault_mutex_hash(h, current->mm, 394b5cec28dSMike Kravetz &pseudo_vma, 395b5cec28dSMike Kravetz mapping, next, 0); 396b5cec28dSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 3971da177e4SLinus Torvalds 3981da177e4SLinus Torvalds lock_page(page); 3991817889eSMike Kravetz if (likely(!page_mapped(page))) { 400b5cec28dSMike Kravetz bool rsv_on_error = !PagePrivate(page); 401b5cec28dSMike Kravetz /* 402b5cec28dSMike Kravetz * We must free the huge page and remove 403b5cec28dSMike Kravetz * from page cache (remove_huge_page) BEFORE 404b5cec28dSMike Kravetz * removing the region/reserve map 405b5cec28dSMike Kravetz * (hugetlb_unreserve_pages). In rare out 406b5cec28dSMike Kravetz * of memory conditions, removal of the 407b5cec28dSMike Kravetz * region/reserve map could fail. Before 408b5cec28dSMike Kravetz * free'ing the page, note PagePrivate which 409b5cec28dSMike Kravetz * is used in case of error. 410b5cec28dSMike Kravetz */ 411b5cec28dSMike Kravetz remove_huge_page(page); 412b5cec28dSMike Kravetz freed++; 413b5cec28dSMike Kravetz if (!truncate_op) { 414b5cec28dSMike Kravetz if (unlikely(hugetlb_unreserve_pages( 415b5cec28dSMike Kravetz inode, next, 416b5cec28dSMike Kravetz next + 1, 1))) 417b5cec28dSMike Kravetz hugetlb_fix_reserve_counts( 418b5cec28dSMike Kravetz inode, rsv_on_error); 419b5cec28dSMike Kravetz } 4201817889eSMike Kravetz } else { 4211817889eSMike Kravetz /* 4221817889eSMike Kravetz * If page is mapped, it was faulted in after 4231817889eSMike Kravetz * being unmapped. It indicates a race between 4241817889eSMike Kravetz * hole punch and page fault. Do nothing in 4251817889eSMike Kravetz * this case. Getting here in a truncate 4261817889eSMike Kravetz * operation is a bug. 4271817889eSMike Kravetz */ 4281817889eSMike Kravetz BUG_ON(truncate_op); 429b5cec28dSMike Kravetz } 430b5cec28dSMike Kravetz 4311da177e4SLinus Torvalds unlock_page(page); 432b5cec28dSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 4331da177e4SLinus Torvalds } 4341817889eSMike Kravetz ++next; 4351da177e4SLinus Torvalds huge_pagevec_release(&pvec); 4361817889eSMike Kravetz cond_resched(); 4371da177e4SLinus Torvalds } 438b5cec28dSMike Kravetz 439b5cec28dSMike Kravetz if (truncate_op) 440b5cec28dSMike Kravetz (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); 4411da177e4SLinus Torvalds } 4421da177e4SLinus Torvalds 4432bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 4441da177e4SLinus Torvalds { 4459119a41eSJoonsoo Kim struct resv_map *resv_map; 4469119a41eSJoonsoo Kim 447b5cec28dSMike Kravetz remove_inode_hugepages(inode, 0, LLONG_MAX); 4489119a41eSJoonsoo Kim resv_map = (struct resv_map *)inode->i_mapping->private_data; 4499119a41eSJoonsoo Kim /* root inode doesn't have the resv_map, so we should check it */ 4509119a41eSJoonsoo Kim if (resv_map) 4519119a41eSJoonsoo Kim resv_map_release(&resv_map->refs); 452dbd5768fSJan Kara clear_inode(inode); 453149f4211SChristoph Hellwig } 454149f4211SChristoph Hellwig 4551da177e4SLinus Torvalds static inline void 4561bfad99aSMike Kravetz hugetlb_vmdelete_list(struct rb_root *root, pgoff_t start, pgoff_t end) 4571da177e4SLinus Torvalds { 4581da177e4SLinus Torvalds struct vm_area_struct *vma; 4591da177e4SLinus Torvalds 4601bfad99aSMike Kravetz /* 4611bfad99aSMike Kravetz * end == 0 indicates that the entire range after 4621bfad99aSMike Kravetz * start should be unmapped. 4631bfad99aSMike Kravetz */ 4641bfad99aSMike Kravetz vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { 4651da177e4SLinus Torvalds unsigned long v_offset; 4661da177e4SLinus Torvalds 4671da177e4SLinus Torvalds /* 468856fc295SHugh Dickins * Can the expression below overflow on 32-bit arches? 4696b2dbba8SMichel Lespinasse * No, because the interval tree returns us only those vmas 470856fc295SHugh Dickins * which overlap the truncated area starting at pgoff, 471856fc295SHugh Dickins * and no vma on a 32-bit arch can span beyond the 4GB. 4721da177e4SLinus Torvalds */ 4731bfad99aSMike Kravetz if (vma->vm_pgoff < start) 4741bfad99aSMike Kravetz v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; 475856fc295SHugh Dickins else 4761da177e4SLinus Torvalds v_offset = 0; 4771da177e4SLinus Torvalds 4781bfad99aSMike Kravetz if (end) { 4791bfad99aSMike Kravetz end = ((end - start) << PAGE_SHIFT) + 4801bfad99aSMike Kravetz vma->vm_start + v_offset; 4811bfad99aSMike Kravetz if (end > vma->vm_end) 4821bfad99aSMike Kravetz end = vma->vm_end; 4831bfad99aSMike Kravetz } else 4841bfad99aSMike Kravetz end = vma->vm_end; 4851bfad99aSMike Kravetz 4861bfad99aSMike Kravetz unmap_hugepage_range(vma, vma->vm_start + v_offset, end, NULL); 4871da177e4SLinus Torvalds } 4881da177e4SLinus Torvalds } 4891da177e4SLinus Torvalds 4901da177e4SLinus Torvalds static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) 4911da177e4SLinus Torvalds { 492856fc295SHugh Dickins pgoff_t pgoff; 4931da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 494a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 4951da177e4SLinus Torvalds 496a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 497856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 4981da177e4SLinus Torvalds 4997aa91e10SKen Chen i_size_write(inode, offset); 50083cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 5016b2dbba8SMichel Lespinasse if (!RB_EMPTY_ROOT(&mapping->i_mmap)) 5021bfad99aSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); 50383cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 504b5cec28dSMike Kravetz remove_inode_hugepages(inode, offset, LLONG_MAX); 5051da177e4SLinus Torvalds return 0; 5061da177e4SLinus Torvalds } 5071da177e4SLinus Torvalds 50870c3547eSMike Kravetz static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) 50970c3547eSMike Kravetz { 51070c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 