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 #include <linux/magic.h>
  34 #include <linux/migrate.h>
  35
  36 #include <asm/uaccess.h>
  37
  38 static const struct super_operations hugetlbfs_ops;
  39 static const struct address_space_operations hugetlbfs_aops;
  40 const struct file_operations hugetlbfs_file_operations;
  41 static const struct inode_operations hugetlbfs_dir_inode_operations;
  42 static const struct inode_operations hugetlbfs_inode_operations;
  43
  44 static struct backing_dev_info hugetlbfs_backing_dev_info = {
  45         .name           = "hugetlbfs",
  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         /*
  98          * Offset passed to mmap (before page shift) could have been
  99          * negative when represented as a (l)off_t.
 100          */
 101         if (((loff_t)vma->vm_pgoff << PAGE_SHIFT) < 0)
 102                 return -EINVAL;
 103
 104         if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
 105                 return -EINVAL;
 106
 107         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 108         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 109         /* check for overflow */
 110         if (len < vma_len)
 111                 return -EINVAL;
 112
 113         mutex_lock(&inode->i_mutex);
 114         file_accessed(file);
 115
 116         ret = -ENOMEM;
 117         if (hugetlb_reserve_pages(inode,
 118                                 vma->vm_pgoff >> huge_page_order(h),
 119                                 len >> huge_page_shift(h), vma,
 120                                 vma->vm_flags))
 121                 goto out;
 122
 123         ret = 0;
 124         hugetlb_prefault_arch_hook(vma->vm_mm);
 125         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 126                 i_size_write(inode, len);
 127 out:
 128         mutex_unlock(&inode->i_mutex);
 129
 130         return ret;
 131 }
 132
 133 /*
 134  * Called under down_write(mmap_sem).
 135  */
 136
 137 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 138 static unsigned long
 139 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 140                 unsigned long len, unsigned long pgoff, unsigned long flags)
 141 {
 142         struct mm_struct *mm = current->mm;
 143         struct vm_area_struct *vma;
 144         unsigned long start_addr;
 145         struct hstate *h = hstate_file(file);
 146
 147         if (len & ~huge_page_mask(h))
 148                 return -EINVAL;
 149         if (len > TASK_SIZE)
 150                 return -ENOMEM;
 151
 152         if (flags & MAP_FIXED) {
 153                 if (prepare_hugepage_range(file, addr, len))
 154                         return -EINVAL;
 155                 return addr;
 156         }
 157
 158         if (addr) {
 159                 addr = ALIGN(addr, huge_page_size(h));
 160                 vma = find_vma(mm, addr);
 161                 if (TASK_SIZE - len >= addr &&
 162                     (!vma || addr + len <= vm_start_gap(vma)))
 163                         return addr;
 164         }
 165
 166         start_addr = mm->free_area_cache;
 167
 168         if (len <= mm->cached_hole_size)
 169                 start_addr = TASK_UNMAPPED_BASE;
 170
 171 full_search:
 172         addr = ALIGN(start_addr, huge_page_size(h));
 173
 174         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 175                 /* At this point:  (!vma || addr < vma->vm_end). */
 176                 if (TASK_SIZE - len < addr) {
 177                         /*
 178                          * Start a new search - just in case we missed
 179                          * some holes.
 180                          */
 181                         if (start_addr != TASK_UNMAPPED_BASE) {
 182                                 start_addr = TASK_UNMAPPED_BASE;
 183                                 goto full_search;
 184                         }
 185                         return -ENOMEM;
 186                 }
 187
 188                 if (!vma || addr + len <= vm_start_gap(vma))
 189                         return addr;
 190                 addr = ALIGN(vma->vm_end, huge_page_size(h));
 191         }
 192 }
 193 #endif
 194
 195 static int
 196 hugetlbfs_read_actor(struct page *page, unsigned long offset,
 197                         char __user *buf, unsigned long count,
 198                         unsigned long size)
 199 {
 200         char *kaddr;
 201         unsigned long left, copied = 0;
 202         int i, chunksize;
 203
 204         if (size > count)
 205                 size = count;
 206
 207         /* Find which 4k chunk and offset with in that chunk */
 208         i = offset >> PAGE_CACHE_SHIFT;
 209         offset = offset & ~PAGE_CACHE_MASK;
 210
 211         while (size) {
 212                 chunksize = PAGE_CACHE_SIZE;
 213                 if (offset)
 214                         chunksize -= offset;
 215                 if (chunksize > size)
 216                         chunksize = size;
 217                 kaddr = kmap(&page[i]);
 218                 left = __copy_to_user(buf, kaddr + offset, chunksize);
 219                 kunmap(&page[i]);
 220                 if (left) {
 221                         copied += (chunksize - left);
 222                         break;
 223                 }
 224                 offset = 0;
 225                 size -= chunksize;
 226                 buf += chunksize;
 227                 copied += chunksize;
 228                 i++;
 229         }
 230         return copied ? copied : -EFAULT;
 231 }
 232
 233 /*
 234  * Support for read() - Find the page attached to f_mapping and copy out the
 235  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
 236  * since it has PAGE_CACHE_SIZE assumptions.
