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. 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 99b857d26SAndrew Morton #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 109b857d26SAndrew Morton 111da177e4SLinus Torvalds #include <linux/module.h> 121da177e4SLinus Torvalds #include <linux/thread_info.h> 131da177e4SLinus Torvalds #include <asm/current.h> 141da177e4SLinus Torvalds #include <linux/sched.h> /* remove ASAP */ 151da177e4SLinus Torvalds #include <linux/fs.h> 161da177e4SLinus Torvalds #include <linux/mount.h> 171da177e4SLinus Torvalds #include <linux/file.h> 18e73a75faSRandy Dunlap #include <linux/kernel.h> 191da177e4SLinus Torvalds #include <linux/writeback.h> 201da177e4SLinus Torvalds #include <linux/pagemap.h> 211da177e4SLinus Torvalds #include <linux/highmem.h> 221da177e4SLinus Torvalds #include <linux/init.h> 231da177e4SLinus Torvalds #include <linux/string.h> 2416f7e0feSRandy Dunlap #include <linux/capability.h> 25e73a75faSRandy Dunlap #include <linux/ctype.h> 261da177e4SLinus Torvalds #include <linux/backing-dev.h> 271da177e4SLinus Torvalds #include <linux/hugetlb.h> 281da177e4SLinus Torvalds #include <linux/pagevec.h> 29e73a75faSRandy Dunlap #include <linux/parser.h> 30036e0856SBenjamin Herrenschmidt #include <linux/mman.h> 311da177e4SLinus Torvalds #include <linux/slab.h> 321da177e4SLinus Torvalds #include <linux/dnotify.h> 331da177e4SLinus Torvalds #include <linux/statfs.h> 341da177e4SLinus Torvalds #include <linux/security.h> 351fd7317dSNick Black #include <linux/magic.h> 36290408d4SNaoya Horiguchi #include <linux/migrate.h> 371da177e4SLinus Torvalds 381da177e4SLinus Torvalds #include <asm/uaccess.h> 391da177e4SLinus Torvalds 40ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops; 41f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops; 424b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations; 4392e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations; 4492e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations; 451da177e4SLinus Torvalds 46a1d776eeSDavid Gibson struct hugetlbfs_config { 47a0eb3a05SEric W. Biederman kuid_t uid; 48a0eb3a05SEric W. Biederman kgid_t gid; 49a1d776eeSDavid Gibson umode_t mode; 507ca02d0aSMike Kravetz long max_hpages; 51a1d776eeSDavid Gibson long nr_inodes; 52a1d776eeSDavid Gibson struct hstate *hstate; 537ca02d0aSMike Kravetz long min_hpages; 54a1d776eeSDavid Gibson }; 55a1d776eeSDavid Gibson 56a1d776eeSDavid Gibson struct hugetlbfs_inode_info { 57a1d776eeSDavid Gibson struct shared_policy policy; 58a1d776eeSDavid Gibson struct inode vfs_inode; 59a1d776eeSDavid Gibson }; 60a1d776eeSDavid Gibson 61a1d776eeSDavid Gibson static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode) 62a1d776eeSDavid Gibson { 63a1d776eeSDavid Gibson return container_of(inode, struct hugetlbfs_inode_info, vfs_inode); 64a1d776eeSDavid Gibson } 65a1d776eeSDavid Gibson 661da177e4SLinus Torvalds int sysctl_hugetlb_shm_group; 671da177e4SLinus Torvalds 68e73a75faSRandy Dunlap enum { 69e73a75faSRandy Dunlap Opt_size, Opt_nr_inodes, 70e73a75faSRandy Dunlap Opt_mode, Opt_uid, Opt_gid, 717ca02d0aSMike Kravetz Opt_pagesize, Opt_min_size, 72e73a75faSRandy Dunlap Opt_err, 73e73a75faSRandy Dunlap }; 74e73a75faSRandy Dunlap 75a447c093SSteven Whitehouse static const match_table_t tokens = { 76e73a75faSRandy Dunlap {Opt_size, "size=%s"}, 77e73a75faSRandy Dunlap {Opt_nr_inodes, "nr_inodes=%s"}, 78e73a75faSRandy Dunlap {Opt_mode, "mode=%o"}, 79e73a75faSRandy Dunlap {Opt_uid, "uid=%u"}, 80e73a75faSRandy Dunlap {Opt_gid, "gid=%u"}, 81a137e1ccSAndi Kleen {Opt_pagesize, "pagesize=%s"}, 827ca02d0aSMike Kravetz {Opt_min_size, "min_size=%s"}, 83e73a75faSRandy Dunlap {Opt_err, NULL}, 84e73a75faSRandy Dunlap }; 85e73a75faSRandy Dunlap 862e9b367cSAdam Litke static void huge_pagevec_release(struct pagevec *pvec) 872e9b367cSAdam Litke { 882e9b367cSAdam Litke int i; 892e9b367cSAdam Litke 902e9b367cSAdam Litke for (i = 0; i < pagevec_count(pvec); ++i) 912e9b367cSAdam Litke put_page(pvec->pages[i]); 922e9b367cSAdam Litke 932e9b367cSAdam Litke pagevec_reinit(pvec); 942e9b367cSAdam Litke } 952e9b367cSAdam Litke 961da177e4SLinus Torvalds static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) 971da177e4SLinus Torvalds { 98496ad9aaSAl Viro struct inode *inode = file_inode(file); 991da177e4SLinus Torvalds loff_t len, vma_len; 1001da177e4SLinus Torvalds int ret; 101a5516438SAndi Kleen struct hstate *h = hstate_file(file); 1021da177e4SLinus Torvalds 10368589bc3SHugh Dickins /* 104dec4ad86SDavid Gibson * vma address alignment (but not the pgoff alignment) has 105dec4ad86SDavid Gibson * already been checked by prepare_hugepage_range. If you add 106dec4ad86SDavid Gibson * any error returns here, do so after setting VM_HUGETLB, so 107dec4ad86SDavid Gibson * is_vm_hugetlb_page tests below unmap_region go the right 108dec4ad86SDavid Gibson * way when do_mmap_pgoff unwinds (may be important on powerpc 109dec4ad86SDavid Gibson * and ia64). 11068589bc3SHugh Dickins */ 111a2fce914SNaoya Horiguchi vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; 11268589bc3SHugh Dickins vma->vm_ops = &hugetlb_vm_ops; 1131da177e4SLinus Torvalds 1142b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 115dec4ad86SDavid Gibson return -EINVAL; 116dec4ad86SDavid Gibson 1171da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 1181da177e4SLinus Torvalds 1191b1dcc1bSJes Sorensen mutex_lock(&inode->i_mutex); 1201da177e4SLinus Torvalds file_accessed(file); 1211da177e4SLinus Torvalds 1221da177e4SLinus Torvalds ret = -ENOMEM; 1231da177e4SLinus Torvalds len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 1241da177e4SLinus Torvalds 125a1e78772SMel Gorman if (hugetlb_reserve_pages(inode, 126a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1275a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1285a6fe125SMel Gorman vma->vm_flags)) 129b45b5bd6SDavid Gibson goto out; 130b45b5bd6SDavid Gibson 1314c887265SAdam Litke ret = 0; 1324c887265SAdam Litke hugetlb_prefault_arch_hook(vma->vm_mm); 133b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 1341da177e4SLinus Torvalds inode->i_size = len; 1351da177e4SLinus Torvalds out: 1361b1dcc1bSJes Sorensen mutex_unlock(&inode->i_mutex); 1371da177e4SLinus Torvalds 1381da177e4SLinus Torvalds return ret; 1391da177e4SLinus Torvalds } 1401da177e4SLinus Torvalds 1411da177e4SLinus Torvalds /* 142508034a3SHugh Dickins * Called under down_write(mmap_sem). 