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/slab.h>
  30 #include <linux/dnotify.h>
  31 #include <linux/statfs.h>
  32 #include <linux/security.h>
  33
  34 #include <asm/uaccess.h>
  35
  36 /* some random number */
  37 #define HUGETLBFS_MAGIC 0x958458f6
  38
  39 static const struct super_operations hugetlbfs_ops;
  40 static const struct address_space_operations hugetlbfs_aops;
  41 const struct file_operations hugetlbfs_file_operations;
  42 static const struct inode_operations hugetlbfs_dir_inode_operations;
  43 static const struct inode_operations hugetlbfs_inode_operations;
  44
  45 static struct backing_dev_info hugetlbfs_backing_dev_info = {
  46         .ra_pages       = 0,    /* No readahead */
  47         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
  48 };
  49
  50 int sysctl_hugetlb_shm_group;
  51
  52 enum {
  53         Opt_size, Opt_nr_inodes,
  54         Opt_mode, Opt_uid, Opt_gid,
  55         Opt_pagesize,
  56         Opt_err,
  57 };
  58
  59 static const 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_pagesize,  "pagesize=%s"},
  66         {Opt_err,       NULL},
  67 };
  68
  69 static void huge_pagevec_release(struct pagevec *pvec)
  70 {
  71         int i;
  72
  73         for (i = 0; i < pagevec_count(pvec); ++i)
  74                 put_page(pvec->pages[i]);
  75
  76         pagevec_reinit(pvec);
  77 }
  78
  79 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  80 {
  81         struct inode *inode = file->f_path.dentry->d_inode;
  82         loff_t len, vma_len;
  83         int ret;
  84         struct hstate *h = hstate_file(file);
  85
  86         /*
  87          * vma address alignment (but not the pgoff alignment) has
  88          * already been checked by prepare_hugepage_range.  If you add
  89          * any error returns here, do so after setting VM_HUGETLB, so
  90          * is_vm_hugetlb_page tests below unmap_region go the right
  91          * way when do_mmap_pgoff unwinds (may be important on powerpc
  92          * and ia64).
  93          */
  94         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
  95         vma->vm_ops = &hugetlb_vm_ops;
  96
  97         if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
  98                 return -EINVAL;
  99
 100         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 101
 102         mutex_lock(&inode->i_mutex);
 103         file_accessed(file);
 104
 105         ret = -ENOMEM;
 106         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 107
 108         if (hugetlb_reserve_pages(inode,
 109                                 vma->vm_pgoff >> huge_page_order(h),
 110                                 len >> huge_page_shift(h), vma,
 111                                 vma->vm_flags))
 112                 goto out;
 113
 114         ret = 0;
 115         hugetlb_prefault_arch_hook(vma->vm_mm);
 116         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 117                 inode->i_size = len;
 118 out:
 119         mutex_unlock(&inode->i_mutex);
 120
 121         return ret;
 122 }
 123
 124 /*
 125  * Called under down_write(mmap_sem).
 126  */
 127
 128 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 129 static unsigned long
 130 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 131                 unsigned long len, unsigned long pgoff, unsigned long flags)
 132 {
 133         struct mm_struct *mm = current->mm;
 134         struct vm_area_struct *vma;
 135         unsigned long start_addr;
 136         struct hstate *h = hstate_file(file);
 137
 138         if (len & ~huge_page_mask(h))
 139                 return -EINVAL;
 140         if (len > TASK_SIZE)
 141                 return -ENOMEM;
 142
 143         if (flags & MAP_FIXED) {
 144                 if (prepare_hugepage_range(file, addr, len))
 145                         return -EINVAL;
 146                 return addr;
 147         }
 148
 149         if (addr) {
 150                 addr = ALIGN(addr, huge_page_size(h));
 151                 vma = find_vma(mm, addr);
 152                 if (TASK_SIZE - len >= addr &&
 153                     (!vma || addr + len <= vma->vm_start))
 154                         return addr;
 155         }
 156
 157         start_addr = mm->free_area_cache;
 158
 159         if (len <= mm->cached_hole_size)
 160                 start_addr = TASK_UNMAPPED_BASE;
 161
 162 full_search:
 163         addr = ALIGN(start_addr, huge_page_size(h));
 164
 165         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 166                 /* At this point:  (!vma || addr < vma->vm_end). */
 167                 if (TASK_SIZE - len < addr) {
 168                         /*
 169                          * Start a new search - just in case we missed
 170                          * some holes.
