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