1431da177e4SLinus Torvalds */ 1441da177e4SLinus Torvalds 145d2ba27e8SAdrian Bunk #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 1461da177e4SLinus Torvalds static unsigned long 1471da177e4SLinus Torvalds hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 1481da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 1491da177e4SLinus Torvalds { 1501da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1511da177e4SLinus Torvalds struct vm_area_struct *vma; 152a5516438SAndi Kleen struct hstate *h = hstate_file(file); 15308659355SMichel Lespinasse struct vm_unmapped_area_info info; 1541da177e4SLinus Torvalds 155a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 1561da177e4SLinus Torvalds return -EINVAL; 1571da177e4SLinus Torvalds if (len > TASK_SIZE) 1581da177e4SLinus Torvalds return -ENOMEM; 1591da177e4SLinus Torvalds 160036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 161a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 162036e0856SBenjamin Herrenschmidt return -EINVAL; 163036e0856SBenjamin Herrenschmidt return addr; 164036e0856SBenjamin Herrenschmidt } 165036e0856SBenjamin Herrenschmidt 1661da177e4SLinus Torvalds if (addr) { 167a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 1681da177e4SLinus Torvalds vma = find_vma(mm, addr); 1691da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 1701da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 1711da177e4SLinus Torvalds return addr; 1721da177e4SLinus Torvalds } 1731da177e4SLinus Torvalds 17408659355SMichel Lespinasse info.flags = 0; 17508659355SMichel Lespinasse info.length = len; 17608659355SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 17708659355SMichel Lespinasse info.high_limit = TASK_SIZE; 17808659355SMichel Lespinasse info.align_mask = PAGE_MASK & ~huge_page_mask(h); 17908659355SMichel Lespinasse info.align_offset = 0; 18008659355SMichel Lespinasse return vm_unmapped_area(&info); 1811da177e4SLinus Torvalds } 1821da177e4SLinus Torvalds #endif 1831da177e4SLinus Torvalds 184e63e1e5aSBadari Pulavarty static int 185e63e1e5aSBadari Pulavarty hugetlbfs_read_actor(struct page *page, unsigned long offset, 186e63e1e5aSBadari Pulavarty char __user *buf, unsigned long count, 187e63e1e5aSBadari Pulavarty unsigned long size) 188e63e1e5aSBadari Pulavarty { 189e63e1e5aSBadari Pulavarty char *kaddr; 190e63e1e5aSBadari Pulavarty unsigned long left, copied = 0; 191e63e1e5aSBadari Pulavarty int i, chunksize; 192e63e1e5aSBadari Pulavarty 193e63e1e5aSBadari Pulavarty if (size > count) 194e63e1e5aSBadari Pulavarty size = count; 195e63e1e5aSBadari Pulavarty 196e63e1e5aSBadari Pulavarty /* Find which 4k chunk and offset with in that chunk */ 197e63e1e5aSBadari Pulavarty i = offset >> PAGE_CACHE_SHIFT; 198e63e1e5aSBadari Pulavarty offset = offset & ~PAGE_CACHE_MASK; 199e63e1e5aSBadari Pulavarty 200e63e1e5aSBadari Pulavarty while (size) { 201e63e1e5aSBadari Pulavarty chunksize = PAGE_CACHE_SIZE; 202e63e1e5aSBadari Pulavarty if (offset) 203e63e1e5aSBadari Pulavarty chunksize -= offset; 204e63e1e5aSBadari Pulavarty if (chunksize > size) 205e63e1e5aSBadari Pulavarty chunksize = size; 206e63e1e5aSBadari Pulavarty kaddr = kmap(&page[i]); 207e63e1e5aSBadari Pulavarty left = __copy_to_user(buf, kaddr + offset, chunksize); 208e63e1e5aSBadari Pulavarty kunmap(&page[i]); 209e63e1e5aSBadari Pulavarty if (left) { 210e63e1e5aSBadari Pulavarty copied += (chunksize - left); 211e63e1e5aSBadari Pulavarty break; 212e63e1e5aSBadari Pulavarty } 213e63e1e5aSBadari Pulavarty offset = 0; 214e63e1e5aSBadari Pulavarty size -= chunksize; 215e63e1e5aSBadari Pulavarty buf += chunksize; 216e63e1e5aSBadari Pulavarty copied += chunksize; 217e63e1e5aSBadari Pulavarty i++; 218e63e1e5aSBadari Pulavarty } 219e63e1e5aSBadari Pulavarty return copied ? copied : -EFAULT; 220e63e1e5aSBadari Pulavarty } 221e63e1e5aSBadari Pulavarty 222e63e1e5aSBadari Pulavarty /* 223e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 224e63e1e5aSBadari Pulavarty * data. Its *very* similar to do_generic_mapping_read(), we can't use that 225e63e1e5aSBadari Pulavarty * since it has PAGE_CACHE_SIZE assumptions. 226e63e1e5aSBadari Pulavarty */ 227e63e1e5aSBadari Pulavarty static ssize_t hugetlbfs_read(struct file *filp, char __user *buf, 228e63e1e5aSBadari Pulavarty size_t len, loff_t *ppos) 229e63e1e5aSBadari Pulavarty { 230a5516438SAndi Kleen struct hstate *h = hstate_file(filp); 231e63e1e5aSBadari Pulavarty struct address_space *mapping = filp->f_mapping; 232e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 233a5516438SAndi Kleen unsigned long index = *ppos >> huge_page_shift(h); 234a5516438SAndi Kleen unsigned long offset = *ppos & ~huge_page_mask(h); 235e63e1e5aSBadari Pulavarty unsigned long end_index; 236e63e1e5aSBadari Pulavarty loff_t isize; 237e63e1e5aSBadari Pulavarty ssize_t retval = 0; 238e63e1e5aSBadari Pulavarty 239e63e1e5aSBadari Pulavarty /* validate length */ 240e63e1e5aSBadari Pulavarty if (len == 0) 241e63e1e5aSBadari Pulavarty goto out; 242e63e1e5aSBadari Pulavarty 243e63e1e5aSBadari Pulavarty for (;;) { 244e63e1e5aSBadari Pulavarty struct page *page; 245a5516438SAndi Kleen unsigned long nr, ret; 24691bf189cSRoel Kluin int ra; 247e63e1e5aSBadari Pulavarty 248e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 249a5516438SAndi Kleen nr = huge_page_size(h); 250a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 251a05b0855SAneesh Kumar K.V if (!isize) 252a05b0855SAneesh Kumar K.V goto out; 253a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 254e63e1e5aSBadari Pulavarty if (index >= end_index) { 255e63e1e5aSBadari Pulavarty if (index > end_index) 256e63e1e5aSBadari Pulavarty goto out; 257a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 258a05b0855SAneesh Kumar K.V if (nr <= offset) 259e63e1e5aSBadari Pulavarty goto out; 260e63e1e5aSBadari Pulavarty } 261e63e1e5aSBadari Pulavarty nr = nr - offset; 262e63e1e5aSBadari Pulavarty 263e63e1e5aSBadari Pulavarty /* Find the page */ 264a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 265e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 266e63e1e5aSBadari Pulavarty /* 267e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 268e63e1e5aSBadari Pulavarty * length of the hole or request. 