 237  */
 238 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
 239                               size_t len, loff_t *ppos)
 240 {
 241         struct hstate *h = hstate_file(filp);
 242         struct address_space *mapping = filp->f_mapping;
 243         struct inode *inode = mapping->host;
 244         unsigned long index = *ppos >> huge_page_shift(h);
 245         unsigned long offset = *ppos & ~huge_page_mask(h);
 246         unsigned long end_index;
 247         loff_t isize;
 248         ssize_t retval = 0;
 249
 250         /* validate length */
 251         if (len == 0)
 252                 goto out;
 253
 254         for (;;) {
 255                 struct page *page;
 256                 unsigned long nr, ret;
 257                 int ra;
 258
 259                 /* nr is the maximum number of bytes to copy from this page */
 260                 nr = huge_page_size(h);
 261                 isize = i_size_read(inode);
 262                 if (!isize)
 263                         goto out;
 264                 end_index = (isize - 1) >> huge_page_shift(h);
 265                 if (index >= end_index) {
 266                         if (index > end_index)
 267                                 goto out;
 268                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
 269                         if (nr <= offset)
 270                                 goto out;
 271                 }
 272                 nr = nr - offset;
 273
 274                 /* Find the page */
 275                 page = find_lock_page(mapping, index);
 276                 if (unlikely(page == NULL)) {
 277                         /*
 278                          * We have a HOLE, zero out the user-buffer for the
 279                          * length of the hole or request.
 280                          */
 281                         ret = len < nr ? len : nr;
 282                         if (clear_user(buf, ret))
 283                                 ra = -EFAULT;
 284                         else
 285                                 ra = 0;
 286                 } else {
 287                         unlock_page(page);
 288
 289                         /*
 290                          * We have the page, copy it to user space buffer.
 291                          */
 292                         ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
 293                         ret = ra;
 294                         page_cache_release(page);
 295                 }
 296                 if (ra < 0) {
 297                         if (retval == 0)
 298                                 retval = ra;
 299                         goto out;
 300                 }
 301
 302                 offset += ret;
 303                 retval += ret;
 304                 len -= ret;
 305                 index += offset >> huge_page_shift(h);
 306                 offset &= ~huge_page_mask(h);
 307
 308                 /* short read or no more work */
 309                 if ((ret != nr) || (len == 0))
 310                         break;
 311         }
 312 out:
 313         *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
 314         return retval;
 315 }
 316
 317 static int hugetlbfs_write_begin(struct file *file,
 318                         struct address_space *mapping,
 319                         loff_t pos, unsigned len, unsigned flags,
 320                         struct page **pagep, void **fsdata)
 321 {
 322         return -EINVAL;
 323 }
 324
 325 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 326                         loff_t pos, unsigned len, unsigned copied,
 327                         struct page *page, void *fsdata)
 328 {
 329         BUG();
 330         return -EINVAL;
 331 }
 332
 333 static void truncate_huge_page(struct page *page)
 334 {
 335         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
 336         ClearPageUptodate(page);
 337         delete_from_page_cache(page);
 338 }
 339
 340 static void truncate_hugepages(struct inode *inode, loff_t lstart)
 341 {
 342         struct hstate *h = hstate_inode(inode);
 343         struct address_space *mapping = &inode->i_data;
 344         const pgoff_t start = lstart >> huge_page_shift(h);
 345         struct pagevec pvec;
 346         pgoff_t next;
 347         int i, freed = 0;
 348
 349         pagevec_init(&pvec, 0);
 350         next = start;
 351         while (1) {
 352                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 353                         if (next == start)
 354                                 break;
 355                         next = start;
 356                         continue;
 357                 }
 358
 359                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 360                         struct page *page = pvec.pages[i];
 361
 362                         lock_page(page);
 363                         if (page->index > next)
 364                                 next = page->index;
 365                         ++next;
 366                         truncate_huge_page(page);
 367                         unlock_page(page);
 368                         freed++;
 369                 }
 370                 huge_pagevec_release(&pvec);
 371         }
 372         BUG_ON(!lstart && mapping->nrpages);
 373         hugetlb_unreserve_pages(inode, start, freed);
 374 }
 375
 376 static void hugetlbfs_evict_inode(struct inode *inode)
 377 {
 378         truncate_hugepages(inode, 0);
 379         end_writeback(inode);
 380 }
 381
 382 static inline void
 383 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
 384 {
 385         struct vm_area_struct *vma;
 386         struct prio_tree_iter iter;
 387
 388         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
 389                 unsigned long v_offset;
 390
 391                 /*
 392                  * Can the expression below overflow on 32-bit arches?
