fs/hugetlbfs/inode.c

   1 /*
   2  * hugetlbpage-backed filesystem.  Based on ramfs.
   3  *
   4  * William Irwin, 2002
   5  *
   6  * Copyright (C) 2002 Linus Torvalds.
   7  */
   8
   9 #include <linux/module.h>
  10 #include <linux/thread_info.h>
  11 #include <asm/current.h>
  12 #include <linux/sched.h>                /* remove ASAP */
  13 #include <linux/fs.h>
  14 #include <linux/mount.h>
  15 #include <linux/file.h>
  16 #include <linux/writeback.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/highmem.h>
  19 #include <linux/init.h>
  20 #include <linux/string.h>
  21 #include <linux/backing-dev.h>
  22 #include <linux/hugetlb.h>
  23 #include <linux/pagevec.h>
  24 #include <linux/quotaops.h>
  25 #include <linux/slab.h>
  26 #include <linux/dnotify.h>
  27 #include <linux/statfs.h>
  28 #include <linux/security.h>
  29
  30 #include <asm/uaccess.h>
  31
  32 /* some random number */
  33 #define HUGETLBFS_MAGIC 0x958458f6
  34
  35 static struct super_operations hugetlbfs_ops;
  36 static struct address_space_operations hugetlbfs_aops;
  37 struct file_operations hugetlbfs_file_operations;
  38 static struct inode_operations hugetlbfs_dir_inode_operations;
  39 static struct inode_operations hugetlbfs_inode_operations;
  40
  41 static struct backing_dev_info hugetlbfs_backing_dev_info = {
  42         .ra_pages       = 0,    /* No readahead */
  43         .memory_backed  = 1,    /* Does not contribute to dirty memory */
  44 };
  45
  46 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  47 {
  48         struct inode *inode = file->f_dentry->d_inode;
  49         struct address_space *mapping = inode->i_mapping;
  50         loff_t len, vma_len;
  51         int ret;
  52
  53         if (vma->vm_start & ~HPAGE_MASK)
  54                 return -EINVAL;
  55
  56         if (vma->vm_end & ~HPAGE_MASK)
  57                 return -EINVAL;
  58
  59         if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
  60                 return -EINVAL;
  61
  62         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
  63
  64         down(&inode->i_sem);
  65         file_accessed(file);
  66         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
  67         vma->vm_ops = &hugetlb_vm_ops;
  68         ret = hugetlb_prefault(mapping, vma);
  69         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
  70         if (ret == 0 && inode->i_size < len)
  71                 inode->i_size = len;
  72         up(&inode->i_sem);
  73
  74         return ret;
  75 }
  76
  77 /*
  78  * Called under down_write(mmap_sem), page_table_lock is not held
  79  */
  80
  81 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
  82 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
  83                 unsigned long len, unsigned long pgoff, unsigned long flags);
  84 #else
  85 static unsigned long
  86 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
  87                 unsigned long len, unsigned long pgoff, unsigned long flags)
  88 {
  89         struct mm_struct *mm = current->mm;
  90         struct vm_area_struct *vma;
  91
  92         if (len & ~HPAGE_MASK)
  93                 return -EINVAL;
  94         if (len > TASK_SIZE)
  95                 return -ENOMEM;
  96
  97         if (addr) {
  98                 addr = ALIGN(addr, HPAGE_SIZE);
  99                 vma = find_vma(mm, addr);
 100                 if (TASK_SIZE - len >= addr &&
 101                     (!vma || addr + len <= vma->vm_start))
 102                         return addr;
 103         }
 104
 105         addr = ALIGN(mm->free_area_cache, HPAGE_SIZE);
 106
 107         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 108                 /* At this point:  (!vma || addr < vma->vm_end). */
 109                 if (TASK_SIZE - len < addr)
 110                         return -ENOMEM;
 111                 if (!vma || addr + len <= vma->vm_start)
 112                         return addr;
 113                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
 114         }
 115 }
 116 #endif
 117
 118 /*
 119  * Read a page. Again trivial. If it didn't already exist
 120  * in the page cache, it is zero-filled.
