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         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         delete_from_page_cache(page);
 336 }
 337
 338 static void truncate_hugepages(struct inode *inode, loff_t lstart)
 339 {
 340         struct hstate *h = hstate_inode(inode);
 341         struct address_space *mapping = &inode->i_data;
 342         const pgoff_t start = lstart >> huge_page_shift(h);
 343         struct pagevec pvec;
 344         pgoff_t next;
 345         int i, freed = 0;
 346
 347         pagevec_init(&pvec, 0);
 348         next = start;
 349         while (1) {
 350                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 351                         if (next == start)
 352                                 break;
 353                         next = start;
 354                         continue;
 355                 }
 356
 357                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 358                         struct page *page = pvec.pages[i];
 359
 360                         lock_page(page);
 361                         if (page->index > next)
 362                                 next = page->index;
 363                         ++next;
 364                         truncate_huge_page(page);
 365                         unlock_page(page);
 366                         freed++;
 367                 }
 368                 huge_pagevec_release(&pvec);
 369         }
 370         BUG_ON(!lstart && mapping->nrpages);
 371         hugetlb_unreserve_pages(inode, start, freed);
 372 }
 373
 374 static void hugetlbfs_evict_inode(struct inode *inode)
 375 {
 376         truncate_hugepages(inode, 0);
 377         end_writeback(inode);
 378 }
 379
 380 static inline void
 381 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
 382 {
 383         struct vm_area_struct *vma;
 384         struct prio_tree_iter iter;
 385
 386         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
 387                 unsigned long v_offset;
 388
 389                 /*
 390                  * Can the expression below overflow on 32-bit arches?
 391                  * No, because the prio_tree returns us only those vmas
 392                  * which overlap the truncated area starting at pgoff,
 393                  * and no vma on a 32-bit arch can span beyond the 4GB.
 394                  */
 395                 if (vma->vm_pgoff < pgoff)
 396                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 397                 else
 398                         v_offset = 0;
 399
 400                 __unmap_hugepage_range(vma,
 401                                 vma->vm_start + v_offset, vma->vm_end, NULL);
 402         }
 403 }
 404
 405 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 406 {
 407         pgoff_t pgoff;
 408         struct address_space *mapping = inode->i_mapping;
 409         struct hstate *h = hstate_inode(inode);
 410
 411         BUG_ON(offset & ~huge_page_mask(h));
 412         pgoff = offset >> PAGE_SHIFT;
 413
 414         i_size_write(inode, offset);
 415         mutex_lock(&mapping->i_mmap_mutex);
 416         if (!prio_tree_empty(&mapping->i_mmap))
 417                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 418         mutex_unlock(&mapping->i_mmap_mutex);
 419         truncate_hugepages(inode, offset);
 420         return 0;
 421 }
 422
 423 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 424 {
 425         struct inode *inode = dentry->d_inode;
 426         struct hstate *h = hstate_inode(inode);
 427         int error;
 428         unsigned int ia_valid = attr->ia_valid;
 429
 430         BUG_ON(!inode);
 431
 432         error = inode_change_ok(inode, attr);
 433         if (error)
 434                 return error;
 435
 436         if (ia_valid & ATTR_SIZE) {
 437                 error = -EINVAL;
 438                 if (attr->ia_size & ~huge_page_mask(h))
 439                         return -EINVAL;
 440                 error = hugetlb_vmtruncate(inode, attr->ia_size);
 441                 if (error)
 442                         return error;
 443         }
 444
 445         setattr_copy(inode, attr);
 446         mark_inode_dirty(inode);
 447         return 0;
 448 }
 449
 450 static struct inode *hugetlbfs_get_root(struct super_block *sb,
 451                                         struct hugetlbfs_config *config)
 452 {
 453         struct inode *inode;
 454
 455         inode = new_inode(sb);
 456         if (inode) {
 457                 struct hugetlbfs_inode_info *info;
 458                 inode->i_ino = get_next_ino();
 459                 inode->i_mode = S_IFDIR | config->mode;
 460                 inode->i_uid = config->uid;
 461                 inode->i_gid = config->gid;
 462                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 463                 info = HUGETLBFS_I(inode);
 464                 mpol_shared_policy_init(&info->policy, NULL);
 465                 inode->i_op = &hugetlbfs_dir_inode_operations;
 466                 inode->i_fop = &simple_dir_operations;
 467                 /* directory inodes start off with i_nlink == 2 (for "." entry) */
 468                 inc_nlink(inode);
 469         }
 470         return inode;
 471 }
 472
 473 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
 474                                         struct inode *dir,
 475                                         umode_t mode, dev_t dev)
 476 {
 477         struct inode *inode;
 478
 479         inode = new_inode(sb);
 480         if (inode) {
 481                 struct hugetlbfs_inode_info *info;
 482                 inode->i_ino = get_next_ino();
 483                 inode_init_owner(inode, dir, mode);
 484                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 485                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 486                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 487                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
 488                 info = HUGETLBFS_I(inode);
 489                 /*
 490                  * The policy is initialized here even if we are creating a
 491                  * private inode because initialization simply creates an
 492                  * an empty rb tree and calls spin_lock_init(), later when we
 493                  * call mpol_free_shared_policy() it will just return because
 494                  * the rb tree will still be empty.
