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