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), 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         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         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(inode->i_sb);
 228
 229         hlist_del_init(&inode->i_hash);
 230         list_del_init(&inode->i_list);
 231         list_del_init(&inode->i_sb_list);
 232         inode->i_state |= I_FREEING;
 233         inodes_stat.nr_inodes--;
 234         spin_unlock(&inode_lock);
 235
 236         if (inode->i_data.nrpages)
 237                 truncate_hugepages(&inode->i_data, 0);
 238
 239         security_inode_delete(inode);
 240
 241         if (sbinfo->free_inodes >= 0) {
 242                 spin_lock(&sbinfo->stat_lock);
 243                 sbinfo->free_inodes++;
 244                 spin_unlock(&sbinfo->stat_lock);
 245         }
 246
 247         clear_inode(inode);
 248         destroy_inode(inode);
 249 }
 250
 251 static void hugetlbfs_forget_inode(struct inode *inode)
 252 {
 253         struct super_block *super_block = inode->i_sb;
 254         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(super_block);
 255
 256         if (hlist_unhashed(&inode->i_hash))
 257                 goto out_truncate;
 258
 259         if (!(inode->i_state & (I_DIRTY|I_LOCK))) {
 260                 list_del(&inode->i_list);
 261                 list_add(&inode->i_list, &inode_unused);
 262         }
 263         inodes_stat.nr_unused++;
 264         if (!super_block || (super_block->s_flags & MS_ACTIVE)) {
 265                 spin_unlock(&inode_lock);
 266                 return;
 267         }
 268
 269         /* write_inode_now() ? */
 270         inodes_stat.nr_unused--;
 271         hlist_del_init(&inode->i_hash);
 272 out_truncate:
 273         list_del_init(&inode->i_list);
 274         list_del_init(&inode->i_sb_list);
 275         inode->i_state |= I_FREEING;
 276         inodes_stat.nr_inodes--;
 277         spin_unlock(&inode_lock);
 278         if (inode->i_data.nrpages)
 279                 truncate_hugepages(&inode->i_data, 0);
 280
 281         if (sbinfo->free_inodes >= 0) {
 282                 spin_lock(&sbinfo->stat_lock);
 283                 sbinfo->free_inodes++;
 284                 spin_unlock(&sbinfo->stat_lock);
 285         }
 286
 287         clear_inode(inode);
 288         destroy_inode(inode);
 289 }
 290
 291 static void hugetlbfs_drop_inode(struct inode *inode)
 292 {
 293         if (!inode->i_nlink)
 294                 hugetlbfs_delete_inode(inode);
 295         else
 296                 hugetlbfs_forget_inode(inode);
 297 }
 298
 299 /*
 300  * h_pgoff is in HPAGE_SIZE units.
 301  * vma->vm_pgoff is in PAGE_SIZE units.
 302  */
 303 static inline void
 304 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
 305 {
 306         struct vm_area_struct *vma;
 307         struct prio_tree_iter iter;
 308
 309         vma_prio_tree_foreach(vma, &iter, root, h_pgoff, ULONG_MAX) {
 310                 unsigned long h_vm_pgoff;
 311                 unsigned long v_length;
 312                 unsigned long v_offset;
 313
 314                 h_vm_pgoff = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
 315                 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
 316                 /*
 317                  * Is this VMA fully outside the truncation point?
 318                  */
 319                 if (h_vm_pgoff >= h_pgoff)
 320                         v_offset = 0;
 321
 322                 v_length = vma->vm_end - vma->vm_start;
 323
 324                 zap_hugepage_range(vma,
 325                                 vma->vm_start + v_offset,
 326                                 v_length - v_offset);
 327         }
 328 }
 329
 330 /*
 331  * Expanding truncates are not allowed.