 495                  */
 496                 mpol_shared_policy_init(&info->policy, NULL);
 497                 switch (mode & S_IFMT) {
 498                 default:
 499                         init_special_inode(inode, mode, dev);
 500                         break;
 501                 case S_IFREG:
 502                         inode->i_op = &hugetlbfs_inode_operations;
 503                         inode->i_fop = &hugetlbfs_file_operations;
 504                         break;
 505                 case S_IFDIR:
 506                         inode->i_op = &hugetlbfs_dir_inode_operations;
 507                         inode->i_fop = &simple_dir_operations;
 508
 509                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 510                         inc_nlink(inode);
 511                         break;
 512                 case S_IFLNK:
 513                         inode->i_op = &page_symlink_inode_operations;
 514                         break;
 515                 }
 516                 lockdep_annotate_inode_mutex_key(inode);
 517         }
 518         return inode;
 519 }
 520
 521 /*
 522  * File creation. Allocate an inode, and we're done..
 523  */
 524 static int hugetlbfs_mknod(struct inode *dir,
 525                         struct dentry *dentry, umode_t mode, dev_t dev)
 526 {
 527         struct inode *inode;
 528         int error = -ENOSPC;
 529
 530         inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
 531         if (inode) {
 532                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 533                 d_instantiate(dentry, inode);
 534                 dget(dentry);   /* Extra count - pin the dentry in core */
 535                 error = 0;
 536         }
 537         return error;
 538 }
 539
 540 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 541 {
 542         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 543         if (!retval)
 544                 inc_nlink(dir);
 545         return retval;
 546 }
 547
 548 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, struct nameidata *nd)
 549 {
 550         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 551 }
 552
 553 static int hugetlbfs_symlink(struct inode *dir,
 554                         struct dentry *dentry, const char *symname)
 555 {
 556         struct inode *inode;
 557         int error = -ENOSPC;
 558
 559         inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
 560         if (inode) {
 561                 int l = strlen(symname)+1;
 562                 error = page_symlink(inode, symname, l);
 563                 if (!error) {
 564                         d_instantiate(dentry, inode);
 565                         dget(dentry);
 566                 } else
 567                         iput(inode);
 568         }
 569         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 570
 571         return error;
 572 }
 573
 574 /*
 575  * mark the head page dirty
 576  */
 577 static int hugetlbfs_set_page_dirty(struct page *page)
 578 {
 579         struct page *head = compound_head(page);
 580
 581         SetPageDirty(head);
 582         return 0;
 583 }
 584
 585 static int hugetlbfs_migrate_page(struct address_space *mapping,
 586                                 struct page *newpage, struct page *page)
 587 {
 588         int rc;
 589
 590         rc = migrate_huge_page_move_mapping(mapping, newpage, page);
 591         if (rc)
 592                 return rc;
 593         migrate_page_copy(newpage, page);
 594
 595         return 0;
 596 }
 597
 598 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 599 {
 600         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 601         struct hstate *h = hstate_inode(dentry->d_inode);
 602
 603         buf->f_type = HUGETLBFS_MAGIC;
 604         buf->f_bsize = huge_page_size(h);
 605         if (sbinfo) {
 606                 spin_lock(&sbinfo->stat_lock);
 607                 /* If no limits set, just report 0 for max/free/used
 608                  * blocks, like simple_statfs() */
 609                 if (sbinfo->max_blocks >= 0) {
 610                         buf->f_blocks = sbinfo->max_blocks;
 611                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 612                         buf->f_files = sbinfo->max_inodes;
 613                         buf->f_ffree = sbinfo->free_inodes;
 614                 }
 615                 spin_unlock(&sbinfo->stat_lock);
 616         }
 617         buf->f_namelen = NAME_MAX;
 618         return 0;
 619 }
 620
 621 static void hugetlbfs_put_super(struct super_block *sb)
 622 {
 623         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 624
 625         if (sbi) {
 626                 sb->s_fs_info = NULL;
 627                 kfree(sbi);
 628         }
 629 }
 630
 631 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 632 {
 633         if (sbinfo->free_inodes >= 0) {
 634                 spin_lock(&sbinfo->stat_lock);
