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/kernel.h>
  17 #include <linux/writeback.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/highmem.h>
  20 #include <linux/init.h>
  21 #include <linux/string.h>
  22 #include <linux/capability.h>
  23 #include <linux/ctype.h>
  24 #include <linux/backing-dev.h>
  25 #include <linux/hugetlb.h>
  26 #include <linux/pagevec.h>
  27 #include <linux/parser.h>
  28 #include <linux/mman.h>
  29 #include <linux/quotaops.h>
  30 #include <linux/slab.h>
  31 #include <linux/dnotify.h>
  32 #include <linux/statfs.h>
  33 #include <linux/security.h>
  34
  35 #include <asm/uaccess.h>
  36
  37 /* some random number */
  38 #define HUGETLBFS_MAGIC 0x958458f6
  39
  40 static const struct super_operations hugetlbfs_ops;
  41 static const struct address_space_operations hugetlbfs_aops;
  42 const struct file_operations hugetlbfs_file_operations;
  43 static const struct inode_operations hugetlbfs_dir_inode_operations;
  44 static const struct inode_operations hugetlbfs_inode_operations;
  45
  46 static struct backing_dev_info hugetlbfs_backing_dev_info = {
  47         .ra_pages       = 0,    /* No readahead */
  48         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
  49 };
  50
  51 int sysctl_hugetlb_shm_group;
  52
  53 enum {
  54         Opt_size, Opt_nr_inodes,
  55         Opt_mode, Opt_uid, Opt_gid,
  56         Opt_err,
  57 };
  58
  59 static match_table_t tokens = {
  60         {Opt_size,      "size=%s"},
  61         {Opt_nr_inodes, "nr_inodes=%s"},
  62         {Opt_mode,      "mode=%o"},
  63         {Opt_uid,       "uid=%u"},
  64         {Opt_gid,       "gid=%u"},
  65         {Opt_err,       NULL},
  66 };
  67
  68 static void huge_pagevec_release(struct pagevec *pvec)
  69 {
  70         int i;
  71
  72         for (i = 0; i < pagevec_count(pvec); ++i)
  73                 put_page(pvec->pages[i]);
  74
  75         pagevec_reinit(pvec);
  76 }
  77
  78 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  79 {
  80         struct inode *inode = file->f_path.dentry->d_inode;
  81         loff_t len, vma_len;
  82         int ret;
  83
  84         /*
  85          * vma address alignment (but not the pgoff alignment) has
  86          * already been checked by prepare_hugepage_range.  If you add
  87          * any error returns here, do so after setting VM_HUGETLB, so
  88          * is_vm_hugetlb_page tests below unmap_region go the right
  89          * way when do_mmap_pgoff unwinds (may be important on powerpc
  90          * and ia64).
  91          */
  92         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
  93         vma->vm_ops = &hugetlb_vm_ops;
  94
  95         if (vma->vm_pgoff & ~(HPAGE_MASK >> PAGE_SHIFT))
  96                 return -EINVAL;
  97
  98         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
  99
 100         mutex_lock(&inode->i_mutex);
 101         file_accessed(file);
 102
 103         ret = -ENOMEM;
 104         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 105
 106         if (vma->vm_flags & VM_MAYSHARE &&
 107             hugetlb_reserve_pages(inode, vma->vm_pgoff >> (HPAGE_SHIFT-PAGE_SHIFT),
 108                                   len >> HPAGE_SHIFT))
 109                 goto out;
 110
 111         ret = 0;
 112         hugetlb_prefault_arch_hook(vma->vm_mm);
 113         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 114                 inode->i_size = len;
 115 out:
 116         mutex_unlock(&inode->i_mutex);
 117
 118         return ret;
 119 }
 120
 121 /*
 122  * Called under down_write(mmap_sem).
