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 /*
  80  * Mask used when checking the page offset value passed in via system
  81  * calls.  This value will be converted to a loff_t which is signed.
  82  * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the
  83  * value.  The extra bit (- 1 in the shift value) is to take the sign
  84  * bit into account.
  85  */
  86 #define PGOFF_LOFFT_MAX \
  87         (((1UL << (PAGE_SHIFT + 1)) - 1) <<  (BITS_PER_LONG - (PAGE_SHIFT + 1)))
  88
  89 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  90 {
  91         struct inode *inode = file->f_path.dentry->d_inode;
  92         loff_t len, vma_len;
  93         int ret;
  94         struct hstate *h = hstate_file(file);
  95
  96         /*
  97          * vma address alignment (but not the pgoff alignment) has
  98          * already been checked by prepare_hugepage_range.  If you add
  99          * any error returns here, do so after setting VM_HUGETLB, so
 100          * is_vm_hugetlb_page tests below unmap_region go the right
 101          * way when do_mmap_pgoff unwinds (may be important on powerpc
 102          * and ia64).
 103          */
 104         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
 105         vma->vm_ops = &hugetlb_vm_ops;
 106
 107         /*
 108          * page based offset in vm_pgoff could be sufficiently large to
 109          * overflow a (l)off_t when converted to byte offset.
 110          */
 111         if (vma->vm_pgoff & PGOFF_LOFFT_MAX)
 112                 return -EINVAL;
 113
 114         /* must be huge page aligned */
 115         if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
 116                 return -EINVAL;
 117
 118         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 119         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 120         /* check for overflow */
 121         if (len < vma_len)
 122                 return -EINVAL;
 123
 124         mutex_lock(&inode->i_mutex);
 125         file_accessed(file);
 126
 127         ret = -ENOMEM;
 128         if (hugetlb_reserve_pages(inode,
 129                                 vma->vm_pgoff >> huge_page_order(h),
 130                                 len >> huge_page_shift(h), vma,
 131                                 vma->vm_flags))
 132                 goto out;
 133
 134         ret = 0;
 135         hugetlb_prefault_arch_hook(vma->vm_mm);
 136         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 137                 i_size_write(inode, len);
 138 out:
 139         mutex_unlock(&inode->i_mutex);
 140
 141         return ret;
 142 }
 143
 144 /*
 145  * Called under down_write(mmap_sem).
 146  */
 147
 148 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 149 static unsigned long
 150 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 151                 unsigned long len, unsigned long pgoff, unsigned long flags)
 152 {
 153         struct mm_struct *mm = current->mm;
 154         struct vm_area_struct *vma;
 155         unsigned long start_addr;
 156         struct hstate *h = hstate_file(file);
 157
 158         if (len & ~huge_page_mask(h))
 159                 return -EINVAL;
 160         if (len > TASK_SIZE)
 161                 return -ENOMEM;
 162
 163         if (flags & MAP_FIXED) {
 164                 if (prepare_hugepage_range(file, addr, len))
 165                         return -EINVAL;
 166                 return addr;
 167         }
 168
 169         if (addr) {
 170                 addr = ALIGN(addr, huge_page_size(h));
 171                 vma = find_vma(mm, addr);
 172                 if (TASK_SIZE - len >= addr &&
 173                     (!vma || addr + len <= vm_start_gap(vma)))
 174                         return addr;
 175         }
 176
 177         start_addr = mm->free_area_cache;
 178
 179         if (len <= mm->cached_hole_size)
 180                 start_addr = TASK_UNMAPPED_BASE;
 181
 182 full_search:
 183         addr = ALIGN(start_addr, huge_page_size(h));
 184
 185         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 186                 /* At this point:  (!vma || addr < vma->vm_end). */
 187                 if (TASK_SIZE - len < addr) {
 188                         /*
 189                          * Start a new search - just in case we missed
 190                          * some holes.
