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/capability.h>
  22 #include <linux/backing-dev.h>
  23 #include <linux/hugetlb.h>
  24 #include <linux/pagevec.h>
  25 #include <linux/quotaops.h>
  26 #include <linux/slab.h>
  27 #include <linux/dnotify.h>
  28 #include <linux/statfs.h>
  29 #include <linux/security.h>
  30
  31 #include <asm/uaccess.h>
  32
  33 /* some random number */
  34 #define HUGETLBFS_MAGIC 0x958458f6
  35
  36 static struct super_operations hugetlbfs_ops;
  37 static struct address_space_operations hugetlbfs_aops;
  38 struct file_operations hugetlbfs_file_operations;
  39 static struct inode_operations hugetlbfs_dir_inode_operations;
  40 static struct inode_operations hugetlbfs_inode_operations;
  41
  42 static struct backing_dev_info hugetlbfs_backing_dev_info = {
  43         .ra_pages       = 0,    /* No readahead */
  44         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
  45 };
  46
  47 int sysctl_hugetlb_shm_group;
  48
  49 static void huge_pagevec_release(struct pagevec *pvec)
  50 {
  51         int i;
  52
  53         for (i = 0; i < pagevec_count(pvec); ++i)
  54                 put_page(pvec->pages[i]);
  55
  56         pagevec_reinit(pvec);
  57 }
  58
  59 /*
  60  * huge_pages_needed tries to determine the number of new huge pages that
  61  * will be required to fully populate this VMA.  This will be equal to
  62  * the size of the VMA in huge pages minus the number of huge pages
  63  * (covered by this VMA) that are found in the page cache.
  64  *
  65  * Result is in bytes to be compatible with is_hugepage_mem_enough()
  66  */
  67 static unsigned long
  68 huge_pages_needed(struct address_space *mapping, struct vm_area_struct *vma)
  69 {
  70         int i;
  71         struct pagevec pvec;
  72         unsigned long start = vma->vm_start;
  73         unsigned long end = vma->vm_end;
  74         unsigned long hugepages = (end - start) >> HPAGE_SHIFT;
  75         pgoff_t next = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
  76         pgoff_t endpg = next + hugepages;
  77
  78         pagevec_init(&pvec, 0);
  79         while (next < endpg) {
  80                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE))
  81                         break;
  82                 for (i = 0; i < pagevec_count(&pvec); i++) {
  83                         struct page *page = pvec.pages[i];
  84                         if (page->index > next)
  85                                 next = page->index;
  86                         if (page->index >= endpg)
  87                                 break;
  88                         next++;
  89                         hugepages--;
  90                 }
  91                 huge_pagevec_release(&pvec);
  92         }
  93         return hugepages << HPAGE_SHIFT;
  94 }
  95
  96 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  97 {
  98         struct inode *inode = file->f_dentry->d_inode;
  99         struct address_space *mapping = inode->i_mapping;
 100         unsigned long bytes;
 101         loff_t len, vma_len;
 102         int ret;
 103
 104         if (vma->vm_pgoff & (HPAGE_SIZE / PAGE_SIZE - 1))
 105                 return -EINVAL;
 106
 107         if (vma->vm_start & ~HPAGE_MASK)
 108                 return -EINVAL;
 109
 110         if (vma->vm_end & ~HPAGE_MASK)
 111                 return -EINVAL;
 112
 113         if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
 114                 return -EINVAL;
 115
 116         bytes = huge_pages_needed(mapping, vma);
 117         if (!is_hugepage_mem_enough(bytes))
 118                 return -ENOMEM;
 119
 120         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 121
 122         mutex_lock(&inode->i_mutex);
 123         file_accessed(file);
 124         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
 125         vma->vm_ops = &hugetlb_vm_ops;
 126
 127         ret = -ENOMEM;
 128         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 129         if (!(vma->vm_flags & VM_WRITE) && len > inode->i_size)
 130                 goto out;
 131
 132         ret = 0;
 133         hugetlb_prefault_arch_hook(vma->vm_mm);
 134         if (inode->i_size < len)
 135                 inode->i_size = len;
 136 out:
 137         mutex_unlock(&inode->i_mutex);
 138
 139         return ret;
 140 }
 141
 142 /*
 143  * Called under down_write(mmap_sem).