51170c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 51270c3547eSMike Kravetz loff_t hole_start, hole_end; 51370c3547eSMike Kravetz 51470c3547eSMike Kravetz /* 51570c3547eSMike Kravetz * For hole punch round up the beginning offset of the hole and 51670c3547eSMike Kravetz * round down the end. 51770c3547eSMike Kravetz */ 51870c3547eSMike Kravetz hole_start = round_up(offset, hpage_size); 51970c3547eSMike Kravetz hole_end = round_down(offset + len, hpage_size); 52070c3547eSMike Kravetz 52170c3547eSMike Kravetz if (hole_end > hole_start) { 52270c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 52370c3547eSMike Kravetz 52470c3547eSMike Kravetz mutex_lock(&inode->i_mutex); 52570c3547eSMike Kravetz i_mmap_lock_write(mapping); 52670c3547eSMike Kravetz if (!RB_EMPTY_ROOT(&mapping->i_mmap)) 52770c3547eSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 52870c3547eSMike Kravetz hole_start >> PAGE_SHIFT, 52970c3547eSMike Kravetz hole_end >> PAGE_SHIFT); 53070c3547eSMike Kravetz i_mmap_unlock_write(mapping); 53170c3547eSMike Kravetz remove_inode_hugepages(inode, hole_start, hole_end); 53270c3547eSMike Kravetz mutex_unlock(&inode->i_mutex); 53370c3547eSMike Kravetz } 53470c3547eSMike Kravetz 53570c3547eSMike Kravetz return 0; 53670c3547eSMike Kravetz } 53770c3547eSMike Kravetz 53870c3547eSMike Kravetz static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, 53970c3547eSMike Kravetz loff_t len) 54070c3547eSMike Kravetz { 54170c3547eSMike Kravetz struct inode *inode = file_inode(file); 54270c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 54370c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 54470c3547eSMike Kravetz struct vm_area_struct pseudo_vma; 54570c3547eSMike Kravetz struct mm_struct *mm = current->mm; 54670c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 54770c3547eSMike Kravetz unsigned long hpage_shift = huge_page_shift(h); 54870c3547eSMike Kravetz pgoff_t start, index, end; 54970c3547eSMike Kravetz int error; 55070c3547eSMike Kravetz u32 hash; 55170c3547eSMike Kravetz 55270c3547eSMike Kravetz if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 55370c3547eSMike Kravetz return -EOPNOTSUPP; 55470c3547eSMike Kravetz 55570c3547eSMike Kravetz if (mode & FALLOC_FL_PUNCH_HOLE) 55670c3547eSMike Kravetz return hugetlbfs_punch_hole(inode, offset, len); 55770c3547eSMike Kravetz 55870c3547eSMike Kravetz /* 55970c3547eSMike Kravetz * Default preallocate case. 56070c3547eSMike Kravetz * For this range, start is rounded down and end is rounded up 56170c3547eSMike Kravetz * as well as being converted to page offsets. 56270c3547eSMike Kravetz */ 56370c3547eSMike Kravetz start = offset >> hpage_shift; 56470c3547eSMike Kravetz end = (offset + len + hpage_size - 1) >> hpage_shift; 56570c3547eSMike Kravetz 56670c3547eSMike Kravetz mutex_lock(&inode->i_mutex); 56770c3547eSMike Kravetz 56870c3547eSMike Kravetz /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ 56970c3547eSMike Kravetz error = inode_newsize_ok(inode, offset + len); 57070c3547eSMike Kravetz if (error) 57170c3547eSMike Kravetz goto out; 57270c3547eSMike Kravetz 57370c3547eSMike Kravetz /* 57470c3547eSMike Kravetz * Initialize a pseudo vma as this is required by the huge page 57570c3547eSMike Kravetz * allocation routines. If NUMA is configured, use page index 57670c3547eSMike Kravetz * as input to create an allocation policy. 57770c3547eSMike Kravetz */ 57870c3547eSMike Kravetz memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); 57970c3547eSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 58070c3547eSMike Kravetz pseudo_vma.vm_file = file; 58170c3547eSMike Kravetz 58270c3547eSMike Kravetz for (index = start; index < end; index++) { 58370c3547eSMike Kravetz /* 58470c3547eSMike Kravetz * This is supposed to be the vaddr where the page is being 58570c3547eSMike Kravetz * faulted in, but we have no vaddr here. 58670c3547eSMike Kravetz */ 58770c3547eSMike Kravetz struct page *page; 58870c3547eSMike Kravetz unsigned long addr; 58970c3547eSMike Kravetz int avoid_reserve = 0; 59070c3547eSMike Kravetz 59170c3547eSMike Kravetz cond_resched(); 59270c3547eSMike Kravetz 59370c3547eSMike Kravetz /* 59470c3547eSMike Kravetz * fallocate(2) manpage permits EINTR; we may have been 59570c3547eSMike Kravetz * interrupted because we are using up too much memory. 59670c3547eSMike Kravetz */ 59770c3547eSMike Kravetz if (signal_pending(current)) { 59870c3547eSMike Kravetz error = -EINTR; 59970c3547eSMike Kravetz break; 60070c3547eSMike Kravetz } 60170c3547eSMike Kravetz 60270c3547eSMike Kravetz /* Set numa allocation policy based on index */ 60370c3547eSMike Kravetz hugetlb_set_vma_policy(&pseudo_vma, inode, index); 60470c3547eSMike Kravetz 60570c3547eSMike Kravetz /* addr is the offset within the file (zero based) */ 60670c3547eSMike Kravetz addr = index * hpage_size; 60770c3547eSMike Kravetz 60870c3547eSMike Kravetz /* mutex taken here, fault path and hole punch */ 60970c3547eSMike Kravetz hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping, 61070c3547eSMike Kravetz index, addr); 61170c3547eSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 61270c3547eSMike Kravetz 61370c3547eSMike Kravetz /* See if already present in mapping to avoid alloc/free */ 61470c3547eSMike Kravetz page = find_get_page(mapping, index); 61570c3547eSMike Kravetz if (page) { 61670c3547eSMike Kravetz put_page(page); 61770c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 61870c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 61970c3547eSMike Kravetz continue; 62070c3547eSMike Kravetz } 62170c3547eSMike Kravetz 62270c3547eSMike Kravetz /* Allocate page and add to page cache */ 62370c3547eSMike Kravetz page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); 62470c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 62570c3547eSMike