 191                          */
 192                         if (start_addr != TASK_UNMAPPED_BASE) {
 193                                 start_addr = TASK_UNMAPPED_BASE;
 194                                 goto full_search;
 195                         }
 196                         return -ENOMEM;
 197                 }
 198
 199                 if (!vma || addr + len <= vm_start_gap(vma))
 200                         return addr;
 201                 addr = ALIGN(vma->vm_end, huge_page_size(h));
 202         }
 203 }
 204 #endif
 205
 206 static int
 207 hugetlbfs_read_actor(struct page *page, unsigned long offset,
 208                         char __user *buf, unsigned long count,
 209                         unsigned long size)
 210 {
 211         char *kaddr;
 212         unsigned long left, copied = 0;
 213         int i, chunksize;
 214
 215         if (size > count)
 216                 size = count;
 217
 218         /* Find which 4k chunk and offset with in that chunk */
 219         i = offset >> PAGE_CACHE_SHIFT;
 220         offset = offset & ~PAGE_CACHE_MASK;
 221
 222         while (size) {
 223                 chunksize = PAGE_CACHE_SIZE;
 224                 if (offset)
 225                         chunksize -= offset;
 226                 if (chunksize > size)
 227                         chunksize = size;
 228                 kaddr = kmap(&page[i]);
 229                 left = __copy_to_user(buf, kaddr + offset, chunksize);
 230                 kunmap(&page[i]);
 231                 if (left) {
 232                         copied += (chunksize - left);
 233                         break;
 234                 }
 235                 offset = 0;
 236                 size -= chunksize;
 237                 buf += chunksize;
 238                 copied += chunksize;
 239                 i++;
 240         }
 241         return copied ? copied : -EFAULT;
 242 }
 243
 244 /*
 245  * Support for read() - Find the page attached to f_mapping and copy out the
 246  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
 247  * since it has PAGE_CACHE_SIZE assumptions.
 248  */
 249 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
 250                               size_t len, loff_t *ppos)
 251 {
 252         struct hstate *h = hstate_file(filp);
 253         struct address_space *mapping = filp->f_mapping;
 254         struct inode *inode = mapping->host;
 255         unsigned long index = *ppos >> huge_page_shift(h);
 256         unsigned long offset = *ppos & ~huge_page_mask(h);
 257         unsigned long end_index;
 258         loff_t isize;
 259         ssize_t retval = 0;
 260
 261         /* validate length */
 262         if (len == 0)
 263                 goto out;
 264
 265         for (;;) {
 266                 struct page *page;
 267                 unsigned long nr, ret;
 268                 int ra;
 269
 270                 /* nr is the maximum number of bytes to copy from this page */
 271                 nr = huge_page_size(h);
 272                 isize = i_size_read(inode);
 273                 if (!isize)
 274                         goto out;
 275                 end_index = (isize - 1) >> huge_page_shift(h);
 276                 if (index >= end_index) {
 277                         if (index > end_index)
 278                                 goto out;
 279                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
 280                         if (nr <= offset)
 281                                 goto out;
 282                 }
 283                 nr = nr - offset;
 284
 285                 /* Find the page */
 286                 page = find_lock_page(mapping, index);
 287                 if (unlikely(page == NULL)) {
 288                         /*
 289                          * We have a HOLE, zero out the user-buffer for the
 290                          * length of the hole or request.
 291                          */
 292                         ret = len < nr ? len : nr;
 293                         if (clear_user(buf, ret))
 294                                 ra = -EFAULT;
 295                         else
 296                                 ra = 0;
 297                 } else {
 298                         unlock_page(page);
 299
 300                         /*
 301                          * We have the page, copy it to user space buffer.
