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         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
  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_flags & (VM_MAYSHARE | VM_WRITE)) == VM_WRITE)
  56                 return -EINVAL;
  57
  58         if (vma->vm_pgoff & (HPAGE_SIZE / PAGE_SIZE - 1))
  59                 return -EINVAL;
  60
  61         if (vma->vm_start & ~HPAGE_MASK)
  62                 return -EINVAL;
  63
  64         if (vma->vm_end & ~HPAGE_MASK)
  65                 return -EINVAL;
  66
  67         if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
  68                 return -EINVAL;
  69
  70         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
  71
  72         down(&inode->i_sem);
  73         file_accessed(file);
  74         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
  75         vma->vm_ops = &hugetlb_vm_ops;
  76
  77         ret = -ENOMEM;
  78         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
  79         if (!(vma->vm_flags & VM_WRITE) && len > inode->i_size)
  80                 goto out;
  81
  82         ret = hugetlb_prefault(mapping, vma);
  83         if (ret)
  84                 goto out;
  85
  86         if (inode->i_size < len)
  87                 inode->i_size = len;
  88 out:
  89         up(&inode->i_sem);
  90
  91         return ret;
  92 }
  93
  94 /*
  95  * Called under down_write(mmap_sem).
  96  */
  97
  98 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
  99 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 100                 unsigned long len, unsigned long pgoff, unsigned long flags);
 101 #else
 102 static unsigned long
 103 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 104                 unsigned long len, unsigned long pgoff, unsigned long flags)
 105 {
 106         struct mm_struct *mm = current->mm;
 107         struct vm_area_struct *vma;
 108         unsigned long start_addr;
 109
 110         if (len & ~HPAGE_MASK)
 111                 return -EINVAL;
 112         if (len > TASK_SIZE)
 113                 return -ENOMEM;
 114
 115         if (addr) {
 116                 addr = ALIGN(addr, HPAGE_SIZE);
 117                 vma = find_vma(mm, addr);
 118                 if (TASK_SIZE - len >= addr &&
 119                     (!vma || addr + len <= vma->vm_start))
 120                         return addr;
 121         }
 122
 123         start_addr = mm->free_area_cache;
 124
 125         if (len <= mm->cached_hole_size)
 126                 start_addr = TASK_UNMAPPED_BASE;
 127
 128 full_search:
 129         addr = ALIGN(start_addr, HPAGE_SIZE);
 130
 131         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 132                 /* At this point:  (!vma || addr < vma->vm_end). */
 133                 if (TASK_SIZE - len < addr) {
 134                         /*
 135                          * Start a new search - just in case we missed
 136                          * some holes.
 137                          */
 138                         if (start_addr != TASK_UNMAPPED_BASE) {
 139                                 start_addr = TASK_UNMAPPED_BASE;
 140                                 goto full_search;
 141                         }
 142                         return -ENOMEM;
 143                 }
 144
 145                 if (!vma || addr + len <= vma->vm_start)
 146                         return addr;
 147                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
 148         }
 149 }
 150 #endif
 151
 152 /*
 153  * Read a page. Again trivial. If it didn't already exist
 154  * in the page cache, it is zero-filled.
