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 581da177e4SLinus Torvalds int sysctl_hugetlb_shm_group; 591da177e4SLinus Torvalds 60e73a75faSRandy Dunlap enum { 61e73a75faSRandy Dunlap Opt_size, Opt_nr_inodes, 62e73a75faSRandy Dunlap Opt_mode, Opt_uid, Opt_gid, 637ca02d0aSMike Kravetz Opt_pagesize, Opt_min_size, 64e73a75faSRandy Dunlap Opt_err, 65e73a75faSRandy Dunlap }; 66e73a75faSRandy Dunlap 67a447c093SSteven Whitehouse static const match_table_t tokens = { 68e73a75faSRandy Dunlap {Opt_size, "size=%s"}, 69e73a75faSRandy Dunlap {Opt_nr_inodes, "nr_inodes=%s"}, 70e73a75faSRandy Dunlap {Opt_mode, "mode=%o"}, 71e73a75faSRandy Dunlap {Opt_uid, "uid=%u"}, 72e73a75faSRandy Dunlap {Opt_gid, "gid=%u"}, 73a137e1ccSAndi Kleen {Opt_pagesize, "pagesize=%s"}, 747ca02d0aSMike Kravetz {Opt_min_size, "min_size=%s"}, 75e73a75faSRandy Dunlap {Opt_err, NULL}, 76e73a75faSRandy Dunlap }; 77e73a75faSRandy Dunlap 7870c3547eSMike Kravetz #ifdef CONFIG_NUMA 7970c3547eSMike Kravetz static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, 8070c3547eSMike Kravetz struct inode *inode, pgoff_t index) 8170c3547eSMike Kravetz { 8270c3547eSMike Kravetz vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy, 8370c3547eSMike Kravetz index); 8470c3547eSMike Kravetz } 8570c3547eSMike Kravetz 8670c3547eSMike Kravetz static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) 8770c3547eSMike Kravetz { 8870c3547eSMike Kravetz mpol_cond_put(vma->vm_policy); 8970c3547eSMike Kravetz } 9070c3547eSMike Kravetz #else 9170c3547eSMike Kravetz static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, 9270c3547eSMike Kravetz struct inode *inode, pgoff_t index) 9370c3547eSMike Kravetz { 9470c3547eSMike Kravetz } 9570c3547eSMike Kravetz 9670c3547eSMike Kravetz static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) 9770c3547eSMike Kravetz { 9870c3547eSMike Kravetz } 9970c3547eSMike Kravetz #endif 10070c3547eSMike Kravetz 1012e9b367cSAdam Litke static void huge_pagevec_release(struct pagevec *pvec) 1022e9b367cSAdam Litke { 1032e9b367cSAdam Litke int i; 1042e9b367cSAdam Litke 1052e9b367cSAdam Litke for (i = 0; i < pagevec_count(pvec); ++i) 1062e9b367cSAdam Litke put_page(pvec->pages[i]); 1072e9b367cSAdam Litke 1082e9b367cSAdam Litke pagevec_reinit(pvec); 1092e9b367cSAdam Litke } 1102e9b367cSAdam Litke 1111da177e4SLinus Torvalds static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) 1121da177e4SLinus Torvalds { 113496ad9aaSAl Viro struct inode *inode = file_inode(file); 1141da177e4SLinus Torvalds loff_t len, vma_len; 1151da177e4SLinus Torvalds int ret; 116a5516438SAndi Kleen struct hstate *h = hstate_file(file); 1171da177e4SLinus Torvalds 11868589bc3SHugh Dickins /* 119dec4ad86SDavid Gibson * vma address alignment (but not the pgoff alignment) has 120dec4ad86SDavid Gibson * already been checked by prepare_hugepage_range. If you add 121dec4ad86SDavid Gibson * any error returns here, do so after setting VM_HUGETLB, so 122dec4ad86SDavid Gibson * is_vm_hugetlb_page tests below unmap_region go the right 123dec4ad86SDavid Gibson * way when do_mmap_pgoff unwinds (may be important on powerpc 124dec4ad86SDavid Gibson * and ia64). 12568589bc3SHugh Dickins */ 126a2fce914SNaoya Horiguchi vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; 12768589bc3SHugh Dickins vma->vm_ops = &hugetlb_vm_ops; 1281da177e4SLinus Torvalds 129045c7a3fSMike Kravetz /* 130045c7a3fSMike Kravetz * Offset passed to mmap (before page shift) could have been 131045c7a3fSMike Kravetz * negative when represented as a (l)off_t. 132045c7a3fSMike Kravetz */ 133045c7a3fSMike Kravetz if (((loff_t)vma->vm_pgoff << PAGE_SHIFT) < 0) 134045c7a3fSMike Kravetz return -EINVAL; 135045c7a3fSMike Kravetz 1362b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 137dec4ad86SDavid Gibson return -EINVAL; 138dec4ad86SDavid Gibson 1391da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 140045c7a3fSMike Kravetz len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 141045c7a3fSMike Kravetz /* check for overflow */ 142045c7a3fSMike Kravetz if (len < vma_len) 143045c7a3fSMike Kravetz return -EINVAL; 1441da177e4SLinus Torvalds 1455955102cSAl Viro inode_lock(inode); 1461da177e4SLinus Torvalds file_accessed(file); 1471da177e4SLinus Torvalds 1481da177e4SLinus Torvalds ret = -ENOMEM; 149a1e78772SMel Gorman if (hugetlb_reserve_pages(inode, 150a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1515a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1525a6fe125SMel Gorman vma->vm_flags)) 153b45b5bd6SDavid Gibson goto out; 154b45b5bd6SDavid Gibson 1554c887265SAdam Litke ret = 0; 156b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 157045c7a3fSMike Kravetz i_size_write(inode, len); 1581da177e4SLinus Torvalds out: 1595955102cSAl Viro inode_unlock(inode); 1601da177e4SLinus Torvalds 1611da177e4SLinus Torvalds return ret; 1621da177e4SLinus Torvalds } 1631da177e4SLinus Torvalds 1641da177e4SLinus Torvalds /* 165508034a3SHugh Dickins * Called under down_write(mmap_sem). 1661da177e4SLinus Torvalds */ 1671da177e4SLinus Torvalds 168d2ba27e8SAdrian Bunk #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 1691da177e4SLinus Torvalds static unsigned long 1701da177e4SLinus Torvalds hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 1711da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 1721da177e4SLinus Torvalds { 1731da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1741da177e4SLinus Torvalds struct vm_area_struct *vma; 175a5516438SAndi Kleen struct hstate *h = hstate_file(file); 17608659355SMichel Lespinasse struct vm_unmapped_area_info info; 1771da177e4SLinus Torvalds 178a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 1791da177e4SLinus Torvalds return -EINVAL; 1801da177e4SLinus Torvalds if (len > TASK_SIZE) 1811da177e4SLinus Torvalds return -ENOMEM; 1821da177e4SLinus Torvalds 183036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 184a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 185036e0856SBenjamin Herrenschmidt return -EINVAL; 186036e0856SBenjamin Herrenschmidt return addr; 187036e0856SBenjamin Herrenschmidt } 188036e0856SBenjamin Herrenschmidt 1891da177e4SLinus Torvalds if (addr) { 190a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 1911da177e4SLinus Torvalds vma = find_vma(mm, addr); 1921da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 1931be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma))) 1941da177e4SLinus Torvalds return addr; 1951da177e4SLinus Torvalds } 1961da177e4SLinus Torvalds 19708659355SMichel Lespinasse info.flags = 0; 19808659355SMichel Lespinasse info.length = len; 19908659355SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 20008659355SMichel Lespinasse info.high_limit = TASK_SIZE; 20108659355SMichel Lespinasse info.align_mask = PAGE_MASK & ~huge_page_mask(h); 20208659355SMichel Lespinasse info.align_offset = 0; 20308659355SMichel Lespinasse return vm_unmapped_area(&info); 2041da177e4SLinus Torvalds } 2051da177e4SLinus Torvalds #endif 2061da177e4SLinus Torvalds 20734d0640eSAl Viro static size_t 208e63e1e5aSBadari Pulavarty hugetlbfs_read_actor(struct page *page, unsigned long offset, 20934d0640eSAl Viro struct iov_iter *to, unsigned long size) 210e63e1e5aSBadari Pulavarty { 21134d0640eSAl Viro size_t copied = 0; 212e63e1e5aSBadari Pulavarty int i, chunksize; 213e63e1e5aSBadari Pulavarty 214e63e1e5aSBadari Pulavarty /* Find which 4k chunk and offset with in that chunk */ 21509cbfeafSKirill A. Shutemov i = offset >> PAGE_SHIFT; 21609cbfeafSKirill A. Shutemov offset = offset & ~PAGE_MASK; 217e63e1e5aSBadari Pulavarty 218e63e1e5aSBadari Pulavarty while (size) { 21934d0640eSAl Viro size_t n; 22009cbfeafSKirill A. Shutemov chunksize = PAGE_SIZE; 221e63e1e5aSBadari Pulavarty if (offset) 222e63e1e5aSBadari Pulavarty chunksize -= offset; 223e63e1e5aSBadari Pulavarty if (chunksize > size) 224e63e1e5aSBadari Pulavarty chunksize = size; 22534d0640eSAl Viro n = copy_page_to_iter(&page[i], offset, chunksize, to); 22634d0640eSAl Viro copied += n; 22734d0640eSAl Viro if (n != chunksize) 22834d0640eSAl Viro return copied; 229e63e1e5aSBadari Pulavarty offset = 0; 230e63e1e5aSBadari Pulavarty size -= chunksize; 231e63e1e5aSBadari Pulavarty i++; 232e63e1e5aSBadari Pulavarty } 23334d0640eSAl Viro return copied; 234e63e1e5aSBadari Pulavarty } 235e63e1e5aSBadari Pulavarty 236e63e1e5aSBadari Pulavarty /* 237e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 238e63e1e5aSBadari Pulavarty * data. Its *very* similar to do_generic_mapping_read(), we can't use that 239ea1754a0SKirill A. Shutemov * since it has PAGE_SIZE assumptions. 