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 inline void
 268 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
 269 {
 270         struct vm_area_struct *vma = NULL;
 271         struct prio_tree_iter iter;
 272
 273         while ((vma = vma_prio_tree_next(vma, root, &iter,
 274                                         h_pgoff, ULONG_MAX)) != NULL) {
 275                 unsigned long h_vm_pgoff;
 276                 unsigned long v_length;
 277                 unsigned long v_offset;
 278
 279                 h_vm_pgoff = vma->vm_pgoff << (HPAGE_SHIFT - PAGE_SHIFT);
 280                 v_length = vma->vm_end - vma->vm_start;
 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                         v_offset = 0;
 288
 289                 zap_hugepage_range(vma,
 290                                 vma->vm_start + v_offset,
 291                                 v_length - v_offset);
 292         }
 293 }
 294
 295 /*
 296  * Expanding truncates are not allowed.
 297  */
 298 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 299 {
 300         unsigned long pgoff;
 301         struct address_space *mapping = inode->i_mapping;
 302
 303         if (offset > inode->i_size)
 304                 return -EINVAL;
 305
 306         BUG_ON(offset & ~HPAGE_MASK);
 307         pgoff = offset >> HPAGE_SHIFT;
 308
 309         inode->i_size = offset;
 310         spin_lock(&mapping->i_mmap_lock);
 311         if (!prio_tree_empty(&mapping->i_mmap))
 312                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 313         spin_unlock(&mapping->i_mmap_lock);
 314         truncate_hugepages(mapping, offset);
 315         return 0;
 316 }
 317
 318 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 319 {
 320         struct inode *inode = dentry->d_inode;
 321         int error;
 322         unsigned int ia_valid = attr->ia_valid;
 323
 324         BUG_ON(!inode);
 325
 326         error = inode_change_ok(inode, attr);
 327         if (error)
 328                 goto out;
 329
 330         if (ia_valid & ATTR_SIZE) {
 331                 error = -EINVAL;
 332                 if (!(attr->ia_size & ~HPAGE_MASK))
 333                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 334                 if (error)
 335                         goto out;
 336                 attr->ia_valid &= ~ATTR_SIZE;
 337         }
 338         error = inode_setattr(inode, attr);
 339 out:
 340         return error;
 341 }
 342
 343 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 344                                         gid_t gid, int mode, dev_t dev)
 345 {
 346         struct inode *inode;
 347         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 348
 349         if (sbinfo->free_inodes >= 0) {
 350                 spin_lock(&sbinfo->stat_lock);
 351                 if (!sbinfo->free_inodes) {
 352                         spin_unlock(&sbinfo->stat_lock);
 353                         return NULL;
 354                 }
 355                 sbinfo->free_inodes--;
 356                 spin_unlock(&sbinfo->stat_lock);
 357         }
 358
 359         inode = new_inode(sb);
 360         if (inode) {
 361                 struct hugetlbfs_inode_info *info;
 362                 inode->i_mode = mode;
 363                 inode->i_uid = uid;
 364                 inode->i_gid = gid;
 365                 inode->i_blksize = HPAGE_SIZE;
 366                 inode->i_blocks = 0;
 367                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 368                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 369                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 370                 info = HUGETLBFS_I(inode);
 371                 mpol_shared_policy_init(&info->policy);
 372                 switch (mode & S_IFMT) {
 373                 default:
 374                         init_special_inode(inode, mode, dev);
 375                         break;
 376                 case S_IFREG:
 377                         inode->i_op = &hugetlbfs_inode_operations;
 378                         inode->i_fop = &hugetlbfs_file_operations;
 379                         break;
 380                 case S_IFDIR:
 381                         inode->i_op = &hugetlbfs_dir_inode_operations;
 382                         inode->i_fop = &simple_dir_operations;
 383
 384                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 385                         inode->i_nlink++;
 386                         break;
 387                 case S_IFLNK:
 388                         inode->i_op = &page_symlink_inode_operations;
 389                         break;
 390                 }
 391         }
 392         return inode;
 393 }
 394
 395 /*
 396  * File creation. Allocate an inode, and we're done..