Kravetz if (IS_ERR(page)) { 62670c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 62770c3547eSMike Kravetz error = PTR_ERR(page); 62870c3547eSMike Kravetz goto out; 62970c3547eSMike Kravetz } 63070c3547eSMike Kravetz clear_huge_page(page, addr, pages_per_huge_page(h)); 63170c3547eSMike Kravetz __SetPageUptodate(page); 63270c3547eSMike Kravetz error = huge_add_to_page_cache(page, mapping, index); 63370c3547eSMike Kravetz if (unlikely(error)) { 63470c3547eSMike Kravetz put_page(page); 63570c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 63670c3547eSMike Kravetz goto out; 63770c3547eSMike Kravetz } 63870c3547eSMike Kravetz 63970c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 64070c3547eSMike Kravetz 64170c3547eSMike Kravetz /* 64270c3547eSMike Kravetz * page_put due to reference from alloc_huge_page() 64370c3547eSMike Kravetz * unlock_page because locked by add_to_page_cache() 64470c3547eSMike Kravetz */ 64570c3547eSMike Kravetz put_page(page); 64670c3547eSMike Kravetz unlock_page(page); 64770c3547eSMike Kravetz } 64870c3547eSMike Kravetz 64970c3547eSMike Kravetz if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) 65070c3547eSMike Kravetz i_size_write(inode, offset + len); 65170c3547eSMike Kravetz inode->i_ctime = CURRENT_TIME; 65270c3547eSMike Kravetz out: 65370c3547eSMike Kravetz mutex_unlock(&inode->i_mutex); 65470c3547eSMike Kravetz return error; 65570c3547eSMike Kravetz } 65670c3547eSMike Kravetz 6571da177e4SLinus Torvalds static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) 6581da177e4SLinus Torvalds { 6592b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 660a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 6611da177e4SLinus Torvalds int error; 6621da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 6631da177e4SLinus Torvalds 6641da177e4SLinus Torvalds BUG_ON(!inode); 6651da177e4SLinus Torvalds 6661da177e4SLinus Torvalds error = inode_change_ok(inode, attr); 6671da177e4SLinus Torvalds if (error) 6681025774cSChristoph Hellwig return error; 6691da177e4SLinus Torvalds 6701da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 6711da177e4SLinus Torvalds error = -EINVAL; 6721025774cSChristoph Hellwig if (attr->ia_size & ~huge_page_mask(h)) 6731025774cSChristoph Hellwig return -EINVAL; 6741da177e4SLinus Torvalds error = hugetlb_vmtruncate(inode, attr->ia_size); 6751da177e4SLinus Torvalds if (error) 6761da177e4SLinus Torvalds return error; 6771da177e4SLinus Torvalds } 6781da177e4SLinus Torvalds 6791025774cSChristoph Hellwig setattr_copy(inode, attr); 6801025774cSChristoph Hellwig mark_inode_dirty(inode); 6811025774cSChristoph Hellwig return 0; 6821025774cSChristoph Hellwig } 6831025774cSChristoph Hellwig 6847d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 6857d54fa64SAl Viro struct hugetlbfs_config *config) 6861da177e4SLinus Torvalds { 6871da177e4SLinus Torvalds struct inode *inode; 6881da177e4SLinus Torvalds 6891da177e4SLinus Torvalds inode = new_inode(sb); 6901da177e4SLinus Torvalds if (inode) { 6911da177e4SLinus Torvalds struct hugetlbfs_inode_info *info; 69285fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 6937d54fa64SAl Viro inode->i_mode = S_IFDIR | config->mode; 6947d54fa64SAl Viro inode->i_uid = config->uid; 6957d54fa64SAl Viro inode->i_gid = config->gid; 6967d54fa64SAl Viro inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 6977d54fa64SAl Viro info = HUGETLBFS_I(inode); 6987d54fa64SAl Viro mpol_shared_policy_init(&info->policy, NULL); 6997d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 7007d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 7017d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 7027d54fa64SAl Viro inc_nlink(inode); 70365ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 7047d54fa64SAl Viro } 7057d54fa64SAl Viro return inode; 7067d54fa64SAl Viro } 7077d54fa64SAl Viro 708b610ded7SMichal Hocko /* 709c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 710b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 71188f306b6SKirill A. Shutemov * annotation because huge_pmd_share() does an allocation under hugetlb's 712c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 713b610ded7SMichal Hocko */ 714c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 715b610ded7SMichal Hocko 7167d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 7177d54fa64SAl Viro struct inode *dir, 71818df2252SAl Viro umode_t mode, dev_t dev) 7197d54fa64SAl Viro { 7207d54fa64SAl Viro struct inode *inode; 7219119a41eSJoonsoo Kim struct resv_map *resv_map; 7229119a41eSJoonsoo Kim 7239119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 7249119a41eSJoonsoo Kim if (!resv_map) 7259119a41eSJoonsoo Kim return NULL; 7267d54fa64SAl Viro 7277d54fa64SAl Viro inode = new_inode(sb); 7287d54fa64SAl Viro if (inode) { 7297d54fa64SAl Viro struct hugetlbfs_inode_info *info; 7307d54fa64SAl Viro inode->i_ino = get_next_ino(); 7317d54fa64SAl Viro inode_init_owner(inode, dir, mode); 732c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 733c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 7341da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 7351da177e4SLinus Torvalds inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 7369119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 7371da177e4SLinus Torvalds info = HUGETLBFS_I(inode); 7386bfde05bSEric B Munson /* 7396bfde05bSEric B Munson * The policy is initialized here even if we are creating a 7406bfde05bSEric B Munson * private inode because initialization simply creates an 7414a8c7bb5SNathan Zimmer * an empty rb tree and calls rwlock_init(), later when we 7426bfde05bSEric B Munson * call mpol_free_shared_policy() it will just return because 7436bfde05bSEric B Munson * the rb tree will still be empty. 