fs/hugetlbfs/inode.c

   1 /*
   2  * hugetlbpage-backed filesystem.  Based on ramfs.
   3  *
   4  * Nadia Yvette Chambers, 2002
   5  *
   6  * Copyright (C) 2002 Linus Torvalds.
   7  */
   8
   9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  10
  11 #include <linux/module.h>
  12 #include <linux/thread_info.h>
  13 #include <asm/current.h>
  14 #include <linux/sched.h>                /* remove ASAP */
  15 #include <linux/fs.h>
  16 #include <linux/mount.h>
  17 #include <linux/file.h>
  18 #include <linux/kernel.h>
  19 #include <linux/writeback.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/highmem.h>
  22 #include <linux/init.h>
  23 #include <linux/string.h>
  24 #include <linux/capability.h>
  25 #include <linux/ctype.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/hugetlb.h>
  28 #include <linux/pagevec.h>
  29 #include <linux/parser.h>
  30 #include <linux/mman.h>
  31 #include <linux/slab.h>
  32 #include <linux/dnotify.h>
  33 #include <linux/statfs.h>
  34 #include <linux/security.h>
  35 #include <linux/magic.h>
  36 #include <linux/migrate.h>
  37 #include <linux/uio.h>
  38
  39 #include <asm/uaccess.h>
  40
  41 static const struct super_operations hugetlbfs_ops;
  42 static const struct address_space_operations hugetlbfs_aops;
  43 const struct file_operations hugetlbfs_file_operations;
  44 static const struct inode_operations hugetlbfs_dir_inode_operations;
  45 static const struct inode_operations hugetlbfs_inode_operations;
  46
  47 struct hugetlbfs_config {
  48         kuid_t   uid;
  49         kgid_t   gid;
  50         umode_t mode;
  51         long    max_hpages;
  52         long    nr_inodes;
  53         struct hstate *hstate;
  54         long    min_hpages;
  55 };
  56
  57 struct hugetlbfs_inode_info {
  58         struct shared_policy policy;
  59         struct inode vfs_inode;
  60 };
  61
  62 static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
  63 {
  64         return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
  65 }
  66
  67 int sysctl_hugetlb_shm_group;
  68
  69 enum {
  70         Opt_size, Opt_nr_inodes,
  71         Opt_mode, Opt_uid, Opt_gid,
  72         Opt_pagesize, Opt_min_size,
  73         Opt_err,
  74 };
  75
  76 static const match_table_t tokens = {
  77         {Opt_size,      "size=%s"},
  78         {Opt_nr_inodes, "nr_inodes=%s"},
  79         {Opt_mode,      "mode=%o"},
  80         {Opt_uid,       "uid=%u"},
  81         {Opt_gid,       "gid=%u"},
  82         {Opt_pagesize,  "pagesize=%s"},
  83         {Opt_min_size,  "min_size=%s"},
  84         {Opt_err,       NULL},
  85 };
  86
  87 static void huge_pagevec_release(struct pagevec *pvec)
  88 {
  89         int i;
  90
  91         for (i = 0; i < pagevec_count(pvec); ++i)
  92                 put_page(pvec->pages[i]);
  93
  94         pagevec_reinit(pvec);
  95 }
  96
  97 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  98 {
  99         struct inode *inode = file_inode(file);
 100         loff_t len, vma_len;
 101         int ret;
 102         struct hstate *h = hstate_file(file);
 103
 104         /*
 105          * vma address alignment (but not the pgoff alignment) has
 106          * already been checked by prepare_hugepage_range.  If you add
 107          * any error returns here, do so after setting VM_HUGETLB, so
 108          * is_vm_hugetlb_page tests below unmap_region go the right
 109          * way when do_mmap_pgoff unwinds (may be important on powerpc
 110          * and ia64).
 111          */
 112         vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
 113         vma->vm_ops = &hugetlb_vm_ops;
 114
 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
 120         mutex_lock(&inode->i_mutex);
 121         file_accessed(file);
 122
 123         ret = -ENOMEM;
 124         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 125
 126         if (hugetlb_reserve_pages(inode,
 127                                 vma->vm_pgoff >> huge_page_order(h),
 128                                 len >> huge_page_shift(h), vma,
 129                                 vma->vm_flags))
 130                 goto out;
 131
 132         ret = 0;
 133         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 134                 inode->i_size = len;
 135 out:
 136         mutex_unlock(&inode->i_mutex);
 137
 138         return ret;
 139 }
 140
 141 /*
 142  * Called under down_write(mmap_sem).