 302                          */
 303                         ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
 304                         ret = ra;
 305                         page_cache_release(page);
 306                 }
 307                 if (ra < 0) {
 308                         if (retval == 0)
 309                                 retval = ra;
 310                         goto out;
 311                 }
 312
 313                 offset += ret;
 314                 retval += ret;
 315                 len -= ret;
 316                 index += offset >> huge_page_shift(h);
 317                 offset &= ~huge_page_mask(h);
 318
 319                 /* short read or no more work */
 320                 if ((ret != nr) || (len == 0))
 321                         break;
 322         }
 323 out:
 324         *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
 325         return retval;
 326 }
 327
 328 static int hugetlbfs_write_begin(struct file *file,
 329                         struct address_space *mapping,
 330                         loff_t pos, unsigned len, unsigned flags,
 331                         struct page **pagep, void **fsdata)
 332 {
 333         return -EINVAL;
 334 }
 335
 336 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 337                         loff_t pos, unsigned len, unsigned copied,
 338                         struct page *page, void *fsdata)
 339 {
 340         BUG();
 341         return -EINVAL;
 342 }
 343
 344 static void truncate_huge_page(struct page *page)
 345 {
 346         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
 347         ClearPageUptodate(page);
 348         delete_from_page_cache(page);
 349 }
 350
 351 static void truncate_hugepages(struct inode *inode, loff_t lstart)
 352 {
 353         struct hstate *h = hstate_inode(inode);
 354         struct address_space *mapping = &inode->i_data;
 355         const pgoff_t start = lstart >> huge_page_shift(h);
 356         struct pagevec pvec;
 357         pgoff_t next;
 358         int i, freed = 0;
 359
 360         pagevec_init(&pvec, 0);
 361         next = start;
 362         while (1) {
 363                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 364                         if (next == start)
 365                                 break;
 366                         next = start;
 367                         continue;
 368                 }
 369
 370                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 371                         struct page *page = pvec.pages[i];
 372
 373                         lock_page(page);
 374                         if (page->index > next)
 375                                 next = page->index;
 376                         ++next;
 377                         truncate_huge_page(page);
 378                         unlock_page(page);
 379                         freed++;
 380                 }
 381                 huge_pagevec_release(&pvec);
 382         }
 383         BUG_ON(!lstart && mapping->nrpages);
 384         hugetlb_unreserve_pages(inode, start, freed);
 385 }
 386
 387 static void hugetlbfs_evict_inode(struct inode *inode)
 388 {
 389         truncate_hugepages(inode, 0);
 390         end_writeback(inode);
 391 }
 392
 393 static inline void
 394 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
 395 {
 396         struct vm_area_struct *vma;
 397         struct prio_tree_iter iter;
 398
 399         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
 400                 unsigned long v_offset;
 401
 402                 /*
 403                  * Can the expression below overflow on 32-bit arches?
 404                  * No, because the prio_tree returns us only those vmas
 405                  * which overlap the truncated area starting at pgoff,
 406                  * and no vma on a 32-bit arch can span beyond the 4GB.
 407                  */
 408                 if (vma->vm_pgoff < pgoff)
 409                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 410                 else
 411                         v_offset = 0;
 412
 413                 __unmap_hugepage_range(vma,
 414                                 vma->vm_start + v_offset, vma->vm_end, NULL);
 415         }
 416 }
 417
 418 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 419 {
 420         pgoff_t pgoff;
 421         struct address_space *mapping = inode->i_mapping;
 422         struct hstate *h = hstate_inode(inode);
 423
 424         BUG_ON(offset & ~huge_page_mask(h));
 425         pgoff = offset >> PAGE_SHIFT;
 426
