fs/hugetlbfs/inode.c

   1 /*
   2  * hugetlbpage-backed filesystem.  Based on ramfs.
   3  *
   4  * William Irwin, 2002
   5  *
   6  * Copyright (C) 2002 Linus Torvalds.
   7  */
   8
   9 #include <linux/module.h>
  10 #include <linux/thread_info.h>
  11 #include <asm/current.h>
  12 #include <linux/sched.h>                /* remove ASAP */
  13 #include <linux/fs.h>
  14 #include <linux/mount.h>
  15 #include <linux/file.h>
  16 #include <linux/writeback.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/highmem.h>
  19 #include <linux/init.h>
  20 #include <linux/string.h>
  21 #include <linux/backing-dev.h>
  22 #include <linux/hugetlb.h>
  23 #include <linux/pagevec.h>
  24 #include <linux/quotaops.h>
  25 #include <linux/slab.h>
  26 #include <linux/dnotify.h>
  27 #include <linux/statfs.h>
  28 #include <linux/security.h>
  29
  30 #include <asm/uaccess.h>
  31
  32 /* some random number */
  33 #define HUGETLBFS_MAGIC 0x958458f6
  34
  35 static struct super_operations hugetlbfs_ops;
  36 static struct address_space_operations hugetlbfs_aops;
  37 struct file_operations hugetlbfs_file_operations;
  38 static struct inode_operations hugetlbfs_dir_inode_operations;
  39 static struct inode_operations hugetlbfs_inode_operations;
  40
  41 static struct backing_dev_info hugetlbfs_backing_dev_info = {
  42         .ra_pages       = 0,    /* No readahead */
  43         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
  44 };
  45
  46 int sysctl_hugetlb_shm_group;
  47
  48 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  49 {
  50         struct inode *inode = file->f_dentry->d_inode;
  51         loff_t len, vma_len;
  52         int ret;
  53
  54         if ((vma->vm_flags & (VM_MAYSHARE | VM_WRITE)) == VM_WRITE)
  55                 return -EINVAL;
  56
  57         if (vma->vm_pgoff & (HPAGE_SIZE / PAGE_SIZE - 1))
  58                 return -EINVAL;
  59
  60         if (vma->vm_start & ~HPAGE_MASK)
  61                 return -EINVAL;
  62
  63         if (vma->vm_end & ~HPAGE_MASK)
  64                 return -EINVAL;
  65
  66         if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
  67                 return -EINVAL;
  68
  69         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
  70
  71         down(&inode->i_sem);
  72         file_accessed(file);
  73         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
  74         vma->vm_ops = &hugetlb_vm_ops;
  75
  76         ret = -ENOMEM;
  77         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
  78         if (!(vma->vm_flags & VM_WRITE) && len > inode->i_size)
  79                 goto out;
  80
  81         ret = 0;
  82         hugetlb_prefault_arch_hook(vma->vm_mm);
  83         if (inode->i_size < len)
  84                 inode->i_size = len;
  85 out:
  86         up(&inode->i_sem);
  87
  88         return ret;
  89 }
  90
  91 /*
  92  * Called under down_write(mmap_sem).
  93  */
  94
  95 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
  96 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
  97                 unsigned long len, unsigned long pgoff, unsigned long flags);
  98 #else
  99 static unsigned long
 100 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 101                 unsigned long len, unsigned long pgoff, unsigned long flags)
 102 {
 103         struct mm_struct *mm = current->mm;
 104         struct vm_area_struct *vma;
 105         unsigned long start_addr;
 106
 107         if (len & ~HPAGE_MASK)
 108                 return -EINVAL;
 109         if (len > TASK_SIZE)
 110                 return -ENOMEM;
 111
 112         if (addr) {
 113                 addr = ALIGN(addr, HPAGE_SIZE);
 114                 vma = find_vma(mm, addr);
 115                 if (TASK_SIZE - len >= addr &&
 116                     (!vma || addr + len <= vma->vm_start))
 117                         return addr;
 118         }
 119
 120         start_addr = mm->free_area_cache;
 121
 122         if (len <= mm->cached_hole_size)
 123                 start_addr = TASK_UNMAPPED_BASE;
 124
 125 full_search:
 126         addr = ALIGN(start_addr, HPAGE_SIZE);
 127
 128         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 129                 /* At this point:  (!vma || addr < vma->vm_end). */
 130                 if (TASK_SIZE - len < addr) {
 131                         /*
 132                          * Start a new search - just in case we missed
 133                          * some holes.