 123  */
 124
 125 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 126 static unsigned long
 127 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 128                 unsigned long len, unsigned long pgoff, unsigned long flags)
 129 {
 130         struct mm_struct *mm = current->mm;
 131         struct vm_area_struct *vma;
 132         unsigned long start_addr;
 133
 134         if (len & ~HPAGE_MASK)
 135                 return -EINVAL;
 136         if (len > TASK_SIZE)
 137                 return -ENOMEM;
 138
 139         if (flags & MAP_FIXED) {
 140                 if (prepare_hugepage_range(addr, len))
 141                         return -EINVAL;
 142                 return addr;
 143         }
 144
 145         if (addr) {
 146                 addr = ALIGN(addr, HPAGE_SIZE);
 147                 vma = find_vma(mm, addr);
 148                 if (TASK_SIZE - len >= addr &&
 149                     (!vma || addr + len <= vma->vm_start))
 150                         return addr;
 151         }
 152
 153         start_addr = mm->free_area_cache;
 154
 155         if (len <= mm->cached_hole_size)
 156                 start_addr = TASK_UNMAPPED_BASE;
 157
 158 full_search:
 159         addr = ALIGN(start_addr, HPAGE_SIZE);
 160
 161         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 162                 /* At this point:  (!vma || addr < vma->vm_end). */
 163                 if (TASK_SIZE - len < addr) {
 164                         /*
 165                          * Start a new search - just in case we missed
 166                          * some holes.
 167                          */
 168                         if (start_addr != TASK_UNMAPPED_BASE) {
 169                                 start_addr = TASK_UNMAPPED_BASE;
 170                                 goto full_search;
 171                         }
 172                         return -ENOMEM;
 173                 }
 174
 175                 if (!vma || addr + len <= vma->vm_start)
 176                         return addr;
 177                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
 178         }
 179 }
 180 #endif
 181
 182 static int
 183 hugetlbfs_read_actor(struct page *page, unsigned long offset,
 184                         char __user *buf, unsigned long count,
 185                         unsigned long size)
 186 {
 187         char *kaddr;
 188         unsigned long left, copied = 0;
 189         int i, chunksize;
 190
 191         if (size > count)
 192                 size = count;
 193
 194         /* Find which 4k chunk and offset with in that chunk */
 195         i = offset >> PAGE_CACHE_SHIFT;
 196         offset = offset & ~PAGE_CACHE_MASK;
 197
 198         while (size) {
 199                 chunksize = PAGE_CACHE_SIZE;
 200                 if (offset)
 201                         chunksize -= offset;
 202                 if (chunksize > size)
 203                         chunksize = size;
 204                 kaddr = kmap(&page[i]);
 205                 left = __copy_to_user(buf, kaddr + offset, chunksize);
 206                 kunmap(&page[i]);
 207                 if (left) {
 208                         copied += (chunksize - left);
 209                         break;
 210                 }
 211                 offset = 0;
 212                 size -= chunksize;
 213                 buf += chunksize;
 214                 copied += chunksize;
 215                 i++;
 216         }
 217         return copied ? copied : -EFAULT;
 218 }
 219
 220 /*
 221  * Support for read() - Find the page attached to f_mapping and copy out the
 222  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
 223  * since it has PAGE_CACHE_SIZE assumptions.
 224  */
 225 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
 226                               size_t len, loff_t *ppos)
 227 {
 228         struct address_space *mapping = filp->f_mapping;
 229         struct inode *inode = mapping->host;
 230         unsigned long index = *ppos >> HPAGE_SHIFT;
 231         unsigned long offset = *ppos & ~HPAGE_MASK;
 232         unsigned long end_index;
 233         loff_t isize;
 234         ssize_t retval = 0;
 235
 236         mutex_lock(&inode->i_mutex);
 237
 238         /* validate length */
 239         if (len == 0)
 240                 goto out;
 241
 242         isize = i_size_read(inode);
 243         if (!isize)
 244                 goto out;
 245
 246         end_index = (isize - 1) >> HPAGE_SHIFT;
 247         for (;;) {
 248                 struct page *page;
 249                 int nr, ret;
 250
 251                 /* nr is the maximum number of bytes to copy from this page */
 252                 nr = HPAGE_SIZE;
 253                 if (index >= end_index) {
 254                         if (index > end_index)
 255                                 goto out;
 256                         nr = ((isize - 1) & ~HPAGE_MASK) + 1;
 257                         if (nr <= offset) {
 258                                 goto out;
 259                         }
 260                 }
 261                 nr = nr - offset;
 262
 263                 /* Find the page */
 264                 page = find_get_page(mapping, index);
 265                 if (unlikely(page == NULL)) {
 266                         /*
 267                          * We have a HOLE, zero out the user-buffer for the
 268                          * length of the hole or request.