 635                 if (unlikely(!sbinfo->free_inodes)) {
 636                         spin_unlock(&sbinfo->stat_lock);
 637                         return 0;
 638                 }
 639                 sbinfo->free_inodes--;
 640                 spin_unlock(&sbinfo->stat_lock);
 641         }
 642
 643         return 1;
 644 }
 645
 646 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 647 {
 648         if (sbinfo->free_inodes >= 0) {
 649                 spin_lock(&sbinfo->stat_lock);
 650                 sbinfo->free_inodes++;
 651                 spin_unlock(&sbinfo->stat_lock);
 652         }
 653 }
 654
 655
 656 static struct kmem_cache *hugetlbfs_inode_cachep;
 657
 658 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 659 {
 660         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 661         struct hugetlbfs_inode_info *p;
 662
 663         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 664                 return NULL;
 665         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 666         if (unlikely(!p)) {
 667                 hugetlbfs_inc_free_inodes(sbinfo);
 668                 return NULL;
 669         }
 670         return &p->vfs_inode;
 671 }
 672
 673 static void hugetlbfs_i_callback(struct rcu_head *head)
 674 {
 675         struct inode *inode = container_of(head, struct inode, i_rcu);
 676         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 677 }
 678
 679 static void hugetlbfs_destroy_inode(struct inode *inode)
 680 {
 681         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 682         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 683         call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
 684 }
 685
 686 static const struct address_space_operations hugetlbfs_aops = {
 687         .write_begin    = hugetlbfs_write_begin,
 688         .write_end      = hugetlbfs_write_end,
 689         .set_page_dirty = hugetlbfs_set_page_dirty,
 690         .migratepage    = hugetlbfs_migrate_page,
 691 };
 692
 693
 694 static void init_once(void *foo)
 695 {
 696         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 697
 698         inode_init_once(&ei->vfs_inode);
 699 }
 700
 701 const struct file_operations hugetlbfs_file_operations = {
 702         .read                   = hugetlbfs_read,
 703         .mmap                   = hugetlbfs_file_mmap,
 704         .fsync                  = noop_fsync,
 705         .get_unmapped_area      = hugetlb_get_unmapped_area,
 706         .llseek         = default_llseek,
 707 };
 708
 709 static const struct inode_operations hugetlbfs_dir_inode_operations = {
 710         .create         = hugetlbfs_create,
 711         .lookup         = simple_lookup,
 712         .link           = simple_link,
 713         .unlink         = simple_unlink,
 714         .symlink        = hugetlbfs_symlink,
 715         .mkdir          = hugetlbfs_mkdir,
 716         .rmdir          = simple_rmdir,
 717         .mknod          = hugetlbfs_mknod,
 718         .rename         = simple_rename,
 719         .setattr        = hugetlbfs_setattr,
 720 };
 721
 722 static const struct inode_operations hugetlbfs_inode_operations = {
 723         .setattr        = hugetlbfs_setattr,
 724 };
 725
 726 static const struct super_operations hugetlbfs_ops = {
 727         .alloc_inode    = hugetlbfs_alloc_inode,
 728         .destroy_inode  = hugetlbfs_destroy_inode,
 729         .evict_inode    = hugetlbfs_evict_inode,
 730         .statfs         = hugetlbfs_statfs,
 731         .put_super      = hugetlbfs_put_super,
 732         .show_options   = generic_show_options,
 733 };
 734
 735 static int
 736 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 737 {
 738         char *p, *rest;
 739         substring_t args[MAX_OPT_ARGS];
 740         int option;
 741         unsigned long long size = 0;
 742         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
 743
 744         if (!options)
 745                 return 0;
 746
 747         while ((p = strsep(&options, ",")) != NULL) {
 748                 int token;
 749                 if (!*p)
 750                         continue;
 751
 752                 token = match_token(p, tokens, args);
 753                 switch (token) {
 754                 case Opt_uid:
 755                         if (match_int(&args[0], &option))
 756                                 goto bad_val;
 757                         pconfig->uid = option;
 758                         break;
 759
 760                 case Opt_gid:
 761                         if (match_int(&args[0], &option))
 762                                 goto bad_val;
 763                         pconfig->gid = option;
 764                         break;
 765
 766                 case Opt_mode:
 767                         if (match_octal(&args[0], &option))
 768                                 goto bad_val;
 769                         pconfig->mode = option & 01777U;
