SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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_AND_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         struct hstate *h = hstate_file(file);
  84
  85         /*
  86          * vma address alignment (but not the pgoff alignment) has
  87          * already been checked by prepare_hugepage_range.  If you add
  88          * any error returns here, do so after setting VM_HUGETLB, so
  89          * is_vm_hugetlb_page tests below unmap_region go the right
  90          * way when do_mmap_pgoff unwinds (may be important on powerpc
  91          * and ia64).
  92          */
  93         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
  94         vma->vm_ops = &hugetlb_vm_ops;
  95
  96         if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
  97                 return -EINVAL;
  98
  99         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 100
 101         mutex_lock(&inode->i_mutex);
 102         file_accessed(file);
 103
 104         ret = -ENOMEM;
 105         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 106
 107         if (hugetlb_reserve_pages(inode,
 108                                 vma->vm_pgoff >> huge_page_order(h),
 109                                 len >> huge_page_shift(h), vma))
 110                 goto out;
 111
 112         ret = 0;
 113         hugetlb_prefault_arch_hook(vma->vm_mm);
 114         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 115                 inode->i_size = len;
 116 out:
 117         mutex_unlock(&inode->i_mutex);
 118
 119         return ret;
 120 }
 121
 122 /*
 123  * Called under down_write(mmap_sem).
 124  */
 125
 126 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 127 static unsigned long
 128 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 129                 unsigned long len, unsigned long pgoff, unsigned long flags)
 130 {
 131         struct mm_struct *mm = current->mm;
 132         struct vm_area_struct *vma;
 133         unsigned long start_addr;
 134         struct hstate *h = hstate_file(file);
 135
 136         if (len & ~huge_page_mask(h))
 137                 return -EINVAL;
 138         if (len > TASK_SIZE)
 139                 return -ENOMEM;
 140
 141         if (flags & MAP_FIXED) {
 142                 if (prepare_hugepage_range(file, addr, len))
 143                         return -EINVAL;
 144                 return addr;
 145         }
 146
 147         if (addr) {
 148                 addr = ALIGN(addr, huge_page_size(h));
 149                 vma = find_vma(mm, addr);
 150                 if (TASK_SIZE - len >= addr &&
 151                     (!vma || addr + len <= vma->vm_start))
 152                         return addr;
 153         }
 154
 155         start_addr = mm->free_area_cache;
 156
 157         if (len <= mm->cached_hole_size)
 158                 start_addr = TASK_UNMAPPED_BASE;
 159
 160 full_search:
 161         addr = ALIGN(start_addr, huge_page_size(h));
 162
 163         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 164                 /* At this point:  (!vma || addr < vma->vm_end). */
 165                 if (TASK_SIZE - len < addr) {
 166                         /*
 167                          * Start a new search - just in case we missed
 168                          * some holes.
 169                          */
 170                         if (start_addr != TASK_UNMAPPED_BASE) {
 171                                 start_addr = TASK_UNMAPPED_BASE;
 172                                 goto full_search;
 173                         }
 174                         return -ENOMEM;
 175                 }
 176
 177                 if (!vma || addr + len <= vma->vm_start)
 178                         return addr;
 179                 addr = ALIGN(vma->vm_end, huge_page_size(h));
 180         }
 181 }
 182 #endif
 183
 184 static int
 185 hugetlbfs_read_actor(struct page *page, unsigned long offset,
 186                         char __user *buf, unsigned long count,
 187                         unsigned long size)
 188 {
 189         char *kaddr;
 190         unsigned long left, copied = 0;
 191         int i, chunksize;
 192
 193         if (size > count)
 194                 size = count;
 195
 196         /* Find which 4k chunk and offset with in that chunk */
 197         i = offset >> PAGE_CACHE_SHIFT;
 198         offset = offset & ~PAGE_CACHE_MASK;
 199
 200         while (size) {
 201                 chunksize = PAGE_CACHE_SIZE;
 202                 if (offset)
 203                         chunksize -= offset;
 204                 if (chunksize > size)
 205                         chunksize = size;
 206                 kaddr = kmap(&page[i]);
 207                 left = __copy_to_user(buf, kaddr + offset, chunksize);
 208                 kunmap(&page[i]);
 209                 if (left) {
 210                         copied += (chunksize - left);
 211                         break;
 212                 }
 213                 offset = 0;
 214                 size -= chunksize;
 215                 buf += chunksize;
 216                 copied += chunksize;
 217                 i++;
 218         }
 219         return copied ? copied : -EFAULT;
 220 }
 221
 222 /*
 223  * Support for read() - Find the page attached to f_mapping and copy out the
 224  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
 225  * since it has PAGE_CACHE_SIZE assumptions.