 427         i_size_write(inode, offset);
 428         mutex_lock(&mapping->i_mmap_mutex);
 429         if (!prio_tree_empty(&mapping->i_mmap))
 430                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 431         mutex_unlock(&mapping->i_mmap_mutex);
 432         truncate_hugepages(inode, offset);
 433         return 0;
 434 }
 435
 436 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 437 {
 438         struct inode *inode = dentry->d_inode;
 439         struct hstate *h = hstate_inode(inode);
 440         int error;
 441         unsigned int ia_valid = attr->ia_valid;
 442
 443         BUG_ON(!inode);
 444
 445         error = setattr_prepare(dentry, attr);
 446         if (error)
 447                 return error;
 448
 449         if (ia_valid & ATTR_SIZE) {
 450                 error = -EINVAL;
 451                 if (attr->ia_size & ~huge_page_mask(h))
 452                         return -EINVAL;
 453                 error = hugetlb_vmtruncate(inode, attr->ia_size);
 454                 if (error)
 455                         return error;
 456         }
 457
 458         setattr_copy(inode, attr);
 459         mark_inode_dirty(inode);
 460         return 0;
 461 }
 462
 463 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 464                                         gid_t gid, int mode, dev_t dev)
 465 {
 466         struct inode *inode;
 467
 468         inode = new_inode(sb);
 469         if (inode) {
 470                 struct hugetlbfs_inode_info *info;
 471                 inode->i_ino = get_next_ino();
 472                 inode->i_mode = mode;
 473                 inode->i_uid = uid;
 474                 inode->i_gid = gid;
 475                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 476                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 477                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 478                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
 479                 info = HUGETLBFS_I(inode);
 480                 /*
 481                  * The policy is initialized here even if we are creating a
 482                  * private inode because initialization simply creates an
 483                  * an empty rb tree and calls spin_lock_init(), later when we
 484                  * call mpol_free_shared_policy() it will just return because
 485                  * the rb tree will still be empty.
 486                  */
 487                 mpol_shared_policy_init(&info->policy, NULL);
 488                 switch (mode & S_IFMT) {
 489                 default:
 490                         init_special_inode(inode, mode, dev);
 491                         break;
 492                 case S_IFREG:
 493                         inode->i_op = &hugetlbfs_inode_operations;
 494                         inode->i_fop = &hugetlbfs_file_operations;
 495                         break;
 496                 case S_IFDIR:
 497                         inode->i_op = &hugetlbfs_dir_inode_operations;
 498                         inode->i_fop = &simple_dir_operations;
 499
 500                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 501                         inc_nlink(inode);
 502                         break;
 503                 case S_IFLNK:
 504                         inode->i_op = &page_symlink_inode_operations;
 505                         break;
 506                 }
 507                 lockdep_annotate_inode_mutex_key(inode);
 508         }
 509         return inode;
 510 }
 511
 512 /*
 513  * File creation. Allocate an inode, and we're done..
 514  */
 515 static int hugetlbfs_mknod(struct inode *dir,
 516                         struct dentry *dentry, int mode, dev_t dev)
 517 {
 518         struct inode *inode;
 519         int error = -ENOSPC;
 520         gid_t gid;
 521
 522         if (dir->i_mode & S_ISGID) {
 523                 gid = dir->i_gid;
 524                 if (S_ISDIR(mode))
 525                         mode |= S_ISGID;
 526         } else {
 527                 gid = current_fsgid();
 528         }
 529         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
 530         if (inode) {
 531                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 532                 d_instantiate(dentry, inode);
 533                 dget(dentry);   /* Extra count - pin the dentry in core */
 534                 error = 0;
 535         }
 536         return error;
 537 }
 538
 539 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 540 {
