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 11163489f8eSMike Kravetz /* 11263489f8eSMike Kravetz * Mask used when checking the page offset value passed in via system 11363489f8eSMike Kravetz * calls. This value will be converted to a loff_t which is signed. 11463489f8eSMike Kravetz * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the 11563489f8eSMike Kravetz * value. The extra bit (- 1 in the shift value) is to take the sign 11663489f8eSMike Kravetz * bit into account. 11763489f8eSMike Kravetz */ 11863489f8eSMike Kravetz #define PGOFF_LOFFT_MAX \ 11963489f8eSMike Kravetz (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1))) 12063489f8eSMike Kravetz 1211da177e4SLinus Torvalds static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) 1221da177e4SLinus Torvalds { 123496ad9aaSAl Viro struct inode *inode = file_inode(file); 1241da177e4SLinus Torvalds loff_t len, vma_len; 1251da177e4SLinus Torvalds int ret; 126a5516438SAndi Kleen struct hstate *h = hstate_file(file); 1271da177e4SLinus Torvalds 12868589bc3SHugh Dickins /* 129dec4ad86SDavid Gibson * vma address alignment (but not the pgoff alignment) has 130dec4ad86SDavid Gibson * already been checked by prepare_hugepage_range. If you add 131dec4ad86SDavid Gibson * any error returns here, do so after setting VM_HUGETLB, so 132dec4ad86SDavid Gibson * is_vm_hugetlb_page tests below unmap_region go the right 133dec4ad86SDavid Gibson * way when do_mmap_pgoff unwinds (may be important on powerpc 134dec4ad86SDavid Gibson * and ia64). 13568589bc3SHugh Dickins */ 136a2fce914SNaoya Horiguchi vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; 13768589bc3SHugh Dickins vma->vm_ops = &hugetlb_vm_ops; 1381da177e4SLinus Torvalds 139045c7a3fSMike Kravetz /* 14063489f8eSMike Kravetz * page based offset in vm_pgoff could be sufficiently large to 1415df63c2aSMike Kravetz * overflow a loff_t when converted to byte offset. This can 1425df63c2aSMike Kravetz * only happen on architectures where sizeof(loff_t) == 1435df63c2aSMike Kravetz * sizeof(unsigned long). So, only check in those instances. 144045c7a3fSMike Kravetz */ 1455df63c2aSMike Kravetz if (sizeof(unsigned long) == sizeof(loff_t)) { 14663489f8eSMike Kravetz if (vma->vm_pgoff & PGOFF_LOFFT_MAX) 147045c7a3fSMike Kravetz return -EINVAL; 1485df63c2aSMike Kravetz } 149045c7a3fSMike Kravetz 15063489f8eSMike Kravetz /* must be huge page aligned */ 1512b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 152dec4ad86SDavid Gibson return -EINVAL; 153dec4ad86SDavid Gibson 1541da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 155045c7a3fSMike Kravetz len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 156045c7a3fSMike Kravetz /* check for overflow */ 157045c7a3fSMike Kravetz if (len < vma_len) 158045c7a3fSMike Kravetz return -EINVAL; 1591da177e4SLinus Torvalds 1605955102cSAl Viro inode_lock(inode); 1611da177e4SLinus Torvalds file_accessed(file); 1621da177e4SLinus Torvalds 1631da177e4SLinus Torvalds ret = -ENOMEM; 164a1e78772SMel Gorman if (hugetlb_reserve_pages(inode, 165a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1665a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1675a6fe125SMel Gorman vma->vm_flags)) 168b45b5bd6SDavid Gibson goto out; 169b45b5bd6SDavid Gibson 1704c887265SAdam Litke ret = 0; 171b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 172045c7a3fSMike Kravetz i_size_write(inode, len); 1731da177e4SLinus Torvalds out: 1745955102cSAl Viro inode_unlock(inode); 1751da177e4SLinus Torvalds 1761da177e4SLinus Torvalds return ret; 1771da177e4SLinus Torvalds } 1781da177e4SLinus Torvalds 1791da177e4SLinus Torvalds /* 180508034a3SHugh Dickins * Called under down_write(mmap_sem). 1811da177e4SLinus Torvalds */ 1821da177e4SLinus Torvalds 183d2ba27e8SAdrian Bunk #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 1841da177e4SLinus Torvalds static unsigned long 1851da177e4SLinus Torvalds hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 1861da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 1871da177e4SLinus Torvalds { 1881da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1891da177e4SLinus Torvalds struct vm_area_struct *vma; 190a5516438SAndi Kleen struct hstate *h = hstate_file(file); 19108659355SMichel Lespinasse struct vm_unmapped_area_info info; 1921da177e4SLinus Torvalds 193a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 1941da177e4SLinus Torvalds return -EINVAL; 1951da177e4SLinus Torvalds if (len > TASK_SIZE) 1961da177e4SLinus Torvalds return -ENOMEM; 1971da177e4SLinus Torvalds 198036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 199a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 200036e0856SBenjamin Herrenschmidt return -EINVAL; 201036e0856SBenjamin Herrenschmidt return addr; 202036e0856SBenjamin Herrenschmidt } 203036e0856SBenjamin Herrenschmidt 2041da177e4SLinus Torvalds if (addr) { 205a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 2061da177e4SLinus Torvalds vma = find_vma(mm, addr); 2071da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 2081be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma))) 2091da177e4SLinus Torvalds return addr; 2101da177e4SLinus Torvalds } 2111da177e4SLinus Torvalds 21208659355SMichel Lespinasse info.flags = 0; 21308659355SMichel Lespinasse info.length = len; 21408659355SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 21508659355SMichel Lespinasse info.high_limit = TASK_SIZE; 21608659355SMichel Lespinasse info.align_mask = PAGE_MASK & ~huge_page_mask(h); 21708659355SMichel Lespinasse info.align_offset = 0; 21808659355SMichel Lespinasse return vm_unmapped_area(&info); 2191da177e4SLinus Torvalds } 2201da177e4SLinus Torvalds #endif 2211da177e4SLinus Torvalds 22234d0640eSAl Viro static size_t 223e63e1e5aSBadari Pulavarty hugetlbfs_read_actor(struct page *page, unsigned long offset, 22434d0640eSAl Viro struct iov_iter *to, unsigned long size) 225e63e1e5aSBadari Pulavarty { 22634d0640eSAl Viro size_t copied = 0; 227e63e1e5aSBadari Pulavarty int i, chunksize; 228e63e1e5aSBadari Pulavarty 229e63e1e5aSBadari Pulavarty /* Find which 4k chunk and offset with in that chunk */ 23009cbfeafSKirill A. Shutemov i = offset >> PAGE_SHIFT; 23109cbfeafSKirill A. Shutemov offset = offset & ~PAGE_MASK; 232e63e1e5aSBadari Pulavarty 233e63e1e5aSBadari Pulavarty while (size) { 23434d0640eSAl Viro size_t n; 23509cbfeafSKirill A. Shutemov chunksize = PAGE_SIZE; 236e63e1e5aSBadari Pulavarty if (offset) 237e63e1e5aSBadari Pulavarty chunksize -= offset; 238e63e1e5aSBadari Pulavarty if (chunksize > size) 239e63e1e5aSBadari Pulavarty chunksize = size; 24034d0640eSAl Viro n = copy_page_to_iter(&page[i], offset, chunksize, to); 24134d0640eSAl Viro copied += n; 24234d0640eSAl Viro if (n != chunksize) 24334d0640eSAl Viro return copied; 244e63e1e5aSBadari Pulavarty offset = 0; 245e63e1e5aSBadari Pulavarty size -= chunksize; 246e63e1e5aSBadari Pulavarty i++; 247e63e1e5aSBadari Pulavarty } 24834d0640eSAl Viro return copied; 249e63e1e5aSBadari Pulavarty } 250e63e1e5aSBadari Pulavarty 251e63e1e5aSBadari Pulavarty /* 252e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 253e63e1e5aSBadari Pulavarty * data. Its *very* similar to do_generic_mapping_read(), we can't use that 254ea1754a0SKirill A. Shutemov * since it has PAGE_SIZE assumptions. 