 134                          */
 135                         if (start_addr != TASK_UNMAPPED_BASE) {
 136                                 start_addr = TASK_UNMAPPED_BASE;
 137                                 goto full_search;
 138                         }
 139                         return -ENOMEM;
 140                 }
 141
 142                 if (!vma || addr + len <= vma->vm_start)
 143                         return addr;
 144                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
 145         }
 146 }
 147 #endif
 148
 149 /*
 150  * Read a page. Again trivial. If it didn't already exist
 151  * in the page cache, it is zero-filled.
 152  */
 153 static int hugetlbfs_readpage(struct file *file, struct page * page)
 154 {
 155         unlock_page(page);
 156         return -EINVAL;
 157 }
 158
 159 static int hugetlbfs_prepare_write(struct file *file,
 160                         struct page *page, unsigned offset, unsigned to)
 161 {
 162         return -EINVAL;
 163 }
 164
 165 static int hugetlbfs_commit_write(struct file *file,
 166                         struct page *page, unsigned offset, unsigned to)
 167 {
 168         return -EINVAL;
 169 }
 170
 171 static void huge_pagevec_release(struct pagevec *pvec)
 172 {
 173         int i;
 174
 175         for (i = 0; i < pagevec_count(pvec); ++i)
 176                 put_page(pvec->pages[i]);
 177
 178         pagevec_reinit(pvec);
 179 }
 180
 181 static void truncate_huge_page(struct page *page)
 182 {
 183         clear_page_dirty(page);
 184         ClearPageUptodate(page);
 185         remove_from_page_cache(page);
 186         put_page(page);
 187 }
 188
 189 static void truncate_hugepages(struct address_space *mapping, loff_t lstart)
 190 {
 191         const pgoff_t start = lstart >> HPAGE_SHIFT;
 192         struct pagevec pvec;
 193         pgoff_t next;
 194         int i;
 195
 196         pagevec_init(&pvec, 0);
 197         next = start;
 198         while (1) {
 199                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 200                         if (next == start)
 201                                 break;
 202                         next = start;
 203                         continue;
 204                 }
 205
 206                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 207                         struct page *page = pvec.pages[i];
 208
 209                         lock_page(page);
 210                         if (page->index > next)
 211                                 next = page->index;
 212                         ++next;
 213                         truncate_huge_page(page);
 214                         unlock_page(page);
 215                         hugetlb_put_quota(mapping);
 216                 }
 217                 huge_pagevec_release(&pvec);
 218         }
 219         BUG_ON(!lstart && mapping->nrpages);
 220 }
 221
 222 static void hugetlbfs_delete_inode(struct inode *inode)
 223 {
 224         if (inode->i_data.nrpages)
 225                 truncate_hugepages(&inode->i_data, 0);
 226         clear_inode(inode);
 227 }
 228
 229 static void hugetlbfs_forget_inode(struct inode *inode)
 230 {
 231         struct super_block *sb = inode->i_sb;
 232
 233         if (!hlist_unhashed(&inode->i_hash)) {
 234                 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
 235                         list_move(&inode->i_list, &inode_unused);
 236                 inodes_stat.nr_unused++;
 237                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
 238                         spin_unlock(&inode_lock);
 239                         return;
 240                 }
 241                 inode->i_state |= I_WILL_FREE;
 242                 spin_unlock(&inode_lock);
 243                 /*
 244                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
 245                  * in our backing_dev_info.