 269                          */
 270                         ret = len < nr ? len : nr;
 271                         if (clear_user(buf, ret))
 272                                 ret = -EFAULT;
 273                 } else {
 274                         /*
 275                          * We have the page, copy it to user space buffer.
 276                          */
 277                         ret = hugetlbfs_read_actor(page, offset, buf, len, nr);
 278                 }
 279                 if (ret < 0) {
 280                         if (retval == 0)
 281                                 retval = ret;
 282                         if (page)
 283                                 page_cache_release(page);
 284                         goto out;
 285                 }
 286
 287                 offset += ret;
 288                 retval += ret;
 289                 len -= ret;
 290                 index += offset >> HPAGE_SHIFT;
 291                 offset &= ~HPAGE_MASK;
 292
 293                 if (page)
 294                         page_cache_release(page);
 295
 296                 /* short read or no more work */
 297                 if ((ret != nr) || (len == 0))
 298                         break;
 299         }
 300 out:
 301         *ppos = ((loff_t)index << HPAGE_SHIFT) + offset;
 302         mutex_unlock(&inode->i_mutex);
 303         return retval;
 304 }
 305
 306 /*
 307  * Read a page. Again trivial. If it didn't already exist
 308  * in the page cache, it is zero-filled.
 309  */
 310 static int hugetlbfs_readpage(struct file *file, struct page * page)
 311 {
 312         unlock_page(page);
 313         return -EINVAL;
 314 }
 315
 316 static int hugetlbfs_write_begin(struct file *file,
 317                         struct address_space *mapping,
 318                         loff_t pos, unsigned len, unsigned flags,
 319                         struct page **pagep, void **fsdata)
 320 {
 321         return -EINVAL;
 322 }
 323
 324 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 325                         loff_t pos, unsigned len, unsigned copied,
 326                         struct page *page, void *fsdata)
 327 {
 328         BUG();
 329         return -EINVAL;
 330 }
 331
 332 static void truncate_huge_page(struct page *page)
 333 {
 334         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
 335         ClearPageUptodate(page);
 336         remove_from_page_cache(page);
 337         put_page(page);
 338 }
 339
 340 static void truncate_hugepages(struct inode *inode, loff_t lstart)
 341 {
 342         struct address_space *mapping = &inode->i_data;
 343         const pgoff_t start = lstart >> HPAGE_SHIFT;
 344         struct pagevec pvec;
 345         pgoff_t next;
 346         int i, freed = 0;
 347
 348         pagevec_init(&pvec, 0);
 349         next = start;
 350         while (1) {
 351                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 352                         if (next == start)
 353                                 break;
 354                         next = start;
 355                         continue;
 356                 }
 357
 358                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 359                         struct page *page = pvec.pages[i];
 360
 361                         lock_page(page);
 362                         if (page->index > next)
 363                                 next = page->index;
 364                         ++next;
 365                         truncate_huge_page(page);
 366                         unlock_page(page);
 367                         freed++;
 368                 }
 369                 huge_pagevec_release(&pvec);
 370         }
 371         BUG_ON(!lstart && mapping->nrpages);
 372         hugetlb_unreserve_pages(inode, start, freed);
 373 }
 374
 375 static void hugetlbfs_delete_inode(struct inode *inode)
 376 {
 377         truncate_hugepages(inode, 0);
 378         clear_inode(inode);
 379 }
 380
 381 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
 382 {
 383         struct super_block *sb = inode->i_sb;
 384
 385         if (!hlist_unhashed(&inode->i_hash)) {
 386                 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
 387                         list_move(&inode->i_list, &inode_unused);
 388                 inodes_stat.nr_unused++;
 389                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
 390                         spin_unlock(&inode_lock);
 391                         return;
 392                 }
 393                 inode->i_state |= I_WILL_FREE;
 394                 spin_unlock(&inode_lock);
 395                 /*
 396                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
 397                  * in our backing_dev_info.