 121  */
 122 static int hugetlbfs_readpage(struct file *file, struct page * page)
 123 {
 124         unlock_page(page);
 125         return -EINVAL;
 126 }
 127
 128 static int hugetlbfs_prepare_write(struct file *file,
 129                         struct page *page, unsigned offset, unsigned to)
 130 {
 131         return -EINVAL;
 132 }
 133
 134 static int hugetlbfs_commit_write(struct file *file,
 135                         struct page *page, unsigned offset, unsigned to)
 136 {
 137         return -EINVAL;
 138 }
 139
 140 void huge_pagevec_release(struct pagevec *pvec)
 141 {
 142         int i;
 143
 144         for (i = 0; i < pagevec_count(pvec); ++i)
 145                 put_page(pvec->pages[i]);
 146
 147         pagevec_reinit(pvec);
 148 }
 149
 150 void truncate_huge_page(struct page *page)
 151 {
 152         clear_page_dirty(page);
 153         ClearPageUptodate(page);
 154         remove_from_page_cache(page);
 155         put_page(page);
 156 }
 157
 158 void truncate_hugepages(struct address_space *mapping, loff_t lstart)
 159 {
 160         const pgoff_t start = lstart >> HPAGE_SHIFT;
 161         struct pagevec pvec;
 162         pgoff_t next;
 163         int i;
 164
 165         pagevec_init(&pvec, 0);
 166         next = start;
 167         while (1) {
 168                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 169                         if (next == start)
 170                                 break;
 171                         next = start;
 172                         continue;
 173                 }
 174
 175                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 176                         struct page *page = pvec.pages[i];
 177
 178                         lock_page(page);
 179                         if (page->index > next)
 180                                 next = page->index;
 181                         ++next;
 182                         truncate_huge_page(page);
 183                         unlock_page(page);
 184                         hugetlb_put_quota(mapping);
 185                 }
 186                 huge_pagevec_release(&pvec);
 187         }
 188         BUG_ON(!lstart && mapping->nrpages);
 189 }
 190
 191 static void hugetlbfs_delete_inode(struct inode *inode)
 192 {
 193         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(inode->i_sb);
 194
 195         hlist_del_init(&inode->i_hash);
 196         list_del_init(&inode->i_list);
 197         inode->i_state |= I_FREEING;
 198         inodes_stat.nr_inodes--;
 199         spin_unlock(&inode_lock);
 200
 201         if (inode->i_data.nrpages)
 202                 truncate_hugepages(&inode->i_data, 0);
 203
 204         security_inode_delete(inode);
 205
 206         if (sbinfo->free_inodes >= 0) {
 207                 spin_lock(&sbinfo->stat_lock);
 208                 sbinfo->free_inodes++;
 209                 spin_unlock(&sbinfo->stat_lock);
 210         }
 211
 212         clear_inode(inode);
 213         destroy_inode(inode);
 214 }
 215
 216 static void hugetlbfs_forget_inode(struct inode *inode)
 217 {
 218         struct super_block *super_block = inode->i_sb;
 219         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(super_block);
 220
 221         if (hlist_unhashed(&inode->i_hash))
 222                 goto out_truncate;
 223
 224         if (!(inode->i_state & (I_DIRTY|I_LOCK))) {
 225                 list_del(&inode->i_list);
 226                 list_add(&inode->i_list, &inode_unused);
 227         }
 228         inodes_stat.nr_unused++;
 229         if (!super_block || (super_block->s_flags & MS_ACTIVE)) {
 230                 spin_unlock(&inode_lock);
 231                 return;
 232         }
 233
 234         /* write_inode_now() ? */
 235         inodes_stat.nr_unused--;
 236         hlist_del_init(&inode->i_hash);
 237 out_truncate:
 238         list_del_init(&inode->i_list);
 239         inode->i_state |= I_FREEING;
 240         inodes_stat.nr_inodes--;
 241         spin_unlock(&inode_lock);
 242         if (inode->i_data.nrpages)
 243                 truncate_hugepages(&inode->i_data, 0);
 244
 245         if (sbinfo->free_inodes >= 0) {
 246                 spin_lock(&sbinfo->stat_lock);
 247                 sbinfo->free_inodes++;
 248                 spin_unlock(&sbinfo->stat_lock);
 249         }
 250
 251         clear_inode(inode);
 252         destroy_inode(inode);
 253 }
 254
 255 static void hugetlbfs_drop_inode(struct inode *inode)
 256 {
 257         if (!inode->i_nlink)
 258                 hugetlbfs_delete_inode(inode);
 259         else
 260                 hugetlbfs_forget_inode(inode);
 261 }
 262
 263 /*
 264  * h_pgoff is in HPAGE_SIZE units.