 246                  */
 247                 write_inode_now(inode, 1);
 248                 spin_lock(&inode_lock);
 249                 inode->i_state &= ~I_WILL_FREE;
 250                 inodes_stat.nr_unused--;
 251                 hlist_del_init(&inode->i_hash);
 252         }
 253         list_del_init(&inode->i_list);
 254         list_del_init(&inode->i_sb_list);
 255         inode->i_state |= I_FREEING;
 256         inodes_stat.nr_inodes--;
 257         spin_unlock(&inode_lock);
 258         if (inode->i_data.nrpages)
 259                 truncate_hugepages(&inode->i_data, 0);
 260         clear_inode(inode);
 261         destroy_inode(inode);
 262 }
 263
 264 static void hugetlbfs_drop_inode(struct inode *inode)
 265 {
 266         if (!inode->i_nlink)
 267                 generic_delete_inode(inode);
 268         else
 269                 hugetlbfs_forget_inode(inode);
 270 }
 271
 272 /*
 273  * h_pgoff is in HPAGE_SIZE units.
 274  * vma->vm_pgoff is in PAGE_SIZE units.
 275  */
 276 static inline void
 277 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
 278 {
 279         struct vm_area_struct *vma;
 280         struct prio_tree_iter iter;
 281
 282         vma_prio_tree_foreach(vma, &iter, root, h_pgoff, ULONG_MAX) {
 283                 unsigned long h_vm_pgoff;
 284                 unsigned long v_offset;
 285
 286                 h_vm_pgoff = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
 287                 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
 288                 /*
 289                  * Is this VMA fully outside the truncation point?
 290                  */
 291                 if (h_vm_pgoff >= h_pgoff)
 292                         v_offset = 0;
 293
 294                 unmap_hugepage_range(vma,
 295                                 vma->vm_start + v_offset, vma->vm_end);
 296         }
 297 }
 298
 299 /*
 300  * Expanding truncates are not allowed.
 301  */
 302 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 303 {
 304         unsigned long pgoff;
 305         struct address_space *mapping = inode->i_mapping;
 306
 307         if (offset > inode->i_size)
 308                 return -EINVAL;
 309
 310         BUG_ON(offset & ~HPAGE_MASK);
 311         pgoff = offset >> HPAGE_SHIFT;
 312
 313         inode->i_size = offset;
 314         spin_lock(&mapping->i_mmap_lock);
 315         if (!prio_tree_empty(&mapping->i_mmap))
 316                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 317         spin_unlock(&mapping->i_mmap_lock);
 318         truncate_hugepages(mapping, offset);
 319         return 0;
 320 }
 321
 322 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 323 {
 324         struct inode *inode = dentry->d_inode;
 325         int error;
 326         unsigned int ia_valid = attr->ia_valid;
 327
 328         BUG_ON(!inode);
 329
 330         error = inode_change_ok(inode, attr);
 331         if (error)
 332                 goto out;
 333
 334         if (ia_valid & ATTR_SIZE) {
 335                 error = -EINVAL;
 336                 if (!(attr->ia_size & ~HPAGE_MASK))
 337                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 338                 if (error)
 339                         goto out;
 340                 attr->ia_valid &= ~ATTR_SIZE;
 341         }
 342         error = inode_setattr(inode, attr);
 343 out:
 344         return error;
 345 }
 346
 347 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 348                                         gid_t gid, int mode, dev_t dev)
 349 {
 350         struct inode *inode;
 351
 352         inode = new_inode(sb);
 353         if (inode) {
 354                 struct hugetlbfs_inode_info *info;
 355                 inode->i_mode = mode;
 356                 inode->i_uid = uid;
 357                 inode->i_gid = gid;
 358                 inode->i_blksize = HPAGE_SIZE;
 359                 inode->i_blocks = 0;
 360                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 361                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 362                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 363                 info = HUGETLBFS_I(inode);
 364                 mpol_shared_policy_init(&info->policy);
 365                 switch (mode & S_IFMT) {
 366                 default:
 367                         init_special_inode(inode, mode, dev);
 368                         break;
 369                 case S_IFREG:
 370                         inode->i_op = &hugetlbfs_inode_operations;
 371                         inode->i_fop = &hugetlbfs_file_operations;
 372                         break;
 373                 case S_IFDIR:
 374                         inode->i_op = &hugetlbfs_dir_inode_operations;
 375                         inode->i_fop = &simple_dir_operations;
 376
 377                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 378                         inode->i_nlink++;
 379                         break;
 380                 case S_IFLNK:
 381                         inode->i_op = &page_symlink_inode_operations;
 382                         break;
 383                 }
 384         }
 385         return inode;
 386 }
 387
 388 /*
 389  * File creation. Allocate an inode, and we're done..
