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/ima.h>
  34
  35 #include <asm/uaccess.h>
  36
  37 /* some random number */
  38 #define HUGETLBFS_MAGIC 0x958458f6
  39
  40 static const struct super_operations hugetlbfs_ops;
  41 static const struct address_space_operations hugetlbfs_aops;
  42 const struct file_operations hugetlbfs_file_operations;
  43 static const struct inode_operations hugetlbfs_dir_inode_operations;
  44 static const struct inode_operations hugetlbfs_inode_operations;
  45
  46 static struct backing_dev_info hugetlbfs_backing_dev_info = {
  47         .name           = "hugetlbfs",
  48         .ra_pages       = 0,    /* No readahead */
  49         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
  50 };
  51
  52 int sysctl_hugetlb_shm_group;
  53
  54 enum {
  55         Opt_size, Opt_nr_inodes,
  56         Opt_mode, Opt_uid, Opt_gid,
  57         Opt_pagesize,
  58         Opt_err,
  59 };
  60
  61 static const match_table_t tokens = {
  62         {Opt_size,      "size=%s"},
  63         {Opt_nr_inodes, "nr_inodes=%s"},
  64         {Opt_mode,      "mode=%o"},
  65         {Opt_uid,       "uid=%u"},
  66         {Opt_gid,       "gid=%u"},
  67         {Opt_pagesize,  "pagesize=%s"},
  68         {Opt_err,       NULL},
  69 };
  70
  71 static void huge_pagevec_release(struct pagevec *pvec)
  72 {
  73         int i;
  74
  75         for (i = 0; i < pagevec_count(pvec); ++i)
  76                 put_page(pvec->pages[i]);
  77
  78         pagevec_reinit(pvec);
  79 }
  80
  81 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  82 {
  83         struct inode *inode = file->f_path.dentry->d_inode;
  84         loff_t len, vma_len;
  85         int ret;
  86         struct hstate *h = hstate_file(file);
  87
  88         /*
  89          * vma address alignment (but not the pgoff alignment) has
  90          * already been checked by prepare_hugepage_range.  If you add
  91          * any error returns here, do so after setting VM_HUGETLB, so
  92          * is_vm_hugetlb_page tests below unmap_region go the right
  93          * way when do_mmap_pgoff unwinds (may be important on powerpc
  94          * and ia64).
  95          */
  96         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
  97         vma->vm_ops = &hugetlb_vm_ops;
  98
  99         if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
 100                 return -EINVAL;
 101
 102         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 103
 104         mutex_lock(&inode->i_mutex);
 105         file_accessed(file);
 106
 107         ret = -ENOMEM;
 108         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 109
 110         if (hugetlb_reserve_pages(inode,
 111                                 vma->vm_pgoff >> huge_page_order(h),
 112                                 len >> huge_page_shift(h), vma,
 113                                 vma->vm_flags))
 114                 goto out;
 115
 116         ret = 0;
 117         hugetlb_prefault_arch_hook(vma->vm_mm);
 118         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 119                 inode->i_size = len;
 120 out:
 121         mutex_unlock(&inode->i_mutex);
 122
 123         return ret;
 124 }
 125
 126 /*
 127  * Called under down_write(mmap_sem).