 143  */
 144
 145 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 146 static unsigned long
 147 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 148                 unsigned long len, unsigned long pgoff, unsigned long flags)
 149 {
 150         struct mm_struct *mm = current->mm;
 151         struct vm_area_struct *vma;
 152         struct hstate *h = hstate_file(file);
 153         struct vm_unmapped_area_info info;
 154
 155         if (len & ~huge_page_mask(h))
 156                 return -EINVAL;
 157         if (len > TASK_SIZE)
 158                 return -ENOMEM;
 159
 160         if (flags & MAP_FIXED) {
 161                 if (prepare_hugepage_range(file, addr, len))
 162                         return -EINVAL;
 163                 return addr;
 164         }
 165
 166         if (addr) {
 167                 addr = ALIGN(addr, huge_page_size(h));
 168                 vma = find_vma(mm, addr);
 169                 if (TASK_SIZE - len >= addr &&
 170                     (!vma || addr + len <= vma->vm_start))
 171                         return addr;
 172         }
 173
 174         info.flags = 0;
 175         info.length = len;
 176         info.low_limit = TASK_UNMAPPED_BASE;
 177         info.high_limit = TASK_SIZE;
 178         info.align_mask = PAGE_MASK & ~huge_page_mask(h);
 179         info.align_offset = 0;
 180         return vm_unmapped_area(&info);
 181 }
 182 #endif
 183
 184 static size_t
 185 hugetlbfs_read_actor(struct page *page, unsigned long offset,
 186                         struct iov_iter *to, unsigned long size)
 187 {
 188         size_t copied = 0;
 189         int i, chunksize;
 190
 191         /* Find which 4k chunk and offset with in that chunk */
 192         i = offset >> PAGE_CACHE_SHIFT;
 193         offset = offset & ~PAGE_CACHE_MASK;
 194
 195         while (size) {
 196                 size_t n;
 197                 chunksize = PAGE_CACHE_SIZE;
 198                 if (offset)
 199                         chunksize -= offset;
 200                 if (chunksize > size)
 201                         chunksize = size;
 202                 n = copy_page_to_iter(&page[i], offset, chunksize, to);
 203                 copied += n;
 204                 if (n != chunksize)
 205                         return copied;
 206                 offset = 0;
 207                 size -= chunksize;
 208                 i++;
 209         }
 210         return copied;
 211 }
 212
 213 /*
 214  * Support for read() - Find the page attached to f_mapping and copy out the
 215  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
 216  * since it has PAGE_CACHE_SIZE assumptions.
 217  */
 218 static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
 219 {
 220         struct file *file = iocb->ki_filp;
 221         struct hstate *h = hstate_file(file);
 222         struct address_space *mapping = file->f_mapping;
 223         struct inode *inode = mapping->host;
 224         unsigned long index = iocb->ki_pos >> huge_page_shift(h);
 225         unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
 226         unsigned long end_index;
 227         loff_t isize;
 228         ssize_t retval = 0;
 229
 230         while (iov_iter_count(to)) {
 231                 struct page *page;
 232                 size_t nr, copied;
 233
 234                 /* nr is the maximum number of bytes to copy from this page */
 235                 nr = huge_page_size(h);
 236                 isize = i_size_read(inode);
 237                 if (!isize)
 238                         break;
 239                 end_index = (isize - 1) >> huge_page_shift(h);
 240                 if (index > end_index)
 241                         break;
 242                 if (index == end_index) {
 243                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
 244                         if (nr <= offset)
 245                                 break;
 246                 }
 247                 nr = nr - offset;
 248
 249                 /* Find the page */
 250                 page = find_lock_page(mapping, index);
 251                 if (unlikely(page == NULL)) {
 252                         /*
 253                          * We have a HOLE, zero out the user-buffer for the
 254                          * length of the hole or request.
 255                          */
 256                         copied = iov_iter_zero(nr, to);
 257                 } else {
 258                         unlock_page(page);
 259
 260                         /*
 261                          * We have the page, copy it to user space buffer.