 390  */
 391 static int hugetlbfs_mknod(struct inode *dir,
 392                         struct dentry *dentry, int mode, dev_t dev)
 393 {
 394         struct inode *inode;
 395         int error = -ENOSPC;
 396         gid_t gid;
 397
 398         if (dir->i_mode & S_ISGID) {
 399                 gid = dir->i_gid;
 400                 if (S_ISDIR(mode))
 401                         mode |= S_ISGID;
 402         } else {
 403                 gid = current->fsgid;
 404         }
 405         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 406         if (inode) {
 407                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 408                 d_instantiate(dentry, inode);
 409                 dget(dentry);   /* Extra count - pin the dentry in core */
 410                 error = 0;
 411         }
 412         return error;
 413 }
 414
 415 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 416 {
 417         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 418         if (!retval)
 419                 dir->i_nlink++;
 420         return retval;
 421 }
 422
 423 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 424 {
 425         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 426 }
 427
 428 static int hugetlbfs_symlink(struct inode *dir,
 429                         struct dentry *dentry, const char *symname)
 430 {
 431         struct inode *inode;
 432         int error = -ENOSPC;
 433         gid_t gid;
 434
 435         if (dir->i_mode & S_ISGID)
 436                 gid = dir->i_gid;
 437         else
 438                 gid = current->fsgid;
 439
 440         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 441                                         gid, S_IFLNK|S_IRWXUGO, 0);
 442         if (inode) {
 443                 int l = strlen(symname)+1;
 444                 error = page_symlink(inode, symname, l);
 445                 if (!error) {
 446                         d_instantiate(dentry, inode);
 447                         dget(dentry);
 448                 } else
 449                         iput(inode);
 450         }
 451         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 452
 453         return error;
 454 }
 455
 456 /*
 457  * For direct-IO reads into hugetlb pages
 458  */
 459 static int hugetlbfs_set_page_dirty(struct page *page)
 460 {
 461         return 0;
 462 }
 463
 464 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
 465 {
 466         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 467
 468         buf->f_type = HUGETLBFS_MAGIC;
 469         buf->f_bsize = HPAGE_SIZE;
 470         if (sbinfo) {
 471                 spin_lock(&sbinfo->stat_lock);
 472                 buf->f_blocks = sbinfo->max_blocks;
 473                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 474                 buf->f_files = sbinfo->max_inodes;
 475                 buf->f_ffree = sbinfo->free_inodes;
 476                 spin_unlock(&sbinfo->stat_lock);
 477         }
 478         buf->f_namelen = NAME_MAX;
 479         return 0;
 480 }
 481
 482 static void hugetlbfs_put_super(struct super_block *sb)
 483 {
 484         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 485
 486         if (sbi) {
 487                 sb->s_fs_info = NULL;
 488                 kfree(sbi);
 489         }
 490 }
 491
 492 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 493 {
 494         if (sbinfo->free_inodes >= 0) {
 495                 spin_lock(&sbinfo->stat_lock);
 496                 if (unlikely(!sbinfo->free_inodes)) {
 497                         spin_unlock(&sbinfo->stat_lock);
 498                         return 0;
 499                 }
 500                 sbinfo->free_inodes--;
 501                 spin_unlock(&sbinfo->stat_lock);
 502         }
 503
 504         return 1;
 505 }
 506
 507 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 508 {
 509         if (sbinfo->free_inodes >= 0) {