 144  */
 145
 146 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 147 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 148                 unsigned long len, unsigned long pgoff, unsigned long flags);
 149 #else
 150 static unsigned long
 151 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 152                 unsigned long len, unsigned long pgoff, unsigned long flags)
 153 {
 154         struct mm_struct *mm = current->mm;
 155         struct vm_area_struct *vma;
 156         unsigned long start_addr;
 157
 158         if (len & ~HPAGE_MASK)
 159                 return -EINVAL;
 160         if (len > TASK_SIZE)
 161                 return -ENOMEM;
 162
 163         if (addr) {
 164                 addr = ALIGN(addr, HPAGE_SIZE);
 165                 vma = find_vma(mm, addr);
 166                 if (TASK_SIZE - len >= addr &&
 167                     (!vma || addr + len <= vma->vm_start))
 168                         return addr;
 169         }
 170
 171         start_addr = mm->free_area_cache;
 172
 173         if (len <= mm->cached_hole_size)
 174                 start_addr = TASK_UNMAPPED_BASE;
 175
 176 full_search:
 177         addr = ALIGN(start_addr, HPAGE_SIZE);
 178
 179         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 180                 /* At this point:  (!vma || addr < vma->vm_end). */
 181                 if (TASK_SIZE - len < addr) {
 182                         /*
 183                          * Start a new search - just in case we missed
 184                          * some holes.
 185                          */
 186                         if (start_addr != TASK_UNMAPPED_BASE) {
 187                                 start_addr = TASK_UNMAPPED_BASE;
 188                                 goto full_search;
 189                         }
 190                         return -ENOMEM;
 191                 }
 192
 193                 if (!vma || addr + len <= vma->vm_start)
 194                         return addr;
 195                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
 196         }
 197 }
 198 #endif
 199
 200 /*
 201  * Read a page. Again trivial. If it didn't already exist
 202  * in the page cache, it is zero-filled.
 203  */
 204 static int hugetlbfs_readpage(struct file *file, struct page * page)
 205 {
 206         unlock_page(page);
 207         return -EINVAL;
 208 }
 209
 210 static int hugetlbfs_prepare_write(struct file *file,
 211                         struct page *page, unsigned offset, unsigned to)
 212 {
 213         return -EINVAL;
 214 }
 215
 216 static int hugetlbfs_commit_write(struct file *file,
 217                         struct page *page, unsigned offset, unsigned to)
 218 {
 219         return -EINVAL;
 220 }
 221
 222 static void truncate_huge_page(struct page *page)
 223 {
 224         clear_page_dirty(page);
 225         ClearPageUptodate(page);
 226         remove_from_page_cache(page);
 227         put_page(page);
 228 }
 229
 230 static void truncate_hugepages(struct address_space *mapping, loff_t lstart)
 231 {
 232         const pgoff_t start = lstart >> HPAGE_SHIFT;
 233         struct pagevec pvec;
 234         pgoff_t next;
 235         int i;
 236
 237         pagevec_init(&pvec, 0);
 238         next = start;
 239         while (1) {
 240                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 241                         if (next == start)
 242                                 break;
 243                         next = start;
 244                         continue;
 245                 }
 246
 247                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 248                         struct page *page = pvec.pages[i];
 249
 250                         lock_page(page);
 251                         if (page->index > next)
 252                                 next = page->index;
 253                         ++next;
 254                         truncate_huge_page(page);
 255                         unlock_page(page);
 256                         hugetlb_put_quota(mapping);
 257                 }
 258                 huge_pagevec_release(&pvec);
 259         }
 260         BUG_ON(!lstart && mapping->nrpages);
 261 }
 262
 263 static void hugetlbfs_delete_inode(struct inode *inode)
 264 {
 265         if (inode->i_data.nrpages)
 266                 truncate_hugepages(&inode->i_data, 0);
 267         clear_inode(inode);
 268 }
 269
 270 static void hugetlbfs_forget_inode(struct inode *inode)
 271 {
 272         struct super_block *sb = inode->i_sb;
 273
 274         if (!hlist_unhashed(&inode->i_hash)) {
 275                 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
 276                         list_move(&inode->i_list, &inode_unused);
 277                 inodes_stat.nr_unused++;
 278                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
 279                         spin_unlock(&inode_lock);
 280                         return;
 281                 }
 282                 inode->i_state |= I_WILL_FREE;
 283                 spin_unlock(&inode_lock);
 284                 /*
 285                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
 286                  * in our backing_dev_info.