255e63e1e5aSBadari Pulavarty */ 25634d0640eSAl Viro static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) 257e63e1e5aSBadari Pulavarty { 25834d0640eSAl Viro struct file *file = iocb->ki_filp; 25934d0640eSAl Viro struct hstate *h = hstate_file(file); 26034d0640eSAl Viro struct address_space *mapping = file->f_mapping; 261e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 26234d0640eSAl Viro unsigned long index = iocb->ki_pos >> huge_page_shift(h); 26334d0640eSAl Viro unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); 264e63e1e5aSBadari Pulavarty unsigned long end_index; 265e63e1e5aSBadari Pulavarty loff_t isize; 266e63e1e5aSBadari Pulavarty ssize_t retval = 0; 267e63e1e5aSBadari Pulavarty 26834d0640eSAl Viro while (iov_iter_count(to)) { 269e63e1e5aSBadari Pulavarty struct page *page; 27034d0640eSAl Viro size_t nr, copied; 271e63e1e5aSBadari Pulavarty 272e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 273a5516438SAndi Kleen nr = huge_page_size(h); 274a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 275a05b0855SAneesh Kumar K.V if (!isize) 27634d0640eSAl Viro break; 277a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 278e63e1e5aSBadari Pulavarty if (index > end_index) 27934d0640eSAl Viro break; 28034d0640eSAl Viro if (index == end_index) { 281a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 282a05b0855SAneesh Kumar K.V if (nr <= offset) 28334d0640eSAl Viro break; 284e63e1e5aSBadari Pulavarty } 285e63e1e5aSBadari Pulavarty nr = nr - offset; 286e63e1e5aSBadari Pulavarty 287e63e1e5aSBadari Pulavarty /* Find the page */ 288a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 289e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 290e63e1e5aSBadari Pulavarty /* 291e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 292e63e1e5aSBadari Pulavarty * length of the hole or request. 293e63e1e5aSBadari Pulavarty */ 29434d0640eSAl Viro copied = iov_iter_zero(nr, to); 295e63e1e5aSBadari Pulavarty } else { 296a05b0855SAneesh Kumar K.V unlock_page(page); 297a05b0855SAneesh Kumar K.V 298e63e1e5aSBadari Pulavarty /* 299e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 300e63e1e5aSBadari Pulavarty */ 30134d0640eSAl Viro copied = hugetlbfs_read_actor(page, offset, to, nr); 30209cbfeafSKirill A. Shutemov put_page(page); 303e63e1e5aSBadari Pulavarty } 30434d0640eSAl Viro offset += copied; 30534d0640eSAl Viro retval += copied; 30634d0640eSAl Viro if (copied != nr && iov_iter_count(to)) { 30734d0640eSAl Viro if (!retval) 30834d0640eSAl Viro retval = -EFAULT; 309e63e1e5aSBadari Pulavarty break; 310e63e1e5aSBadari Pulavarty } 31134d0640eSAl Viro index += offset >> huge_page_shift(h); 31234d0640eSAl Viro offset &= ~huge_page_mask(h); 31334d0640eSAl Viro } 31434d0640eSAl Viro iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; 315e63e1e5aSBadari Pulavarty return retval; 316e63e1e5aSBadari Pulavarty } 317e63e1e5aSBadari Pulavarty 318800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 319800d15a5SNick Piggin struct address_space *mapping, 320800d15a5SNick Piggin loff_t pos, unsigned len, unsigned flags, 321800d15a5SNick Piggin struct page **pagep, void **fsdata) 3221da177e4SLinus Torvalds { 3231da177e4SLinus Torvalds return -EINVAL; 3241da177e4SLinus Torvalds } 3251da177e4SLinus Torvalds 326800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 327800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 328800d15a5SNick Piggin struct page *page, void *fsdata) 3291da177e4SLinus Torvalds { 330800d15a5SNick Piggin BUG(); 3311da177e4SLinus Torvalds return -EINVAL; 3321da177e4SLinus Torvalds } 3331da177e4SLinus Torvalds 334b5cec28dSMike Kravetz static void remove_huge_page(struct page *page) 3351da177e4SLinus Torvalds { 336b9ea2515SKonstantin Khlebnikov ClearPageDirty(page); 3371da177e4SLinus Torvalds ClearPageUptodate(page); 338bd65cb86SMinchan Kim delete_from_page_cache(page); 3391da177e4SLinus Torvalds } 3401da177e4SLinus Torvalds 3414aae8d1cSMike Kravetz static void 342f808c13fSDavidlohr Bueso hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end) 3434aae8d1cSMike Kravetz { 3444aae8d1cSMike Kravetz struct vm_area_struct *vma; 3454aae8d1cSMike Kravetz 3464aae8d1cSMike Kravetz /* 3474aae8d1cSMike Kravetz * end == 0 indicates that the entire range after 3484aae8d1cSMike Kravetz * start should be unmapped. 3494aae8d1cSMike Kravetz */ 3504aae8d1cSMike Kravetz vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { 3514aae8d1cSMike Kravetz unsigned long v_offset; 3524aae8d1cSMike Kravetz unsigned long v_end; 3534aae8d1cSMike Kravetz 3544aae8d1cSMike Kravetz /* 3554aae8d1cSMike Kravetz * Can the expression below overflow on 32-bit arches? 3564aae8d1cSMike Kravetz * No, because the interval tree returns us only those vmas 3574aae8d1cSMike Kravetz * which overlap the truncated area starting at pgoff, 3584aae8d1cSMike Kravetz * and no vma on a 32-bit arch can span beyond the 4GB. 3594aae8d1cSMike Kravetz */ 3604aae8d1cSMike Kravetz if (vma->vm_pgoff < start) 3614aae8d1cSMike Kravetz v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; 3624aae8d1cSMike Kravetz else 3634aae8d1cSMike Kravetz v_offset = 0; 3644aae8d1cSMike Kravetz 3654aae8d1cSMike Kravetz if (!end) 3664aae8d1cSMike Kravetz v_end = vma->vm_end; 3674aae8d1cSMike Kravetz else { 3684aae8d1cSMike Kravetz v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) 3694aae8d1cSMike Kravetz + vma->vm_start; 3704aae8d1cSMike Kravetz if (v_end > vma->vm_end) 3714aae8d1cSMike Kravetz v_end = vma->vm_end; 3724aae8d1cSMike Kravetz } 3734aae8d1cSMike Kravetz 3744aae8d1cSMike Kravetz unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, 3754aae8d1cSMike Kravetz NULL); 3764aae8d1cSMike Kravetz } 3774aae8d1cSMike Kravetz } 378b5cec28dSMike Kravetz 379b5cec28dSMike Kravetz /* 380b5cec28dSMike Kravetz * remove_inode_hugepages handles two distinct cases: truncation and hole 381b5cec28dSMike Kravetz * punch. There are subtle differences in operation for each case. 3824aae8d1cSMike Kravetz * 383b5cec28dSMike Kravetz * truncation is indicated by end of range being LLONG_MAX 384b5cec28dSMike Kravetz * In this case, we first scan the range and release found pages. 385b5cec28dSMike Kravetz * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv 386c86aa7bbSMike Kravetz * maps and global counts. 387b5cec28dSMike Kravetz * hole punch is indicated if end is not LLONG_MAX 388b5cec28dSMike Kravetz * In the hole punch case we scan the range and release found pages. 389b5cec28dSMike Kravetz * Only when releasing a page is the associated region/reserv map 390b5cec28dSMike Kravetz * deleted. The region/reserv map for ranges without associated 391c86aa7bbSMike Kravetz * pages are not modified. 392c86aa7bbSMike Kravetz * 393c86aa7bbSMike Kravetz * Callers of this routine must hold the i_mmap_rwsem in write mode to prevent 394c86aa7bbSMike Kravetz * races with page faults. 395c86aa7bbSMike Kravetz * 396b5cec28dSMike Kravetz * Note: If the passed end of range value is beyond the end of file, but 397b5cec28dSMike Kravetz * not LLONG_MAX this routine still performs a hole punch operation. 398b5cec28dSMike Kravetz */ 399b5cec28dSMike Kravetz static void remove_inode_hugepages(struct inode *inode, loff_t lstart, 400b5cec28dSMike Kravetz loff_t lend) 4011da177e4SLinus Torvalds { 402a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 403b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 404a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 405b5cec28dSMike Kravetz const pgoff_t end = lend >> huge_page_shift(h); 406b5cec28dSMike Kravetz struct vm_area_struct pseudo_vma; 4071da177e4SLinus Torvalds struct pagevec pvec; 408d72dc8a2SJan Kara pgoff_t next, index; 409a43a8c39SChen, Kenneth W int i, freed = 0; 410b5cec28dSMike Kravetz bool truncate_op = (lend == LLONG_MAX); 4111da177e4SLinus Torvalds 4122c4541e2SKirill A. Shutemov vma_init(&pseudo_vma, current->mm); 413b5cec28dSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 41486679820SMel Gorman pagevec_init(&pvec); 4151da177e4SLinus Torvalds next = start; 416b5cec28dSMike Kravetz while (next < end) { 417b5cec28dSMike Kravetz /* 4181817889eSMike Kravetz * When no more pages are found, we are done. 419b5cec28dSMike Kravetz */ 420397162ffSJan Kara if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1)) 4211da177e4SLinus Torvalds break; 4221da177e4SLinus Torvalds 4231da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); ++i) { 4241da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 425b5cec28dSMike Kravetz 426d72dc8a2SJan Kara index = page->index; 427b5cec28dSMike Kravetz /* 428c86aa7bbSMike Kravetz * A mapped page is impossible as callers should unmap 429c86aa7bbSMike Kravetz * all references before calling. And, i_mmap_rwsem 430c86aa7bbSMike Kravetz * prevents the creation of additional mappings. 