 398                  */
 399                 write_inode_now(inode, 1);
 400                 spin_lock(&inode_lock);
 401                 inode->i_state &= ~I_WILL_FREE;
 402                 inodes_stat.nr_unused--;
 403                 hlist_del_init(&inode->i_hash);
 404         }
 405         list_del_init(&inode->i_list);
 406         list_del_init(&inode->i_sb_list);
 407         inode->i_state |= I_FREEING;
 408         inodes_stat.nr_inodes--;
 409         spin_unlock(&inode_lock);
 410         truncate_hugepages(inode, 0);
 411         clear_inode(inode);
 412         destroy_inode(inode);
 413 }
 414
 415 static void hugetlbfs_drop_inode(struct inode *inode)
 416 {
 417         if (!inode->i_nlink)
 418                 generic_delete_inode(inode);
 419         else
 420                 hugetlbfs_forget_inode(inode);
 421 }
 422
 423 static inline void
 424 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
 425 {
 426         struct vm_area_struct *vma;
 427         struct prio_tree_iter iter;
 428
 429         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
 430                 unsigned long v_offset;
 431
 432                 /*
 433                  * Can the expression below overflow on 32-bit arches?
 434                  * No, because the prio_tree returns us only those vmas
 435                  * which overlap the truncated area starting at pgoff,
 436                  * and no vma on a 32-bit arch can span beyond the 4GB.
 437                  */
 438                 if (vma->vm_pgoff < pgoff)
 439                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 440                 else
 441                         v_offset = 0;
 442
 443                 __unmap_hugepage_range(vma,
 444                                 vma->vm_start + v_offset, vma->vm_end);
 445         }
 446 }
 447
 448 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 449 {
 450         pgoff_t pgoff;
 451         struct address_space *mapping = inode->i_mapping;
 452
 453         BUG_ON(offset & ~HPAGE_MASK);
 454         pgoff = offset >> PAGE_SHIFT;
 455
 456         i_size_write(inode, offset);
 457         spin_lock(&mapping->i_mmap_lock);
 458         if (!prio_tree_empty(&mapping->i_mmap))
 459                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 460         spin_unlock(&mapping->i_mmap_lock);
 461         truncate_hugepages(inode, offset);
 462         return 0;
 463 }
 464
 465 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 466 {
 467         struct inode *inode = dentry->d_inode;
 468         int error;
 469         unsigned int ia_valid = attr->ia_valid;
 470
 471         BUG_ON(!inode);
 472
 473         error = inode_change_ok(inode, attr);
 474         if (error)
 475                 goto out;
 476
 477         if (ia_valid & ATTR_SIZE) {
 478                 error = -EINVAL;
 479                 if (!(attr->ia_size & ~HPAGE_MASK))
 480                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 481                 if (error)
 482                         goto out;
 483                 attr->ia_valid &= ~ATTR_SIZE;
 484         }
 485         error = inode_setattr(inode, attr);
 486 out:
 487         return error;
 488 }
 489
 490 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 491                                         gid_t gid, int mode, dev_t dev)
 492 {
 493         struct inode *inode;
 494
 495         inode = new_inode(sb);
 496         if (inode) {
 497                 struct hugetlbfs_inode_info *info;
 498                 inode->i_mode = mode;
 499                 inode->i_uid = uid;
 500                 inode->i_gid = gid;
 501                 inode->i_blocks = 0;
 502                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 503                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 504                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 505                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
 506                 info = HUGETLBFS_I(inode);
 507                 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
 508                 switch (mode & S_IFMT) {
 509                 default:
 510                         init_special_inode(inode, mode, dev);
 511                         break;
 512                 case S_IFREG:
 513                         inode->i_op = &hugetlbfs_inode_operations;
 514                         inode->i_fop = &hugetlbfs_file_operations;
 515                         break;
 516                 case S_IFDIR:
 517                         inode->i_op = &hugetlbfs_dir_inode_operations;
 518                         inode->i_fop = &simple_dir_operations;
 519
 520                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 521                         inc_nlink(inode);
 522                         break;
 523                 case S_IFLNK:
 524                         inode->i_op = &page_symlink_inode_operations;
 525                         break;
 526                 }
 527         }
 528         return inode;
 529 }
 530
 531 /*
 532  * File creation. Allocate an inode, and we're done..