 262                          */
 263                         copied = hugetlbfs_read_actor(page, offset, to, nr);
 264                         page_cache_release(page);
 265                 }
 266                 offset += copied;
 267                 retval += copied;
 268                 if (copied != nr && iov_iter_count(to)) {
 269                         if (!retval)
 270                                 retval = -EFAULT;
 271                         break;
 272                 }
 273                 index += offset >> huge_page_shift(h);
 274                 offset &= ~huge_page_mask(h);
 275         }
 276         iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
 277         return retval;
 278 }
 279
 280 static int hugetlbfs_write_begin(struct file *file,
 281                         struct address_space *mapping,
 282                         loff_t pos, unsigned len, unsigned flags,
 283                         struct page **pagep, void **fsdata)
 284 {
 285         return -EINVAL;
 286 }
 287
 288 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 289                         loff_t pos, unsigned len, unsigned copied,
 290                         struct page *page, void *fsdata)
 291 {
 292         BUG();
 293         return -EINVAL;
 294 }
 295
 296 static void truncate_huge_page(struct page *page)
 297 {
 298         ClearPageDirty(page);
 299         ClearPageUptodate(page);
 300         delete_from_page_cache(page);
 301 }
 302
 303 static void truncate_hugepages(struct inode *inode, loff_t lstart)
 304 {
 305         struct hstate *h = hstate_inode(inode);
 306         struct address_space *mapping = &inode->i_data;
 307         const pgoff_t start = lstart >> huge_page_shift(h);
 308         struct pagevec pvec;
 309         pgoff_t next;
 310         int i, freed = 0;
 311
 312         pagevec_init(&pvec, 0);
 313         next = start;
 314         while (1) {
 315                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 316                         if (next == start)
 317                                 break;
 318                         next = start;
 319                         continue;
 320                 }
 321
 322                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 323                         struct page *page = pvec.pages[i];
 324
 325                         lock_page(page);
 326                         if (page->index > next)
 327                                 next = page->index;
 328                         ++next;
 329                         truncate_huge_page(page);
 330                         unlock_page(page);
 331                         freed++;
 332                 }
 333                 huge_pagevec_release(&pvec);
 334         }
 335         BUG_ON(!lstart && mapping->nrpages);
 336         hugetlb_unreserve_pages(inode, start, freed);
 337 }
 338
 339 static void hugetlbfs_evict_inode(struct inode *inode)
 340 {
 341         struct resv_map *resv_map;
 342
 343         truncate_hugepages(inode, 0);
 344         resv_map = (struct resv_map *)inode->i_mapping->private_data;
 345         /* root inode doesn't have the resv_map, so we should check it */
 346         if (resv_map)
 347                 resv_map_release(&resv_map->refs);
 348         clear_inode(inode);
 349 }
 350
 351 static inline void
 352 hugetlb_vmtruncate_list(struct rb_root *root, pgoff_t pgoff)
 353 {
 354         struct vm_area_struct *vma;
 355
 356         vma_interval_tree_foreach(vma, root, pgoff, ULONG_MAX) {
 357                 unsigned long v_offset;
 358
 359                 /*
 360                  * Can the expression below overflow on 32-bit arches?
 361                  * No, because the interval tree returns us only those vmas
 362                  * which overlap the truncated area starting at pgoff,
 363                  * and no vma on a 32-bit arch can span beyond the 4GB.