 171                          */
 172                         if (start_addr != TASK_UNMAPPED_BASE) {
 173                                 start_addr = TASK_UNMAPPED_BASE;
 174                                 goto full_search;
 175                         }
 176                         return -ENOMEM;
 177                 }
 178
 179                 if (!vma || addr + len <= vma->vm_start)
 180                         return addr;
 181                 addr = ALIGN(vma->vm_end, huge_page_size(h));
 182         }
 183 }
 184 #endif
 185
 186 static int
 187 hugetlbfs_read_actor(struct page *page, unsigned long offset,
 188                         char __user *buf, unsigned long count,
 189                         unsigned long size)
 190 {
 191         char *kaddr;
 192         unsigned long left, copied = 0;
 193         int i, chunksize;
 194
 195         if (size > count)
 196                 size = count;
 197
 198         /* Find which 4k chunk and offset with in that chunk */
 199         i = offset >> PAGE_CACHE_SHIFT;
 200         offset = offset & ~PAGE_CACHE_MASK;
 201
 202         while (size) {
 203                 chunksize = PAGE_CACHE_SIZE;
 204                 if (offset)
 205                         chunksize -= offset;
 206                 if (chunksize > size)
 207                         chunksize = size;
 208                 kaddr = kmap(&page[i]);
 209                 left = __copy_to_user(buf, kaddr + offset, chunksize);
 210                 kunmap(&page[i]);
 211                 if (left) {
 212                         copied += (chunksize - left);
 213                         break;
 214                 }
 215                 offset = 0;
 216                 size -= chunksize;
 217                 buf += chunksize;
 218                 copied += chunksize;
 219                 i++;
 220         }
 221         return copied ? copied : -EFAULT;
 222 }
 223
 224 /*
 225  * Support for read() - Find the page attached to f_mapping and copy out the
 226  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
 227  * since it has PAGE_CACHE_SIZE assumptions.
 228  */
 229 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
 230                               size_t len, loff_t *ppos)
 231 {
 232         struct hstate *h = hstate_file(filp);
 233         struct address_space *mapping = filp->f_mapping;
 234         struct inode *inode = mapping->host;
 235         unsigned long index = *ppos >> huge_page_shift(h);
 236         unsigned long offset = *ppos & ~huge_page_mask(h);
 237         unsigned long end_index;
 238         loff_t isize;
 239         ssize_t retval = 0;
 240
 241         mutex_lock(&inode->i_mutex);
 242
 243         /* validate length */
 244         if (len == 0)
 245                 goto out;
 246
 247         isize = i_size_read(inode);
 248         if (!isize)
 249                 goto out;
 250
 251         end_index = (isize - 1) >> huge_page_shift(h);
 252         for (;;) {
 253                 struct page *page;
 254                 unsigned long nr, ret;
 255                 int ra;
 256
 257                 /* nr is the maximum number of bytes to copy from this page */
 258                 nr = huge_page_size(h);
 259                 if (index >= end_index) {
 260                         if (index > end_index)
 261                                 goto out;
 262                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
 263                         if (nr <= offset) {
 264                                 goto out;
 265                         }
 266                 }
 267                 nr = nr - offset;
 268
 269                 /* Find the page */
 270                 page = find_get_page(mapping, index);
 271                 if (unlikely(page == NULL)) {
 272                         /*
 273                          * We have a HOLE, zero out the user-buffer for the
 274                          * length of the hole or request.
 275                          */
 276                         ret = len < nr ? len : nr;
 277                         if (clear_user(buf, ret))
 278                                 ra = -EFAULT;
 279                         else
 280                                 ra = 0;
 281                 } else {
 282                         /*
 283                          * We have the page, copy it to user space buffer.