 393                  * No, because the prio_tree returns us only those vmas
 394                  * which overlap the truncated area starting at pgoff,
 395                  * and no vma on a 32-bit arch can span beyond the 4GB.
 396                  */
 397                 if (vma->vm_pgoff < pgoff)
 398                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 399                 else
 400                         v_offset = 0;
 401
 402                 __unmap_hugepage_range(vma,
 403                                 vma->vm_start + v_offset, vma->vm_end, NULL);
 404         }
 405 }
 406
 407 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 408 {
 409         pgoff_t pgoff;
 410         struct address_space *mapping = inode->i_mapping;
 411         struct hstate *h = hstate_inode(inode);
 412
 413         BUG_ON(offset & ~huge_page_mask(h));
 414         pgoff = offset >> PAGE_SHIFT;
 415
 416         i_size_write(inode, offset);
 417         mutex_lock(&mapping->i_mmap_mutex);
 418         if (!prio_tree_empty(&mapping->i_mmap))
 419                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 420         mutex_unlock(&mapping->i_mmap_mutex);
 421         truncate_hugepages(inode, offset);
 422         return 0;
 423 }
 424
 425 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 426 {
 427         struct inode *inode = dentry->d_inode;
 428         struct hstate *h = hstate_inode(inode);
 429         int error;
 430         unsigned int ia_valid = attr->ia_valid;
 431
 432         BUG_ON(!inode);
 433
 434         error = setattr_prepare(dentry, attr);
 435         if (error)
 436                 return error;
 437
 438         if (ia_valid & ATTR_SIZE) {
 439                 error = -EINVAL;
 440                 if (attr->ia_size & ~huge_page_mask(h))
 441                         return -EINVAL;
 442                 error = hugetlb_vmtruncate(inode, attr->ia_size);
 443                 if (error)
 444                         return error;
 445         }
 446
 447         setattr_copy(inode, attr);
 448         mark_inode_dirty(inode);
 449         return 0;
 450 }
 451
 452 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 453                                         gid_t gid, int mode, dev_t dev)
 454 {
 455         struct inode *inode;
 456
 457         inode = new_inode(sb);
 458         if (inode) {
 459                 struct hugetlbfs_inode_info *info;
 460                 inode->i_ino = get_next_ino();
 461                 inode->i_mode = mode;
 462                 inode->i_uid = uid;
 463                 inode->i_gid = gid;
 464                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 465                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 466                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 467                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
 468                 info = HUGETLBFS_I(inode);
 469                 /*
 470                  * The policy is initialized here even if we are creating a
 471                  * private inode because initialization simply creates an
 472                  * an empty rb tree and calls spin_lock_init(), later when we
 473                  * call mpol_free_shared_policy() it will just return because
 474                  * the rb tree will still be empty.
 475                  */
 476                 mpol_shared_policy_init(&info->policy, NULL);
 477                 switch (mode & S_IFMT) {
 478                 default:
 479                         init_special_inode(inode, mode, dev);
 480                         break;
 481                 case S_IFREG:
 482                         inode->i_op = &hugetlbfs_inode_operations;
 483                         inode->i_fop = &hugetlbfs_file_operations;
 484                         break;
 485                 case S_IFDIR:
 486                         inode->i_op = &hugetlbfs_dir_inode_operations;
 487                         inode->i_fop = &simple_dir_operations;
 488
 489                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 490                         inc_nlink(inode);
 491                         break;
 492                 case S_IFLNK:
 493                         inode->i_op = &page_symlink_inode_operations;
 494                         break;
 495                 }
 496                 lockdep_annotate_inode_mutex_key(inode);
 497         }
 498         return inode;
 499 }
 500
 501 /*
 502  * File creation. Allocate an inode, and we're done..
