11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * hugetlbpage-backed filesystem. Based on ramfs. 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * William Irwin, 2002 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 2002 Linus Torvalds. 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/module.h> 101da177e4SLinus Torvalds #include <linux/thread_info.h> 111da177e4SLinus Torvalds #include <asm/current.h> 121da177e4SLinus Torvalds #include <linux/sched.h> /* remove ASAP */ 131da177e4SLinus Torvalds #include <linux/fs.h> 141da177e4SLinus Torvalds #include <linux/mount.h> 151da177e4SLinus Torvalds #include <linux/file.h> 16e73a75faSRandy Dunlap #include <linux/kernel.h> 171da177e4SLinus Torvalds #include <linux/writeback.h> 181da177e4SLinus Torvalds #include <linux/pagemap.h> 191da177e4SLinus Torvalds #include <linux/highmem.h> 201da177e4SLinus Torvalds #include <linux/init.h> 211da177e4SLinus Torvalds #include <linux/string.h> 2216f7e0feSRandy Dunlap #include <linux/capability.h> 23e73a75faSRandy Dunlap #include <linux/ctype.h> 241da177e4SLinus Torvalds #include <linux/backing-dev.h> 251da177e4SLinus Torvalds #include <linux/hugetlb.h> 261da177e4SLinus Torvalds #include <linux/pagevec.h> 27e73a75faSRandy Dunlap #include <linux/parser.h> 28036e0856SBenjamin Herrenschmidt #include <linux/mman.h> 291da177e4SLinus Torvalds #include <linux/slab.h> 301da177e4SLinus Torvalds #include <linux/dnotify.h> 311da177e4SLinus Torvalds #include <linux/statfs.h> 321da177e4SLinus Torvalds #include <linux/security.h> 331fd7317dSNick Black #include <linux/magic.h> 34290408d4SNaoya Horiguchi #include <linux/migrate.h> 351da177e4SLinus Torvalds 361da177e4SLinus Torvalds #include <asm/uaccess.h> 371da177e4SLinus Torvalds 38ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops; 39f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops; 404b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations; 4192e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations; 4292e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations; 431da177e4SLinus Torvalds 44a1d776eeSDavid Gibson struct hugetlbfs_config { 45a0eb3a05SEric W. Biederman kuid_t uid; 46a0eb3a05SEric W. Biederman kgid_t gid; 47a1d776eeSDavid Gibson umode_t mode; 48a1d776eeSDavid Gibson long nr_blocks; 49a1d776eeSDavid Gibson long nr_inodes; 50a1d776eeSDavid Gibson struct hstate *hstate; 51a1d776eeSDavid Gibson }; 52a1d776eeSDavid Gibson 53a1d776eeSDavid Gibson struct hugetlbfs_inode_info { 54a1d776eeSDavid Gibson struct shared_policy policy; 55a1d776eeSDavid Gibson struct inode vfs_inode; 56a1d776eeSDavid Gibson }; 57a1d776eeSDavid Gibson 58a1d776eeSDavid Gibson static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode) 59a1d776eeSDavid Gibson { 60a1d776eeSDavid Gibson return container_of(inode, struct hugetlbfs_inode_info, vfs_inode); 61a1d776eeSDavid Gibson } 62a1d776eeSDavid Gibson 631da177e4SLinus Torvalds static struct backing_dev_info hugetlbfs_backing_dev_info = { 64d993831fSJens Axboe .name = "hugetlbfs", 651da177e4SLinus Torvalds .ra_pages = 0, /* No readahead */ 66e4ad08feSMiklos Szeredi .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, 671da177e4SLinus Torvalds }; 681da177e4SLinus Torvalds 691da177e4SLinus Torvalds int sysctl_hugetlb_shm_group; 701da177e4SLinus Torvalds 71e73a75faSRandy Dunlap enum { 72e73a75faSRandy Dunlap Opt_size, Opt_nr_inodes, 73e73a75faSRandy Dunlap Opt_mode, Opt_uid, Opt_gid, 74a137e1ccSAndi Kleen Opt_pagesize, 75e73a75faSRandy Dunlap Opt_err, 76e73a75faSRandy Dunlap }; 77e73a75faSRandy Dunlap 78a447c093SSteven Whitehouse static const match_table_t tokens = { 79e73a75faSRandy Dunlap {Opt_size, "size=%s"}, 80e73a75faSRandy Dunlap {Opt_nr_inodes, "nr_inodes=%s"}, 81e73a75faSRandy Dunlap {Opt_mode, "mode=%o"}, 82e73a75faSRandy Dunlap {Opt_uid, "uid=%u"}, 83e73a75faSRandy Dunlap {Opt_gid, "gid=%u"}, 84a137e1ccSAndi Kleen {Opt_pagesize, "pagesize=%s"}, 85e73a75faSRandy Dunlap {Opt_err, NULL}, 86e73a75faSRandy Dunlap }; 87e73a75faSRandy Dunlap 882e9b367cSAdam Litke static void huge_pagevec_release(struct pagevec *pvec) 892e9b367cSAdam Litke { 902e9b367cSAdam Litke int i; 912e9b367cSAdam Litke 922e9b367cSAdam Litke for (i = 0; i < pagevec_count(pvec); ++i) 932e9b367cSAdam Litke put_page(pvec->pages[i]); 942e9b367cSAdam Litke 952e9b367cSAdam Litke pagevec_reinit(pvec); 962e9b367cSAdam Litke } 972e9b367cSAdam Litke 981da177e4SLinus Torvalds static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) 991da177e4SLinus Torvalds { 100b39424e2SJosef Sipek struct inode *inode = file->f_path.dentry->d_inode; 1011da177e4SLinus Torvalds loff_t len, vma_len; 1021da177e4SLinus Torvalds int ret; 103a5516438SAndi Kleen struct hstate *h = hstate_file(file); 1041da177e4SLinus Torvalds 10568589bc3SHugh Dickins /* 106dec4ad86SDavid Gibson * vma address alignment (but not the pgoff alignment) has 107dec4ad86SDavid Gibson * already been checked by prepare_hugepage_range. If you add 108dec4ad86SDavid Gibson * any error returns here, do so after setting VM_HUGETLB, so 109dec4ad86SDavid Gibson * is_vm_hugetlb_page tests below unmap_region go the right 110dec4ad86SDavid Gibson * way when do_mmap_pgoff unwinds (may be important on powerpc 111dec4ad86SDavid Gibson * and ia64). 11268589bc3SHugh Dickins */ 113314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND | VM_DONTDUMP; 11468589bc3SHugh Dickins vma->vm_ops = &hugetlb_vm_ops; 1151da177e4SLinus Torvalds 1162b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 117dec4ad86SDavid Gibson return -EINVAL; 118dec4ad86SDavid Gibson 1191da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 1201da177e4SLinus Torvalds 1211b1dcc1bSJes Sorensen mutex_lock(&inode->i_mutex); 1221da177e4SLinus Torvalds file_accessed(file); 1231da177e4SLinus Torvalds 1241da177e4SLinus Torvalds ret = -ENOMEM; 1251da177e4SLinus Torvalds len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 1261da177e4SLinus Torvalds 127a1e78772SMel Gorman if (hugetlb_reserve_pages(inode, 128a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1295a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1305a6fe125SMel Gorman vma->vm_flags)) 131b45b5bd6SDavid Gibson goto out; 132b45b5bd6SDavid Gibson 1334c887265SAdam Litke ret = 0; 1344c887265SAdam Litke hugetlb_prefault_arch_hook(vma->vm_mm); 135b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 1361da177e4SLinus Torvalds inode->i_size = len; 1371da177e4SLinus Torvalds out: 1381b1dcc1bSJes Sorensen mutex_unlock(&inode->i_mutex); 1391da177e4SLinus Torvalds 1401da177e4SLinus Torvalds return ret; 1411da177e4SLinus Torvalds } 1421da177e4SLinus Torvalds 1431da177e4SLinus Torvalds /* 144508034a3SHugh Dickins * Called under down_write(mmap_sem). 