 510                 spin_lock(&sbinfo->stat_lock);
 511                 sbinfo->free_inodes++;
 512                 spin_unlock(&sbinfo->stat_lock);
 513         }
 514 }
 515
 516
 517 static kmem_cache_t *hugetlbfs_inode_cachep;
 518
 519 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 520 {
 521         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 522         struct hugetlbfs_inode_info *p;
 523
 524         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 525                 return NULL;
 526         p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
 527         if (unlikely(!p)) {
 528                 hugetlbfs_inc_free_inodes(sbinfo);
 529                 return NULL;
 530         }
 531         return &p->vfs_inode;
 532 }
 533
 534 static void hugetlbfs_destroy_inode(struct inode *inode)
 535 {
 536         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 537         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 538         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 539 }
 540
 541 static struct address_space_operations hugetlbfs_aops = {
 542         .readpage       = hugetlbfs_readpage,
 543         .prepare_write  = hugetlbfs_prepare_write,
 544         .commit_write   = hugetlbfs_commit_write,
 545         .set_page_dirty = hugetlbfs_set_page_dirty,
 546 };
 547
 548
 549 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
 550 {
 551         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 552
 553         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 554             SLAB_CTOR_CONSTRUCTOR)
 555                 inode_init_once(&ei->vfs_inode);
 556 }
 557
 558 struct file_operations hugetlbfs_file_operations = {
 559         .mmap                   = hugetlbfs_file_mmap,
 560         .fsync                  = simple_sync_file,
 561         .get_unmapped_area      = hugetlb_get_unmapped_area,
 562 };
 563
 564 static struct inode_operations hugetlbfs_dir_inode_operations = {
 565         .create         = hugetlbfs_create,
 566         .lookup         = simple_lookup,
 567         .link           = simple_link,
 568         .unlink         = simple_unlink,
 569         .symlink        = hugetlbfs_symlink,
 570         .mkdir          = hugetlbfs_mkdir,
 571         .rmdir          = simple_rmdir,
 572         .mknod          = hugetlbfs_mknod,
 573         .rename         = simple_rename,
 574         .setattr        = hugetlbfs_setattr,
 575 };
 576
 577 static struct inode_operations hugetlbfs_inode_operations = {
 578         .setattr        = hugetlbfs_setattr,
 579 };
 580
 581 static struct super_operations hugetlbfs_ops = {
 582         .alloc_inode    = hugetlbfs_alloc_inode,
 583         .destroy_inode  = hugetlbfs_destroy_inode,
 584         .statfs         = hugetlbfs_statfs,
 585         .delete_inode   = hugetlbfs_delete_inode,
 586         .drop_inode     = hugetlbfs_drop_inode,
 587         .put_super      = hugetlbfs_put_super,
 588 };
 589
 590 static int
 591 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 592 {
 593         char *opt, *value, *rest;
 594
 595         if (!options)
 596                 return 0;
 597         while ((opt = strsep(&options, ",")) != NULL) {
 598                 if (!*opt)
 599                         continue;
 600
 601                 value = strchr(opt, '=');
 602                 if (!value || !*value)
 603                         return -EINVAL;
 604                 else
 605                         *value++ = '\0';
 606
 607                 if (!strcmp(opt, "uid"))
 608                         pconfig->uid = simple_strtoul(value, &value, 0);
 609                 else if (!strcmp(opt, "gid"))
 610                         pconfig->gid = simple_strtoul(value, &value, 0);
 611                 else if (!strcmp(opt, "mode"))
 612                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
 613                 else if (!strcmp(opt, "size")) {