 397  */
 398 static int hugetlbfs_mknod(struct inode *dir,
 399                         struct dentry *dentry, int mode, dev_t dev)
 400 {
 401         struct inode *inode;
 402         int error = -ENOSPC;
 403         gid_t gid;
 404
 405         if (dir->i_mode & S_ISGID) {
 406                 gid = dir->i_gid;
 407                 if (S_ISDIR(mode))
 408                         mode |= S_ISGID;
 409         } else {
 410                 gid = current->fsgid;
 411         }
 412         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 413         if (inode) {
 414                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 415                 d_instantiate(dentry, inode);
 416                 dget(dentry);   /* Extra count - pin the dentry in core */
 417                 error = 0;
 418         }
 419         return error;
 420 }
 421
 422 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 423 {
 424         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 425         if (!retval)
 426                 dir->i_nlink++;
 427         return retval;
 428 }
 429
 430 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 431 {
 432         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 433 }
 434
 435 static int hugetlbfs_symlink(struct inode *dir,
 436                         struct dentry *dentry, const char *symname)
 437 {
 438         struct inode *inode;
 439         int error = -ENOSPC;
 440         gid_t gid;
 441
 442         if (dir->i_mode & S_ISGID)
 443                 gid = dir->i_gid;
 444         else
 445                 gid = current->fsgid;
 446
 447         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 448                                         gid, S_IFLNK|S_IRWXUGO, 0);
 449         if (inode) {
 450                 int l = strlen(symname)+1;
 451                 error = page_symlink(inode, symname, l);
 452                 if (!error) {
 453                         d_instantiate(dentry, inode);
 454                         dget(dentry);
 455                 } else
 456                         iput(inode);
 457         }
 458         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 459
 460         return error;
 461 }
 462
 463 /*
 464  * For direct-IO reads into hugetlb pages
 465  */
 466 int hugetlbfs_set_page_dirty(struct page *page)
 467 {
 468         return 0;
 469 }
 470
 471 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
 472 {
 473         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 474
 475         buf->f_type = HUGETLBFS_MAGIC;
 476         buf->f_bsize = HPAGE_SIZE;
 477         if (sbinfo) {
 478                 spin_lock(&sbinfo->stat_lock);
 479                 buf->f_blocks = sbinfo->max_blocks;
 480                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 481                 buf->f_files = sbinfo->max_inodes;
 482                 buf->f_ffree = sbinfo->free_inodes;
 483                 spin_unlock(&sbinfo->stat_lock);
 484         }
 485         buf->f_namelen = NAME_MAX;
 486         return 0;
 487 }
 488
 489 static void hugetlbfs_put_super(struct super_block *sb)
 490 {
 491         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 492
 493         if (sbi) {
 494                 sb->s_fs_info = NULL;
 495                 kfree(sbi);
 496         }
 497 }
 498
 499 static kmem_cache_t *hugetlbfs_inode_cachep;
 500
 501 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 502 {
 503         struct hugetlbfs_inode_info *p;
 504
 505         p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
 506         if (!p)
 507                 return NULL;
 508         return &p->vfs_inode;
 509 }
 510
 511 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
 512 {
 513         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 514
 515         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 516             SLAB_CTOR_CONSTRUCTOR)
 517                 inode_init_once(&ei->vfs_inode);
 518 }
 519
 520 static void hugetlbfs_destroy_inode(struct inode *inode)
 521 {
 522         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 523         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 524 }
 525
 526 static struct address_space_operations hugetlbfs_aops = {
 527         .readpage       = hugetlbfs_readpage,
 528         .prepare_write  = hugetlbfs_prepare_write,
 529         .commit_write   = hugetlbfs_commit_write,
 530         .set_page_dirty = hugetlbfs_set_page_dirty,
 531 };
 532
 533 struct file_operations hugetlbfs_file_operations = {
 534         .mmap                   = hugetlbfs_file_mmap,
 535         .fsync                  = simple_sync_file,
 536         .get_unmapped_area      = hugetlb_get_unmapped_area,