 332  */
 333 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 334 {
 335         unsigned long pgoff;
 336         struct address_space *mapping = inode->i_mapping;
 337
 338         if (offset > inode->i_size)
 339                 return -EINVAL;
 340
 341         BUG_ON(offset & ~HPAGE_MASK);
 342         pgoff = offset >> HPAGE_SHIFT;
 343
 344         inode->i_size = offset;
 345         spin_lock(&mapping->i_mmap_lock);
 346         if (!prio_tree_empty(&mapping->i_mmap))
 347                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 348         spin_unlock(&mapping->i_mmap_lock);
 349         truncate_hugepages(mapping, offset);
 350         return 0;
 351 }
 352
 353 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 354 {
 355         struct inode *inode = dentry->d_inode;
 356         int error;
 357         unsigned int ia_valid = attr->ia_valid;
 358
 359         BUG_ON(!inode);
 360
 361         error = inode_change_ok(inode, attr);
 362         if (error)
 363                 goto out;
 364
 365         if (ia_valid & ATTR_SIZE) {
 366                 error = -EINVAL;
 367                 if (!(attr->ia_size & ~HPAGE_MASK))
 368                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 369                 if (error)
 370                         goto out;
 371                 attr->ia_valid &= ~ATTR_SIZE;
 372         }
 373         error = inode_setattr(inode, attr);
 374 out:
 375         return error;
 376 }
 377
 378 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 379                                         gid_t gid, int mode, dev_t dev)
 380 {
 381         struct inode *inode;
 382         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 383
 384         if (sbinfo->free_inodes >= 0) {
 385                 spin_lock(&sbinfo->stat_lock);
 386                 if (!sbinfo->free_inodes) {
 387                         spin_unlock(&sbinfo->stat_lock);
 388                         return NULL;
 389                 }
 390                 sbinfo->free_inodes--;
 391                 spin_unlock(&sbinfo->stat_lock);
 392         }
 393
 394         inode = new_inode(sb);
 395         if (inode) {
 396                 struct hugetlbfs_inode_info *info;
 397                 inode->i_mode = mode;
 398                 inode->i_uid = uid;
 399                 inode->i_gid = gid;
 400                 inode->i_blksize = HPAGE_SIZE;
 401                 inode->i_blocks = 0;
 402                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 403                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 404                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 405                 info = HUGETLBFS_I(inode);
 406                 mpol_shared_policy_init(&info->policy);
 407                 switch (mode & S_IFMT) {
 408                 default:
 409                         init_special_inode(inode, mode, dev);
 410                         break;
 411                 case S_IFREG:
 412                         inode->i_op = &hugetlbfs_inode_operations;
 413                         inode->i_fop = &hugetlbfs_file_operations;
 414                         break;
 415                 case S_IFDIR:
 416                         inode->i_op = &hugetlbfs_dir_inode_operations;
 417                         inode->i_fop = &simple_dir_operations;
 418
 419                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 420                         inode->i_nlink++;
 421                         break;
 422                 case S_IFLNK:
 423                         inode->i_op = &page_symlink_inode_operations;
 424                         break;
 425                 }
 426         }
 427         return inode;
 428 }
 429
 430 /*
 431  * File creation. Allocate an inode, and we're done..
 432  */
 433 static int hugetlbfs_mknod(struct inode *dir,
 434                         struct dentry *dentry, int mode, dev_t dev)
 435 {
 436         struct inode *inode;
 437         int error = -ENOSPC;
 438         gid_t gid;
 439
 440         if (dir->i_mode & S_ISGID) {
 441                 gid = dir->i_gid;
 442                 if (S_ISDIR(mode))
 443                         mode |= S_ISGID;
 444         } else {
 445                 gid = current->fsgid;
 446         }
 447         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 448         if (inode) {
 449                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 450                 d_instantiate(dentry, inode);
 451                 dget(dentry);   /* Extra count - pin the dentry in core */
 452                 error = 0;