 503  */
 504 static int hugetlbfs_mknod(struct inode *dir,
 505                         struct dentry *dentry, int mode, dev_t dev)
 506 {
 507         struct inode *inode;
 508         int error = -ENOSPC;
 509         gid_t gid;
 510
 511         if (dir->i_mode & S_ISGID) {
 512                 gid = dir->i_gid;
 513                 if (S_ISDIR(mode))
 514                         mode |= S_ISGID;
 515         } else {
 516                 gid = current_fsgid();
 517         }
 518         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
 519         if (inode) {
 520                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 521                 d_instantiate(dentry, inode);
 522                 dget(dentry);   /* Extra count - pin the dentry in core */
 523                 error = 0;
 524         }
 525         return error;
 526 }
 527
 528 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 529 {
 530         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 531         if (!retval)
 532                 inc_nlink(dir);
 533         return retval;
 534 }
 535
 536 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 537 {
 538         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 539 }
 540
 541 static int hugetlbfs_symlink(struct inode *dir,
 542                         struct dentry *dentry, const char *symname)
 543 {
 544         struct inode *inode;
 545         int error = -ENOSPC;
 546         gid_t gid;
 547
 548         if (dir->i_mode & S_ISGID)
 549                 gid = dir->i_gid;
 550         else
 551                 gid = current_fsgid();
 552
 553         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
 554                                         gid, S_IFLNK|S_IRWXUGO, 0);
 555         if (inode) {
 556                 int l = strlen(symname)+1;
 557                 error = page_symlink(inode, symname, l);
 558                 if (!error) {
 559                         d_instantiate(dentry, inode);
 560                         dget(dentry);
 561                 } else
 562                         iput(inode);
 563         }
 564         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 565
 566         return error;
 567 }
 568
 569 /*
 570  * mark the head page dirty
 571  */
 572 static int hugetlbfs_set_page_dirty(struct page *page)
 573 {
 574         struct page *head = compound_head(page);
 575
 576         SetPageDirty(head);
 577         return 0;
 578 }
 579
 580 static int hugetlbfs_migrate_page(struct address_space *mapping,
 581                                 struct page *newpage, struct page *page,
 582                                 enum migrate_mode mode)
 583 {
 584         int rc;
 585
 586         rc = migrate_huge_page_move_mapping(mapping, newpage, page);
 587         if (rc)
 588                 return rc;
 589         migrate_page_copy(newpage, page);
 590
 591         return 0;
 592 }
 593
 594 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 595 {
 596         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 597         struct hstate *h = hstate_inode(dentry->d_inode);
 598
 599         buf->f_type = HUGETLBFS_MAGIC;
 600         buf->f_bsize = huge_page_size(h);
 601         if (sbinfo) {
 602                 spin_lock(&sbinfo->stat_lock);
 603                 /* If no limits set, just report 0 for max/free/used
 604                  * blocks, like simple_statfs() */
 605                 if (sbinfo->spool) {
 606                         long free_pages;
 607
 608                         spin_lock(&sbinfo->spool->lock);
 609                         buf->f_blocks = sbinfo->spool->max_hpages;
 610                         free_pages = sbinfo->spool->max_hpages
 611                                 - sbinfo->spool->used_hpages;
 612                         buf->f_bavail = buf->f_bfree = free_pages;
 613                         spin_unlock(&sbinfo->spool->lock);
 614                         buf->f_files = sbinfo->max_inodes;
 615                         buf->f_ffree = sbinfo->free_inodes;
 616                 }
 617                 spin_unlock(&sbinfo->stat_lock);
 618         }
 619         buf->f_namelen = NAME_MAX;
 620         return 0;
 621 }
 622
 623 static void hugetlbfs_put_super(struct super_block *sb)
 624 {
 625         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 626
 627         if (sbi) {
 628                 sb->s_fs_info = NULL;
 629
 630                 if (sbi->spool)
 631                         hugepage_put_subpool(sbi->spool);
 632
 633                 kfree(sbi);
 634         }
 635 }
 636
 637 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 638 {
 639         if (sbinfo->free_inodes >= 0) {
 640                 spin_lock(&sbinfo->stat_lock);