1451da177e4SLinus Torvalds */ 1461da177e4SLinus Torvalds 147d2ba27e8SAdrian Bunk #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 1481da177e4SLinus Torvalds static unsigned long 1491da177e4SLinus Torvalds hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 1501da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 1511da177e4SLinus Torvalds { 1521da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1531da177e4SLinus Torvalds struct vm_area_struct *vma; 154a5516438SAndi Kleen struct hstate *h = hstate_file(file); 15508659355SMichel Lespinasse struct vm_unmapped_area_info info; 1561da177e4SLinus Torvalds 157a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 1581da177e4SLinus Torvalds return -EINVAL; 1591da177e4SLinus Torvalds if (len > TASK_SIZE) 1601da177e4SLinus Torvalds return -ENOMEM; 1611da177e4SLinus Torvalds 162036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 163a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 164036e0856SBenjamin Herrenschmidt return -EINVAL; 165036e0856SBenjamin Herrenschmidt return addr; 166036e0856SBenjamin Herrenschmidt } 167036e0856SBenjamin Herrenschmidt 1681da177e4SLinus Torvalds if (addr) { 169a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 1701da177e4SLinus Torvalds vma = find_vma(mm, addr); 1711da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 1721da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 1731da177e4SLinus Torvalds return addr; 1741da177e4SLinus Torvalds } 1751da177e4SLinus Torvalds 17608659355SMichel Lespinasse info.flags = 0; 17708659355SMichel Lespinasse info.length = len; 17808659355SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 17908659355SMichel Lespinasse info.high_limit = TASK_SIZE; 18008659355SMichel Lespinasse info.align_mask = PAGE_MASK & ~huge_page_mask(h); 18108659355SMichel Lespinasse info.align_offset = 0; 18208659355SMichel Lespinasse return vm_unmapped_area(&info); 1831da177e4SLinus Torvalds } 1841da177e4SLinus Torvalds #endif 1851da177e4SLinus Torvalds 186e63e1e5aSBadari Pulavarty static int 187e63e1e5aSBadari Pulavarty hugetlbfs_read_actor(struct page *page, unsigned long offset, 188e63e1e5aSBadari Pulavarty char __user *buf, unsigned long count, 189e63e1e5aSBadari Pulavarty unsigned long size) 190e63e1e5aSBadari Pulavarty { 191e63e1e5aSBadari Pulavarty char *kaddr; 192e63e1e5aSBadari Pulavarty unsigned long left, copied = 0; 193e63e1e5aSBadari Pulavarty int i, chunksize; 194e63e1e5aSBadari Pulavarty 195e63e1e5aSBadari Pulavarty if (size > count) 196e63e1e5aSBadari Pulavarty size = count; 197e63e1e5aSBadari Pulavarty 198e63e1e5aSBadari Pulavarty /* Find which 4k chunk and offset with in that chunk */ 199e63e1e5aSBadari Pulavarty i = offset >> PAGE_CACHE_SHIFT; 200e63e1e5aSBadari Pulavarty offset = offset & ~PAGE_CACHE_MASK; 201e63e1e5aSBadari Pulavarty 202e63e1e5aSBadari Pulavarty while (size) { 203e63e1e5aSBadari Pulavarty chunksize = PAGE_CACHE_SIZE; 204e63e1e5aSBadari Pulavarty if (offset) 205e63e1e5aSBadari Pulavarty chunksize -= offset; 206e63e1e5aSBadari Pulavarty if (chunksize > size) 207e63e1e5aSBadari Pulavarty chunksize = size; 208e63e1e5aSBadari Pulavarty kaddr = kmap(&page[i]); 209e63e1e5aSBadari Pulavarty left = __copy_to_user(buf, kaddr + offset, chunksize); 210e63e1e5aSBadari Pulavarty kunmap(&page[i]); 211e63e1e5aSBadari Pulavarty if (left) { 212e63e1e5aSBadari Pulavarty copied += (chunksize - left); 213e63e1e5aSBadari Pulavarty break; 214e63e1e5aSBadari Pulavarty } 215e63e1e5aSBadari Pulavarty offset = 0; 216e63e1e5aSBadari Pulavarty size -= chunksize; 217e63e1e5aSBadari Pulavarty buf += chunksize; 218e63e1e5aSBadari Pulavarty copied += chunksize; 219e63e1e5aSBadari Pulavarty i++; 220e63e1e5aSBadari Pulavarty } 221e63e1e5aSBadari Pulavarty return copied ? copied : -EFAULT; 222e63e1e5aSBadari Pulavarty } 223e63e1e5aSBadari Pulavarty 224e63e1e5aSBadari Pulavarty /* 225e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 226e63e1e5aSBadari Pulavarty * data. Its *very* similar to do_generic_mapping_read(), we can't use that 227e63e1e5aSBadari Pulavarty * since it has PAGE_CACHE_SIZE assumptions. 228e63e1e5aSBadari Pulavarty */ 229e63e1e5aSBadari Pulavarty static ssize_t hugetlbfs_read(struct file *filp, char __user *buf, 230e63e1e5aSBadari Pulavarty size_t len, loff_t *ppos) 231e63e1e5aSBadari Pulavarty { 232a5516438SAndi Kleen struct hstate *h = hstate_file(filp); 233e63e1e5aSBadari Pulavarty struct address_space *mapping = filp->f_mapping; 234e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 235a5516438SAndi Kleen unsigned long index = *ppos >> huge_page_shift(h); 236a5516438SAndi Kleen unsigned long offset = *ppos & ~huge_page_mask(h); 237e63e1e5aSBadari Pulavarty unsigned long end_index; 238e63e1e5aSBadari Pulavarty loff_t isize; 239e63e1e5aSBadari Pulavarty ssize_t retval = 0; 240e63e1e5aSBadari Pulavarty 241e63e1e5aSBadari Pulavarty /* validate length */ 242e63e1e5aSBadari Pulavarty if (len == 0) 243e63e1e5aSBadari Pulavarty goto out; 244e63e1e5aSBadari Pulavarty 245e63e1e5aSBadari Pulavarty for (;;) { 246e63e1e5aSBadari Pulavarty struct page *page; 247a5516438SAndi Kleen unsigned long nr, ret; 24891bf189cSRoel Kluin int ra; 249e63e1e5aSBadari Pulavarty 250e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 251a5516438SAndi Kleen nr = huge_page_size(h); 252a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 253a05b0855SAneesh Kumar K.V if (!isize) 254a05b0855SAneesh Kumar K.V goto out; 255a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 256e63e1e5aSBadari Pulavarty if (index >= end_index) { 257e63e1e5aSBadari Pulavarty if (index > end_index) 258e63e1e5aSBadari Pulavarty goto out; 259a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 260a05b0855SAneesh Kumar K.V if (nr <= offset) 261e63e1e5aSBadari Pulavarty goto out; 262e63e1e5aSBadari Pulavarty } 263e63e1e5aSBadari Pulavarty nr = nr - offset; 264e63e1e5aSBadari Pulavarty 265e63e1e5aSBadari Pulavarty /* Find the page */ 266a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 267e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 268e63e1e5aSBadari Pulavarty /* 269e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 270e63e1e5aSBadari Pulavarty * length of the hole or request. 