 537 };
 538
 539 static struct inode_operations hugetlbfs_dir_inode_operations = {
 540         .create         = hugetlbfs_create,
 541         .lookup         = simple_lookup,
 542         .link           = simple_link,
 543         .unlink         = simple_unlink,
 544         .symlink        = hugetlbfs_symlink,
 545         .mkdir          = hugetlbfs_mkdir,
 546         .rmdir          = simple_rmdir,
 547         .mknod          = hugetlbfs_mknod,
 548         .rename         = simple_rename,
 549         .setattr        = hugetlbfs_setattr,
 550 };
 551
 552 static struct inode_operations hugetlbfs_inode_operations = {
 553         .setattr        = hugetlbfs_setattr,
 554 };
 555
 556 static struct super_operations hugetlbfs_ops = {
 557         .alloc_inode    = hugetlbfs_alloc_inode,
 558         .destroy_inode  = hugetlbfs_destroy_inode,
 559         .statfs         = hugetlbfs_statfs,
 560         .drop_inode     = hugetlbfs_drop_inode,
 561         .put_super      = hugetlbfs_put_super,
 562 };
 563
 564 static int
 565 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 566 {
 567         char *opt, *value, *rest;
 568
 569         if (!options)
 570                 return 0;
 571         while ((opt = strsep(&options, ",")) != NULL) {
 572                 if (!*opt)
 573                         continue;
 574
 575                 value = strchr(opt, '=');
 576                 if (!value || !*value)
 577                         return -EINVAL;
 578                 else
 579                         *value++ = '\0';
 580
 581                 if (!strcmp(opt, "uid"))
 582                         pconfig->uid = simple_strtoul(value, &value, 0);
 583                 else if (!strcmp(opt, "gid"))
 584                         pconfig->gid = simple_strtoul(value, &value, 0);
 585                 else if (!strcmp(opt, "mode"))
 586                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
 587                 else if (!strcmp(opt, "size")) {
 588                         unsigned long long size = memparse(value, &rest);
 589                         if (*rest == '%') {
 590                                 size <<= HPAGE_SHIFT;
 591                                 size *= max_huge_pages;
 592                                 do_div(size, 100);
 593                                 rest++;
 594                         }
 595                         size &= HPAGE_MASK;
 596                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 597                         value = rest;
 598                 } else if (!strcmp(opt,"nr_inodes")) {
 599                         pconfig->nr_inodes = memparse(value, &rest);
 600                         value = rest;
 601                 } else
 602                         return -EINVAL;
 603
 604                 if (*value)
 605                         return -EINVAL;
 606         }
 607         return 0;
 608 }
 609
 610 static int
 611 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 612 {
 613         struct inode * inode;
 614         struct dentry * root;
 615         int ret;
 616         struct hugetlbfs_config config;
 617         struct hugetlbfs_sb_info *sbinfo;
 618
 619         config.nr_blocks = -1; /* No limit on size by default */
 620         config.nr_inodes = -1; /* No limit on number of inodes by default */
 621         config.uid = current->fsuid;
 622         config.gid = current->fsgid;
 623         config.mode = 0755;
 624         ret = hugetlbfs_parse_options(data, &config);
 625
 626         if (ret)
 627                 return ret;
 628
 629         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 630         if (!sbinfo)
 631                 return -ENOMEM;
 632         sb->s_fs_info = sbinfo;
 633         spin_lock_init(&sbinfo->stat_lock);
 634         sbinfo->max_blocks = config.nr_blocks;
 635         sbinfo->free_blocks = config.nr_blocks;
 636         sbinfo->max_inodes = config.nr_inodes;
 637         sbinfo->free_inodes = config.nr_inodes;
 638         sb->s_blocksize = HPAGE_SIZE;
 639         sb->s_blocksize_bits = HPAGE_SHIFT;
 640         sb->s_magic = HUGETLBFS_MAGIC;
 641         sb->s_op = &hugetlbfs_ops;
 642         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 643                                         S_IFDIR | config.mode, 0);
 644         if (!inode)
 645                 goto out_free;
 646
 647         root = d_alloc_root(inode);
 648         if (!root) {
 649                 iput(inode);
 650                 goto out_free;
 651         }
 652         sb->s_root = root;
 653         return 0;
 654 out_free:
 655         kfree(sbinfo);
 656         return -ENOMEM;
 657 }
 658
 659 int hugetlb_get_quota(struct address_space *mapping)
