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_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), page_table_lock is not held
  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 full_search:
 126         addr = ALIGN(start_addr, HPAGE_SIZE);
 127
 128         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 129                 /* At this point:  (!vma || addr < vma->vm_end). */
 130                 if (TASK_SIZE - len < addr) {
 131                         /*
 132                          * Start a new search - just in case we missed
 133                          * some holes.
 134                          */
 135                         if (start_addr != TASK_UNMAPPED_BASE) {
 136                                 start_addr = TASK_UNMAPPED_BASE;
 137                                 goto full_search;
 138                         }
 139                         return -ENOMEM;
 140                 }
 141
 142                 if (!vma || addr + len <= vma->vm_start)
 143                         return addr;
 144                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
 145         }
 146 }
 147 #endif
 148
 149 /*
 150  * Read a page. Again trivial. If it didn't already exist
 151  * in the page cache, it is zero-filled.
 152  */
 153 static int hugetlbfs_readpage(struct file *file, struct page * page)
 154 {
 155         unlock_page(page);
 156         return -EINVAL;
 157 }
 158
 159 static int hugetlbfs_prepare_write(struct file *file,
 160                         struct page *page, unsigned offset, unsigned to)
 161 {
 162         return -EINVAL;
 163 }
 164
 165 static int hugetlbfs_commit_write(struct file *file,
 166                         struct page *page, unsigned offset, unsigned to)
 167 {
 168         return -EINVAL;
 169 }
 170
 171 void huge_pagevec_release(struct pagevec *pvec)
 172 {
 173         int i;
 174
 175         for (i = 0; i < pagevec_count(pvec); ++i)
 176                 put_page(pvec->pages[i]);
 177
 178         pagevec_reinit(pvec);
 179 }
 180
 181 void truncate_huge_page(struct page *page)
 182 {
 183         clear_page_dirty(page);
 184         ClearPageUptodate(page);
 185         remove_from_page_cache(page);
 186         put_page(page);
 187 }
 188
 189 void truncate_hugepages(struct address_space *mapping, loff_t lstart)
 190 {
 191         const pgoff_t start = lstart >> HPAGE_SHIFT;
 192         struct pagevec pvec;
 193         pgoff_t next;
 194         int i;
 195
 196         pagevec_init(&pvec, 0);
 197         next = start;
 198         while (1) {
 199                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 200                         if (next == start)
 201                                 break;
 202                         next = start;
 203                         continue;
 204                 }
 205
 206                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 207                         struct page *page = pvec.pages[i];
 208
 209                         lock_page(page);
 210                         if (page->index > next)
 211                                 next = page->index;
 212                         ++next;
 213                         truncate_huge_page(page);
 214                         unlock_page(page);
 215                         hugetlb_put_quota(mapping);
 216                 }
 217                 huge_pagevec_release(&pvec);
 218         }
 219         BUG_ON(!lstart && mapping->nrpages);
 220 }
 221
 222 static void hugetlbfs_delete_inode(struct inode *inode)
 223 {
 224         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(inode->i_sb);
 225
 226         hlist_del_init(&inode->i_hash);
 227         list_del_init(&inode->i_list);
 228         inode->i_state |= I_FREEING;
 229         inodes_stat.nr_inodes--;
 230         spin_unlock(&inode_lock);
 231
 232         if (inode->i_data.nrpages)
 233                 truncate_hugepages(&inode->i_data, 0);
 234
 235         security_inode_delete(inode);
 236
 237         if (sbinfo->free_inodes >= 0) {
 238                 spin_lock(&sbinfo->stat_lock);
 239                 sbinfo->free_inodes++;
 240                 spin_unlock(&sbinfo->stat_lock);
 241         }
 242
 243         clear_inode(inode);
 244         destroy_inode(inode);
 245 }
 246
 247 static void hugetlbfs_forget_inode(struct inode *inode)
 248 {
 249         struct super_block *super_block = inode->i_sb;
 250         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(super_block);
 251
 252         if (hlist_unhashed(&inode->i_hash))
 253                 goto out_truncate;
 254
 255         if (!(inode->i_state & (I_DIRTY|I_LOCK))) {
 256                 list_del(&inode->i_list);
 257                 list_add(&inode->i_list, &inode_unused);
 258         }
 259         inodes_stat.nr_unused++;
 260         if (!super_block || (super_block->s_flags & MS_ACTIVE)) {
 261                 spin_unlock(&inode_lock);
 262                 return;
 263         }
 264
 265         /* write_inode_now() ? */
 266         inodes_stat.nr_unused--;
 267         hlist_del_init(&inode->i_hash);
 268 out_truncate:
 269         list_del_init(&inode->i_list);
 270         inode->i_state |= I_FREEING;
 271         inodes_stat.nr_inodes--;
 272         spin_unlock(&inode_lock);
 273         if (inode->i_data.nrpages)
 274                 truncate_hugepages(&inode->i_data, 0);
 275
 276         if (sbinfo->free_inodes >= 0) {
 277                 spin_lock(&sbinfo->stat_lock);
 278                 sbinfo->free_inodes++;
 279                 spin_unlock(&sbinfo->stat_lock);
 280         }
 281
 282         clear_inode(inode);
 283         destroy_inode(inode);
 284 }
 285
 286 static void hugetlbfs_drop_inode(struct inode *inode)
 287 {
 288         if (!inode->i_nlink)
 289                 hugetlbfs_delete_inode(inode);
 290         else
 291                 hugetlbfs_forget_inode(inode);
 292 }
 293
 294 /*
 295  * h_pgoff is in HPAGE_SIZE units.