240e63e1e5aSBadari Pulavarty */ 24134d0640eSAl Viro static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) 242e63e1e5aSBadari Pulavarty { 24334d0640eSAl Viro struct file *file = iocb->ki_filp; 24434d0640eSAl Viro struct hstate *h = hstate_file(file); 24534d0640eSAl Viro struct address_space *mapping = file->f_mapping; 246e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 24734d0640eSAl Viro unsigned long index = iocb->ki_pos >> huge_page_shift(h); 24834d0640eSAl Viro unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); 249e63e1e5aSBadari Pulavarty unsigned long end_index; 250e63e1e5aSBadari Pulavarty loff_t isize; 251e63e1e5aSBadari Pulavarty ssize_t retval = 0; 252e63e1e5aSBadari Pulavarty 25334d0640eSAl Viro while (iov_iter_count(to)) { 254e63e1e5aSBadari Pulavarty struct page *page; 25534d0640eSAl Viro size_t nr, copied; 256e63e1e5aSBadari Pulavarty 257e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 258a5516438SAndi Kleen nr = huge_page_size(h); 259a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 260a05b0855SAneesh Kumar K.V if (!isize) 26134d0640eSAl Viro break; 262a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 263e63e1e5aSBadari Pulavarty if (index > end_index) 26434d0640eSAl Viro break; 26534d0640eSAl Viro if (index == end_index) { 266a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 267a05b0855SAneesh Kumar K.V if (nr <= offset) 26834d0640eSAl Viro break; 269e63e1e5aSBadari Pulavarty } 270e63e1e5aSBadari Pulavarty nr = nr - offset; 271e63e1e5aSBadari Pulavarty 272e63e1e5aSBadari Pulavarty /* Find the page */ 273a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 274e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 275e63e1e5aSBadari Pulavarty /* 276e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 277e63e1e5aSBadari Pulavarty * length of the hole or request. 278e63e1e5aSBadari Pulavarty */ 27934d0640eSAl Viro copied = iov_iter_zero(nr, to); 280e63e1e5aSBadari Pulavarty } else { 281a05b0855SAneesh Kumar K.V unlock_page(page); 282a05b0855SAneesh Kumar K.V 283e63e1e5aSBadari Pulavarty /* 284e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 285e63e1e5aSBadari Pulavarty */ 28634d0640eSAl Viro copied = hugetlbfs_read_actor(page, offset, to, nr); 28709cbfeafSKirill A. Shutemov put_page(page); 288e63e1e5aSBadari Pulavarty } 28934d0640eSAl Viro offset += copied; 29034d0640eSAl Viro retval += copied; 29134d0640eSAl Viro if (copied != nr && iov_iter_count(to)) { 29234d0640eSAl Viro if (!retval) 29334d0640eSAl Viro retval = -EFAULT; 294e63e1e5aSBadari Pulavarty break; 295e63e1e5aSBadari Pulavarty } 29634d0640eSAl Viro index += offset >> huge_page_shift(h); 29734d0640eSAl Viro offset &= ~huge_page_mask(h); 29834d0640eSAl Viro } 29934d0640eSAl Viro iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; 300e63e1e5aSBadari Pulavarty return retval; 301e63e1e5aSBadari Pulavarty } 302e63e1e5aSBadari Pulavarty 303800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 304800d15a5SNick Piggin struct address_space *mapping, 305800d15a5SNick Piggin loff_t pos, unsigned len, unsigned flags, 306800d15a5SNick Piggin struct page **pagep, void **fsdata) 3071da177e4SLinus Torvalds { 3081da177e4SLinus Torvalds return -EINVAL; 3091da177e4SLinus Torvalds } 3101da177e4SLinus Torvalds 311800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 312800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 313800d15a5SNick Piggin struct page *page, void *fsdata) 3141da177e4SLinus Torvalds { 315800d15a5SNick Piggin BUG(); 3161da177e4SLinus Torvalds return -EINVAL; 3171da177e4SLinus Torvalds } 3181da177e4SLinus Torvalds 319b5cec28dSMike Kravetz static void remove_huge_page(struct page *page) 3201da177e4SLinus Torvalds { 321b9ea2515SKonstantin Khlebnikov ClearPageDirty(page); 3221da177e4SLinus Torvalds ClearPageUptodate(page); 323bd65cb86SMinchan Kim delete_from_page_cache(page); 3241da177e4SLinus Torvalds } 3251da177e4SLinus Torvalds 3264aae8d1cSMike Kravetz static void 327f808c13fSDavidlohr Bueso hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end) 3284aae8d1cSMike Kravetz { 3294aae8d1cSMike Kravetz struct vm_area_struct *vma; 3304aae8d1cSMike Kravetz 3314aae8d1cSMike Kravetz /* 3324aae8d1cSMike Kravetz * end == 0 indicates that the entire range after 3334aae8d1cSMike Kravetz * start should be unmapped. 3344aae8d1cSMike Kravetz */ 3354aae8d1cSMike Kravetz vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { 3364aae8d1cSMike Kravetz unsigned long v_offset; 3374aae8d1cSMike Kravetz unsigned long v_end; 3384aae8d1cSMike Kravetz 3394aae8d1cSMike Kravetz /* 3404aae8d1cSMike Kravetz * Can the expression below overflow on 32-bit arches? 3414aae8d1cSMike Kravetz * No, because the interval tree returns us only those vmas 3424aae8d1cSMike Kravetz * which overlap the truncated area starting at pgoff, 3434aae8d1cSMike Kravetz * and no vma on a 32-bit arch can span beyond the 4GB. 3444aae8d1cSMike Kravetz */ 3454aae8d1cSMike Kravetz if (vma->vm_pgoff < start) 3464aae8d1cSMike Kravetz v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; 3474aae8d1cSMike Kravetz else 3484aae8d1cSMike Kravetz v_offset = 0; 3494aae8d1cSMike Kravetz 3504aae8d1cSMike Kravetz if (!end) 3514aae8d1cSMike Kravetz v_end = vma->vm_end; 3524aae8d1cSMike Kravetz else { 3534aae8d1cSMike Kravetz v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) 3544aae8d1cSMike Kravetz + vma->vm_start; 3554aae8d1cSMike Kravetz if (v_end > vma->vm_end) 3564aae8d1cSMike Kravetz v_end = vma->vm_end; 3574aae8d1cSMike Kravetz } 3584aae8d1cSMike Kravetz 3594aae8d1cSMike Kravetz unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, 3604aae8d1cSMike Kravetz NULL); 3614aae8d1cSMike Kravetz } 3624aae8d1cSMike Kravetz } 363b5cec28dSMike Kravetz 364b5cec28dSMike Kravetz /* 365b5cec28dSMike Kravetz * remove_inode_hugepages handles two distinct cases: truncation and hole 366b5cec28dSMike Kravetz * punch. There are subtle differences in operation for each case. 3674aae8d1cSMike Kravetz * 368b5cec28dSMike Kravetz * truncation is indicated by end of range being LLONG_MAX 369b5cec28dSMike Kravetz * In this case, we first scan the range and release found pages. 370b5cec28dSMike Kravetz * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv 3711817889eSMike Kravetz * maps and global counts. Page faults can not race with truncation 3721817889eSMike Kravetz * in this routine. hugetlb_no_page() prevents page faults in the 3731817889eSMike Kravetz * truncated range. It checks i_size before allocation, and again after 3741817889eSMike Kravetz * with the page table lock for the page held. The same lock must be 3751817889eSMike Kravetz * acquired to unmap a page. 376b5cec28dSMike Kravetz * hole punch is indicated if end is not LLONG_MAX 377b5cec28dSMike Kravetz * In the hole punch case we scan the range and release found pages. 378b5cec28dSMike Kravetz * Only when releasing a page is the associated region/reserv map 379b5cec28dSMike Kravetz * deleted. The region/reserv map for ranges without associated 3801817889eSMike Kravetz * pages are not modified. Page faults can race with hole punch. 3811817889eSMike Kravetz * This is indicated if we find a mapped page. 382b5cec28dSMike Kravetz * Note: If the passed end of range value is beyond the end of file, but 383b5cec28dSMike Kravetz * not LLONG_MAX this routine still performs a hole punch operation. 