 364                  */
 365                 if (vma->vm_pgoff < pgoff)
 366                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 367                 else
 368                         v_offset = 0;
 369
 370                 unmap_hugepage_range(vma, vma->vm_start + v_offset,
 371                                      vma->vm_end, NULL);
 372         }
 373 }
 374
 375 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 376 {
 377         pgoff_t pgoff;
 378         struct address_space *mapping = inode->i_mapping;
 379         struct hstate *h = hstate_inode(inode);
 380
 381         BUG_ON(offset & ~huge_page_mask(h));
 382         pgoff = offset >> PAGE_SHIFT;
 383
 384         i_size_write(inode, offset);
 385         i_mmap_lock_write(mapping);
 386         if (!RB_EMPTY_ROOT(&mapping->i_mmap))
 387                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 388         i_mmap_unlock_write(mapping);
 389         truncate_hugepages(inode, offset);
 390         return 0;
 391 }
 392
 393 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 394 {
 395         struct inode *inode = d_inode(dentry);
 396         struct hstate *h = hstate_inode(inode);
 397         int error;
 398         unsigned int ia_valid = attr->ia_valid;
 399
 400         BUG_ON(!inode);
 401
 402         error = inode_change_ok(inode, attr);
 403         if (error)
 404                 return error;
 405
 406         if (ia_valid & ATTR_SIZE) {
 407                 error = -EINVAL;
 408                 if (attr->ia_size & ~huge_page_mask(h))
 409                         return -EINVAL;
 410                 error = hugetlb_vmtruncate(inode, attr->ia_size);
 411                 if (error)
 412                         return error;
 413         }
 414
 415         setattr_copy(inode, attr);
 416         mark_inode_dirty(inode);
 417         return 0;
 418 }
 419
 420 static struct inode *hugetlbfs_get_root(struct super_block *sb,
 421                                         struct hugetlbfs_config *config)
 422 {
 423         struct inode *inode;
 424
 425         inode = new_inode(sb);
 426         if (inode) {
 427                 struct hugetlbfs_inode_info *info;
 428                 inode->i_ino = get_next_ino();
 429                 inode->i_mode = S_IFDIR | config->mode;
 430                 inode->i_uid = config->uid;
 431                 inode->i_gid = config->gid;
 432                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 433                 info = HUGETLBFS_I(inode);
 434                 mpol_shared_policy_init(&info->policy, NULL);
 435                 inode->i_op = &hugetlbfs_dir_inode_operations;
 436                 inode->i_fop = &simple_dir_operations;
 437                 /* directory inodes start off with i_nlink == 2 (for "." entry) */
 438                 inc_nlink(inode);
 439                 lockdep_annotate_inode_mutex_key(inode);
 440         }
 441         return inode;
 442 }
 443
 444 /*
 445  * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
 446  * be taken from reclaim -- unlike regular filesystems. This needs an
 447  * annotation because huge_pmd_share() does an allocation under
 448  * i_mmap_rwsem.
 449  */
 450 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
 451
 452 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
 453                                         struct inode *dir,
 454                                         umode_t mode, dev_t dev)
 455 {
 456         struct inode *inode;
 457         struct resv_map *resv_map;
 458
 459         resv_map = resv_map_alloc();
 460         if (!resv_map)
 461                 return NULL;
 462
 463         inode = new_inode(sb);
 464         if (inode) {
 465                 struct hugetlbfs_inode_info *info;
 466                 inode->i_ino = get_next_ino();
 467                 inode_init_owner(inode, dir, mode);
 468                 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
 469                                 &hugetlbfs_i_mmap_rwsem_key);
 470                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 471                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 472                 inode->i_mapping->private_data = resv_map;
 473                 info = HUGETLBFS_I(inode);
 474                 /*
 475                  * The policy is initialized here even if we are creating a
 476                  * private inode because initialization simply creates an
 477                  * an empty rb tree and calls spin_lock_init(), later when we
 478                  * call mpol_free_shared_policy() it will just return because
 479                  * the rb tree will still be empty.
 480                  */
 481                 mpol_shared_policy_init(&info->policy, NULL);
 482                 switch (mode & S_IFMT) {
 483                 default:
 484                         init_special_inode(inode, mode, dev);
 485                         break;
 486                 case S_IFREG:
 487                         inode->i_op = &hugetlbfs_inode_operations;
 488                         inode->i_fop = &hugetlbfs_file_operations;
 489                         break;
 490                 case S_IFDIR:
 491                         inode->i_op = &hugetlbfs_dir_inode_operations;
 492                         inode->i_fop = &simple_dir_operations;
 493
 494                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 495                         inc_nlink(inode);
 496                         break;
 497                 case S_IFLNK:
 498                         inode->i_op = &page_symlink_inode_operations;
 499                         break;
 500                 }
 501                 lockdep_annotate_inode_mutex_key(inode);
 502         } else
 503                 kref_put(&resv_map->refs, resv_map_release);
 504
 505         return inode;
 506 }
 507
 508 /*
 509  * File creation. Allocate an inode, and we're done..