 128  */
 129
 130 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 131 static unsigned long
 132 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 133                 unsigned long len, unsigned long pgoff, unsigned long flags)
 134 {
 135         struct mm_struct *mm = current->mm;
 136         struct vm_area_struct *vma;
 137         unsigned long start_addr;
 138         struct hstate *h = hstate_file(file);
 139
 140         if (len & ~huge_page_mask(h))
 141                 return -EINVAL;
 142         if (len > TASK_SIZE)
 143                 return -ENOMEM;
 144
 145         if (flags & MAP_FIXED) {
 146                 if (prepare_hugepage_range(file, addr, len))
 147                         return -EINVAL;
 148                 return addr;
 149         }
 150
 151         if (addr) {
 152                 addr = ALIGN(addr, huge_page_size(h));
 153                 vma = find_vma(mm, addr);
 154                 if (TASK_SIZE - len >= addr &&
 155                     (!vma || addr + len <= vma->vm_start))
 156                         return addr;
 157         }
 158
 159         start_addr = mm->free_area_cache;
 160
 161         if (len <= mm->cached_hole_size)
 162                 start_addr = TASK_UNMAPPED_BASE;
 163
 164 full_search:
 165         addr = ALIGN(start_addr, huge_page_size(h));
 166
 167         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 168                 /* At this point:  (!vma || addr < vma->vm_end). */
 169                 if (TASK_SIZE - len < addr) {
 170                         /*
 171                          * Start a new search - just in case we missed
 172                          * some holes.
 173                          */
 174                         if (start_addr != TASK_UNMAPPED_BASE) {
 175                                 start_addr = TASK_UNMAPPED_BASE;
 176                                 goto full_search;
 177                         }
 178                         return -ENOMEM;
 179                 }
 180
 181                 if (!vma || addr + len <= vma->vm_start)
 182                         return addr;
 183                 addr = ALIGN(vma->vm_end, huge_page_size(h));
 184         }
 185 }
 186 #endif
 187
 188 static int
 189 hugetlbfs_read_actor(struct page *page, unsigned long offset,
 190                         char __user *buf, unsigned long count,
 191                         unsigned long size)
 192 {
 193         char *kaddr;
 194         unsigned long left, copied = 0;
 195         int i, chunksize;
 196
 197         if (size > count)
 198                 size = count;
 199
 200         /* Find which 4k chunk and offset with in that chunk */
 201         i = offset >> PAGE_CACHE_SHIFT;
 202         offset = offset & ~PAGE_CACHE_MASK;
 203
 204         while (size) {
 205                 chunksize = PAGE_CACHE_SIZE;
 206                 if (offset)
 207                         chunksize -= offset;
 208                 if (chunksize > size)
 209                         chunksize = size;
 210                 kaddr = kmap(&page[i]);
 211                 left = __copy_to_user(buf, kaddr + offset, chunksize);
 212                 kunmap(&page[i]);
 213                 if (left) {
 214                         copied += (chunksize - left);
 215                         break;
 216                 }
 217                 offset = 0;
 218                 size -= chunksize;
 219                 buf += chunksize;
 220                 copied += chunksize;
 221                 i++;
 222         }
 223         return copied ? copied : -EFAULT;
 224 }
 225
 226 /*
 227  * Support for read() - Find the page attached to f_mapping and copy out the
 228  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
 229  * since it has PAGE_CACHE_SIZE assumptions.
 230  */
 231 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
 232                               size_t len, loff_t *ppos)
 233 {
 234         struct hstate *h = hstate_file(filp);
 235         struct address_space *mapping = filp->f_mapping;
 236         struct inode *inode = mapping->host;
 237         unsigned long index = *ppos >> huge_page_shift(h);
 238         unsigned long offset = *ppos & ~huge_page_mask(h);
 239         unsigned long end_index;
 240         loff_t isize;
 241         ssize_t retval = 0;
 242
 243         mutex_lock(&inode->i_mutex);
 244
 245         /* validate length */
 246         if (len == 0)
 247                 goto out;
 248
 249         isize = i_size_read(inode);
 250         if (!isize)
 251                 goto out;
 252
 253         end_index = (isize - 1) >> huge_page_shift(h);
 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                 if (index >= end_index) {
 262                         if (index > end_index)
 263                                 goto out;
 264                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
 265                         if (nr <= offset) {
 266                                 goto out;
 267                         }
 268                 }
 269                 nr = nr - offset;
 270
 271                 /* Find the page */
 272                 page = find_get_page(mapping, index);
 273                 if (unlikely(page == NULL)) {
 274                         /*
 275                          * We have a HOLE, zero out the user-buffer for the
 276                          * length of the hole or request.