384b5cec28dSMike Kravetz */ 385b5cec28dSMike Kravetz static void remove_inode_hugepages(struct inode *inode, loff_t lstart, 386b5cec28dSMike Kravetz loff_t lend) 3871da177e4SLinus Torvalds { 388a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 389b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 390a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 391b5cec28dSMike Kravetz const pgoff_t end = lend >> huge_page_shift(h); 392b5cec28dSMike Kravetz struct vm_area_struct pseudo_vma; 3931da177e4SLinus Torvalds struct pagevec pvec; 394d72dc8a2SJan Kara pgoff_t next, index; 395a43a8c39SChen, Kenneth W int i, freed = 0; 396b5cec28dSMike Kravetz bool truncate_op = (lend == LLONG_MAX); 3971da177e4SLinus Torvalds 398b5cec28dSMike Kravetz memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); 399b5cec28dSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 40086679820SMel Gorman pagevec_init(&pvec); 4011da177e4SLinus Torvalds next = start; 402b5cec28dSMike Kravetz while (next < end) { 403b5cec28dSMike Kravetz /* 4041817889eSMike Kravetz * When no more pages are found, we are done. 405b5cec28dSMike Kravetz */ 406397162ffSJan Kara if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1)) 4071da177e4SLinus Torvalds break; 4081da177e4SLinus Torvalds 4091da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); ++i) { 4101da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 411b5cec28dSMike Kravetz u32 hash; 412b5cec28dSMike Kravetz 413d72dc8a2SJan Kara index = page->index; 414b5cec28dSMike Kravetz hash = hugetlb_fault_mutex_hash(h, current->mm, 415b5cec28dSMike Kravetz &pseudo_vma, 416d72dc8a2SJan Kara mapping, index, 0); 417b5cec28dSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 4181da177e4SLinus Torvalds 419b5cec28dSMike Kravetz /* 4204aae8d1cSMike Kravetz * If page is mapped, it was faulted in after being 4214aae8d1cSMike Kravetz * unmapped in caller. Unmap (again) now after taking 4224aae8d1cSMike Kravetz * the fault mutex. The mutex will prevent faults 4234aae8d1cSMike Kravetz * until we finish removing the page. 4244aae8d1cSMike Kravetz * 4254aae8d1cSMike Kravetz * This race can only happen in the hole punch case. 4264aae8d1cSMike Kravetz * Getting here in a truncate operation is a bug. 427b5cec28dSMike Kravetz */ 4284aae8d1cSMike Kravetz if (unlikely(page_mapped(page))) { 4294aae8d1cSMike Kravetz BUG_ON(truncate_op); 4304aae8d1cSMike Kravetz 4314aae8d1cSMike Kravetz i_mmap_lock_write(mapping); 4324aae8d1cSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 433d72dc8a2SJan Kara index * pages_per_huge_page(h), 434d72dc8a2SJan Kara (index + 1) * pages_per_huge_page(h)); 4354aae8d1cSMike Kravetz i_mmap_unlock_write(mapping); 4364aae8d1cSMike Kravetz } 4374aae8d1cSMike Kravetz 4384aae8d1cSMike Kravetz lock_page(page); 4394aae8d1cSMike Kravetz /* 4404aae8d1cSMike Kravetz * We must free the huge page and remove from page 4414aae8d1cSMike Kravetz * cache (remove_huge_page) BEFORE removing the 4424aae8d1cSMike Kravetz * region/reserve map (hugetlb_unreserve_pages). In 4434aae8d1cSMike Kravetz * rare out of memory conditions, removal of the 44472e2936cSzhong jiang * region/reserve map could fail. Correspondingly, 44572e2936cSzhong jiang * the subpool and global reserve usage count can need 44672e2936cSzhong jiang * to be adjusted. 4474aae8d1cSMike Kravetz */ 44872e2936cSzhong jiang VM_BUG_ON(PagePrivate(page)); 449b5cec28dSMike Kravetz remove_huge_page(page); 450b5cec28dSMike Kravetz freed++; 451b5cec28dSMike Kravetz if (!truncate_op) { 4524aae8d1cSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, 453d72dc8a2SJan Kara index, index + 1, 1))) 45472e2936cSzhong jiang hugetlb_fix_reserve_counts(inode); 455b5cec28dSMike Kravetz } 456b5cec28dSMike Kravetz 4571da177e4SLinus Torvalds unlock_page(page); 458b5cec28dSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 4591da177e4SLinus Torvalds } 4601da177e4SLinus Torvalds huge_pagevec_release(&pvec); 4611817889eSMike Kravetz cond_resched(); 4621da177e4SLinus Torvalds } 463b5cec28dSMike Kravetz 464b5cec28dSMike Kravetz if (truncate_op) 465b5cec28dSMike Kravetz (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); 4661da177e4SLinus Torvalds } 4671da177e4SLinus Torvalds 4682bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 4691da177e4SLinus Torvalds { 4709119a41eSJoonsoo Kim struct resv_map *resv_map; 4719119a41eSJoonsoo Kim 472b5cec28dSMike Kravetz remove_inode_hugepages(inode, 0, LLONG_MAX); 4739119a41eSJoonsoo Kim resv_map = (struct resv_map *)inode->i_mapping->private_data; 4749119a41eSJoonsoo Kim /* root inode doesn't have the resv_map, so we should check it */ 4759119a41eSJoonsoo Kim if (resv_map) 4769119a41eSJoonsoo Kim resv_map_release(&resv_map->refs); 477dbd5768fSJan Kara clear_inode(inode); 478149f4211SChristoph Hellwig } 479149f4211SChristoph Hellwig 4801da177e4SLinus Torvalds static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) 4811da177e4SLinus Torvalds { 482856fc295SHugh Dickins pgoff_t pgoff; 4831da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 484a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 4851da177e4SLinus Torvalds 486a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 487856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 4881da177e4SLinus Torvalds 4897aa91e10SKen Chen i_size_write(inode, offset); 49083cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 491f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 4921bfad99aSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); 49383cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 494b5cec28dSMike Kravetz remove_inode_hugepages(inode, offset, LLONG_MAX); 4951da177e4SLinus Torvalds return 0; 4961da177e4SLinus Torvalds } 4971da177e4SLinus Torvalds 49870c3547eSMike Kravetz static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) 49970c3547eSMike Kravetz { 50070c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 50170c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 50270c3547eSMike Kravetz loff_t hole_start, hole_end; 50370c3547eSMike Kravetz 50470c3547eSMike Kravetz /* 50570c3547eSMike Kravetz * For hole punch round up the beginning offset of the hole and 50670c3547eSMike Kravetz * round down the end. 50770c3547eSMike Kravetz */ 50870c3547eSMike Kravetz hole_start = round_up(offset, hpage_size); 50970c3547eSMike Kravetz hole_end = round_down(offset + len, hpage_size); 51070c3547eSMike Kravetz 51170c3547eSMike Kravetz if (hole_end > hole_start) { 51270c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 513ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 51470c3547eSMike Kravetz 5155955102cSAl Viro inode_lock(inode); 516ff62a342SMarc-André Lureau 517ff62a342SMarc-André Lureau /* protected by i_mutex */ 518ff62a342SMarc-André Lureau if (info->seals & F_SEAL_WRITE) { 519ff62a342SMarc-André Lureau inode_unlock(inode); 520ff62a342SMarc-André Lureau return -EPERM; 521ff62a342SMarc-André Lureau } 522ff62a342SMarc-André Lureau 52370c3547eSMike Kravetz i_mmap_lock_write(mapping); 524f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 52570c3547eSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 52670c3547eSMike Kravetz hole_start >> PAGE_SHIFT, 52770c3547eSMike Kravetz hole_end >> PAGE_SHIFT); 52870c3547eSMike Kravetz i_mmap_unlock_write(mapping); 52970c3547eSMike Kravetz remove_inode_hugepages(inode, hole_start, hole_end); 5305955102cSAl Viro inode_unlock(inode); 53170c3547eSMike Kravetz } 53270c3547eSMike Kravetz 53370c3547eSMike Kravetz return 0; 53470c3547eSMike Kravetz } 53570c3547eSMike Kravetz 53670c3547eSMike Kravetz static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, 53770c3547eSMike Kravetz loff_t len) 53870c3547eSMike Kravetz { 53970c3547eSMike Kravetz struct inode *inode = file_inode(file); 540ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 54170c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 54270c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 54370c3547eSMike Kravetz struct vm_area_struct pseudo_vma; 54470c3547eSMike Kravetz struct mm_struct *mm = current->mm; 54570c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 54670c3547eSMike Kravetz unsigned long hpage_shift = huge_page_shift(h); 54770c3547eSMike Kravetz pgoff_t start, index, end; 54870c3547eSMike Kravetz int error; 54970c3547eSMike Kravetz u32 hash; 55070c3547eSMike Kravetz 55170c3547eSMike Kravetz if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 55270c3547eSMike Kravetz return -EOPNOTSUPP; 55370c3547eSMike Kravetz 55470c3547eSMike Kravetz if (mode & FALLOC_FL_PUNCH_HOLE) 55570c3547eSMike Kravetz return hugetlbfs_punch_hole(inode, offset, len); 55670c3547eSMike Kravetz 55770c3547eSMike Kravetz /* 55870c3547eSMike Kravetz * Default preallocate case. 55970c3547eSMike Kravetz * For this range, start is rounded down and end is rounded up 56070c3547eSMike Kravetz * as well as being converted to page offsets. 