 296  * vma->vm_pgoff is in PAGE_SIZE units.
 297  */
 298 static inline void
 299 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
 300 {
 301         struct vm_area_struct *vma;
 302         struct prio_tree_iter iter;
 303
 304         vma_prio_tree_foreach(vma, &iter, root, h_pgoff, ULONG_MAX) {
 305                 unsigned long h_vm_pgoff;
 306                 unsigned long v_length;
 307                 unsigned long v_offset;
 308
 309                 h_vm_pgoff = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
 310                 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
 311                 /*
 312                  * Is this VMA fully outside the truncation point?
 313                  */
 314                 if (h_vm_pgoff >= h_pgoff)
 315                         v_offset = 0;
 316
 317                 v_length = vma->vm_end - vma->vm_start;
 318
 319                 zap_hugepage_range(vma,
 320                                 vma->vm_start + v_offset,
 321                                 v_length - v_offset);
 322         }
 323 }
 324
 325 /*
 326  * Expanding truncates are not allowed.
 327  */
 328 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 329 {
 330         unsigned long pgoff;
 331         struct address_space *mapping = inode->i_mapping;
 332
 333         if (offset > inode->i_size)
 334                 return -EINVAL;
 335
 336         BUG_ON(offset & ~HPAGE_MASK);
 337         pgoff = offset >> HPAGE_SHIFT;
 338
 339         inode->i_size = offset;
 340         spin_lock(&mapping->i_mmap_lock);
 341         if (!prio_tree_empty(&mapping->i_mmap))
 342                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 343         spin_unlock(&mapping->i_mmap_lock);
 344         truncate_hugepages(mapping, offset);
 345         return 0;
 346 }
 347
 348 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 349 {
 350         struct inode *inode = dentry->d_inode;
 351         int error;
 352         unsigned int ia_valid = attr->ia_valid;
 353
 354         BUG_ON(!inode);
 355
 356         error = inode_change_ok(inode, attr);
 357         if (error)
 358                 goto out;
 359
 360         if (ia_valid & ATTR_SIZE) {
 361                 error = -EINVAL;
 362                 if (!(attr->ia_size & ~HPAGE_MASK))
 363                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 364                 if (error)
 365                         goto out;
 366                 attr->ia_valid &= ~ATTR_SIZE;
 367         }
 368         error = inode_setattr(inode, attr);
 369 out:
 370         return error;
 371 }
 372
 373 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 374                                         gid_t gid, int mode, dev_t dev)
 375 {
 376         struct inode *inode;
 377         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 378
 379         if (sbinfo->free_inodes >= 0) {
 380                 spin_lock(&sbinfo->stat_lock);
 381                 if (!sbinfo->free_inodes) {
 382                         spin_unlock(&sbinfo->stat_lock);
 383                         return NULL;
 384                 }
 385                 sbinfo->free_inodes--;
 386                 spin_unlock(&sbinfo->stat_lock);
 387         }
 388
 389         inode = new_inode(sb);
 390         if (inode) {
 391                 struct hugetlbfs_inode_info *info;
 392                 inode->i_mode = mode;
 393                 inode->i_uid = uid;
 394                 inode->i_gid = gid;
 395                 inode->i_blksize = HPAGE_SIZE;
 396                 inode->i_blocks = 0;
 397                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 398                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 399                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 400                 info = HUGETLBFS_I(inode);
 401                 mpol_shared_policy_init(&info->policy);
 402                 switch (mode & S_IFMT) {
 403                 default:
 404                         init_special_inode(inode, mode, dev);
 405                         break;
 406                 case S_IFREG:
 407                         inode->i_op = &hugetlbfs_inode_operations;
 408                         inode->i_fop = &hugetlbfs_file_operations;
 409                         break;
 410                 case S_IFDIR:
 411                         inode->i_op = &hugetlbfs_dir_inode_operations;
 412                         inode->i_fop = &simple_dir_operations;
 413
 414                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 415                         inode->i_nlink++;
 416                         break;
 417                 case S_IFLNK:
 418                         inode->i_op = &page_symlink_inode_operations;
 419                         break;
 420                 }
 421         }
 422         return inode;
 423 }
 424
 425 /*
 426  * File creation. Allocate an inode, and we're done..