269e63e1e5aSBadari Pulavarty */ 270e63e1e5aSBadari Pulavarty ret = len < nr ? len : nr; 271e63e1e5aSBadari Pulavarty if (clear_user(buf, ret)) 27291bf189cSRoel Kluin ra = -EFAULT; 27391bf189cSRoel Kluin else 27491bf189cSRoel Kluin ra = 0; 275e63e1e5aSBadari Pulavarty } else { 276a05b0855SAneesh Kumar K.V unlock_page(page); 277a05b0855SAneesh Kumar K.V 278e63e1e5aSBadari Pulavarty /* 279e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 280e63e1e5aSBadari Pulavarty */ 28191bf189cSRoel Kluin ra = hugetlbfs_read_actor(page, offset, buf, len, nr); 28291bf189cSRoel Kluin ret = ra; 283a05b0855SAneesh Kumar K.V page_cache_release(page); 284e63e1e5aSBadari Pulavarty } 28591bf189cSRoel Kluin if (ra < 0) { 286e63e1e5aSBadari Pulavarty if (retval == 0) 28791bf189cSRoel Kluin retval = ra; 288e63e1e5aSBadari Pulavarty goto out; 289e63e1e5aSBadari Pulavarty } 290e63e1e5aSBadari Pulavarty 291e63e1e5aSBadari Pulavarty offset += ret; 292e63e1e5aSBadari Pulavarty retval += ret; 293e63e1e5aSBadari Pulavarty len -= ret; 294a5516438SAndi Kleen index += offset >> huge_page_shift(h); 295a5516438SAndi Kleen offset &= ~huge_page_mask(h); 296e63e1e5aSBadari Pulavarty 297e63e1e5aSBadari Pulavarty /* short read or no more work */ 298e63e1e5aSBadari Pulavarty if ((ret != nr) || (len == 0)) 299e63e1e5aSBadari Pulavarty break; 300e63e1e5aSBadari Pulavarty } 301e63e1e5aSBadari Pulavarty out: 302a5516438SAndi Kleen *ppos = ((loff_t)index << huge_page_shift(h)) + offset; 303e63e1e5aSBadari Pulavarty return retval; 304e63e1e5aSBadari Pulavarty } 305e63e1e5aSBadari Pulavarty 306800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 307800d15a5SNick Piggin struct address_space *mapping, 308800d15a5SNick Piggin loff_t pos, unsigned len, unsigned flags, 309800d15a5SNick Piggin struct page **pagep, void **fsdata) 3101da177e4SLinus Torvalds { 3111da177e4SLinus Torvalds return -EINVAL; 3121da177e4SLinus Torvalds } 3131da177e4SLinus Torvalds 314800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 315800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 316800d15a5SNick Piggin struct page *page, void *fsdata) 3171da177e4SLinus Torvalds { 318800d15a5SNick Piggin BUG(); 3191da177e4SLinus Torvalds return -EINVAL; 3201da177e4SLinus Torvalds } 3211da177e4SLinus Torvalds 3221da177e4SLinus Torvalds static void truncate_huge_page(struct page *page) 3231da177e4SLinus Torvalds { 324b9ea2515SKonstantin Khlebnikov ClearPageDirty(page); 3251da177e4SLinus Torvalds ClearPageUptodate(page); 326bd65cb86SMinchan Kim delete_from_page_cache(page); 3271da177e4SLinus Torvalds } 3281da177e4SLinus Torvalds 329b45b5bd6SDavid Gibson static void truncate_hugepages(struct inode *inode, loff_t lstart) 3301da177e4SLinus Torvalds { 331a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 332b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 333a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 3341da177e4SLinus Torvalds struct pagevec pvec; 3351da177e4SLinus Torvalds pgoff_t next; 336a43a8c39SChen, Kenneth W int i, freed = 0; 3371da177e4SLinus Torvalds 3381da177e4SLinus Torvalds pagevec_init(&pvec, 0); 3391da177e4SLinus Torvalds next = start; 3401da177e4SLinus Torvalds while (1) { 3411da177e4SLinus Torvalds if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 3421da177e4SLinus Torvalds if (next == start) 3431da177e4SLinus Torvalds break; 3441da177e4SLinus Torvalds next = start; 3451da177e4SLinus Torvalds continue; 3461da177e4SLinus Torvalds } 3471da177e4SLinus Torvalds 3481da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); ++i) { 3491da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3501da177e4SLinus Torvalds 3511da177e4SLinus Torvalds lock_page(page); 3521da177e4SLinus Torvalds if (page->index > next) 3531da177e4SLinus Torvalds next = page->index; 3541da177e4SLinus Torvalds ++next; 3551da177e4SLinus Torvalds truncate_huge_page(page); 3561da177e4SLinus Torvalds unlock_page(page); 357a43a8c39SChen, Kenneth W freed++; 3581da177e4SLinus Torvalds } 3591da177e4SLinus Torvalds huge_pagevec_release(&pvec); 3601da177e4SLinus Torvalds } 3611da177e4SLinus Torvalds BUG_ON(!lstart && mapping->nrpages); 362a43a8c39SChen, Kenneth W hugetlb_unreserve_pages(inode, start, freed); 3631da177e4SLinus Torvalds } 3641da177e4SLinus Torvalds 3652bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 3661da177e4SLinus Torvalds { 3679119a41eSJoonsoo Kim struct resv_map *resv_map; 3689119a41eSJoonsoo Kim 369b45b5bd6SDavid Gibson truncate_hugepages(inode, 0); 3709119a41eSJoonsoo Kim resv_map = (struct resv_map *)inode->i_mapping->private_data; 3719119a41eSJoonsoo Kim /* root inode doesn't have the resv_map, so we should check it */ 3729119a41eSJoonsoo Kim if (resv_map) 3739119a41eSJoonsoo Kim resv_map_release(&resv_map->refs); 374dbd5768fSJan Kara clear_inode(inode); 375149f4211SChristoph Hellwig } 376149f4211SChristoph Hellwig 3771da177e4SLinus Torvalds static inline void 3786b2dbba8SMichel Lespinasse hugetlb_vmtruncate_list(struct rb_root *root, pgoff_t pgoff) 3791da177e4SLinus Torvalds { 3801da177e4SLinus Torvalds struct vm_area_struct *vma; 3811da177e4SLinus Torvalds 3826b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, root, pgoff, ULONG_MAX) { 3831da177e4SLinus Torvalds unsigned long v_offset; 3841da177e4SLinus Torvalds 3851da177e4SLinus Torvalds /* 386856fc295SHugh Dickins * Can the expression below overflow on 32-bit arches? 3876b2dbba8SMichel Lespinasse * No, because the interval tree returns us only those vmas 388856fc295SHugh Dickins * which overlap the truncated area starting at pgoff, 389856fc295SHugh Dickins * and no vma on a 32-bit arch can span beyond the 4GB. 3901da177e4SLinus Torvalds */ 391856fc295SHugh Dickins if (vma->vm_pgoff < pgoff) 392856fc295SHugh Dickins v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT; 393856fc295SHugh Dickins else 3941da177e4SLinus Torvalds v_offset = 0; 3951da177e4SLinus Torvalds 39624669e58SAneesh Kumar K.V unmap_hugepage_range(vma, vma->vm_start + v_offset, 39724669e58SAneesh Kumar K.V vma->vm_end, NULL); 3981da177e4SLinus Torvalds } 3991da177e4SLinus Torvalds } 4001da177e4SLinus Torvalds 4011da177e4SLinus Torvalds static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) 4021da177e4SLinus Torvalds { 403856fc295SHugh Dickins pgoff_t pgoff; 4041da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 405a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 4061da177e4SLinus Torvalds 407a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 408856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 4091da177e4SLinus Torvalds 4107aa91e10SKen Chen i_size_write(inode, offset); 41183cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 4126b2dbba8SMichel Lespinasse if (!RB_EMPTY_ROOT(&mapping->i_mmap)) 4131da177e4SLinus Torvalds hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff); 41483cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 415b45b5bd6SDavid Gibson truncate_hugepages(inode, offset); 4161da177e4SLinus Torvalds return 0; 4171da177e4SLinus Torvalds } 4181da177e4SLinus Torvalds 4191da177e4SLinus Torvalds static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) 4201da177e4SLinus Torvalds { 4211da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 422a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 4231da177e4SLinus Torvalds int error; 4241da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds BUG_ON(!inode); 4271da177e4SLinus Torvalds 4281da177e4SLinus Torvalds error = inode_change_ok(inode, attr); 4291da177e4SLinus Torvalds if (error) 4301025774cSChristoph Hellwig return error; 4311da177e4SLinus Torvalds 4321da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 4331da177e4SLinus Torvalds error = -EINVAL; 4341025774cSChristoph Hellwig if (attr->ia_size & ~huge_page_mask(h)) 4351025774cSChristoph Hellwig return -EINVAL; 4361da177e4SLinus Torvalds error = hugetlb_vmtruncate(inode, attr->ia_size); 4371da177e4SLinus Torvalds if (error) 4381da177e4SLinus Torvalds return error; 4391da177e4SLinus Torvalds } 4401da177e4SLinus Torvalds 4411025774cSChristoph Hellwig setattr_copy(inode, attr); 4421025774cSChristoph Hellwig mark_inode_dirty(inode); 4431025774cSChristoph Hellwig return 0; 4441025774cSChristoph Hellwig } 4451025774cSChristoph Hellwig 4467d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 4477d54fa64SAl Viro struct hugetlbfs_config *config) 4481da177e4SLinus Torvalds { 4491da177e4SLinus Torvalds struct inode *inode; 4501da177e4SLinus Torvalds 4511da177e4SLinus Torvalds inode = new_inode(sb); 4521da177e4SLinus Torvalds if (inode) { 4531da177e4SLinus Torvalds struct hugetlbfs_inode_info *info; 45485fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 4557d54fa64SAl Viro inode->i_mode = S_IFDIR | config->mode; 4567d54fa64SAl Viro inode->i_uid = config->uid; 4577d54fa64SAl Viro inode->i_gid = config->gid; 4587d54fa64SAl Viro inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 4597d54fa64SAl Viro info = HUGETLBFS_I(inode); 4607d54fa64SAl Viro mpol_shared_policy_init(&info->policy, NULL); 4617d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 4627d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 4637d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 4647d54fa64SAl Viro inc_nlink(inode); 46565ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 4667d54fa64SAl Viro } 4677d54fa64SAl Viro return inode; 4687d54fa64SAl Viro } 4697d54fa64SAl Viro 470b610ded7SMichal Hocko /* 471c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 472b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 473b610ded7SMichal Hocko * annotation because huge_pmd_share() does an allocation under 474c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 475b610ded7SMichal Hocko */ 476c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 477b610ded7SMichal Hocko 4787d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 4797d54fa64SAl Viro struct inode *dir, 48018df2252SAl Viro umode_t mode, dev_t dev) 4817d54fa64SAl Viro { 4827d54fa64SAl Viro struct inode *inode; 4839119a41eSJoonsoo Kim struct resv_map *resv_map; 4849119a41eSJoonsoo Kim 4859119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 4869119a41eSJoonsoo Kim if (!resv_map) 4879119a41eSJoonsoo Kim return NULL; 4887d54fa64SAl Viro 4897d54fa64SAl Viro inode = new_inode(sb); 4907d54fa64SAl Viro if (inode) { 4917d54fa64SAl Viro struct hugetlbfs_inode_info *info; 4927d54fa64SAl Viro inode->i_ino = get_next_ino(); 4937d54fa64SAl Viro inode_init_owner(inode, dir, mode); 494c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 495c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 4961da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 4971da177e4SLinus Torvalds inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 4989119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 4991da177e4SLinus Torvalds info = HUGETLBFS_I(inode); 5006bfde05bSEric B Munson /* 5016bfde05bSEric B Munson * The policy is initialized here even if we are creating a 5026bfde05bSEric B Munson * private inode because initialization simply creates an 5036bfde05bSEric B Munson * an empty rb tree and calls spin_lock_init(), later when we 5046bfde05bSEric B Munson * call mpol_free_shared_policy() it will just return because 5056bfde05bSEric B Munson * the rb tree will still be empty. 5066bfde05bSEric B Munson */ 50771fe804bSLee Schermerhorn mpol_shared_policy_init(&info->policy, NULL); 5081da177e4SLinus Torvalds switch (mode & S_IFMT) { 5091da177e4SLinus Torvalds default: 5101da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 5111da177e4SLinus Torvalds break; 5121da177e4SLinus Torvalds case S_IFREG: 5131da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 5141da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 5151da177e4SLinus Torvalds break; 5161da177e4SLinus Torvalds case S_IFDIR: 5171da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 5181da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 5191da177e4SLinus Torvalds 5201da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 521d8c76e6fSDave Hansen inc_nlink(inode); 5221da177e4SLinus Torvalds break; 5231da177e4SLinus Torvalds case S_IFLNK: 5241da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 5251da177e4SLinus Torvalds break; 5261da177e4SLinus Torvalds } 527e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 5289119a41eSJoonsoo Kim } else 5299119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 5309119a41eSJoonsoo Kim 5311da177e4SLinus Torvalds return inode; 5321da177e4SLinus Torvalds } 5331da177e4SLinus Torvalds 5341da177e4SLinus Torvalds /* 5351da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 5361da177e4SLinus Torvalds */ 5371da177e4SLinus Torvalds static int hugetlbfs_mknod(struct inode *dir, 5381a67aafbSAl Viro struct dentry *dentry, umode_t mode, dev_t dev) 5391da177e4SLinus Torvalds { 5401da177e4SLinus Torvalds struct inode *inode; 5411da177e4SLinus Torvalds int error = -ENOSPC; 5421da177e4SLinus Torvalds 5437d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 5441da177e4SLinus Torvalds if (inode) { 5451da177e4SLinus Torvalds dir->i_ctime = dir->i_mtime = CURRENT_TIME; 5461da177e4SLinus Torvalds d_instantiate(dentry, inode); 5471da177e4SLinus Torvalds dget(dentry); /* Extra count - pin the dentry in core */ 5481da177e4SLinus Torvalds error = 0; 5491da177e4SLinus Torvalds } 5501da177e4SLinus Torvalds return error; 5511da177e4SLinus Torvalds } 5521da177e4SLinus Torvalds 55318bb1db3SAl Viro static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 5541da177e4SLinus Torvalds { 5551da177e4SLinus Torvalds int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); 5561da177e4SLinus Torvalds if (!retval) 557d8c76e6fSDave Hansen inc_nlink(dir); 5581da177e4SLinus Torvalds return retval; 5591da177e4SLinus Torvalds } 5601da177e4SLinus Torvalds 561ebfc3b49SAl Viro static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) 5621da177e4SLinus Torvalds { 5631da177e4SLinus Torvalds return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); 5641da177e4SLinus Torvalds } 5651da177e4SLinus Torvalds 5661da177e4SLinus Torvalds static int hugetlbfs_symlink(struct inode *dir, 5671da177e4SLinus Torvalds struct dentry *dentry, const char *symname) 5681da177e4SLinus Torvalds { 5691da177e4SLinus Torvalds struct inode *inode; 5701da177e4SLinus Torvalds int error = -ENOSPC; 5711da177e4SLinus Torvalds 5727d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 5731da177e4SLinus Torvalds if (inode) { 5741da177e4SLinus Torvalds int l = strlen(symname)+1; 5751da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 5761da177e4SLinus Torvalds if (!error) { 5771da177e4SLinus Torvalds d_instantiate(dentry, inode); 5781da177e4SLinus Torvalds dget(dentry); 5791da177e4SLinus Torvalds } else 5801da177e4SLinus Torvalds iput(inode); 5811da177e4SLinus Torvalds } 5821da177e4SLinus Torvalds dir->i_ctime = dir->i_mtime = CURRENT_TIME; 5831da177e4SLinus Torvalds 5841da177e4SLinus Torvalds return error; 5851da177e4SLinus Torvalds } 5861da177e4SLinus Torvalds 5871da177e4SLinus Torvalds /* 5886649a386SKen Chen * mark the head page dirty 5891da177e4SLinus Torvalds */ 5901da177e4SLinus Torvalds static int hugetlbfs_set_page_dirty(struct page *page) 5911da177e4SLinus Torvalds { 592d85f3385SChristoph Lameter struct page *head = compound_head(page); 5936649a386SKen Chen 5946649a386SKen Chen SetPageDirty(head); 5951da177e4SLinus Torvalds return 0; 5961da177e4SLinus Torvalds } 5971da177e4SLinus Torvalds 598290408d4SNaoya Horiguchi static int hugetlbfs_migrate_page(struct address_space *mapping, 599b969c4abSMel Gorman struct page *newpage, struct page *page, 600a6bc32b8SMel Gorman enum migrate_mode mode) 601290408d4SNaoya Horiguchi { 602290408d4SNaoya Horiguchi int rc; 603290408d4SNaoya Horiguchi 604290408d4SNaoya Horiguchi rc = migrate_huge_page_move_mapping(mapping, newpage, page); 60578bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 606290408d4SNaoya Horiguchi return rc; 607290408d4SNaoya Horiguchi migrate_page_copy(newpage, page); 608290408d4SNaoya Horiguchi 60978bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 610290408d4SNaoya Horiguchi } 611290408d4SNaoya Horiguchi 612726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 6131da177e4SLinus Torvalds { 614726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 615a5516438SAndi Kleen struct hstate *h = hstate_inode(dentry->d_inode); 6161da177e4SLinus Torvalds 6171da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 618a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 6191da177e4SLinus Torvalds if (sbinfo) { 6201da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 62174a8a65cSDavid Gibson /* If no limits set, just report 0 for max/free/used 62274a8a65cSDavid Gibson * blocks, like simple_statfs() */ 62390481622SDavid Gibson if (sbinfo->spool) { 62490481622SDavid Gibson long free_pages; 62590481622SDavid Gibson 62690481622SDavid Gibson spin_lock(&sbinfo->spool->lock); 62790481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 62890481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 62990481622SDavid Gibson - sbinfo->spool->used_hpages; 63090481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 63190481622SDavid Gibson spin_unlock(&sbinfo->spool->lock); 6321da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 6331da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 63474a8a65cSDavid Gibson } 6351da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 6361da177e4SLinus Torvalds } 6371da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 6381da177e4SLinus Torvalds return 0; 6391da177e4SLinus Torvalds } 6401da177e4SLinus Torvalds 6411da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 6421da177e4SLinus Torvalds { 6431da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 6441da177e4SLinus Torvalds 6451da177e4SLinus Torvalds if (sbi) { 6461da177e4SLinus Torvalds sb->s_fs_info = NULL; 64790481622SDavid Gibson 64890481622SDavid Gibson if (sbi->spool) 64990481622SDavid Gibson hugepage_put_subpool(sbi->spool); 65090481622SDavid Gibson 6511da177e4SLinus Torvalds kfree(sbi); 6521da177e4SLinus Torvalds } 6531da177e4SLinus Torvalds } 6541da177e4SLinus Torvalds 65596527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 65696527980SChristoph Hellwig { 65796527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 65896527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 65996527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 66096527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 66196527980SChristoph Hellwig return 0; 66296527980SChristoph Hellwig } 66396527980SChristoph Hellwig sbinfo->free_inodes--; 66496527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 66596527980SChristoph Hellwig } 66696527980SChristoph Hellwig 66796527980SChristoph Hellwig return 1; 66896527980SChristoph Hellwig } 66996527980SChristoph Hellwig 67096527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 67196527980SChristoph Hellwig { 67296527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 67396527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 67496527980SChristoph Hellwig sbinfo->free_inodes++; 67596527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 67696527980SChristoph Hellwig } 67796527980SChristoph Hellwig } 67896527980SChristoph Hellwig 67996527980SChristoph Hellwig 680e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 6811da177e4SLinus Torvalds 6821da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 6831da177e4SLinus Torvalds { 68496527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 6851da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 6861da177e4SLinus Torvalds 68796527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 6881da177e4SLinus Torvalds return NULL; 689e94b1766SChristoph Lameter p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); 69096527980SChristoph Hellwig if (unlikely(!p)) { 69196527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 69296527980SChristoph Hellwig return NULL; 6931da177e4SLinus Torvalds } 69496527980SChristoph Hellwig return &p->vfs_inode; 6951da177e4SLinus Torvalds } 6961da177e4SLinus Torvalds 697fa0d7e3dSNick Piggin static void hugetlbfs_i_callback(struct rcu_head *head) 698fa0d7e3dSNick Piggin { 699fa0d7e3dSNick Piggin struct inode *inode = container_of(head, struct inode, i_rcu); 700fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 701fa0d7e3dSNick Piggin } 702fa0d7e3dSNick Piggin 7031da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 7041da177e4SLinus Torvalds { 70596527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 7061da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 707fa0d7e3dSNick Piggin call_rcu(&inode->i_rcu, hugetlbfs_i_callback); 7081da177e4SLinus Torvalds } 7091da177e4SLinus Torvalds 710f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 711800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 