 541         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 542         if (!retval)
 543                 inc_nlink(dir);
 544         return retval;
 545 }
 546
 547 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 548 {
 549         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 550 }
 551
 552 static int hugetlbfs_symlink(struct inode *dir,
 553                         struct dentry *dentry, const char *symname)
 554 {
 555         struct inode *inode;
 556         int error = -ENOSPC;
 557         gid_t gid;
 558
 559         if (dir->i_mode & S_ISGID)
 560                 gid = dir->i_gid;
 561         else
 562                 gid = current_fsgid();
 563
 564         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
 565                                         gid, S_IFLNK|S_IRWXUGO, 0);
 566         if (inode) {
 567                 int l = strlen(symname)+1;
 568                 error = page_symlink(inode, symname, l);
 569                 if (!error) {
 570                         d_instantiate(dentry, inode);
 571                         dget(dentry);
 572                 } else
 573                         iput(inode);
 574         }
 575         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 576
 577         return error;
 578 }
 579
 580 /*
 581  * mark the head page dirty
 582  */
 583 static int hugetlbfs_set_page_dirty(struct page *page)
 584 {
 585         struct page *head = compound_head(page);
 586
 587         SetPageDirty(head);
 588         return 0;
 589 }
 590
 591 static int hugetlbfs_migrate_page(struct address_space *mapping,
 592                                 struct page *newpage, struct page *page,
 593                                 enum migrate_mode mode)
 594 {
 595         int rc;
 596
 597         rc = migrate_huge_page_move_mapping(mapping, newpage, page);
 598         if (rc)
 599                 return rc;
 600         migrate_page_copy(newpage, page);
 601
 602         return 0;
 603 }
 604
 605 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 606 {
 607         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 608         struct hstate *h = hstate_inode(dentry->d_inode);
 609
 610         buf->f_type = HUGETLBFS_MAGIC;
 611         buf->f_bsize = huge_page_size(h);
 612         if (sbinfo) {
 613                 spin_lock(&sbinfo->stat_lock);
 614                 /* If no limits set, just report 0 for max/free/used
 615                  * blocks, like simple_statfs() */
 616                 if (sbinfo->spool) {
 617                         long free_pages;
 618
 619                         spin_lock(&sbinfo->spool->lock);
 620                         buf->f_blocks = sbinfo->spool->max_hpages;
 621                         free_pages = sbinfo->spool->max_hpages
 622                                 - sbinfo->spool->used_hpages;
 623                         buf->f_bavail = buf->f_bfree = free_pages;
 624                         spin_unlock(&sbinfo->spool->lock);
 625                         buf->f_files = sbinfo->max_inodes;
 626                         buf->f_ffree = sbinfo->free_inodes;
 627                 }
 628                 spin_unlock(&sbinfo->stat_lock);
 629         }
 630         buf->f_namelen = NAME_MAX;
 631         return 0;
 632 }
 633
 634 static void hugetlbfs_put_super(struct super_block *sb)
 635 {
 636         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 637
 638         if (sbi) {
 639                 sb->s_fs_info = NULL;
 640
 641                 if (sbi->spool)
 642                         hugepage_put_subpool(sbi->spool);
 643
 644                 kfree(sbi);
 645         }
 646 }
 647
 648 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 649 {
 650         if (sbinfo->free_inodes >= 0) {
 651                 spin_lock(&sbinfo->stat_lock);
 652                 if (unlikely(!sbinfo->free_inodes)) {
 653                         spin_unlock(&sbinfo->stat_lock);
 654                         return 0;
 655                 }
 656                 sbinfo->free_inodes--;
 657                 spin_unlock(&sbinfo->stat_lock);
 658         }
 659
 660         return 1;
 661 }
 662
 663 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 664 {
 665         if (sbinfo->free_inodes >= 0) {
 666                 spin_lock(&sbinfo->stat_lock);
 667                 sbinfo->free_inodes++;
 668                 spin_unlock(&sbinfo->stat_lock);
 669         }
 670 }
 671
 672
 673 static struct kmem_cache *hugetlbfs_inode_cachep;
 674
 675 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 676 {