 533  */
 534 static int hugetlbfs_mknod(struct inode *dir,
 535                         struct dentry *dentry, int mode, dev_t dev)
 536 {
 537         struct inode *inode;
 538         int error = -ENOSPC;
 539         gid_t gid;
 540
 541         if (dir->i_mode & S_ISGID) {
 542                 gid = dir->i_gid;
 543                 if (S_ISDIR(mode))
 544                         mode |= S_ISGID;
 545         } else {
 546                 gid = current->fsgid;
 547         }
 548         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 549         if (inode) {
 550                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 551                 d_instantiate(dentry, inode);
 552                 dget(dentry);   /* Extra count - pin the dentry in core */
 553                 error = 0;
 554         }
 555         return error;
 556 }
 557
 558 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 559 {
 560         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 561         if (!retval)
 562                 inc_nlink(dir);
 563         return retval;
 564 }
 565
 566 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 567 {
 568         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 569 }
 570
 571 static int hugetlbfs_symlink(struct inode *dir,
 572                         struct dentry *dentry, const char *symname)
 573 {
 574         struct inode *inode;
 575         int error = -ENOSPC;
 576         gid_t gid;
 577
 578         if (dir->i_mode & S_ISGID)
 579                 gid = dir->i_gid;
 580         else
 581                 gid = current->fsgid;
 582
 583         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 584                                         gid, S_IFLNK|S_IRWXUGO, 0);
 585         if (inode) {
 586                 int l = strlen(symname)+1;
 587                 error = page_symlink(inode, symname, l);
 588                 if (!error) {
 589                         d_instantiate(dentry, inode);
 590                         dget(dentry);
 591                 } else
 592                         iput(inode);
 593         }
 594         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 595
 596         return error;
 597 }
 598
 599 /*
 600  * mark the head page dirty
 601  */
 602 static int hugetlbfs_set_page_dirty(struct page *page)
 603 {
 604         struct page *head = compound_head(page);
 605
 606         SetPageDirty(head);
 607         return 0;
 608 }
 609
 610 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 611 {
 612         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 613
 614         buf->f_type = HUGETLBFS_MAGIC;
 615         buf->f_bsize = HPAGE_SIZE;
 616         if (sbinfo) {
 617                 spin_lock(&sbinfo->stat_lock);
 618                 /* If no limits set, just report 0 for max/free/used
 619                  * blocks, like simple_statfs() */
 620                 if (sbinfo->max_blocks >= 0) {
 621                         buf->f_blocks = sbinfo->max_blocks;
 622                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 623                         buf->f_files = sbinfo->max_inodes;
 624                         buf->f_ffree = sbinfo->free_inodes;
 625                 }
 626                 spin_unlock(&sbinfo->stat_lock);
 627         }
 628         buf->f_namelen = NAME_MAX;
 629         return 0;
 630 }
 631
 632 static void hugetlbfs_put_super(struct super_block *sb)
 633 {
 634         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 635
 636         if (sbi) {
 637                 sb->s_fs_info = NULL;
 638                 kfree(sbi);
 639         }
 640 }
 641
 642 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 643 {
 644         if (sbinfo->free_inodes >= 0) {
 645                 spin_lock(&sbinfo->stat_lock);
 646                 if (unlikely(!sbinfo->free_inodes)) {
 647                         spin_unlock(&sbinfo->stat_lock);
 648                         return 0;
 649                 }
 650                 sbinfo->free_inodes--;
 651                 spin_unlock(&sbinfo->stat_lock);
 652         }
 653
 654         return 1;
 655 }
 656
 657 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 658 {