 284                          */
 285                         ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
 286                         ret = ra;
 287                 }
 288                 if (ra < 0) {
 289                         if (retval == 0)
 290                                 retval = ra;
 291                         if (page)
 292                                 page_cache_release(page);
 293                         goto out;
 294                 }
 295
 296                 offset += ret;
 297                 retval += ret;
 298                 len -= ret;
 299                 index += offset >> huge_page_shift(h);
 300                 offset &= ~huge_page_mask(h);
 301
 302                 if (page)
 303                         page_cache_release(page);
 304
 305                 /* short read or no more work */
 306                 if ((ret != nr) || (len == 0))
 307                         break;
 308         }
 309 out:
 310         *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
 311         mutex_unlock(&inode->i_mutex);
 312         return retval;
 313 }
 314
 315 static int hugetlbfs_write_begin(struct file *file,
 316                         struct address_space *mapping,
 317                         loff_t pos, unsigned len, unsigned flags,
 318                         struct page **pagep, void **fsdata)
 319 {
 320         return -EINVAL;
 321 }
 322
 323 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 324                         loff_t pos, unsigned len, unsigned copied,
 325                         struct page *page, void *fsdata)
 326 {
 327         BUG();
 328         return -EINVAL;
 329 }
 330
 331 static void truncate_huge_page(struct page *page)
 332 {
 333         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
 334         ClearPageUptodate(page);
 335         remove_from_page_cache(page);
 336         put_page(page);
 337 }
 338
 339 static void truncate_hugepages(struct inode *inode, loff_t lstart)
 340 {
 341         struct hstate *h = hstate_inode(inode);
 342         struct address_space *mapping = &inode->i_data;
 343         const pgoff_t start = lstart >> huge_page_shift(h);
 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, NULL);
 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         struct hstate *h = hstate_inode(inode);
 453
 454         BUG_ON(offset & ~huge_page_mask(h));
 455         pgoff = offset >> PAGE_SHIFT;
 456
 457         i_size_write(inode, offset);
 458         spin_lock(&mapping->i_mmap_lock);
 459         if (!prio_tree_empty(&mapping->i_mmap))
 460                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 461         spin_unlock(&mapping->i_mmap_lock);
 462         truncate_hugepages(inode, offset);
 463         return 0;
 464 }
 465
 466 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 467 {
 468         struct inode *inode = dentry->d_inode;
 469         struct hstate *h = hstate_inode(inode);
 470         int error;
 471         unsigned int ia_valid = attr->ia_valid;
 472
 473         BUG_ON(!inode);
 474
 475         error = inode_change_ok(inode, attr);
 476         if (error)
 477                 goto out;
 478
 479         if (ia_valid & ATTR_SIZE) {
 480                 error = -EINVAL;
 481                 if (!(attr->ia_size & ~huge_page_mask(h)))
 482                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 483                 if (error)
 484                         goto out;
 485                 attr->ia_valid &= ~ATTR_SIZE;
 486         }
 487         error = inode_setattr(inode, attr);
 488 out:
 489         return error;
 490 }
 491
 492 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 493                                         gid_t gid, int mode, dev_t dev)
 494 {
 495         struct inode *inode;
 496
 497         inode = new_inode(sb);
 498         if (inode) {
 499                 struct hugetlbfs_inode_info *info;
 500                 inode->i_mode = mode;
 501                 inode->i_uid = uid;
 502                 inode->i_gid = gid;
 503                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 504                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 505                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 506                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
 507                 info = HUGETLBFS_I(inode);
 508                 mpol_shared_policy_init(&info->policy, NULL);
 509                 switch (mode & S_IFMT) {
 510                 default:
 511                         init_special_inode(inode, mode, dev);
 512                         break;
 513                 case S_IFREG:
 514                         inode->i_op = &hugetlbfs_inode_operations;
 515                         inode->i_fop = &hugetlbfs_file_operations;
 516                         break;
 517                 case S_IFDIR:
 518                         inode->i_op = &hugetlbfs_dir_inode_operations;
 519                         inode->i_fop = &simple_dir_operations;
 520
 521                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 522                         inc_nlink(inode);
 523                         break;
 524                 case S_IFLNK:
 525                         inode->i_op = &page_symlink_inode_operations;
 526                         break;
 527                 }
 528         }
 529         return inode;
 530 }
 531
 532 /*
 533  * File creation. Allocate an inode, and we're done..