 770                         break;
 771
 772                 case Opt_size: {
 773                         /* memparse() will accept a K/M/G without a digit */
 774                         if (!isdigit(*args[0].from))
 775                                 goto bad_val;
 776                         size = memparse(args[0].from, &rest);
 777                         setsize = SIZE_STD;
 778                         if (*rest == '%')
 779                                 setsize = SIZE_PERCENT;
 780                         break;
 781                 }
 782
 783                 case Opt_nr_inodes:
 784                         /* memparse() will accept a K/M/G without a digit */
 785                         if (!isdigit(*args[0].from))
 786                                 goto bad_val;
 787                         pconfig->nr_inodes = memparse(args[0].from, &rest);
 788                         break;
 789
 790                 case Opt_pagesize: {
 791                         unsigned long ps;
 792                         ps = memparse(args[0].from, &rest);
 793                         pconfig->hstate = size_to_hstate(ps);
 794                         if (!pconfig->hstate) {
 795                                 printk(KERN_ERR
 796                                 "hugetlbfs: Unsupported page size %lu MB\n",
 797                                         ps >> 20);
 798                                 return -EINVAL;
 799                         }
 800                         break;
 801                 }
 802
 803                 default:
 804                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
 805                                  p);
 806                         return -EINVAL;
 807                         break;
 808                 }
 809         }
 810
 811         /* Do size after hstate is set up */
 812         if (setsize > NO_SIZE) {
 813                 struct hstate *h = pconfig->hstate;
 814                 if (setsize == SIZE_PERCENT) {
 815                         size <<= huge_page_shift(h);
 816                         size *= h->max_huge_pages;
 817                         do_div(size, 100);
 818                 }
 819                 pconfig->nr_blocks = (size >> huge_page_shift(h));
 820         }
 821
 822         return 0;
 823
 824 bad_val:
 825         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
 826                args[0].from, p);
 827         return -EINVAL;
 828 }
 829
 830 static int
 831 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 832 {
 833         struct inode * inode;
 834         struct dentry * root;
 835         int ret;
 836         struct hugetlbfs_config config;
 837         struct hugetlbfs_sb_info *sbinfo;
 838
 839         save_mount_options(sb, data);
 840
 841         config.nr_blocks = -1; /* No limit on size by default */
 842         config.nr_inodes = -1; /* No limit on number of inodes by default */
 843         config.uid = current_fsuid();
 844         config.gid = current_fsgid();
 845         config.mode = 0755;
 846         config.hstate = &default_hstate;
 847         ret = hugetlbfs_parse_options(data, &config);
 848         if (ret)
 849                 return ret;
 850
 851         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 852         if (!sbinfo)
 853                 return -ENOMEM;
 854         sb->s_fs_info = sbinfo;
 855         sbinfo->hstate = config.hstate;
 856         spin_lock_init(&sbinfo->stat_lock);
 857         sbinfo->max_blocks = config.nr_blocks;
 858         sbinfo->free_blocks = config.nr_blocks;
 859         sbinfo->max_inodes = config.nr_inodes;
 860         sbinfo->free_inodes = config.nr_inodes;
 861         sb->s_maxbytes = MAX_LFS_FILESIZE;
 862         sb->s_blocksize = huge_page_size(config.hstate);
 863         sb->s_blocksize_bits = huge_page_shift(config.hstate);
 864         sb->s_magic = HUGETLBFS_MAGIC;
 865         sb->s_op = &hugetlbfs_ops;
 866         sb->s_time_gran = 1;
 867         inode = hugetlbfs_get_root(sb, &config);
 868         if (!inode)
 869                 goto out_free;
 870
 871         root = d_alloc_root(inode);
 872         if (!root) {
 873                 iput(inode);
 874                 goto out_free;
 875         }
 876         sb->s_root = root;
 877         return 0;
 878 out_free:
 879         kfree(sbinfo);
 880         return -ENOMEM;
 881 }
 882
 883 int hugetlb_get_quota(struct address_space *mapping, long delta)
 884 {
 885         int ret = 0;
 886         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 887
 888         if (sbinfo->free_blocks > -1) {
 889                 spin_lock(&sbinfo->stat_lock);
 890                 if (sbinfo->free_blocks - delta >= 0)
 891                         sbinfo->free_blocks -= delta;
 892                 else
 893                         ret = -ENOMEM;
 894                 spin_unlock(&sbinfo->stat_lock);
 895         }
 896
 897         return ret;
 898 }
 899
