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 void huge_pagevec_release(struct pagevec *pvec)
  49 {
  50         int i;
  51
  52         for (i = 0; i < pagevec_count(pvec); ++i)
  53                 put_page(pvec->pages[i]);
  54
  55         pagevec_reinit(pvec);
  56 }
  57
  58 /*
  59  * huge_pages_needed tries to determine the number of new huge pages that
  60  * will be required to fully populate this VMA.  This will be equal to
  61  * the size of the VMA in huge pages minus the number of huge pages
  62  * (covered by this VMA) that are found in the page cache.
  63  *
  64  * Result is in bytes to be compatible with is_hugepage_mem_enough()
  65  */
  66 static unsigned long
  67 huge_pages_needed(struct address_space *mapping, struct vm_area_struct *vma)
  68 {
  69         int i;
  70         struct pagevec pvec;
  71         unsigned long start = vma->vm_start;
  72         unsigned long end = vma->vm_end;
  73         unsigned long hugepages = (end - start) >> HPAGE_SHIFT;
  74         pgoff_t next = vma->vm_pgoff;
  75         pgoff_t endpg = next + ((end - start) >> PAGE_SHIFT);
  76
  77         pagevec_init(&pvec, 0);
  78         while (next < endpg) {
  79                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE))
  80                         break;
  81                 for (i = 0; i < pagevec_count(&pvec); i++) {
  82                         struct page *page = pvec.pages[i];
  83                         if (page->index > next)
  84                                 next = page->index;
  85                         if (page->index >= endpg)
  86                                 break;
  87                         next++;
  88                         hugepages--;
  89                 }
  90                 huge_pagevec_release(&pvec);
  91         }
  92         return hugepages << HPAGE_SHIFT;
  93 }
  94
  95 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  96 {
  97         struct inode *inode = file->f_dentry->d_inode;
  98         struct address_space *mapping = inode->i_mapping;
  99         unsigned long bytes;
 100         loff_t len, vma_len;
 101         int ret;
 102
 103         if ((vma->vm_flags & (VM_MAYSHARE | VM_WRITE)) == VM_WRITE)
 104                 return -EINVAL;
 105
 106         if (vma->vm_pgoff & (HPAGE_SIZE / PAGE_SIZE - 1))
 107                 return -EINVAL;
 108
 109         if (vma->vm_start & ~HPAGE_MASK)
 110                 return -EINVAL;
 111
 112         if (vma->vm_end & ~HPAGE_MASK)
 113                 return -EINVAL;
 114
 115         if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
 116                 return -EINVAL;
 117
 118         bytes = huge_pages_needed(mapping, vma);
 119         if (!is_hugepage_mem_enough(bytes))
 120                 return -ENOMEM;
 121
 122         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 123
 124         down(&inode->i_sem);
 125         file_accessed(file);
 126         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
 127         vma->vm_ops = &hugetlb_vm_ops;
 128
 129         ret = -ENOMEM;
 130         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 131         if (!(vma->vm_flags & VM_WRITE) && len > inode->i_size)
 132                 goto out;
 133
 134         ret = 0;
 135         hugetlb_prefault_arch_hook(vma->vm_mm);
 136         if (inode->i_size < len)
 137                 inode->i_size = len;
 138 out:
 139         up(&inode->i_sem);
 140
 141         return ret;
 142 }
 143
 144 /*
 145  * Called under down_write(mmap_sem).