 677         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 678         struct hugetlbfs_inode_info *p;
 679
 680         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 681                 return NULL;
 682         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 683         if (unlikely(!p)) {
 684                 hugetlbfs_inc_free_inodes(sbinfo);
 685                 return NULL;
 686         }
 687         return &p->vfs_inode;
 688 }
 689
 690 static void hugetlbfs_i_callback(struct rcu_head *head)
 691 {
 692         struct inode *inode = container_of(head, struct inode, i_rcu);
 693         INIT_LIST_HEAD(&inode->i_dentry);
 694         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 695 }
 696
 697 static void hugetlbfs_destroy_inode(struct inode *inode)
 698 {
 699         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 700         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 701         call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
 702 }
 703
 704 static const struct address_space_operations hugetlbfs_aops = {
 705         .write_begin    = hugetlbfs_write_begin,
 706         .write_end      = hugetlbfs_write_end,
 707         .set_page_dirty = hugetlbfs_set_page_dirty,
 708         .migratepage    = hugetlbfs_migrate_page,
 709 };
 710
 711
 712 static void init_once(void *foo)
 713 {
 714         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 715
 716         inode_init_once(&ei->vfs_inode);
 717 }
 718
 719 const struct file_operations hugetlbfs_file_operations = {
 720         .read                   = hugetlbfs_read,
 721         .mmap                   = hugetlbfs_file_mmap,
 722         .fsync                  = noop_fsync,
 723         .get_unmapped_area      = hugetlb_get_unmapped_area,
 724         .llseek         = default_llseek,
 725 };
 726
 727 static const struct inode_operations hugetlbfs_dir_inode_operations = {
 728         .create         = hugetlbfs_create,
 729         .lookup         = simple_lookup,
 730         .link           = simple_link,
 731         .unlink         = simple_unlink,
 732         .symlink        = hugetlbfs_symlink,
 733         .mkdir          = hugetlbfs_mkdir,
 734         .rmdir          = simple_rmdir,
 735         .mknod          = hugetlbfs_mknod,
 736         .rename         = simple_rename,
 737         .setattr        = hugetlbfs_setattr,
 738 };
 739
 740 static const struct inode_operations hugetlbfs_inode_operations = {
 741         .setattr        = hugetlbfs_setattr,
 742 };
 743
 744 static const struct super_operations hugetlbfs_ops = {
 745         .alloc_inode    = hugetlbfs_alloc_inode,
 746         .destroy_inode  = hugetlbfs_destroy_inode,
 747         .evict_inode    = hugetlbfs_evict_inode,
 748         .statfs         = hugetlbfs_statfs,
 749         .put_super      = hugetlbfs_put_super,
 750         .show_options   = generic_show_options,
 751 };
 752
 753 static int
 754 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 755 {
 756         char *p, *rest;
 757         substring_t args[MAX_OPT_ARGS];
 758         int option;
 759         unsigned long long size = 0;
 760         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
 761
 762         if (!options)
 763                 return 0;
 764
 765         while ((p = strsep(&options, ",")) != NULL) {
 766                 int token;
 767                 if (!*p)
 768                         continue;
 769
 770                 token = match_token(p, tokens, args);
 771                 switch (token) {
 772                 case Opt_uid:
 773                         if (match_int(&args[0], &option))
 774                                 goto bad_val;
 775                         pconfig->uid = option;
 776                         break;
 777
 778                 case Opt_gid:
 779                         if (match_int(&args[0], &option))
 780                                 goto bad_val;
 781                         pconfig->gid = option;
 782                         break;
 783
 784                 case Opt_mode:
 785                         if (match_octal(&args[0], &option))
 786                                 goto bad_val;
 787                         pconfig->mode = option & 01777U;
 788                         break;
 789
 790                 case Opt_size: {
 791                         /* memparse() will accept a K/M/G without a digit */
 792                         if (!isdigit(*args[0].from))
 793                                 goto bad_val;
 794                         size = memparse(args[0].from, &rest);
 795                         setsize = SIZE_STD;
