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