 155  */
 156 static int hugetlbfs_readpage(struct file *file, struct page * page)
 157 {
 158         unlock_page(page);
 159         return -EINVAL;
 160 }
 161
 162 static int hugetlbfs_prepare_write(struct file *file,
 163                         struct page *page, unsigned offset, unsigned to)
 164 {
 165         return -EINVAL;
 166 }
 167
 168 static int hugetlbfs_commit_write(struct file *file,
 169                         struct page *page, unsigned offset, unsigned to)
 170 {
 171         return -EINVAL;
 172 }
 173
 174 static void huge_pagevec_release(struct pagevec *pvec)
 175 {
 176         int i;
 177
 178         for (i = 0; i < pagevec_count(pvec); ++i)
 179                 put_page(pvec->pages[i]);
 180
 181         pagevec_reinit(pvec);
 182 }
 183
 184 static void truncate_huge_page(struct page *page)
 185 {
 186         clear_page_dirty(page);
 187         ClearPageUptodate(page);
 188         remove_from_page_cache(page);
 189         put_page(page);
 190 }
 191
 192 static void truncate_hugepages(struct address_space *mapping, loff_t lstart)
 193 {
 194         const pgoff_t start = lstart >> HPAGE_SHIFT;
 195         struct pagevec pvec;
 196         pgoff_t next;
 197         int i;
 198
 199         pagevec_init(&pvec, 0);
 200         next = start;
 201         while (1) {
 202                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 203                         if (next == start)
 204                                 break;
 205                         next = start;
 206                         continue;
 207                 }
 208
 209                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 210                         struct page *page = pvec.pages[i];
 211
 212                         lock_page(page);
 213                         if (page->index > next)
 214                                 next = page->index;
 215                         ++next;
 216                         truncate_huge_page(page);
 217                         unlock_page(page);
 218                         hugetlb_put_quota(mapping);
 219                 }
 220                 huge_pagevec_release(&pvec);
 221         }
 222         BUG_ON(!lstart && mapping->nrpages);
 223 }
 224
 225 static void hugetlbfs_delete_inode(struct inode *inode)
 226 {
 227         if (inode->i_data.nrpages)
 228                 truncate_hugepages(&inode->i_data, 0);
 229         clear_inode(inode);
 230 }
 231
 232 static void hugetlbfs_forget_inode(struct inode *inode)
 233 {
 234         struct super_block *sb = inode->i_sb;
 235
 236         if (!hlist_unhashed(&inode->i_hash)) {
 237                 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
 238                         list_move(&inode->i_list, &inode_unused);
 239                 inodes_stat.nr_unused++;
 240                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
 241                         spin_unlock(&inode_lock);
 242                         return;
 243                 }
 244                 inode->i_state |= I_WILL_FREE;
 245                 spin_unlock(&inode_lock);
 246                 /*
 247                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
 248                  * in our backing_dev_info.
 249                  */
 250                 write_inode_now(inode, 1);
 251                 spin_lock(&inode_lock);
 252                 inode->i_state &= ~I_WILL_FREE;
 253                 inodes_stat.nr_unused--;
 254                 hlist_del_init(&inode->i_hash);
 255         }
 256         list_del_init(&inode->i_list);
 257         list_del_init(&inode->i_sb_list);
 258         inode->i_state |= I_FREEING;
 259         inodes_stat.nr_inodes--;
 260         spin_unlock(&inode_lock);
 261         if (inode->i_data.nrpages)
 262                 truncate_hugepages(&inode->i_data, 0);
 263         clear_inode(inode);
 264         destroy_inode(inode);
 265 }
 266
 267 static void hugetlbfs_drop_inode(struct inode *inode)
 268 {
 269         if (!inode->i_nlink)
 270                 generic_delete_inode(inode);
 271         else
 272                 hugetlbfs_forget_inode(inode);
 273 }
 274
 275 /*
 276  * h_pgoff is in HPAGE_SIZE units.
 277  * vma->vm_pgoff is in PAGE_SIZE units.
 278  */
 279 static inline void
 280 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
 281 {
 282         struct vm_area_struct *vma;
 283         struct prio_tree_iter iter;
 284
 285         vma_prio_tree_foreach(vma, &iter, root, h_pgoff, ULONG_MAX) {
 286                 unsigned long h_vm_pgoff;
 287                 unsigned long v_offset;
 288
 289                 h_vm_pgoff = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
 290                 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
 291                 /*
 292                  * Is this VMA fully outside the truncation point?
 293                  */
 294                 if (h_vm_pgoff >= h_pgoff)
 295                         v_offset = 0;
 296
 297                 unmap_hugepage_range(vma,
 298                                 vma->vm_start + v_offset, vma->vm_end);
 299         }
 300 }
 301
 302 /*
 303  * Expanding truncates are not allowed.