 510  */
 511 static int hugetlbfs_mknod(struct inode *dir,
 512                         struct dentry *dentry, umode_t mode, dev_t dev)
 513 {
 514         struct inode *inode;
 515         int error = -ENOSPC;
 516
 517         inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
 518         if (inode) {
 519                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 520                 d_instantiate(dentry, inode);
 521                 dget(dentry);   /* Extra count - pin the dentry in core */
 522                 error = 0;
 523         }
 524         return error;
 525 }
 526
 527 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 528 {
 529         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 530         if (!retval)
 531                 inc_nlink(dir);
 532         return retval;
 533 }
 534
 535 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
 536 {
 537         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 538 }
 539
 540 static int hugetlbfs_symlink(struct inode *dir,
 541                         struct dentry *dentry, const char *symname)
 542 {
 543         struct inode *inode;
 544         int error = -ENOSPC;
 545
 546         inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
 547         if (inode) {
 548                 int l = strlen(symname)+1;
 549                 error = page_symlink(inode, symname, l);
 550                 if (!error) {
 551                         d_instantiate(dentry, inode);
 552                         dget(dentry);
 553                 } else
 554                         iput(inode);
 555         }
 556         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 557
 558         return error;
 559 }
 560
 561 /*
 562  * mark the head page dirty
 563  */
 564 static int hugetlbfs_set_page_dirty(struct page *page)
 565 {
 566         struct page *head = compound_head(page);
 567
 568         SetPageDirty(head);
 569         return 0;
 570 }
 571
 572 static int hugetlbfs_migrate_page(struct address_space *mapping,
 573                                 struct page *newpage, struct page *page,
 574                                 enum migrate_mode mode)
 575 {
 576         int rc;
 577
 578         rc = migrate_huge_page_move_mapping(mapping, newpage, page);
 579         if (rc != MIGRATEPAGE_SUCCESS)
 580                 return rc;
 581         migrate_page_copy(newpage, page);
 582
 583         return MIGRATEPAGE_SUCCESS;
 584 }
 585
 586 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 587 {
 588         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 589         struct hstate *h = hstate_inode(d_inode(dentry));
 590
 591         buf->f_type = HUGETLBFS_MAGIC;
 592         buf->f_bsize = huge_page_size(h);
 593         if (sbinfo) {
 594                 spin_lock(&sbinfo->stat_lock);
 595                 /* If no limits set, just report 0 for max/free/used
 596                  * blocks, like simple_statfs() */
 597                 if (sbinfo->spool) {
 598                         long free_pages;
 599
 600                         spin_lock(&sbinfo->spool->lock);
 601                         buf->f_blocks = sbinfo->spool->max_hpages;
 602                         free_pages = sbinfo->spool->max_hpages
 603                                 - sbinfo->spool->used_hpages;
 604                         buf->f_bavail = buf->f_bfree = free_pages;
 605                         spin_unlock(&sbinfo->spool->lock);
 606                         buf->f_files = sbinfo->max_inodes;
 607                         buf->f_ffree = sbinfo->free_inodes;
 608                 }
 609                 spin_unlock(&sbinfo->stat_lock);
 610         }
 611         buf->f_namelen = NAME_MAX;
 612         return 0;
 613 }
 614
 615 static void hugetlbfs_put_super(struct super_block *sb)
 616 {
 617         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 618
 619         if (sbi) {
 620                 sb->s_fs_info = NULL;
 621
 622                 if (sbi->spool)
 623                         hugepage_put_subpool(sbi->spool);
 624
 625                 kfree(sbi);
 626         }
 627 }
 628
 629 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 630 {
 631         if (sbinfo->free_inodes >= 0) {
 632                 spin_lock(&sbinfo->stat_lock);
 633                 if (unlikely(!sbinfo->free_inodes)) {
 634                         spin_unlock(&sbinfo->stat_lock);
 635                         return 0;
 636                 }
 637                 sbinfo->free_inodes--;
 638                 spin_unlock(&sbinfo->stat_lock);
 639         }
 640
 641         return 1;
 642 }
 643
 644 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 645 {
 646         if (sbinfo->free_inodes >= 0) {
 647                 spin_lock(&sbinfo->stat_lock);
 648                 sbinfo->free_inodes++;
 649                 spin_unlock(&sbinfo->stat_lock);
 650         }
 651 }
 652
 653
 654 static struct kmem_cache *hugetlbfs_inode_cachep;
 655
 656 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 657 {
 658         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 659         struct hugetlbfs_inode_info *p;
 660
 661         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 662                 return NULL;
 663         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 664         if (unlikely(!p)) {