 453         }
 454         return error;
 455 }
 456
 457 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 458 {
 459         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 460         if (!retval)
 461                 dir->i_nlink++;
 462         return retval;
 463 }
 464
 465 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 466 {
 467         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 468 }
 469
 470 static int hugetlbfs_symlink(struct inode *dir,
 471                         struct dentry *dentry, const char *symname)
 472 {
 473         struct inode *inode;
 474         int error = -ENOSPC;
 475         gid_t gid;
 476
 477         if (dir->i_mode & S_ISGID)
 478                 gid = dir->i_gid;
 479         else
 480                 gid = current->fsgid;
 481
 482         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 483                                         gid, S_IFLNK|S_IRWXUGO, 0);
 484         if (inode) {
 485                 int l = strlen(symname)+1;
 486                 error = page_symlink(inode, symname, l);
 487                 if (!error) {
 488                         d_instantiate(dentry, inode);
 489                         dget(dentry);
 490                 } else
 491                         iput(inode);
 492         }
 493         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 494
 495         return error;
 496 }
 497
 498 /*
 499  * For direct-IO reads into hugetlb pages
 500  */
 501 static int hugetlbfs_set_page_dirty(struct page *page)
 502 {
 503         return 0;
 504 }
 505
 506 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
 507 {
 508         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 509
 510         buf->f_type = HUGETLBFS_MAGIC;
 511         buf->f_bsize = HPAGE_SIZE;
 512         if (sbinfo) {
 513                 spin_lock(&sbinfo->stat_lock);
 514                 buf->f_blocks = sbinfo->max_blocks;
 515                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 516                 buf->f_files = sbinfo->max_inodes;
 517                 buf->f_ffree = sbinfo->free_inodes;
 518                 spin_unlock(&sbinfo->stat_lock);
 519         }
 520         buf->f_namelen = NAME_MAX;
 521         return 0;
 522 }
 523
 524 static void hugetlbfs_put_super(struct super_block *sb)
 525 {
 526         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 527
 528         if (sbi) {
 529                 sb->s_fs_info = NULL;
 530                 kfree(sbi);
 531         }
 532 }
 533
 534 static kmem_cache_t *hugetlbfs_inode_cachep;
 535
 536 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 537 {
 538         struct hugetlbfs_inode_info *p;
 539
 540         p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
 541         if (!p)
 542                 return NULL;
 543         return &p->vfs_inode;
 544 }
 545
 546 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
 547 {
 548         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 549
 550         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 551             SLAB_CTOR_CONSTRUCTOR)
 552                 inode_init_once(&ei->vfs_inode);
 553 }
 554
 555 static void hugetlbfs_destroy_inode(struct inode *inode)
 556 {
 557         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 558         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 559 }
 560
 561 static struct address_space_operations hugetlbfs_aops = {
 562         .readpage       = hugetlbfs_readpage,
 563         .prepare_write  = hugetlbfs_prepare_write,
 564         .commit_write   = hugetlbfs_commit_write,
 565         .set_page_dirty = hugetlbfs_set_page_dirty,
 566 };
 567
 568 struct file_operations hugetlbfs_file_operations = {
 569         .mmap                   = hugetlbfs_file_mmap,
 570         .fsync                  = simple_sync_file,
 571         .get_unmapped_area      = hugetlb_get_unmapped_area,
 572 };
 573
 574 static struct inode_operations hugetlbfs_dir_inode_operations = {
 575         .create         = hugetlbfs_create,
 576         .lookup         = simple_lookup,
 577         .link           = simple_link,
 578         .unlink         = simple_unlink,
 579         .symlink        = hugetlbfs_symlink,
 580         .mkdir          = hugetlbfs_mkdir,
 581         .rmdir          = simple_rmdir,
 582         .mknod          = hugetlbfs_mknod,
 583         .rename         = simple_rename,
 584         .setattr        = hugetlbfs_setattr,
 585 };
 586
 587 static struct inode_operations hugetlbfs_inode_operations = {
 588         .setattr        = hugetlbfs_setattr,