 659         if (sbinfo->free_inodes >= 0) {
 660                 spin_lock(&sbinfo->stat_lock);
 661                 sbinfo->free_inodes++;
 662                 spin_unlock(&sbinfo->stat_lock);
 663         }
 664 }
 665
 666
 667 static struct kmem_cache *hugetlbfs_inode_cachep;
 668
 669 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 670 {
 671         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 672         struct hugetlbfs_inode_info *p;
 673
 674         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 675                 return NULL;
 676         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 677         if (unlikely(!p)) {
 678                 hugetlbfs_inc_free_inodes(sbinfo);
 679                 return NULL;
 680         }
 681         return &p->vfs_inode;
 682 }
 683
 684 static void hugetlbfs_destroy_inode(struct inode *inode)
 685 {
 686         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 687         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 688         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 689 }
 690
 691 static const struct address_space_operations hugetlbfs_aops = {
 692         .readpage       = hugetlbfs_readpage,
 693         .write_begin    = hugetlbfs_write_begin,
 694         .write_end      = hugetlbfs_write_end,
 695         .set_page_dirty = hugetlbfs_set_page_dirty,
 696 };
 697
 698
 699 static void init_once(struct kmem_cache *cachep, void *foo)
 700 {
 701         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 702
 703         inode_init_once(&ei->vfs_inode);
 704 }
 705
 706 const struct file_operations hugetlbfs_file_operations = {
 707         .read                   = hugetlbfs_read,
 708         .mmap                   = hugetlbfs_file_mmap,
 709         .fsync                  = simple_sync_file,
 710         .get_unmapped_area      = hugetlb_get_unmapped_area,
 711 };
 712
 713 static const struct inode_operations hugetlbfs_dir_inode_operations = {
 714         .create         = hugetlbfs_create,
 715         .lookup         = simple_lookup,
 716         .link           = simple_link,
 717         .unlink         = simple_unlink,
 718         .symlink        = hugetlbfs_symlink,
 719         .mkdir          = hugetlbfs_mkdir,
 720         .rmdir          = simple_rmdir,
 721         .mknod          = hugetlbfs_mknod,
 722         .rename         = simple_rename,
 723         .setattr        = hugetlbfs_setattr,
 724 };
 725
 726 static const struct inode_operations hugetlbfs_inode_operations = {
 727         .setattr        = hugetlbfs_setattr,
 728 };
 729
 730 static const struct super_operations hugetlbfs_ops = {
 731         .alloc_inode    = hugetlbfs_alloc_inode,
 732         .destroy_inode  = hugetlbfs_destroy_inode,
 733         .statfs         = hugetlbfs_statfs,
 734         .delete_inode   = hugetlbfs_delete_inode,
 735         .drop_inode     = hugetlbfs_drop_inode,
 736         .put_super      = hugetlbfs_put_super,
 737 };
 738
 739 static int
 740 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 741 {
 742         char *p, *rest;
 743         substring_t args[MAX_OPT_ARGS];
 744         int option;
 745
 746         if (!options)
 747                 return 0;
 748
 749         while ((p = strsep(&options, ",")) != NULL) {
 750                 int token;
 751                 if (!*p)
 752                         continue;
 753
 754                 token = match_token(p, tokens, args);
 755                 switch (token) {
 756                 case Opt_uid:
 757                         if (match_int(&args[0], &option))
 758                                 goto bad_val;
 759                         pconfig->uid = option;
 760                         break;
 761
 762                 case Opt_gid:
 763                         if (match_int(&args[0], &option))
 764                                 goto bad_val;
 765                         pconfig->gid = option;
 766                         break;
 767
 768                 case Opt_mode:
 769                         if (match_octal(&args[0], &option))
 770                                 goto bad_val;
 771                         pconfig->mode = option & 01777U;