 277                          */
 278                         ret = len < nr ? len : nr;
 279                         if (clear_user(buf, ret))
 280                                 ra = -EFAULT;
 281                         else
 282                                 ra = 0;
 283                 } else {
 284                         /*
 285                          * We have the page, copy it to user space buffer.
 286                          */
 287                         ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
 288                         ret = ra;
 289                 }
 290                 if (ra < 0) {
 291                         if (retval == 0)
 292                                 retval = ra;
 293                         if (page)
 294                                 page_cache_release(page);
 295                         goto out;
 296                 }
 297
 298                 offset += ret;
 299                 retval += ret;
 300                 len -= ret;
 301                 index += offset >> huge_page_shift(h);
 302                 offset &= ~huge_page_mask(h);
 303
 304                 if (page)
 305                         page_cache_release(page);
 306
 307                 /* short read or no more work */
 308                 if ((ret != nr) || (len == 0))
 309                         break;
 310         }
 311 out:
 312         *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
 313         mutex_unlock(&inode->i_mutex);
 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         remove_from_page_cache(page);
 338         put_page(page);
 339 }
 340
 341 static void truncate_hugepages(struct inode *inode, loff_t lstart)
 342 {
 343         struct hstate *h = hstate_inode(inode);
 344         struct address_space *mapping = &inode->i_data;
 345         const pgoff_t start = lstart >> huge_page_shift(h);
 346         struct pagevec pvec;
 347         pgoff_t next;
 348         int i, freed = 0;
 349
 350         pagevec_init(&pvec, 0);
 351         next = start;
 352         while (1) {
 353                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 354                         if (next == start)
 355                                 break;
 356                         next = start;
 357                         continue;
 358                 }
 359
 360                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 361                         struct page *page = pvec.pages[i];
 362
 363                         lock_page(page);
 364                         if (page->index > next)
 365                                 next = page->index;
 366                         ++next;
 367                         truncate_huge_page(page);
 368                         unlock_page(page);
 369                         freed++;
 370                 }
 371                 huge_pagevec_release(&pvec);
 372         }
 373         BUG_ON(!lstart && mapping->nrpages);
 374         hugetlb_unreserve_pages(inode, start, freed);
 375 }
 376
 377 static void hugetlbfs_delete_inode(struct inode *inode)
 378 {
 379         truncate_hugepages(inode, 0);
 380         clear_inode(inode);
 381 }
 382
 383 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
 384 {
 385         struct super_block *sb = inode->i_sb;
 386
 387         if (!hlist_unhashed(&inode->i_hash)) {
 388                 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
 389                         list_move(&inode->i_list, &inode_unused);
 390                 inodes_stat.nr_unused++;
 391                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
 392                         spin_unlock(&inode_lock);
 393                         return;
 394                 }
 395                 inode->i_state |= I_WILL_FREE;
 396                 spin_unlock(&inode_lock);
 397                 /*
 398                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
 399                  * in our backing_dev_info.
 400                  */
 401                 write_inode_now(inode, 1);
 402                 spin_lock(&inode_lock);
 403                 inode->i_state &= ~I_WILL_FREE;
 404                 inodes_stat.nr_unused--;
 405                 hlist_del_init(&inode->i_hash);
 406         }
 407         list_del_init(&inode->i_list);
 408         list_del_init(&inode->i_sb_list);
 409         inode->i_state |= I_FREEING;
 410         inodes_stat.nr_inodes--;
 411         spin_unlock(&inode_lock);
 412         truncate_hugepages(inode, 0);
 413         clear_inode(inode);
 414         destroy_inode(inode);
 415 }
 416
 417 static void hugetlbfs_drop_inode(struct inode *inode)
 418 {
 419         if (!inode->i_nlink)
 420                 generic_delete_inode(inode);
 421         else
 422                 hugetlbfs_forget_inode(inode);
 423 }
 424
 425 static inline void
 426 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
 427 {
 428         struct vm_area_struct *vma;
 429         struct prio_tree_iter iter;
 430
 431         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
 432                 unsigned long v_offset;
 433
 434                 /*
 435                  * Can the expression below overflow on 32-bit arches?
