fs/hugetlbfs/inode.c

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