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 int sysctl_hugetlb_shm_group;
  47
  48 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  49 {
  50         struct inode *inode = file->f_dentry->d_inode;
  51         struct address_space *mapping = inode->i_mapping;
  52         loff_t len, vma_len;
  53         int ret;
  54
  55         if (vma->vm_start & ~HPAGE_MASK)
  56                 return -EINVAL;
  57
  58         if (vma->vm_end & ~HPAGE_MASK)
  59                 return -EINVAL;
  60
  61         if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
  62                 return -EINVAL;
  63
  64         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
  65
  66         down(&inode->i_sem);
  67         file_accessed(file);
  68         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
  69         vma->vm_ops = &hugetlb_vm_ops;
  70         ret = hugetlb_prefault(mapping, vma);
  71         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
  72         if (ret == 0 && inode->i_size < len)
  73                 inode->i_size = len;
  74         up(&inode->i_sem);
  75
  76         return ret;
  77 }
  78
  79 /*
  80  * Called under down_write(mmap_sem), page_table_lock is not held
  81  */
  82
  83 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
  84 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
  85                 unsigned long len, unsigned long pgoff, unsigned long flags);
  86 #else
  87 static unsigned long
  88 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
  89                 unsigned long len, unsigned long pgoff, unsigned long flags)
  90 {
  91         struct mm_struct *mm = current->mm;
  92         struct vm_area_struct *vma;
  93
  94         if (len & ~HPAGE_MASK)
  95                 return -EINVAL;
  96         if (len > TASK_SIZE)
  97                 return -ENOMEM;
  98
  99         if (addr) {
 100                 addr = ALIGN(addr, HPAGE_SIZE);
 101                 vma = find_vma(mm, addr);
 102                 if (TASK_SIZE - len >= addr &&
 103                     (!vma || addr + len <= vma->vm_start))
 104                         return addr;
 105         }
 106
 107         addr = ALIGN(mm->free_area_cache, HPAGE_SIZE);
 108
 109         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 110                 /* At this point:  (!vma || addr < vma->vm_end). */
 111                 if (TASK_SIZE - len < addr)
 112                         return -ENOMEM;
 113                 if (!vma || addr + len <= vma->vm_start)
 114                         return addr;
 115                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
 116         }
 117 }
 118 #endif
 119
 120 /*
 121  * Read a page. Again trivial. If it didn't already exist
 122  * in the page cache, it is zero-filled.
 123  */
 124 static int hugetlbfs_readpage(struct file *file, struct page * page)
 125 {
 126         unlock_page(page);
 127         return -EINVAL;
 128 }
 129
 130 static int hugetlbfs_prepare_write(struct file *file,
 131                         struct page *page, unsigned offset, unsigned to)
 132 {
 133         return -EINVAL;
 134 }
 135
 136 static int hugetlbfs_commit_write(struct file *file,
 137                         struct page *page, unsigned offset, unsigned to)
 138 {
 139         return -EINVAL;
 140 }
 141
 142 void huge_pagevec_release(struct pagevec *pvec)
 143 {
 144         int i;
 145
 146         for (i = 0; i < pagevec_count(pvec); ++i)
 147                 put_page(pvec->pages[i]);
 148
 149         pagevec_reinit(pvec);
 150 }
 151
 152 void truncate_huge_page(struct page *page)
 153 {
 154         clear_page_dirty(page);
 155         ClearPageUptodate(page);
 156         remove_from_page_cache(page);
 157         put_page(page);
 158 }
 159
 160 void truncate_hugepages(struct address_space *mapping, loff_t lstart)
 161 {
 162         const pgoff_t start = lstart >> HPAGE_SHIFT;
 163         struct pagevec pvec;
 164         pgoff_t next;
 165         int i;
 166
 167         pagevec_init(&pvec, 0);
 168         next = start;
 169         while (1) {
 170                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 171                         if (next == start)
 172                                 break;
 173                         next = start;
 174                         continue;