 287                  */
 288                 write_inode_now(inode, 1);
 289                 spin_lock(&inode_lock);
 290                 inode->i_state &= ~I_WILL_FREE;
 291                 inodes_stat.nr_unused--;
 292                 hlist_del_init(&inode->i_hash);
 293         }
 294         list_del_init(&inode->i_list);
 295         list_del_init(&inode->i_sb_list);
 296         inode->i_state |= I_FREEING;
 297         inodes_stat.nr_inodes--;
 298         spin_unlock(&inode_lock);
 299         if (inode->i_data.nrpages)
 300                 truncate_hugepages(&inode->i_data, 0);
 301         clear_inode(inode);
 302         destroy_inode(inode);
 303 }
 304
 305 static void hugetlbfs_drop_inode(struct inode *inode)
 306 {
 307         if (!inode->i_nlink)
 308                 generic_delete_inode(inode);
 309         else
 310                 hugetlbfs_forget_inode(inode);
 311 }
 312
 313 static inline void
 314 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
 315 {
 316         struct vm_area_struct *vma;
 317         struct prio_tree_iter iter;
 318
 319         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
 320                 unsigned long v_offset;
 321
 322                 /*
 323                  * Can the expression below overflow on 32-bit arches?
 324                  * No, because the prio_tree returns us only those vmas
 325                  * which overlap the truncated area starting at pgoff,
 326                  * and no vma on a 32-bit arch can span beyond the 4GB.
 327                  */
 328                 if (vma->vm_pgoff < pgoff)
 329                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 330                 else
 331                         v_offset = 0;
 332
 333                 unmap_hugepage_range(vma,
 334                                 vma->vm_start + v_offset, vma->vm_end);
 335         }
 336 }
 337
 338 /*
 339  * Expanding truncates are not allowed.
 340  */
 341 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 342 {
 343         pgoff_t pgoff;
 344         struct address_space *mapping = inode->i_mapping;
 345
 346         if (offset > inode->i_size)
 347                 return -EINVAL;
 348
 349         BUG_ON(offset & ~HPAGE_MASK);
 350         pgoff = offset >> PAGE_SHIFT;
 351
 352         inode->i_size = offset;
 353         spin_lock(&mapping->i_mmap_lock);
 354         if (!prio_tree_empty(&mapping->i_mmap))
 355                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 356         spin_unlock(&mapping->i_mmap_lock);
 357         truncate_hugepages(mapping, offset);
 358         return 0;
 359 }
 360
 361 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 362 {
 363         struct inode *inode = dentry->d_inode;
 364         int error;
 365         unsigned int ia_valid = attr->ia_valid;
 366
 367         BUG_ON(!inode);
 368
 369         error = inode_change_ok(inode, attr);
 370         if (error)
 371                 goto out;
 372
 373         if (ia_valid & ATTR_SIZE) {
 374                 error = -EINVAL;
 375                 if (!(attr->ia_size & ~HPAGE_MASK))
 376                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 377                 if (error)
 378                         goto out;
 379                 attr->ia_valid &= ~ATTR_SIZE;
 380         }
 381         error = inode_setattr(inode, attr);
 382 out:
 383         return error;
 384 }
 385
 386 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 387                                         gid_t gid, int mode, dev_t dev)
 388 {
 389         struct inode *inode;
 390
 391         inode = new_inode(sb);
 392         if (inode) {
 393                 struct hugetlbfs_inode_info *info;
 394                 inode->i_mode = mode;
 395                 inode->i_uid = uid;
 396                 inode->i_gid = gid;
 397                 inode->i_blksize = HPAGE_SIZE;
 398                 inode->i_blocks = 0;
 399                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 400                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 401                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 402                 info = HUGETLBFS_I(inode);