271e63e1e5aSBadari Pulavarty */ 272e63e1e5aSBadari Pulavarty ret = len < nr ? len : nr; 273e63e1e5aSBadari Pulavarty if (clear_user(buf, ret)) 27491bf189cSRoel Kluin ra = -EFAULT; 27591bf189cSRoel Kluin else 27691bf189cSRoel Kluin ra = 0; 277e63e1e5aSBadari Pulavarty } else { 278a05b0855SAneesh Kumar K.V unlock_page(page); 279a05b0855SAneesh Kumar K.V 280e63e1e5aSBadari Pulavarty /* 281e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 282e63e1e5aSBadari Pulavarty */ 28391bf189cSRoel Kluin ra = hugetlbfs_read_actor(page, offset, buf, len, nr); 28491bf189cSRoel Kluin ret = ra; 285a05b0855SAneesh Kumar K.V page_cache_release(page); 286e63e1e5aSBadari Pulavarty } 28791bf189cSRoel Kluin if (ra < 0) { 288e63e1e5aSBadari Pulavarty if (retval == 0) 28991bf189cSRoel Kluin retval = ra; 290e63e1e5aSBadari Pulavarty goto out; 291e63e1e5aSBadari Pulavarty } 292e63e1e5aSBadari Pulavarty 293e63e1e5aSBadari Pulavarty offset += ret; 294e63e1e5aSBadari Pulavarty retval += ret; 295e63e1e5aSBadari Pulavarty len -= ret; 296a5516438SAndi Kleen index += offset >> huge_page_shift(h); 297a5516438SAndi Kleen offset &= ~huge_page_mask(h); 298e63e1e5aSBadari Pulavarty 299e63e1e5aSBadari Pulavarty /* short read or no more work */ 300e63e1e5aSBadari Pulavarty if ((ret != nr) || (len == 0)) 301e63e1e5aSBadari Pulavarty break; 302e63e1e5aSBadari Pulavarty } 303e63e1e5aSBadari Pulavarty out: 304a5516438SAndi Kleen *ppos = ((loff_t)index << huge_page_shift(h)) + offset; 305e63e1e5aSBadari Pulavarty return retval; 306e63e1e5aSBadari Pulavarty } 307e63e1e5aSBadari Pulavarty 308800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 309800d15a5SNick Piggin struct address_space *mapping, 310800d15a5SNick Piggin loff_t pos, unsigned len, unsigned flags, 311800d15a5SNick Piggin struct page **pagep, void **fsdata) 3121da177e4SLinus Torvalds { 3131da177e4SLinus Torvalds return -EINVAL; 3141da177e4SLinus Torvalds } 3151da177e4SLinus Torvalds 316800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 317800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 318800d15a5SNick Piggin struct page *page, void *fsdata) 3191da177e4SLinus Torvalds { 320800d15a5SNick Piggin BUG(); 3211da177e4SLinus Torvalds return -EINVAL; 3221da177e4SLinus Torvalds } 3231da177e4SLinus Torvalds 3241da177e4SLinus Torvalds static void truncate_huge_page(struct page *page) 3251da177e4SLinus Torvalds { 326fba2591bSLinus Torvalds cancel_dirty_page(page, /* No IO accounting for huge pages? */0); 3271da177e4SLinus Torvalds ClearPageUptodate(page); 328bd65cb86SMinchan Kim delete_from_page_cache(page); 3291da177e4SLinus Torvalds } 3301da177e4SLinus Torvalds 331b45b5bd6SDavid Gibson static void truncate_hugepages(struct inode *inode, loff_t lstart) 3321da177e4SLinus Torvalds { 333a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 334b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 335a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 3361da177e4SLinus Torvalds struct pagevec pvec; 3371da177e4SLinus Torvalds pgoff_t next; 338a43a8c39SChen, Kenneth W int i, freed = 0; 3391da177e4SLinus Torvalds 3401da177e4SLinus Torvalds pagevec_init(&pvec, 0); 3411da177e4SLinus Torvalds next = start; 3421da177e4SLinus Torvalds while (1) { 3431da177e4SLinus Torvalds if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 3441da177e4SLinus Torvalds if (next == start) 3451da177e4SLinus Torvalds break; 3461da177e4SLinus Torvalds next = start; 3471da177e4SLinus Torvalds continue; 3481da177e4SLinus Torvalds } 3491da177e4SLinus Torvalds 3501da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); ++i) { 3511da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3521da177e4SLinus Torvalds 3531da177e4SLinus Torvalds lock_page(page); 3541da177e4SLinus Torvalds if (page->index > next) 3551da177e4SLinus Torvalds next = page->index; 3561da177e4SLinus Torvalds ++next; 3571da177e4SLinus Torvalds truncate_huge_page(page); 3581da177e4SLinus Torvalds unlock_page(page); 359a43a8c39SChen, Kenneth W freed++; 3601da177e4SLinus Torvalds } 3611da177e4SLinus Torvalds huge_pagevec_release(&pvec); 3621da177e4SLinus Torvalds } 3631da177e4SLinus Torvalds BUG_ON(!lstart && mapping->nrpages); 364a43a8c39SChen, Kenneth W hugetlb_unreserve_pages(inode, start, freed); 3651da177e4SLinus Torvalds } 3661da177e4SLinus Torvalds 3672bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 3681da177e4SLinus Torvalds { 369b45b5bd6SDavid Gibson truncate_hugepages(inode, 0); 370dbd5768fSJan Kara clear_inode(inode); 371149f4211SChristoph Hellwig } 372149f4211SChristoph Hellwig 3731da177e4SLinus Torvalds static inline void 3746b2dbba8SMichel Lespinasse hugetlb_vmtruncate_list(struct rb_root *root, pgoff_t pgoff) 3751da177e4SLinus Torvalds { 3761da177e4SLinus Torvalds struct vm_area_struct *vma; 3771da177e4SLinus Torvalds 3786b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, root, pgoff, ULONG_MAX) { 3791da177e4SLinus Torvalds unsigned long v_offset; 3801da177e4SLinus Torvalds 3811da177e4SLinus Torvalds /* 382856fc295SHugh Dickins * Can the expression below overflow on 32-bit arches? 3836b2dbba8SMichel Lespinasse * No, because the interval tree returns us only those vmas 384856fc295SHugh Dickins * which overlap the truncated area starting at pgoff, 385856fc295SHugh Dickins * and no vma on a 32-bit arch can span beyond the 4GB. 3861da177e4SLinus Torvalds */ 387856fc295SHugh Dickins if (vma->vm_pgoff < pgoff) 388856fc295SHugh Dickins v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT; 389856fc295SHugh Dickins else 3901da177e4SLinus Torvalds v_offset = 0; 3911da177e4SLinus Torvalds 39224669e58SAneesh Kumar K.V unmap_hugepage_range(vma, vma->vm_start + v_offset, 39324669e58SAneesh Kumar K.V vma->vm_end, NULL); 3941da177e4SLinus Torvalds } 3951da177e4SLinus Torvalds } 3961da177e4SLinus Torvalds 3971da177e4SLinus Torvalds static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) 3981da177e4SLinus Torvalds { 399856fc295SHugh Dickins pgoff_t pgoff; 4001da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 401a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 4021da177e4SLinus Torvalds 403a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 