 146  */
 147
 148 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 149 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 150                 unsigned long len, unsigned long pgoff, unsigned long flags);
 151 #else
 152 static unsigned long
 153 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 154                 unsigned long len, unsigned long pgoff, unsigned long flags)
 155 {
 156         struct mm_struct *mm = current->mm;
 157         struct vm_area_struct *vma;
 158         unsigned long start_addr;
 159
 160         if (len & ~HPAGE_MASK)
 161                 return -EINVAL;
 162         if (len > TASK_SIZE)
 163                 return -ENOMEM;
 164
 165         if (addr) {
 166                 addr = ALIGN(addr, HPAGE_SIZE);
 167                 vma = find_vma(mm, addr);
 168                 if (TASK_SIZE - len >= addr &&
 169                     (!vma || addr + len <= vma->vm_start))
 170                         return addr;
 171         }
 172
 173         start_addr = mm->free_area_cache;
 174
 175         if (len <= mm->cached_hole_size)
 176                 start_addr = TASK_UNMAPPED_BASE;
 177
 178 full_search:
 179         addr = ALIGN(start_addr, HPAGE_SIZE);
 180
 181         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 182                 /* At this point:  (!vma || addr < vma->vm_end). */
 183                 if (TASK_SIZE - len < addr) {
 184                         /*
 185                          * Start a new search - just in case we missed
 186                          * some holes.
 187                          */
 188                         if (start_addr != TASK_UNMAPPED_BASE) {
 189                                 start_addr = TASK_UNMAPPED_BASE;
 190                                 goto full_search;
 191                         }
 192                         return -ENOMEM;
 193                 }
 194
 195                 if (!vma || addr + len <= vma->vm_start)
 196                         return addr;
 197                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
 198         }
 199 }
 200 #endif
 201
 202 /*
 203  * Read a page. Again trivial. If it didn't already exist
 204  * in the page cache, it is zero-filled.
 205  */
 206 static int hugetlbfs_readpage(struct file *file, struct page * page)
 207 {
 208         unlock_page(page);
 209         return -EINVAL;
 210 }
 211
 212 static int hugetlbfs_prepare_write(struct file *file,
 213                         struct page *page, unsigned offset, unsigned to)
 214 {
 215         return -EINVAL;
 216 }
 217
 218 static int hugetlbfs_commit_write(struct file *file,
 219                         struct page *page, unsigned offset, unsigned to)
 220 {
 221         return -EINVAL;
 222 }
 223
 224 static void truncate_huge_page(struct page *page)
 225 {
 226         clear_page_dirty(page);
 227         ClearPageUptodate(page);
 228         remove_from_page_cache(page);
 229         put_page(page);
 230 }
 231
 232 static void truncate_hugepages(struct address_space *mapping, loff_t lstart)
 233 {
 234         const pgoff_t start = lstart >> HPAGE_SHIFT;
 235         struct pagevec pvec;
 236         pgoff_t next;
 237         int i;
 238
 239         pagevec_init(&pvec, 0);
 240         next = start;
 241         while (1) {
 242                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 243                         if (next == start)
 244                                 break;
 245                         next = start;
 246                         continue;
 247                 }
 248
 249                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 250                         struct page *page = pvec.pages[i];
 251
 252                         lock_page(page);
 253                         if (page->index > next)
 254                                 next = page->index;
 255                         ++next;
 256                         truncate_huge_page(page);
 257                         unlock_page(page);
 258                         hugetlb_put_quota(mapping);
 259                 }
 260                 huge_pagevec_release(&pvec);
 261         }
 262         BUG_ON(!lstart && mapping->nrpages);
 263 }
 264
 265 static void hugetlbfs_delete_inode(struct inode *inode)
 266 {
 267         if (inode->i_data.nrpages)
 268                 truncate_hugepages(&inode->i_data, 0);
 269         clear_inode(inode);
 270 }
 271
 272 static void hugetlbfs_forget_inode(struct inode *inode)
 273 {
 274         struct super_block *sb = inode->i_sb;
 275
 276         if (!hlist_unhashed(&inode->i_hash)) {
 277                 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
 278                         list_move(&inode->i_list, &inode_unused);
 279                 inodes_stat.nr_unused++;
 280                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
 281                         spin_unlock(&inode_lock);
 282                         return;
 283                 }
 284                 inode->i_state |= I_WILL_FREE;
 285                 spin_unlock(&inode_lock);
 286                 /*
 287                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
 288                  * in our backing_dev_info.