 304  */
 305 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 306 {
 307         unsigned long pgoff;
 308         struct address_space *mapping = inode->i_mapping;
 309
 310         if (offset > inode->i_size)
 311                 return -EINVAL;
 312
 313         BUG_ON(offset & ~HPAGE_MASK);
 314         pgoff = offset >> HPAGE_SHIFT;
 315
 316         inode->i_size = offset;
 317         spin_lock(&mapping->i_mmap_lock);
 318         if (!prio_tree_empty(&mapping->i_mmap))
 319                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 320         spin_unlock(&mapping->i_mmap_lock);
 321         truncate_hugepages(mapping, offset);
 322         return 0;
 323 }
 324
 325 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 326 {
 327         struct inode *inode = dentry->d_inode;
 328         int error;
 329         unsigned int ia_valid = attr->ia_valid;
 330
 331         BUG_ON(!inode);
 332
 333         error = inode_change_ok(inode, attr);
 334         if (error)
 335                 goto out;
 336
 337         if (ia_valid & ATTR_SIZE) {
 338                 error = -EINVAL;
 339                 if (!(attr->ia_size & ~HPAGE_MASK))
 340                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 341                 if (error)
 342                         goto out;
 343                 attr->ia_valid &= ~ATTR_SIZE;
 344         }
 345         error = inode_setattr(inode, attr);
 346 out:
 347         return error;
 348 }
 349
 350 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 351                                         gid_t gid, int mode, dev_t dev)
 352 {
 353         struct inode *inode;
 354
 355         inode = new_inode(sb);
 356         if (inode) {
 357                 struct hugetlbfs_inode_info *info;
 358                 inode->i_mode = mode;
 359                 inode->i_uid = uid;
 360                 inode->i_gid = gid;
 361                 inode->i_blksize = HPAGE_SIZE;
 362                 inode->i_blocks = 0;
 363                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 364                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 365                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 366                 info = HUGETLBFS_I(inode);
 367                 mpol_shared_policy_init(&info->policy);
 368                 switch (mode & S_IFMT) {
 369                 default:
 370                         init_special_inode(inode, mode, dev);
 371                         break;
 372                 case S_IFREG:
 373                         inode->i_op = &hugetlbfs_inode_operations;
 374                         inode->i_fop = &hugetlbfs_file_operations;
 375                         break;
 376                 case S_IFDIR:
 377                         inode->i_op = &hugetlbfs_dir_inode_operations;
 378                         inode->i_fop = &simple_dir_operations;
 379
 380                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 381                         inode->i_nlink++;
 382                         break;
 383                 case S_IFLNK:
 384                         inode->i_op = &page_symlink_inode_operations;
 385                         break;
 386                 }
 387         }
 388         return inode;
 389 }
 390
 391 /*
 392  * File creation. Allocate an inode, and we're done..
 393  */
 394 static int hugetlbfs_mknod(struct inode *dir,
 395                         struct dentry *dentry, int mode, dev_t dev)
 396 {
 397         struct inode *inode;
 398         int error = -ENOSPC;
 399         gid_t gid;
 400
 401         if (dir->i_mode & S_ISGID) {
 402                 gid = dir->i_gid;
 403                 if (S_ISDIR(mode))
 404                         mode |= S_ISGID;
 405         } else {
 406                 gid = current->fsgid;
 407         }
 408         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 409         if (inode) {
 410                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 411                 d_instantiate(dentry, inode);
 412                 dget(dentry);   /* Extra count - pin the dentry in core */
 413                 error = 0;
 414         }
 415         return error;
 416 }
 417
 418 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 419 {
 420         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 421         if (!retval)
 422                 dir->i_nlink++;
 423         return retval;
 424 }
 425
 426 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 427 {
 428         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 429 }
 430
 431 static int hugetlbfs_symlink(struct inode *dir,
 432                         struct dentry *dentry, const char *symname)
 433 {
 434         struct inode *inode;
 435         int error = -ENOSPC;