404856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 4051da177e4SLinus Torvalds 4067aa91e10SKen Chen i_size_write(inode, offset); 4073d48ae45SPeter Zijlstra mutex_lock(&mapping->i_mmap_mutex); 4086b2dbba8SMichel Lespinasse if (!RB_EMPTY_ROOT(&mapping->i_mmap)) 4091da177e4SLinus Torvalds hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff); 4103d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 411b45b5bd6SDavid Gibson truncate_hugepages(inode, offset); 4121da177e4SLinus Torvalds return 0; 4131da177e4SLinus Torvalds } 4141da177e4SLinus Torvalds 4151da177e4SLinus Torvalds static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) 4161da177e4SLinus Torvalds { 4171da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 418a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 4191da177e4SLinus Torvalds int error; 4201da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 4211da177e4SLinus Torvalds 4221da177e4SLinus Torvalds BUG_ON(!inode); 4231da177e4SLinus Torvalds 4241da177e4SLinus Torvalds error = inode_change_ok(inode, attr); 4251da177e4SLinus Torvalds if (error) 4261025774cSChristoph Hellwig return error; 4271da177e4SLinus Torvalds 4281da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 4291da177e4SLinus Torvalds error = -EINVAL; 4301025774cSChristoph Hellwig if (attr->ia_size & ~huge_page_mask(h)) 4311025774cSChristoph Hellwig return -EINVAL; 4321da177e4SLinus Torvalds error = hugetlb_vmtruncate(inode, attr->ia_size); 4331da177e4SLinus Torvalds if (error) 4341da177e4SLinus Torvalds return error; 4351da177e4SLinus Torvalds } 4361da177e4SLinus Torvalds 4371025774cSChristoph Hellwig setattr_copy(inode, attr); 4381025774cSChristoph Hellwig mark_inode_dirty(inode); 4391025774cSChristoph Hellwig return 0; 4401025774cSChristoph Hellwig } 4411025774cSChristoph Hellwig 4427d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 4437d54fa64SAl Viro struct hugetlbfs_config *config) 4441da177e4SLinus Torvalds { 4451da177e4SLinus Torvalds struct inode *inode; 4461da177e4SLinus Torvalds 4471da177e4SLinus Torvalds inode = new_inode(sb); 4481da177e4SLinus Torvalds if (inode) { 4491da177e4SLinus Torvalds struct hugetlbfs_inode_info *info; 45085fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 4517d54fa64SAl Viro inode->i_mode = S_IFDIR | config->mode; 4527d54fa64SAl Viro inode->i_uid = config->uid; 4537d54fa64SAl Viro inode->i_gid = config->gid; 4547d54fa64SAl Viro inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 4557d54fa64SAl Viro info = HUGETLBFS_I(inode); 4567d54fa64SAl Viro mpol_shared_policy_init(&info->policy, NULL); 4577d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 4587d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 4597d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 4607d54fa64SAl Viro inc_nlink(inode); 46165ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 4627d54fa64SAl Viro } 4637d54fa64SAl Viro return inode; 4647d54fa64SAl Viro } 4657d54fa64SAl Viro 4667d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 4677d54fa64SAl Viro struct inode *dir, 46818df2252SAl Viro umode_t mode, dev_t dev) 4697d54fa64SAl Viro { 4707d54fa64SAl Viro struct inode *inode; 4717d54fa64SAl Viro 4727d54fa64SAl Viro inode = new_inode(sb); 4737d54fa64SAl Viro if (inode) { 4747d54fa64SAl Viro struct hugetlbfs_inode_info *info; 4757d54fa64SAl Viro inode->i_ino = get_next_ino(); 4767d54fa64SAl Viro inode_init_owner(inode, dir, mode); 4771da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 4781da177e4SLinus Torvalds inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info; 4791da177e4SLinus Torvalds inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 480a43a8c39SChen, Kenneth W INIT_LIST_HEAD(&inode->i_mapping->private_list); 4811da177e4SLinus Torvalds info = HUGETLBFS_I(inode); 4826bfde05bSEric B Munson /* 4836bfde05bSEric B Munson * The policy is initialized here even if we are creating a 4846bfde05bSEric B Munson * private inode because initialization simply creates an 4856bfde05bSEric B Munson * an empty rb tree and calls spin_lock_init(), later when we 4866bfde05bSEric B Munson * call mpol_free_shared_policy() it will just return because 4876bfde05bSEric B Munson * the rb tree will still be empty. 4886bfde05bSEric B Munson */ 48971fe804bSLee Schermerhorn mpol_shared_policy_init(&info->policy, NULL); 4901da177e4SLinus Torvalds switch (mode & S_IFMT) { 4911da177e4SLinus Torvalds default: 4921da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 4931da177e4SLinus Torvalds break; 4941da177e4SLinus Torvalds case S_IFREG: 4951da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 4961da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 4971da177e4SLinus Torvalds break; 4981da177e4SLinus Torvalds case S_IFDIR: 4991da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 5001da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 5011da177e4SLinus Torvalds 5021da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 503d8c76e6fSDave Hansen inc_nlink(inode); 5041da177e4SLinus Torvalds break; 5051da177e4SLinus Torvalds case S_IFLNK: 5061da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 5071da177e4SLinus Torvalds break; 5081da177e4SLinus Torvalds } 509e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 5101da177e4SLinus Torvalds } 5111da177e4SLinus Torvalds return inode; 5121da177e4SLinus Torvalds } 5131da177e4SLinus Torvalds 5141da177e4SLinus Torvalds /* 5151da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 5161da177e4SLinus Torvalds */ 5171da177e4SLinus Torvalds static int hugetlbfs_mknod(struct inode *dir, 5181a67aafbSAl Viro struct dentry *dentry, umode_t mode, dev_t dev) 5191da177e4SLinus Torvalds { 5201da177e4SLinus Torvalds struct inode *inode; 5211da177e4SLinus Torvalds int error = -ENOSPC; 5221da177e4SLinus Torvalds 5237d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 5241da177e4SLinus Torvalds if (inode) { 5251da177e4SLinus Torvalds dir->i_ctime = dir->i_mtime = CURRENT_TIME; 5261da177e4SLinus Torvalds d_instantiate(dentry, inode); 5271da177e4SLinus Torvalds dget(dentry); /* Extra count - pin the dentry in core */ 5281da177e4SLinus Torvalds error = 0; 5291da177e4SLinus Torvalds } 5301da177e4SLinus Torvalds return error; 5311da177e4SLinus Torvalds } 5321da177e4SLinus Torvalds 53318bb1db3SAl Viro static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 