 589 };
 590
 591 static struct super_operations hugetlbfs_ops = {
 592         .alloc_inode    = hugetlbfs_alloc_inode,
 593         .destroy_inode  = hugetlbfs_destroy_inode,
 594         .statfs         = hugetlbfs_statfs,
 595         .drop_inode     = hugetlbfs_drop_inode,
 596         .put_super      = hugetlbfs_put_super,
 597 };
 598
 599 static int
 600 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 601 {
 602         char *opt, *value, *rest;
 603
 604         if (!options)
 605                 return 0;
 606         while ((opt = strsep(&options, ",")) != NULL) {
 607                 if (!*opt)
 608                         continue;
 609
 610                 value = strchr(opt, '=');
 611                 if (!value || !*value)
 612                         return -EINVAL;
 613                 else
 614                         *value++ = '\0';
 615
 616                 if (!strcmp(opt, "uid"))
 617                         pconfig->uid = simple_strtoul(value, &value, 0);
 618                 else if (!strcmp(opt, "gid"))
 619                         pconfig->gid = simple_strtoul(value, &value, 0);
 620                 else if (!strcmp(opt, "mode"))
 621                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
 622                 else if (!strcmp(opt, "size")) {
 623                         unsigned long long size = memparse(value, &rest);
 624                         if (*rest == '%') {
 625                                 size <<= HPAGE_SHIFT;
 626                                 size *= max_huge_pages;
 627                                 do_div(size, 100);
 628                                 rest++;
 629                         }
 630                         size &= HPAGE_MASK;
 631                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 632                         value = rest;
 633                 } else if (!strcmp(opt,"nr_inodes")) {
 634                         pconfig->nr_inodes = memparse(value, &rest);
 635                         value = rest;
 636                 } else
 637                         return -EINVAL;
 638
 639                 if (*value)
 640                         return -EINVAL;
 641         }
 642         return 0;
 643 }
 644
 645 static int
 646 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 647 {
 648         struct inode * inode;
 649         struct dentry * root;
 650         int ret;
 651         struct hugetlbfs_config config;
 652         struct hugetlbfs_sb_info *sbinfo;
 653
 654         config.nr_blocks = -1; /* No limit on size by default */
 655         config.nr_inodes = -1; /* No limit on number of inodes by default */
 656         config.uid = current->fsuid;
 657         config.gid = current->fsgid;
 658         config.mode = 0755;
 659         ret = hugetlbfs_parse_options(data, &config);
 660
 661         if (ret)
 662                 return ret;
 663
 664         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 665         if (!sbinfo)
 666                 return -ENOMEM;
 667         sb->s_fs_info = sbinfo;
 668         spin_lock_init(&sbinfo->stat_lock);
 669         sbinfo->max_blocks = config.nr_blocks;
 670         sbinfo->free_blocks = config.nr_blocks;
 671         sbinfo->max_inodes = config.nr_inodes;
 672         sbinfo->free_inodes = config.nr_inodes;
 673         sb->s_maxbytes = MAX_LFS_FILESIZE;
 674         sb->s_blocksize = HPAGE_SIZE;
 675         sb->s_blocksize_bits = HPAGE_SHIFT;
 676         sb->s_magic = HUGETLBFS_MAGIC;
 677         sb->s_op = &hugetlbfs_ops;
 678         sb->s_time_gran = 1;
 679         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 680                                         S_IFDIR | config.mode, 0);
 681         if (!inode)
 682                 goto out_free;
 683
 684         root = d_alloc_root(inode);
 685         if (!root) {
 686                 iput(inode);
 687                 goto out_free;
 688         }
 689         sb->s_root = root;
 690         return 0;
 691 out_free:
 692         kfree(sbinfo);
 693         return -ENOMEM;
 694 }
 695
 696 int hugetlb_get_quota(struct address_space *mapping)
 697 {
 698         int ret = 0;
 699         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 700
 701         if (sbinfo->free_blocks > -1) {
 702                 spin_lock(&sbinfo->stat_lock);
 703                 if (sbinfo->free_blocks > 0)
 704                         sbinfo->free_blocks--;
 705                 else
 706                         ret = -ENOMEM;
 707                 spin_unlock(&sbinfo->stat_lock);
 708         }
 709