 436                  * No, because the prio_tree returns us only those vmas
 437                  * which overlap the truncated area starting at pgoff,
 438                  * and no vma on a 32-bit arch can span beyond the 4GB.
 439                  */
 440                 if (vma->vm_pgoff < pgoff)
 441                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 442                 else
 443                         v_offset = 0;
 444
 445                 __unmap_hugepage_range(vma,
 446                                 vma->vm_start + v_offset, vma->vm_end, NULL);
 447         }
 448 }
 449
 450 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 451 {
 452         pgoff_t pgoff;
 453         struct address_space *mapping = inode->i_mapping;
 454         struct hstate *h = hstate_inode(inode);
 455
 456         BUG_ON(offset & ~huge_page_mask(h));
 457         pgoff = offset >> PAGE_SHIFT;
 458
 459         i_size_write(inode, offset);
 460         spin_lock(&mapping->i_mmap_lock);
 461         if (!prio_tree_empty(&mapping->i_mmap))
 462                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 463         spin_unlock(&mapping->i_mmap_lock);
 464         truncate_hugepages(inode, offset);
 465         return 0;
 466 }
 467
 468 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 469 {
 470         struct inode *inode = dentry->d_inode;
 471         struct hstate *h = hstate_inode(inode);
 472         int error;
 473         unsigned int ia_valid = attr->ia_valid;
 474
 475         BUG_ON(!inode);
 476
 477         error = inode_change_ok(inode, attr);
 478         if (error)
 479                 goto out;
 480
 481         if (ia_valid & ATTR_SIZE) {
 482                 error = -EINVAL;
 483                 if (!(attr->ia_size & ~huge_page_mask(h)))
 484                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 485                 if (error)
 486                         goto out;
 487                 attr->ia_valid &= ~ATTR_SIZE;
 488         }
 489         error = inode_setattr(inode, attr);
 490 out:
 491         return error;
 492 }
 493
 494 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 495                                         gid_t gid, int mode, dev_t dev)
 496 {
 497         struct inode *inode;
 498
 499         inode = new_inode(sb);
 500         if (inode) {
 501                 struct hugetlbfs_inode_info *info;
 502                 inode->i_mode = mode;
 503                 inode->i_uid = uid;
 504                 inode->i_gid = gid;
 505                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 506                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 507                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 508                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
 509                 info = HUGETLBFS_I(inode);
 510                 mpol_shared_policy_init(&info->policy, NULL);
 511                 switch (mode & S_IFMT) {
 512                 default:
 513                         init_special_inode(inode, mode, dev);
 514                         break;
 515                 case S_IFREG:
 516                         inode->i_op = &hugetlbfs_inode_operations;
 517                         inode->i_fop = &hugetlbfs_file_operations;
 518                         break;
 519                 case S_IFDIR:
 520                         inode->i_op = &hugetlbfs_dir_inode_operations;
 521                         inode->i_fop = &simple_dir_operations;
 522
 523                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 524                         inc_nlink(inode);
 525                         break;
 526                 case S_IFLNK:
 527                         inode->i_op = &page_symlink_inode_operations;
 528                         break;
 529                 }
 530         }
 531         return inode;
 532 }
 533
 534 /*
 535  * File creation. Allocate an inode, and we're done..