 796                         if (*rest == '%')
 797                                 setsize = SIZE_PERCENT;
 798                         break;
 799                 }
 800
 801                 case Opt_nr_inodes:
 802                         /* memparse() will accept a K/M/G without a digit */
 803                         if (!isdigit(*args[0].from))
 804                                 goto bad_val;
 805                         pconfig->nr_inodes = memparse(args[0].from, &rest);
 806                         break;
 807
 808                 case Opt_pagesize: {
 809                         unsigned long ps;
 810                         ps = memparse(args[0].from, &rest);
 811                         pconfig->hstate = size_to_hstate(ps);
 812                         if (!pconfig->hstate) {
 813                                 printk(KERN_ERR
 814                                 "hugetlbfs: Unsupported page size %lu MB\n",
 815                                         ps >> 20);
 816                                 return -EINVAL;
 817                         }
 818                         break;
 819                 }
 820
 821                 default:
 822                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
 823                                  p);
 824                         return -EINVAL;
 825                         break;
 826                 }
 827         }
 828
 829         /* Do size after hstate is set up */
 830         if (setsize > NO_SIZE) {
 831                 struct hstate *h = pconfig->hstate;
 832                 if (setsize == SIZE_PERCENT) {
 833                         size <<= huge_page_shift(h);
 834                         size *= h->max_huge_pages;
 835                         do_div(size, 100);
 836                 }
 837                 pconfig->nr_blocks = (size >> huge_page_shift(h));
 838         }
 839
 840         return 0;
 841
 842 bad_val:
 843         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
 844                args[0].from, p);
 845         return -EINVAL;
 846 }
 847
 848 static int
 849 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 850 {
 851         struct inode * inode;
 852         struct dentry * root;
 853         int ret;
 854         struct hugetlbfs_config config;
 855         struct hugetlbfs_sb_info *sbinfo;
 856
 857         save_mount_options(sb, data);
 858
 859         config.nr_blocks = -1; /* No limit on size by default */
 860         config.nr_inodes = -1; /* No limit on number of inodes by default */
 861         config.uid = current_fsuid();
 862         config.gid = current_fsgid();
 863         config.mode = 0755;
 864         config.hstate = &default_hstate;
 865         ret = hugetlbfs_parse_options(data, &config);
 866         if (ret)
 867                 return ret;
 868
 869         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 870         if (!sbinfo)
 871                 return -ENOMEM;
 872         sb->s_fs_info = sbinfo;
 873         sbinfo->hstate = config.hstate;
 874         spin_lock_init(&sbinfo->stat_lock);
 875         sbinfo->max_inodes = config.nr_inodes;
 876         sbinfo->free_inodes = config.nr_inodes;
 877         sbinfo->spool = NULL;
 878         if (config.nr_blocks != -1) {
 879                 sbinfo->spool = hugepage_new_subpool(config.nr_blocks);
 880                 if (!sbinfo->spool)
 881                         goto out_free;
 882         }
 883         sb->s_maxbytes = MAX_LFS_FILESIZE;
 884         sb->s_blocksize = huge_page_size(config.hstate);
 885         sb->s_blocksize_bits = huge_page_shift(config.hstate);
 886         sb->s_magic = HUGETLBFS_MAGIC;
 887         sb->s_op = &hugetlbfs_ops;
 888         sb->s_time_gran = 1;
 889         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 890                                         S_IFDIR | config.mode, 0);
 891         if (!inode)
 892                 goto out_free;
 893
 894         root = d_alloc_root(inode);
 895         if (!root) {
 896                 iput(inode);
 897                 goto out_free;
 898         }
 899         sb->s_root = root;
 900         return 0;
 901 out_free:
 902         if (sbinfo->spool)
 903                 kfree(sbinfo->spool);
 904         kfree(sbinfo);
 905         return -ENOMEM;
 906 }
 907
 908 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
 909         int flags, const char *dev_name, void *data)
 910 {
 911         return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 912 }
 913
 914 static struct file_system_type hugetlbfs_fs_type = {