 772                         break;
 773
 774                 case Opt_size: {
 775                         unsigned long long size;
 776                         /* memparse() will accept a K/M/G without a digit */
 777                         if (!isdigit(*args[0].from))
 778                                 goto bad_val;
 779                         size = memparse(args[0].from, &rest);
 780                         if (*rest == '%') {
 781                                 size <<= HPAGE_SHIFT;
 782                                 size *= max_huge_pages;
 783                                 do_div(size, 100);
 784                         }
 785                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 786                         break;
 787                 }
 788
 789                 case Opt_nr_inodes:
 790                         /* memparse() will accept a K/M/G without a digit */
 791                         if (!isdigit(*args[0].from))
 792                                 goto bad_val;
 793                         pconfig->nr_inodes = memparse(args[0].from, &rest);
 794                         break;
 795
 796                 default:
 797                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
 798                                  p);
 799                         return -EINVAL;
 800                         break;
 801                 }
 802         }
 803         return 0;
 804
 805 bad_val:
 806         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
 807                args[0].from, p);
 808         return 1;
 809 }
 810
 811 static int
 812 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 813 {
 814         struct inode * inode;
 815         struct dentry * root;
 816         int ret;
 817         struct hugetlbfs_config config;
 818         struct hugetlbfs_sb_info *sbinfo;
 819
 820         config.nr_blocks = -1; /* No limit on size by default */
 821         config.nr_inodes = -1; /* No limit on number of inodes by default */
 822         config.uid = current->fsuid;
 823         config.gid = current->fsgid;
 824         config.mode = 0755;
 825         ret = hugetlbfs_parse_options(data, &config);
 826         if (ret)
 827                 return ret;
 828
 829         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 830         if (!sbinfo)
 831                 return -ENOMEM;
 832         sb->s_fs_info = sbinfo;
 833         spin_lock_init(&sbinfo->stat_lock);
 834         sbinfo->max_blocks = config.nr_blocks;
 835         sbinfo->free_blocks = config.nr_blocks;
 836         sbinfo->max_inodes = config.nr_inodes;
 837         sbinfo->free_inodes = config.nr_inodes;
 838         sb->s_maxbytes = MAX_LFS_FILESIZE;
 839         sb->s_blocksize = HPAGE_SIZE;
 840         sb->s_blocksize_bits = HPAGE_SHIFT;
 841         sb->s_magic = HUGETLBFS_MAGIC;
 842         sb->s_op = &hugetlbfs_ops;
 843         sb->s_time_gran = 1;
 844         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 845                                         S_IFDIR | config.mode, 0);
 846         if (!inode)
 847                 goto out_free;
 848
 849         root = d_alloc_root(inode);
 850         if (!root) {
 851                 iput(inode);
 852                 goto out_free;
 853         }
 854         sb->s_root = root;
 855         return 0;
 856 out_free:
 857         kfree(sbinfo);
 858         return -ENOMEM;
 859 }
 860
 861 int hugetlb_get_quota(struct address_space *mapping, long delta)
 862 {
 863         int ret = 0;
 864         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 865
 866         if (sbinfo->free_blocks > -1) {
 867                 spin_lock(&sbinfo->stat_lock);
 868                 if (sbinfo->free_blocks - delta >= 0)
 869                         sbinfo->free_blocks -= delta;
 870                 else
 871                         ret = -ENOMEM;
 872                 spin_unlock(&sbinfo->stat_lock);
 873         }
 874
 875         return ret;
 876 }
 877
 878 void hugetlb_put_quota(struct address_space *mapping, long delta)
 879 {
 880         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 881
 882         if (sbinfo->free_blocks > -1) {