 436         gid_t gid;
 437
 438         if (dir->i_mode & S_ISGID)
 439                 gid = dir->i_gid;
 440         else
 441                 gid = current->fsgid;
 442
 443         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 444                                         gid, S_IFLNK|S_IRWXUGO, 0);
 445         if (inode) {
 446                 int l = strlen(symname)+1;
 447                 error = page_symlink(inode, symname, l);
 448                 if (!error) {
 449                         d_instantiate(dentry, inode);
 450                         dget(dentry);
 451                 } else
 452                         iput(inode);
 453         }
 454         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 455
 456         return error;
 457 }
 458
 459 /*
 460  * For direct-IO reads into hugetlb pages
 461  */
 462 static int hugetlbfs_set_page_dirty(struct page *page)
 463 {
 464         return 0;
 465 }
 466
 467 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
 468 {
 469         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 470
 471         buf->f_type = HUGETLBFS_MAGIC;
 472         buf->f_bsize = HPAGE_SIZE;
 473         if (sbinfo) {
 474                 spin_lock(&sbinfo->stat_lock);
 475                 buf->f_blocks = sbinfo->max_blocks;
 476                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 477                 buf->f_files = sbinfo->max_inodes;
 478                 buf->f_ffree = sbinfo->free_inodes;
 479                 spin_unlock(&sbinfo->stat_lock);
 480         }
 481         buf->f_namelen = NAME_MAX;
 482         return 0;
 483 }
 484
 485 static void hugetlbfs_put_super(struct super_block *sb)
 486 {
 487         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 488
 489         if (sbi) {
 490                 sb->s_fs_info = NULL;
 491                 kfree(sbi);
 492         }
 493 }
 494
 495 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 496 {
 497         if (sbinfo->free_inodes >= 0) {
 498                 spin_lock(&sbinfo->stat_lock);
 499                 if (unlikely(!sbinfo->free_inodes)) {
 500                         spin_unlock(&sbinfo->stat_lock);
 501                         return 0;
 502                 }
 503                 sbinfo->free_inodes--;
 504                 spin_unlock(&sbinfo->stat_lock);
 505         }
 506
 507         return 1;
 508 }
 509
 510 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 511 {
 512         if (sbinfo->free_inodes >= 0) {
 513                 spin_lock(&sbinfo->stat_lock);
 514                 sbinfo->free_inodes++;
 515                 spin_unlock(&sbinfo->stat_lock);
 516         }
 517 }
 518
 519
 520 static kmem_cache_t *hugetlbfs_inode_cachep;
 521
 522 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 523 {
 524         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 525         struct hugetlbfs_inode_info *p;
 526
 527         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 528                 return NULL;
 529         p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
 530         if (unlikely(!p)) {
 531                 hugetlbfs_inc_free_inodes(sbinfo);
 532                 return NULL;
 533         }
 534         return &p->vfs_inode;
 535 }
 536
 537 static void hugetlbfs_destroy_inode(struct inode *inode)
 538 {
 539         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 540         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 541         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 542 }
 543
 544 static struct address_space_operations hugetlbfs_aops = {
 545         .readpage       = hugetlbfs_readpage,
 546         .prepare_write  = hugetlbfs_prepare_write,
 547         .commit_write   = hugetlbfs_commit_write,
 548         .set_page_dirty = hugetlbfs_set_page_dirty,
 549 };
 550
 551
 552 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
 553 {
 554         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 555
 556         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 557             SLAB_CTOR_CONSTRUCTOR)
 558                 inode_init_once(&ei->vfs_inode);
 559 }
 560
 561 struct file_operations hugetlbfs_file_operations = {
 562         .mmap                   = hugetlbfs_file_mmap,
 563         .fsync                  = simple_sync_file,
 564         .get_unmapped_area      = hugetlb_get_unmapped_area,
 565 };
 566
 567 static struct inode_operations hugetlbfs_dir_inode_operations = {
 568         .create         = hugetlbfs_create,
 569         .lookup         = simple_lookup,
 570         .link           = simple_link,
 571         .unlink         = simple_unlink,
 572         .symlink        = hugetlbfs_symlink,
 573         .mkdir          = hugetlbfs_mkdir,