56170c3547eSMike Kravetz */ 56270c3547eSMike Kravetz start = offset >> hpage_shift; 56370c3547eSMike Kravetz end = (offset + len + hpage_size - 1) >> hpage_shift; 56470c3547eSMike Kravetz 5655955102cSAl Viro inode_lock(inode); 56670c3547eSMike Kravetz 56770c3547eSMike Kravetz /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ 56870c3547eSMike Kravetz error = inode_newsize_ok(inode, offset + len); 56970c3547eSMike Kravetz if (error) 57070c3547eSMike Kravetz goto out; 57170c3547eSMike Kravetz 572ff62a342SMarc-André Lureau if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { 573ff62a342SMarc-André Lureau error = -EPERM; 574ff62a342SMarc-André Lureau goto out; 575ff62a342SMarc-André Lureau } 576ff62a342SMarc-André Lureau 57770c3547eSMike Kravetz /* 57870c3547eSMike Kravetz * Initialize a pseudo vma as this is required by the huge page 57970c3547eSMike Kravetz * allocation routines. If NUMA is configured, use page index 58070c3547eSMike Kravetz * as input to create an allocation policy. 58170c3547eSMike Kravetz */ 58270c3547eSMike Kravetz memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); 58370c3547eSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 58470c3547eSMike Kravetz pseudo_vma.vm_file = file; 58570c3547eSMike Kravetz 58670c3547eSMike Kravetz for (index = start; index < end; index++) { 58770c3547eSMike Kravetz /* 58870c3547eSMike Kravetz * This is supposed to be the vaddr where the page is being 58970c3547eSMike Kravetz * faulted in, but we have no vaddr here. 59070c3547eSMike Kravetz */ 59170c3547eSMike Kravetz struct page *page; 59270c3547eSMike Kravetz unsigned long addr; 59370c3547eSMike Kravetz int avoid_reserve = 0; 59470c3547eSMike Kravetz 59570c3547eSMike Kravetz cond_resched(); 59670c3547eSMike Kravetz 59770c3547eSMike Kravetz /* 59870c3547eSMike Kravetz * fallocate(2) manpage permits EINTR; we may have been 59970c3547eSMike Kravetz * interrupted because we are using up too much memory. 60070c3547eSMike Kravetz */ 60170c3547eSMike Kravetz if (signal_pending(current)) { 60270c3547eSMike Kravetz error = -EINTR; 60370c3547eSMike Kravetz break; 60470c3547eSMike Kravetz } 60570c3547eSMike Kravetz 60670c3547eSMike Kravetz /* Set numa allocation policy based on index */ 60770c3547eSMike Kravetz hugetlb_set_vma_policy(&pseudo_vma, inode, index); 60870c3547eSMike Kravetz 60970c3547eSMike Kravetz /* addr is the offset within the file (zero based) */ 61070c3547eSMike Kravetz addr = index * hpage_size; 61170c3547eSMike Kravetz 61270c3547eSMike Kravetz /* mutex taken here, fault path and hole punch */ 61370c3547eSMike Kravetz hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping, 61470c3547eSMike Kravetz index, addr); 61570c3547eSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 61670c3547eSMike Kravetz 61770c3547eSMike Kravetz /* See if already present in mapping to avoid alloc/free */ 61870c3547eSMike Kravetz page = find_get_page(mapping, index); 61970c3547eSMike Kravetz if (page) { 62070c3547eSMike Kravetz put_page(page); 62170c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 62270c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 62370c3547eSMike Kravetz continue; 62470c3547eSMike Kravetz } 62570c3547eSMike Kravetz 62670c3547eSMike Kravetz /* Allocate page and add to page cache */ 62770c3547eSMike Kravetz page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); 62870c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 62970c3547eSMike Kravetz if (IS_ERR(page)) { 63070c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 63170c3547eSMike Kravetz error = PTR_ERR(page); 63270c3547eSMike Kravetz goto out; 63370c3547eSMike Kravetz } 63470c3547eSMike Kravetz clear_huge_page(page, addr, pages_per_huge_page(h)); 63570c3547eSMike Kravetz __SetPageUptodate(page); 63670c3547eSMike Kravetz error = huge_add_to_page_cache(page, mapping, index); 63770c3547eSMike Kravetz if (unlikely(error)) { 63870c3547eSMike Kravetz put_page(page); 63970c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 64070c3547eSMike Kravetz goto out; 64170c3547eSMike Kravetz } 64270c3547eSMike Kravetz 64370c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 64470c3547eSMike Kravetz 64570c3547eSMike Kravetz /* 64670c3547eSMike Kravetz * unlock_page because locked by add_to_page_cache() 64772639e6dSNadav Amit * page_put due to reference from alloc_huge_page() 64870c3547eSMike Kravetz */ 64970c3547eSMike Kravetz unlock_page(page); 65072639e6dSNadav Amit put_page(page); 65170c3547eSMike Kravetz } 65270c3547eSMike Kravetz 65370c3547eSMike Kravetz if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) 65470c3547eSMike Kravetz i_size_write(inode, offset + len); 655078cd827SDeepa Dinamani inode->i_ctime = current_time(inode); 65670c3547eSMike Kravetz out: 6575955102cSAl Viro inode_unlock(inode); 65870c3547eSMike Kravetz return error; 65970c3547eSMike Kravetz } 66070c3547eSMike Kravetz 6611da177e4SLinus Torvalds static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) 6621da177e4SLinus Torvalds { 6632b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 664a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 6651da177e4SLinus Torvalds int error; 6661da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 667ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 6681da177e4SLinus Torvalds 6691da177e4SLinus Torvalds BUG_ON(!inode); 6701da177e4SLinus Torvalds 67131051c85SJan Kara error = setattr_prepare(dentry, attr); 6721da177e4SLinus Torvalds if (error) 6731025774cSChristoph Hellwig return error; 6741da177e4SLinus Torvalds 6751da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 676ff62a342SMarc-André Lureau loff_t oldsize = inode->i_size; 677ff62a342SMarc-André Lureau loff_t newsize = attr->ia_size; 678ff62a342SMarc-André Lureau 679ff62a342SMarc-André Lureau if (newsize & ~huge_page_mask(h)) 6801025774cSChristoph Hellwig return -EINVAL; 681ff62a342SMarc-André Lureau /* protected by i_mutex */ 682ff62a342SMarc-André Lureau if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || 683ff62a342SMarc-André Lureau (newsize > oldsize && (info->seals & F_SEAL_GROW))) 684ff62a342SMarc-André Lureau return -EPERM; 685ff62a342SMarc-André Lureau error = hugetlb_vmtruncate(inode, newsize); 6861da177e4SLinus Torvalds if (error) 6871da177e4SLinus Torvalds return error; 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds 6901025774cSChristoph Hellwig setattr_copy(inode, attr); 6911025774cSChristoph Hellwig mark_inode_dirty(inode); 6921025774cSChristoph Hellwig return 0; 6931025774cSChristoph Hellwig } 6941025774cSChristoph Hellwig 6957d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 6967d54fa64SAl Viro struct hugetlbfs_config *config) 6971da177e4SLinus Torvalds { 6981da177e4SLinus Torvalds struct inode *inode; 6991da177e4SLinus Torvalds 7001da177e4SLinus Torvalds inode = new_inode(sb); 7011da177e4SLinus Torvalds if (inode) { 70285fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 7037d54fa64SAl Viro inode->i_mode = S_IFDIR | config->mode; 7047d54fa64SAl Viro inode->i_uid = config->uid; 7057d54fa64SAl Viro inode->i_gid = config->gid; 706078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 7077d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 7087d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 7097d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 7107d54fa64SAl Viro inc_nlink(inode); 71165ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 7127d54fa64SAl Viro } 7137d54fa64SAl Viro return inode; 7147d54fa64SAl Viro } 7157d54fa64SAl Viro 716b610ded7SMichal Hocko /* 717c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 718b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 71988f306b6SKirill A. Shutemov * annotation because huge_pmd_share() does an allocation under hugetlb's 720c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 721b610ded7SMichal Hocko */ 722c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 723b610ded7SMichal Hocko 7247d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 7257d54fa64SAl Viro struct inode *dir, 72618df2252SAl Viro umode_t mode, dev_t dev) 7277d54fa64SAl Viro { 7287d54fa64SAl Viro struct inode *inode; 7299119a41eSJoonsoo Kim struct resv_map *resv_map; 7309119a41eSJoonsoo Kim 7319119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 7329119a41eSJoonsoo Kim if (!resv_map) 7339119a41eSJoonsoo Kim return NULL; 7347d54fa64SAl Viro 7357d54fa64SAl Viro inode = new_inode(sb); 7367d54fa64SAl Viro if (inode) { 737ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 738ff62a342SMarc-André Lureau 7397d54fa64SAl Viro inode->i_ino = get_next_ino(); 