 427  */
 428 static int hugetlbfs_mknod(struct inode *dir,
 429                         struct dentry *dentry, int mode, dev_t dev)
 430 {
 431         struct inode *inode;
 432         int error = -ENOSPC;
 433         gid_t gid;
 434
 435         if (dir->i_mode & S_ISGID) {
 436                 gid = dir->i_gid;
 437                 if (S_ISDIR(mode))
 438                         mode |= S_ISGID;
 439         } else {
 440                 gid = current->fsgid;
 441         }
 442         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 443         if (inode) {
 444                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 445                 d_instantiate(dentry, inode);
 446                 dget(dentry);   /* Extra count - pin the dentry in core */
 447                 error = 0;
 448         }
 449         return error;
 450 }
 451
 452 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 453 {
 454         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 455         if (!retval)
 456                 dir->i_nlink++;
 457         return retval;
 458 }
 459
 460 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 461 {
 462         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 463 }
 464
 465 static int hugetlbfs_symlink(struct inode *dir,
 466                         struct dentry *dentry, const char *symname)
 467 {
 468         struct inode *inode;
 469         int error = -ENOSPC;
 470         gid_t gid;
 471
 472         if (dir->i_mode & S_ISGID)
 473                 gid = dir->i_gid;
 474         else
 475                 gid = current->fsgid;
 476
 477         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 478                                         gid, S_IFLNK|S_IRWXUGO, 0);
 479         if (inode) {
 480                 int l = strlen(symname)+1;
 481                 error = page_symlink(inode, symname, l);
 482                 if (!error) {
 483                         d_instantiate(dentry, inode);
 484                         dget(dentry);
 485                 } else
 486                         iput(inode);
 487         }
 488         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 489
 490         return error;
 491 }
 492
 493 /*
 494  * For direct-IO reads into hugetlb pages
 495  */
 496 int hugetlbfs_set_page_dirty(struct page *page)
 497 {
 498         return 0;
 499 }
 500
 501 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
 502 {
 503         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 504
 505         buf->f_type = HUGETLBFS_MAGIC;
 506         buf->f_bsize = HPAGE_SIZE;
 507         if (sbinfo) {
 508                 spin_lock(&sbinfo->stat_lock);
 509                 buf->f_blocks = sbinfo->max_blocks;
 510                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 511                 buf->f_files = sbinfo->max_inodes;
 512                 buf->f_ffree = sbinfo->free_inodes;
 513                 spin_unlock(&sbinfo->stat_lock);
 514         }
 515         buf->f_namelen = NAME_MAX;
 516         return 0;
 517 }
 518
 519 static void hugetlbfs_put_super(struct super_block *sb)
 520 {
 521         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 522
 523         if (sbi) {
 524                 sb->s_fs_info = NULL;
 525                 kfree(sbi);
 526         }
 527 }
 528
 529 static kmem_cache_t *hugetlbfs_inode_cachep;
 530
 531 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 532 {
 533         struct hugetlbfs_inode_info *p;
 534
 535         p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
 536         if (!p)
 537                 return NULL;
 538         return &p->vfs_inode;
 539 }
 540
 541 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
 542 {
 543         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 544
 545         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 546             SLAB_CTOR_CONSTRUCTOR)
 547                 inode_init_once(&ei->vfs_inode);
 548 }
 549
 550 static void hugetlbfs_destroy_inode(struct inode *inode)
 551 {
 552         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 553         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 554 }
 555
 556 static struct address_space_operations hugetlbfs_aops = {
 557         .readpage       = hugetlbfs_readpage,
 558         .prepare_write  = hugetlbfs_prepare_write,
 559         .commit_write   = hugetlbfs_commit_write,
 560         .set_page_dirty = hugetlbfs_set_page_dirty,
 561 };
 562
 563 struct file_operations hugetlbfs_file_operations = {