 710         return ret;
 711 }
 712
 713 void hugetlb_put_quota(struct address_space *mapping)
 714 {
 715         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 716
 717         if (sbinfo->free_blocks > -1) {
 718                 spin_lock(&sbinfo->stat_lock);
 719                 sbinfo->free_blocks++;
 720                 spin_unlock(&sbinfo->stat_lock);
 721         }
 722 }
 723
 724 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
 725         int flags, const char *dev_name, void *data)
 726 {
 727         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 728 }
 729
 730 static struct file_system_type hugetlbfs_fs_type = {
 731         .name           = "hugetlbfs",
 732         .get_sb         = hugetlbfs_get_sb,
 733         .kill_sb        = kill_litter_super,
 734 };
 735
 736 static struct vfsmount *hugetlbfs_vfsmount;
 737
 738 /*
 739  * Return the next identifier for a shm file
 740  */
 741 static unsigned long hugetlbfs_counter(void)
 742 {
 743         static DEFINE_SPINLOCK(lock);
 744         static unsigned long counter;
 745         unsigned long ret;
 746
 747         spin_lock(&lock);
 748         ret = ++counter;
 749         spin_unlock(&lock);
 750         return ret;
 751 }
 752
 753 static int can_do_hugetlb_shm(void)
 754 {
 755         return likely(capable(CAP_IPC_LOCK) ||
 756                         in_group_p(sysctl_hugetlb_shm_group) ||
 757                         can_do_mlock());
 758 }
 759
 760 struct file *hugetlb_zero_setup(size_t size)
 761 {
 762         int error = -ENOMEM;
 763         struct file *file;
 764         struct inode *inode;
 765         struct dentry *dentry, *root;
 766         struct qstr quick_string;
 767         char buf[16];
 768
 769         if (!can_do_hugetlb_shm())
 770                 return ERR_PTR(-EPERM);
 771
 772         if (!is_hugepage_mem_enough(size))
 773                 return ERR_PTR(-ENOMEM);
 774
 775         if (!user_shm_lock(size, current->user))
 776                 return ERR_PTR(-ENOMEM);
 777
 778         root = hugetlbfs_vfsmount->mnt_root;
 779         snprintf(buf, 16, "%lu", hugetlbfs_counter());
 780         quick_string.name = buf;
 781         quick_string.len = strlen(quick_string.name);
 782         quick_string.hash = 0;
 783         dentry = d_alloc(root, &quick_string);
 784         if (!dentry)
 785                 goto out_shm_unlock;
 786
 787         error = -ENFILE;
 788         file = get_empty_filp();
 789         if (!file)
 790                 goto out_dentry;
 791
 792         error = -ENOSPC;
 793         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
 794                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
 795         if (!inode)
 796                 goto out_file;
 797
 798         d_instantiate(dentry, inode);
 799         inode->i_size = size;
 800         inode->i_nlink = 0;
 801         file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
 802         file->f_dentry = dentry;
 803         file->f_mapping = inode->i_mapping;
 804         file->f_op = &hugetlbfs_file_operations;
 805         file->f_mode = FMODE_WRITE | FMODE_READ;
 806         return file;
 807
 808 out_file:
 809         put_filp(file);
 810 out_dentry:
 811         dput(dentry);
 812 out_shm_unlock:
 813         user_shm_unlock(size, current->user);
 814         return ERR_PTR(error);
 815 }
 816
 817 static int __init init_hugetlbfs_fs(void)
 818 {
 819         int error;
 820         struct vfsmount *vfsmount;
 821
 822         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
 823                                         sizeof(struct hugetlbfs_inode_info),
 824                                         0, 0, init_once, NULL);
 825         if (hugetlbfs_inode_cachep == NULL)
 826                 return -ENOMEM;
 827
 828         error = register_filesystem(&hugetlbfs_fs_type);
 829         if (error)
 830                 goto out;
 831
 832         vfsmount = kern_mount(&hugetlbfs_fs_type);
 833
 834         if (!IS_ERR(vfsmount)) {
 835                 hugetlbfs_vfsmount = vfsmount;
 836                 return 0;
 837         }
 838
 839         error = PTR_ERR(vfsmount);
 840
 841  out:
 842         if (error)
 843                 kmem_cache_destroy(hugetlbfs_inode_cachep);
 844         return error;
 845 }
 846
 847 static void __exit exit_hugetlbfs_fs(void)
 848 {
 849         kmem_cache_destroy(hugetlbfs_inode_cachep);
 850         unregister_filesystem(&hugetlbfs_fs_type);
 851 }
 852
 853 module_init(init_hugetlbfs_fs)
 854 module_exit(exit_hugetlbfs_fs)
 855
 856 MODULE_LICENSE("GPL");