431b5cec28dSMike Kravetz */ 432c86aa7bbSMike Kravetz VM_BUG_ON(page_mapped(page)); 4334aae8d1cSMike Kravetz 4344aae8d1cSMike Kravetz lock_page(page); 4354aae8d1cSMike Kravetz /* 4364aae8d1cSMike Kravetz * We must free the huge page and remove from page 4374aae8d1cSMike Kravetz * cache (remove_huge_page) BEFORE removing the 4384aae8d1cSMike Kravetz * region/reserve map (hugetlb_unreserve_pages). In 4394aae8d1cSMike Kravetz * rare out of memory conditions, removal of the 44072e2936cSzhong jiang * region/reserve map could fail. Correspondingly, 44172e2936cSzhong jiang * the subpool and global reserve usage count can need 44272e2936cSzhong jiang * to be adjusted. 4434aae8d1cSMike Kravetz */ 44472e2936cSzhong jiang VM_BUG_ON(PagePrivate(page)); 445b5cec28dSMike Kravetz remove_huge_page(page); 446b5cec28dSMike Kravetz freed++; 447b5cec28dSMike Kravetz if (!truncate_op) { 4484aae8d1cSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, 449d72dc8a2SJan Kara index, index + 1, 1))) 45072e2936cSzhong jiang hugetlb_fix_reserve_counts(inode); 451b5cec28dSMike Kravetz } 452b5cec28dSMike Kravetz 4531da177e4SLinus Torvalds unlock_page(page); 4541da177e4SLinus Torvalds } 4551da177e4SLinus Torvalds huge_pagevec_release(&pvec); 4561817889eSMike Kravetz cond_resched(); 4571da177e4SLinus Torvalds } 458b5cec28dSMike Kravetz 459b5cec28dSMike Kravetz if (truncate_op) 460b5cec28dSMike Kravetz (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); 4611da177e4SLinus Torvalds } 4621da177e4SLinus Torvalds 4632bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 4641da177e4SLinus Torvalds { 465c86aa7bbSMike Kravetz struct address_space *mapping = inode->i_mapping; 4669119a41eSJoonsoo Kim struct resv_map *resv_map; 4679119a41eSJoonsoo Kim 468c86aa7bbSMike Kravetz /* 469c86aa7bbSMike Kravetz * The vfs layer guarantees that there are no other users of this 470c86aa7bbSMike Kravetz * inode. Therefore, it would be safe to call remove_inode_hugepages 471c86aa7bbSMike Kravetz * without holding i_mmap_rwsem. We acquire and hold here to be 472c86aa7bbSMike Kravetz * consistent with other callers. Since there will be no contention 473c86aa7bbSMike Kravetz * on the semaphore, overhead is negligible. 474c86aa7bbSMike Kravetz */ 475c86aa7bbSMike Kravetz i_mmap_lock_write(mapping); 476b5cec28dSMike Kravetz remove_inode_hugepages(inode, 0, LLONG_MAX); 477c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 478c86aa7bbSMike Kravetz 4799119a41eSJoonsoo Kim resv_map = (struct resv_map *)inode->i_mapping->private_data; 4809119a41eSJoonsoo Kim /* root inode doesn't have the resv_map, so we should check it */ 4819119a41eSJoonsoo Kim if (resv_map) 4829119a41eSJoonsoo Kim resv_map_release(&resv_map->refs); 483dbd5768fSJan Kara clear_inode(inode); 484149f4211SChristoph Hellwig } 485149f4211SChristoph Hellwig 4861da177e4SLinus Torvalds static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) 4871da177e4SLinus Torvalds { 488856fc295SHugh Dickins pgoff_t pgoff; 4891da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 490a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 4911da177e4SLinus Torvalds 492a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 493856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 4941da177e4SLinus Torvalds 4957aa91e10SKen Chen i_size_write(inode, offset); 49683cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 497f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 4981bfad99aSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); 499b5cec28dSMike Kravetz remove_inode_hugepages(inode, offset, LLONG_MAX); 500c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 5011da177e4SLinus Torvalds return 0; 5021da177e4SLinus Torvalds } 5031da177e4SLinus Torvalds 50470c3547eSMike Kravetz static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) 50570c3547eSMike Kravetz { 50670c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 50770c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 50870c3547eSMike Kravetz loff_t hole_start, hole_end; 50970c3547eSMike Kravetz 51070c3547eSMike Kravetz /* 51170c3547eSMike Kravetz * For hole punch round up the beginning offset of the hole and 51270c3547eSMike Kravetz * round down the end. 51370c3547eSMike Kravetz */ 51470c3547eSMike Kravetz hole_start = round_up(offset, hpage_size); 51570c3547eSMike Kravetz hole_end = round_down(offset + len, hpage_size); 51670c3547eSMike Kravetz 51770c3547eSMike Kravetz if (hole_end > hole_start) { 51870c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 519ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 52070c3547eSMike Kravetz 5215955102cSAl Viro inode_lock(inode); 522ff62a342SMarc-André Lureau 523ff62a342SMarc-André Lureau /* protected by i_mutex */ 524ff62a342SMarc-André Lureau if (info->seals & F_SEAL_WRITE) { 525ff62a342SMarc-André Lureau inode_unlock(inode); 526ff62a342SMarc-André Lureau return -EPERM; 527ff62a342SMarc-André Lureau } 528ff62a342SMarc-André Lureau 52970c3547eSMike Kravetz i_mmap_lock_write(mapping); 530f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 53170c3547eSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 53270c3547eSMike Kravetz hole_start >> PAGE_SHIFT, 53370c3547eSMike Kravetz hole_end >> PAGE_SHIFT); 53470c3547eSMike Kravetz remove_inode_hugepages(inode, hole_start, hole_end); 535c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 5365955102cSAl Viro inode_unlock(inode); 53770c3547eSMike Kravetz } 53870c3547eSMike Kravetz 53970c3547eSMike Kravetz return 0; 54070c3547eSMike Kravetz } 54170c3547eSMike Kravetz 54270c3547eSMike Kravetz static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, 54370c3547eSMike Kravetz loff_t len) 54470c3547eSMike Kravetz { 54570c3547eSMike Kravetz struct inode *inode = file_inode(file); 546ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 54770c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 54870c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 54970c3547eSMike Kravetz struct vm_area_struct pseudo_vma; 55070c3547eSMike Kravetz struct mm_struct *mm = current->mm; 55170c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 55270c3547eSMike Kravetz unsigned long hpage_shift = huge_page_shift(h); 55370c3547eSMike Kravetz pgoff_t start, index, end; 55470c3547eSMike Kravetz int error; 55570c3547eSMike Kravetz u32 hash; 55670c3547eSMike Kravetz 55770c3547eSMike Kravetz if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 55870c3547eSMike Kravetz return -EOPNOTSUPP; 55970c3547eSMike Kravetz 56070c3547eSMike Kravetz if (mode & FALLOC_FL_PUNCH_HOLE) 56170c3547eSMike Kravetz return hugetlbfs_punch_hole(inode, offset, len); 56270c3547eSMike Kravetz 56370c3547eSMike Kravetz /* 56470c3547eSMike Kravetz * Default preallocate case. 56570c3547eSMike Kravetz * For this range, start is rounded down and end is rounded up 56670c3547eSMike Kravetz * as well as being converted to page offsets. 56770c3547eSMike Kravetz */ 56870c3547eSMike Kravetz start = offset >> hpage_shift; 56970c3547eSMike Kravetz end = (offset + len + hpage_size - 1) >> hpage_shift; 57070c3547eSMike Kravetz 5715955102cSAl Viro inode_lock(inode); 57270c3547eSMike Kravetz 57370c3547eSMike Kravetz /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ 57470c3547eSMike Kravetz error = inode_newsize_ok(inode, offset + len); 57570c3547eSMike Kravetz if (error) 57670c3547eSMike Kravetz goto out; 57770c3547eSMike Kravetz 578ff62a342SMarc-André Lureau if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { 579ff62a342SMarc-André Lureau error = -EPERM; 580ff62a342SMarc-André Lureau goto out; 581ff62a342SMarc-André Lureau } 582ff62a342SMarc-André Lureau 58370c3547eSMike Kravetz /* 58470c3547eSMike Kravetz * Initialize a pseudo vma as this is required by the huge page 58570c3547eSMike Kravetz * allocation routines. If NUMA is configured, use page index 58670c3547eSMike Kravetz * as input to create an allocation policy. 