 403                 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
 404                 switch (mode & S_IFMT) {
 405                 default:
 406                         init_special_inode(inode, mode, dev);
 407                         break;
 408                 case S_IFREG:
 409                         inode->i_op = &hugetlbfs_inode_operations;
 410                         inode->i_fop = &hugetlbfs_file_operations;
 411                         break;
 412                 case S_IFDIR:
 413                         inode->i_op = &hugetlbfs_dir_inode_operations;
 414                         inode->i_fop = &simple_dir_operations;
 415
 416                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 417                         inode->i_nlink++;
 418                         break;
 419                 case S_IFLNK:
 420                         inode->i_op = &page_symlink_inode_operations;
 421                         break;
 422                 }
 423         }
 424         return inode;
 425 }
 426
 427 /*
 428  * File creation. Allocate an inode, and we're done..
 429  */
 430 static int hugetlbfs_mknod(struct inode *dir,
 431                         struct dentry *dentry, int mode, dev_t dev)
 432 {
 433         struct inode *inode;
 434         int error = -ENOSPC;
 435         gid_t gid;
 436
 437         if (dir->i_mode & S_ISGID) {
 438                 gid = dir->i_gid;
 439                 if (S_ISDIR(mode))
 440                         mode |= S_ISGID;
 441         } else {
 442                 gid = current->fsgid;
 443         }
 444         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 445         if (inode) {
 446                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 447                 d_instantiate(dentry, inode);
 448                 dget(dentry);   /* Extra count - pin the dentry in core */
 449                 error = 0;
 450         }
 451         return error;
 452 }
 453
 454 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 455 {
 456         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 457         if (!retval)
 458                 dir->i_nlink++;
 459         return retval;
 460 }
 461
 462 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 463 {
 464         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 465 }
 466
 467 static int hugetlbfs_symlink(struct inode *dir,
 468                         struct dentry *dentry, const char *symname)
 469 {
 470         struct inode *inode;
 471         int error = -ENOSPC;
 472         gid_t gid;
 473
 474         if (dir->i_mode & S_ISGID)
 475                 gid = dir->i_gid;
 476         else
 477                 gid = current->fsgid;
 478
 479         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 480                                         gid, S_IFLNK|S_IRWXUGO, 0);
 481         if (inode) {
 482                 int l = strlen(symname)+1;
 483                 error = page_symlink(inode, symname, l);
 484                 if (!error) {
 485                         d_instantiate(dentry, inode);
 486                         dget(dentry);
 487                 } else
 488                         iput(inode);
 489         }
 490         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 491
 492         return error;
 493 }
 494
 495 /*
 496  * For direct-IO reads into hugetlb pages
 497  */
 498 static int hugetlbfs_set_page_dirty(struct page *page)
 499 {
 500         return 0;
 501 }
 502
 503 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
 504 {
 505         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 506
 507         buf->f_type = HUGETLBFS_MAGIC;
 508         buf->f_bsize = HPAGE_SIZE;
 509         if (sbinfo) {
 510                 spin_lock(&sbinfo->stat_lock);
 511                 /* If no limits set, just report 0 for max/free/used
 512                  * blocks, like simple_statfs() */
 513                 if (sbinfo->max_blocks >= 0) {
 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                 }
 519                 spin_unlock(&sbinfo->stat_lock);
 520         }
 521         buf->f_namelen = NAME_MAX;
 522         return 0;
 523 }
 524