 641                 if (unlikely(!sbinfo->free_inodes)) {
 642                         spin_unlock(&sbinfo->stat_lock);
 643                         return 0;
 644                 }
 645                 sbinfo->free_inodes--;
 646                 spin_unlock(&sbinfo->stat_lock);
 647         }
 648
 649         return 1;
 650 }
 651
 652 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 653 {
 654         if (sbinfo->free_inodes >= 0) {
 655                 spin_lock(&sbinfo->stat_lock);
 656                 sbinfo->free_inodes++;
 657                 spin_unlock(&sbinfo->stat_lock);
 658         }
 659 }
 660
 661
 662 static struct kmem_cache *hugetlbfs_inode_cachep;
 663
 664 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 665 {
 666         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 667         struct hugetlbfs_inode_info *p;
 668
 669         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 670                 return NULL;
 671         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 672         if (unlikely(!p)) {
 673                 hugetlbfs_inc_free_inodes(sbinfo);
 674                 return NULL;
 675         }
 676         return &p->vfs_inode;
 677 }
 678
 679 static void hugetlbfs_i_callback(struct rcu_head *head)
 680 {
 681         struct inode *inode = container_of(head, struct inode, i_rcu);
 682         INIT_LIST_HEAD(&inode->i_dentry);
 683         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(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         call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
 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         .migratepage    = hugetlbfs_migrate_page,
 698 };
 699
 700
 701 static void init_once(void *foo)
 702 {
 703         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 704
 705         inode_init_once(&ei->vfs_inode);
 706 }
 707
 708 const struct file_operations hugetlbfs_file_operations = {
 709         .read                   = hugetlbfs_read,
 710         .mmap                   = hugetlbfs_file_mmap,
 711         .fsync                  = noop_fsync,
 712         .get_unmapped_area      = hugetlb_get_unmapped_area,
 713         .llseek         = default_llseek,
 714 };
 715
 716 static const struct inode_operations hugetlbfs_dir_inode_operations = {
 717         .create         = hugetlbfs_create,
 718         .lookup         = simple_lookup,
 719         .link           = simple_link,
 720         .unlink         = simple_unlink,
 721         .symlink        = hugetlbfs_symlink,
 722         .mkdir          = hugetlbfs_mkdir,
 723         .rmdir          = simple_rmdir,
 724         .mknod          = hugetlbfs_mknod,
 725         .rename         = simple_rename,
 726         .setattr        = hugetlbfs_setattr,
 727 };
 728
 729 static const struct inode_operations hugetlbfs_inode_operations = {
 730         .setattr        = hugetlbfs_setattr,
 731 };
 732
 733 static const struct super_operations hugetlbfs_ops = {
 734         .alloc_inode    = hugetlbfs_alloc_inode,
 735         .destroy_inode  = hugetlbfs_destroy_inode,
 736         .evict_inode    = hugetlbfs_evict_inode,
 737         .statfs         = hugetlbfs_statfs,
 738         .put_super      = hugetlbfs_put_super,
 739         .show_options   = generic_show_options,
 740 };
 741
 742 static int
 743 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 744 {
 745         char *p, *rest;
 746         substring_t args[MAX_OPT_ARGS];
 747         int option;
 748         unsigned long long size = 0;
 749         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
 750
 751         if (!options)
 752                 return 0;
 753
 754         while ((p = strsep(&options, ",")) != NULL) {
 755                 int token;
 756                 if (!*p)
 757                         continue;
 758
 759                 token = match_token(p, tokens, args);
 760                 switch (token) {
 761                 case Opt_uid:
 762                         if (match_int(&args[0], &option))
 763                                 goto bad_val;
 764                         pconfig->uid = option;
 765                         break;
 766
 767                 case Opt_gid:
 768                         if (match_int(&args[0], &option))
 769                                 goto bad_val;
 770                         pconfig->gid = option;
 771                         break;
 772
 773                 case Opt_mode:
 774                         if (match_octal(&args[0], &option))
 775                                 goto bad_val;
 776                         pconfig->mode = option & 01777U;
 777                         break;
 778
 779                 case Opt_size: {