 660 {
 661         int ret = 0;
 662         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 663
 664         if (sbinfo->free_blocks > -1) {
 665                 spin_lock(&sbinfo->stat_lock);
 666                 if (sbinfo->free_blocks > 0)
 667                         sbinfo->free_blocks--;
 668                 else
 669                         ret = -ENOMEM;
 670                 spin_unlock(&sbinfo->stat_lock);
 671         }
 672
 673         return ret;
 674 }
 675
 676 void hugetlb_put_quota(struct address_space *mapping)
 677 {
 678         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 679
 680         if (sbinfo->free_blocks > -1) {
 681                 spin_lock(&sbinfo->stat_lock);
 682                 sbinfo->free_blocks++;
 683                 spin_unlock(&sbinfo->stat_lock);
 684         }
 685 }
 686
 687 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
 688         int flags, const char *dev_name, void *data)
 689 {
 690         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 691 }
 692
 693 static struct file_system_type hugetlbfs_fs_type = {
 694         .name           = "hugetlbfs",
 695         .get_sb         = hugetlbfs_get_sb,
 696         .kill_sb        = kill_litter_super,
 697 };
 698
 699 static struct vfsmount *hugetlbfs_vfsmount;
 700
 701 /*
 702  * Return the next identifier for a shm file
 703  */
 704 static unsigned long hugetlbfs_counter(void)
 705 {
 706         static spinlock_t lock = SPIN_LOCK_UNLOCKED;
 707         static unsigned long counter;
 708         unsigned long ret;
 709
 710         spin_lock(&lock);
 711         ret = ++counter;
 712         spin_unlock(&lock);
 713         return ret;
 714 }
 715
 716 struct file *hugetlb_zero_setup(size_t size)
 717 {
 718         int error;
 719         struct file *file;
 720         struct inode *inode;
 721         struct dentry *dentry, *root;
 722         struct qstr quick_string;
 723         char buf[16];
 724
 725         if (!can_do_mlock())
 726                 return ERR_PTR(-EPERM);
 727
 728         if (!is_hugepage_mem_enough(size))
 729                 return ERR_PTR(-ENOMEM);
 730
 731         root = hugetlbfs_vfsmount->mnt_root;
 732         snprintf(buf, 16, "%lu", hugetlbfs_counter());
 733         quick_string.name = buf;
 734         quick_string.len = strlen(quick_string.name);
 735         quick_string.hash = 0;
 736         dentry = d_alloc(root, &quick_string);
 737         if (!dentry)
 738                 return ERR_PTR(-ENOMEM);
 739
 740         error = -ENFILE;
 741         file = get_empty_filp();
 742         if (!file)
 743                 goto out_dentry;
 744
 745         error = -ENOSPC;
 746         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
 747                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
 748         if (!inode)
 749                 goto out_file;
 750
 751         d_instantiate(dentry, inode);
 752         inode->i_size = size;
 753         inode->i_nlink = 0;
 754         file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
 755         file->f_dentry = dentry;
 756         file->f_mapping = inode->i_mapping;
 757         file->f_op = &hugetlbfs_file_operations;
 758         file->f_mode = FMODE_WRITE | FMODE_READ;
 759         return file;
 760
 761 out_file:
 762         put_filp(file);
 763 out_dentry:
 764         dput(dentry);
 765         return ERR_PTR(error);
 766 }
 767
 768 static int __init init_hugetlbfs_fs(void)
 769 {
 770         int error;
 771         struct vfsmount *vfsmount;
 772
 773         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
 774                                         sizeof(struct hugetlbfs_inode_info),
 775                                         0, SLAB_RECLAIM_ACCOUNT,
 776                                         init_once, NULL);
 777         if (hugetlbfs_inode_cachep == NULL)
 778                 return -ENOMEM;
 779
 780         error = register_filesystem(&hugetlbfs_fs_type);
 781         if (error)
 782                 goto out;
 783
 784         vfsmount = kern_mount(&hugetlbfs_fs_type);
 785
 786         if (!IS_ERR(vfsmount)) {
 787                 hugetlbfs_vfsmount = vfsmount;
 788                 return 0;
 789         }
 790
 791         error = PTR_ERR(vfsmount);
 792
 793  out:
 794         if (error)
 795                 kmem_cache_destroy(hugetlbfs_inode_cachep);
 796         return error;
 797 }
 798
 799 static void __exit exit_hugetlbfs_fs(void)
 800 {
 801         kmem_cache_destroy(hugetlbfs_inode_cachep);
 802         unregister_filesystem(&hugetlbfs_fs_type);
 803 }
 804
 805 module_init(init_hugetlbfs_fs)
 806 module_exit(exit_hugetlbfs_fs)
 807
 808 MODULE_LICENSE("GPL");