 614                         unsigned long long size = memparse(value, &rest);
 615                         if (*rest == '%') {
 616                                 size <<= HPAGE_SHIFT;
 617                                 size *= max_huge_pages;
 618                                 do_div(size, 100);
 619                                 rest++;
 620                         }
 621                         size &= HPAGE_MASK;
 622                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 623                         value = rest;
 624                 } else if (!strcmp(opt,"nr_inodes")) {
 625                         pconfig->nr_inodes = memparse(value, &rest);
 626                         value = rest;
 627                 } else
 628                         return -EINVAL;
 629
 630                 if (*value)
 631                         return -EINVAL;
 632         }
 633         return 0;
 634 }
 635
 636 static int
 637 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 638 {
 639         struct inode * inode;
 640         struct dentry * root;
 641         int ret;
 642         struct hugetlbfs_config config;
 643         struct hugetlbfs_sb_info *sbinfo;
 644
 645         config.nr_blocks = -1; /* No limit on size by default */
 646         config.nr_inodes = -1; /* No limit on number of inodes by default */
 647         config.uid = current->fsuid;
 648         config.gid = current->fsgid;
 649         config.mode = 0755;
 650         ret = hugetlbfs_parse_options(data, &config);
 651
 652         if (ret)
 653                 return ret;
 654
 655         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 656         if (!sbinfo)
 657                 return -ENOMEM;
 658         sb->s_fs_info = sbinfo;
 659         spin_lock_init(&sbinfo->stat_lock);
 660         sbinfo->max_blocks = config.nr_blocks;
 661         sbinfo->free_blocks = config.nr_blocks;
 662         sbinfo->max_inodes = config.nr_inodes;
 663         sbinfo->free_inodes = config.nr_inodes;
 664         sb->s_maxbytes = MAX_LFS_FILESIZE;
 665         sb->s_blocksize = HPAGE_SIZE;
 666         sb->s_blocksize_bits = HPAGE_SHIFT;
 667         sb->s_magic = HUGETLBFS_MAGIC;
 668         sb->s_op = &hugetlbfs_ops;
 669         sb->s_time_gran = 1;
 670         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 671                                         S_IFDIR | config.mode, 0);
 672         if (!inode)
 673                 goto out_free;
 674
 675         root = d_alloc_root(inode);
 676         if (!root) {
 677                 iput(inode);
 678                 goto out_free;
 679         }
 680         sb->s_root = root;
 681         return 0;
 682 out_free:
 683         kfree(sbinfo);
 684         return -ENOMEM;
 685 }
 686
 687 int hugetlb_get_quota(struct address_space *mapping)
 688 {
 689         int ret = 0;
 690         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 691
 692         if (sbinfo->free_blocks > -1) {
 693                 spin_lock(&sbinfo->stat_lock);
 694                 if (sbinfo->free_blocks > 0)
 695                         sbinfo->free_blocks--;
 696                 else
 697                         ret = -ENOMEM;
 698                 spin_unlock(&sbinfo->stat_lock);
 699         }
 700
 701         return ret;
 702 }
 703
 704 void hugetlb_put_quota(struct address_space *mapping)
 705 {
 706         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 707
 708         if (sbinfo->free_blocks > -1) {
 709                 spin_lock(&sbinfo->stat_lock);
 710                 sbinfo->free_blocks++;
 711                 spin_unlock(&sbinfo->stat_lock);
 712         }
 713 }
 714
 715 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
 716         int flags, const char *dev_name, void *data)
 717 {
 718         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 719 }
 720
 721 static struct file_system_type hugetlbfs_fs_type = {