58770c3547eSMike Kravetz */ 5882c4541e2SKirill A. Shutemov vma_init(&pseudo_vma, mm); 58970c3547eSMike Kravetz pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 59070c3547eSMike Kravetz pseudo_vma.vm_file = file; 59170c3547eSMike Kravetz 59270c3547eSMike Kravetz for (index = start; index < end; index++) { 59370c3547eSMike Kravetz /* 59470c3547eSMike Kravetz * This is supposed to be the vaddr where the page is being 59570c3547eSMike Kravetz * faulted in, but we have no vaddr here. 59670c3547eSMike Kravetz */ 59770c3547eSMike Kravetz struct page *page; 59870c3547eSMike Kravetz unsigned long addr; 59970c3547eSMike Kravetz int avoid_reserve = 0; 60070c3547eSMike Kravetz 60170c3547eSMike Kravetz cond_resched(); 60270c3547eSMike Kravetz 60370c3547eSMike Kravetz /* 60470c3547eSMike Kravetz * fallocate(2) manpage permits EINTR; we may have been 60570c3547eSMike Kravetz * interrupted because we are using up too much memory. 60670c3547eSMike Kravetz */ 60770c3547eSMike Kravetz if (signal_pending(current)) { 60870c3547eSMike Kravetz error = -EINTR; 60970c3547eSMike Kravetz break; 61070c3547eSMike Kravetz } 61170c3547eSMike Kravetz 61270c3547eSMike Kravetz /* Set numa allocation policy based on index */ 61370c3547eSMike Kravetz hugetlb_set_vma_policy(&pseudo_vma, inode, index); 61470c3547eSMike Kravetz 61570c3547eSMike Kravetz /* addr is the offset within the file (zero based) */ 61670c3547eSMike Kravetz addr = index * hpage_size; 61770c3547eSMike Kravetz 618c86aa7bbSMike Kravetz /* 619c86aa7bbSMike Kravetz * fault mutex taken here, protects against fault path 620c86aa7bbSMike Kravetz * and hole punch. inode_lock previously taken protects 621c86aa7bbSMike Kravetz * against truncation. 622c86aa7bbSMike Kravetz */ 62370c3547eSMike Kravetz hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping, 62470c3547eSMike Kravetz index, addr); 62570c3547eSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 62670c3547eSMike Kravetz 62770c3547eSMike Kravetz /* See if already present in mapping to avoid alloc/free */ 62870c3547eSMike Kravetz page = find_get_page(mapping, index); 62970c3547eSMike Kravetz if (page) { 63070c3547eSMike Kravetz put_page(page); 63170c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 63270c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 63370c3547eSMike Kravetz continue; 63470c3547eSMike Kravetz } 63570c3547eSMike Kravetz 63670c3547eSMike Kravetz /* Allocate page and add to page cache */ 63770c3547eSMike Kravetz page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); 63870c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 63970c3547eSMike Kravetz if (IS_ERR(page)) { 64070c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 64170c3547eSMike Kravetz error = PTR_ERR(page); 64270c3547eSMike Kravetz goto out; 64370c3547eSMike Kravetz } 64470c3547eSMike Kravetz clear_huge_page(page, addr, pages_per_huge_page(h)); 64570c3547eSMike Kravetz __SetPageUptodate(page); 64670c3547eSMike Kravetz error = huge_add_to_page_cache(page, mapping, index); 64770c3547eSMike Kravetz if (unlikely(error)) { 64870c3547eSMike Kravetz put_page(page); 64970c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 65070c3547eSMike Kravetz goto out; 65170c3547eSMike Kravetz } 65270c3547eSMike Kravetz 65370c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 65470c3547eSMike Kravetz 65570c3547eSMike Kravetz /* 65670c3547eSMike Kravetz * unlock_page because locked by add_to_page_cache() 65772639e6dSNadav Amit * page_put due to reference from alloc_huge_page() 65870c3547eSMike Kravetz */ 65970c3547eSMike Kravetz unlock_page(page); 66072639e6dSNadav Amit put_page(page); 66170c3547eSMike Kravetz } 66270c3547eSMike Kravetz 66370c3547eSMike Kravetz if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) 66470c3547eSMike Kravetz i_size_write(inode, offset + len); 665078cd827SDeepa Dinamani inode->i_ctime = current_time(inode); 66670c3547eSMike Kravetz out: 6675955102cSAl Viro inode_unlock(inode); 66870c3547eSMike Kravetz return error; 66970c3547eSMike Kravetz } 67070c3547eSMike Kravetz 6711da177e4SLinus Torvalds static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) 6721da177e4SLinus Torvalds { 6732b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 674a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 6751da177e4SLinus Torvalds int error; 6761da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 677ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 6781da177e4SLinus Torvalds 6791da177e4SLinus Torvalds BUG_ON(!inode); 6801da177e4SLinus Torvalds 68131051c85SJan Kara error = setattr_prepare(dentry, attr); 6821da177e4SLinus Torvalds if (error) 6831025774cSChristoph Hellwig return error; 6841da177e4SLinus Torvalds 6851da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 686ff62a342SMarc-André Lureau loff_t oldsize = inode->i_size; 687ff62a342SMarc-André Lureau loff_t newsize = attr->ia_size; 688ff62a342SMarc-André Lureau 689ff62a342SMarc-André Lureau if (newsize & ~huge_page_mask(h)) 6901025774cSChristoph Hellwig return -EINVAL; 691ff62a342SMarc-André Lureau /* protected by i_mutex */ 692ff62a342SMarc-André Lureau if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || 693ff62a342SMarc-André Lureau (newsize > oldsize && (info->seals & F_SEAL_GROW))) 694ff62a342SMarc-André Lureau return -EPERM; 695ff62a342SMarc-André Lureau error = hugetlb_vmtruncate(inode, newsize); 6961da177e4SLinus Torvalds if (error) 6971da177e4SLinus Torvalds return error; 6981da177e4SLinus Torvalds } 6991da177e4SLinus Torvalds 7001025774cSChristoph Hellwig setattr_copy(inode, attr); 7011025774cSChristoph Hellwig mark_inode_dirty(inode); 7021025774cSChristoph Hellwig return 0; 7031025774cSChristoph Hellwig } 7041025774cSChristoph Hellwig 7057d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 7067d54fa64SAl Viro struct hugetlbfs_config *config) 7071da177e4SLinus Torvalds { 7081da177e4SLinus Torvalds struct inode *inode; 7091da177e4SLinus Torvalds 7101da177e4SLinus Torvalds inode = new_inode(sb); 7111da177e4SLinus Torvalds if (inode) { 71285fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 7137d54fa64SAl Viro inode->i_mode = S_IFDIR | config->mode; 7147d54fa64SAl Viro inode->i_uid = config->uid; 7157d54fa64SAl Viro inode->i_gid = config->gid; 716078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 7177d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 7187d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 7197d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 7207d54fa64SAl Viro inc_nlink(inode); 72165ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 7227d54fa64SAl Viro } 7237d54fa64SAl Viro return inode; 7247d54fa64SAl Viro } 7257d54fa64SAl Viro 726b610ded7SMichal Hocko /* 727c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 728b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 72988f306b6SKirill A. Shutemov * annotation because huge_pmd_share() does an allocation under hugetlb's 730c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 731b610ded7SMichal Hocko */ 732c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 733b610ded7SMichal Hocko 7347d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 7357d54fa64SAl Viro struct inode *dir, 73618df2252SAl Viro umode_t mode, dev_t dev) 7377d54fa64SAl Viro { 7387d54fa64SAl Viro struct inode *inode; 7399119a41eSJoonsoo Kim struct resv_map *resv_map; 7409119a41eSJoonsoo Kim 7419119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 7429119a41eSJoonsoo Kim if (!resv_map) 7439119a41eSJoonsoo Kim return NULL; 7447d54fa64SAl Viro 7457d54fa64SAl Viro inode = new_inode(sb); 7467d54fa64SAl Viro if (inode) { 747ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 748ff62a342SMarc-André Lureau 7497d54fa64SAl Viro inode->i_ino = get_next_ino(); 7507d54fa64SAl Viro inode_init_owner(inode, dir, mode); 751c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 752c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 