 175                 }
 176
 177                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 178                         struct page *page = pvec.pages[i];
 179
 180                         lock_page(page);
 181                         if (page->index > next)
 182                                 next = page->index;
 183                         ++next;
 184                         truncate_huge_page(page);
 185                         unlock_page(page);
 186                         hugetlb_put_quota(mapping);
 187                 }
 188                 huge_pagevec_release(&pvec);
 189         }
 190         BUG_ON(!lstart && mapping->nrpages);
 191 }
 192
 193 static void hugetlbfs_delete_inode(struct inode *inode)
 194 {
 195         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(inode->i_sb);
 196
 197         hlist_del_init(&inode->i_hash);
 198         list_del_init(&inode->i_list);
 199         inode->i_state |= I_FREEING;
 200         inodes_stat.nr_inodes--;
 201         spin_unlock(&inode_lock);
 202
 203         if (inode->i_data.nrpages)
 204                 truncate_hugepages(&inode->i_data, 0);
 205
 206         security_inode_delete(inode);
 207
 208         if (sbinfo->free_inodes >= 0) {
 209                 spin_lock(&sbinfo->stat_lock);
 210                 sbinfo->free_inodes++;
 211                 spin_unlock(&sbinfo->stat_lock);
 212         }
 213
 214         clear_inode(inode);
 215         destroy_inode(inode);
 216 }
 217
 218 static void hugetlbfs_forget_inode(struct inode *inode)
 219 {
 220         struct super_block *super_block = inode->i_sb;
 221         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(super_block);
 222
 223         if (hlist_unhashed(&inode->i_hash))
 224                 goto out_truncate;
 225
 226         if (!(inode->i_state & (I_DIRTY|I_LOCK))) {
 227                 list_del(&inode->i_list);
 228                 list_add(&inode->i_list, &inode_unused);
 229         }
 230         inodes_stat.nr_unused++;
 231         if (!super_block || (super_block->s_flags & MS_ACTIVE)) {
 232                 spin_unlock(&inode_lock);
 233                 return;
 234         }
 235
 236         /* write_inode_now() ? */
 237         inodes_stat.nr_unused--;
 238         hlist_del_init(&inode->i_hash);
 239 out_truncate:
 240         list_del_init(&inode->i_list);
 241         inode->i_state |= I_FREEING;
 242         inodes_stat.nr_inodes--;
 243         spin_unlock(&inode_lock);
 244         if (inode->i_data.nrpages)
 245                 truncate_hugepages(&inode->i_data, 0);
 246
 247         if (sbinfo->free_inodes >= 0) {
 248                 spin_lock(&sbinfo->stat_lock);
 249                 sbinfo->free_inodes++;
 250                 spin_unlock(&sbinfo->stat_lock);
 251         }
 252
 253         clear_inode(inode);
 254         destroy_inode(inode);
 255 }
 256
 257 static void hugetlbfs_drop_inode(struct inode *inode)
 258 {
 259         if (!inode->i_nlink)
 260                 hugetlbfs_delete_inode(inode);
 261         else
 262                 hugetlbfs_forget_inode(inode);
 263 }
 264
 265 /*
 266  * h_pgoff is in HPAGE_SIZE units.
 267  * vma->vm_pgoff is in PAGE_SIZE units.
 268  */
 269 static inline void
 270 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
 271 {
 272         struct vm_area_struct *vma = NULL;
 273         struct prio_tree_iter iter;
 274
 275         while ((vma = vma_prio_tree_next(vma, root, &iter,
 276                                         h_pgoff, ULONG_MAX)) != NULL) {
 277                 unsigned long h_vm_pgoff;
 278                 unsigned long v_length;
 279                 unsigned long v_offset;
 280
 281                 h_vm_pgoff = vma->vm_pgoff << (HPAGE_SHIFT - PAGE_SHIFT);
 282                 v_length = vma->vm_end - vma->vm_start;
 283                 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
 284
 285                 /*
 286                  * Is this VMA fully outside the truncation point?
 287                  */
 288                 if (h_vm_pgoff >= h_pgoff)
 289                         v_offset = 0;
 290
 291                 zap_hugepage_range(vma,
 292                                 vma->vm_start + v_offset,
 293                                 v_length - v_offset);
 294         }
 295 }
 296
 297 /*
 298  * Expanding truncates are not allowed.