7446bfde05bSEric B Munson */ 74571fe804bSLee Schermerhorn mpol_shared_policy_init(&info->policy, NULL); 7461da177e4SLinus Torvalds switch (mode & S_IFMT) { 7471da177e4SLinus Torvalds default: 7481da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 7491da177e4SLinus Torvalds break; 7501da177e4SLinus Torvalds case S_IFREG: 7511da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 7521da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 7531da177e4SLinus Torvalds break; 7541da177e4SLinus Torvalds case S_IFDIR: 7551da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 7561da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 7571da177e4SLinus Torvalds 7581da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 759d8c76e6fSDave Hansen inc_nlink(inode); 7601da177e4SLinus Torvalds break; 7611da177e4SLinus Torvalds case S_IFLNK: 7621da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 76321fc61c7SAl Viro inode_nohighmem(inode); 7641da177e4SLinus Torvalds break; 7651da177e4SLinus Torvalds } 766e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 7679119a41eSJoonsoo Kim } else 7689119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 7699119a41eSJoonsoo Kim 7701da177e4SLinus Torvalds return inode; 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds 7731da177e4SLinus Torvalds /* 7741da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 7751da177e4SLinus Torvalds */ 7761da177e4SLinus Torvalds static int hugetlbfs_mknod(struct inode *dir, 7771a67aafbSAl Viro struct dentry *dentry, umode_t mode, dev_t dev) 7781da177e4SLinus Torvalds { 7791da177e4SLinus Torvalds struct inode *inode; 7801da177e4SLinus Torvalds int error = -ENOSPC; 7811da177e4SLinus Torvalds 7827d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 7831da177e4SLinus Torvalds if (inode) { 7841da177e4SLinus Torvalds dir->i_ctime = dir->i_mtime = CURRENT_TIME; 7851da177e4SLinus Torvalds d_instantiate(dentry, inode); 7861da177e4SLinus Torvalds dget(dentry); /* Extra count - pin the dentry in core */ 7871da177e4SLinus Torvalds error = 0; 7881da177e4SLinus Torvalds } 7891da177e4SLinus Torvalds return error; 7901da177e4SLinus Torvalds } 7911da177e4SLinus Torvalds 79218bb1db3SAl Viro static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 7931da177e4SLinus Torvalds { 7941da177e4SLinus Torvalds int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); 7951da177e4SLinus Torvalds if (!retval) 796d8c76e6fSDave Hansen inc_nlink(dir); 7971da177e4SLinus Torvalds return retval; 7981da177e4SLinus Torvalds } 7991da177e4SLinus Torvalds 800ebfc3b49SAl Viro static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) 8011da177e4SLinus Torvalds { 8021da177e4SLinus Torvalds return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); 8031da177e4SLinus Torvalds } 8041da177e4SLinus Torvalds 8051da177e4SLinus Torvalds static int hugetlbfs_symlink(struct inode *dir, 8061da177e4SLinus Torvalds struct dentry *dentry, const char *symname) 8071da177e4SLinus Torvalds { 8081da177e4SLinus Torvalds struct inode *inode; 8091da177e4SLinus Torvalds int error = -ENOSPC; 8101da177e4SLinus Torvalds 8117d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 8121da177e4SLinus Torvalds if (inode) { 8131da177e4SLinus Torvalds int l = strlen(symname)+1; 8141da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 8151da177e4SLinus Torvalds if (!error) { 8161da177e4SLinus Torvalds d_instantiate(dentry, inode); 8171da177e4SLinus Torvalds dget(dentry); 8181da177e4SLinus Torvalds } else 8191da177e4SLinus Torvalds iput(inode); 8201da177e4SLinus Torvalds } 8211da177e4SLinus Torvalds dir->i_ctime = dir->i_mtime = CURRENT_TIME; 8221da177e4SLinus Torvalds 8231da177e4SLinus Torvalds return error; 8241da177e4SLinus Torvalds } 8251da177e4SLinus Torvalds 8261da177e4SLinus Torvalds /* 8276649a386SKen Chen * mark the head page dirty 8281da177e4SLinus Torvalds */ 8291da177e4SLinus Torvalds static int hugetlbfs_set_page_dirty(struct page *page) 8301da177e4SLinus Torvalds { 831d85f3385SChristoph Lameter struct page *head = compound_head(page); 8326649a386SKen Chen 8336649a386SKen Chen SetPageDirty(head); 8341da177e4SLinus Torvalds return 0; 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 837290408d4SNaoya Horiguchi static int hugetlbfs_migrate_page(struct address_space *mapping, 838b969c4abSMel Gorman struct page *newpage, struct page *page, 839a6bc32b8SMel Gorman enum migrate_mode mode) 840290408d4SNaoya Horiguchi { 841290408d4SNaoya Horiguchi int rc; 842290408d4SNaoya Horiguchi 843290408d4SNaoya Horiguchi rc = migrate_huge_page_move_mapping(mapping, newpage, page); 84478bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 845290408d4SNaoya Horiguchi return rc; 846290408d4SNaoya Horiguchi migrate_page_copy(newpage, page); 847290408d4SNaoya Horiguchi 84878bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 849290408d4SNaoya Horiguchi } 850290408d4SNaoya Horiguchi 851726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 8521da177e4SLinus Torvalds { 853726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 8542b0143b5SDavid Howells struct hstate *h = hstate_inode(d_inode(dentry)); 8551da177e4SLinus Torvalds 8561da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 857a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 8581da177e4SLinus Torvalds if (sbinfo) { 8591da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 86074a8a65cSDavid Gibson /* If no limits set, just report 0 for max/free/used 86174a8a65cSDavid Gibson * blocks, like simple_statfs() */ 86290481622SDavid Gibson if (sbinfo->spool) { 86390481622SDavid Gibson long free_pages; 86490481622SDavid Gibson 86590481622SDavid Gibson spin_lock(&sbinfo->spool->lock); 86690481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 86790481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 86890481622SDavid Gibson - sbinfo->spool->used_hpages; 