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    nr_blocks;
  52         long    nr_inodes;
  53         struct hstate *hstate;
  54 };
  55
  56 struct hugetlbfs_inode_info {
  57         struct shared_policy policy;
  58         struct inode vfs_inode;
  59 };
  60
  61 static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
  62 {
  63         return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
  64 }
  65
  66 int sysctl_hugetlb_shm_group;
  67
  68 enum {
  69         Opt_size, Opt_nr_inodes,
  70         Opt_mode, Opt_uid, Opt_gid,
  71         Opt_pagesize,
  72         Opt_err,
  73 };
  74
  75 static const match_table_t tokens = {
  76         {Opt_size,      "size=%s"},
  77         {Opt_nr_inodes, "nr_inodes=%s"},
  78         {Opt_mode,      "mode=%o"},
  79         {Opt_uid,       "uid=%u"},
  80         {Opt_gid,       "gid=%u"},
  81         {Opt_pagesize,  "pagesize=%s"},
  82         {Opt_err,       NULL},
  83 };
  84
  85 static void huge_pagevec_release(struct pagevec *pvec)
  86 {
  87         int i;
  88
  89         for (i = 0; i < pagevec_count(pvec); ++i)
  90                 put_page(pvec->pages[i]);
  91
  92         pagevec_reinit(pvec);
  93 }
  94
  95 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  96 {
  97         struct inode *inode = file_inode(file);
  98         loff_t len, vma_len;
  99         int ret;
 100         struct hstate *h = hstate_file(file);
 101
 102         /*
 103          * vma address alignment (but not the pgoff alignment) has
 104          * already been checked by prepare_hugepage_range.  If you add
 105          * any error returns here, do so after setting VM_HUGETLB, so
 106          * is_vm_hugetlb_page tests below unmap_region go the right
 107          * way when do_mmap_pgoff unwinds (may be important on powerpc
 108          * and ia64).
 109          */
 110         vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
 111         vma->vm_ops = &hugetlb_vm_ops;
 112
 113         if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
 114                 return -EINVAL;
 115
 116         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 117
 118         mutex_lock(&inode->i_mutex);
 119         file_accessed(file);
 120
 121         ret = -ENOMEM;
 122         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 123
 124         if (hugetlb_reserve_pages(inode,
 125                                 vma->vm_pgoff >> huge_page_order(h),
 126                                 len >> huge_page_shift(h), vma,
 127                                 vma->vm_flags))
 128                 goto out;
 129
 130         ret = 0;
 131         hugetlb_prefault_arch_hook(vma->vm_mm);
 132         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 133                 inode->i_size = len;
 134 out:
 135         mutex_unlock(&inode->i_mutex);
 136
 137         return ret;
 138 }
 139
 140 /*
 141  * Called under down_write(mmap_sem).
 142  */
 143
 144 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 145 static unsigned long
 146 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 147                 unsigned long len, unsigned long pgoff, unsigned long flags)
 148 {
 149         struct mm_struct *mm = current->mm;
 150         struct vm_area_struct *vma;
 151         struct hstate *h = hstate_file(file);
 152         struct vm_unmapped_area_info info;
 153
 154         if (len & ~huge_page_mask(h))
 155                 return -EINVAL;
 156         if (len > TASK_SIZE)
 157                 return -ENOMEM;
 158
 159         if (flags & MAP_FIXED) {
 160                 if (prepare_hugepage_range(file, addr, len))
 161                         return -EINVAL;
 162                 return addr;
 163         }
 164
 165         if (addr) {
 166                 addr = ALIGN(addr, huge_page_size(h));
 167                 vma = find_vma(mm, addr);
 168                 if (TASK_SIZE - len >= addr &&
 169                     (!vma || addr + len <= vma->vm_start))
 170                         return addr;
 171         }
 172
 173         info.flags = 0;
 174         info.length = len;
 175         info.low_limit = TASK_UNMAPPED_BASE;
 176         info.high_limit = TASK_SIZE;
 177         info.align_mask = PAGE_MASK & ~huge_page_mask(h);
 178         info.align_offset = 0;
 179         return vm_unmapped_area(&info);
 180 }
 181 #endif
 182
 183 static size_t
 184 hugetlbfs_read_actor(struct page *page, unsigned long offset,
 185                         struct iov_iter *to, unsigned long size)
 186 {
 187         size_t copied = 0;
 188         int i, chunksize;
 189
 190         /* Find which 4k chunk and offset with in that chunk */
 191         i = offset >> PAGE_CACHE_SHIFT;
 192         offset = offset & ~PAGE_CACHE_MASK;
 193
 194         while (size) {
 195                 size_t n;
 196                 chunksize = PAGE_CACHE_SIZE;
 197                 if (offset)
 198                         chunksize -= offset;
 199                 if (chunksize > size)
 200                         chunksize = size;
 201                 n = copy_page_to_iter(&page[i], offset, chunksize, to);
 202                 copied += n;
 203                 if (n != chunksize)
 204                         return copied;
 205                 offset = 0;
 206                 size -= chunksize;
 207                 i++;
 208         }
 209         return copied;
 210 }
 211
 212 /*
 213  * Support for read() - Find the page attached to f_mapping and copy out the
 214  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
 215  * since it has PAGE_CACHE_SIZE assumptions.