 722         .name           = "hugetlbfs",
 723         .get_sb         = hugetlbfs_get_sb,
 724         .kill_sb        = kill_litter_super,
 725 };
 726
 727 static struct vfsmount *hugetlbfs_vfsmount;
 728
 729 /*
 730  * Return the next identifier for a shm file
 731  */
 732 static unsigned long hugetlbfs_counter(void)
 733 {
 734         static DEFINE_SPINLOCK(lock);
 735         static unsigned long counter;
 736         unsigned long ret;
 737
 738         spin_lock(&lock);
 739         ret = ++counter;
 740         spin_unlock(&lock);
 741         return ret;
 742 }
 743
 744 static int can_do_hugetlb_shm(void)
 745 {
 746         return likely(capable(CAP_IPC_LOCK) ||
 747                         in_group_p(sysctl_hugetlb_shm_group) ||
 748                         can_do_mlock());
 749 }
 750
 751 struct file *hugetlb_zero_setup(size_t size)
 752 {
 753         int error = -ENOMEM;
 754         struct file *file;
 755         struct inode *inode;
 756         struct dentry *dentry, *root;
 757         struct qstr quick_string;
 758         char buf[16];
 759
 760         if (!can_do_hugetlb_shm())
 761                 return ERR_PTR(-EPERM);
 762
 763         if (!is_hugepage_mem_enough(size))
 764                 return ERR_PTR(-ENOMEM);
 765
 766         if (!user_shm_lock(size, current->user))
 767                 return ERR_PTR(-ENOMEM);
 768
 769         root = hugetlbfs_vfsmount->mnt_root;
 770         snprintf(buf, 16, "%lu", hugetlbfs_counter());
 771         quick_string.name = buf;
 772         quick_string.len = strlen(quick_string.name);
 773         quick_string.hash = 0;
 774         dentry = d_alloc(root, &quick_string);
 775         if (!dentry)
 776                 goto out_shm_unlock;
 777
 778         error = -ENFILE;
 779         file = get_empty_filp();
 780         if (!file)
 781                 goto out_dentry;
 782
 783         error = -ENOSPC;
 784         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
 785                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
 786         if (!inode)
 787                 goto out_file;
 788
 789         d_instantiate(dentry, inode);
 790         inode->i_size = size;
 791         inode->i_nlink = 0;
 792         file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
 793         file->f_dentry = dentry;
 794         file->f_mapping = inode->i_mapping;
 795         file->f_op = &hugetlbfs_file_operations;
 796         file->f_mode = FMODE_WRITE | FMODE_READ;
 797         return file;
 798
 799 out_file:
 800         put_filp(file);
 801 out_dentry:
 802         dput(dentry);
 803 out_shm_unlock:
 804         user_shm_unlock(size, current->user);
 805         return ERR_PTR(error);
 806 }
 807
 808 static int __init init_hugetlbfs_fs(void)
 809 {
 810         int error;
 811         struct vfsmount *vfsmount;
 812
 813         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
 814                                         sizeof(struct hugetlbfs_inode_info),
 815                                         0, 0, init_once, NULL);
 816         if (hugetlbfs_inode_cachep == NULL)
 817                 return -ENOMEM;
 818
 819         error = register_filesystem(&hugetlbfs_fs_type);
 820         if (error)
 821                 goto out;
 822
 823         vfsmount = kern_mount(&hugetlbfs_fs_type);
 824
 825         if (!IS_ERR(vfsmount)) {
 826                 hugetlbfs_vfsmount = vfsmount;
 827                 return 0;
 828         }
 829
 830         error = PTR_ERR(vfsmount);
 831
 832  out:
 833         if (error)
 834                 kmem_cache_destroy(hugetlbfs_inode_cachep);
 835         return error;
 836 }
 837
 838 static void __exit exit_hugetlbfs_fs(void)
 839 {
 840         kmem_cache_destroy(hugetlbfs_inode_cachep);
 841         unregister_filesystem(&hugetlbfs_fs_type);
 842 }
 843
 844 module_init(init_hugetlbfs_fs)
 845 module_exit(exit_hugetlbfs_fs)
 846
 847 MODULE_LICENSE("GPL");