 299  */
 300 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 301 {
 302         unsigned long pgoff;
 303         struct address_space *mapping = inode->i_mapping;
 304
 305         if (offset > inode->i_size)
 306                 return -EINVAL;
 307
 308         BUG_ON(offset & ~HPAGE_MASK);
 309         pgoff = offset >> HPAGE_SHIFT;
 310
 311         inode->i_size = offset;
 312         spin_lock(&mapping->i_mmap_lock);
 313         if (!prio_tree_empty(&mapping->i_mmap))
 314                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 315         spin_unlock(&mapping->i_mmap_lock);
 316         truncate_hugepages(mapping, offset);
 317         return 0;
 318 }
 319
 320 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 321 {
 322         struct inode *inode = dentry->d_inode;
 323         int error;
 324         unsigned int ia_valid = attr->ia_valid;
 325
 326         BUG_ON(!inode);
 327
 328         error = inode_change_ok(inode, attr);
 329         if (error)
 330                 goto out;
 331
 332         error = security_inode_setattr(dentry, attr);
 333         if (error)
 334                 goto out;
 335         if (ia_valid & ATTR_SIZE) {
 336                 error = -EINVAL;
 337                 if (!(attr->ia_size & ~HPAGE_MASK))
 338                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 339                 if (error)
 340                         goto out;
 341                 attr->ia_valid &= ~ATTR_SIZE;
 342         }
 343         error = inode_setattr(inode, attr);
 344 out:
 345         return error;
 346 }
 347
 348 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 349                                         gid_t gid, int mode, dev_t dev)
 350 {
 351         struct inode *inode;
 352         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 353
 354         if (sbinfo->free_inodes >= 0) {
 355                 spin_lock(&sbinfo->stat_lock);
 356                 if (!sbinfo->free_inodes) {
 357                         spin_unlock(&sbinfo->stat_lock);
 358                         return NULL;
 359                 }
 360                 sbinfo->free_inodes--;
 361                 spin_unlock(&sbinfo->stat_lock);
 362         }
 363
 364         inode = new_inode(sb);
 365         if (inode) {
 366                 struct hugetlbfs_inode_info *info;
 367                 inode->i_mode = mode;
 368                 inode->i_uid = uid;
 369                 inode->i_gid = gid;
 370                 inode->i_blksize = HPAGE_SIZE;
 371                 inode->i_blocks = 0;
 372                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 373                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 374                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 375                 info = HUGETLBFS_I(inode);
 376                 mpol_shared_policy_init(&info->policy);
 377                 switch (mode & S_IFMT) {
 378                 default:
 379                         init_special_inode(inode, mode, dev);
 380                         break;
 381                 case S_IFREG:
 382                         inode->i_op = &hugetlbfs_inode_operations;
 383                         inode->i_fop = &hugetlbfs_file_operations;
 384                         break;
 385                 case S_IFDIR:
 386                         inode->i_op = &hugetlbfs_dir_inode_operations;
 387                         inode->i_fop = &simple_dir_operations;
 388
 389                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 390                         inode->i_nlink++;
 391                         break;
 392                 case S_IFLNK:
 393                         inode->i_op = &page_symlink_inode_operations;
 394                         break;
 395                 }
 396         }
 397         return inode;
 398 }
 399
 400 /*
 401  * File creation. Allocate an inode, and we're done..