7531da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 754078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 7559119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 756ff62a342SMarc-André Lureau info->seals = F_SEAL_SEAL; 7571da177e4SLinus Torvalds switch (mode & S_IFMT) { 7581da177e4SLinus Torvalds default: 7591da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 7601da177e4SLinus Torvalds break; 7611da177e4SLinus Torvalds case S_IFREG: 7621da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 7631da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 7641da177e4SLinus Torvalds break; 7651da177e4SLinus Torvalds case S_IFDIR: 7661da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 7671da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 7681da177e4SLinus Torvalds 7691da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 770d8c76e6fSDave Hansen inc_nlink(inode); 7711da177e4SLinus Torvalds break; 7721da177e4SLinus Torvalds case S_IFLNK: 7731da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 77421fc61c7SAl Viro inode_nohighmem(inode); 7751da177e4SLinus Torvalds break; 7761da177e4SLinus Torvalds } 777e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 7789119a41eSJoonsoo Kim } else 7799119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 7809119a41eSJoonsoo Kim 7811da177e4SLinus Torvalds return inode; 7821da177e4SLinus Torvalds } 7831da177e4SLinus Torvalds 7841da177e4SLinus Torvalds /* 7851da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 7861da177e4SLinus Torvalds */ 7871da177e4SLinus Torvalds static int hugetlbfs_mknod(struct inode *dir, 7881a67aafbSAl Viro struct dentry *dentry, umode_t mode, dev_t dev) 7891da177e4SLinus Torvalds { 7901da177e4SLinus Torvalds struct inode *inode; 7911da177e4SLinus Torvalds int error = -ENOSPC; 7921da177e4SLinus Torvalds 7937d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 7941da177e4SLinus Torvalds if (inode) { 795078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 7961da177e4SLinus Torvalds d_instantiate(dentry, inode); 7971da177e4SLinus Torvalds dget(dentry); /* Extra count - pin the dentry in core */ 7981da177e4SLinus Torvalds error = 0; 7991da177e4SLinus Torvalds } 8001da177e4SLinus Torvalds return error; 8011da177e4SLinus Torvalds } 8021da177e4SLinus Torvalds 80318bb1db3SAl Viro static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 8041da177e4SLinus Torvalds { 8051da177e4SLinus Torvalds int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); 8061da177e4SLinus Torvalds if (!retval) 807d8c76e6fSDave Hansen inc_nlink(dir); 8081da177e4SLinus Torvalds return retval; 8091da177e4SLinus Torvalds } 8101da177e4SLinus Torvalds 811ebfc3b49SAl Viro static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) 8121da177e4SLinus Torvalds { 8131da177e4SLinus Torvalds return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); 8141da177e4SLinus Torvalds } 8151da177e4SLinus Torvalds 8161da177e4SLinus Torvalds static int hugetlbfs_symlink(struct inode *dir, 8171da177e4SLinus Torvalds struct dentry *dentry, const char *symname) 8181da177e4SLinus Torvalds { 8191da177e4SLinus Torvalds struct inode *inode; 8201da177e4SLinus Torvalds int error = -ENOSPC; 8211da177e4SLinus Torvalds 8227d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 8231da177e4SLinus Torvalds if (inode) { 8241da177e4SLinus Torvalds int l = strlen(symname)+1; 8251da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 8261da177e4SLinus Torvalds if (!error) { 8271da177e4SLinus Torvalds d_instantiate(dentry, inode); 8281da177e4SLinus Torvalds dget(dentry); 8291da177e4SLinus Torvalds } else 8301da177e4SLinus Torvalds iput(inode); 8311da177e4SLinus Torvalds } 832078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 8331da177e4SLinus Torvalds 8341da177e4SLinus Torvalds return error; 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 8371da177e4SLinus Torvalds /* 8386649a386SKen Chen * mark the head page dirty 8391da177e4SLinus Torvalds */ 8401da177e4SLinus Torvalds static int hugetlbfs_set_page_dirty(struct page *page) 8411da177e4SLinus Torvalds { 842d85f3385SChristoph Lameter struct page *head = compound_head(page); 8436649a386SKen Chen 8446649a386SKen Chen SetPageDirty(head); 8451da177e4SLinus Torvalds return 0; 8461da177e4SLinus Torvalds } 8471da177e4SLinus Torvalds 848290408d4SNaoya Horiguchi static int hugetlbfs_migrate_page(struct address_space *mapping, 849b969c4abSMel Gorman struct page *newpage, struct page *page, 850a6bc32b8SMel Gorman enum migrate_mode mode) 851290408d4SNaoya Horiguchi { 852290408d4SNaoya Horiguchi int rc; 853290408d4SNaoya Horiguchi 854290408d4SNaoya Horiguchi rc = migrate_huge_page_move_mapping(mapping, newpage, page); 85578bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 856290408d4SNaoya Horiguchi return rc; 8572916ecc0SJérôme Glisse if (mode != MIGRATE_SYNC_NO_COPY) 858290408d4SNaoya Horiguchi migrate_page_copy(newpage, page); 8592916ecc0SJérôme Glisse else 8602916ecc0SJérôme Glisse migrate_page_states(newpage, page); 861290408d4SNaoya Horiguchi 86278bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 863290408d4SNaoya Horiguchi } 864290408d4SNaoya Horiguchi 86578bb9203SNaoya Horiguchi static int hugetlbfs_error_remove_page(struct address_space *mapping, 86678bb9203SNaoya Horiguchi struct page *page) 86778bb9203SNaoya Horiguchi { 86878bb9203SNaoya Horiguchi struct inode *inode = mapping->host; 869ab615a5bSMike Kravetz pgoff_t index = page->index; 87078bb9203SNaoya Horiguchi 87178bb9203SNaoya Horiguchi remove_huge_page(page); 872ab615a5bSMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) 87378bb9203SNaoya Horiguchi hugetlb_fix_reserve_counts(inode); 874ab615a5bSMike Kravetz 87578bb9203SNaoya Horiguchi return 0; 87678bb9203SNaoya Horiguchi } 87778bb9203SNaoya Horiguchi 8784a25220dSDavid Howells /* 8794a25220dSDavid Howells * Display the mount options in /proc/mounts. 8804a25220dSDavid Howells */ 8814a25220dSDavid Howells static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) 8824a25220dSDavid Howells { 8834a25220dSDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); 8844a25220dSDavid Howells struct hugepage_subpool *spool = sbinfo->spool; 8854a25220dSDavid Howells unsigned long hpage_size = huge_page_size(sbinfo->hstate); 8864a25220dSDavid Howells unsigned hpage_shift = huge_page_shift(sbinfo->hstate); 8874a25220dSDavid Howells char mod; 8884a25220dSDavid Howells 8894a25220dSDavid Howells if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) 8904a25220dSDavid Howells seq_printf(m, ",uid=%u", 8914a25220dSDavid Howells from_kuid_munged(&init_user_ns, sbinfo->uid)); 8924a25220dSDavid Howells if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) 8934a25220dSDavid Howells seq_printf(m, ",gid=%u", 8944a25220dSDavid Howells from_kgid_munged(&init_user_ns, sbinfo->gid)); 8954a25220dSDavid Howells if (sbinfo->mode != 0755) 8964a25220dSDavid Howells seq_printf(m, ",mode=%o", sbinfo->mode); 8974a25220dSDavid Howells if (sbinfo->max_inodes != -1) 8984a25220dSDavid Howells seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); 8994a25220dSDavid Howells 9004a25220dSDavid Howells hpage_size /= 1024; 9014a25220dSDavid Howells mod = 'K'; 9024a25220dSDavid Howells if (hpage_size >= 1024) { 9034a25220dSDavid Howells hpage_size /= 1024; 9044a25220dSDavid Howells mod = 'M'; 9054a25220dSDavid Howells } 9064a25220dSDavid Howells seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); 9074a25220dSDavid Howells if (spool) { 9084a25220dSDavid Howells if (spool->max_hpages != -1) 9094a25220dSDavid Howells seq_printf(m, ",size=%llu", 9104a25220dSDavid Howells (unsigned long long)spool->max_hpages << hpage_shift); 9114a25220dSDavid Howells if (spool->min_hpages != -1) 9124a25220dSDavid Howells seq_printf(m, ",min_size=%llu", 9134a25220dSDavid Howells (unsigned long long)spool->min_hpages << hpage_shift); 9144a25220dSDavid Howells } 9154a25220dSDavid Howells return 0; 9164a25220dSDavid Howells } 