 289                  */
 290                 write_inode_now(inode, 1);
 291                 spin_lock(&inode_lock);
 292                 inode->i_state &= ~I_WILL_FREE;
 293                 inodes_stat.nr_unused--;
 294                 hlist_del_init(&inode->i_hash);
 295         }
 296         list_del_init(&inode->i_list);
 297         list_del_init(&inode->i_sb_list);
 298         inode->i_state |= I_FREEING;
 299         inodes_stat.nr_inodes--;
 300         spin_unlock(&inode_lock);
 301         if (inode->i_data.nrpages)
 302                 truncate_hugepages(&inode->i_data, 0);
 303         clear_inode(inode);
 304         destroy_inode(inode);
 305 }
 306
 307 static void hugetlbfs_drop_inode(struct inode *inode)
 308 {
 309         if (!inode->i_nlink)
 310                 generic_delete_inode(inode);
 311         else
 312                 hugetlbfs_forget_inode(inode);
 313 }
 314
 315 /*
 316  * h_pgoff is in HPAGE_SIZE units.
 317  * vma->vm_pgoff is in PAGE_SIZE units.
 318  */
 319 static inline void
 320 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
 321 {
 322         struct vm_area_struct *vma;
 323         struct prio_tree_iter iter;
 324
 325         vma_prio_tree_foreach(vma, &iter, root, h_pgoff, ULONG_MAX) {
 326                 unsigned long h_vm_pgoff;
 327                 unsigned long v_offset;
 328
 329                 h_vm_pgoff = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
 330                 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
 331                 /*
 332                  * Is this VMA fully outside the truncation point?
 333                  */
 334                 if (h_vm_pgoff >= h_pgoff)
 335                         v_offset = 0;
 336
 337                 unmap_hugepage_range(vma,
 338                                 vma->vm_start + v_offset, vma->vm_end);
 339         }
 340 }
 341
 342 /*
 343  * Expanding truncates are not allowed.
 344  */
 345 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 346 {
 347         unsigned long pgoff;
 348         struct address_space *mapping = inode->i_mapping;
 349
 350         if (offset > inode->i_size)
 351                 return -EINVAL;
 352
 353         BUG_ON(offset & ~HPAGE_MASK);
 354         pgoff = offset >> HPAGE_SHIFT;
 355
 356         inode->i_size = offset;
 357         spin_lock(&mapping->i_mmap_lock);
 358         if (!prio_tree_empty(&mapping->i_mmap))
 359                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 360         spin_unlock(&mapping->i_mmap_lock);
 361         truncate_hugepages(mapping, offset);
 362         return 0;
 363 }
 364
 365 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 366 {
 367         struct inode *inode = dentry->d_inode;
 368         int error;
 369         unsigned int ia_valid = attr->ia_valid;
 370
 371         BUG_ON(!inode);
 372
 373         error = inode_change_ok(inode, attr);
 374         if (error)
 375                 goto out;
 376
 377         if (ia_valid & ATTR_SIZE) {
 378                 error = -EINVAL;
 379                 if (!(attr->ia_size & ~HPAGE_MASK))
 380                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 381                 if (error)
 382                         goto out;
 383                 attr->ia_valid &= ~ATTR_SIZE;
 384         }
 385         error = inode_setattr(inode, attr);
 386 out:
 387         return error;
 388 }
 389
 390 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 391                                         gid_t gid, int mode, dev_t dev)
 392 {
 393         struct inode *inode;
 394
 395         inode = new_inode(sb);
 396         if (inode) {
 397                 struct hugetlbfs_inode_info *info;
 398                 inode->i_mode = mode;
 399                 inode->i_uid = uid;
 400                 inode->i_gid = gid;
 401                 inode->i_blksize = HPAGE_SIZE;
 402                 inode->i_blocks = 0;
 403                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 404                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 405                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 406                 info = HUGETLBFS_I(inode);
 407                 mpol_shared_policy_init(&info->policy);
 408                 switch (mode & S_IFMT) {
 409                 default:
 410                         init_special_inode(inode, mode, dev);
 411                         break;
 412                 case S_IFREG:
 413                         inode->i_op = &hugetlbfs_inode_operations;
 414                         inode->i_fop = &hugetlbfs_file_operations;
 415                         break;
 416                 case S_IFDIR:
 417                         inode->i_op = &hugetlbfs_dir_inode_operations;
 418                         inode->i_fop = &simple_dir_operations;
 419
 420                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 421                         inode->i_nlink++;
 422                         break;
 423                 case S_IFLNK:
 424                         inode->i_op = &page_symlink_inode_operations;
 425                         break;
 426                 }
 427         }
 428         return inode;
 429 }
 430
 431 /*
 432  * File creation. Allocate an inode, and we're done..