 574         .rmdir          = simple_rmdir,
 575         .mknod          = hugetlbfs_mknod,
 576         .rename         = simple_rename,
 577         .setattr        = hugetlbfs_setattr,
 578 };
 579
 580 static struct inode_operations hugetlbfs_inode_operations = {
 581         .setattr        = hugetlbfs_setattr,
 582 };
 583
 584 static struct super_operations hugetlbfs_ops = {
 585         .alloc_inode    = hugetlbfs_alloc_inode,
 586         .destroy_inode  = hugetlbfs_destroy_inode,
 587         .statfs         = hugetlbfs_statfs,
 588         .delete_inode   = hugetlbfs_delete_inode,
 589         .drop_inode     = hugetlbfs_drop_inode,
 590         .put_super      = hugetlbfs_put_super,
 591 };
 592
 593 static int
 594 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 595 {
 596         char *opt, *value, *rest;
 597
 598         if (!options)
 599                 return 0;
 600         while ((opt = strsep(&options, ",")) != NULL) {
 601                 if (!*opt)
 602                         continue;
 603
 604                 value = strchr(opt, '=');
 605                 if (!value || !*value)
 606                         return -EINVAL;
 607                 else
 608                         *value++ = '\0';
 609
 610                 if (!strcmp(opt, "uid"))
 611                         pconfig->uid = simple_strtoul(value, &value, 0);
 612                 else if (!strcmp(opt, "gid"))
 613                         pconfig->gid = simple_strtoul(value, &value, 0);
 614                 else if (!strcmp(opt, "mode"))
 615                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
 616                 else if (!strcmp(opt, "size")) {
 617                         unsigned long long size = memparse(value, &rest);
 618                         if (*rest == '%') {
 619                                 size <<= HPAGE_SHIFT;
 620                                 size *= max_huge_pages;
 621                                 do_div(size, 100);
 622                                 rest++;
 623                         }
 624                         size &= HPAGE_MASK;
 625                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 626                         value = rest;
 627                 } else if (!strcmp(opt,"nr_inodes")) {
 628                         pconfig->nr_inodes = memparse(value, &rest);
 629                         value = rest;
 630                 } else
 631                         return -EINVAL;
 632
 633                 if (*value)
 634                         return -EINVAL;
 635         }
 636         return 0;
 637 }
 638
 639 static int
 640 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 641 {
 642         struct inode * inode;
 643         struct dentry * root;
 644         int ret;
 645         struct hugetlbfs_config config;
 646         struct hugetlbfs_sb_info *sbinfo;
 647
 648         config.nr_blocks = -1; /* No limit on size by default */
 649         config.nr_inodes = -1; /* No limit on number of inodes by default */
 650         config.uid = current->fsuid;
 651         config.gid = current->fsgid;
 652         config.mode = 0755;
 653         ret = hugetlbfs_parse_options(data, &config);
 654
 655         if (ret)
 656                 return ret;
 657
 658         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 659         if (!sbinfo)
 660                 return -ENOMEM;
 661         sb->s_fs_info = sbinfo;
 662         spin_lock_init(&sbinfo->stat_lock);
 663         sbinfo->max_blocks = config.nr_blocks;
 664         sbinfo->free_blocks = config.nr_blocks;
 665         sbinfo->max_inodes = config.nr_inodes;
 666         sbinfo->free_inodes = config.nr_inodes;
 667         sb->s_maxbytes = MAX_LFS_FILESIZE;
 668         sb->s_blocksize = HPAGE_SIZE;
 669         sb->s_blocksize_bits = HPAGE_SHIFT;
 670         sb->s_magic = HUGETLBFS_MAGIC;
 671         sb->s_op = &hugetlbfs_ops;
 672         sb->s_time_gran = 1;
 673         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 674                                         S_IFDIR | config.mode, 0);
 675         if (!inode)
 676                 goto out_free;
 677
 678         root = d_alloc_root(inode);
 679         if (!root) {
 680                 iput(inode);
 681                 goto out_free;
 682         }
 683         sb->s_root = root;
 684         return 0;
 685 out_free:
 686         kfree(sbinfo);
 687         return -ENOMEM;
 688 }
 689
 690 int hugetlb_get_quota(struct address_space *mapping)
 691 {
 692         int ret = 0;
 693         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 694
 695         if (sbinfo->free_blocks > -1) {
 696                 spin_lock(&sbinfo->stat_lock);
 697                 if (sbinfo->free_blocks > 0)