86990481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 87090481622SDavid Gibson spin_unlock(&sbinfo->spool->lock); 8711da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 8721da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 87374a8a65cSDavid Gibson } 8741da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 8751da177e4SLinus Torvalds } 8761da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 8771da177e4SLinus Torvalds return 0; 8781da177e4SLinus Torvalds } 8791da177e4SLinus Torvalds 8801da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 8811da177e4SLinus Torvalds { 8821da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 8831da177e4SLinus Torvalds 8841da177e4SLinus Torvalds if (sbi) { 8851da177e4SLinus Torvalds sb->s_fs_info = NULL; 88690481622SDavid Gibson 88790481622SDavid Gibson if (sbi->spool) 88890481622SDavid Gibson hugepage_put_subpool(sbi->spool); 88990481622SDavid Gibson 8901da177e4SLinus Torvalds kfree(sbi); 8911da177e4SLinus Torvalds } 8921da177e4SLinus Torvalds } 8931da177e4SLinus Torvalds 89496527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 89596527980SChristoph Hellwig { 89696527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 89796527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 89896527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 89996527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 90096527980SChristoph Hellwig return 0; 90196527980SChristoph Hellwig } 90296527980SChristoph Hellwig sbinfo->free_inodes--; 90396527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 90496527980SChristoph Hellwig } 90596527980SChristoph Hellwig 90696527980SChristoph Hellwig return 1; 90796527980SChristoph Hellwig } 90896527980SChristoph Hellwig 90996527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 91096527980SChristoph Hellwig { 91196527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 91296527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 91396527980SChristoph Hellwig sbinfo->free_inodes++; 91496527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 91596527980SChristoph Hellwig } 91696527980SChristoph Hellwig } 91796527980SChristoph Hellwig 91896527980SChristoph Hellwig 919e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 9201da177e4SLinus Torvalds 9211da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 9221da177e4SLinus Torvalds { 92396527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 9241da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 9251da177e4SLinus Torvalds 92696527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 9271da177e4SLinus Torvalds return NULL; 928e94b1766SChristoph Lameter p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); 92996527980SChristoph Hellwig if (unlikely(!p)) { 93096527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 93196527980SChristoph Hellwig return NULL; 9321da177e4SLinus Torvalds } 93396527980SChristoph Hellwig return &p->vfs_inode; 9341da177e4SLinus Torvalds } 9351da177e4SLinus Torvalds 936fa0d7e3dSNick Piggin static void hugetlbfs_i_callback(struct rcu_head *head) 937fa0d7e3dSNick Piggin { 938fa0d7e3dSNick Piggin struct inode *inode = container_of(head, struct inode, i_rcu); 939fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 940fa0d7e3dSNick Piggin } 941fa0d7e3dSNick Piggin 9421da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 9431da177e4SLinus Torvalds { 94496527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 9451da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 946fa0d7e3dSNick Piggin call_rcu(&inode->i_rcu, hugetlbfs_i_callback); 9471da177e4SLinus Torvalds } 9481da177e4SLinus Torvalds 949f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 950800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 951800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 9521da177e4SLinus Torvalds .set_page_dirty = hugetlbfs_set_page_dirty, 953290408d4SNaoya Horiguchi .migratepage = hugetlbfs_migrate_page, 9541da177e4SLinus Torvalds }; 9551da177e4SLinus Torvalds 95696527980SChristoph Hellwig 95751cc5068SAlexey Dobriyan static void init_once(void *foo) 95896527980SChristoph Hellwig { 95996527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 96096527980SChristoph Hellwig 96196527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 96296527980SChristoph Hellwig } 96396527980SChristoph Hellwig 9644b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 96534d0640eSAl Viro .read_iter = hugetlbfs_read_iter, 9661da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 9671b061d92SChristoph Hellwig .fsync = noop_fsync, 9681da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 9696038f373SArnd Bergmann .llseek = default_llseek, 97070c3547eSMike Kravetz .fallocate = hugetlbfs_fallocate, 9711da177e4SLinus Torvalds }; 9721da177e4SLinus Torvalds 97392e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 9741da177e4SLinus Torvalds .create = hugetlbfs_create, 9751da177e4SLinus Torvalds .lookup = simple_lookup, 9761da177e4SLinus Torvalds .link = simple_link, 9771da177e4SLinus Torvalds .unlink = simple_unlink, 9781da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 9791da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 9801da177e4SLinus Torvalds .rmdir = simple_rmdir, 9811da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 9821da177e4SLinus Torvalds .rename = simple_rename, 9831da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 9841da177e4SLinus Torvalds }; 9851da177e4SLinus Torvalds 98692e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 9871da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 9881da177e4SLinus Torvalds }; 9891da177e4SLinus Torvalds 990ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 9911da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 9921da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 