 525 static void hugetlbfs_put_super(struct super_block *sb)
 526 {
 527         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 528
 529         if (sbi) {
 530                 sb->s_fs_info = NULL;
 531                 kfree(sbi);
 532         }
 533 }
 534
 535 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 536 {
 537         if (sbinfo->free_inodes >= 0) {
 538                 spin_lock(&sbinfo->stat_lock);
 539                 if (unlikely(!sbinfo->free_inodes)) {
 540                         spin_unlock(&sbinfo->stat_lock);
 541                         return 0;
 542                 }
 543                 sbinfo->free_inodes--;
 544                 spin_unlock(&sbinfo->stat_lock);
 545         }
 546
 547         return 1;
 548 }
 549
 550 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 551 {
 552         if (sbinfo->free_inodes >= 0) {
 553                 spin_lock(&sbinfo->stat_lock);
 554                 sbinfo->free_inodes++;
 555                 spin_unlock(&sbinfo->stat_lock);
 556         }
 557 }
 558
 559
 560 static kmem_cache_t *hugetlbfs_inode_cachep;
 561
 562 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 563 {
 564         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 565         struct hugetlbfs_inode_info *p;
 566
 567         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 568                 return NULL;
 569         p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
 570         if (unlikely(!p)) {
 571                 hugetlbfs_inc_free_inodes(sbinfo);
 572                 return NULL;
 573         }
 574         return &p->vfs_inode;
 575 }
 576
 577 static void hugetlbfs_destroy_inode(struct inode *inode)
 578 {
 579         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 580         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 581         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 582 }
 583
 584 static struct address_space_operations hugetlbfs_aops = {
 585         .readpage       = hugetlbfs_readpage,
 586         .prepare_write  = hugetlbfs_prepare_write,
 587         .commit_write   = hugetlbfs_commit_write,
 588         .set_page_dirty = hugetlbfs_set_page_dirty,
 589 };
 590
 591
 592 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
 593 {
 594         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 595
 596         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 597             SLAB_CTOR_CONSTRUCTOR)
 598                 inode_init_once(&ei->vfs_inode);
 599 }
 600
 601 struct file_operations hugetlbfs_file_operations = {
 602         .mmap                   = hugetlbfs_file_mmap,
 603         .fsync                  = simple_sync_file,
 604         .get_unmapped_area      = hugetlb_get_unmapped_area,
 605 };
 606
 607 static struct inode_operations hugetlbfs_dir_inode_operations = {
 608         .create         = hugetlbfs_create,
 609         .lookup         = simple_lookup,
 610         .link           = simple_link,
 611         .unlink         = simple_unlink,
 612         .symlink        = hugetlbfs_symlink,
 613         .mkdir          = hugetlbfs_mkdir,
 614         .rmdir          = simple_rmdir,
 615         .mknod          = hugetlbfs_mknod,
 616         .rename         = simple_rename,
 617         .setattr        = hugetlbfs_setattr,
 618 };
 619
 620 static struct inode_operations hugetlbfs_inode_operations = {
 621         .setattr        = hugetlbfs_setattr,
 622 };
 623
 624 static struct super_operations hugetlbfs_ops = {
 625         .alloc_inode    = hugetlbfs_alloc_inode,
 626         .destroy_inode  = hugetlbfs_destroy_inode,
 627         .statfs         = hugetlbfs_statfs,
 628         .delete_inode   = hugetlbfs_delete_inode,
 629         .drop_inode     = hugetlbfs_drop_inode,
 630         .put_super      = hugetlbfs_put_super,
 631 };
 632
 633 static int
 634 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 635 {
 636         char *opt, *value, *rest;
 637
 638         if (!options)
 639                 return 0;
 640         while ((opt = strsep(&options, ",")) != NULL) {
 641                 if (!*opt)
 642                         continue;
 643
 644                 value = strchr(opt, '=');