 883                 spin_lock(&sbinfo->stat_lock);
 884                 sbinfo->free_blocks += delta;
 885                 spin_unlock(&sbinfo->stat_lock);
 886         }
 887 }
 888
 889 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
 890         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 891 {
 892         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
 893 }
 894
 895 static struct file_system_type hugetlbfs_fs_type = {
 896         .name           = "hugetlbfs",
 897         .get_sb         = hugetlbfs_get_sb,
 898         .kill_sb        = kill_litter_super,
 899 };
 900
 901 static struct vfsmount *hugetlbfs_vfsmount;
 902
 903 static int can_do_hugetlb_shm(void)
 904 {
 905         return likely(capable(CAP_IPC_LOCK) ||
 906                         in_group_p(sysctl_hugetlb_shm_group) ||
 907                         can_do_mlock());
 908 }
 909
 910 struct file *hugetlb_file_setup(const char *name, size_t size)
 911 {
 912         int error = -ENOMEM;
 913         struct file *file;
 914         struct inode *inode;
 915         struct dentry *dentry, *root;
 916         struct qstr quick_string;
 917
 918         if (!hugetlbfs_vfsmount)
 919                 return ERR_PTR(-ENOENT);
 920
 921         if (!can_do_hugetlb_shm())
 922                 return ERR_PTR(-EPERM);
 923
 924         if (!user_shm_lock(size, current->user))
 925                 return ERR_PTR(-ENOMEM);
 926
 927         root = hugetlbfs_vfsmount->mnt_root;
 928         quick_string.name = name;
 929         quick_string.len = strlen(quick_string.name);
 930         quick_string.hash = 0;
 931         dentry = d_alloc(root, &quick_string);
 932         if (!dentry)
 933                 goto out_shm_unlock;
 934
 935         error = -ENOSPC;
 936         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
 937                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
 938         if (!inode)
 939                 goto out_dentry;
 940
 941         error = -ENOMEM;
 942         if (hugetlb_reserve_pages(inode, 0, size >> HPAGE_SHIFT))
 943                 goto out_inode;
 944
 945         d_instantiate(dentry, inode);
 946         inode->i_size = size;
 947         inode->i_nlink = 0;
 948
 949         error = -ENFILE;
 950         file = alloc_file(hugetlbfs_vfsmount, dentry,
 951                         FMODE_WRITE | FMODE_READ,
 952                         &hugetlbfs_file_operations);
 953         if (!file)
 954                 goto out_inode;
 955
 956         return file;
 957
 958 out_inode:
 959         iput(inode);
 960 out_dentry:
 961         dput(dentry);
 962 out_shm_unlock:
 963         user_shm_unlock(size, current->user);
 964         return ERR_PTR(error);
 965 }
 966
 967 static int __init init_hugetlbfs_fs(void)
 968 {
 969         int error;
 970         struct vfsmount *vfsmount;
 971
 972         error = bdi_init(&hugetlbfs_backing_dev_info);
 973         if (error)
 974                 return error;
 975
 976         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
 977                                         sizeof(struct hugetlbfs_inode_info),
 978                                         0, 0, init_once);
 979         if (hugetlbfs_inode_cachep == NULL)
 980                 goto out2;
 981
 982         error = register_filesystem(&hugetlbfs_fs_type);
 983         if (error)
 984                 goto out;
 985
 986         vfsmount = kern_mount(&hugetlbfs_fs_type);
 987
 988         if (!IS_ERR(vfsmount)) {
 989                 hugetlbfs_vfsmount = vfsmount;
 990                 return 0;
 991         }
 992
 993         error = PTR_ERR(vfsmount);
 994
 995  out:
 996         if (error)
 997                 kmem_cache_destroy(hugetlbfs_inode_cachep);
 998  out2:
 999         bdi_destroy(&hugetlbfs_backing_dev_info);
1000         return error;
1001 }
1002
1003 static void __exit exit_hugetlbfs_fs(void)
1004 {
1005         kmem_cache_destroy(hugetlbfs_inode_cachep);
1006         unregister_filesystem(&hugetlbfs_fs_type);
1007         bdi_destroy(&hugetlbfs_backing_dev_info);
1008 }
1009
1010 module_init(init_hugetlbfs_fs)
1011 module_exit(exit_hugetlbfs_fs)
1012
1013 MODULE_LICENSE("GPL");