 698                         sbinfo->free_blocks--;
 699                 else
 700                         ret = -ENOMEM;
 701                 spin_unlock(&sbinfo->stat_lock);
 702         }
 703
 704         return ret;
 705 }
 706
 707 void hugetlb_put_quota(struct address_space *mapping)
 708 {
 709         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 710
 711         if (sbinfo->free_blocks > -1) {
 712                 spin_lock(&sbinfo->stat_lock);
 713                 sbinfo->free_blocks++;
 714                 spin_unlock(&sbinfo->stat_lock);
 715         }
 716 }
 717
 718 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
 719         int flags, const char *dev_name, void *data)
 720 {
 721         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 722 }
 723
 724 static struct file_system_type hugetlbfs_fs_type = {
 725         .name           = "hugetlbfs",
 726         .get_sb         = hugetlbfs_get_sb,
 727         .kill_sb        = kill_litter_super,
 728 };
 729
 730 static struct vfsmount *hugetlbfs_vfsmount;
 731
 732 /*
 733  * Return the next identifier for a shm file
 734  */
 735 static unsigned long hugetlbfs_counter(void)
 736 {
 737         static DEFINE_SPINLOCK(lock);
 738         static unsigned long counter;
 739         unsigned long ret;
 740
 741         spin_lock(&lock);
 742         ret = ++counter;
 743         spin_unlock(&lock);
 744         return ret;
 745 }
 746
 747 static int can_do_hugetlb_shm(void)
 748 {
 749         return likely(capable(CAP_IPC_LOCK) ||
 750                         in_group_p(sysctl_hugetlb_shm_group) ||
 751                         can_do_mlock());
 752 }
 753
 754 struct file *hugetlb_zero_setup(size_t size)
 755 {
 756         int error = -ENOMEM;
 757         struct file *file;
 758         struct inode *inode;
 759         struct dentry *dentry, *root;
 760         struct qstr quick_string;
 761         char buf[16];
 762
 763         if (!can_do_hugetlb_shm())
 764                 return ERR_PTR(-EPERM);
 765
 766         if (!is_hugepage_mem_enough(size))
 767                 return ERR_PTR(-ENOMEM);
 768
 769         if (!user_shm_lock(size, current->user))
 770                 return ERR_PTR(-ENOMEM);
 771
 772         root = hugetlbfs_vfsmount->mnt_root;
 773         snprintf(buf, 16, "%lu", hugetlbfs_counter());
 774         quick_string.name = buf;
 775         quick_string.len = strlen(quick_string.name);
 776         quick_string.hash = 0;
 777         dentry = d_alloc(root, &quick_string);
 778         if (!dentry)
 779                 goto out_shm_unlock;
 780
 781         error = -ENFILE;
 782         file = get_empty_filp();
 783         if (!file)
 784                 goto out_dentry;
 785
 786         error = -ENOSPC;
 787         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
 788                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
 789         if (!inode)
 790                 goto out_file;
 791
 792         d_instantiate(dentry, inode);
 793         inode->i_size = size;
 794         inode->i_nlink = 0;
 795         file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
 796         file->f_dentry = dentry;
 797         file->f_mapping = inode->i_mapping;
 798         file->f_op = &hugetlbfs_file_operations;
 799         file->f_mode = FMODE_WRITE | FMODE_READ;
 800         return file;
 801
 802 out_file:
 803         put_filp(file);
 804 out_dentry:
 805         dput(dentry);
 806 out_shm_unlock:
 807         user_shm_unlock(size, current->user);
 808         return ERR_PTR(error);
 809 }
 810
 811 static int __init init_hugetlbfs_fs(void)
 812 {
 813         int error;
 814         struct vfsmount *vfsmount;
 815
 816         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
 817                                         sizeof(struct hugetlbfs_inode_info),
 818                                         0, 0, init_once, NULL);
 819         if (hugetlbfs_inode_cachep == NULL)
 820                 return -ENOMEM;
 821
 822         error = register_filesystem(&hugetlbfs_fs_type);
 823         if (error)
 824                 goto out;
 825
 826         vfsmount = kern_mount(&hugetlbfs_fs_type);
 827
 828         if (!IS_ERR(vfsmount)) {
 829                 hugetlbfs_vfsmount = vfsmount;
 830                 return 0;
 831         }
 832
 833         error = PTR_ERR(vfsmount);
 834
 835  out:
 836         if (error)
 837                 kmem_cache_destroy(hugetlbfs_inode_cachep);
 838         return error;
 839 }
 840
 841 static void __exit exit_hugetlbfs_fs(void)
 842 {
 843         kmem_cache_destroy(hugetlbfs_inode_cachep);
 844         unregister_filesystem(&hugetlbfs_fs_type);
 845 }
 846
 847 module_init(init_hugetlbfs_fs)
 848 module_exit(exit_hugetlbfs_fs)
 849
 850 MODULE_LICENSE("GPL");