 645                 if (!value || !*value)
 646                         return -EINVAL;
 647                 else
 648                         *value++ = '\0';
 649
 650                 if (!strcmp(opt, "uid"))
 651                         pconfig->uid = simple_strtoul(value, &value, 0);
 652                 else if (!strcmp(opt, "gid"))
 653                         pconfig->gid = simple_strtoul(value, &value, 0);
 654                 else if (!strcmp(opt, "mode"))
 655                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
 656                 else if (!strcmp(opt, "size")) {
 657                         unsigned long long size = memparse(value, &rest);
 658                         if (*rest == '%') {
 659                                 size <<= HPAGE_SHIFT;
 660                                 size *= max_huge_pages;
 661                                 do_div(size, 100);
 662                                 rest++;
 663                         }
 664                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 665                         value = rest;
 666                 } else if (!strcmp(opt,"nr_inodes")) {
 667                         pconfig->nr_inodes = memparse(value, &rest);
 668                         value = rest;
 669                 } else
 670                         return -EINVAL;
 671
 672                 if (*value)
 673                         return -EINVAL;
 674         }
 675         return 0;
 676 }
 677
 678 static int
 679 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 680 {
 681         struct inode * inode;
 682         struct dentry * root;
 683         int ret;
 684         struct hugetlbfs_config config;
 685         struct hugetlbfs_sb_info *sbinfo;
 686
 687         config.nr_blocks = -1; /* No limit on size by default */
 688         config.nr_inodes = -1; /* No limit on number of inodes by default */
 689         config.uid = current->fsuid;
 690         config.gid = current->fsgid;
 691         config.mode = 0755;
 692         ret = hugetlbfs_parse_options(data, &config);
 693
 694         if (ret)
 695                 return ret;
 696
 697         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 698         if (!sbinfo)
 699                 return -ENOMEM;
 700         sb->s_fs_info = sbinfo;
 701         spin_lock_init(&sbinfo->stat_lock);
 702         sbinfo->max_blocks = config.nr_blocks;
 703         sbinfo->free_blocks = config.nr_blocks;
 704         sbinfo->max_inodes = config.nr_inodes;
 705         sbinfo->free_inodes = config.nr_inodes;
 706         sb->s_maxbytes = MAX_LFS_FILESIZE;
 707         sb->s_blocksize = HPAGE_SIZE;
 708         sb->s_blocksize_bits = HPAGE_SHIFT;
 709         sb->s_magic = HUGETLBFS_MAGIC;
 710         sb->s_op = &hugetlbfs_ops;
 711         sb->s_time_gran = 1;
 712         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 713                                         S_IFDIR | config.mode, 0);
 714         if (!inode)
 715                 goto out_free;
 716
 717         root = d_alloc_root(inode);
 718         if (!root) {
 719                 iput(inode);
 720                 goto out_free;
 721         }
 722         sb->s_root = root;
 723         return 0;
 724 out_free:
 725         kfree(sbinfo);
 726         return -ENOMEM;
 727 }
 728
 729 int hugetlb_get_quota(struct address_space *mapping)
 730 {
 731         int ret = 0;
 732         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 733
 734         if (sbinfo->free_blocks > -1) {
 735                 spin_lock(&sbinfo->stat_lock);
 736                 if (sbinfo->free_blocks > 0)
 737                         sbinfo->free_blocks--;
 738                 else
 739                         ret = -ENOMEM;
 740                 spin_unlock(&sbinfo->stat_lock);
 741         }
 742
 743         return ret;
 744 }
 745
 746 void hugetlb_put_quota(struct address_space *mapping)
 747 {
 748         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 749
 750         if (sbinfo->free_blocks > -1) {
 751                 spin_lock(&sbinfo->stat_lock);
 752                 sbinfo->free_blocks++;
 753                 spin_unlock(&sbinfo->stat_lock);
 754         }
 755 }
 756