 536  */
 537 static int hugetlbfs_mknod(struct inode *dir,
 538                         struct dentry *dentry, int mode, dev_t dev)
 539 {
 540         struct inode *inode;
 541         int error = -ENOSPC;
 542         gid_t gid;
 543
 544         if (dir->i_mode & S_ISGID) {
 545                 gid = dir->i_gid;
 546                 if (S_ISDIR(mode))
 547                         mode |= S_ISGID;
 548         } else {
 549                 gid = current_fsgid();
 550         }
 551         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
 552         if (inode) {
 553                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 554                 d_instantiate(dentry, inode);
 555                 dget(dentry);   /* Extra count - pin the dentry in core */
 556                 error = 0;
 557         }
 558         return error;
 559 }
 560
 561 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 562 {
 563         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 564         if (!retval)
 565                 inc_nlink(dir);
 566         return retval;
 567 }
 568
 569 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 570 {
 571         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 572 }
 573
 574 static int hugetlbfs_symlink(struct inode *dir,
 575                         struct dentry *dentry, const char *symname)
 576 {
 577         struct inode *inode;
 578         int error = -ENOSPC;
 579         gid_t gid;
 580
 581         if (dir->i_mode & S_ISGID)
 582                 gid = dir->i_gid;
 583         else
 584                 gid = current_fsgid();
 585
 586         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
 587                                         gid, S_IFLNK|S_IRWXUGO, 0);
 588         if (inode) {
 589                 int l = strlen(symname)+1;
 590                 error = page_symlink(inode, symname, l);
 591                 if (!error) {
 592                         d_instantiate(dentry, inode);
 593                         dget(dentry);
 594                 } else
 595                         iput(inode);
 596         }
 597         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 598
 599         return error;
 600 }
 601
 602 /*
 603  * mark the head page dirty
 604  */
 605 static int hugetlbfs_set_page_dirty(struct page *page)
 606 {
 607         struct page *head = compound_head(page);
 608
 609         SetPageDirty(head);
 610         return 0;
 611 }
 612
 613 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 614 {
 615         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 616         struct hstate *h = hstate_inode(dentry->d_inode);
 617
 618         buf->f_type = HUGETLBFS_MAGIC;
 619         buf->f_bsize = huge_page_size(h);
 620         if (sbinfo) {
 621                 spin_lock(&sbinfo->stat_lock);
 622                 /* If no limits set, just report 0 for max/free/used
 623                  * blocks, like simple_statfs() */
 624                 if (sbinfo->max_blocks >= 0) {
 625                         buf->f_blocks = sbinfo->max_blocks;
 626                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 627                         buf->f_files = sbinfo->max_inodes;
 628                         buf->f_ffree = sbinfo->free_inodes;
 629                 }
 630                 spin_unlock(&sbinfo->stat_lock);
 631         }
 632         buf->f_namelen = NAME_MAX;
 633         return 0;
 634 }
 635
 636 static void hugetlbfs_put_super(struct super_block *sb)
 637 {
 638         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 639
 640         if (sbi) {
 641                 sb->s_fs_info = NULL;
 642                 kfree(sbi);
 643         }
 644 }
 645
 646 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 647 {
 648         if (sbinfo->free_inodes >= 0) {
 649                 spin_lock(&sbinfo->stat_lock);
 650                 if (unlikely(!sbinfo->free_inodes)) {
 651                         spin_unlock(&sbinfo->stat_lock);
 652                         return 0;
 653                 }
 654                 sbinfo->free_inodes--;
 655                 spin_unlock(&sbinfo->stat_lock);
 656         }
 657
 658         return 1;
 659 }
 660
 661 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 662 {
 663         if (sbinfo->free_inodes >= 0) {
 664                 spin_lock(&sbinfo->stat_lock);
 665                 sbinfo->free_inodes++;