7407d54fa64SAl Viro inode_init_owner(inode, dir, mode); 741c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 742c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 7431da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 744078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 7459119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 746ff62a342SMarc-André Lureau info->seals = F_SEAL_SEAL; 7471da177e4SLinus Torvalds switch (mode & S_IFMT) { 7481da177e4SLinus Torvalds default: 7491da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 7501da177e4SLinus Torvalds break; 7511da177e4SLinus Torvalds case S_IFREG: 7521da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 7531da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 7541da177e4SLinus Torvalds break; 7551da177e4SLinus Torvalds case S_IFDIR: 7561da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 7571da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 7581da177e4SLinus Torvalds 7591da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 760d8c76e6fSDave Hansen inc_nlink(inode); 7611da177e4SLinus Torvalds break; 7621da177e4SLinus Torvalds case S_IFLNK: 7631da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 76421fc61c7SAl Viro inode_nohighmem(inode); 7651da177e4SLinus Torvalds break; 7661da177e4SLinus Torvalds } 767e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 7689119a41eSJoonsoo Kim } else 7699119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 7709119a41eSJoonsoo Kim 7711da177e4SLinus Torvalds return inode; 7721da177e4SLinus Torvalds } 7731da177e4SLinus Torvalds 7741da177e4SLinus Torvalds /* 7751da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 7761da177e4SLinus Torvalds */ 7771da177e4SLinus Torvalds static int hugetlbfs_mknod(struct inode *dir, 7781a67aafbSAl Viro struct dentry *dentry, umode_t mode, dev_t dev) 7791da177e4SLinus Torvalds { 7801da177e4SLinus Torvalds struct inode *inode; 7811da177e4SLinus Torvalds int error = -ENOSPC; 7821da177e4SLinus Torvalds 7837d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 7841da177e4SLinus Torvalds if (inode) { 785078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 7861da177e4SLinus Torvalds d_instantiate(dentry, inode); 7871da177e4SLinus Torvalds dget(dentry); /* Extra count - pin the dentry in core */ 7881da177e4SLinus Torvalds error = 0; 7891da177e4SLinus Torvalds } 7901da177e4SLinus Torvalds return error; 7911da177e4SLinus Torvalds } 7921da177e4SLinus Torvalds 79318bb1db3SAl Viro static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 7941da177e4SLinus Torvalds { 7951da177e4SLinus Torvalds int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); 7961da177e4SLinus Torvalds if (!retval) 797d8c76e6fSDave Hansen inc_nlink(dir); 7981da177e4SLinus Torvalds return retval; 7991da177e4SLinus Torvalds } 8001da177e4SLinus Torvalds 801ebfc3b49SAl Viro static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) 8021da177e4SLinus Torvalds { 8031da177e4SLinus Torvalds return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); 8041da177e4SLinus Torvalds } 8051da177e4SLinus Torvalds 8061da177e4SLinus Torvalds static int hugetlbfs_symlink(struct inode *dir, 8071da177e4SLinus Torvalds struct dentry *dentry, const char *symname) 8081da177e4SLinus Torvalds { 8091da177e4SLinus Torvalds struct inode *inode; 8101da177e4SLinus Torvalds int error = -ENOSPC; 8111da177e4SLinus Torvalds 8127d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 8131da177e4SLinus Torvalds if (inode) { 8141da177e4SLinus Torvalds int l = strlen(symname)+1; 8151da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 8161da177e4SLinus Torvalds if (!error) { 8171da177e4SLinus Torvalds d_instantiate(dentry, inode); 8181da177e4SLinus Torvalds dget(dentry); 8191da177e4SLinus Torvalds } else 8201da177e4SLinus Torvalds iput(inode); 8211da177e4SLinus Torvalds } 822078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 8231da177e4SLinus Torvalds 8241da177e4SLinus Torvalds return error; 8251da177e4SLinus Torvalds } 8261da177e4SLinus Torvalds 8271da177e4SLinus Torvalds /* 8286649a386SKen Chen * mark the head page dirty 8291da177e4SLinus Torvalds */ 8301da177e4SLinus Torvalds static int hugetlbfs_set_page_dirty(struct page *page) 8311da177e4SLinus Torvalds { 832d85f3385SChristoph Lameter struct page *head = compound_head(page); 8336649a386SKen Chen 8346649a386SKen Chen SetPageDirty(head); 8351da177e4SLinus Torvalds return 0; 8361da177e4SLinus Torvalds } 8371da177e4SLinus Torvalds 838290408d4SNaoya Horiguchi static int hugetlbfs_migrate_page(struct address_space *mapping, 839b969c4abSMel Gorman struct page *newpage, struct page *page, 840a6bc32b8SMel Gorman enum migrate_mode mode) 841290408d4SNaoya Horiguchi { 842290408d4SNaoya Horiguchi int rc; 843290408d4SNaoya Horiguchi 844290408d4SNaoya Horiguchi rc = migrate_huge_page_move_mapping(mapping, newpage, page); 84578bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 846290408d4SNaoya Horiguchi return rc; 8472916ecc0SJérôme Glisse if (mode != MIGRATE_SYNC_NO_COPY) 848290408d4SNaoya Horiguchi migrate_page_copy(newpage, page); 8492916ecc0SJérôme Glisse else 8502916ecc0SJérôme Glisse migrate_page_states(newpage, page); 851290408d4SNaoya Horiguchi 85278bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 853290408d4SNaoya Horiguchi } 854290408d4SNaoya Horiguchi 85578bb9203SNaoya Horiguchi static int hugetlbfs_error_remove_page(struct address_space *mapping, 85678bb9203SNaoya Horiguchi struct page *page) 85778bb9203SNaoya Horiguchi { 85878bb9203SNaoya Horiguchi struct inode *inode = mapping->host; 859ab615a5bSMike Kravetz pgoff_t index = page->index; 86078bb9203SNaoya Horiguchi 86178bb9203SNaoya Horiguchi remove_huge_page(page); 862ab615a5bSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) 86378bb9203SNaoya Horiguchi hugetlb_fix_reserve_counts(inode); 864ab615a5bSMike Kravetz 86578bb9203SNaoya Horiguchi return 0; 86678bb9203SNaoya Horiguchi } 86778bb9203SNaoya Horiguchi 8684a25220dSDavid Howells /* 8694a25220dSDavid Howells * Display the mount options in /proc/mounts. 8704a25220dSDavid Howells */ 8714a25220dSDavid Howells static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) 8724a25220dSDavid Howells { 8734a25220dSDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); 8744a25220dSDavid Howells struct hugepage_subpool *spool = sbinfo->spool; 8754a25220dSDavid Howells unsigned long hpage_size = huge_page_size(sbinfo->hstate); 8764a25220dSDavid Howells unsigned hpage_shift = huge_page_shift(sbinfo->hstate); 8774a25220dSDavid Howells char mod; 8784a25220dSDavid Howells 8794a25220dSDavid Howells if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) 8804a25220dSDavid Howells seq_printf(m, ",uid=%u", 8814a25220dSDavid Howells from_kuid_munged(&init_user_ns, sbinfo->uid)); 8824a25220dSDavid Howells if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) 8834a25220dSDavid Howells seq_printf(m, ",gid=%u", 8844a25220dSDavid Howells from_kgid_munged(&init_user_ns, sbinfo->gid)); 8854a25220dSDavid Howells if (sbinfo->mode != 0755) 8864a25220dSDavid Howells seq_printf(m, ",mode=%o", sbinfo->mode); 8874a25220dSDavid Howells if (sbinfo->max_inodes != -1) 8884a25220dSDavid Howells seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); 8894a25220dSDavid Howells 8904a25220dSDavid Howells hpage_size /= 1024; 8914a25220dSDavid Howells mod = 'K'; 8924a25220dSDavid Howells if (hpage_size >= 1024) { 8934a25220dSDavid Howells hpage_size /= 1024; 8944a25220dSDavid Howells mod = 'M'; 8954a25220dSDavid Howells } 8964a25220dSDavid Howells seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); 8974a25220dSDavid Howells if (spool) { 8984a25220dSDavid Howells if (spool->max_hpages != -1) 8994a25220dSDavid Howells seq_printf(m, ",size=%llu", 9004a25220dSDavid Howells (unsigned long long)spool->max_hpages << hpage_shift); 9014a25220dSDavid Howells if (spool->min_hpages != -1) 9024a25220dSDavid Howells seq_printf(m, ",min_size=%llu", 9034a25220dSDavid Howells (unsigned long long)spool->min_hpages << hpage_shift); 9044a25220dSDavid Howells } 9054a25220dSDavid Howells return 0; 9064a25220dSDavid Howells } 9074a25220dSDavid Howells 908726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 9091da177e4SLinus Torvalds { 910726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 9112b0143b5SDavid Howells struct hstate *h = hstate_inode(d_inode(dentry)); 9121da177e4SLinus Torvalds 9131da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 914a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 9151da177e4SLinus Torvalds if (sbinfo) { 9161da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 91774a8a65cSDavid Gibson /* If no limits set, just report 0 for max/free/used 91874a8a65cSDavid Gibson * blocks, like simple_statfs() */ 91990481622SDavid Gibson if (sbinfo->spool) { 92090481622SDavid Gibson long free_pages; 92190481622SDavid Gibson 92290481622SDavid Gibson spin_lock(&sbinfo->spool->lock); 92390481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 92490481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 92590481622SDavid Gibson - sbinfo->spool->used_hpages; 92690481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 92790481622SDavid Gibson spin_unlock(&sbinfo->spool->lock); 9281da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 9291da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 93074a8a65cSDavid Gibson } 9311da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 9321da177e4SLinus Torvalds } 9331da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 9341da177e4SLinus Torvalds return 0; 9351da177e4SLinus Torvalds } 9361da177e4SLinus Torvalds 9371da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 9381da177e4SLinus Torvalds { 9391da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 9401da177e4SLinus Torvalds 9411da177e4SLinus Torvalds if (sbi) { 9421da177e4SLinus Torvalds sb->s_fs_info = NULL; 94390481622SDavid Gibson 94490481622SDavid Gibson if (sbi->spool) 94590481622SDavid Gibson hugepage_put_subpool(sbi->spool); 94690481622SDavid Gibson 9471da177e4SLinus Torvalds kfree(sbi); 9481da177e4SLinus Torvalds } 9491da177e4SLinus Torvalds } 9501da177e4SLinus Torvalds 95196527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 95296527980SChristoph Hellwig { 95396527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 95496527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 95596527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 95696527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 95796527980SChristoph Hellwig return 0; 95896527980SChristoph Hellwig } 95996527980SChristoph Hellwig sbinfo->free_inodes--; 96096527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 96196527980SChristoph Hellwig } 96296527980SChristoph Hellwig 96396527980SChristoph Hellwig return 1; 96496527980SChristoph Hellwig } 96596527980SChristoph Hellwig 96696527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 96796527980SChristoph Hellwig { 96896527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 96996527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 97096527980SChristoph Hellwig sbinfo->free_inodes++; 97196527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 97296527980SChristoph Hellwig } 97396527980SChristoph Hellwig } 97496527980SChristoph Hellwig 97596527980SChristoph Hellwig 976e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 9771da177e4SLinus Torvalds 9781da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 9791da177e4SLinus Torvalds { 98096527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 9811da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 9821da177e4SLinus Torvalds 98396527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 9841da177e4SLinus Torvalds return NULL; 985e94b1766SChristoph Lameter p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); 98696527980SChristoph Hellwig if (unlikely(!p)) { 98796527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 98896527980SChristoph Hellwig return NULL; 9891da177e4SLinus Torvalds } 9904742a35dSMike Kravetz 9914742a35dSMike Kravetz /* 9924742a35dSMike Kravetz * Any time after allocation, hugetlbfs_destroy_inode can be called 9934742a35dSMike Kravetz * for the inode. mpol_free_shared_policy is unconditionally called 9944742a35dSMike Kravetz * as part of hugetlbfs_destroy_inode. So, initialize policy here 9954742a35dSMike Kravetz * in case of a quick call to destroy. 9964742a35dSMike Kravetz * 9974742a35dSMike Kravetz * Note that the policy is initialized even if we are creating a 9984742a35dSMike Kravetz * private inode. This simplifies hugetlbfs_destroy_inode. 9994742a35dSMike Kravetz */ 10004742a35dSMike Kravetz mpol_shared_policy_init(&p->policy, NULL); 10014742a35dSMike Kravetz 100296527980SChristoph Hellwig return &p->vfs_inode; 10031da177e4SLinus Torvalds } 10041da177e4SLinus Torvalds 1005fa0d7e3dSNick Piggin static void hugetlbfs_i_callback(struct rcu_head *head) 1006fa0d7e3dSNick Piggin { 1007fa0d7e3dSNick Piggin struct inode *inode = container_of(head, struct inode, i_rcu); 1008fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 1009fa0d7e3dSNick Piggin } 1010fa0d7e3dSNick Piggin 10111da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 10121da177e4SLinus Torvalds { 101396527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 10141da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 1015fa0d7e3dSNick Piggin call_rcu(&inode->i_rcu, hugetlbfs_i_callback); 10161da177e4SLinus Torvalds } 10171da177e4SLinus Torvalds 1018f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 1019800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 1020800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 10211da177e4SLinus Torvalds .set_page_dirty = hugetlbfs_set_page_dirty, 1022290408d4SNaoya Horiguchi .migratepage = hugetlbfs_migrate_page, 102378bb9203SNaoya Horiguchi .error_remove_page = hugetlbfs_error_remove_page, 10241da177e4SLinus Torvalds }; 10251da177e4SLinus Torvalds 102696527980SChristoph Hellwig 102751cc5068SAlexey Dobriyan static void init_once(void *foo) 102896527980SChristoph Hellwig { 102996527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 103096527980SChristoph Hellwig 103196527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 103296527980SChristoph Hellwig } 103396527980SChristoph Hellwig 10344b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 103534d0640eSAl Viro .read_iter = hugetlbfs_read_iter, 10361da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 10371b061d92SChristoph Hellwig .fsync = noop_fsync, 10381da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 10396038f373SArnd Bergmann .llseek = default_llseek, 104070c3547eSMike Kravetz .fallocate = hugetlbfs_fallocate, 10411da177e4SLinus Torvalds }; 10421da177e4SLinus Torvalds 104392e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 10441da177e4SLinus Torvalds .create = hugetlbfs_create, 10451da177e4SLinus Torvalds .lookup = simple_lookup, 10461da177e4SLinus Torvalds .link = simple_link, 10471da177e4SLinus Torvalds .unlink = simple_unlink, 10481da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 10491da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 10501da177e4SLinus Torvalds .rmdir = simple_rmdir, 10511da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 10521da177e4SLinus Torvalds .rename = simple_rename, 10531da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 10541da177e4SLinus Torvalds }; 10551da177e4SLinus Torvalds 105692e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 10571da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 10581da177e4SLinus Torvalds }; 10591da177e4SLinus Torvalds 1060ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 10611da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 10621da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 10632bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 10641da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 10651da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 10664a25220dSDavid Howells .show_options = hugetlbfs_show_options, 10671da177e4SLinus Torvalds }; 10681da177e4SLinus Torvalds 10694a25220dSDavid Howells enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT }; 10707ca02d0aSMike Kravetz 10717ca02d0aSMike Kravetz /* 10727ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 10737ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 10747ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 10757ca02d0aSMike Kravetz */ 10764a25220dSDavid Howells static long 10777ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 10784a25220dSDavid Howells enum hugetlbfs_size_type val_type) 10797ca02d0aSMike Kravetz { 10807ca02d0aSMike Kravetz if (val_type == NO_SIZE) 10817ca02d0aSMike Kravetz return -1; 10827ca02d0aSMike Kravetz 10837ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 10847ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 10857ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 10867ca02d0aSMike Kravetz do_div(size_opt, 100); 10877ca02d0aSMike Kravetz } 10887ca02d0aSMike Kravetz 10897ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 10907ca02d0aSMike Kravetz return size_opt; 10917ca02d0aSMike Kravetz } 10927ca02d0aSMike Kravetz 10931da177e4SLinus Torvalds static int 10941da177e4SLinus Torvalds hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) 10951da177e4SLinus Torvalds { 1096e73a75faSRandy Dunlap char *p, *rest; 1097e73a75faSRandy Dunlap substring_t args[MAX_OPT_ARGS]; 1098e73a75faSRandy Dunlap int option; 10997ca02d0aSMike Kravetz unsigned long long max_size_opt = 0, min_size_opt = 0; 11004a25220dSDavid Howells enum hugetlbfs_size_type max_val_type = NO_SIZE, min_val_type = NO_SIZE; 11011da177e4SLinus Torvalds 11021da177e4SLinus Torvalds if (!options) 11031da177e4SLinus Torvalds return 0; 11041da177e4SLinus Torvalds 1105e73a75faSRandy Dunlap while ((p = strsep(&options, ",")) != NULL) { 1106e73a75faSRandy Dunlap int token; 1107b4c07bceSLee Schermerhorn if (!