9932bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 9941da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 9951da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 99610f19a86SMiklos Szeredi .show_options = generic_show_options, 9971da177e4SLinus Torvalds }; 9981da177e4SLinus Torvalds 9997ca02d0aSMike Kravetz enum { NO_SIZE, SIZE_STD, SIZE_PERCENT }; 10007ca02d0aSMike Kravetz 10017ca02d0aSMike Kravetz /* 10027ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 10037ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 10047ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 10057ca02d0aSMike Kravetz */ 10067ca02d0aSMike Kravetz static long long 10077ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 10087ca02d0aSMike Kravetz int val_type) 10097ca02d0aSMike Kravetz { 10107ca02d0aSMike Kravetz if (val_type == NO_SIZE) 10117ca02d0aSMike Kravetz return -1; 10127ca02d0aSMike Kravetz 10137ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 10147ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 10157ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 10167ca02d0aSMike Kravetz do_div(size_opt, 100); 10177ca02d0aSMike Kravetz } 10187ca02d0aSMike Kravetz 10197ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 10207ca02d0aSMike Kravetz return size_opt; 10217ca02d0aSMike Kravetz } 10227ca02d0aSMike Kravetz 10231da177e4SLinus Torvalds static int 10241da177e4SLinus Torvalds hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) 10251da177e4SLinus Torvalds { 1026e73a75faSRandy Dunlap char *p, *rest; 1027e73a75faSRandy Dunlap substring_t args[MAX_OPT_ARGS]; 1028e73a75faSRandy Dunlap int option; 10297ca02d0aSMike Kravetz unsigned long long max_size_opt = 0, min_size_opt = 0; 10307ca02d0aSMike Kravetz int max_val_type = NO_SIZE, min_val_type = NO_SIZE; 10311da177e4SLinus Torvalds 10321da177e4SLinus Torvalds if (!options) 10331da177e4SLinus Torvalds return 0; 10341da177e4SLinus Torvalds 1035e73a75faSRandy Dunlap while ((p = strsep(&options, ",")) != NULL) { 1036e73a75faSRandy Dunlap int token; 1037b4c07bceSLee Schermerhorn if (!*p) 1038b4c07bceSLee Schermerhorn continue; 10391da177e4SLinus Torvalds 1040e73a75faSRandy Dunlap token = match_token(p, tokens, args); 1041e73a75faSRandy Dunlap switch (token) { 1042e73a75faSRandy Dunlap case Opt_uid: 1043e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 1044e73a75faSRandy Dunlap goto bad_val; 1045a0eb3a05SEric W. Biederman pconfig->uid = make_kuid(current_user_ns(), option); 1046a0eb3a05SEric W. Biederman if (!uid_valid(pconfig->uid)) 1047a0eb3a05SEric W. Biederman goto bad_val; 1048e73a75faSRandy Dunlap break; 1049e73a75faSRandy Dunlap 1050e73a75faSRandy Dunlap case Opt_gid: 1051e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 1052e73a75faSRandy Dunlap goto bad_val; 1053a0eb3a05SEric W. Biederman pconfig->gid = make_kgid(current_user_ns(), option); 1054a0eb3a05SEric W. Biederman if (!gid_valid(pconfig->gid)) 1055a0eb3a05SEric W. Biederman goto bad_val; 1056e73a75faSRandy Dunlap break; 1057e73a75faSRandy Dunlap 1058e73a75faSRandy Dunlap case Opt_mode: 1059e73a75faSRandy Dunlap if (match_octal(&args[0], &option)) 1060e73a75faSRandy Dunlap goto bad_val; 106175897d60SKen Chen pconfig->mode = option & 01777U; 1062e73a75faSRandy Dunlap break; 1063e73a75faSRandy Dunlap 1064e73a75faSRandy Dunlap case Opt_size: { 1065e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 1066e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 1067e73a75faSRandy Dunlap goto bad_val; 10687ca02d0aSMike Kravetz max_size_opt = memparse(args[0].from, &rest); 10697ca02d0aSMike Kravetz max_val_type = SIZE_STD; 1070a137e1ccSAndi Kleen if (*rest == '%') 10717ca02d0aSMike Kravetz max_val_type = SIZE_PERCENT; 1072e73a75faSRandy Dunlap break; 1073e73a75faSRandy Dunlap } 10741da177e4SLinus Torvalds 1075e73a75faSRandy Dunlap case Opt_nr_inodes: 1076e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 1077e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 1078e73a75faSRandy Dunlap goto bad_val; 1079e73a75faSRandy Dunlap pconfig->nr_inodes = memparse(args[0].from, &rest); 1080e73a75faSRandy Dunlap break; 1081e73a75faSRandy Dunlap 1082a137e1ccSAndi Kleen case Opt_pagesize: { 1083a137e1ccSAndi Kleen unsigned long ps; 1084a137e1ccSAndi Kleen ps = memparse(args[0].from, &rest); 1085a137e1ccSAndi Kleen pconfig->hstate = size_to_hstate(ps); 1086a137e1ccSAndi Kleen if (!pconfig->hstate) { 10879b857d26SAndrew Morton pr_err("Unsupported page size %lu MB\n", 1088a137e1ccSAndi Kleen ps >> 20); 1089a137e1ccSAndi Kleen return -EINVAL; 1090a137e1ccSAndi Kleen } 1091a137e1ccSAndi Kleen break; 1092a137e1ccSAndi Kleen } 1093a137e1ccSAndi Kleen 10947ca02d0aSMike Kravetz case Opt_min_size: { 10957ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 10967ca02d0aSMike Kravetz if (!isdigit(*args[0].from)) 10977ca02d0aSMike Kravetz goto bad_val; 10987ca02d0aSMike Kravetz min_size_opt = memparse(args[0].from, &rest); 10997ca02d0aSMike Kravetz min_val_type = SIZE_STD; 11007ca02d0aSMike Kravetz if (*rest == '%') 11017ca02d0aSMike Kravetz min_val_type = SIZE_PERCENT; 11027ca02d0aSMike Kravetz break; 11037ca02d0aSMike Kravetz } 11047ca02d0aSMike Kravetz 1105e73a75faSRandy Dunlap default: 11069b857d26SAndrew Morton pr_err("Bad mount option: \"%s\"\n", p); 1107b4c07bceSLee Schermerhorn return -EINVAL; 1108e73a75faSRandy Dunlap break; 1109e73a75faSRandy Dunlap } 11101da177e4SLinus Torvalds } 1111a137e1ccSAndi Kleen 11127ca02d0aSMike Kravetz /* 11137ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 11147ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 11157ca02d0aSMike Kravetz */ 11167ca02d0aSMike Kravetz pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, 11177ca02d0aSMike Kravetz max_size_opt, max_val_type); 11187ca02d0aSMike Kravetz pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, 11197ca02d0aSMike Kravetz min_size_opt, min_val_type); 11207ca02d0aSMike Kravetz 11217ca02d0aSMike Kravetz /* 11227ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 11237ca02d0aSMike Kravetz */ 11247ca02d0aSMike Kravetz if (max_val_type > NO_SIZE && 11257ca02d0aSMike Kravetz pconfig->min_hpages > pconfig->max_hpages) { 11267ca02d0aSMike Kravetz pr_err("minimum size can not be greater than maximum size\n"); 11277ca02d0aSMike Kravetz return -EINVAL; 1128a137e1ccSAndi Kleen } 1129a137e1ccSAndi Kleen 11301da177e4SLinus Torvalds return 0; 1131e73a75faSRandy Dunlap 1132e73a75faSRandy Dunlap bad_val: 11339b857d26SAndrew Morton pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); 1134c12ddba0SAkinobu Mita return -EINVAL; 11351da177e4SLinus Torvalds } 11361da177e4SLinus Torvalds 11371da177e4SLinus Torvalds static int 11381da177e4SLinus Torvalds hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) 11391da177e4SLinus Torvalds { 11401da177e4SLinus Torvalds int ret; 11411da177e4SLinus Torvalds struct hugetlbfs_config config; 11421da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 11431da177e4SLinus Torvalds 114410f19a86SMiklos Szeredi save_mount_options(sb, data); 114510f19a86SMiklos Szeredi 11467ca02d0aSMike Kravetz config.max_hpages = -1; /* No limit on size by default */ 11471da177e4SLinus Torvalds config.nr_inodes = -1; /* No limit on number of inodes by default */ 114877c70de1SDavid Howells config.uid = current_fsuid(); 114977c70de1SDavid Howells config.gid = current_fsgid(); 11501da177e4SLinus Torvalds config.mode = 0755; 1151a137e1ccSAndi Kleen config.hstate = &default_hstate; 11527ca02d0aSMike Kravetz config.min_hpages = -1; /* No default minimum size */ 11531da177e4SLinus Torvalds ret = hugetlbfs_parse_options(data, &config); 11541da177e4SLinus Torvalds if (ret) 11551da177e4SLinus Torvalds return ret; 11561da177e4SLinus Torvalds 11571da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 11581da177e4SLinus Torvalds if (!sbinfo) 11591da177e4SLinus Torvalds return -ENOMEM; 11601da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 1161a137e1ccSAndi Kleen sbinfo->hstate = config.hstate; 11621da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 11631da177e4SLinus Torvalds sbinfo->max_inodes = config.nr_inodes; 11641da177e4SLinus Torvalds sbinfo->free_inodes = config.nr_inodes; 116590481622SDavid Gibson sbinfo->spool = NULL; 11667ca02d0aSMike Kravetz /* 11677ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 11687ca02d0aSMike Kravetz * specified. Any needed reservations (for minimim size) are taken 11697ca02d0aSMike Kravetz * taken when the subpool is created. 11707ca02d0aSMike Kravetz */ 11717ca02d0aSMike Kravetz if (config.max_hpages != -1 || config.min_hpages != -1) { 11727ca02d0aSMike Kravetz sbinfo->spool = hugepage_new_subpool(config.hstate, 11737ca02d0aSMike Kravetz config.max_hpages, 11747ca02d0aSMike Kravetz config.min_hpages); 117590481622SDavid Gibson if (!sbinfo->spool) 117690481622SDavid Gibson goto out_free; 117790481622SDavid Gibson } 11781da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 1179a137e1ccSAndi Kleen sb->s_blocksize = huge_page_size(config.hstate); 1180a137e1ccSAndi Kleen sb->s_blocksize_bits = huge_page_shift(config.hstate); 11811da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 11821da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 11831da177e4SLinus Torvalds sb->s_time_gran = 1; 118448fde701SAl Viro sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); 118548fde701SAl Viro if (!sb->s_root) 11861da177e4SLinus Torvalds goto out_free; 11871da177e4SLinus Torvalds return 0; 11881da177e4SLinus Torvalds out_free: 118990481622SDavid Gibson kfree(sbinfo->spool); 11901da177e4SLinus Torvalds kfree(sbinfo); 11911da177e4SLinus Torvalds return -ENOMEM; 11921da177e4SLinus Torvalds } 11931da177e4SLinus Torvalds 11943c26ff6eSAl Viro static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, 11953c26ff6eSAl Viro int flags, const char *dev_name, void *data) 11961da177e4SLinus Torvalds { 11973c26ff6eSAl Viro return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); 11981da177e4SLinus Torvalds } 11991da177e4SLinus Torvalds 12001da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 12011da177e4SLinus Torvalds .name = "hugetlbfs", 12023c26ff6eSAl Viro .mount = hugetlbfs_mount, 12031da177e4SLinus Torvalds .kill_sb = kill_litter_super, 12041da177e4SLinus Torvalds }; 12051da177e4SLinus Torvalds 120642d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 12071da177e4SLinus Torvalds 1208ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 12091da177e4SLinus Torvalds { 1210a0eb3a05SEric W. Biederman kgid_t shm_group; 1211a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 1212a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 12131da177e4SLinus Torvalds } 12141da177e4SLinus Torvalds 121542d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 121642d7395fSAndi Kleen { 1217af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 121842d7395fSAndi Kleen 121942d7395fSAndi Kleen if (!h) 122042d7395fSAndi Kleen return -1; 122142d7395fSAndi Kleen return h - hstates; 122242d7395fSAndi Kleen } 122342d7395fSAndi Kleen 1224be1d2cf5SFabian Frederick static const struct dentry_operations anon_ops = { 1225118b2302SAl Viro .d_dname = simple_dname 12260df4d6e5SAl Viro }; 12270df4d6e5SAl Viro 1228af73e4d9SNaoya Horiguchi /* 1229af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 1230af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 1231af73e4d9SNaoya Horiguchi */ 1232af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 1233af73e4d9SNaoya Horiguchi vm_flags_t acctflag, struct user_struct **user, 123442d7395fSAndi Kleen int creat_flags, int page_size_log) 12351da177e4SLinus Torvalds { 123639b65252SAnatol