 226  */
 227 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
 228                               size_t len, loff_t *ppos)
 229 {
 230         struct hstate *h = hstate_file(filp);
 231         struct address_space *mapping = filp->f_mapping;
 232         struct inode *inode = mapping->host;
 233         unsigned long index = *ppos >> huge_page_shift(h);
 234         unsigned long offset = *ppos & ~huge_page_mask(h);
 235         unsigned long end_index;
 236         loff_t isize;
 237         ssize_t retval = 0;
 238
 239         mutex_lock(&inode->i_mutex);
 240
 241         /* validate length */
 242         if (len == 0)
 243                 goto out;
 244
 245         isize = i_size_read(inode);
 246         if (!isize)
 247                 goto out;
 248
 249         end_index = (isize - 1) >> huge_page_shift(h);
 250         for (;;) {
 251                 struct page *page;
 252                 unsigned long nr, ret;
 253
 254                 /* nr is the maximum number of bytes to copy from this page */
 255                 nr = huge_page_size(h);
 256                 if (index >= end_index) {
 257                         if (index > end_index)
 258                                 goto out;
 259                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
 260                         if (nr <= offset) {
 261                                 goto out;
 262                         }
 263                 }
 264                 nr = nr - offset;
 265
 266                 /* Find the page */
 267                 page = find_get_page(mapping, index);
 268                 if (unlikely(page == NULL)) {
 269                         /*
 270                          * We have a HOLE, zero out the user-buffer for the
 271                          * length of the hole or request.
 272                          */
 273                         ret = len < nr ? len : nr;
 274                         if (clear_user(buf, ret))
 275                                 ret = -EFAULT;
 276                 } else {
 277                         /*
 278                          * We have the page, copy it to user space buffer.
 279                          */
 280                         ret = hugetlbfs_read_actor(page, offset, buf, len, nr);
 281                 }
 282                 if (ret < 0) {
 283                         if (retval == 0)
 284                                 retval = ret;
 285                         if (page)
 286                                 page_cache_release(page);
 287                         goto out;
 288                 }
 289
 290                 offset += ret;
 291                 retval += ret;
 292                 len -= ret;
 293                 index += offset >> huge_page_shift(h);
 294                 offset &= ~huge_page_mask(h);
 295
 296                 if (page)
 297                         page_cache_release(page);
 298
 299                 /* short read or no more work */
 300                 if ((ret != nr) || (len == 0))
 301                         break;
 302         }
 303 out:
 304         *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
 305         mutex_unlock(&inode->i_mutex);
 306         return retval;
 307 }
 308
 309 /*
 310  * Read a page. Again trivial. If it didn't already exist
 311  * in the page cache, it is zero-filled.