 915         .name           = "hugetlbfs",
 916         .mount          = hugetlbfs_mount,
 917         .kill_sb        = kill_litter_super,
 918 };
 919
 920 static struct vfsmount *hugetlbfs_vfsmount;
 921
 922 static int can_do_hugetlb_shm(void)
 923 {
 924         return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
 925 }
 926
 927 struct file *hugetlb_file_setup(const char *name, size_t size,
 928                                 vm_flags_t acctflag,
 929                                 struct user_struct **user, int creat_flags)
 930 {
 931         int error = -ENOMEM;
 932         struct file *file;
 933         struct inode *inode;
 934         struct path path;
 935         struct dentry *root;
 936         struct qstr quick_string;
 937
 938         *user = NULL;
 939         if (!hugetlbfs_vfsmount)
 940                 return ERR_PTR(-ENOENT);
 941
 942         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
 943                 *user = current_user();
 944                 if (user_shm_lock(size, *user)) {
 945                         printk_once(KERN_WARNING "Using mlock ulimits for SHM_HUGETLB is deprecated\n");
 946                 } else {
 947                         *user = NULL;
 948                         return ERR_PTR(-EPERM);
 949                 }
 950         }
 951
 952         root = hugetlbfs_vfsmount->mnt_root;
 953         quick_string.name = name;
 954         quick_string.len = strlen(quick_string.name);
 955         quick_string.hash = 0;
 956         path.dentry = d_alloc(root, &quick_string);
 957         if (!path.dentry)
 958                 goto out_shm_unlock;
 959
 960         path.mnt = mntget(hugetlbfs_vfsmount);
 961         error = -ENOSPC;
 962         inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
 963                                 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
 964         if (!inode)
 965                 goto out_dentry;
 966
 967         error = -ENOMEM;
 968         if (hugetlb_reserve_pages(inode, 0,
 969                         size >> huge_page_shift(hstate_inode(inode)), NULL,
 970                         acctflag))
 971                 goto out_inode;
 972
 973         d_instantiate(path.dentry, inode);
 974         inode->i_size = size;
 975         clear_nlink(inode);
 976
 977         error = -ENFILE;
 978         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
 979                         &hugetlbfs_file_operations);
 980         if (!file)
 981                 goto out_dentry; /* inode is already attached */
 982
 983         return file;
 984
 985 out_inode:
 986         iput(inode);
 987 out_dentry:
 988         path_put(&path);
 989 out_shm_unlock:
 990         if (*user) {
 991                 user_shm_unlock(size, *user);
 992                 *user = NULL;
 993         }
 994         return ERR_PTR(error);
 995 }
 996
 997 static int __init init_hugetlbfs_fs(void)
 998 {
 999         int error;
1000         struct vfsmount *vfsmount;
1001
1002         if (!hugepages_supported()) {
1003                 pr_info("hugetlbfs: disabling because there are no supported hugepage sizes\n");
1004                 return -ENOTSUPP;
1005         }
1006
1007         error = bdi_init(&hugetlbfs_backing_dev_info);
1008         if (error)
1009                 return error;
1010
1011         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1012                                         sizeof(struct hugetlbfs_inode_info),
1013                                         0, 0, init_once);
1014         if (hugetlbfs_inode_cachep == NULL)
1015                 goto out2;
1016
1017         error = register_filesystem(&hugetlbfs_fs_type);
1018         if (error)
1019                 goto out;
1020
1021         vfsmount = kern_mount(&hugetlbfs_fs_type);
1022
1023         if (!IS_ERR(vfsmount)) {
1024                 hugetlbfs_vfsmount = vfsmount;
1025                 return 0;
1026         }
1027
1028         error = PTR_ERR(vfsmount);
1029
1030  out:
1031         if (error)
1032                 kmem_cache_destroy(hugetlbfs_inode_cachep);
1033  out2:
1034         bdi_destroy(&hugetlbfs_backing_dev_info);
1035         return error;
1036 }
1037
1038 static void __exit exit_hugetlbfs_fs(void)
1039 {
1040         kmem_cache_destroy(hugetlbfs_inode_cachep);
1041         kern_unmount(hugetlbfs_vfsmount);
1042         unregister_filesystem(&hugetlbfs_fs_type);
1043         bdi_destroy(&hugetlbfs_backing_dev_info);
1044 }
1045
1046 module_init(init_hugetlbfs_fs)
1047 module_exit(exit_hugetlbfs_fs)
1048
1049 MODULE_LICENSE("GPL");