*p) 1108b4c07bceSLee Schermerhorn continue; 11091da177e4SLinus Torvalds 1110e73a75faSRandy Dunlap token = match_token(p, tokens, args); 1111e73a75faSRandy Dunlap switch (token) { 1112e73a75faSRandy Dunlap case Opt_uid: 1113e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 1114e73a75faSRandy Dunlap goto bad_val; 1115a0eb3a05SEric W. Biederman pconfig->uid = make_kuid(current_user_ns(), option); 1116a0eb3a05SEric W. Biederman if (!uid_valid(pconfig->uid)) 1117a0eb3a05SEric W. Biederman goto bad_val; 1118e73a75faSRandy Dunlap break; 1119e73a75faSRandy Dunlap 1120e73a75faSRandy Dunlap case Opt_gid: 1121e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 1122e73a75faSRandy Dunlap goto bad_val; 1123a0eb3a05SEric W. Biederman pconfig->gid = make_kgid(current_user_ns(), option); 1124a0eb3a05SEric W. Biederman if (!gid_valid(pconfig->gid)) 1125a0eb3a05SEric W. Biederman goto bad_val; 1126e73a75faSRandy Dunlap break; 1127e73a75faSRandy Dunlap 1128e73a75faSRandy Dunlap case Opt_mode: 1129e73a75faSRandy Dunlap if (match_octal(&args[0], &option)) 1130e73a75faSRandy Dunlap goto bad_val; 113175897d60SKen Chen pconfig->mode = option & 01777U; 1132e73a75faSRandy Dunlap break; 1133e73a75faSRandy Dunlap 1134e73a75faSRandy Dunlap case Opt_size: { 1135e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 1136e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 1137e73a75faSRandy Dunlap goto bad_val; 11387ca02d0aSMike Kravetz max_size_opt = memparse(args[0].from, &rest); 11397ca02d0aSMike Kravetz max_val_type = SIZE_STD; 1140a137e1ccSAndi Kleen if (*rest == '%') 11417ca02d0aSMike Kravetz max_val_type = SIZE_PERCENT; 1142e73a75faSRandy Dunlap break; 1143e73a75faSRandy Dunlap } 11441da177e4SLinus Torvalds 1145e73a75faSRandy Dunlap case Opt_nr_inodes: 1146e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 1147e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 1148e73a75faSRandy Dunlap goto bad_val; 1149e73a75faSRandy Dunlap pconfig->nr_inodes = memparse(args[0].from, &rest); 1150e73a75faSRandy Dunlap break; 1151e73a75faSRandy Dunlap 1152a137e1ccSAndi Kleen case Opt_pagesize: { 1153a137e1ccSAndi Kleen unsigned long ps; 1154a137e1ccSAndi Kleen ps = memparse(args[0].from, &rest); 1155a137e1ccSAndi Kleen pconfig->hstate = size_to_hstate(ps); 1156a137e1ccSAndi Kleen if (!pconfig->hstate) { 11579b857d26SAndrew Morton pr_err("Unsupported page size %lu MB\n", 1158a137e1ccSAndi Kleen ps >> 20); 1159a137e1ccSAndi Kleen return -EINVAL; 1160a137e1ccSAndi Kleen } 1161a137e1ccSAndi Kleen break; 1162a137e1ccSAndi Kleen } 1163a137e1ccSAndi Kleen 11647ca02d0aSMike Kravetz case Opt_min_size: { 11657ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 11667ca02d0aSMike Kravetz if (!isdigit(*args[0].from)) 11677ca02d0aSMike Kravetz goto bad_val; 11687ca02d0aSMike Kravetz min_size_opt = memparse(args[0].from, &rest); 11697ca02d0aSMike Kravetz min_val_type = SIZE_STD; 11707ca02d0aSMike Kravetz if (*rest == '%') 11717ca02d0aSMike Kravetz min_val_type = SIZE_PERCENT; 11727ca02d0aSMike Kravetz break; 11737ca02d0aSMike Kravetz } 11747ca02d0aSMike Kravetz 1175e73a75faSRandy Dunlap default: 11769b857d26SAndrew Morton pr_err("Bad mount option: \"%s\"\n", p); 1177b4c07bceSLee Schermerhorn return -EINVAL; 1178e73a75faSRandy Dunlap break; 1179e73a75faSRandy Dunlap } 11801da177e4SLinus Torvalds } 1181a137e1ccSAndi Kleen 11827ca02d0aSMike Kravetz /* 11837ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 11847ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 11857ca02d0aSMike Kravetz */ 11867ca02d0aSMike Kravetz pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, 11877ca02d0aSMike Kravetz max_size_opt, max_val_type); 11887ca02d0aSMike Kravetz pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, 11897ca02d0aSMike Kravetz min_size_opt, min_val_type); 11907ca02d0aSMike Kravetz 11917ca02d0aSMike Kravetz /* 11927ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 11937ca02d0aSMike Kravetz */ 11947ca02d0aSMike Kravetz if (max_val_type > NO_SIZE && 11957ca02d0aSMike Kravetz pconfig->min_hpages > pconfig->max_hpages) { 11967ca02d0aSMike Kravetz pr_err("minimum size can not be greater than maximum size\n"); 11977ca02d0aSMike Kravetz return -EINVAL; 1198a137e1ccSAndi Kleen } 1199a137e1ccSAndi Kleen 12001da177e4SLinus Torvalds return 0; 1201e73a75faSRandy Dunlap 1202e73a75faSRandy Dunlap bad_val: 12039b857d26SAndrew Morton pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); 1204c12ddba0SAkinobu Mita return -EINVAL; 12051da177e4SLinus Torvalds } 12061da177e4SLinus Torvalds 12071da177e4SLinus Torvalds static int 12081da177e4SLinus Torvalds hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) 12091da177e4SLinus Torvalds { 12101da177e4SLinus Torvalds int ret; 12111da177e4SLinus Torvalds struct hugetlbfs_config config; 12121da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 12131da177e4SLinus Torvalds 12147ca02d0aSMike Kravetz config.max_hpages = -1; /* No limit on size by default */ 12151da177e4SLinus Torvalds config.nr_inodes = -1; /* No limit on number of inodes by default */ 121677c70de1SDavid Howells config.uid = current_fsuid(); 121777c70de1SDavid Howells config.gid = current_fsgid(); 12181da177e4SLinus Torvalds config.mode = 0755; 1219a137e1ccSAndi Kleen config.hstate = &default_hstate; 12207ca02d0aSMike Kravetz config.min_hpages = -1; /* No default minimum size */ 12211da177e4SLinus Torvalds ret = hugetlbfs_parse_options(data, &config); 12221da177e4SLinus Torvalds if (ret) 12231da177e4SLinus Torvalds return ret; 12241da177e4SLinus Torvalds 12251da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 12261da177e4SLinus Torvalds if (!sbinfo) 12271da177e4SLinus Torvalds return -ENOMEM; 12281da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 1229a137e1ccSAndi Kleen sbinfo->hstate = config.hstate; 12301da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 12311da177e4SLinus Torvalds sbinfo->max_inodes = config.nr_inodes; 12321da177e4SLinus Torvalds sbinfo->free_inodes = config.nr_inodes; 123390481622SDavid Gibson sbinfo->spool = NULL; 12344a25220dSDavid Howells sbinfo->uid = config.uid; 12354a25220dSDavid Howells sbinfo->gid = config.gid; 12364a25220dSDavid Howells sbinfo->mode = config.mode; 12374a25220dSDavid Howells 12387ca02d0aSMike Kravetz /* 12397ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 12407ca02d0aSMike Kravetz * specified. Any needed reservations (for minimim size) are taken 12417ca02d0aSMike Kravetz * taken when the subpool is created. 