 265  * vma->vm_pgoff is in PAGE_SIZE units.
 266  */
 267 static void
 268 hugetlb_vmtruncate_list(struct list_head *list, unsigned long h_pgoff)
 269 {
 270         struct vm_area_struct *vma;
 271
 272         list_for_each_entry(vma, list, shared) {
 273                 unsigned long h_vm_pgoff;
 274                 unsigned long v_length;
 275                 unsigned long h_length;
 276                 unsigned long v_offset;
 277
 278                 h_vm_pgoff = vma->vm_pgoff << (HPAGE_SHIFT - PAGE_SHIFT);
 279                 v_length = vma->vm_end - vma->vm_start;
 280                 h_length = v_length >> HPAGE_SHIFT;
 281                 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
 282
 283                 /*
 284                  * Is this VMA fully outside the truncation point?
 285                  */
 286                 if (h_vm_pgoff >= h_pgoff) {
 287                         zap_hugepage_range(vma, vma->vm_start, v_length);
 288                         continue;
 289                 }
 290
 291                 /*
 292                  * Is this VMA fully inside the truncaton point?
 293                  */
 294                 if (h_vm_pgoff + (v_length >> HPAGE_SHIFT) <= h_pgoff)
 295                         continue;
 296
 297                 /*
 298                  * The VMA straddles the truncation point.  v_offset is the
 299                  * offset (in bytes) into the VMA where the point lies.
 300                  */
 301                 zap_hugepage_range(vma,
 302                                 vma->vm_start + v_offset,
 303                                 v_length - v_offset);
 304         }
 305 }
 306
 307 /*
 308  * Expanding truncates are not allowed.
 309  */
 310 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 311 {
 312         unsigned long pgoff;
 313         struct address_space *mapping = inode->i_mapping;
 314
 315         if (offset > inode->i_size)
 316                 return -EINVAL;
 317
 318         BUG_ON(offset & ~HPAGE_MASK);
 319         pgoff = offset >> HPAGE_SHIFT;
 320
 321         inode->i_size = offset;
 322         down(&mapping->i_shared_sem);
 323         if (!list_empty(&mapping->i_mmap))
 324                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 325         if (!list_empty(&mapping->i_mmap_shared))
 326                 hugetlb_vmtruncate_list(&mapping->i_mmap_shared, pgoff);
 327         up(&mapping->i_shared_sem);
 328         truncate_hugepages(mapping, offset);
 329         return 0;
 330 }
 331
 332 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 333 {
 334         struct inode *inode = dentry->d_inode;
 335         int error;
 336         unsigned int ia_valid = attr->ia_valid;
 337
 338         BUG_ON(!inode);
 339
 340         error = inode_change_ok(inode, attr);
 341         if (error)
 342                 goto out;
 343
 344         error = security_inode_setattr(dentry, attr);
 345         if (error)
 346                 goto out;
 347         if (ia_valid & ATTR_SIZE) {
 348                 error = -EINVAL;
 349                 if (!(attr->ia_size & ~HPAGE_MASK))
 350                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 351                 if (error)
 352                         goto out;
 353                 attr->ia_valid &= ~ATTR_SIZE;
 354         }
 355         error = inode_setattr(inode, attr);
 356 out:
 357         return error;
 358 }
 359
 360 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 361                                         gid_t gid, int mode, dev_t dev)
 362 {
 363         struct inode *inode;
 364         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 365
 366         if (sbinfo->free_inodes >= 0) {
 367                 spin_lock(&sbinfo->stat_lock);
 368                 if (!sbinfo->free_inodes) {
 369                         spin_unlock(&sbinfo->stat_lock);
 370                         return NULL;
 371                 }
 372                 sbinfo->free_inodes--;
 373                 spin_unlock(&sbinfo->stat_lock);
 374         }
 375
 376         inode = new_inode(sb);
 377         if (inode) {
 378                 inode->i_mode = mode;
 379                 inode->i_uid = uid;
 380                 inode->i_gid = gid;
 381                 inode->i_blksize = HPAGE_SIZE;
 382                 inode->i_blocks = 0;
 383                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 384                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 385                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 386                 switch (mode & S_IFMT) {
 387                 default:
 388                         init_special_inode(inode, mode, dev);
 389                         break;
 390                 case S_IFREG:
 391                         inode->i_op = &hugetlbfs_inode_operations;
 392                         inode->i_fop = &hugetlbfs_file_operations;
 393                         break;
 394                 case S_IFDIR:
 395                         inode->i_op = &hugetlbfs_dir_inode_operations;
 396                         inode->i_fop = &simple_dir_operations;
 397
 398                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 399                         inode->i_nlink++;
 400                         break;
 401                 case S_IFLNK:
 402                         inode->i_op = &page_symlink_inode_operations;
 403                         break;
 404                 }
 405         }
 406         return inode;
 407 }
 408
 409 /*
 410  * File creation. Allocate an inode, and we're done..