 666                 spin_unlock(&sbinfo->stat_lock);
 667         }
 668 }
 669
 670
 671 static struct kmem_cache *hugetlbfs_inode_cachep;
 672
 673 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 674 {
 675         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 676         struct hugetlbfs_inode_info *p;
 677
 678         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 679                 return NULL;
 680         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 681         if (unlikely(!p)) {
 682                 hugetlbfs_inc_free_inodes(sbinfo);
 683                 return NULL;
 684         }
 685         return &p->vfs_inode;
 686 }
 687
 688 static void hugetlbfs_destroy_inode(struct inode *inode)
 689 {
 690         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 691         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 692         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 693 }
 694
 695 static const struct address_space_operations hugetlbfs_aops = {
 696         .write_begin    = hugetlbfs_write_begin,
 697         .write_end      = hugetlbfs_write_end,
 698         .set_page_dirty = hugetlbfs_set_page_dirty,
 699 };
 700
 701
 702 static void init_once(void *foo)
 703 {
 704         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 705
 706         inode_init_once(&ei->vfs_inode);
 707 }
 708
 709 const struct file_operations hugetlbfs_file_operations = {
 710         .read                   = hugetlbfs_read,
 711         .mmap                   = hugetlbfs_file_mmap,
 712         .fsync                  = simple_sync_file,
 713         .get_unmapped_area      = hugetlb_get_unmapped_area,
 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         .statfs         = hugetlbfs_statfs,
 737         .delete_inode   = hugetlbfs_delete_inode,
 738         .drop_inode     = hugetlbfs_drop_inode,
 739         .put_super      = hugetlbfs_put_super,
 740         .show_options   = generic_show_options,
 741 };
 742
 743 static int
 744 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 745 {
 746         char *p, *rest;
 747         substring_t args[MAX_OPT_ARGS];
 748         int option;
 749         unsigned long long size = 0;
 750         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
 751
 752         if (!options)
 753                 return 0;
 754
 755         while ((p = strsep(&options, ",")) != NULL) {
 756                 int token;
 757                 if (!*p)
 758                         continue;
 759
 760                 token = match_token(p, tokens, args);
 761                 switch (token) {
 762                 case Opt_uid:
 763                         if (match_int(&args[0], &option))
 764                                 goto bad_val;
 765                         pconfig->uid = option;
 766                         break;
 767
 768                 case Opt_gid:
 769                         if (match_int(&args[0], &option))
 770                                 goto bad_val;
 771                         pconfig->gid = option;
 772                         break;
 773
 774                 case Opt_mode:
 775                         if (match_octal(&args[0], &option))
 776                                 goto bad_val;
 777                         pconfig->mode = option & 01777U;
 778                         break;
 779
 780                 case Opt_size: {
 781                         /* memparse() will accept a K/M/G without a digit */
 782                         if (!isdigit(*args[0].from))
 783                                 goto bad_val;
 784                         size = memparse(args[0].from, &rest);
 785                         setsize = SIZE_STD;
 786                         if (*rest == '%')
 787                                 setsize = SIZE_PERCENT;
 788                         break;
 789                 }
 790
 791                 case Opt_nr_inodes:
 792                         /* memparse() will accept a K/M/G without a digit */
 793                         if (!isdigit(*args[0].from))
 794                                 goto bad_val;
 795                         pconfig->nr_inodes = memparse(args[0].from, &rest);
 796                         break;
 797
 798                 case Opt_pagesize: {
 799                         unsigned long ps;
 800                         ps = memparse(args[0].from, &rest);
 801                         pconfig->hstate = size_to_hstate(ps);