5341da177e4SLinus Torvalds { 5351da177e4SLinus Torvalds int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); 5361da177e4SLinus Torvalds if (!retval) 537d8c76e6fSDave Hansen inc_nlink(dir); 5381da177e4SLinus Torvalds return retval; 5391da177e4SLinus Torvalds } 5401da177e4SLinus Torvalds 541ebfc3b49SAl Viro static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) 5421da177e4SLinus Torvalds { 5431da177e4SLinus Torvalds return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); 5441da177e4SLinus Torvalds } 5451da177e4SLinus Torvalds 5461da177e4SLinus Torvalds static int hugetlbfs_symlink(struct inode *dir, 5471da177e4SLinus Torvalds struct dentry *dentry, const char *symname) 5481da177e4SLinus Torvalds { 5491da177e4SLinus Torvalds struct inode *inode; 5501da177e4SLinus Torvalds int error = -ENOSPC; 5511da177e4SLinus Torvalds 5527d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 5531da177e4SLinus Torvalds if (inode) { 5541da177e4SLinus Torvalds int l = strlen(symname)+1; 5551da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 5561da177e4SLinus Torvalds if (!error) { 5571da177e4SLinus Torvalds d_instantiate(dentry, inode); 5581da177e4SLinus Torvalds dget(dentry); 5591da177e4SLinus Torvalds } else 5601da177e4SLinus Torvalds iput(inode); 5611da177e4SLinus Torvalds } 5621da177e4SLinus Torvalds dir->i_ctime = dir->i_mtime = CURRENT_TIME; 5631da177e4SLinus Torvalds 5641da177e4SLinus Torvalds return error; 5651da177e4SLinus Torvalds } 5661da177e4SLinus Torvalds 5671da177e4SLinus Torvalds /* 5686649a386SKen Chen * mark the head page dirty 5691da177e4SLinus Torvalds */ 5701da177e4SLinus Torvalds static int hugetlbfs_set_page_dirty(struct page *page) 5711da177e4SLinus Torvalds { 572d85f3385SChristoph Lameter struct page *head = compound_head(page); 5736649a386SKen Chen 5746649a386SKen Chen SetPageDirty(head); 5751da177e4SLinus Torvalds return 0; 5761da177e4SLinus Torvalds } 5771da177e4SLinus Torvalds 578290408d4SNaoya Horiguchi static int hugetlbfs_migrate_page(struct address_space *mapping, 579b969c4abSMel Gorman struct page *newpage, struct page *page, 580a6bc32b8SMel Gorman enum migrate_mode mode) 581290408d4SNaoya Horiguchi { 582290408d4SNaoya Horiguchi int rc; 583290408d4SNaoya Horiguchi 584290408d4SNaoya Horiguchi rc = migrate_huge_page_move_mapping(mapping, newpage, page); 58578bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 586290408d4SNaoya Horiguchi return rc; 587290408d4SNaoya Horiguchi migrate_page_copy(newpage, page); 588290408d4SNaoya Horiguchi 58978bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 590290408d4SNaoya Horiguchi } 591290408d4SNaoya Horiguchi 592726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 5931da177e4SLinus Torvalds { 594726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 595a5516438SAndi Kleen struct hstate *h = hstate_inode(dentry->d_inode); 5961da177e4SLinus Torvalds 5971da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 598a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 5991da177e4SLinus Torvalds if (sbinfo) { 6001da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 60174a8a65cSDavid Gibson /* If no limits set, just report 0 for max/free/used 60274a8a65cSDavid Gibson * blocks, like simple_statfs() */ 60390481622SDavid Gibson if (sbinfo->spool) { 60490481622SDavid Gibson long free_pages; 60590481622SDavid Gibson 60690481622SDavid Gibson spin_lock(&sbinfo->spool->lock); 60790481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 60890481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 60990481622SDavid Gibson - sbinfo->spool->used_hpages; 61090481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 61190481622SDavid Gibson spin_unlock(&sbinfo->spool->lock); 6121da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 6131da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 61474a8a65cSDavid Gibson } 6151da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 6161da177e4SLinus Torvalds } 6171da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 6181da177e4SLinus Torvalds return 0; 6191da177e4SLinus Torvalds } 6201da177e4SLinus Torvalds 6211da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 6221da177e4SLinus Torvalds { 6231da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 6241da177e4SLinus Torvalds 6251da177e4SLinus Torvalds if (sbi) { 6261da177e4SLinus Torvalds sb->s_fs_info = NULL; 62790481622SDavid Gibson 62890481622SDavid Gibson if (sbi->spool) 62990481622SDavid Gibson hugepage_put_subpool(sbi->spool); 63090481622SDavid Gibson 6311da177e4SLinus Torvalds kfree(sbi); 6321da177e4SLinus Torvalds } 6331da177e4SLinus Torvalds } 6341da177e4SLinus Torvalds 63596527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 63696527980SChristoph Hellwig { 63796527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 63896527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 63996527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 64096527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 64196527980SChristoph Hellwig return 0; 64296527980SChristoph Hellwig } 64396527980SChristoph Hellwig sbinfo->free_inodes--; 64496527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 64596527980SChristoph Hellwig } 64696527980SChristoph Hellwig 64796527980SChristoph Hellwig return 1; 64896527980SChristoph Hellwig } 64996527980SChristoph Hellwig 65096527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 65196527980SChristoph Hellwig { 65296527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 65396527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 65496527980SChristoph Hellwig sbinfo->free_inodes++; 65596527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 65696527980SChristoph Hellwig } 65796527980SChristoph Hellwig } 65896527980SChristoph Hellwig 65996527980SChristoph Hellwig 660e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 6611da177e4SLinus Torvalds 6621da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 6631da177e4SLinus Torvalds { 66496527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 6651da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 6661da177e4SLinus Torvalds 66796527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 6681da177e4SLinus Torvalds return NULL; 669e94b1766SChristoph Lameter p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); 67096527980SChristoph Hellwig if (unlikely(!p)) { 67196527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 67296527980SChristoph Hellwig return NULL; 6731da177e4SLinus Torvalds } 67496527980SChristoph Hellwig return &p->vfs_inode; 6751da177e4SLinus Torvalds } 6761da177e4SLinus Torvalds 677fa0d7e3dSNick Piggin static void hugetlbfs_i_callback(struct rcu_head *head) 678fa0d7e3dSNick Piggin { 679fa0d7e3dSNick Piggin struct inode *inode = container_of(head, struct inode, i_rcu); 680fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 681fa0d7e3dSNick Piggin } 682fa0d7e3dSNick Piggin 6831da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 6841da177e4SLinus Torvalds { 68596527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 6861da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 687fa0d7e3dSNick Piggin call_rcu(&inode->i_rcu, hugetlbfs_i_callback); 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds 690f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 691800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 692800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 6931da177e4SLinus Torvalds .set_page_dirty = hugetlbfs_set_page_dirty, 694290408d4SNaoya Horiguchi .migratepage = hugetlbfs_migrate_page, 6951da177e4SLinus Torvalds }; 6961da177e4SLinus Torvalds 69796527980SChristoph Hellwig 69851cc5068SAlexey Dobriyan static void init_once(void *foo) 69996527980SChristoph Hellwig { 70096527980SChristoph Hellwig struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; 70196527980SChristoph Hellwig 70296527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 70396527980SChristoph Hellwig } 70496527980SChristoph Hellwig 7054b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 706e63e1e5aSBadari Pulavarty .read = hugetlbfs_read, 7071da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 7081b061d92SChristoph Hellwig .fsync = noop_fsync, 7091da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 7106038f373SArnd Bergmann .llseek = default_llseek, 7111da177e4SLinus Torvalds }; 7121da177e4SLinus Torvalds 71392e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 7141da177e4SLinus Torvalds .create = hugetlbfs_create, 7151da177e4SLinus Torvalds .lookup = simple_lookup, 7161da177e4SLinus Torvalds .link = simple_link, 7171da177e4SLinus Torvalds .unlink = simple_unlink, 7181da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 7191da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 7201da177e4SLinus Torvalds .rmdir = simple_rmdir, 7211da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 7221da177e4SLinus Torvalds .rename = simple_rename, 7231da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 7241da177e4SLinus Torvalds }; 7251da177e4SLinus Torvalds 72692e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 7271da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 7281da177e4SLinus Torvalds }; 7291da177e4SLinus Torvalds 730ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 7311da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 7321da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 7332bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 7341da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 7351da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 73610f19a86SMiklos Szeredi .show_options = generic_show_options, 7371da177e4SLinus Torvalds }; 7381da177e4SLinus Torvalds 7391da177e4SLinus Torvalds static int 7401da177e4SLinus Torvalds hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) 7411da177e4SLinus Torvalds { 742e73a75faSRandy Dunlap char *p, *rest; 743e73a75faSRandy Dunlap substring_t args[MAX_OPT_ARGS]; 744e73a75faSRandy Dunlap int option; 745a137e1ccSAndi Kleen unsigned long long size = 0; 746a137e1ccSAndi Kleen enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE; 7471da177e4SLinus Torvalds 7481da177e4SLinus Torvalds if (!options) 7491da177e4SLinus Torvalds return 0; 7501da177e4SLinus Torvalds 751e73a75faSRandy Dunlap while ((p = strsep(&options, ",")) != NULL) { 752e73a75faSRandy Dunlap int token; 753b4c07bceSLee Schermerhorn if (!*p) 754b4c07bceSLee Schermerhorn continue; 7551da177e4SLinus Torvalds 756e73a75faSRandy Dunlap token = match_token(p, tokens, args); 757e73a75faSRandy Dunlap switch (token) { 758e73a75faSRandy Dunlap case Opt_uid: 759e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 760e73a75faSRandy Dunlap goto bad_val; 761a0eb3a05SEric W. Biederman pconfig->uid = make_kuid(current_user_ns(), option); 762a0eb3a05SEric W. Biederman if (!uid_valid(pconfig->uid)) 763a0eb3a05SEric W. Biederman goto bad_val; 764e73a75faSRandy Dunlap break; 765e73a75faSRandy Dunlap 766e73a75faSRandy Dunlap case Opt_gid: 767e73a75faSRandy Dunlap if (match_int(&args[0], &option)) 768e73a75faSRandy Dunlap goto bad_val; 769a0eb3a05SEric W. Biederman pconfig->gid = make_kgid(current_user_ns(), option); 770a0eb3a05SEric W. Biederman if (!gid_valid(pconfig->gid)) 771a0eb3a05SEric W. Biederman goto bad_val; 772e73a75faSRandy Dunlap break; 773e73a75faSRandy Dunlap 774e73a75faSRandy Dunlap case Opt_mode: 775e73a75faSRandy Dunlap if (match_octal(&args[0], &option)) 776e73a75faSRandy Dunlap goto bad_val; 77775897d60SKen Chen pconfig->mode = option & 01777U; 778e73a75faSRandy Dunlap break; 779e73a75faSRandy Dunlap 780e73a75faSRandy Dunlap case Opt_size: { 781e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 782e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 783e73a75faSRandy Dunlap goto bad_val; 784e73a75faSRandy Dunlap size = memparse(args[0].from, &rest); 785a137e1ccSAndi Kleen setsize = SIZE_STD; 786a137e1ccSAndi Kleen if (*rest == '%') 787a137e1ccSAndi Kleen setsize = SIZE_PERCENT; 788e73a75faSRandy Dunlap break; 789e73a75faSRandy Dunlap } 7901da177e4SLinus Torvalds 791e73a75faSRandy Dunlap case Opt_nr_inodes: 792e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 793e73a75faSRandy Dunlap if (!isdigit(*args[0].from)) 794e73a75faSRandy Dunlap goto bad_val; 795e73a75faSRandy Dunlap pconfig->nr_inodes = memparse(args[0].from, &rest); 796e73a75faSRandy Dunlap break; 797e73a75faSRandy Dunlap 798a137e1ccSAndi Kleen case Opt_pagesize: { 799a137e1ccSAndi Kleen unsigned long ps; 800a137e1ccSAndi Kleen ps = memparse(args[0].from, &rest); 801a137e1ccSAndi Kleen pconfig->hstate = size_to_hstate(ps); 802a137e1ccSAndi Kleen if (!pconfig->hstate) { 803a137e1ccSAndi Kleen printk(KERN_ERR 804a137e1ccSAndi Kleen "hugetlbfs: Unsupported page size %lu MB\n", 805a137e1ccSAndi Kleen ps >> 20); 806a137e1ccSAndi Kleen return -EINVAL; 807a137e1ccSAndi Kleen } 808a137e1ccSAndi Kleen break; 809a137e1ccSAndi Kleen } 810a137e1ccSAndi Kleen 811e73a75faSRandy Dunlap default: 812b4c07bceSLee Schermerhorn printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n", 813b4c07bceSLee Schermerhorn p); 814b4c07bceSLee Schermerhorn return -EINVAL; 815e73a75faSRandy Dunlap break; 816e73a75faSRandy Dunlap } 8171da177e4SLinus Torvalds } 818a137e1ccSAndi Kleen 819a137e1ccSAndi Kleen /* Do size after hstate is set up */ 820a137e1ccSAndi Kleen if (setsize > NO_SIZE) { 821a137e1ccSAndi Kleen struct hstate *h = pconfig->hstate; 822a137e1ccSAndi Kleen if (setsize == SIZE_PERCENT) { 823a137e1ccSAndi Kleen size <<= huge_page_shift(h); 824a137e1ccSAndi Kleen size *= h->max_huge_pages; 825a137e1ccSAndi Kleen do_div(size, 100); 826a137e1ccSAndi Kleen } 827a137e1ccSAndi Kleen pconfig->nr_blocks = (size >> huge_page_shift(h)); 828a137e1ccSAndi Kleen } 829a137e1ccSAndi Kleen 8301da177e4SLinus Torvalds return 0; 831e73a75faSRandy Dunlap 832e73a75faSRandy Dunlap bad_val: 833e73a75faSRandy Dunlap printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n", 834e73a75faSRandy Dunlap args[0].from, p); 835c12ddba0SAkinobu Mita return -EINVAL; 8361da177e4SLinus Torvalds } 8371da177e4SLinus Torvalds 8381da177e4SLinus Torvalds static int 8391da177e4SLinus Torvalds hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) 8401da177e4SLinus Torvalds { 8411da177e4SLinus Torvalds int ret; 8421da177e4SLinus Torvalds struct hugetlbfs_config config; 8431da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 8441da177e4SLinus Torvalds 84510f19a86SMiklos Szeredi save_mount_options(sb, data); 84610f19a86SMiklos Szeredi 8471da177e4SLinus Torvalds config.nr_blocks = -1; /* No limit on size by default */ 8481da177e4SLinus Torvalds config.nr_inodes = -1; /* No limit on number of inodes by default */ 84977c70de1SDavid Howells config.uid = current_fsuid(); 85077c70de1SDavid Howells config.gid = current_fsgid(); 8511da177e4SLinus Torvalds config.mode = 0755; 852a137e1ccSAndi Kleen config.hstate = &default_hstate; 8531da177e4SLinus Torvalds ret = hugetlbfs_parse_options(data, &config); 8541da177e4SLinus Torvalds if (ret) 8551da177e4SLinus Torvalds return ret; 8561da177e4SLinus Torvalds 8571da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 8581da177e4SLinus Torvalds if (!sbinfo) 8591da177e4SLinus Torvalds return -ENOMEM; 8601da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 861a137e1ccSAndi Kleen sbinfo->hstate = config.hstate; 8621da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 8631da177e4SLinus Torvalds sbinfo->max_inodes = config.nr_inodes; 8641da177e4SLinus Torvalds sbinfo->free_inodes = config.nr_inodes; 86590481622SDavid Gibson sbinfo->spool = NULL; 86690481622SDavid Gibson if (config.nr_blocks != -1) { 86790481622SDavid Gibson sbinfo->spool = hugepage_new_subpool(config.nr_blocks); 86890481622SDavid Gibson if (!sbinfo->spool) 86990481622SDavid Gibson goto out_free; 87090481622SDavid Gibson } 8711da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 872a137e1ccSAndi Kleen sb->s_blocksize = huge_page_size(config.hstate); 873a137e1ccSAndi Kleen sb->s_blocksize_bits = huge_page_shift(config.hstate); 8741da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 8751da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 8761da177e4SLinus Torvalds sb->s_time_gran = 1; 87748fde701SAl Viro sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); 87848fde701SAl Viro if (!sb->s_root) 8791da177e4SLinus Torvalds goto out_free; 8801da177e4SLinus Torvalds return 0; 8811da177e4SLinus Torvalds out_free: 88290481622SDavid Gibson if (sbinfo->spool) 88390481622SDavid Gibson kfree(sbinfo->spool); 8841da177e4SLinus Torvalds kfree(sbinfo); 8851da177e4SLinus Torvalds return -ENOMEM; 8861da177e4SLinus Torvalds } 8871da177e4SLinus Torvalds 8883c26ff6eSAl Viro static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, 8893c26ff6eSAl Viro int flags, const char *dev_name, void *data) 8901da177e4SLinus Torvalds { 8913c26ff6eSAl Viro return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); 8921da177e4SLinus Torvalds } 8931da177e4SLinus Torvalds 8941da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 8951da177e4SLinus Torvalds .name = "hugetlbfs", 8963c26ff6eSAl Viro .mount = hugetlbfs_mount, 8971da177e4SLinus Torvalds .kill_sb = kill_litter_super, 8981da177e4SLinus Torvalds }; 8991da177e4SLinus Torvalds 90042d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 9011da177e4SLinus Torvalds 902ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 9031da177e4SLinus Torvalds { 904a0eb3a05SEric W. Biederman kgid_t shm_group; 905a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 906a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 9071da177e4SLinus Torvalds } 9081da177e4SLinus Torvalds 90942d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 91042d7395fSAndi Kleen { 91142d7395fSAndi Kleen struct hstate *h; 91242d7395fSAndi Kleen 91342d7395fSAndi Kleen if (!page_size_log) 91442d7395fSAndi Kleen return default_hstate_idx; 91542d7395fSAndi Kleen h = size_to_hstate(1 << page_size_log); 91642d7395fSAndi Kleen if (!h) 91742d7395fSAndi Kleen return -1; 91842d7395fSAndi Kleen return h - hstates; 91942d7395fSAndi Kleen } 92042d7395fSAndi Kleen 92140716e29SSteven Truelove struct file *hugetlb_file_setup(const char *name, unsigned long addr, 92240716e29SSteven Truelove size_t size, vm_flags_t acctflag, 92342d7395fSAndi Kleen struct user_struct **user, 92442d7395fSAndi Kleen int creat_flags, int page_size_log) 9251da177e4SLinus Torvalds { 9261da177e4SLinus Torvalds int error = -ENOMEM; 9271da177e4SLinus Torvalds struct file *file; 9281da177e4SLinus Torvalds struct inode *inode; 9292c48b9c4SAl Viro struct path path; 9302c48b9c4SAl Viro struct dentry *root; 9311da177e4SLinus Torvalds struct qstr quick_string; 93240716e29SSteven Truelove struct hstate *hstate; 93340716e29SSteven Truelove unsigned long num_pages; 93442d7395fSAndi Kleen int hstate_idx; 93542d7395fSAndi Kleen 93642d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 93742d7395fSAndi Kleen if (hstate_idx < 0) 93842d7395fSAndi Kleen return ERR_PTR(-ENODEV); 9391da177e4SLinus Torvalds 940353d5c30SHugh Dickins *user = NULL; 94142d7395fSAndi