 402  */
 403 static int hugetlbfs_mknod(struct inode *dir,
 404                         struct dentry *dentry, int mode, dev_t dev)
 405 {
 406         struct inode *inode;
 407         int error = -ENOSPC;
 408         gid_t gid;
 409
 410         if (dir->i_mode & S_ISGID) {
 411                 gid = dir->i_gid;
 412                 if (S_ISDIR(mode))
 413                         mode |= S_ISGID;
 414         } else {
 415                 gid = current->fsgid;
 416         }
 417         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 418         if (inode) {
 419                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 420                 d_instantiate(dentry, inode);
 421                 dget(dentry);   /* Extra count - pin the dentry in core */
 422                 error = 0;
 423         }
 424         return error;
 425 }
 426
 427 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 428 {
 429         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 430         if (!retval)
 431                 dir->i_nlink++;
 432         return retval;
 433 }
 434
 435 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 436 {
 437         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 438 }
 439
 440 static int hugetlbfs_symlink(struct inode *dir,
 441                         struct dentry *dentry, const char *symname)
 442 {
 443         struct inode *inode;
 444         int error = -ENOSPC;
 445         gid_t gid;
 446
 447         if (dir->i_mode & S_ISGID)
 448                 gid = dir->i_gid;
 449         else
 450                 gid = current->fsgid;
 451
 452         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 453                                         gid, S_IFLNK|S_IRWXUGO, 0);
 454         if (inode) {
 455                 int l = strlen(symname)+1;
 456                 error = page_symlink(inode, symname, l);
 457                 if (!error) {
 458                         d_instantiate(dentry, inode);
 459                         dget(dentry);
 460                 } else
 461                         iput(inode);
 462         }
 463         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 464
 465         return error;
 466 }
 467
 468 /*
 469  * For direct-IO reads into hugetlb pages
 470  */
 471 int hugetlbfs_set_page_dirty(struct page *page)
 472 {
 473         return 0;
 474 }
 475
 476 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
 477 {
 478         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 479
 480         buf->f_type = HUGETLBFS_MAGIC;
 481         buf->f_bsize = HPAGE_SIZE;
 482         if (sbinfo) {
 483                 spin_lock(&sbinfo->stat_lock);
 484                 buf->f_blocks = sbinfo->max_blocks;
 485                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 486                 buf->f_files = sbinfo->max_inodes;
 487                 buf->f_ffree = sbinfo->free_inodes;
 488                 spin_unlock(&sbinfo->stat_lock);
 489         }
 490         buf->f_namelen = NAME_MAX;
 491         return 0;
 492 }
 493
 494 static void hugetlbfs_put_super(struct super_block *sb)
 495 {
 496         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 497
 498         if (sbi) {
 499                 sb->s_fs_info = NULL;
 500                 kfree(sbi);
 501         }
 502 }
 503
 504 static kmem_cache_t *hugetlbfs_inode_cachep;
 505
 506 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 507 {
 508         struct hugetlbfs_inode_info *p;
 509
 510         p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
 511         if (!p)
 512                 return NULL;
 513         return &p->vfs_inode;
 514 }
 515
 516 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
 517 {
 518         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 519
 520         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 521             SLAB_CTOR_CONSTRUCTOR)
 522                 inode_init_once(&ei->vfs_inode);
 523 }
 524
 525 static void hugetlbfs_destroy_inode(struct inode *inode)
 526 {
 527         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 528         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 529 }
 530
 531 static struct address_space_operations hugetlbfs_aops = {
 532         .readpage       = hugetlbfs_readpage,
 533         .prepare_write  = hugetlbfs_prepare_write,
 534         .commit_write   = hugetlbfs_commit_write,
 535         .set_page_dirty = hugetlbfs_set_page_dirty,
 536 };
 537
 538 struct file_operations hugetlbfs_file_operations = {
 539         .mmap                   = hugetlbfs_file_mmap,
 540         .fsync                  = simple_sync_file,
 541         .get_unmapped_area      = hugetlb_get_unmapped_area,
 542 };
 543
 544 static struct inode_operations hugetlbfs_dir_inode_operations = {