 534  */
 535 static int hugetlbfs_mknod(struct inode *dir,
 536                         struct dentry *dentry, int mode, dev_t dev)
 537 {
 538         struct inode *inode;
 539         int error = -ENOSPC;
 540         gid_t gid;
 541
 542         if (dir->i_mode & S_ISGID) {
 543                 gid = dir->i_gid;
 544                 if (S_ISDIR(mode))
 545                         mode |= S_ISGID;
 546         } else {
 547                 gid = current_fsgid();
 548         }
 549         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
 550         if (inode) {
 551                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 552                 d_instantiate(dentry, inode);
 553                 dget(dentry);   /* Extra count - pin the dentry in core */
 554                 error = 0;
 555         }
 556         return error;
 557 }
 558
 559 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 560 {
 561         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 562         if (!retval)
 563                 inc_nlink(dir);
 564         return retval;
 565 }
 566
 567 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 568 {
 569         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 570 }
 571
 572 static int hugetlbfs_symlink(struct inode *dir,
 573                         struct dentry *dentry, const char *symname)
 574 {
 575         struct inode *inode;
 576         int error = -ENOSPC;
 577         gid_t gid;
 578
 579         if (dir->i_mode & S_ISGID)
 580                 gid = dir->i_gid;
 581         else
 582                 gid = current_fsgid();
 583
 584         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
 585                                         gid, S_IFLNK|S_IRWXUGO, 0);
 586         if (inode) {
 587                 int l = strlen(symname)+1;
 588                 error = page_symlink(inode, symname, l);
 589                 if (!error) {
 590                         d_instantiate(dentry, inode);
 591                         dget(dentry);
 592                 } else
 593                         iput(inode);
 594         }
 595         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 596
 597         return error;
 598 }
 599
 600 /*
 601  * mark the head page dirty
 602  */
 603 static int hugetlbfs_set_page_dirty(struct page *page)
 604 {
 605         struct page *head = compound_head(page);
 606
 607         SetPageDirty(head);
 608         return 0;
 609 }
 610
 611 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 612 {
 613         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 614         struct hstate *h = hstate_inode(dentry->d_inode);
 615
 616         buf->f_type = HUGETLBFS_MAGIC;
 617         buf->f_bsize = huge_page_size(h);
 618         if (sbinfo) {
 619                 spin_lock(&sbinfo->stat_lock);
 620                 /* If no limits set, just report 0 for max/free/used
 621                  * blocks, like simple_statfs() */
 622                 if (sbinfo->max_blocks >= 0) {
 623                         buf->f_blocks = sbinfo->max_blocks;
 624                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 625                         buf->f_files = sbinfo->max_inodes;
 626                         buf->f_ffree = sbinfo->free_inodes;
 627                 }
 628                 spin_unlock(&sbinfo->stat_lock);
 629         }
 630         buf->f_namelen = NAME_MAX;
 631         return 0;
 632 }
 633
 634 static void hugetlbfs_put_super(struct super_block *sb)
 635 {
 636         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 637
 638         if (sbi) {
 639                 sb->s_fs_info = NULL;
 640                 kfree(sbi);
 641         }
 642 }
 643
 644 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 645 {
 646         if (sbinfo->free_inodes >= 0) {
 647                 spin_lock(&sbinfo->stat_lock);
 648                 if (unlikely(!sbinfo->free_inodes)) {
 649                         spin_unlock(&sbinfo->stat_lock);
 650                         return 0;
 651                 }
 652                 sbinfo->free_inodes--;
 653                 spin_unlock(&sbinfo->stat_lock);
 654         }
 655
 656         return 1;
 657 }
 658
 659 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 660 {
 661         if (sbinfo->free_inodes >= 0) {
 662                 spin_lock(&sbinfo->stat_lock);
 663                 sbinfo->free_inodes++;
 664                 spin_unlock(&sbinfo->stat_lock);
 665         }
 666 }