 312  */
 313 static int hugetlbfs_readpage(struct file *file, struct page * page)
 314 {
 315         unlock_page(page);
 316         return -EINVAL;
 317 }
 318
 319 static int hugetlbfs_write_begin(struct file *file,
 320                         struct address_space *mapping,
 321                         loff_t pos, unsigned len, unsigned flags,
 322                         struct page **pagep, void **fsdata)
 323 {
 324         return -EINVAL;
 325 }
 326
 327 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 328                         loff_t pos, unsigned len, unsigned copied,
 329                         struct page *page, void *fsdata)
 330 {
 331         BUG();
 332         return -EINVAL;
 333 }
 334
 335 static void truncate_huge_page(struct page *page)
 336 {
 337         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
 338         ClearPageUptodate(page);
 339         remove_from_page_cache(page);
 340         put_page(page);
 341 }
 342
 343 static void truncate_hugepages(struct inode *inode, loff_t lstart)
 344 {
 345         struct hstate *h = hstate_inode(inode);
 346         struct address_space *mapping = &inode->i_data;
 347         const pgoff_t start = lstart >> huge_page_shift(h);
 348         struct pagevec pvec;
 349         pgoff_t next;
 350         int i, freed = 0;
 351
 352         pagevec_init(&pvec, 0);
 353         next = start;
 354         while (1) {
 355                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 356                         if (next == start)
 357                                 break;
 358                         next = start;
 359                         continue;
 360                 }
 361
 362                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 363                         struct page *page = pvec.pages[i];
 364
 365                         lock_page(page);
 366                         if (page->index > next)
 367                                 next = page->index;
 368                         ++next;
 369                         truncate_huge_page(page);
 370                         unlock_page(page);
 371                         freed++;
 372                 }
 373                 huge_pagevec_release(&pvec);
 374         }
 375         BUG_ON(!lstart && mapping->nrpages);
 376         hugetlb_unreserve_pages(inode, start, freed);
 377 }
 378
 379 static void hugetlbfs_delete_inode(struct inode *inode)
 380 {
 381         truncate_hugepages(inode, 0);
 382         clear_inode(inode);
 383 }
 384
 385 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
 386 {
 387         struct super_block *sb = inode->i_sb;
 388
 389         if (!hlist_unhashed(&inode->i_hash)) {
 390                 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
 391                         list_move(&inode->i_list, &inode_unused);
 392                 inodes_stat.nr_unused++;
 393                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
 394                         spin_unlock(&inode_lock);
 395                         return;
 396                 }
 397                 inode->i_state |= I_WILL_FREE;
 398                 spin_unlock(&inode_lock);
 399                 /*
 400                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
 401                  * in our backing_dev_info.
 402                  */
 403                 write_inode_now(inode, 1);
 404                 spin_lock(&inode_lock);
 405                 inode->i_state &= ~I_WILL_FREE;
 406                 inodes_stat.nr_unused--;
 407                 hlist_del_init(&inode->i_hash);
 408         }
 409         list_del_init(&inode->i_list);
 410         list_del_init(&inode->i_sb_list);
 411         inode->i_state |= I_FREEING;
 412         inodes_stat.nr_inodes--;
 413         spin_unlock(&inode_lock);
 414         truncate_hugepages(inode, 0);
 415         clear_inode(inode);
 416         destroy_inode(inode);
 417 }
 418
 419 static void hugetlbfs_drop_inode(struct inode *inode)
 420 {
 421         if (!inode->i_nlink)
 422                 generic_delete_inode(inode);
 423         else
 424                 hugetlbfs_forget_inode(inode);
 425 }
 426
 427 static inline void
 428 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
 429 {
 430         struct vm_area_struct *vma;
 431         struct prio_tree_iter iter;
 432
 433         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
 434                 unsigned long v_offset;
 435
 436                 /*
 437                  * Can the expression below overflow on 32-bit arches?
 438                  * No, because the prio_tree returns us only those vmas
 439                  * which overlap the truncated area starting at pgoff,
 440                  * and no vma on a 32-bit arch can span beyond the 4GB.
 441                  */
 442                 if (vma->vm_pgoff < pgoff)
 443                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 444                 else
 445                         v_offset = 0;
 446
 447                 __unmap_hugepage_range(vma,
 448                                 vma->vm_start + v_offset, vma->vm_end, NULL);