 433  */
 434 static int hugetlbfs_mknod(struct inode *dir,
 435                         struct dentry *dentry, int mode, dev_t dev)
 436 {
 437         struct inode *inode;
 438         int error = -ENOSPC;
 439         gid_t gid;
 440
 441         if (dir->i_mode & S_ISGID) {
 442                 gid = dir->i_gid;
 443                 if (S_ISDIR(mode))
 444                         mode |= S_ISGID;
 445         } else {
 446                 gid = current->fsgid;
 447         }
 448         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 449         if (inode) {
 450                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 451                 d_instantiate(dentry, inode);
 452                 dget(dentry);   /* Extra count - pin the dentry in core */
 453                 error = 0;
 454         }
 455         return error;
 456 }
 457
 458 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 459 {
 460         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 461         if (!retval)
 462                 dir->i_nlink++;
 463         return retval;
 464 }
 465
 466 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 467 {
 468         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 469 }
 470
 471 static int hugetlbfs_symlink(struct inode *dir,
 472                         struct dentry *dentry, const char *symname)
 473 {
 474         struct inode *inode;
 475         int error = -ENOSPC;
 476         gid_t gid;
 477
 478         if (dir->i_mode & S_ISGID)
 479                 gid = dir->i_gid;
 480         else
 481                 gid = current->fsgid;
 482
 483         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 484                                         gid, S_IFLNK|S_IRWXUGO, 0);
 485         if (inode) {
 486                 int l = strlen(symname)+1;
 487                 error = page_symlink(inode, symname, l);
 488                 if (!error) {
 489                         d_instantiate(dentry, inode);
 490                         dget(dentry);
 491                 } else
 492                         iput(inode);
 493         }
 494         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 495
 496         return error;
 497 }
 498
 499 /*
 500  * For direct-IO reads into hugetlb pages
 501  */
 502 static int hugetlbfs_set_page_dirty(struct page *page)
 503 {
 504         return 0;
 505 }
 506
 507 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
 508 {
 509         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 510
 511         buf->f_type = HUGETLBFS_MAGIC;
 512         buf->f_bsize = HPAGE_SIZE;
 513         if (sbinfo) {
 514                 spin_lock(&sbinfo->stat_lock);
 515                 /* If no limits set, just report 0 for max/free/used
 516                  * blocks, like simple_statfs() */
 517                 if (sbinfo->max_blocks >= 0) {
 518                         buf->f_blocks = sbinfo->max_blocks;
 519                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 520                         buf->f_files = sbinfo->max_inodes;
 521                         buf->f_ffree = sbinfo->free_inodes;
 522                 }
 523                 spin_unlock(&sbinfo->stat_lock);
 524         }
 525         buf->f_namelen = NAME_MAX;
 526         return 0;
 527 }
 528
 529 static void hugetlbfs_put_super(struct super_block *sb)
 530 {
 531         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 532
 533         if (sbi) {
 534                 sb->s_fs_info = NULL;
 535                 kfree(sbi);
 536         }
 537 }
 538
 539 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 540 {
 541         if (sbinfo->free_inodes >= 0) {
 542                 spin_lock(&sbinfo->stat_lock);
 543                 if (unlikely(!sbinfo->free_inodes)) {
 544                         spin_unlock(&sbinfo->stat_lock);
 545                         return 0;
 546                 }
 547                 sbinfo->free_inodes--;
 548                 spin_unlock(&sbinfo->stat_lock);