712800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 7131da177e4SLinus Torvalds .set_page_dirty = hugetlbfs_set_page_dirty, 714290408d4SNaoya Horiguchi .migratepage = hugetlbfs_migrate_page, 7151da177e4SLinus Torvalds }; 7161da177e4SLinus Torvalds 71796527980SChristoph Hellwig 71851cc5068SAlexey Dobriyan static void init_once(void *foo) 71996527980SChristoph Hellwig { 72096527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 72196527980SChristoph Hellwig 72296527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 72396527980SChristoph Hellwig } 72496527980SChristoph Hellwig 7254b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 726e63e1e5aSBadari Pulavarty .read = hugetlbfs_read, 7271da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 7281b061d92SChristoph Hellwig .fsync = noop_fsync, 7291da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 7306038f373SArnd Bergmann .llseek = default_llseek, 7311da177e4SLinus Torvalds }; 7321da177e4SLinus Torvalds 73392e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 7341da177e4SLinus Torvalds .create = hugetlbfs_create, 7351da177e4SLinus Torvalds .lookup = simple_lookup, 7361da177e4SLinus Torvalds .link = simple_link, 7371da177e4SLinus Torvalds .unlink = simple_unlink, 7381da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 7391da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 7401da177e4SLinus Torvalds .rmdir = simple_rmdir, 7411da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 7421da177e4SLinus Torvalds .rename = simple_rename, 7431da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 7441da177e4SLinus Torvalds }; 7451da177e4SLinus Torvalds 74692e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 7471da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 7481da177e4SLinus Torvalds }; 7491da177e4SLinus Torvalds 750ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 7511da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 7521da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 7532bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 7541da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 7551da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 75610f19a86SMiklos Szeredi .show_options = generic_show_options, 7571da177e4SLinus Torvalds }; 7581da177e4SLinus Torvalds 7597ca02d0aSMike Kravetz enum { NO_SIZE, SIZE_STD, SIZE_PERCENT }; 7607ca02d0aSMike Kravetz 7617ca02d0aSMike Kravetz /* 7627ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 7637ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 7647ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 7657ca02d0aSMike Kravetz */ 7667ca02d0aSMike Kravetz static long long 7677ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 7687ca02d0aSMike Kravetz int val_type) 7697ca02d0aSMike Kravetz { 7707ca02d0aSMike Kravetz if (val_type == NO_SIZE) 7717ca02d0aSMike Kravetz return -1; 7727ca02d0aSMike Kravetz 7737ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 7747ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 7757ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 7767ca02d0aSMike Kravetz do_div(size_opt, 100); 7777ca02d0aSMike Kravetz } 7787ca02d0aSMike Kravetz 7797ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 7807ca02d0aSMike Kravetz return size_opt; 7817ca02d0aSMike Kravetz } 7827ca02d0aSMike Kravetz 7831da177e4SLinus Torvalds static int 7841da177e4SLinus Torvalds hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) 7851da177e4SLinus Torvalds { 786e73a75faSRandy Dunlap char *p, *rest; 787e73a75faSRandy Dunlap substring_t args[MAX_OPT_ARGS]; 788e73a75faSRandy Dunlap int option; 7897ca02d0aSMike Kravetz unsigned long long max_size_opt = 0, min_size_opt = 0; 7907ca02d0aSMike Kravetz int max_val_type = NO_SIZE, min_val_type = NO_SIZE; 7911da177e4SLinus Torvalds 7921da177e4SLinus Torvalds if (!options) 7931da177e4SLinus Torvalds return 0; 7941da177e4SLinus Torvalds 795e73a75faSRandy Dunlap while ((p = strsep(&options, ",")) != NULL) { 796e73a75faSRandy Dunlap int token; 797b4c07bceSLee Schermerhorn if (!*p) 798b4c07bceSLee Schermerhorn continue; 7991da177e4SLinus Torvalds 800e73a75faSRandy Dunlap token = match_token(p, tokens, args); 801e73a75faSRandy Dunlap switch (token) { 802e73a75faSRandy Dunlap case Opt_uid: 803e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 804e73a75faSRandy Dunlap goto bad_val; 805a0eb3a05SEric W. Biederman pconfig->uid = make_kuid(current_user_ns(), option); 806a0eb3a05SEric W. Biederman if (!uid_valid(pconfig->uid)) 807a0eb3a05SEric W. Biederman goto bad_val; 808e73a75faSRandy Dunlap break; 809e73a75faSRandy Dunlap 810e73a75faSRandy Dunlap case Opt_gid: 811e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 812e73a75faSRandy Dunlap goto bad_val; 813a0eb3a05SEric W. Biederman pconfig->gid = make_kgid(current_user_ns(), option); 814a0eb3a05SEric W. Biederman if (!gid_valid(pconfig->gid)) 815a0eb3a05SEric W. Biederman goto bad_val; 816e73a75faSRandy Dunlap break; 817e73a75faSRandy Dunlap 818e73a75faSRandy Dunlap case Opt_mode: 819e73a75faSRandy Dunlap if (match_octal(&args[0], &option)) 820e73a75faSRandy Dunlap goto bad_val; 82175897d60SKen Chen pconfig->mode = option & 01777U; 822e73a75faSRandy Dunlap break; 823e73a75faSRandy Dunlap 824e73a75faSRandy Dunlap case Opt_size: { 825e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 826e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 827e73a75faSRandy Dunlap goto bad_val; 8287ca02d0aSMike Kravetz max_size_opt = memparse(args[0].from, &rest); 8297ca02d0aSMike Kravetz max_val_type = SIZE_STD; 830a137e1ccSAndi Kleen if (*rest == '%') 8317ca02d0aSMike Kravetz max_val_type = SIZE_PERCENT; 832e73a75faSRandy Dunlap break; 833e73a75faSRandy Dunlap } 8341da177e4SLinus Torvalds 835e73a75faSRandy Dunlap case Opt_nr_inodes: 836e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 837e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 838e73a75faSRandy Dunlap goto bad_val; 839e73a75faSRandy Dunlap pconfig->nr_inodes = memparse(args[0].from, &rest); 840e73a75faSRandy Dunlap break; 841e73a75faSRandy Dunlap 842a137e1ccSAndi Kleen case Opt_pagesize: { 843a137e1ccSAndi Kleen unsigned long ps; 844a137e1ccSAndi Kleen ps = memparse(args[0].from, &rest); 845a137e1ccSAndi Kleen pconfig->hstate = size_to_hstate(ps); 846a137e1ccSAndi Kleen if (!pconfig->hstate) { 8479b857d26SAndrew Morton pr_err("Unsupported page size %lu MB\n", 848a137e1ccSAndi Kleen ps >> 20); 849a137e1ccSAndi Kleen return -EINVAL; 850a137e1ccSAndi Kleen } 851a137e1ccSAndi Kleen break; 852a137e1ccSAndi Kleen } 853a137e1ccSAndi Kleen 8547ca02d0aSMike Kravetz case Opt_min_size: { 8557ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 8567ca02d0aSMike Kravetz