 802                         if (!pconfig->hstate) {
 803                                 printk(KERN_ERR
 804                                 "hugetlbfs: Unsupported page size %lu MB\n",
 805                                         ps >> 20);
 806                                 return -EINVAL;
 807                         }
 808                         break;
 809                 }
 810
 811                 default:
 812                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
 813                                  p);
 814                         return -EINVAL;
 815                         break;
 816                 }
 817         }
 818
 819         /* Do size after hstate is set up */
 820         if (setsize > NO_SIZE) {
 821                 struct hstate *h = pconfig->hstate;
 822                 if (setsize == SIZE_PERCENT) {
 823                         size <<= huge_page_shift(h);
 824                         size *= h->max_huge_pages;
 825                         do_div(size, 100);
 826                 }
 827                 pconfig->nr_blocks = (size >> huge_page_shift(h));
 828         }
 829
 830         return 0;
 831
 832 bad_val:
 833         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
 834                args[0].from, p);
 835         return -EINVAL;
 836 }
 837
 838 static int
 839 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 840 {
 841         struct inode * inode;
 842         struct dentry * root;
 843         int ret;
 844         struct hugetlbfs_config config;
 845         struct hugetlbfs_sb_info *sbinfo;
 846
 847         save_mount_options(sb, data);
 848
 849         config.nr_blocks = -1; /* No limit on size by default */
 850         config.nr_inodes = -1; /* No limit on number of inodes by default */
 851         config.uid = current_fsuid();
 852         config.gid = current_fsgid();
 853         config.mode = 0755;
 854         config.hstate = &default_hstate;
 855         ret = hugetlbfs_parse_options(data, &config);
 856         if (ret)
 857                 return ret;
 858
 859         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 860         if (!sbinfo)
 861                 return -ENOMEM;
 862         sb->s_fs_info = sbinfo;
 863         sbinfo->hstate = config.hstate;
 864         spin_lock_init(&sbinfo->stat_lock);
 865         sbinfo->max_blocks = config.nr_blocks;
 866         sbinfo->free_blocks = config.nr_blocks;
 867         sbinfo->max_inodes = config.nr_inodes;
 868         sbinfo->free_inodes = config.nr_inodes;
 869         sb->s_maxbytes = MAX_LFS_FILESIZE;
 870         sb->s_blocksize = huge_page_size(config.hstate);
 871         sb->s_blocksize_bits = huge_page_shift(config.hstate);
 872         sb->s_magic = HUGETLBFS_MAGIC;
 873         sb->s_op = &hugetlbfs_ops;
 874         sb->s_time_gran = 1;
 875         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 876                                         S_IFDIR | config.mode, 0);
 877         if (!inode)
 878                 goto out_free;
 879
 880         root = d_alloc_root(inode);
 881         if (!root) {
 882                 iput(inode);
 883                 goto out_free;
 884         }
 885         sb->s_root = root;
 886         return 0;
 887 out_free:
 888         kfree(sbinfo);
 889         return -ENOMEM;
 890 }
 891
 892 int hugetlb_get_quota(struct address_space *mapping, long delta)
 893 {
 894         int ret = 0;
 895         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 896
 897         if (sbinfo->free_blocks > -1) {
 898                 spin_lock(&sbinfo->stat_lock);
 899                 if (sbinfo->free_blocks - delta >= 0)
 900                         sbinfo->free_blocks -= delta;
 901                 else
 902                         ret = -ENOMEM;
 903                 spin_unlock(&sbinfo->stat_lock);
 904         }
 905
 906         return ret;
 907 }
 908
 909 void hugetlb_put_quota(struct address_space *mapping, long delta)
 910 {
 911         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 912
 913         if (sbinfo->free_blocks > -1) {
 914                 spin_lock(&sbinfo->stat_lock);
 915                 sbinfo->free_blocks += delta;
 916                 spin_unlock(&sbinfo->stat_lock);
 917         }
 918 }
 919
 920 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
 921         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 922 {