 780                         /* memparse() will accept a K/M/G without a digit */
 781                         if (!isdigit(*args[0].from))
 782                                 goto bad_val;
 783                         size = memparse(args[0].from, &rest);
 784                         setsize = SIZE_STD;
 785                         if (*rest == '%')
 786                                 setsize = SIZE_PERCENT;
 787                         break;
 788                 }
 789
 790                 case Opt_nr_inodes:
 791                         /* memparse() will accept a K/M/G without a digit */
 792                         if (!isdigit(*args[0].from))
 793                                 goto bad_val;
 794                         pconfig->nr_inodes = memparse(args[0].from, &rest);
 795                         break;
 796
 797                 case Opt_pagesize: {
 798                         unsigned long ps;
 799                         ps = memparse(args[0].from, &rest);
 800                         pconfig->hstate = size_to_hstate(ps);
 801                         if (!pconfig->hstate) {
 802                                 printk(KERN_ERR
 803                                 "hugetlbfs: Unsupported page size %lu MB\n",
 804                                         ps >> 20);
 805                                 return -EINVAL;
 806                         }
 807                         break;
 808                 }
 809
 810                 default:
 811                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
 812                                  p);
 813                         return -EINVAL;
 814                         break;
 815                 }
 816         }
 817
 818         /* Do size after hstate is set up */
 819         if (setsize > NO_SIZE) {
 820                 struct hstate *h = pconfig->hstate;
 821                 if (setsize == SIZE_PERCENT) {
 822                         size <<= huge_page_shift(h);
 823                         size *= h->max_huge_pages;
 824                         do_div(size, 100);
 825                 }
 826                 pconfig->nr_blocks = (size >> huge_page_shift(h));
 827         }
 828
 829         return 0;
 830
 831 bad_val:
 832         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
 833                args[0].from, p);
 834         return -EINVAL;
 835 }
 836
 837 static int
 838 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 839 {
 840         struct inode * inode;
 841         struct dentry * root;
 842         int ret;
 843         struct hugetlbfs_config config;
 844         struct hugetlbfs_sb_info *sbinfo;
 845
 846         save_mount_options(sb, data);
 847
 848         config.nr_blocks = -1; /* No limit on size by default */
 849         config.nr_inodes = -1; /* No limit on number of inodes by default */
 850         config.uid = current_fsuid();
 851         config.gid = current_fsgid();
 852         config.mode = 0755;
 853         config.hstate = &default_hstate;
 854         ret = hugetlbfs_parse_options(data, &config);
 855         if (ret)
 856                 return ret;
 857
 858         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 859         if (!sbinfo)
 860                 return -ENOMEM;
 861         sb->s_fs_info = sbinfo;
 862         sbinfo->hstate = config.hstate;
 863         spin_lock_init(&sbinfo->stat_lock);
 864         sbinfo->max_inodes = config.nr_inodes;
 865         sbinfo->free_inodes = config.nr_inodes;
 866         sbinfo->spool = NULL;
 867         if (config.nr_blocks != -1) {
 868                 sbinfo->spool = hugepage_new_subpool(config.nr_blocks);
 869                 if (!sbinfo->spool)
 870                         goto out_free;
 871         }
 872         sb->s_maxbytes = MAX_LFS_FILESIZE;
 873         sb->s_blocksize = huge_page_size(config.hstate);
 874         sb->s_blocksize_bits = huge_page_shift(config.hstate);
 875         sb->s_magic = HUGETLBFS_MAGIC;
 876         sb->s_op = &hugetlbfs_ops;
 877         sb->s_time_gran = 1;
 878         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 879                                         S_IFDIR | config.mode, 0);
 880         if (!inode)
 881                 goto out_free;
 882
 883         root = d_alloc_root(inode);
 884         if (!root) {
 885                 iput(inode);
 886                 goto out_free;
 887         }
 888         sb->s_root = root;
 889         return 0;
 890 out_free:
 891         if (sbinfo->spool)
 892                 kfree(sbinfo->spool);
 893         kfree(sbinfo);
 894         return -ENOMEM;
 895 }
 896
 897 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
 898         int flags, const char *dev_name, void *data)