 667
 668
 669 static struct kmem_cache *hugetlbfs_inode_cachep;
 670
 671 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 672 {
 673         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 674         struct hugetlbfs_inode_info *p;
 675
 676         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 677                 return NULL;
 678         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 679         if (unlikely(!p)) {
 680                 hugetlbfs_inc_free_inodes(sbinfo);
 681                 return NULL;
 682         }
 683         return &p->vfs_inode;
 684 }
 685
 686 static void hugetlbfs_destroy_inode(struct inode *inode)
 687 {
 688         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 689         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 690         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 691 }
 692
 693 static const struct address_space_operations hugetlbfs_aops = {
 694         .write_begin    = hugetlbfs_write_begin,
 695         .write_end      = hugetlbfs_write_end,
 696         .set_page_dirty = hugetlbfs_set_page_dirty,
 697 };
 698
 699
 700 static void init_once(void *foo)
 701 {
 702         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 703
 704         inode_init_once(&ei->vfs_inode);
 705 }
 706
 707 const struct file_operations hugetlbfs_file_operations = {
 708         .read                   = hugetlbfs_read,
 709         .mmap                   = hugetlbfs_file_mmap,
 710         .fsync                  = simple_sync_file,
 711         .get_unmapped_area      = hugetlb_get_unmapped_area,
 712 };
 713
 714 static const struct inode_operations hugetlbfs_dir_inode_operations = {
 715         .create         = hugetlbfs_create,
 716         .lookup         = simple_lookup,
 717         .link           = simple_link,
 718         .unlink         = simple_unlink,
 719         .symlink        = hugetlbfs_symlink,
 720         .mkdir          = hugetlbfs_mkdir,
 721         .rmdir          = simple_rmdir,
 722         .mknod          = hugetlbfs_mknod,
 723         .rename         = simple_rename,
 724         .setattr        = hugetlbfs_setattr,
 725 };
 726
 727 static const struct inode_operations hugetlbfs_inode_operations = {
 728         .setattr        = hugetlbfs_setattr,
 729 };
 730
 731 static const struct super_operations hugetlbfs_ops = {
 732         .alloc_inode    = hugetlbfs_alloc_inode,
 733         .destroy_inode  = hugetlbfs_destroy_inode,
 734         .statfs         = hugetlbfs_statfs,
 735         .delete_inode   = hugetlbfs_delete_inode,
 736         .drop_inode     = hugetlbfs_drop_inode,
 737         .put_super      = hugetlbfs_put_super,
 738         .show_options   = generic_show_options,
 739 };
 740
 741 static int
 742 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 743 {
 744         char *p, *rest;
 745         substring_t args[MAX_OPT_ARGS];
 746         int option;
 747         unsigned long long size = 0;
 748         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
 749
 750         if (!options)
 751                 return 0;
 752
 753         while ((p = strsep(&options, ",")) != NULL) {
 754                 int token;
 755                 if (!*p)
 756                         continue;
 757
 758                 token = match_token(p, tokens, args);
 759                 switch (token) {
 760                 case Opt_uid:
 761                         if (match_int(&args[0], &option))
 762                                 goto bad_val;
 763                         pconfig->uid = option;
 764                         break;
 765
 766                 case Opt_gid:
 767                         if (match_int(&args[0], &option))
 768                                 goto bad_val;
 769                         pconfig->gid = option;
 770                         break;
 771
 772                 case Opt_mode:
 773                         if (match_octal(&args[0], &option))
 774                                 goto bad_val;
 775                         pconfig->mode = option & 01777U;
 776                         break;
 777
 778                 case Opt_size: {
 779                         /* memparse() will accept a K/M/G without a digit */
 780                         if (!isdigit(*args[0].from))
 781                                 goto bad_val;
 782                         size = memparse(args[0].from, &rest);
 783                         setsize = SIZE_STD;
 784                         if (*rest == '%')