9174a25220dSDavid Howells 918726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 9191da177e4SLinus Torvalds { 920726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 9212b0143b5SDavid Howells struct hstate *h = hstate_inode(d_inode(dentry)); 9221da177e4SLinus Torvalds 9231da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 924a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 9251da177e4SLinus Torvalds if (sbinfo) { 9261da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 92774a8a65cSDavid Gibson /* If no limits set, just report 0 for max/free/used 92874a8a65cSDavid Gibson * blocks, like simple_statfs() */ 92990481622SDavid Gibson if (sbinfo->spool) { 93090481622SDavid Gibson long free_pages; 93190481622SDavid Gibson 93290481622SDavid Gibson spin_lock(&sbinfo->spool->lock); 93390481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 93490481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 93590481622SDavid Gibson - sbinfo->spool->used_hpages; 93690481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 93790481622SDavid Gibson spin_unlock(&sbinfo->spool->lock); 9381da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 9391da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 94074a8a65cSDavid Gibson } 9411da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 9441da177e4SLinus Torvalds return 0; 9451da177e4SLinus Torvalds } 9461da177e4SLinus Torvalds 9471da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 9481da177e4SLinus Torvalds { 9491da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 9501da177e4SLinus Torvalds 9511da177e4SLinus Torvalds if (sbi) { 9521da177e4SLinus Torvalds sb->s_fs_info = NULL; 95390481622SDavid Gibson 95490481622SDavid Gibson if (sbi->spool) 95590481622SDavid Gibson hugepage_put_subpool(sbi->spool); 95690481622SDavid Gibson 9571da177e4SLinus Torvalds kfree(sbi); 9581da177e4SLinus Torvalds } 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds 96196527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 96296527980SChristoph Hellwig { 96396527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 96496527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 96596527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 96696527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 96796527980SChristoph Hellwig return 0; 96896527980SChristoph Hellwig } 96996527980SChristoph Hellwig sbinfo->free_inodes--; 97096527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 97196527980SChristoph Hellwig } 97296527980SChristoph Hellwig 97396527980SChristoph Hellwig return 1; 97496527980SChristoph Hellwig } 97596527980SChristoph Hellwig 97696527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 97796527980SChristoph Hellwig { 97896527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 97996527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 98096527980SChristoph Hellwig sbinfo->free_inodes++; 98196527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 98296527980SChristoph Hellwig } 98396527980SChristoph Hellwig } 98496527980SChristoph Hellwig 98596527980SChristoph Hellwig 986e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 9871da177e4SLinus Torvalds 9881da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 9891da177e4SLinus Torvalds { 99096527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 9911da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 9921da177e4SLinus Torvalds 99396527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 9941da177e4SLinus Torvalds return NULL; 995e94b1766SChristoph Lameter p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); 99696527980SChristoph Hellwig if (unlikely(!p)) { 99796527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 99896527980SChristoph Hellwig return NULL; 9991da177e4SLinus Torvalds } 10004742a35dSMike Kravetz 10014742a35dSMike Kravetz /* 10024742a35dSMike Kravetz * Any time after allocation, hugetlbfs_destroy_inode can be called 10034742a35dSMike Kravetz * for the inode. mpol_free_shared_policy is unconditionally called 10044742a35dSMike Kravetz * as part of hugetlbfs_destroy_inode. So, initialize policy here 10054742a35dSMike Kravetz * in case of a quick call to destroy. 10064742a35dSMike Kravetz * 10074742a35dSMike Kravetz * Note that the policy is initialized even if we are creating a 10084742a35dSMike Kravetz * private inode. This simplifies hugetlbfs_destroy_inode. 10094742a35dSMike Kravetz */ 10104742a35dSMike Kravetz mpol_shared_policy_init(&p->policy, NULL); 10114742a35dSMike Kravetz 101296527980SChristoph Hellwig return &p->vfs_inode; 10131da177e4SLinus Torvalds } 10141da177e4SLinus Torvalds 1015fa0d7e3dSNick Piggin static void hugetlbfs_i_callback(struct rcu_head *head) 1016fa0d7e3dSNick Piggin { 1017fa0d7e3dSNick Piggin struct inode *inode = container_of(head, struct inode, i_rcu); 1018fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 1019fa0d7e3dSNick Piggin } 1020fa0d7e3dSNick Piggin 10211da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 10221da177e4SLinus Torvalds { 102396527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 10241da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 1025fa0d7e3dSNick Piggin call_rcu(&inode->i_rcu, hugetlbfs_i_callback); 10261da177e4SLinus Torvalds } 10271da177e4SLinus Torvalds 1028f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 1029800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 1030800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 10311da177e4SLinus Torvalds .set_page_dirty = hugetlbfs_set_page_dirty, 1032290408d4SNaoya Horiguchi .migratepage = hugetlbfs_migrate_page, 103378bb9203SNaoya Horiguchi .error_remove_page = hugetlbfs_error_remove_page, 10341da177e4SLinus Torvalds }; 10351da177e4SLinus Torvalds 103696527980SChristoph Hellwig 103751cc5068SAlexey Dobriyan static void init_once(void *foo) 103896527980SChristoph Hellwig { 103996527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 104096527980SChristoph Hellwig 104196527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 104296527980SChristoph Hellwig } 104396527980SChristoph Hellwig 10444b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 104534d0640eSAl Viro .read_iter = hugetlbfs_read_iter, 10461da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 10471b061d92SChristoph Hellwig .fsync = noop_fsync, 10481da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 10496038f373SArnd Bergmann .llseek = default_llseek, 105070c3547eSMike Kravetz .fallocate = hugetlbfs_fallocate, 10511da177e4SLinus Torvalds }; 10521da177e4SLinus Torvalds 105392e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 10541da177e4SLinus Torvalds .create = hugetlbfs_create, 10551da177e4SLinus Torvalds .lookup = simple_lookup, 10561da177e4SLinus Torvalds .link = simple_link, 10571da177e4SLinus Torvalds .unlink = simple_unlink, 10581da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 10591da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 10601da177e4SLinus Torvalds .rmdir = simple_rmdir, 10611da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 10621da177e4SLinus Torvalds .rename = simple_rename, 10631da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 10641da177e4SLinus Torvalds }; 10651da177e4SLinus Torvalds 106692e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 10671da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 10681da177e4SLinus Torvalds }; 10691da177e4SLinus Torvalds 1070ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 10711da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 10721da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 10732bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 10741da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 10751da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 10764a25220dSDavid Howells .show_options = hugetlbfs_show_options, 10771da177e4SLinus Torvalds }; 10781da177e4SLinus Torvalds 10794a25220dSDavid Howells enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT }; 10807ca02d0aSMike Kravetz 10817ca02d0aSMike Kravetz /* 10827ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 10837ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 10847ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 10857ca02d0aSMike Kravetz */ 10864a25220dSDavid Howells static long 10877ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 10884a25220dSDavid Howells enum hugetlbfs_size_type val_type) 10897ca02d0aSMike Kravetz { 10907ca02d0aSMike Kravetz if (val_type == NO_SIZE) 10917ca02d0aSMike Kravetz return -1; 10927ca02d0aSMike Kravetz 10937ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 10947ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 10957ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 10967ca02d0aSMike Kravetz do_div(size_opt, 100); 10977ca02d0aSMike Kravetz } 10987ca02d0aSMike Kravetz 10997ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 11007ca02d0aSMike Kravetz return size_opt; 11017ca02d0aSMike Kravetz } 11027ca02d0aSMike Kravetz 11031da177e4SLinus Torvalds static int 11041da177e4SLinus Torvalds hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) 11051da177e4SLinus Torvalds { 1106e73a75faSRandy Dunlap char *p, *rest; 1107e73a75faSRandy Dunlap substring_t args[MAX_OPT_ARGS]; 1108e73a75faSRandy Dunlap int option; 11097ca02d0aSMike Kravetz unsigned long long max_size_opt = 0, min_size_opt = 0; 11104a25220dSDavid Howells enum hugetlbfs_size_type max_val_type = NO_SIZE, min_val_type = NO_SIZE; 11111da177e4SLinus Torvalds 11121da177e4SLinus Torvalds if (!options) 11131da177e4SLinus Torvalds return 0; 11141da177e4SLinus Torvalds 1115e73a75faSRandy Dunlap while ((p = strsep(&options, ",")) != NULL) { 1116e73a75faSRandy Dunlap int token; 1117b4c07bceSLee Schermerhorn if (!*p) 1118b4c07bceSLee Schermerhorn continue; 11191da177e4SLinus Torvalds 1120e73a75faSRandy Dunlap token = match_token(p, tokens, args); 1121e73a75faSRandy Dunlap switch (token) { 1122e73a75faSRandy Dunlap case Opt_uid: 1123e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 1124e73a75faSRandy Dunlap goto bad_val; 1125a0eb3a05SEric W. Biederman pconfig->uid = make_kuid(current_user_ns(), option); 1126a0eb3a05SEric W. Biederman if (!uid_valid(pconfig->uid)) 1127a0eb3a05SEric W. Biederman goto bad_val; 1128e73a75faSRandy Dunlap break; 1129e73a75faSRandy Dunlap 1130e73a75faSRandy Dunlap case Opt_gid: 1131e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 1132e73a75faSRandy Dunlap goto bad_val; 1133a0eb3a05SEric W. Biederman pconfig->gid = make_kgid(current_user_ns(), option); 1134a0eb3a05SEric W. Biederman if (!gid_valid(pconfig->gid)) 1135a0eb3a05SEric W. Biederman goto bad_val; 1136e73a75faSRandy Dunlap break; 1137e73a75faSRandy Dunlap 1138e73a75faSRandy Dunlap case Opt_mode: 1139e73a75faSRandy Dunlap if (match_octal(&args[0], &option)) 1140e73a75faSRandy Dunlap goto bad_val; 114175897d60SKen Chen pconfig->mode = option & 01777U; 1142e73a75faSRandy Dunlap break; 1143e73a75faSRandy Dunlap 1144e73a75faSRandy Dunlap case Opt_size: { 1145e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 1146e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 1147e73a75faSRandy Dunlap goto bad_val; 11487ca02d0aSMike Kravetz max_size_opt = memparse(args[0].from, &rest); 11497ca02d0aSMike Kravetz max_val_type = SIZE_STD; 1150a137e1ccSAndi Kleen if (*rest == '%') 11517ca02d0aSMike Kravetz max_val_type = SIZE_PERCENT; 1152e73a75faSRandy Dunlap break; 1153e73a75faSRandy Dunlap } 11541da177e4SLinus Torvalds 1155e73a75faSRandy Dunlap case Opt_nr_inodes: 1156e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 1157e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 1158e73a75faSRandy Dunlap goto bad_val; 1159e73a75faSRandy Dunlap pconfig->nr_inodes = memparse(args[0].from, &rest); 1160e73a75faSRandy Dunlap break; 1161e73a75faSRandy Dunlap 1162a137e1ccSAndi Kleen case Opt_pagesize: { 1163a137e1ccSAndi Kleen unsigned long ps; 1164a137e1ccSAndi Kleen ps = memparse(args[0].from, &rest); 1165a137e1ccSAndi Kleen pconfig->hstate = size_to_hstate(ps); 1166a137e1ccSAndi Kleen if (!pconfig->hstate) { 11679b857d26SAndrew Morton pr_err("Unsupported page size %lu MB\n", 1168a137e1ccSAndi Kleen ps >> 20); 1169a137e1ccSAndi Kleen return -EINVAL; 1170a137e1ccSAndi Kleen } 1171a137e1ccSAndi Kleen break; 1172a137e1ccSAndi Kleen } 1173a137e1ccSAndi Kleen 11747ca02d0aSMike Kravetz case Opt_min_size: { 11757ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 11767ca02d0aSMike Kravetz if (!isdigit(*args[0].from)) 11777ca02d0aSMike Kravetz goto bad_val; 11787ca02d0aSMike Kravetz min_size_opt = memparse(args[0].from, &rest); 11797ca02d0aSMike Kravetz min_val_type = SIZE_STD; 11807ca02d0aSMike Kravetz if (*rest == '%') 11817ca02d0aSMike Kravetz min_val_type = SIZE_PERCENT; 11827ca02d0aSMike Kravetz break; 11837ca02d0aSMike Kravetz } 11847ca02d0aSMike Kravetz 1185e73a75faSRandy Dunlap default: 11869b857d26SAndrew Morton pr_err("Bad mount option: \"%s\"\n", p); 1187b4c07bceSLee Schermerhorn return -EINVAL; 1188e73a75faSRandy Dunlap break; 1189e73a75faSRandy Dunlap } 11901da177e4SLinus Torvalds } 1191a137e1ccSAndi Kleen 11927ca02d0aSMike Kravetz /* 11937ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 11947ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 11957ca02d0aSMike Kravetz */ 11967ca02d0aSMike Kravetz pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, 11977ca02d0aSMike Kravetz max_size_opt, max_val_type); 11987ca02d0aSMike Kravetz pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, 11997ca02d0aSMike Kravetz min_size_opt, min_val_type); 12007ca02d0aSMike Kravetz 12017ca02d0aSMike Kravetz /* 12027ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 12037ca02d0aSMike Kravetz */ 12047ca02d0aSMike Kravetz if (max_val_type > NO_SIZE && 12057ca02d0aSMike Kravetz pconfig->min_hpages > pconfig->max_hpages) { 12067ca02d0aSMike Kravetz pr_err("minimum size can not be greater than maximum size\n"); 12077ca02d0aSMike Kravetz return -EINVAL; 1208a137e1ccSAndi Kleen } 1209a137e1ccSAndi Kleen 12101da177e4SLinus Torvalds return 0; 1211e73a75faSRandy Dunlap 1212e73a75faSRandy Dunlap bad_val: 12139b857d26SAndrew Morton pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); 1214c12ddba0SAkinobu Mita return -EINVAL; 12151da177e4SLinus Torvalds } 12161da177e4SLinus Torvalds 12171da177e4SLinus Torvalds static int 12181da177e4SLinus Torvalds hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) 12191da177e4SLinus Torvalds { 12201da177e4SLinus Torvalds int ret; 12211da177e4SLinus Torvalds struct hugetlbfs_config config; 12221da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 12231da177e4SLinus Torvalds 12247ca02d0aSMike Kravetz config.max_hpages = -1; /* No limit on size by default */ 12251da177e4SLinus Torvalds config.nr_inodes = -1; /* No limit on number of inodes by default */ 122677c70de1SDavid Howells config.uid = current_fsuid(); 122777c70de1SDavid Howells config.gid = current_fsgid(); 12281da177e4SLinus Torvalds config.mode = 0755; 1229a137e1ccSAndi Kleen config.hstate = &default_hstate; 12307ca02d0aSMike Kravetz config.min_hpages = -1; /* No default minimum size */ 12311da177e4SLinus Torvalds ret = hugetlbfs_parse_options(data, &config); 12321da177e4SLinus Torvalds if (ret) 12331da177e4SLinus Torvalds return ret; 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 12361da177e4SLinus Torvalds if (!sbinfo) 12371da177e4SLinus Torvalds return -ENOMEM; 12381da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 1239a137e1ccSAndi Kleen sbinfo->hstate = config.hstate; 12401da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 12411da177e4SLinus Torvalds sbinfo->max_inodes = config.nr_inodes; 12421da177e4SLinus Torvalds sbinfo->free_inodes = config.nr_inodes; 124390481622SDavid Gibson sbinfo->spool = NULL; 12444a25220dSDavid Howells sbinfo->uid = config.uid; 12454a25220dSDavid Howells sbinfo->gid = config.gid; 12464a25220dSDavid Howells sbinfo->mode = config.mode; 12474a25220dSDavid Howells 12487ca02d0aSMike Kravetz /* 12497ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 12507ca02d0aSMike Kravetz * specified. Any needed reservations (for minimim size) are taken 12517ca02d0aSMike Kravetz * taken when the subpool is created. 