if (!isdigit(*args[0].from)) 8577ca02d0aSMike Kravetz goto bad_val; 8587ca02d0aSMike Kravetz min_size_opt = memparse(args[0].from, &rest); 8597ca02d0aSMike Kravetz min_val_type = SIZE_STD; 8607ca02d0aSMike Kravetz if (*rest == '%') 8617ca02d0aSMike Kravetz min_val_type = SIZE_PERCENT; 8627ca02d0aSMike Kravetz break; 8637ca02d0aSMike Kravetz } 8647ca02d0aSMike Kravetz 865e73a75faSRandy Dunlap default: 8669b857d26SAndrew Morton pr_err("Bad mount option: \"%s\"\n", p); 867b4c07bceSLee Schermerhorn return -EINVAL; 868e73a75faSRandy Dunlap break; 869e73a75faSRandy Dunlap } 8701da177e4SLinus Torvalds } 871a137e1ccSAndi Kleen 8727ca02d0aSMike Kravetz /* 8737ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 8747ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 8757ca02d0aSMike Kravetz */ 8767ca02d0aSMike Kravetz pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, 8777ca02d0aSMike Kravetz max_size_opt, max_val_type); 8787ca02d0aSMike Kravetz pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, 8797ca02d0aSMike Kravetz min_size_opt, min_val_type); 8807ca02d0aSMike Kravetz 8817ca02d0aSMike Kravetz /* 8827ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 8837ca02d0aSMike Kravetz */ 8847ca02d0aSMike Kravetz if (max_val_type > NO_SIZE && 8857ca02d0aSMike Kravetz pconfig->min_hpages > pconfig->max_hpages) { 8867ca02d0aSMike Kravetz pr_err("minimum size can not be greater than maximum size\n"); 8877ca02d0aSMike Kravetz return -EINVAL; 888a137e1ccSAndi Kleen } 889a137e1ccSAndi Kleen 8901da177e4SLinus Torvalds return 0; 891e73a75faSRandy Dunlap 892e73a75faSRandy Dunlap bad_val: 8939b857d26SAndrew Morton pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); 894c12ddba0SAkinobu Mita return -EINVAL; 8951da177e4SLinus Torvalds } 8961da177e4SLinus Torvalds 8971da177e4SLinus Torvalds static int 8981da177e4SLinus Torvalds hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) 8991da177e4SLinus Torvalds { 9001da177e4SLinus Torvalds int ret; 9011da177e4SLinus Torvalds struct hugetlbfs_config config; 9021da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 9031da177e4SLinus Torvalds 90410f19a86SMiklos Szeredi save_mount_options(sb, data); 90510f19a86SMiklos Szeredi 9067ca02d0aSMike Kravetz config.max_hpages = -1; /* No limit on size by default */ 9071da177e4SLinus Torvalds config.nr_inodes = -1; /* No limit on number of inodes by default */ 90877c70de1SDavid Howells config.uid = current_fsuid(); 90977c70de1SDavid Howells config.gid = current_fsgid(); 9101da177e4SLinus Torvalds config.mode = 0755; 911a137e1ccSAndi Kleen config.hstate = &default_hstate; 9127ca02d0aSMike Kravetz config.min_hpages = -1; /* No default minimum size */ 9131da177e4SLinus Torvalds ret = hugetlbfs_parse_options(data, &config); 9141da177e4SLinus Torvalds if (ret) 9151da177e4SLinus Torvalds return ret; 9161da177e4SLinus Torvalds 9171da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 9181da177e4SLinus Torvalds if (!sbinfo) 9191da177e4SLinus Torvalds return -ENOMEM; 9201da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 921a137e1ccSAndi Kleen sbinfo->hstate = config.hstate; 9221da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 9231da177e4SLinus Torvalds sbinfo->max_inodes = config.nr_inodes; 9241da177e4SLinus Torvalds sbinfo->free_inodes = config.nr_inodes; 92590481622SDavid Gibson sbinfo->spool = NULL; 9267ca02d0aSMike Kravetz /* 9277ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 9287ca02d0aSMike Kravetz * specified. Any needed reservations (for minimim size) are taken 9297ca02d0aSMike Kravetz * taken when the subpool is created. 9307ca02d0aSMike Kravetz */ 9317ca02d0aSMike Kravetz if (config.max_hpages != -1 || config.min_hpages != -1) { 9327ca02d0aSMike Kravetz sbinfo->spool = hugepage_new_subpool(config.hstate, 9337ca02d0aSMike Kravetz config.max_hpages, 9347ca02d0aSMike Kravetz config.min_hpages); 93590481622SDavid Gibson if (!sbinfo->spool) 93690481622SDavid Gibson goto out_free; 93790481622SDavid Gibson } 9381da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 939a137e1ccSAndi Kleen sb->s_blocksize = huge_page_size(config.hstate); 940a137e1ccSAndi Kleen sb->s_blocksize_bits = huge_page_shift(config.hstate); 9411da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 9421da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 9431da177e4SLinus Torvalds sb->s_time_gran = 1; 94448fde701SAl Viro sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); 94548fde701SAl Viro if (!sb->s_root) 9461da177e4SLinus Torvalds goto out_free; 9471da177e4SLinus Torvalds return 0; 9481da177e4SLinus Torvalds out_free: 94990481622SDavid Gibson kfree(sbinfo->spool); 9501da177e4SLinus Torvalds kfree(sbinfo); 9511da177e4SLinus Torvalds return -ENOMEM; 9521da177e4SLinus Torvalds } 9531da177e4SLinus Torvalds 9543c26ff6eSAl Viro static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, 9553c26ff6eSAl Viro int flags, const char *dev_name, void *data) 9561da177e4SLinus Torvalds { 9573c26ff6eSAl Viro return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); 9581da177e4SLinus Torvalds } 9591da177e4SLinus Torvalds 9601da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 9611da177e4SLinus Torvalds .name = "hugetlbfs", 9623c26ff6eSAl Viro .mount = hugetlbfs_mount, 9631da177e4SLinus Torvalds .kill_sb = kill_litter_super, 9641da177e4SLinus Torvalds }; 9657f78e035SEric W. Biederman MODULE_ALIAS_FS("hugetlbfs"); 9661da177e4SLinus Torvalds 96742d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 9681da177e4SLinus Torvalds 969ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 9701da177e4SLinus Torvalds { 971a0eb3a05SEric W. Biederman kgid_t shm_group; 972a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 973a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 9741da177e4SLinus Torvalds } 9751da177e4SLinus Torvalds 97642d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 97742d7395fSAndi Kleen { 978af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 97942d7395fSAndi Kleen 98042d7395fSAndi Kleen if (!h) 98142d7395fSAndi Kleen return -1; 98242d7395fSAndi Kleen return h - hstates; 98342d7395fSAndi Kleen } 98442d7395fSAndi Kleen 985be1d2cf5SFabian Frederick static const struct dentry_operations anon_ops = { 986118b2302SAl Viro .d_dname = simple_dname 9870df4d6e5SAl Viro }; 9880df4d6e5SAl Viro 989af73e4d9SNaoya Horiguchi /* 990af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 991af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 992af73e4d9SNaoya Horiguchi */ 993af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 994af73e4d9SNaoya Horiguchi vm_flags_t acctflag, struct user_struct **user, 99542d7395fSAndi Kleen int creat_flags, int page_size_log) 9961da177e4SLinus