Pomozov struct file *file = ERR_PTR(-ENOMEM); 12371da177e4SLinus Torvalds struct inode *inode; 12382c48b9c4SAl Viro struct path path; 12390df4d6e5SAl Viro struct super_block *sb; 12401da177e4SLinus Torvalds struct qstr quick_string; 124142d7395fSAndi Kleen int hstate_idx; 124242d7395fSAndi Kleen 124342d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 124442d7395fSAndi Kleen if (hstate_idx < 0) 124542d7395fSAndi Kleen return ERR_PTR(-ENODEV); 12461da177e4SLinus Torvalds 1247353d5c30SHugh Dickins *user = NULL; 124842d7395fSAndi Kleen if (!hugetlbfs_vfsmount[hstate_idx]) 12495bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 12505bc98594SAkinobu Mita 1251ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 1252353d5c30SHugh Dickins *user = current_user(); 1253353d5c30SHugh Dickins if (user_shm_lock(size, *user)) { 125421a3c273SDavid Rientjes task_lock(current); 12559b857d26SAndrew Morton pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", 125621a3c273SDavid Rientjes current->comm, current->pid); 125721a3c273SDavid Rientjes task_unlock(current); 1258353d5c30SHugh Dickins } else { 1259353d5c30SHugh Dickins *user = NULL; 12601da177e4SLinus Torvalds return ERR_PTR(-EPERM); 12612584e517SRavikiran G Thirumalai } 1262353d5c30SHugh Dickins } 12631da177e4SLinus Torvalds 12640df4d6e5SAl Viro sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb; 12659d66586fSEric W. Biederman quick_string.name = name; 12661da177e4SLinus Torvalds quick_string.len = strlen(quick_string.name); 12671da177e4SLinus Torvalds quick_string.hash = 0; 12680df4d6e5SAl Viro path.dentry = d_alloc_pseudo(sb, &quick_string); 12692c48b9c4SAl Viro if (!path.dentry) 12701da177e4SLinus Torvalds goto out_shm_unlock; 12711da177e4SLinus Torvalds 12720df4d6e5SAl Viro d_set_d_op(path.dentry, &anon_ops); 127342d7395fSAndi Kleen path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); 127439b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 12750df4d6e5SAl Viro inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0); 12761da177e4SLinus Torvalds if (!inode) 1277ce8d2cdfSDave Hansen goto out_dentry; 1278e1832f29SStephen Smalley if (creat_flags == HUGETLB_SHMFS_INODE) 1279e1832f29SStephen Smalley inode->i_flags |= S_PRIVATE; 12801da177e4SLinus Torvalds 128139b65252SAnatol Pomozov file = ERR_PTR(-ENOMEM); 1282af73e4d9SNaoya Horiguchi if (hugetlb_reserve_pages(inode, 0, 1283af73e4d9SNaoya Horiguchi size >> huge_page_shift(hstate_inode(inode)), NULL, 1284af73e4d9SNaoya Horiguchi acctflag)) 1285b45b5bd6SDavid Gibson goto out_inode; 1286b45b5bd6SDavid Gibson 12872c48b9c4SAl Viro d_instantiate(path.dentry, inode); 12881da177e4SLinus Torvalds inode->i_size = size; 12896d6b77f1SMiklos Szeredi clear_nlink(inode); 1290ce8d2cdfSDave Hansen 12912c48b9c4SAl Viro file = alloc_file(&path, FMODE_WRITE | FMODE_READ, 1292ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 129339b65252SAnatol Pomozov if (IS_ERR(file)) 1294b4d232e6SAl Viro goto out_dentry; /* inode is already attached */ 1295ce8d2cdfSDave Hansen 12961da177e4SLinus Torvalds return file; 12971da177e4SLinus Torvalds 1298b45b5bd6SDavid Gibson out_inode: 1299b45b5bd6SDavid Gibson iput(inode); 13001da177e4SLinus Torvalds out_dentry: 13012c48b9c4SAl Viro path_put(&path); 13021da177e4SLinus Torvalds out_shm_unlock: 1303353d5c30SHugh Dickins if (*user) { 1304353d5c30SHugh Dickins user_shm_unlock(size, *user); 1305353d5c30SHugh Dickins *user = NULL; 1306353d5c30SHugh Dickins } 130739b65252SAnatol Pomozov return file; 13081da177e4SLinus Torvalds } 13091da177e4SLinus Torvalds 13101da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 13111da177e4SLinus Torvalds { 131242d7395fSAndi Kleen struct hstate *h; 13131da177e4SLinus Torvalds int error; 131442d7395fSAndi Kleen int i; 13151da177e4SLinus Torvalds 1316457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 13179b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1318457c1b27SNishanth Aravamudan return -ENOTSUPP; 1319457c1b27SNishanth Aravamudan } 1320457c1b27SNishanth Aravamudan 1321d1d5e05fSHillf Danton error = -ENOMEM; 13221da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 13231da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 13245d097056SVladimir Davydov 0, SLAB_ACCOUNT, init_once); 13251da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 1326e0bf68ddSPeter Zijlstra goto out2; 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 13291da177e4SLinus Torvalds if (error) 13301da177e4SLinus Torvalds goto out; 13311da177e4SLinus Torvalds 133242d7395fSAndi Kleen i = 0; 133342d7395fSAndi Kleen for_each_hstate(h) { 133442d7395fSAndi Kleen char buf[50]; 133542d7395fSAndi Kleen unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); 13361da177e4SLinus Torvalds 133742d7395fSAndi Kleen snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); 133842d7395fSAndi Kleen hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, 133942d7395fSAndi Kleen buf); 134042d7395fSAndi Kleen 134142d7395fSAndi Kleen if (IS_ERR(hugetlbfs_vfsmount[i])) { 13429b857d26SAndrew Morton pr_err("Cannot mount internal hugetlbfs for " 134342d7395fSAndi Kleen "page size %uK", ps_kb); 134442d7395fSAndi Kleen error = PTR_ERR(hugetlbfs_vfsmount[i]); 134542d7395fSAndi Kleen hugetlbfs_vfsmount[i] = NULL; 13461da177e4SLinus Torvalds } 134742d7395fSAndi Kleen i++; 134842d7395fSAndi Kleen } 134942d7395fSAndi Kleen /* Non default hstates are optional */ 135042d7395fSAndi Kleen if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) 135142d7395fSAndi Kleen return 0; 13521da177e4SLinus Torvalds 13531da177e4SLinus Torvalds out: 13541da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 1355e0bf68ddSPeter Zijlstra out2: 13561da177e4SLinus Torvalds return error; 13571da177e4SLinus Torvalds } 13583e89e1c5SPaul Gortmaker fs_initcall(init_hugetlbfs_fs) 1359