 545         .create         = hugetlbfs_create,
 546         .lookup         = simple_lookup,
 547         .link           = simple_link,
 548         .unlink         = simple_unlink,
 549         .symlink        = hugetlbfs_symlink,
 550         .mkdir          = hugetlbfs_mkdir,
 551         .rmdir          = simple_rmdir,
 552         .mknod          = hugetlbfs_mknod,
 553         .rename         = simple_rename,
 554         .setattr        = hugetlbfs_setattr,
 555 };
 556
 557 static struct inode_operations hugetlbfs_inode_operations = {
 558         .setattr        = hugetlbfs_setattr,
 559 };
 560
 561 static struct super_operations hugetlbfs_ops = {
 562         .alloc_inode    = hugetlbfs_alloc_inode,
 563         .destroy_inode  = hugetlbfs_destroy_inode,
 564         .statfs         = hugetlbfs_statfs,
 565         .drop_inode     = hugetlbfs_drop_inode,
 566         .put_super      = hugetlbfs_put_super,
 567 };
 568
 569 static int
 570 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 571 {
 572         char *opt, *value, *rest;
 573
 574         if (!options)
 575                 return 0;
 576         while ((opt = strsep(&options, ",")) != NULL) {
 577                 if (!*opt)
 578                         continue;
 579
 580                 value = strchr(opt, '=');
 581                 if (!value || !*value)
 582                         return -EINVAL;
 583                 else
 584                         *value++ = '\0';
 585
 586                 if (!strcmp(opt, "uid"))
 587                         pconfig->uid = simple_strtoul(value, &value, 0);
 588                 else if (!strcmp(opt, "gid"))
 589                         pconfig->gid = simple_strtoul(value, &value, 0);
 590                 else if (!strcmp(opt, "mode"))
 591                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
 592                 else if (!strcmp(opt, "size")) {
 593                         unsigned long long size = memparse(value, &rest);
 594                         if (*rest == '%') {
 595                                 size <<= HPAGE_SHIFT;
 596                                 size *= max_huge_pages;
 597                                 do_div(size, 100);
 598                                 rest++;
 599                         }
 600                         size &= HPAGE_MASK;
 601                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 602                         value = rest;
 603                 } else if (!strcmp(opt,"nr_inodes")) {
 604                         pconfig->nr_inodes = memparse(value, &rest);
 605                         value = rest;
 606                 } else
 607                         return -EINVAL;
 608
 609                 if (*value)
 610                         return -EINVAL;
 611         }
 612         return 0;
 613 }
 614
 615 static int
 616 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 617 {
 618         struct inode * inode;
 619         struct dentry * root;
 620         int ret;
 621         struct hugetlbfs_config config;
 622         struct hugetlbfs_sb_info *sbinfo;
 623
 624         config.nr_blocks = -1; /* No limit on size by default */
 625         config.nr_inodes = -1; /* No limit on number of inodes by default */
 626         config.uid = current->fsuid;
 627         config.gid = current->fsgid;
 628         config.mode = 0755;
 629         ret = hugetlbfs_parse_options(data, &config);
 630
 631         if (ret)
 632                 return ret;
 633
 634         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 635         if (!sbinfo)
 636                 return -ENOMEM;
 637         sb->s_fs_info = sbinfo;
 638         spin_lock_init(&sbinfo->stat_lock);
 639         sbinfo->max_blocks = config.nr_blocks;
 640         sbinfo->free_blocks = config.nr_blocks;
 641         sbinfo->max_inodes = config.nr_inodes;
 642         sbinfo->free_inodes = config.nr_inodes;
 643         sb->s_blocksize = HPAGE_SIZE;
 644         sb->s_blocksize_bits = HPAGE_SHIFT;
 645         sb->s_magic = HUGETLBFS_MAGIC;
 646         sb->s_op = &hugetlbfs_ops;
 647         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 648                                         S_IFDIR | config.mode, 0);
 649         if (!inode)
 650                 goto out_free;
 651
 652         root = d_alloc_root(inode);
 653         if (!root) {
 654                 iput(inode);
 655                 goto out_free;
 656         }
 657         sb->s_root = root;
 658         return 0;
 659 out_free:
 660         kfree(sbinfo);
 661         return -ENOMEM;
 662 }
 663
 664 int hugetlb_get_quota(struct address_space *mapping)
 665 {
 666         int ret = 0;
 667         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 668