Torvalds { 99739b65252SAnatol Pomozov struct file *file = ERR_PTR(-ENOMEM); 9981da177e4SLinus Torvalds struct inode *inode; 9992c48b9c4SAl Viro struct path path; 10000df4d6e5SAl Viro struct super_block *sb; 10011da177e4SLinus Torvalds struct qstr quick_string; 100242d7395fSAndi Kleen int hstate_idx; 100342d7395fSAndi Kleen 100442d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 100542d7395fSAndi Kleen if (hstate_idx < 0) 100642d7395fSAndi Kleen return ERR_PTR(-ENODEV); 10071da177e4SLinus Torvalds 1008353d5c30SHugh Dickins *user = NULL; 100942d7395fSAndi Kleen if (!hugetlbfs_vfsmount[hstate_idx]) 10105bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 10115bc98594SAkinobu Mita 1012ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 1013353d5c30SHugh Dickins *user = current_user(); 1014353d5c30SHugh Dickins if (user_shm_lock(size, *user)) { 101521a3c273SDavid Rientjes task_lock(current); 10169b857d26SAndrew Morton pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", 101721a3c273SDavid Rientjes current->comm, current->pid); 101821a3c273SDavid Rientjes task_unlock(current); 1019353d5c30SHugh Dickins } else { 1020353d5c30SHugh Dickins *user = NULL; 10211da177e4SLinus Torvalds return ERR_PTR(-EPERM); 10222584e517SRavikiran G Thirumalai } 1023353d5c30SHugh Dickins } 10241da177e4SLinus Torvalds 10250df4d6e5SAl Viro sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb; 10269d66586fSEric W. Biederman quick_string.name = name; 10271da177e4SLinus Torvalds quick_string.len = strlen(quick_string.name); 10281da177e4SLinus Torvalds quick_string.hash = 0; 10290df4d6e5SAl Viro path.dentry = d_alloc_pseudo(sb, &quick_string); 10302c48b9c4SAl Viro if (!path.dentry) 10311da177e4SLinus Torvalds goto out_shm_unlock; 10321da177e4SLinus Torvalds 10330df4d6e5SAl Viro d_set_d_op(path.dentry, &anon_ops); 103442d7395fSAndi Kleen path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); 103539b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 10360df4d6e5SAl Viro inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0); 10371da177e4SLinus Torvalds if (!inode) 1038ce8d2cdfSDave Hansen goto out_dentry; 10391da177e4SLinus Torvalds 104039b65252SAnatol Pomozov file = ERR_PTR(-ENOMEM); 1041af73e4d9SNaoya Horiguchi if (hugetlb_reserve_pages(inode, 0, 1042af73e4d9SNaoya Horiguchi size >> huge_page_shift(hstate_inode(inode)), NULL, 1043af73e4d9SNaoya Horiguchi acctflag)) 1044b45b5bd6SDavid Gibson goto out_inode; 1045b45b5bd6SDavid Gibson 10462c48b9c4SAl Viro d_instantiate(path.dentry, inode); 10471da177e4SLinus Torvalds inode->i_size = size; 10486d6b77f1SMiklos Szeredi clear_nlink(inode); 1049ce8d2cdfSDave Hansen 10502c48b9c4SAl Viro file = alloc_file(&path, FMODE_WRITE | FMODE_READ, 1051ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 105239b65252SAnatol Pomozov if (IS_ERR(file)) 1053b4d232e6SAl Viro goto out_dentry; /* inode is already attached */ 1054ce8d2cdfSDave Hansen 10551da177e4SLinus Torvalds return file; 10561da177e4SLinus Torvalds 1057b45b5bd6SDavid Gibson out_inode: 1058b45b5bd6SDavid Gibson iput(inode); 10591da177e4SLinus Torvalds out_dentry: 10602c48b9c4SAl Viro path_put(&path); 10611da177e4SLinus Torvalds out_shm_unlock: 1062353d5c30SHugh Dickins if (*user) { 1063353d5c30SHugh Dickins user_shm_unlock(size, *user); 1064353d5c30SHugh Dickins *user = NULL; 1065353d5c30SHugh Dickins } 106639b65252SAnatol Pomozov return file; 10671da177e4SLinus Torvalds } 10681da177e4SLinus Torvalds 10691da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 10701da177e4SLinus Torvalds { 107142d7395fSAndi Kleen struct hstate *h; 10721da177e4SLinus Torvalds int error; 107342d7395fSAndi Kleen int i; 10741da177e4SLinus Torvalds 1075457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 10769b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1077457c1b27SNishanth Aravamudan return -ENOTSUPP; 1078457c1b27SNishanth Aravamudan } 1079457c1b27SNishanth Aravamudan 1080d1d5e05fSHillf Danton error = -ENOMEM; 10811da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 10821da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 108320c2df83SPaul Mundt 0, 0, init_once); 10841da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 1085e0bf68ddSPeter Zijlstra goto out2; 10861da177e4SLinus Torvalds 10871da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 10881da177e4SLinus Torvalds if (error) 10891da177e4SLinus Torvalds goto out; 10901da177e4SLinus Torvalds 109142d7395fSAndi Kleen i = 0; 109242d7395fSAndi Kleen for_each_hstate(h) { 109342d7395fSAndi Kleen char buf[50]; 109442d7395fSAndi Kleen unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); 10951da177e4SLinus Torvalds 109642d7395fSAndi Kleen snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); 109742d7395fSAndi Kleen hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, 109842d7395fSAndi Kleen buf); 109942d7395fSAndi Kleen 110042d7395fSAndi Kleen if (IS_ERR(hugetlbfs_vfsmount[i])) { 11019b857d26SAndrew Morton pr_err("Cannot mount internal hugetlbfs for " 110242d7395fSAndi Kleen "page size %uK", ps_kb); 110342d7395fSAndi Kleen error = PTR_ERR(hugetlbfs_vfsmount[i]); 110442d7395fSAndi Kleen hugetlbfs_vfsmount[i] = NULL; 11051da177e4SLinus Torvalds } 110642d7395fSAndi Kleen i++; 110742d7395fSAndi Kleen } 110842d7395fSAndi Kleen /* Non default hstates are optional */ 110942d7395fSAndi Kleen if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) 111042d7395fSAndi Kleen return 0; 11111da177e4SLinus Torvalds 11121da177e4SLinus Torvalds out: 11131da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 1114e0bf68ddSPeter Zijlstra out2: 11151da177e4SLinus Torvalds return error; 11161da177e4SLinus Torvalds } 11171da177e4SLinus Torvalds 11181da177e4SLinus Torvalds static void __exit exit_hugetlbfs_fs(void) 11191da177e4SLinus Torvalds { 112042d7395fSAndi Kleen struct hstate *h; 112142d7395fSAndi Kleen int i; 112242d7395fSAndi Kleen 112342d7395fSAndi Kleen 11248c0a8537SKirill A. Shutemov /* 11258c0a8537SKirill A. Shutemov * Make sure all delayed rcu free inodes are flushed before we 11268c0a8537SKirill A. Shutemov * destroy cache. 11278c0a8537SKirill A. Shutemov */ 11288c0a8537SKirill A. Shutemov rcu_barrier(); 11291da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 113042d7395fSAndi Kleen i = 0; 113142d7395fSAndi Kleen for_each_hstate(h) 113242d7395fSAndi Kleen kern_unmount(hugetlbfs_vfsmount[i++]); 11331da177e4SLinus Torvalds unregister_filesystem(&hugetlbfs_fs_type); 11341da177e4SLinus Torvalds } 11351da177e4SLinus Torvalds 11361da177e4SLinus Torvalds module_init(init_hugetlbfs_fs) 11371da177e4SLinus Torvalds module_exit(exit_hugetlbfs_fs) 11381da177e4SLinus Torvalds 11391da177e4SLinus Torvalds MODULE_LICENSE("GPL"); 1140