 899 {
 900         return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 901 }
 902
 903 static struct file_system_type hugetlbfs_fs_type = {
 904         .name           = "hugetlbfs",
 905         .mount          = hugetlbfs_mount,
 906         .kill_sb        = kill_litter_super,
 907 };
 908
 909 static struct vfsmount *hugetlbfs_vfsmount;
 910
 911 static int can_do_hugetlb_shm(void)
 912 {
 913         return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
 914 }
 915
 916 struct file *hugetlb_file_setup(const char *name, size_t size,
 917                                 vm_flags_t acctflag,
 918                                 struct user_struct **user, int creat_flags)
 919 {
 920         int error = -ENOMEM;
 921         struct file *file;
 922         struct inode *inode;
 923         struct path path;
 924         struct dentry *root;
 925         struct qstr quick_string;
 926
 927         *user = NULL;
 928         if (!hugetlbfs_vfsmount)
 929                 return ERR_PTR(-ENOENT);
 930
 931         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
 932                 *user = current_user();
 933                 if (user_shm_lock(size, *user)) {
 934                         printk_once(KERN_WARNING "Using mlock ulimits for SHM_HUGETLB is deprecated\n");
 935                 } else {
 936                         *user = NULL;
 937                         return ERR_PTR(-EPERM);
 938                 }
 939         }
 940
 941         root = hugetlbfs_vfsmount->mnt_root;
 942         quick_string.name = name;
 943         quick_string.len = strlen(quick_string.name);
 944         quick_string.hash = 0;
 945         path.dentry = d_alloc(root, &quick_string);
 946         if (!path.dentry)
 947                 goto out_shm_unlock;
 948
 949         path.mnt = mntget(hugetlbfs_vfsmount);
 950         error = -ENOSPC;
 951         inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
 952                                 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
 953         if (!inode)
 954                 goto out_dentry;
 955
 956         error = -ENOMEM;
 957         if (hugetlb_reserve_pages(inode, 0,
 958                         size >> huge_page_shift(hstate_inode(inode)), NULL,
 959                         acctflag))
 960                 goto out_inode;
 961
 962         d_instantiate(path.dentry, inode);
 963         inode->i_size = size;
 964         clear_nlink(inode);
 965
 966         error = -ENFILE;
 967         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
 968                         &hugetlbfs_file_operations);
 969         if (!file)
 970                 goto out_dentry; /* inode is already attached */
 971
 972         return file;
 973
 974 out_inode:
 975         iput(inode);
 976 out_dentry:
 977         path_put(&path);
 978 out_shm_unlock:
 979         if (*user) {
 980                 user_shm_unlock(size, *user);
 981                 *user = NULL;
 982         }
 983         return ERR_PTR(error);
 984 }
 985
 986 static int __init init_hugetlbfs_fs(void)
 987 {
 988         int error;
 989         struct vfsmount *vfsmount;
 990
 991         if (!hugepages_supported()) {
 992                 pr_info("hugetlbfs: disabling because there are no supported hugepage sizes\n");
 993                 return -ENOTSUPP;
 994         }
 995
 996         error = bdi_init(&hugetlbfs_backing_dev_info);
 997         if (error)
 998                 return error;
 999
1000         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1001                                         sizeof(struct hugetlbfs_inode_info),
1002                                         0, 0, init_once);
1003         if (hugetlbfs_inode_cachep == NULL)
1004                 goto out2;
1005
1006         error = register_filesystem(&hugetlbfs_fs_type);
1007         if (error)
1008                 goto out;
1009
1010         vfsmount = kern_mount(&hugetlbfs_fs_type);
1011
1012         if (!IS_ERR(vfsmount)) {
1013                 hugetlbfs_vfsmount = vfsmount;
1014                 return 0;
1015         }
1016
1017         error = PTR_ERR(vfsmount);
1018
1019  out:
1020         if (error)
1021                 kmem_cache_destroy(hugetlbfs_inode_cachep);
1022  out2:
1023         bdi_destroy(&hugetlbfs_backing_dev_info);
1024         return error;
1025 }
1026
1027 static void __exit exit_hugetlbfs_fs(void)
1028 {
1029         kmem_cache_destroy(hugetlbfs_inode_cachep);
1030         kern_unmount(hugetlbfs_vfsmount);
1031         unregister_filesystem(&hugetlbfs_fs_type);
1032         bdi_destroy(&hugetlbfs_backing_dev_info);
1033 }
1034
1035 module_init(init_hugetlbfs_fs)
1036 module_exit(exit_hugetlbfs_fs)
1037
1038 MODULE_LICENSE("GPL");