 665                 hugetlbfs_inc_free_inodes(sbinfo);
 666                 return NULL;
 667         }
 668         return &p->vfs_inode;
 669 }
 670
 671 static void hugetlbfs_i_callback(struct rcu_head *head)
 672 {
 673         struct inode *inode = container_of(head, struct inode, i_rcu);
 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_iter              = hugetlbfs_read_iter,
 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 enum { NO_SIZE, SIZE_STD, SIZE_PERCENT };
 734
 735 /*
 736  * Convert size option passed from command line to number of huge pages
 737  * in the pool specified by hstate.  Size option could be in bytes
 738  * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
 739  */
 740 static long long
 741 hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
 742                                                                 int val_type)
 743 {
 744         if (val_type == NO_SIZE)
 745                 return -1;
 746
 747         if (val_type == SIZE_PERCENT) {
 748                 size_opt <<= huge_page_shift(h);
 749                 size_opt *= h->max_huge_pages;
 750                 do_div(size_opt, 100);
 751         }
 752
 753         size_opt >>= huge_page_shift(h);
 754         return size_opt;
 755 }
 756
 757 static int
 758 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 759 {
 760         char *p, *rest;
 761         substring_t args[MAX_OPT_ARGS];
 762         int option;
 763         unsigned long long max_size_opt = 0, min_size_opt = 0;
 764         int max_val_type = NO_SIZE, min_val_type = NO_SIZE;
 765
 766         if (!options)
 767                 return 0;
 768
 769         while ((p = strsep(&options, ",")) != NULL) {
 770                 int token;
 771                 if (!*p)
 772                         continue;
 773
 774                 token = match_token(p, tokens, args);
 775                 switch (token) {
 776                 case Opt_uid:
 777                         if (match_int(&args[0], &option))
 778                                 goto bad_val;
 779                         pconfig->uid = make_kuid(current_user_ns(), option);
 780                         if (!uid_valid(pconfig->uid))
 781                                 goto bad_val;
 782                         break;
 783
 784                 case Opt_gid:
 785                         if (match_int(&args[0], &option))
 786                                 goto bad_val;
 787                         pconfig->gid = make_kgid(current_user_ns(), option);
 788                         if (!gid_valid(pconfig->gid))
 789                                 goto bad_val;
 790                         break;
 791
 792                 case Opt_mode:
 793                         if (match_octal(&args[0], &option))
 794                                 goto bad_val;
 795                         pconfig->mode = option & 01777U;
 796                         break;
 797
 798                 case Opt_size: {
 799                         /* memparse() will accept a K/M/G without a digit */
 800                         if (!isdigit(*args[0].from))
 801                                 goto bad_val;
 802                         max_size_opt = memparse(args[0].from, &rest);
 803                         max_val_type = SIZE_STD;
 804                         if (*rest == '%')
 805                                 max_val_type = SIZE_PERCENT;
 806                         break;
 807                 }
 808
 809                 case Opt_nr_inodes:
 810                         /* memparse() will accept a K/M/G without a digit */
 811                         if (!isdigit(*args[0].from))
 812                                 goto bad_val;
 813                         pconfig->nr_inodes = memparse(args[0].from, &rest);
 814                         break;
 815
 816                 case Opt_pagesize: {
 817                         unsigned long ps;
 818                         ps = memparse(args[0].from, &rest);
 819                         pconfig->hstate = size_to_hstate(ps);
 820                         if (!pconfig->hstate) {
 821                                 pr_err("Unsupported page size %lu MB\n",
 822                                         ps >> 20);
 823                                 return -EINVAL;
 824                         }
 825                         break;
 826                 }
 827
 828                 case Opt_min_size: {
 829                         /* memparse() will accept a K/M/G without a digit */
 830                         if (!isdigit(*args[0].from))
 831                                 goto bad_val;
 832                         min_size_opt = memparse(args[0].from, &rest);
 833                         min_val_type = SIZE_STD;
 834                         if (*rest == '%')
 835                                 min_val_type = SIZE_PERCENT;
 836                         break;
 837                 }
 838
 839                 default:
 840                         pr_err("Bad mount option: \"%s\"\n", p);
 841                         return -EINVAL;
 842                         break;
 843                 }
 844         }
 845
 846         /*
 847          * Use huge page pool size (in hstate) to convert the size
 848          * options to number of huge pages.  If NO_SIZE, -1 is returned.