 411  */
 412 static int hugetlbfs_mknod(struct inode *dir,
 413                         struct dentry *dentry, int mode, dev_t dev)
 414 {
 415         struct inode *inode;
 416         int error = -ENOSPC;
 417         gid_t gid;
 418
 419         if (dir->i_mode & S_ISGID) {
 420                 gid = dir->i_gid;
 421                 if (S_ISDIR(mode))
 422                         mode |= S_ISGID;
 423         } else {
 424                 gid = current->fsgid;
 425         }
 426         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 427         if (inode) {
 428                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 429                 d_instantiate(dentry, inode);
 430                 dget(dentry);   /* Extra count - pin the dentry in core */
 431                 error = 0;
 432         }
 433         return error;
 434 }
 435
 436 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 437 {
 438         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 439         if (!retval)
 440                 dir->i_nlink++;
 441         return retval;
 442 }
 443
 444 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 445 {
 446         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 447 }
 448
 449 static int hugetlbfs_symlink(struct inode *dir,
 450                         struct dentry *dentry, const char *symname)
 451 {
 452         struct inode *inode;
 453         int error = -ENOSPC;
 454         gid_t gid;
 455
 456         if (dir->i_mode & S_ISGID)
 457                 gid = dir->i_gid;
 458         else
 459                 gid = current->fsgid;
 460
 461         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 462                                         gid, S_IFLNK|S_IRWXUGO, 0);
 463         if (inode) {
 464                 int l = strlen(symname)+1;
 465                 error = page_symlink(inode, symname, l);
 466                 if (!error) {
 467                         d_instantiate(dentry, inode);
 468                         dget(dentry);
 469                 } else
 470                         iput(inode);
 471         }
 472         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 473
 474         return error;
 475 }
 476
 477 /*
 478  * For direct-IO reads into hugetlb pages
 479  */
 480 int hugetlbfs_set_page_dirty(struct page *page)
 481 {
 482         return 0;
 483 }
 484
 485 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
 486 {
 487         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 488
 489         buf->f_type = HUGETLBFS_MAGIC;
 490         buf->f_bsize = HPAGE_SIZE;
 491         if (sbinfo) {
 492                 spin_lock(&sbinfo->stat_lock);
 493                 buf->f_blocks = sbinfo->max_blocks;
 494                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 495                 buf->f_files = sbinfo->max_inodes;
 496                 buf->f_ffree = sbinfo->free_inodes;
 497                 spin_unlock(&sbinfo->stat_lock);
 498         }
 499         buf->f_namelen = NAME_MAX;
 500         return 0;
 501 }
 502
 503 static void hugetlbfs_put_super(struct super_block *sb)
 504 {
 505         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 506
 507         if (sbi) {
 508                 sb->s_fs_info = NULL;
 509                 kfree(sbi);
 510         }
 511 }
 512
 513 static struct address_space_operations hugetlbfs_aops = {
 514         .readpage       = hugetlbfs_readpage,
 515         .prepare_write  = hugetlbfs_prepare_write,
 516         .commit_write   = hugetlbfs_commit_write,
 517         .set_page_dirty = hugetlbfs_set_page_dirty,
 518 };
 519
 520 struct file_operations hugetlbfs_file_operations = {
 521         .mmap                   = hugetlbfs_file_mmap,
 522         .fsync                  = simple_sync_file,
 523         .get_unmapped_area      = hugetlb_get_unmapped_area,
 524 };
 525
 526 static struct inode_operations hugetlbfs_dir_inode_operations = {
 527         .create         = hugetlbfs_create,
 528         .lookup         = simple_lookup,
 529         .link           = simple_link,
 530         .unlink         = simple_unlink,
 531         .symlink        = hugetlbfs_symlink,
 532         .mkdir          = hugetlbfs_mkdir,