 785                                 setsize = SIZE_PERCENT;
 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                 case Opt_pagesize: {
 797                         unsigned long ps;
 798                         ps = memparse(args[0].from, &rest);
 799                         pconfig->hstate = size_to_hstate(ps);
 800                         if (!pconfig->hstate) {
 801                                 printk(KERN_ERR
 802                                 "hugetlbfs: Unsupported page size %lu MB\n",
 803                                         ps >> 20);
 804                                 return -EINVAL;
 805                         }
 806                         break;
 807                 }
 808
 809                 default:
 810                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
 811                                  p);
 812                         return -EINVAL;
 813                         break;
 814                 }
 815         }
 816
 817         /* Do size after hstate is set up */
 818         if (setsize > NO_SIZE) {
 819                 struct hstate *h = pconfig->hstate;
 820                 if (setsize == SIZE_PERCENT) {
 821                         size <<= huge_page_shift(h);
 822                         size *= h->max_huge_pages;
 823                         do_div(size, 100);
 824                 }
 825                 pconfig->nr_blocks = (size >> huge_page_shift(h));
 826         }
 827
 828         return 0;
 829
 830 bad_val:
 831         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
 832                args[0].from, p);
 833         return -EINVAL;
 834 }
 835
 836 static int
 837 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 838 {
 839         struct inode * inode;
 840         struct dentry * root;
 841         int ret;
 842         struct hugetlbfs_config config;
 843         struct hugetlbfs_sb_info *sbinfo;
 844
 845         save_mount_options(sb, data);
 846
 847         config.nr_blocks = -1; /* No limit on size by default */
 848         config.nr_inodes = -1; /* No limit on number of inodes by default */
 849         config.uid = current_fsuid();
 850         config.gid = current_fsgid();
 851         config.mode = 0755;
 852         config.hstate = &default_hstate;
 853         ret = hugetlbfs_parse_options(data, &config);
 854         if (ret)
 855                 return ret;
 856
 857         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 858         if (!sbinfo)
 859                 return -ENOMEM;
 860         sb->s_fs_info = sbinfo;
 861         sbinfo->hstate = config.hstate;
 862         spin_lock_init(&sbinfo->stat_lock);
 863         sbinfo->max_blocks = config.nr_blocks;
 864         sbinfo->free_blocks = config.nr_blocks;
 865         sbinfo->max_inodes = config.nr_inodes;
 866         sbinfo->free_inodes = config.nr_inodes;
 867         sb->s_maxbytes = MAX_LFS_FILESIZE;
 868         sb->s_blocksize = huge_page_size(config.hstate);
 869         sb->s_blocksize_bits = huge_page_shift(config.hstate);
 870         sb->s_magic = HUGETLBFS_MAGIC;
 871         sb->s_op = &hugetlbfs_ops;
 872         sb->s_time_gran = 1;
 873         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 874                                         S_IFDIR | config.mode, 0);
 875         if (!inode)
 876                 goto out_free;
 877
 878         root = d_alloc_root(inode);
 879         if (!root) {
 880                 iput(inode);
 881                 goto out_free;
 882         }
 883         sb->s_root = root;
 884         return 0;
 885 out_free:
 886         kfree(sbinfo);
 887         return -ENOMEM;
 888 }
 889
 890 int hugetlb_get_quota(struct address_space *mapping, long delta)
 891 {
 892         int ret = 0;
 893         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 894
 895         if (sbinfo->free_blocks > -1) {
 896                 spin_lock(&sbinfo->stat_lock);
 897                 if (sbinfo->free_blocks - delta >= 0)
 898                         sbinfo->free_blocks -= delta;
 899                 else
 900                         ret = -ENOMEM;
 901                 spin_unlock(&sbinfo->stat_lock);
 902         }
 903
 904         return ret;
 905 }
 906
 907 void hugetlb_put_quota(struct address_space *mapping, long delta)
 908 {
 909         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 910
 911         if (sbinfo->free_blocks > -1) {
 912                 spin_lock(&sbinfo->stat_lock);