 849          */
 850         pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
 851                                                 max_size_opt, max_val_type);
 852         pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
 853                                                 min_size_opt, min_val_type);
 854
 855         /*
 856          * If max_size was specified, then min_size must be smaller
 857          */
 858         if (max_val_type > NO_SIZE &&
 859             pconfig->min_hpages > pconfig->max_hpages) {
 860                 pr_err("minimum size can not be greater than maximum size\n");
 861                 return -EINVAL;
 862         }
 863
 864         return 0;
 865
 866 bad_val:
 867         pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p);
 868         return -EINVAL;
 869 }
 870
 871 static int
 872 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 873 {
 874         int ret;
 875         struct hugetlbfs_config config;
 876         struct hugetlbfs_sb_info *sbinfo;
 877
 878         save_mount_options(sb, data);
 879
 880         config.max_hpages = -1; /* No limit on size by default */
 881         config.nr_inodes = -1; /* No limit on number of inodes by default */
 882         config.uid = current_fsuid();
 883         config.gid = current_fsgid();
 884         config.mode = 0755;
 885         config.hstate = &default_hstate;
 886         config.min_hpages = -1; /* No default minimum size */
 887         ret = hugetlbfs_parse_options(data, &config);
 888         if (ret)
 889                 return ret;
 890
 891         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 892         if (!sbinfo)
 893                 return -ENOMEM;
 894         sb->s_fs_info = sbinfo;
 895         sbinfo->hstate = config.hstate;
 896         spin_lock_init(&sbinfo->stat_lock);
 897         sbinfo->max_inodes = config.nr_inodes;
 898         sbinfo->free_inodes = config.nr_inodes;
 899         sbinfo->spool = NULL;
 900         /*
 901          * Allocate and initialize subpool if maximum or minimum size is
 902          * specified.  Any needed reservations (for minimim size) are taken
 903          * taken when the subpool is created.
 904          */
 905         if (config.max_hpages != -1 || config.min_hpages != -1) {
 906                 sbinfo->spool = hugepage_new_subpool(config.hstate,
 907                                                         config.max_hpages,
 908                                                         config.min_hpages);
 909                 if (!sbinfo->spool)
 910                         goto out_free;
 911         }
 912         sb->s_maxbytes = MAX_LFS_FILESIZE;
 913         sb->s_blocksize = huge_page_size(config.hstate);
 914         sb->s_blocksize_bits = huge_page_shift(config.hstate);
 915         sb->s_magic = HUGETLBFS_MAGIC;
 916         sb->s_op = &hugetlbfs_ops;
 917         sb->s_time_gran = 1;
 918         sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
 919         if (!sb->s_root)
 920                 goto out_free;
 921         return 0;
 922 out_free:
 923         kfree(sbinfo->spool);
 924         kfree(sbinfo);
 925         return -ENOMEM;
 926 }
 927
 928 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
 929         int flags, const char *dev_name, void *data)
 930 {
 931         return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 932 }
 933
 934 static struct file_system_type hugetlbfs_fs_type = {
 935         .name           = "hugetlbfs",
 936         .mount          = hugetlbfs_mount,
 937         .kill_sb        = kill_litter_super,
 938 };
 939 MODULE_ALIAS_FS("hugetlbfs");
 940
 941 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
 942
 943 static int can_do_hugetlb_shm(void)
 944 {
 945         kgid_t shm_group;
 946         shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
 947         return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
 948 }
 949
 950 static int get_hstate_idx(int page_size_log)
 951 {
 952         struct hstate *h = hstate_sizelog(page_size_log);
 953
 954         if (!h)
 955                 return -1;
 956         return h - hstates;
 957 }
 958
 959 static const struct dentry_operations anon_ops = {
 960         .d_dname = simple_dname
 961 };
 962
 963 /*
 964  * Note that size should be aligned to proper hugepage size in caller side,
 965  * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
 966  */
 967 struct file *hugetlb_file_setup(const char *name, size_t size,
 968                                 vm_flags_t acctflag, struct user_struct **user,
 969                                 int creat_flags, int page_size_log)
 970 {
 971         struct file *file = ERR_PTR(-ENOMEM);
 972         struct inode *inode;
 973         struct path path;
 974         struct super_block *sb;
 975         struct qstr quick_string;
 976         int hstate_idx;