 533         .rmdir          = simple_rmdir,
 534         .mknod          = hugetlbfs_mknod,
 535         .rename         = simple_rename,
 536         .setattr        = hugetlbfs_setattr,
 537 };
 538
 539 static struct inode_operations hugetlbfs_inode_operations = {
 540         .setattr        = hugetlbfs_setattr,
 541 };
 542
 543 static struct super_operations hugetlbfs_ops = {
 544         .statfs         = hugetlbfs_statfs,
 545         .drop_inode     = hugetlbfs_drop_inode,
 546         .put_super      = hugetlbfs_put_super,
 547 };
 548
 549 static int
 550 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 551 {
 552         char *opt, *value, *rest;
 553
 554         if (!options)
 555                 return 0;
 556         while ((opt = strsep(&options, ",")) != NULL) {
 557                 if (!*opt)
 558                         continue;
 559
 560                 value = strchr(opt, '=');
 561                 if (!value || !*value)
 562                         return -EINVAL;
 563                 else
 564                         *value++ = '\0';
 565
 566                 if (!strcmp(opt, "uid"))
 567                         pconfig->uid = simple_strtoul(value, &value, 0);
 568                 else if (!strcmp(opt, "gid"))
 569                         pconfig->gid = simple_strtoul(value, &value, 0);
 570                 else if (!strcmp(opt, "mode"))
 571                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
 572                 else if (!strcmp(opt, "size")) {
 573                         unsigned long long size = memparse(value, &rest);
 574                         if (*rest == '%') {
 575                                 size <<= HPAGE_SHIFT;
 576                                 size *= max_huge_pages;
 577                                 do_div(size, 100);
 578                                 rest++;
 579                         }
 580                         size &= HPAGE_MASK;
 581                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 582                         value = rest;
 583                 } else if (!strcmp(opt,"nr_inodes")) {
 584                         pconfig->nr_inodes = memparse(value, &rest);
 585                         value = rest;
 586                 } else
 587                         return -EINVAL;
 588
 589                 if (*value)
 590                         return -EINVAL;
 591         }
 592         return 0;
 593 }
 594
 595 static int
 596 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 597 {
 598         struct inode * inode;
 599         struct dentry * root;
 600         int ret;
 601         struct hugetlbfs_config config;
 602         struct hugetlbfs_sb_info *sbinfo;
 603
 604         config.nr_blocks = -1; /* No limit on size by default */
 605         config.nr_inodes = -1; /* No limit on number of inodes by default */
 606         config.uid = current->fsuid;
 607         config.gid = current->fsgid;
 608         config.mode = 0755;
 609         ret = hugetlbfs_parse_options(data, &config);
 610
 611         if (ret)
 612                 return ret;
 613
 614         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 615         if (!sbinfo)
 616                 return -ENOMEM;
 617         sb->s_fs_info = sbinfo;
 618         spin_lock_init(&sbinfo->stat_lock);
 619         sbinfo->max_blocks = config.nr_blocks;
 620         sbinfo->free_blocks = config.nr_blocks;
 621         sbinfo->max_inodes = config.nr_inodes;
 622         sbinfo->free_inodes = config.nr_inodes;
 623         sb->s_blocksize = HPAGE_SIZE;
 624         sb->s_blocksize_bits = HPAGE_SHIFT;
 625         sb->s_magic = HUGETLBFS_MAGIC;
 626         sb->s_op = &hugetlbfs_ops;
 627         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 628                                         S_IFDIR | config.mode, 0);
 629         if (!inode)
 630                 goto out_free;
 631
 632         root = d_alloc_root(inode);
 633         if (!root) {
 634                 iput(inode);
 635                 goto out_free;
 636         }
 637         sb->s_root = root;
 638         return 0;
 639 out_free:
 640         kfree(sbinfo);
 641         return -ENOMEM;
 642 }
 643