 449         }
 450 }
 451
 452 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 453 {
 454         pgoff_t pgoff;
 455         struct address_space *mapping = inode->i_mapping;
 456         struct hstate *h = hstate_inode(inode);
 457
 458         BUG_ON(offset & ~huge_page_mask(h));
 459         pgoff = offset >> PAGE_SHIFT;
 460
 461         i_size_write(inode, offset);
 462         spin_lock(&mapping->i_mmap_lock);
 463         if (!prio_tree_empty(&mapping->i_mmap))
 464                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 465         spin_unlock(&mapping->i_mmap_lock);
 466         truncate_hugepages(inode, offset);
 467         return 0;
 468 }
 469
 470 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 471 {
 472         struct inode *inode = dentry->d_inode;
 473         struct hstate *h = hstate_inode(inode);
 474         int error;
 475         unsigned int ia_valid = attr->ia_valid;
 476
 477         BUG_ON(!inode);
 478
 479         error = inode_change_ok(inode, attr);
 480         if (error)
 481                 goto out;
 482
 483         if (ia_valid & ATTR_SIZE) {
 484                 error = -EINVAL;
 485                 if (!(attr->ia_size & ~huge_page_mask(h)))
 486                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 487                 if (error)
 488                         goto out;
 489                 attr->ia_valid &= ~ATTR_SIZE;
 490         }
 491         error = inode_setattr(inode, attr);
 492 out:
 493         return error;
 494 }
 495
 496 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 497                                         gid_t gid, int mode, dev_t dev)
 498 {
 499         struct inode *inode;
 500
 501         inode = new_inode(sb);
 502         if (inode) {
 503                 struct hugetlbfs_inode_info *info;
 504                 inode->i_mode = mode;
 505                 inode->i_uid = uid;
 506                 inode->i_gid = gid;
 507                 inode->i_blocks = 0;
 508                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 509                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 510                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 511                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
 512                 info = HUGETLBFS_I(inode);
 513                 mpol_shared_policy_init(&info->policy, NULL);
 514                 switch (mode & S_IFMT) {
 515                 default:
 516                         init_special_inode(inode, mode, dev);
 517                         break;
 518                 case S_IFREG:
 519                         inode->i_op = &hugetlbfs_inode_operations;
 520                         inode->i_fop = &hugetlbfs_file_operations;
 521                         break;
 522                 case S_IFDIR:
 523                         inode->i_op = &hugetlbfs_dir_inode_operations;
 524                         inode->i_fop = &simple_dir_operations;
 525
 526                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 527                         inc_nlink(inode);
 528                         break;
 529                 case S_IFLNK:
 530                         inode->i_op = &page_symlink_inode_operations;
 531                         break;
 532                 }
 533         }
 534         return inode;
 535 }
 536
 537 /*
 538  * File creation. Allocate an inode, and we're done..
 539  */
 540 static int hugetlbfs_mknod(struct inode *dir,
 541                         struct dentry *dentry, int mode, dev_t dev)
 542 {
 543         struct inode *inode;
 544         int error = -ENOSPC;
 545         gid_t gid;
 546
 547         if (dir->i_mode & S_ISGID) {
 548                 gid = dir->i_gid;
 549                 if (S_ISDIR(mode))
 550                         mode |= S_ISGID;
 551         } else {
 552                 gid = current->fsgid;
 553         }
 554         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 555         if (inode) {
 556                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 557                 d_instantiate(dentry, inode);
 558                 dget(dentry);   /* Extra count - pin the dentry in core */
 559                 error = 0;
 560         }
 561         return error;
 562 }
 563
 564 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 565 {
 566         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 567         if (!retval)
 568                 inc_nlink(dir);
 569         return retval;
 570 }
 571
 572 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 573 {
 574         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 575 }
 576
 577 static int hugetlbfs_symlink(struct inode *dir,
 578                         struct dentry *dentry, const char *symname)
 579 {
 580         struct inode *inode;
 581         int error = -ENOSPC;
 582         gid_t gid;
 583
 584         if (dir->i_mode & S_ISGID)
 585                 gid = dir->i_gid;
 586         else
 587                 gid = current->fsgid;
 588