 977
 978         hstate_idx = get_hstate_idx(page_size_log);
 979         if (hstate_idx < 0)
 980                 return ERR_PTR(-ENODEV);
 981
 982         *user = NULL;
 983         if (!hugetlbfs_vfsmount[hstate_idx])
 984                 return ERR_PTR(-ENOENT);
 985
 986         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
 987                 *user = current_user();
 988                 if (user_shm_lock(size, *user)) {
 989                         task_lock(current);
 990                         pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
 991                                 current->comm, current->pid);
 992                         task_unlock(current);
 993                 } else {
 994                         *user = NULL;
 995                         return ERR_PTR(-EPERM);
 996                 }
 997         }
 998
 999         sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb;
1000         quick_string.name = name;
1001         quick_string.len = strlen(quick_string.name);
1002         quick_string.hash = 0;
1003         path.dentry = d_alloc_pseudo(sb, &quick_string);
1004         if (!path.dentry)
1005                 goto out_shm_unlock;
1006
1007         d_set_d_op(path.dentry, &anon_ops);
1008         path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]);
1009         file = ERR_PTR(-ENOSPC);
1010         inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0);
1011         if (!inode)
1012                 goto out_dentry;
1013
1014         file = ERR_PTR(-ENOMEM);
1015         if (hugetlb_reserve_pages(inode, 0,
1016                         size >> huge_page_shift(hstate_inode(inode)), NULL,
1017                         acctflag))
1018                 goto out_inode;
1019
1020         d_instantiate(path.dentry, inode);
1021         inode->i_size = size;
1022         clear_nlink(inode);
1023
1024         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
1025                         &hugetlbfs_file_operations);
1026         if (IS_ERR(file))
1027                 goto out_dentry; /* inode is already attached */
1028
1029         return file;
1030
1031 out_inode:
1032         iput(inode);
1033 out_dentry:
1034         path_put(&path);
1035 out_shm_unlock:
1036         if (*user) {
1037                 user_shm_unlock(size, *user);
1038                 *user = NULL;
1039         }
1040         return file;
1041 }
1042
1043 static int __init init_hugetlbfs_fs(void)
1044 {
1045         struct hstate *h;
1046         int error;
1047         int i;
1048
1049         if (!hugepages_supported()) {
1050                 pr_info("disabling because there are no supported hugepage sizes\n");
1051                 return -ENOTSUPP;
1052         }
1053
1054         error = -ENOMEM;
1055         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1056                                         sizeof(struct hugetlbfs_inode_info),
1057                                         0, 0, init_once);
1058         if (hugetlbfs_inode_cachep == NULL)
1059                 goto out2;
1060
1061         error = register_filesystem(&hugetlbfs_fs_type);
1062         if (error)
1063                 goto out;
1064
1065         i = 0;
1066         for_each_hstate(h) {
1067                 char buf[50];
1068                 unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10);
1069
1070                 snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb);
1071                 hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type,
1072                                                         buf);
1073
1074                 if (IS_ERR(hugetlbfs_vfsmount[i])) {
1075                         pr_err("Cannot mount internal hugetlbfs for "
1076                                 "page size %uK", ps_kb);
1077                         error = PTR_ERR(hugetlbfs_vfsmount[i]);
1078                         hugetlbfs_vfsmount[i] = NULL;
1079                 }
1080                 i++;
1081         }
1082         /* Non default hstates are optional */
1083         if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx]))
1084                 return 0;
1085
1086  out:
1087         kmem_cache_destroy(hugetlbfs_inode_cachep);
1088  out2:
1089         return error;
1090 }
1091
1092 static void __exit exit_hugetlbfs_fs(void)
1093 {
1094         struct hstate *h;
1095         int i;
1096
1097
1098         /*
1099          * Make sure all delayed rcu free inodes are flushed before we
1100          * destroy cache.
1101          */
1102         rcu_barrier();
1103         kmem_cache_destroy(hugetlbfs_inode_cachep);
1104         i = 0;
1105         for_each_hstate(h)
1106                 kern_unmount(hugetlbfs_vfsmount[i++]);
1107         unregister_filesystem(&hugetlbfs_fs_type);
1108 }
1109
1110 module_init(init_hugetlbfs_fs)
1111 module_exit(exit_hugetlbfs_fs)
1112
1113 MODULE_LICENSE("GPL");