 900 void hugetlb_put_quota(struct address_space *mapping, long delta)
 901 {
 902         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 903
 904         if (sbinfo->free_blocks > -1) {
 905                 spin_lock(&sbinfo->stat_lock);
 906                 sbinfo->free_blocks += delta;
 907                 spin_unlock(&sbinfo->stat_lock);
 908         }
 909 }
 910
 911 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
 912         int flags, const char *dev_name, void *data)
 913 {
 914         return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 915 }
 916
 917 static struct file_system_type hugetlbfs_fs_type = {
 918         .name           = "hugetlbfs",
 919         .mount          = hugetlbfs_mount,
 920         .kill_sb        = kill_litter_super,
 921 };
 922
 923 static struct vfsmount *hugetlbfs_vfsmount;
 924
 925 static int can_do_hugetlb_shm(void)
 926 {
 927         return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
 928 }
 929
 930 struct file *hugetlb_file_setup(const char *name, size_t size,
 931                                 vm_flags_t acctflag,
 932                                 struct user_struct **user, int creat_flags)
 933 {
 934         int error = -ENOMEM;
 935         struct file *file;
 936         struct inode *inode;
 937         struct path path;
 938         struct dentry *root;
 939         struct qstr quick_string;
 940
 941         *user = NULL;
 942         if (!hugetlbfs_vfsmount)
 943                 return ERR_PTR(-ENOENT);
 944
 945         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
 946                 *user = current_user();
 947                 if (user_shm_lock(size, *user)) {
 948                         printk_once(KERN_WARNING "Using mlock ulimits for SHM_HUGETLB is deprecated\n");
 949                 } else {
 950                         *user = NULL;
 951                         return ERR_PTR(-EPERM);
 952                 }
 953         }
 954
 955         root = hugetlbfs_vfsmount->mnt_root;
 956         quick_string.name = name;
 957         quick_string.len = strlen(quick_string.name);
 958         quick_string.hash = 0;
 959         path.dentry = d_alloc(root, &quick_string);
 960         if (!path.dentry)
 961                 goto out_shm_unlock;
 962
 963         path.mnt = mntget(hugetlbfs_vfsmount);
 964         error = -ENOSPC;
 965         inode = hugetlbfs_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0);
 966         if (!inode)
 967                 goto out_dentry;
 968
 969         error = -ENOMEM;
 970         if (hugetlb_reserve_pages(inode, 0,
 971                         size >> huge_page_shift(hstate_inode(inode)), NULL,
 972                         acctflag))
 973                 goto out_inode;
 974
 975         d_instantiate(path.dentry, inode);
 976         inode->i_size = size;
 977         clear_nlink(inode);
 978
 979         error = -ENFILE;
 980         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
 981                         &hugetlbfs_file_operations);
 982         if (!file)
 983                 goto out_dentry; /* inode is already attached */
 984
 985         return file;
 986
 987 out_inode:
 988         iput(inode);
 989 out_dentry:
 990         path_put(&path);
 991 out_shm_unlock:
 992         if (*user) {
 993                 user_shm_unlock(size, *user);
 994                 *user = NULL;
 995         }
 996         return ERR_PTR(error);
 997 }
 998
 999 static int __init init_hugetlbfs_fs(void)
1000 {
1001         int error;
1002         struct vfsmount *vfsmount;
1003
1004         error = bdi_init(&hugetlbfs_backing_dev_info);
1005         if (error)
1006                 return error;
1007
1008         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1009                                         sizeof(struct hugetlbfs_inode_info),
1010                                         0, 0, init_once);
1011         if (hugetlbfs_inode_cachep == NULL)
1012                 goto out2;
1013
1014         error = register_filesystem(&hugetlbfs_fs_type);
1015         if (error)
1016                 goto out;
1017
1018         vfsmount = kern_mount(&hugetlbfs_fs_type);
1019
1020         if (!IS_ERR(vfsmount)) {
1021                 hugetlbfs_vfsmount = vfsmount;
1022                 return 0;
1023         }
1024
1025         error = PTR_ERR(vfsmount);
1026
1027  out:
1028         if (error)
1029                 kmem_cache_destroy(hugetlbfs_inode_cachep);
1030  out2:
1031         bdi_destroy(&hugetlbfs_backing_dev_info);
1032         return error;
1033 }
1034
1035 static void __exit exit_hugetlbfs_fs(void)
1036 {
1037         kmem_cache_destroy(hugetlbfs_inode_cachep);
1038         kern_unmount(hugetlbfs_vfsmount);
1039         unregister_filesystem(&hugetlbfs_fs_type);
1040         bdi_destroy(&hugetlbfs_backing_dev_info);
1041 }
1042
1043 module_init(init_hugetlbfs_fs)
1044 module_exit(exit_hugetlbfs_fs)
1045
1046 MODULE_LICENSE("GPL");