 913                 sbinfo->free_blocks += delta;
 914                 spin_unlock(&sbinfo->stat_lock);
 915         }
 916 }
 917
 918 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
 919         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 920 {
 921         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
 922 }
 923
 924 static struct file_system_type hugetlbfs_fs_type = {
 925         .name           = "hugetlbfs",
 926         .get_sb         = hugetlbfs_get_sb,
 927         .kill_sb        = kill_litter_super,
 928 };
 929
 930 static struct vfsmount *hugetlbfs_vfsmount;
 931
 932 static int can_do_hugetlb_shm(void)
 933 {
 934         return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
 935 }
 936
 937 struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag)
 938 {
 939         int error = -ENOMEM;
 940         int unlock_shm = 0;
 941         struct file *file;
 942         struct inode *inode;
 943         struct dentry *dentry, *root;
 944         struct qstr quick_string;
 945         struct user_struct *user = current_user();
 946
 947         if (!hugetlbfs_vfsmount)
 948                 return ERR_PTR(-ENOENT);
 949
 950         if (!can_do_hugetlb_shm()) {
 951                 if (user_shm_lock(size, user)) {
 952                         unlock_shm = 1;
 953                         WARN_ONCE(1,
 954                           "Using mlock ulimits for SHM_HUGETLB deprecated\n");
 955                 } else
 956                         return ERR_PTR(-EPERM);
 957         }
 958
 959         root = hugetlbfs_vfsmount->mnt_root;
 960         quick_string.name = name;
 961         quick_string.len = strlen(quick_string.name);
 962         quick_string.hash = 0;
 963         dentry = d_alloc(root, &quick_string);
 964         if (!dentry)
 965                 goto out_shm_unlock;
 966
 967         error = -ENOSPC;
 968         inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
 969                                 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
 970         if (!inode)
 971                 goto out_dentry;
 972
 973         error = -ENOMEM;
 974         if (hugetlb_reserve_pages(inode, 0,
 975                         size >> huge_page_shift(hstate_inode(inode)), NULL,
 976                         acctflag))
 977                 goto out_inode;
 978
 979         d_instantiate(dentry, inode);
 980         inode->i_size = size;
 981         inode->i_nlink = 0;
 982
 983         error = -ENFILE;
 984         file = alloc_file(hugetlbfs_vfsmount, dentry,
 985                         FMODE_WRITE | FMODE_READ,
 986                         &hugetlbfs_file_operations);
 987         if (!file)
 988                 goto out_dentry; /* inode is already attached */
 989
 990         return file;
 991
 992 out_inode:
 993         iput(inode);
 994 out_dentry:
 995         dput(dentry);
 996 out_shm_unlock:
 997         if (unlock_shm)
 998                 user_shm_unlock(size, user);
 999         return ERR_PTR(error);
1000 }
1001
1002 static int __init init_hugetlbfs_fs(void)
1003 {
1004         int error;
1005         struct vfsmount *vfsmount;
1006
1007         error = bdi_init(&hugetlbfs_backing_dev_info);
1008         if (error)
1009                 return error;
1010
1011         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1012                                         sizeof(struct hugetlbfs_inode_info),
1013                                         0, 0, init_once);
1014         if (hugetlbfs_inode_cachep == NULL)
1015                 goto out2;
1016
1017         error = register_filesystem(&hugetlbfs_fs_type);
1018         if (error)
1019                 goto out;
1020
1021         vfsmount = kern_mount(&hugetlbfs_fs_type);
1022
1023         if (!IS_ERR(vfsmount)) {
1024                 hugetlbfs_vfsmount = vfsmount;
1025                 return 0;
1026         }
1027
1028         error = PTR_ERR(vfsmount);
1029
1030  out:
1031         if (error)
1032                 kmem_cache_destroy(hugetlbfs_inode_cachep);
1033  out2:
1034         bdi_destroy(&hugetlbfs_backing_dev_info);
1035         return error;
1036 }
1037
1038 static void __exit exit_hugetlbfs_fs(void)
1039 {
1040         kmem_cache_destroy(hugetlbfs_inode_cachep);
1041         unregister_filesystem(&hugetlbfs_fs_type);
1042         bdi_destroy(&hugetlbfs_backing_dev_info);
1043 }
1044
1045 module_init(init_hugetlbfs_fs)
1046 module_exit(exit_hugetlbfs_fs)
1047
1048 MODULE_LICENSE("GPL");