 669         if (sbinfo->free_blocks > -1) {
 670                 spin_lock(&sbinfo->stat_lock);
 671                 if (sbinfo->free_blocks > 0)
 672                         sbinfo->free_blocks--;
 673                 else
 674                         ret = -ENOMEM;
 675                 spin_unlock(&sbinfo->stat_lock);
 676         }
 677
 678         return ret;
 679 }
 680
 681 void hugetlb_put_quota(struct address_space *mapping)
 682 {
 683         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 684
 685         if (sbinfo->free_blocks > -1) {
 686                 spin_lock(&sbinfo->stat_lock);
 687                 sbinfo->free_blocks++;
 688                 spin_unlock(&sbinfo->stat_lock);
 689         }
 690 }
 691
 692 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
 693         int flags, const char *dev_name, void *data)
 694 {
 695         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 696 }
 697
 698 static struct file_system_type hugetlbfs_fs_type = {
 699         .name           = "hugetlbfs",
 700         .get_sb         = hugetlbfs_get_sb,
 701         .kill_sb        = kill_litter_super,
 702 };
 703
 704 static struct vfsmount *hugetlbfs_vfsmount;
 705
 706 /*
 707  * Return the next identifier for a shm file
 708  */
 709 static unsigned long hugetlbfs_counter(void)
 710 {
 711         static spinlock_t lock = SPIN_LOCK_UNLOCKED;
 712         static unsigned long counter;
 713         unsigned long ret;
 714
 715         spin_lock(&lock);
 716         ret = ++counter;
 717         spin_unlock(&lock);
 718         return ret;
 719 }
 720
 721 static int can_do_hugetlb_shm(void)
 722 {
 723         return likely(capable(CAP_IPC_LOCK) ||
 724                         in_group_p(sysctl_hugetlb_shm_group));
 725 }
 726
 727 struct file *hugetlb_zero_setup(size_t size)
 728 {
 729         int error;
 730         struct file *file;
 731         struct inode *inode;
 732         struct dentry *dentry, *root;
 733         struct qstr quick_string;
 734         char buf[16];
 735
 736         if (!can_do_hugetlb_shm())
 737                 return ERR_PTR(-EPERM);
 738
 739         if (!is_hugepage_mem_enough(size))
 740                 return ERR_PTR(-ENOMEM);
 741
 742         root = hugetlbfs_vfsmount->mnt_root;
 743         snprintf(buf, 16, "%lu", hugetlbfs_counter());
 744         quick_string.name = buf;
 745         quick_string.len = strlen(quick_string.name);
 746         quick_string.hash = 0;
 747         dentry = d_alloc(root, &quick_string);
 748         if (!dentry)
 749                 return ERR_PTR(-ENOMEM);
 750
 751         error = -ENFILE;
 752         file = get_empty_filp();
 753         if (!file)
 754                 goto out_dentry;
 755
 756         error = -ENOSPC;
 757         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
 758                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
 759         if (!inode)
 760                 goto out_file;
 761
 762         d_instantiate(dentry, inode);
 763         inode->i_size = size;
 764         inode->i_nlink = 0;
 765         file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
 766         file->f_dentry = dentry;
 767         file->f_mapping = inode->i_mapping;
 768         file->f_op = &hugetlbfs_file_operations;
 769         file->f_mode = FMODE_WRITE | FMODE_READ;
 770         return file;
 771
 772 out_file:
 773         put_filp(file);
 774 out_dentry:
 775         dput(dentry);
 776         return ERR_PTR(error);
 777 }
 778
 779 static int __init init_hugetlbfs_fs(void)
 780 {
 781         int error;
 782         struct vfsmount *vfsmount;
 783
 784         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
 785                                         sizeof(struct hugetlbfs_inode_info),
 786                                         0, SLAB_RECLAIM_ACCOUNT,
 787                                         init_once, NULL);
 788         if (hugetlbfs_inode_cachep == NULL)
 789                 return -ENOMEM;
 790
 791         error = register_filesystem(&hugetlbfs_fs_type);
 792         if (error)
 793                 goto out;
 794
 795         vfsmount = kern_mount(&hugetlbfs_fs_type);
 796
 797         if (!IS_ERR(vfsmount)) {
 798                 hugetlbfs_vfsmount = vfsmount;
 799                 return 0;
 800         }
 801
 802         error = PTR_ERR(vfsmount);
 803
 804  out:
 805         if (error)
 806                 kmem_cache_destroy(hugetlbfs_inode_cachep);
 807         return error;
 808 }
 809
 810 static void __exit exit_hugetlbfs_fs(void)
 811 {
 812         kmem_cache_destroy(hugetlbfs_inode_cachep);
 813         unregister_filesystem(&hugetlbfs_fs_type);
 814 }
 815
 816 module_init(init_hugetlbfs_fs)
 817 module_exit(exit_hugetlbfs_fs)
 818
 819 MODULE_LICENSE("GPL");