Kleen if (!hugetlbfs_vfsmount[hstate_idx]) 9425bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 9435bc98594SAkinobu Mita 944ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 945353d5c30SHugh Dickins *user = current_user(); 946353d5c30SHugh Dickins if (user_shm_lock(size, *user)) { 94721a3c273SDavid Rientjes task_lock(current); 94821a3c273SDavid Rientjes printk_once(KERN_WARNING 94921a3c273SDavid Rientjes "%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", 95021a3c273SDavid Rientjes current->comm, current->pid); 95121a3c273SDavid Rientjes task_unlock(current); 952353d5c30SHugh Dickins } else { 953353d5c30SHugh Dickins *user = NULL; 9541da177e4SLinus Torvalds return ERR_PTR(-EPERM); 9552584e517SRavikiran G Thirumalai } 956353d5c30SHugh Dickins } 9571da177e4SLinus Torvalds 95842d7395fSAndi Kleen root = hugetlbfs_vfsmount[hstate_idx]->mnt_root; 9599d66586fSEric W. Biederman quick_string.name = name; 9601da177e4SLinus Torvalds quick_string.len = strlen(quick_string.name); 9611da177e4SLinus Torvalds quick_string.hash = 0; 9622c48b9c4SAl Viro path.dentry = d_alloc(root, &quick_string); 9632c48b9c4SAl Viro if (!path.dentry) 9641da177e4SLinus Torvalds goto out_shm_unlock; 9651da177e4SLinus Torvalds 96642d7395fSAndi Kleen path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); 9671da177e4SLinus Torvalds error = -ENOSPC; 9687d54fa64SAl Viro inode = hugetlbfs_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0); 9691da177e4SLinus Torvalds if (!inode) 970ce8d2cdfSDave Hansen goto out_dentry; 9711da177e4SLinus Torvalds 97240716e29SSteven Truelove hstate = hstate_inode(inode); 97340716e29SSteven Truelove size += addr & ~huge_page_mask(hstate); 97440716e29SSteven Truelove num_pages = ALIGN(size, huge_page_size(hstate)) >> 97540716e29SSteven Truelove huge_page_shift(hstate); 976b45b5bd6SDavid Gibson error = -ENOMEM; 97740716e29SSteven Truelove if (hugetlb_reserve_pages(inode, 0, num_pages, NULL, acctflag)) 978b45b5bd6SDavid Gibson goto out_inode; 979b45b5bd6SDavid Gibson 9802c48b9c4SAl Viro d_instantiate(path.dentry, inode); 9811da177e4SLinus Torvalds inode->i_size = size; 9826d6b77f1SMiklos Szeredi clear_nlink(inode); 983ce8d2cdfSDave Hansen 984ce8d2cdfSDave Hansen error = -ENFILE; 9852c48b9c4SAl Viro file = alloc_file(&path, FMODE_WRITE | FMODE_READ, 986ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 987ce8d2cdfSDave Hansen if (!file) 988b4d232e6SAl Viro goto out_dentry; /* inode is already attached */ 989ce8d2cdfSDave Hansen 9901da177e4SLinus Torvalds return file; 9911da177e4SLinus Torvalds 992b45b5bd6SDavid Gibson out_inode: 993b45b5bd6SDavid Gibson iput(inode); 9941da177e4SLinus Torvalds out_dentry: 9952c48b9c4SAl Viro path_put(&path); 9961da177e4SLinus Torvalds out_shm_unlock: 997353d5c30SHugh Dickins if (*user) { 998353d5c30SHugh Dickins user_shm_unlock(size, *user); 999353d5c30SHugh Dickins *user = NULL; 1000353d5c30SHugh Dickins } 10011da177e4SLinus Torvalds return ERR_PTR(error); 10021da177e4SLinus Torvalds } 10031da177e4SLinus Torvalds 10041da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 10051da177e4SLinus Torvalds { 100642d7395fSAndi Kleen struct hstate *h; 10071da177e4SLinus Torvalds int error; 100842d7395fSAndi Kleen int i; 10091da177e4SLinus Torvalds 1010e0bf68ddSPeter Zijlstra error = bdi_init(&hugetlbfs_backing_dev_info); 1011e0bf68ddSPeter Zijlstra if (error) 1012e0bf68ddSPeter Zijlstra return error; 1013e0bf68ddSPeter Zijlstra 1014d1d5e05fSHillf Danton error = -ENOMEM; 10151da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 10161da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 101720c2df83SPaul Mundt 0, 0, init_once); 10181da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 1019e0bf68ddSPeter Zijlstra goto out2; 10201da177e4SLinus Torvalds 10211da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 10221da177e4SLinus Torvalds if (error) 10231da177e4SLinus Torvalds goto out; 10241da177e4SLinus Torvalds 102542d7395fSAndi Kleen i = 0; 102642d7395fSAndi Kleen for_each_hstate(h) { 102742d7395fSAndi Kleen char buf[50]; 102842d7395fSAndi Kleen unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); 10291da177e4SLinus Torvalds 103042d7395fSAndi Kleen snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); 103142d7395fSAndi Kleen hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, 103242d7395fSAndi Kleen buf); 103342d7395fSAndi Kleen 103442d7395fSAndi Kleen if (IS_ERR(hugetlbfs_vfsmount[i])) { 103542d7395fSAndi Kleen pr_err("hugetlb: Cannot mount internal hugetlbfs for " 103642d7395fSAndi Kleen "page size %uK", ps_kb); 103742d7395fSAndi Kleen error = PTR_ERR(hugetlbfs_vfsmount[i]); 103842d7395fSAndi Kleen hugetlbfs_vfsmount[i] = NULL; 10391da177e4SLinus Torvalds } 104042d7395fSAndi Kleen i++; 104142d7395fSAndi Kleen } 104242d7395fSAndi Kleen /* Non default hstates are optional */ 104342d7395fSAndi Kleen if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) 104442d7395fSAndi Kleen return 0; 10451da177e4SLinus Torvalds 10461da177e4SLinus Torvalds out: 10471da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 1048e0bf68ddSPeter Zijlstra out2: 1049e0bf68ddSPeter Zijlstra bdi_destroy(&hugetlbfs_backing_dev_info); 10501da177e4SLinus Torvalds return error; 10511da177e4SLinus Torvalds } 10521da177e4SLinus Torvalds 10531da177e4SLinus Torvalds static void __exit exit_hugetlbfs_fs(void) 10541da177e4SLinus Torvalds { 105542d7395fSAndi Kleen struct hstate *h; 105642d7395fSAndi Kleen int i; 105742d7395fSAndi Kleen 105842d7395fSAndi Kleen 10598c0a8537SKirill A. Shutemov /* 10608c0a8537SKirill A. Shutemov * Make sure all delayed rcu free inodes are flushed before we 10618c0a8537SKirill A. Shutemov * destroy cache. 10628c0a8537SKirill A. Shutemov */ 10638c0a8537SKirill A. Shutemov rcu_barrier(); 10641da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 106542d7395fSAndi Kleen i = 0; 106642d7395fSAndi Kleen for_each_hstate(h) 106742d7395fSAndi Kleen kern_unmount(hugetlbfs_vfsmount[i++]); 10681da177e4SLinus Torvalds unregister_filesystem(&hugetlbfs_fs_type); 1069e0bf68ddSPeter Zijlstra bdi_destroy(&hugetlbfs_backing_dev_info); 10701da177e4SLinus Torvalds } 10711da177e4SLinus Torvalds 10721da177e4SLinus Torvalds module_init(init_hugetlbfs_fs) 10731da177e4SLinus Torvalds module_exit(exit_hugetlbfs_fs) 10741da177e4SLinus Torvalds 10751da177e4SLinus Torvalds MODULE_LICENSE("GPL"); 1076