 549         }
 550
 551         return 1;
 552 }
 553
 554 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 555 {
 556         if (sbinfo->free_inodes >= 0) {
 557                 spin_lock(&sbinfo->stat_lock);
 558                 sbinfo->free_inodes++;
 559                 spin_unlock(&sbinfo->stat_lock);
 560         }
 561 }
 562
 563
 564 static kmem_cache_t *hugetlbfs_inode_cachep;
 565
 566 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 567 {
 568         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 569         struct hugetlbfs_inode_info *p;
 570
 571         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 572                 return NULL;
 573         p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
 574         if (unlikely(!p)) {
 575                 hugetlbfs_inc_free_inodes(sbinfo);
 576                 return NULL;
 577         }
 578         return &p->vfs_inode;
 579 }
 580
 581 static void hugetlbfs_destroy_inode(struct inode *inode)
 582 {
 583         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 584         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 585         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 586 }
 587
 588 static struct address_space_operations hugetlbfs_aops = {
 589         .readpage       = hugetlbfs_readpage,
 590         .prepare_write  = hugetlbfs_prepare_write,
 591         .commit_write   = hugetlbfs_commit_write,
 592         .set_page_dirty = hugetlbfs_set_page_dirty,
 593 };
 594
 595
 596 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
 597 {
 598         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 599
 600         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 601             SLAB_CTOR_CONSTRUCTOR)
 602                 inode_init_once(&ei->vfs_inode);
 603 }
 604
 605 struct file_operations hugetlbfs_file_operations = {
 606         .mmap                   = hugetlbfs_file_mmap,
 607         .fsync                  = simple_sync_file,
 608         .get_unmapped_area      = hugetlb_get_unmapped_area,
 609 };
 610
 611 static struct inode_operations hugetlbfs_dir_inode_operations = {
 612         .create         = hugetlbfs_create,
 613         .lookup         = simple_lookup,
 614         .link           = simple_link,
 615         .unlink         = simple_unlink,
 616         .symlink        = hugetlbfs_symlink,
 617         .mkdir          = hugetlbfs_mkdir,
 618         .rmdir          = simple_rmdir,
 619         .mknod          = hugetlbfs_mknod,
 620         .rename         = simple_rename,
 621         .setattr        = hugetlbfs_setattr,
 622 };
 623
 624 static struct inode_operations hugetlbfs_inode_operations = {
 625         .setattr        = hugetlbfs_setattr,
 626 };
 627
 628 static struct super_operations hugetlbfs_ops = {
 629         .alloc_inode    = hugetlbfs_alloc_inode,
 630         .destroy_inode  = hugetlbfs_destroy_inode,
 631         .statfs         = hugetlbfs_statfs,
 632         .delete_inode   = hugetlbfs_delete_inode,
 633         .drop_inode     = hugetlbfs_drop_inode,
 634         .put_super      = hugetlbfs_put_super,
 635 };
 636
 637 static int
 638 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 639 {
 640         char *opt, *value, *rest;
 641
 642         if (!options)
 643                 return 0;
 644         while ((opt = strsep(&options, ",")) != NULL) {
 645                 if (!*opt)
 646                         continue;
 647
 648                 value = strchr(opt, '=');
 649                 if (!value || !*value)
 650                         return -EINVAL;
 651                 else
 652                         *value++ = '\0';
 653
 654                 if (!strcmp(opt, "uid"))
 655                         pconfig->uid = simple_strtoul(value, &value, 0);
 656                 else if (!strcmp(opt, "gid"))
 657                         pconfig->gid = simple_strtoul(value, &value, 0);
 658                 else if (!strcmp(opt, "mode"))