 564         .mmap                   = hugetlbfs_file_mmap,
 565         .fsync                  = simple_sync_file,
 566         .get_unmapped_area      = hugetlb_get_unmapped_area,
 567 };
 568
 569 static struct inode_operations hugetlbfs_dir_inode_operations = {
 570         .create         = hugetlbfs_create,
 571         .lookup         = simple_lookup,
 572         .link           = simple_link,
 573         .unlink         = simple_unlink,
 574         .symlink        = hugetlbfs_symlink,
 575         .mkdir          = hugetlbfs_mkdir,
 576         .rmdir          = simple_rmdir,
 577         .mknod          = hugetlbfs_mknod,
 578         .rename         = simple_rename,
 579         .setattr        = hugetlbfs_setattr,
 580 };
 581
 582 static struct inode_operations hugetlbfs_inode_operations = {
 583         .setattr        = hugetlbfs_setattr,
 584 };
 585
 586 static struct super_operations hugetlbfs_ops = {
 587         .alloc_inode    = hugetlbfs_alloc_inode,
 588         .destroy_inode  = hugetlbfs_destroy_inode,
 589         .statfs         = hugetlbfs_statfs,
 590         .drop_inode     = hugetlbfs_drop_inode,
 591         .put_super      = hugetlbfs_put_super,
 592 };
 593
 594 static int
 595 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 596 {
 597         char *opt, *value, *rest;
 598
 599         if (!options)
 600                 return 0;
 601         while ((opt = strsep(&options, ",")) != NULL) {
 602                 if (!*opt)
 603                         continue;
 604
 605                 value = strchr(opt, '=');
 606                 if (!value || !*value)
 607                         return -EINVAL;
 608                 else
 609                         *value++ = '\0';
 610
 611                 if (!strcmp(opt, "uid"))
 612                         pconfig->uid = simple_strtoul(value, &value, 0);
 613                 else if (!strcmp(opt, "gid"))
 614                         pconfig->gid = simple_strtoul(value, &value, 0);
 615                 else if (!strcmp(opt, "mode"))
 616                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
 617                 else if (!strcmp(opt, "size")) {
 618                         unsigned long long size = memparse(value, &rest);
 619                         if (*rest == '%') {
 620                                 size <<= HPAGE_SHIFT;
 621                                 size *= max_huge_pages;
 622                                 do_div(size, 100);
 623                                 rest++;
 624                         }
 625                         size &= HPAGE_MASK;
 626                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 627                         value = rest;
 628                 } else if (!strcmp(opt,"nr_inodes")) {
 629                         pconfig->nr_inodes = memparse(value, &rest);
 630                         value = rest;
 631                 } else
 632                         return -EINVAL;
 633
 634                 if (*value)
 635                         return -EINVAL;
 636         }
 637         return 0;
 638 }
 639
 640 static int
 641 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 642 {
 643         struct inode * inode;
 644         struct dentry * root;
 645         int ret;
 646         struct hugetlbfs_config config;
 647         struct hugetlbfs_sb_info *sbinfo;
 648
 649         config.nr_blocks = -1; /* No limit on size by default */
 650         config.nr_inodes = -1; /* No limit on number of inodes by default */
 651         config.uid = current->fsuid;
 652         config.gid = current->fsgid;
 653         config.mode = 0755;
 654         ret = hugetlbfs_parse_options(data, &config);
 655
 656         if (ret)
 657                 return ret;
 658
 659         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 660         if (!sbinfo)
 661                 return -ENOMEM;
 662         sb->s_fs_info = sbinfo;
 663         spin_lock_init(&sbinfo->stat_lock);
 664         sbinfo->max_blocks = config.nr_blocks;
 665         sbinfo->free_blocks = config.nr_blocks;
 666         sbinfo->max_inodes = config.nr_inodes;
 667         sbinfo->free_inodes = config.nr_inodes;
 668         sb->s_maxbytes = MAX_LFS_FILESIZE;
 669         sb->s_blocksize = HPAGE_SIZE;
 670         sb->s_blocksize_bits = HPAGE_SHIFT;
 671         sb->s_magic = HUGETLBFS_MAGIC;
 672         sb->s_op = &hugetlbfs_ops;
 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 spinlock_t lock = SPIN_LOCK_UNLOCKED;
 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");