 923         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
 924 }
 925
 926 static struct file_system_type hugetlbfs_fs_type = {
 927         .name           = "hugetlbfs",
 928         .get_sb         = hugetlbfs_get_sb,
 929         .kill_sb        = kill_litter_super,
 930 };
 931
 932 static struct vfsmount *hugetlbfs_vfsmount;
 933
 934 static int can_do_hugetlb_shm(void)
 935 {
 936         return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
 937 }
 938
 939 struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag,
 940                                                 struct user_struct **user)
 941 {
 942         int error = -ENOMEM;
 943         struct file *file;
 944         struct inode *inode;
 945         struct dentry *dentry, *root;
 946         struct qstr quick_string;
 947
 948         *user = NULL;
 949         if (!hugetlbfs_vfsmount)
 950                 return ERR_PTR(-ENOENT);
 951
 952         if (!can_do_hugetlb_shm()) {
 953                 *user = current_user();
 954                 if (user_shm_lock(size, *user)) {
 955                         WARN_ONCE(1,
 956                           "Using mlock ulimits for SHM_HUGETLB deprecated\n");
 957                 } else {
 958                         *user = NULL;
 959                         return ERR_PTR(-EPERM);
 960                 }
 961         }
 962
 963         root = hugetlbfs_vfsmount->mnt_root;
 964         quick_string.name = name;
 965         quick_string.len = strlen(quick_string.name);
 966         quick_string.hash = 0;
 967         dentry = d_alloc(root, &quick_string);
 968         if (!dentry)
 969                 goto out_shm_unlock;
 970
 971         error = -ENOSPC;
 972         inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
 973                                 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
 974         if (!inode)
 975                 goto out_dentry;
 976
 977         error = -ENOMEM;
 978         if (hugetlb_reserve_pages(inode, 0,
 979                         size >> huge_page_shift(hstate_inode(inode)), NULL,
 980                         acctflag))
 981                 goto out_inode;
 982
 983         d_instantiate(dentry, inode);
 984         inode->i_size = size;
 985         inode->i_nlink = 0;
 986
 987         error = -ENFILE;
 988         file = alloc_file(hugetlbfs_vfsmount, dentry,
 989                         FMODE_WRITE | FMODE_READ,
 990                         &hugetlbfs_file_operations);
 991         if (!file)
 992                 goto out_dentry; /* inode is already attached */
 993         ima_counts_get(file);
 994
 995         return file;
 996
 997 out_inode:
 998         iput(inode);
 999 out_dentry:
1000         dput(dentry);
1001 out_shm_unlock:
1002         if (*user) {
1003                 user_shm_unlock(size, *user);
1004                 *user = NULL;
1005         }
1006         return ERR_PTR(error);
1007 }
1008
1009 static int __init init_hugetlbfs_fs(void)
1010 {
1011         int error;
1012         struct vfsmount *vfsmount;
1013
1014         error = bdi_init(&hugetlbfs_backing_dev_info);
1015         if (error)
1016                 return error;
1017
1018         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1019                                         sizeof(struct hugetlbfs_inode_info),
1020                                         0, 0, init_once);
1021         if (hugetlbfs_inode_cachep == NULL)
1022                 goto out2;
1023
1024         error = register_filesystem(&hugetlbfs_fs_type);
1025         if (error)
1026                 goto out;
1027
1028         vfsmount = kern_mount(&hugetlbfs_fs_type);
1029
1030         if (!IS_ERR(vfsmount)) {
1031                 hugetlbfs_vfsmount = vfsmount;
1032                 return 0;
1033         }
1034
1035         error = PTR_ERR(vfsmount);
1036
1037  out:
1038         if (error)
1039                 kmem_cache_destroy(hugetlbfs_inode_cachep);
1040  out2:
1041         bdi_destroy(&hugetlbfs_backing_dev_info);
1042         return error;
1043 }
1044
1045 static void __exit exit_hugetlbfs_fs(void)
1046 {
1047         kmem_cache_destroy(hugetlbfs_inode_cachep);
1048         unregister_filesystem(&hugetlbfs_fs_type);
1049         bdi_destroy(&hugetlbfs_backing_dev_info);
1050 }
1051
1052 module_init(init_hugetlbfs_fs)
1053 module_exit(exit_hugetlbfs_fs)
1054
1055 MODULE_LICENSE("GPL");