 659                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
 660                 else if (!strcmp(opt, "size")) {
 661                         unsigned long long size = memparse(value, &rest);
 662                         if (*rest == '%') {
 663                                 size <<= HPAGE_SHIFT;
 664                                 size *= max_huge_pages;
 665                                 do_div(size, 100);
 666                                 rest++;
 667                         }
 668                         size &= HPAGE_MASK;
 669                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 670                         value = rest;
 671                 } else if (!strcmp(opt,"nr_inodes")) {
 672                         pconfig->nr_inodes = memparse(value, &rest);
 673                         value = rest;
 674                 } else
 675                         return -EINVAL;
 676
 677                 if (*value)
 678                         return -EINVAL;
 679         }
 680         return 0;
 681 }
 682
 683 static int
 684 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 685 {
 686         struct inode * inode;
 687         struct dentry * root;
 688         int ret;
 689         struct hugetlbfs_config config;
 690         struct hugetlbfs_sb_info *sbinfo;
 691
 692         config.nr_blocks = -1; /* No limit on size by default */
 693         config.nr_inodes = -1; /* No limit on number of inodes by default */
 694         config.uid = current->fsuid;
 695         config.gid = current->fsgid;
 696         config.mode = 0755;
 697         ret = hugetlbfs_parse_options(data, &config);
 698
 699         if (ret)
 700                 return ret;
 701
 702         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 703         if (!sbinfo)
 704                 return -ENOMEM;
 705         sb->s_fs_info = sbinfo;
 706         spin_lock_init(&sbinfo->stat_lock);
 707         sbinfo->max_blocks = config.nr_blocks;
 708         sbinfo->free_blocks = config.nr_blocks;
 709         sbinfo->max_inodes = config.nr_inodes;
 710         sbinfo->free_inodes = config.nr_inodes;
 711         sb->s_maxbytes = MAX_LFS_FILESIZE;
 712         sb->s_blocksize = HPAGE_SIZE;
 713         sb->s_blocksize_bits = HPAGE_SHIFT;
 714         sb->s_magic = HUGETLBFS_MAGIC;
 715         sb->s_op = &hugetlbfs_ops;
 716         sb->s_time_gran = 1;
 717         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 718                                         S_IFDIR | config.mode, 0);
 719         if (!inode)
 720                 goto out_free;
 721
 722         root = d_alloc_root(inode);
 723         if (!root) {
 724                 iput(inode);
 725                 goto out_free;
 726         }
 727         sb->s_root = root;
 728         return 0;
 729 out_free:
 730         kfree(sbinfo);
 731         return -ENOMEM;
 732 }
 733
 734 int hugetlb_get_quota(struct address_space *mapping)
 735 {
 736         int ret = 0;
 737         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 738
 739         if (sbinfo->free_blocks > -1) {
 740                 spin_lock(&sbinfo->stat_lock);
 741                 if (sbinfo->free_blocks > 0)
 742                         sbinfo->free_blocks--;
 743                 else
 744                         ret = -ENOMEM;
 745                 spin_unlock(&sbinfo->stat_lock);
 746         }
 747
 748         return ret;
 749 }
 750
 751 void hugetlb_put_quota(struct address_space *mapping)
 752 {
 753         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 754
 755         if (sbinfo->free_blocks > -1) {
 756                 spin_lock(&sbinfo->stat_lock);
 757                 sbinfo->free_blocks++;
 758                 spin_unlock(&sbinfo->stat_lock);
 759         }
 760 }
 761
 762 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
 763         int flags, const char *dev_name, void *data)
 764 {
 765         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 766 }
 767