 757 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
 758         int flags, const char *dev_name, void *data)
 759 {
 760         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 761 }
 762
 763 static struct file_system_type hugetlbfs_fs_type = {
 764         .name           = "hugetlbfs",
 765         .get_sb         = hugetlbfs_get_sb,
 766         .kill_sb        = kill_litter_super,
 767 };
 768
 769 static struct vfsmount *hugetlbfs_vfsmount;
 770
 771 /*
 772  * Return the next identifier for a shm file
 773  */
 774 static unsigned long hugetlbfs_counter(void)
 775 {
 776         static DEFINE_SPINLOCK(lock);
 777         static unsigned long counter;
 778         unsigned long ret;
 779
 780         spin_lock(&lock);
 781         ret = ++counter;
 782         spin_unlock(&lock);
 783         return ret;
 784 }
 785
 786 static int can_do_hugetlb_shm(void)
 787 {
 788         return likely(capable(CAP_IPC_LOCK) ||
 789                         in_group_p(sysctl_hugetlb_shm_group) ||
 790                         can_do_mlock());
 791 }
 792
 793 struct file *hugetlb_zero_setup(size_t size)
 794 {
 795         int error = -ENOMEM;
 796         struct file *file;
 797         struct inode *inode;
 798         struct dentry *dentry, *root;
 799         struct qstr quick_string;
 800         char buf[16];
 801
 802         if (!can_do_hugetlb_shm())
 803                 return ERR_PTR(-EPERM);
 804
 805         if (!is_hugepage_mem_enough(size))
 806                 return ERR_PTR(-ENOMEM);
 807
 808         if (!user_shm_lock(size, current->user))
 809                 return ERR_PTR(-ENOMEM);
 810
 811         root = hugetlbfs_vfsmount->mnt_root;
 812         snprintf(buf, 16, "%lu", hugetlbfs_counter());
 813         quick_string.name = buf;
 814         quick_string.len = strlen(quick_string.name);
 815         quick_string.hash = 0;
 816         dentry = d_alloc(root, &quick_string);
 817         if (!dentry)
 818                 goto out_shm_unlock;
 819
 820         error = -ENFILE;
 821         file = get_empty_filp();
 822         if (!file)
 823                 goto out_dentry;
 824
 825         error = -ENOSPC;
 826         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
 827                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
 828         if (!inode)
 829                 goto out_file;
 830
 831         d_instantiate(dentry, inode);
 832         inode->i_size = size;
 833         inode->i_nlink = 0;
 834         file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
 835         file->f_dentry = dentry;
 836         file->f_mapping = inode->i_mapping;
 837         file->f_op = &hugetlbfs_file_operations;
 838         file->f_mode = FMODE_WRITE | FMODE_READ;
 839         return file;
 840
 841 out_file:
 842         put_filp(file);
 843 out_dentry:
 844         dput(dentry);
 845 out_shm_unlock:
 846         user_shm_unlock(size, current->user);
 847         return ERR_PTR(error);
 848 }
 849
 850 static int __init init_hugetlbfs_fs(void)
 851 {
 852         int error;
 853         struct vfsmount *vfsmount;
 854
 855         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
 856                                         sizeof(struct hugetlbfs_inode_info),
 857                                         0, 0, init_once, NULL);
 858         if (hugetlbfs_inode_cachep == NULL)
 859                 return -ENOMEM;
 860
 861         error = register_filesystem(&hugetlbfs_fs_type);
 862         if (error)
 863                 goto out;
 864
 865         vfsmount = kern_mount(&hugetlbfs_fs_type);
 866
 867         if (!IS_ERR(vfsmount)) {
 868                 hugetlbfs_vfsmount = vfsmount;
 869                 return 0;
 870         }
 871
 872         error = PTR_ERR(vfsmount);
 873
 874  out:
 875         if (error)
 876                 kmem_cache_destroy(hugetlbfs_inode_cachep);
 877         return error;
 878 }
 879
 880 static void __exit exit_hugetlbfs_fs(void)
 881 {
 882         kmem_cache_destroy(hugetlbfs_inode_cachep);
 883         unregister_filesystem(&hugetlbfs_fs_type);
 884 }
 885
 886 module_init(init_hugetlbfs_fs)
 887 module_exit(exit_hugetlbfs_fs)
 888
 889 MODULE_LICENSE("GPL");