 644 int hugetlb_get_quota(struct address_space *mapping)
 645 {
 646         int ret = 0;
 647         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 648
 649         if (sbinfo->free_blocks > -1) {
 650                 spin_lock(&sbinfo->stat_lock);
 651                 if (sbinfo->free_blocks > 0)
 652                         sbinfo->free_blocks--;
 653                 else
 654                         ret = -ENOMEM;
 655                 spin_unlock(&sbinfo->stat_lock);
 656         }
 657
 658         return ret;
 659 }
 660
 661 void hugetlb_put_quota(struct address_space *mapping)
 662 {
 663         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 664
 665         if (sbinfo->free_blocks > -1) {
 666                 spin_lock(&sbinfo->stat_lock);
 667                 sbinfo->free_blocks++;
 668                 spin_unlock(&sbinfo->stat_lock);
 669         }
 670 }
 671
 672 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
 673         int flags, const char *dev_name, void *data)
 674 {
 675         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 676 }
 677
 678 static struct file_system_type hugetlbfs_fs_type = {
 679         .name           = "hugetlbfs",
 680         .get_sb         = hugetlbfs_get_sb,
 681         .kill_sb        = kill_litter_super,
 682 };
 683
 684 static struct vfsmount *hugetlbfs_vfsmount;
 685
 686 /*
 687  * Return the next identifier for a shm file
 688  */
 689 static unsigned long hugetlbfs_counter(void)
 690 {
 691         static spinlock_t lock = SPIN_LOCK_UNLOCKED;
 692         static unsigned long counter;
 693         unsigned long ret;
 694
 695         spin_lock(&lock);
 696         ret = ++counter;
 697         spin_unlock(&lock);
 698         return ret;
 699 }
 700
 701 struct file *hugetlb_zero_setup(size_t size)
 702 {
 703         int error;
 704         struct file *file;
 705         struct inode *inode;
 706         struct dentry *dentry, *root;
 707         struct qstr quick_string;
 708         char buf[16];
 709
 710         if (!capable(CAP_IPC_LOCK))
 711                 return ERR_PTR(-EPERM);
 712
 713         if (!is_hugepage_mem_enough(size))
 714                 return ERR_PTR(-ENOMEM);
 715
 716         root = hugetlbfs_vfsmount->mnt_root;
 717         snprintf(buf, 16, "%lu", hugetlbfs_counter());
 718         quick_string.name = buf;
 719         quick_string.len = strlen(quick_string.name);
 720         quick_string.hash = 0;
 721         dentry = d_alloc(root, &quick_string);
 722         if (!dentry)
 723                 return ERR_PTR(-ENOMEM);
 724
 725         error = -ENFILE;
 726         file = get_empty_filp();
 727         if (!file)
 728                 goto out_dentry;
 729
 730         error = -ENOSPC;
 731         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
 732                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
 733         if (!inode)
 734                 goto out_file;
 735
 736         d_instantiate(dentry, inode);
 737         inode->i_size = size;
 738         inode->i_nlink = 0;
 739         file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
 740         file->f_dentry = dentry;
 741         file->f_mapping = inode->i_mapping;
 742         file->f_op = &hugetlbfs_file_operations;
 743         file->f_mode = FMODE_WRITE | FMODE_READ;
 744         return file;
 745
 746 out_file:
 747         put_filp(file);
 748 out_dentry:
 749         dput(dentry);
 750         return ERR_PTR(error);
 751 }
 752
 753 static int __init init_hugetlbfs_fs(void)
 754 {
 755         int error;
 756         struct vfsmount *vfsmount;
 757
 758         error = register_filesystem(&hugetlbfs_fs_type);
 759         if (error)
 760                 return error;
 761
 762         vfsmount = kern_mount(&hugetlbfs_fs_type);
 763
 764         if (!IS_ERR(vfsmount)) {
 765                 hugetlbfs_vfsmount = vfsmount;
 766                 return 0;
 767         }
 768
 769         error = PTR_ERR(vfsmount);
 770         return error;
 771 }
 772
 773 static void __exit exit_hugetlbfs_fs(void)
 774 {
 775         unregister_filesystem(&hugetlbfs_fs_type);
 776 }
 777
 778 module_init(init_hugetlbfs_fs)
 779 module_exit(exit_hugetlbfs_fs)
 780
 781 MODULE_LICENSE("GPL");