 589         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 590                                         gid, S_IFLNK|S_IRWXUGO, 0);
 591         if (inode) {
 592                 int l = strlen(symname)+1;
 593                 error = page_symlink(inode, symname, l);
 594                 if (!error) {
 595                         d_instantiate(dentry, inode);
 596                         dget(dentry);
 597                 } else
 598                         iput(inode);
 599         }
 600         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 601
 602         return error;
 603 }
 604
 605 /*
 606  * mark the head page dirty
 607  */
 608 static int hugetlbfs_set_page_dirty(struct page *page)
 609 {
 610         struct page *head = compound_head(page);
 611
 612         SetPageDirty(head);
 613         return 0;
 614 }
 615
 616 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 617 {
 618         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 619         struct hstate *h = hstate_inode(dentry->d_inode);
 620
 621         buf->f_type = HUGETLBFS_MAGIC;
 622         buf->f_bsize = huge_page_size(h);
 623         if (sbinfo) {
 624                 spin_lock(&sbinfo->stat_lock);
 625                 /* If no limits set, just report 0 for max/free/used
 626                  * blocks, like simple_statfs() */
 627                 if (sbinfo->max_blocks >= 0) {
 628                         buf->f_blocks = sbinfo->max_blocks;
 629                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 630                         buf->f_files = sbinfo->max_inodes;
 631                         buf->f_ffree = sbinfo->free_inodes;
 632                 }
 633                 spin_unlock(&sbinfo->stat_lock);
 634         }
 635         buf->f_namelen = NAME_MAX;
 636         return 0;
 637 }
 638
 639 static void hugetlbfs_put_super(struct super_block *sb)
 640 {
 641         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 642
 643         if (sbi) {
 644                 sb->s_fs_info = NULL;
 645                 kfree(sbi);
 646         }
 647 }
 648
 649 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 650 {
 651         if (sbinfo->free_inodes >= 0) {
 652                 spin_lock(&sbinfo->stat_lock);
 653                 if (unlikely(!sbinfo->free_inodes)) {
 654                         spin_unlock(&sbinfo->stat_lock);
 655                         return 0;
 656                 }
 657                 sbinfo->free_inodes--;
 658                 spin_unlock(&sbinfo->stat_lock);
 659         }
 660
 661         return 1;
 662 }
 663
 664 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 665 {
 666         if (sbinfo->free_inodes >= 0) {
 667                 spin_lock(&sbinfo->stat_lock);
 668                 sbinfo->free_inodes++;
 669                 spin_unlock(&sbinfo->stat_lock);
 670         }
 671 }
 672
 673
 674 static struct kmem_cache *hugetlbfs_inode_cachep;
 675
 676 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 677 {
 678         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 679         struct hugetlbfs_inode_info *p;
 680
 681         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 682                 return NULL;
 683         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 684         if (unlikely(!p)) {
 685                 hugetlbfs_inc_free_inodes(sbinfo);
 686                 return NULL;
 687         }
 688         return &p->vfs_inode;
 689 }
 690
 691 static void hugetlbfs_destroy_inode(struct inode *inode)
 692 {
 693         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 694         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 695         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 696 }
 697
 698 static const struct address_space_operations hugetlbfs_aops = {
 699         .readpage       = hugetlbfs_readpage,
 700         .write_begin    = hugetlbfs_write_begin,
 701         .write_end      = hugetlbfs_write_end,
 702         .set_page_dirty = hugetlbfs_set_page_dirty,
 703 };
 704
 705
 706 static void init_once(struct kmem_cache *cachep, void *foo)
 707 {
 708         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 709
 710         inode_init_once(&ei->vfs_inode);
 711 }
 712
 713 const struct file_operations hugetlbfs_file_operations = {
 714         .read                   = hugetlbfs_read,
 715         .mmap                   = hugetlbfs_file_mmap,
 716         .fsync                  = simple_sync_file,
 717         .get_unmapped_area      = hugetlb_get_unmapped_area,
 718 };
 719
 720 static const struct inode_operations hugetlbfs_dir_inode_operations = {
 721         .create         = hugetlbfs_create,
 722         .lookup         = simple_lookup,
 723         .link           = simple_link,
 724         .unlink         = simple_unlink,
 725         .symlink        = hugetlbfs_symlink,
 726         .mkdir          = hugetlbfs_mkdir,
 727         .rmdir          = simple_rmdir,
 728         .mknod          = hugetlbfs_mknod,
 729         .rename         = simple_rename,
 730         .setattr        = hugetlbfs_setattr,
 731 };
 732
 733 static const struct inode_operations hugetlbfs_inode_operations = {