12527ca02d0aSMike Kravetz */ 12537ca02d0aSMike Kravetz if (config.max_hpages != -1 || config.min_hpages != -1) { 12547ca02d0aSMike Kravetz sbinfo->spool = hugepage_new_subpool(config.hstate, 12557ca02d0aSMike Kravetz config.max_hpages, 12567ca02d0aSMike Kravetz config.min_hpages); 125790481622SDavid Gibson if (!sbinfo->spool) 125890481622SDavid Gibson goto out_free; 125990481622SDavid Gibson } 12601da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 1261a137e1ccSAndi Kleen sb->s_blocksize = huge_page_size(config.hstate); 1262a137e1ccSAndi Kleen sb->s_blocksize_bits = huge_page_shift(config.hstate); 12631da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 12641da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 12651da177e4SLinus Torvalds sb->s_time_gran = 1; 126648fde701SAl Viro sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); 126748fde701SAl Viro if (!sb->s_root) 12681da177e4SLinus Torvalds goto out_free; 12691da177e4SLinus Torvalds return 0; 12701da177e4SLinus Torvalds out_free: 127190481622SDavid Gibson kfree(sbinfo->spool); 12721da177e4SLinus Torvalds kfree(sbinfo); 12731da177e4SLinus Torvalds return -ENOMEM; 12741da177e4SLinus Torvalds } 12751da177e4SLinus Torvalds 12763c26ff6eSAl Viro static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, 12773c26ff6eSAl Viro int flags, const char *dev_name, void *data) 12781da177e4SLinus Torvalds { 12793c26ff6eSAl Viro return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); 12801da177e4SLinus Torvalds } 12811da177e4SLinus Torvalds 12821da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 12831da177e4SLinus Torvalds .name = "hugetlbfs", 12843c26ff6eSAl Viro .mount = hugetlbfs_mount, 12851da177e4SLinus Torvalds .kill_sb = kill_litter_super, 12861da177e4SLinus Torvalds }; 12871da177e4SLinus Torvalds 128842d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 12891da177e4SLinus Torvalds 1290ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 12911da177e4SLinus Torvalds { 1292a0eb3a05SEric W. Biederman kgid_t shm_group; 1293a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 1294a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 12951da177e4SLinus Torvalds } 12961da177e4SLinus Torvalds 129742d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 129842d7395fSAndi Kleen { 1299af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 130042d7395fSAndi Kleen 130142d7395fSAndi Kleen if (!h) 130242d7395fSAndi Kleen return -1; 130342d7395fSAndi Kleen return h - hstates; 130442d7395fSAndi Kleen } 130542d7395fSAndi Kleen 1306af73e4d9SNaoya Horiguchi /* 1307af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 1308af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 1309af73e4d9SNaoya Horiguchi */ 1310af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 1311af73e4d9SNaoya Horiguchi vm_flags_t acctflag, struct user_struct **user, 131242d7395fSAndi Kleen int creat_flags, int page_size_log) 13131da177e4SLinus Torvalds { 13141da177e4SLinus Torvalds struct inode *inode; 1315e68375c8SAl Viro struct vfsmount *mnt; 131642d7395fSAndi Kleen int hstate_idx; 1317e68375c8SAl Viro struct file *file; 131842d7395fSAndi Kleen 131942d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 132042d7395fSAndi Kleen if (hstate_idx < 0) 132142d7395fSAndi Kleen return ERR_PTR(-ENODEV); 13221da177e4SLinus Torvalds 1323353d5c30SHugh Dickins *user = NULL; 1324e68375c8SAl Viro mnt = hugetlbfs_vfsmount[hstate_idx]; 1325e68375c8SAl Viro if (!mnt) 13265bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 13275bc98594SAkinobu Mita 1328ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 1329353d5c30SHugh Dickins *user = current_user(); 1330353d5c30SHugh Dickins if (user_shm_lock(size, *user)) { 133121a3c273SDavid Rientjes task_lock(current); 13329b857d26SAndrew Morton pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", 133321a3c273SDavid Rientjes current->comm, current->pid); 133421a3c273SDavid Rientjes task_unlock(current); 1335353d5c30SHugh Dickins } else { 1336353d5c30SHugh Dickins *user = NULL; 13371da177e4SLinus Torvalds return ERR_PTR(-EPERM); 13382584e517SRavikiran G Thirumalai } 1339353d5c30SHugh Dickins } 13401da177e4SLinus Torvalds 134139b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 1342e68375c8SAl Viro inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0); 13431da177e4SLinus Torvalds if (!inode) 1344e68375c8SAl Viro goto out; 1345e1832f29SStephen Smalley if (creat_flags == HUGETLB_SHMFS_INODE) 1346e1832f29SStephen Smalley inode->i_flags |= S_PRIVATE; 13471da177e4SLinus Torvalds 13481da177e4SLinus Torvalds inode->i_size = size; 13496d6b77f1SMiklos Szeredi clear_nlink(inode); 1350ce8d2cdfSDave Hansen 1351e68375c8SAl Viro if (hugetlb_reserve_pages(inode, 0, 1352e68375c8SAl Viro size >> huge_page_shift(hstate_inode(inode)), NULL, 1353e68375c8SAl Viro acctflag)) 1354e68375c8SAl Viro file = ERR_PTR(-ENOMEM); 1355e68375c8SAl Viro else 1356e68375c8SAl Viro file = alloc_file_pseudo(inode, mnt, name, O_RDWR, 1357ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 1358e68375c8SAl Viro if (!IS_ERR(file)) 13591da177e4SLinus Torvalds return file; 13601da177e4SLinus Torvalds 1361b45b5bd6SDavid Gibson iput(inode); 1362e68375c8SAl Viro out: 1363353d5c30SHugh Dickins if (*user) { 1364353d5c30SHugh Dickins user_shm_unlock(size, *user); 1365353d5c30SHugh Dickins *user = NULL; 1366353d5c30SHugh Dickins } 136739b65252SAnatol Pomozov return file; 13681da177e4SLinus Torvalds } 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 13711da177e4SLinus Torvalds { 137242d7395fSAndi Kleen struct hstate *h; 13731da177e4SLinus Torvalds int error; 137442d7395fSAndi Kleen int i; 13751da177e4SLinus Torvalds 1376457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 13779b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1378457c1b27SNishanth Aravamudan return -ENOTSUPP; 1379457c1b27SNishanth Aravamudan } 1380457c1b27SNishanth Aravamudan 1381d1d5e05fSHillf Danton error = -ENOMEM; 13821da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 13831da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 13845d097056SVladimir Davydov 0, SLAB_ACCOUNT, init_once); 13851da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 1386e0bf68ddSPeter Zijlstra goto out2; 13871da177e4SLinus Torvalds 13881da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 13891da177e4SLinus Torvalds if (error) 13901da177e4SLinus Torvalds goto out; 13911da177e4SLinus Torvalds 139242d7395fSAndi Kleen i = 0; 139342d7395fSAndi Kleen for_each_hstate(h) { 139442d7395fSAndi Kleen char buf[50]; 139542d7395fSAndi Kleen unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); 13961da177e4SLinus Torvalds 139742d7395fSAndi Kleen snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); 139842d7395fSAndi Kleen hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, 139942d7395fSAndi Kleen buf); 140042d7395fSAndi Kleen 140142d7395fSAndi Kleen if (IS_ERR(hugetlbfs_vfsmount[i])) { 14029b857d26SAndrew Morton pr_err("Cannot mount internal hugetlbfs for " 140342d7395fSAndi Kleen "page size %uK", ps_kb); 140442d7395fSAndi Kleen error = PTR_ERR(hugetlbfs_vfsmount[i]); 140542d7395fSAndi Kleen hugetlbfs_vfsmount[i] = NULL; 14061da177e4SLinus Torvalds } 140742d7395fSAndi Kleen i++; 140842d7395fSAndi Kleen } 140942d7395fSAndi Kleen /* Non default hstates are optional */ 141042d7395fSAndi Kleen if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) 141142d7395fSAndi Kleen return 0; 14121da177e4SLinus Torvalds 14131da177e4SLinus Torvalds out: 14141da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 1415e0bf68ddSPeter Zijlstra out2: 14161da177e4SLinus Torvalds return error; 14171da177e4SLinus Torvalds } 14183e89e1c5SPaul Gortmaker fs_initcall(init_hugetlbfs_fs) 1419