12427ca02d0aSMike Kravetz */ 12437ca02d0aSMike Kravetz if (config.max_hpages != -1 || config.min_hpages != -1) { 12447ca02d0aSMike Kravetz sbinfo->spool = hugepage_new_subpool(config.hstate, 12457ca02d0aSMike Kravetz config.max_hpages, 12467ca02d0aSMike Kravetz config.min_hpages); 124790481622SDavid Gibson if (!sbinfo->spool) 124890481622SDavid Gibson goto out_free; 124990481622SDavid Gibson } 12501da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 1251a137e1ccSAndi Kleen sb->s_blocksize = huge_page_size(config.hstate); 1252a137e1ccSAndi Kleen sb->s_blocksize_bits = huge_page_shift(config.hstate); 12531da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 12541da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 12551da177e4SLinus Torvalds sb->s_time_gran = 1; 125648fde701SAl Viro sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); 125748fde701SAl Viro if (!sb->s_root) 12581da177e4SLinus Torvalds goto out_free; 12591da177e4SLinus Torvalds return 0; 12601da177e4SLinus Torvalds out_free: 126190481622SDavid Gibson kfree(sbinfo->spool); 12621da177e4SLinus Torvalds kfree(sbinfo); 12631da177e4SLinus Torvalds return -ENOMEM; 12641da177e4SLinus Torvalds } 12651da177e4SLinus Torvalds 12663c26ff6eSAl Viro static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, 12673c26ff6eSAl Viro int flags, const char *dev_name, void *data) 12681da177e4SLinus Torvalds { 12693c26ff6eSAl Viro return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); 12701da177e4SLinus Torvalds } 12711da177e4SLinus Torvalds 12721da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 12731da177e4SLinus Torvalds .name = "hugetlbfs", 12743c26ff6eSAl Viro .mount = hugetlbfs_mount, 12751da177e4SLinus Torvalds .kill_sb = kill_litter_super, 12761da177e4SLinus Torvalds }; 12771da177e4SLinus Torvalds 127842d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 12791da177e4SLinus Torvalds 1280ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 12811da177e4SLinus Torvalds { 1282a0eb3a05SEric W. Biederman kgid_t shm_group; 1283a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 1284a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 12851da177e4SLinus Torvalds } 12861da177e4SLinus Torvalds 128742d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 128842d7395fSAndi Kleen { 1289af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 129042d7395fSAndi Kleen 129142d7395fSAndi Kleen if (!h) 129242d7395fSAndi Kleen return -1; 129342d7395fSAndi Kleen return h - hstates; 129442d7395fSAndi Kleen } 129542d7395fSAndi Kleen 1296be1d2cf5SFabian Frederick static const struct dentry_operations anon_ops = { 1297118b2302SAl Viro .d_dname = simple_dname 12980df4d6e5SAl Viro }; 12990df4d6e5SAl Viro 1300af73e4d9SNaoya Horiguchi /* 1301af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 1302af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 1303af73e4d9SNaoya Horiguchi */ 1304af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 1305af73e4d9SNaoya Horiguchi vm_flags_t acctflag, struct user_struct **user, 130642d7395fSAndi Kleen int creat_flags, int page_size_log) 13071da177e4SLinus Torvalds { 130839b65252SAnatol Pomozov struct file *file = ERR_PTR(-ENOMEM); 13091da177e4SLinus Torvalds struct inode *inode; 13102c48b9c4SAl Viro struct path path; 13110df4d6e5SAl Viro struct super_block *sb; 13121da177e4SLinus Torvalds struct qstr quick_string; 131342d7395fSAndi Kleen int hstate_idx; 131442d7395fSAndi Kleen 131542d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 131642d7395fSAndi Kleen if (hstate_idx < 0) 131742d7395fSAndi Kleen return ERR_PTR(-ENODEV); 13181da177e4SLinus Torvalds 1319353d5c30SHugh Dickins *user = NULL; 132042d7395fSAndi Kleen if (!hugetlbfs_vfsmount[hstate_idx]) 13215bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 13225bc98594SAkinobu Mita 1323ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 1324353d5c30SHugh Dickins *user = current_user(); 1325353d5c30SHugh Dickins if (user_shm_lock(size, *user)) { 132621a3c273SDavid Rientjes task_lock(current); 13279b857d26SAndrew Morton pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", 132821a3c273SDavid Rientjes current->comm, current->pid); 132921a3c273SDavid Rientjes task_unlock(current); 1330353d5c30SHugh Dickins } else { 1331353d5c30SHugh Dickins *user = NULL; 13321da177e4SLinus Torvalds return ERR_PTR(-EPERM); 13332584e517SRavikiran G Thirumalai } 1334353d5c30SHugh Dickins } 13351da177e4SLinus Torvalds 13360df4d6e5SAl Viro sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb; 13379d66586fSEric W. Biederman quick_string.name = name; 13381da177e4SLinus Torvalds quick_string.len = strlen(quick_string.name); 13391da177e4SLinus Torvalds quick_string.hash = 0; 13400df4d6e5SAl Viro path.dentry = d_alloc_pseudo(sb, &quick_string); 13412c48b9c4SAl Viro if (!path.dentry) 13421da177e4SLinus Torvalds goto out_shm_unlock; 13431da177e4SLinus Torvalds 13440df4d6e5SAl Viro d_set_d_op(path.dentry, &anon_ops); 134542d7395fSAndi Kleen path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); 134639b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 13470df4d6e5SAl Viro inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0); 13481da177e4SLinus Torvalds if (!inode) 1349ce8d2cdfSDave Hansen goto out_dentry; 1350e1832f29SStephen Smalley if (creat_flags == HUGETLB_SHMFS_INODE) 1351e1832f29SStephen Smalley inode->i_flags |= S_PRIVATE; 13521da177e4SLinus Torvalds 135339b65252SAnatol Pomozov file = ERR_PTR(-ENOMEM); 1354af73e4d9SNaoya Horiguchi if (hugetlb_reserve_pages(inode, 0, 1355af73e4d9SNaoya Horiguchi size >> huge_page_shift(hstate_inode(inode)), NULL, 1356af73e4d9SNaoya Horiguchi acctflag)) 1357b45b5bd6SDavid Gibson goto out_inode; 1358b45b5bd6SDavid Gibson 13592c48b9c4SAl Viro d_instantiate(path.dentry, inode); 13601da177e4SLinus Torvalds inode->i_size = size; 13616d6b77f1SMiklos Szeredi clear_nlink(inode); 1362ce8d2cdfSDave Hansen 13632c48b9c4SAl Viro file = alloc_file(&path, FMODE_WRITE | FMODE_READ, 1364ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 136539b65252SAnatol Pomozov if (IS_ERR(file)) 1366b4d232e6SAl Viro goto out_dentry; /* inode is already attached */ 1367ce8d2cdfSDave Hansen 13681da177e4SLinus Torvalds return file; 13691da177e4SLinus Torvalds 1370b45b5bd6SDavid Gibson out_inode: 1371b45b5bd6SDavid Gibson iput(inode); 13721da177e4SLinus Torvalds out_dentry: 13732c48b9c4SAl Viro path_put(&path); 13741da177e4SLinus Torvalds out_shm_unlock: 1375353d5c30SHugh Dickins if (*user) { 1376353d5c30SHugh Dickins user_shm_unlock(size, *user); 1377353d5c30SHugh Dickins *user = NULL; 1378353d5c30SHugh Dickins } 137939b65252SAnatol Pomozov return file; 13801da177e4SLinus Torvalds } 13811da177e4SLinus Torvalds 13821da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 13831da177e4SLinus Torvalds { 138442d7395fSAndi Kleen struct hstate *h; 13851da177e4SLinus Torvalds int error; 138642d7395fSAndi Kleen int i; 13871da177e4SLinus Torvalds 1388457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 13899b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1390457c1b27SNishanth Aravamudan return -ENOTSUPP; 1391457c1b27SNishanth Aravamudan } 1392457c1b27SNishanth Aravamudan 1393d1d5e05fSHillf Danton error = -ENOMEM; 13941da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 13951da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 13965d097056SVladimir Davydov 0, SLAB_ACCOUNT, init_once); 13971da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 1398e0bf68ddSPeter Zijlstra goto out2; 13991da177e4SLinus Torvalds 14001da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 14011da177e4SLinus Torvalds if (error) 14021da177e4SLinus Torvalds goto out; 14031da177e4SLinus Torvalds 140442d7395fSAndi Kleen i = 0; 140542d7395fSAndi Kleen for_each_hstate(h) { 140642d7395fSAndi Kleen char buf[50]; 140742d7395fSAndi Kleen unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); 14081da177e4SLinus Torvalds 140942d7395fSAndi Kleen snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); 141042d7395fSAndi Kleen hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, 141142d7395fSAndi Kleen buf); 141242d7395fSAndi Kleen 141342d7395fSAndi Kleen if (IS_ERR(hugetlbfs_vfsmount[i])) { 14149b857d26SAndrew Morton pr_err("Cannot mount internal hugetlbfs for " 141542d7395fSAndi Kleen "page size %uK", ps_kb); 141642d7395fSAndi Kleen error = PTR_ERR(hugetlbfs_vfsmount[i]); 141742d7395fSAndi Kleen hugetlbfs_vfsmount[i] = NULL; 14181da177e4SLinus Torvalds } 141942d7395fSAndi Kleen i++; 142042d7395fSAndi Kleen } 142142d7395fSAndi Kleen /* Non default hstates are optional */ 142242d7395fSAndi Kleen if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) 142342d7395fSAndi Kleen return 0; 14241da177e4SLinus Torvalds 14251da177e4SLinus Torvalds out: 14261da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 1427e0bf68ddSPeter Zijlstra out2: 14281da177e4SLinus Torvalds return error; 14291da177e4SLinus Torvalds } 14303e89e1c5SPaul Gortmaker fs_initcall(init_hugetlbfs_fs) 1431