 734         .setattr        = hugetlbfs_setattr,
 735 };
 736
 737 static const struct super_operations hugetlbfs_ops = {
 738         .alloc_inode    = hugetlbfs_alloc_inode,
 739         .destroy_inode  = hugetlbfs_destroy_inode,
 740         .statfs         = hugetlbfs_statfs,
 741         .delete_inode   = hugetlbfs_delete_inode,
 742         .drop_inode     = hugetlbfs_drop_inode,
 743         .put_super      = hugetlbfs_put_super,
 744         .show_options   = generic_show_options,
 745 };
 746
 747 static int
 748 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 749 {
 750         char *p, *rest;
 751         substring_t args[MAX_OPT_ARGS];
 752         int option;
 753
 754         if (!options)
 755                 return 0;
 756
 757         while ((p = strsep(&options, ",")) != NULL) {
 758                 int token;
 759                 if (!*p)
 760                         continue;
 761
 762                 token = match_token(p, tokens, args);
 763                 switch (token) {
 764                 case Opt_uid:
 765                         if (match_int(&args[0], &option))
 766                                 goto bad_val;
 767                         pconfig->uid = option;
 768                         break;
 769
 770                 case Opt_gid:
 771                         if (match_int(&args[0], &option))
 772                                 goto bad_val;
 773                         pconfig->gid = option;
 774                         break;
 775
 776                 case Opt_mode:
 777                         if (match_octal(&args[0], &option))
 778                                 goto bad_val;
 779                         pconfig->mode = option & 01777U;
 780                         break;
 781
 782                 case Opt_size: {
 783                         unsigned long long size;
 784                         /* memparse() will accept a K/M/G without a digit */
 785                         if (!isdigit(*args[0].from))
 786                                 goto bad_val;
 787                         size = memparse(args[0].from, &rest);
 788                         if (*rest == '%') {
 789                                 size <<= HPAGE_SHIFT;
 790                                 size *= max_huge_pages;
 791                                 do_div(size, 100);
 792                         }
 793                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 794                         break;
 795                 }
 796
 797                 case Opt_nr_inodes:
 798                         /* memparse() will accept a K/M/G without a digit */
 799                         if (!isdigit(*args[0].from))
 800                                 goto bad_val;
 801                         pconfig->nr_inodes = memparse(args[0].from, &rest);
 802                         break;
 803
 804                 default:
 805                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
 806                                  p);
 807                         return -EINVAL;
 808                         break;
 809                 }
 810         }
 811         return 0;
 812
 813 bad_val:
 814         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
 815                args[0].from, p);
 816         return 1;
 817 }
 818
 819 static int
 820 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 821 {
 822         struct inode * inode;
 823         struct dentry * root;
 824         int ret;
 825         struct hugetlbfs_config config;
 826         struct hugetlbfs_sb_info *sbinfo;
 827
 828         save_mount_options(sb, data);
 829
 830         config.nr_blocks = -1; /* No limit on size by default */
 831         config.nr_inodes = -1; /* No limit on number of inodes by default */
 832         config.uid = current->fsuid;
 833         config.gid = current->fsgid;
 834         config.mode = 0755;
 835         ret = hugetlbfs_parse_options(data, &config);
 836         if (ret)
 837                 return ret;
 838
 839         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 840         if (!sbinfo)
 841                 return -ENOMEM;
 842         sb->s_fs_info = sbinfo;
 843         spin_lock_init(&sbinfo->stat_lock);
 844         sbinfo->max_blocks = config.nr_blocks;
 845         sbinfo->free_blocks = config.nr_blocks;
 846         sbinfo->max_inodes = config.nr_inodes;
 847         sbinfo->free_inodes = config.nr_inodes;
 848         sb->s_maxbytes = MAX_LFS_FILESIZE;
 849         sb->s_blocksize = HPAGE_SIZE;
 850         sb->s_blocksize_bits = HPAGE_SHIFT;
 851         sb->s_magic = HUGETLBFS_MAGIC;
 852         sb->s_op = &hugetlbfs_ops;
 853         sb->s_time_gran = 1;
 854         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 855                                         S_IFDIR | config.mode, 0);
 856         if (!inode)
 857                 goto out_free;
 858
 859         root = d_alloc_root(inode);
 860         if (!root) {
 861                 iput(inode);
 862                 goto out_free;
 863         }
 864         sb->s_root = root;
 865         return 0;
 866 out_free:
 867         kfree(sbinfo);
 868         return -ENOMEM;
 869 }
 870
 871 int hugetlb_get_quota(struct address_space *mapping, long delta)
 872 {
 873         int ret = 0;
 874         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 875
 876         if (sbinfo->free_blocks > -1) {
 877                 spin_lock(&sbinfo->stat_lock);
 878                 if (sbinfo->free_blocks - delta >= 0)
 879                         sbinfo->free_blocks -= delta;
 880                 else
 881                         ret = -ENOMEM;
 882                 spin_unlock(&sbinfo->stat_lock);
 883         }
 884
 885         return ret;
 886 }
 887
 888 void hugetlb_put_quota(struct address_space *mapping, long delta)
 889 {
 890         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 891
 892         if (sbinfo->free_blocks > -1) {
 893                 spin_lock(&sbinfo->stat_lock);
 894                 sbinfo->free_blocks += delta;
 895                 spin_unlock(&sbinfo->stat_lock);
 896         }
 897 }
 898
 899 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
 900         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 901 {
 902         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
 903 }
 904
 905 static struct file_system_type hugetlbfs_fs_type = {
 906         .name           = "hugetlbfs",
 907         .get_sb         = hugetlbfs_get_sb,
 908         .kill_sb        = kill_litter_super,
 909 };
 910
 911 static struct vfsmount *hugetlbfs_vfsmount;
 912
 913 static int can_do_hugetlb_shm(void)
 914 {
 915         return likely(capable(CAP_IPC_LOCK) ||
 916                         in_group_p(sysctl_hugetlb_shm_group) ||
 917                         can_do_mlock());
 918 }
 919
 920 struct file *hugetlb_file_setup(const char *name, size_t size)
 921 {
 922         int error = -ENOMEM;
 923         struct file *file;
 924         struct inode *inode;
 925         struct dentry *dentry, *root;
 926         struct qstr quick_string;
 927
 928         if (!hugetlbfs_vfsmount)
 929                 return ERR_PTR(-ENOENT);
 930
 931         if (!can_do_hugetlb_shm())
 932                 return ERR_PTR(-EPERM);
 933
 934         if (!user_shm_lock(size, current->user))
 935                 return ERR_PTR(-ENOMEM);
 936
 937         root = hugetlbfs_vfsmount->mnt_root;
 938         quick_string.name = name;
 939         quick_string.len = strlen(quick_string.name);
 940         quick_string.hash = 0;
 941         dentry = d_alloc(root, &quick_string);
 942         if (!dentry)
 943                 goto out_shm_unlock;
 944
 945         error = -ENOSPC;
 946         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
 947                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
 948         if (!inode)
 949                 goto out_dentry;
 950
 951         error = -ENOMEM;
 952         if (hugetlb_reserve_pages(inode, 0,
 953                         size >> huge_page_shift(hstate_inode(inode)), NULL))
 954                 goto out_inode;
 955
 956         d_instantiate(dentry, inode);
 957         inode->i_size = size;
 958         inode->i_nlink = 0;
 959
 960         error = -ENFILE;
 961         file = alloc_file(hugetlbfs_vfsmount, dentry,
 962                         FMODE_WRITE | FMODE_READ,
 963                         &hugetlbfs_file_operations);
 964         if (!file)
 965                 goto out_dentry; /* inode is already attached */
 966
 967         return file;
 968
 969 out_inode:
 970         iput(inode);
 971 out_dentry:
 972         dput(dentry);
 973 out_shm_unlock:
 974         user_shm_unlock(size, current->user);
 975         return ERR_PTR(error);
 976 }
 977
 978 static int __init init_hugetlbfs_fs(void)
 979 {
 980         int error;
 981         struct vfsmount *vfsmount;
 982
 983         error = bdi_init(&hugetlbfs_backing_dev_info);
 984         if (error)
 985                 return error;
 986
 987         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
 988                                         sizeof(struct hugetlbfs_inode_info),
 989                                         0, 0, init_once);
 990         if (hugetlbfs_inode_cachep == NULL)
 991                 goto out2;
 992
 993         error = register_filesystem(&hugetlbfs_fs_type);
 994         if (error)
 995                 goto out;
 996
 997         vfsmount = kern_mount(&hugetlbfs_fs_type);
 998
 999         if (!IS_ERR(vfsmount)) {
1000                 hugetlbfs_vfsmount = vfsmount;
1001                 return 0;
1002         }
1003
1004         error = PTR_ERR(vfsmount);
1005
1006  out:
1007         if (error)
1008                 kmem_cache_destroy(hugetlbfs_inode_cachep);
1009  out2:
1010         bdi_destroy(&hugetlbfs_backing_dev_info);
1011         return error;
1012 }
1013
1014 static void __exit exit_hugetlbfs_fs(void)
1015 {
1016         kmem_cache_destroy(hugetlbfs_inode_cachep);
1017         unregister_filesystem(&hugetlbfs_fs_type);
1018         bdi_destroy(&hugetlbfs_backing_dev_info);
1019 }
1020
1021 module_init(init_hugetlbfs_fs)
1022 module_exit(exit_hugetlbfs_fs)
1023
1024 MODULE_LICENSE("GPL");