 768 static struct file_system_type hugetlbfs_fs_type = {
 769         .name           = "hugetlbfs",
 770         .get_sb         = hugetlbfs_get_sb,
 771         .kill_sb        = kill_litter_super,
 772 };
 773
 774 static struct vfsmount *hugetlbfs_vfsmount;
 775
 776 /*
 777  * Return the next identifier for a shm file
 778  */
 779 static unsigned long hugetlbfs_counter(void)
 780 {
 781         static DEFINE_SPINLOCK(lock);
 782         static unsigned long counter;
 783         unsigned long ret;
 784
 785         spin_lock(&lock);
 786         ret = ++counter;
 787         spin_unlock(&lock);
 788         return ret;
 789 }
 790
 791 static int can_do_hugetlb_shm(void)
 792 {
 793         return likely(capable(CAP_IPC_LOCK) ||
 794                         in_group_p(sysctl_hugetlb_shm_group) ||
 795                         can_do_mlock());
 796 }
 797
 798 struct file *hugetlb_zero_setup(size_t size)
 799 {
 800         int error = -ENOMEM;
 801         struct file *file;
 802         struct inode *inode;
 803         struct dentry *dentry, *root;
 804         struct qstr quick_string;
 805         char buf[16];
 806
 807         if (!can_do_hugetlb_shm())
 808                 return ERR_PTR(-EPERM);
 809
 810         if (!is_hugepage_mem_enough(size))
 811                 return ERR_PTR(-ENOMEM);
 812
 813         if (!user_shm_lock(size, current->user))
 814                 return ERR_PTR(-ENOMEM);
 815
 816         root = hugetlbfs_vfsmount->mnt_root;
 817         snprintf(buf, 16, "%lu", hugetlbfs_counter());
 818         quick_string.name = buf;
 819         quick_string.len = strlen(quick_string.name);
 820         quick_string.hash = 0;
 821         dentry = d_alloc(root, &quick_string);
 822         if (!dentry)
 823                 goto out_shm_unlock;
 824
 825         error = -ENFILE;
 826         file = get_empty_filp();
 827         if (!file)
 828                 goto out_dentry;
 829
 830         error = -ENOSPC;
 831         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
 832                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
 833         if (!inode)
 834                 goto out_file;
 835
 836         d_instantiate(dentry, inode);
 837         inode->i_size = size;
 838         inode->i_nlink = 0;
 839         file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
 840         file->f_dentry = dentry;
 841         file->f_mapping = inode->i_mapping;
 842         file->f_op = &hugetlbfs_file_operations;
 843         file->f_mode = FMODE_WRITE | FMODE_READ;
 844         return file;
 845
 846 out_file:
 847         put_filp(file);
 848 out_dentry:
 849         dput(dentry);
 850 out_shm_unlock:
 851         user_shm_unlock(size, current->user);
 852         return ERR_PTR(error);
 853 }
 854
 855 static int __init init_hugetlbfs_fs(void)
 856 {
 857         int error;
 858         struct vfsmount *vfsmount;
 859
 860         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
 861                                         sizeof(struct hugetlbfs_inode_info),
 862                                         0, 0, init_once, NULL);
 863         if (hugetlbfs_inode_cachep == NULL)
 864                 return -ENOMEM;
 865
 866         error = register_filesystem(&hugetlbfs_fs_type);
 867         if (error)
 868                 goto out;
 869
 870         vfsmount = kern_mount(&hugetlbfs_fs_type);
 871
 872         if (!IS_ERR(vfsmount)) {
 873                 hugetlbfs_vfsmount = vfsmount;
 874                 return 0;
 875         }
 876
 877         error = PTR_ERR(vfsmount);
 878
 879  out:
 880         if (error)
 881                 kmem_cache_destroy(hugetlbfs_inode_cachep);
 882         return error;
 883 }
 884
 885 static void __exit exit_hugetlbfs_fs(void)
 886 {
 887         kmem_cache_destroy(hugetlbfs_inode_cachep);
 888         unregister_filesystem(&hugetlbfs_fs_type);
 889 }
 890
 891 module_init(init_hugetlbfs_fs)
 892 module_exit(exit_hugetlbfs_fs)
 893
 894 MODULE_LICENSE("GPL");