 216  */
 217 static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
 218 {
 219         struct file *file = iocb->ki_filp;
 220         struct hstate *h = hstate_file(file);
 221         struct address_space *mapping = file->f_mapping;
 222         struct inode *inode = mapping->host;
 223         unsigned long index = iocb->ki_pos >> huge_page_shift(h);
 224         unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
 225         unsigned long end_index;
 226         loff_t isize;
 227         ssize_t retval = 0;
 228
 229         while (iov_iter_count(to)) {
 230                 struct page *page;
 231                 size_t nr, copied;
 232
 233                 /* nr is the maximum number of bytes to copy from this page */
 234                 nr = huge_page_size(h);
 235                 isize = i_size_read(inode);
 236                 if (!isize)
 237                         break;
 238                 end_index = (isize - 1) >> huge_page_shift(h);
 239                 if (index > end_index)
 240                         break;
 241                 if (index == end_index) {
 242                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
 243                         if (nr <= offset)
 244                                 break;
 245                 }
 246                 nr = nr - offset;
 247
 248                 /* Find the page */
 249                 page = find_lock_page(mapping, index);
 250                 if (unlikely(page == NULL)) {
 251                         /*
 252                          * We have a HOLE, zero out the user-buffer for the
 253                          * length of the hole or request.
 254                          */
 255                         copied = iov_iter_zero(nr, to);
 256                 } else {
 257                         unlock_page(page);
 258
 259                         /*
 260                          * We have the page, copy it to user space buffer.
 261                          */
 262                         copied = hugetlbfs_read_actor(page, offset, to, nr);
 263                         page_cache_release(page);
 264                 }
 265                 offset += copied;
 266                 retval += copied;
 267                 if (copied != nr && iov_iter_count(to)) {
 268                         if (!retval)
 269                                 retval = -EFAULT;
 270                         break;
 271                 }
 272                 index += offset >> huge_page_shift(h);
 273                 offset &= ~huge_page_mask(h);
 274         }
 275         iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
 276         return retval;
 277 }
 278
 279 static int hugetlbfs_write_begin(struct file *file,
 280                         struct address_space *mapping,
 281                         loff_t pos, unsigned len, unsigned flags,
 282                         struct page **pagep, void **fsdata)
 283 {
 284         return -EINVAL;
 285 }
 286
 287 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 288                         loff_t pos, unsigned len, unsigned copied,
 289                         struct page *page, void *fsdata)
 290 {
 291         BUG();
 292         return -EINVAL;
 293 }
 294
 295 static void truncate_huge_page(struct page *page)
 296 {
 297         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
 298         ClearPageUptodate(page);
 299         delete_from_page_cache(page);
 300 }
 301
 302 static void truncate_hugepages(struct inode *inode, loff_t lstart)
 303 {
 304         struct hstate *h = hstate_inode(inode);
 305         struct address_space *mapping = &inode->i_data;
 306         const pgoff_t start = lstart >> huge_page_shift(h);
 307         struct pagevec pvec;
 308         pgoff_t next;
 309         int i, freed = 0;
 310
 311         pagevec_init(&pvec, 0);
 312         next = start;
 313         while (1) {
 314                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 315                         if (next == start)
 316                                 break;
 317                         next = start;
 318                         continue;
 319                 }
 320
 321                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 322                         struct page *page = pvec.pages[i];
 323
 324                         lock_page(page);
 325                         if (page->index > next)
 326                                 next = page->index;
 327                         ++next;
 328                         truncate_huge_page(page);
 329                         unlock_page(page);
 330                         freed++;
 331                 }
 332                 huge_pagevec_release(&pvec);
 333         }
 334         BUG_ON(!lstart && mapping->nrpages);
 335         hugetlb_unreserve_pages(inode, start, freed);
 336 }
 337
 338 static void hugetlbfs_evict_inode(struct inode *inode)
 339 {
 340         struct resv_map *resv_map;
 341
 342         truncate_hugepages(inode, 0);
 343         resv_map = (struct resv_map *)inode->i_mapping->private_data;
 344         /* root inode doesn't have the resv_map, so we should check it */
 345         if (resv_map)
 346                 resv_map_release(&resv_map->refs);
 347         clear_inode(inode);
 348 }
 349
 350 static inline void
 351 hugetlb_vmtruncate_list(struct rb_root *root, pgoff_t pgoff)
 352 {
 353         struct vm_area_struct *vma;
 354
 355         vma_interval_tree_foreach(vma, root, pgoff, ULONG_MAX) {
 356                 unsigned long v_offset;
 357
 358                 /*
 359                  * Can the expression below overflow on 32-bit arches?
 360                  * No, because the interval tree returns us only those vmas
 361                  * which overlap the truncated area starting at pgoff,
 362                  * and no vma on a 32-bit arch can span beyond the 4GB.
 363                  */
 364                 if (vma->vm_pgoff < pgoff)
 365                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 366                 else
 367                         v_offset = 0;
 368
 369                 unmap_hugepage_range(vma, vma->vm_start + v_offset,
 370                                      vma->vm_end, NULL);
 371         }
 372 }
 373
 374 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 375 {
 376         pgoff_t pgoff;
 377         struct address_space *mapping = inode->i_mapping;
 378         struct hstate *h = hstate_inode(inode);
 379
 380         BUG_ON(offset & ~huge_page_mask(h));
 381         pgoff = offset >> PAGE_SHIFT;
 382
 383         i_size_write(inode, offset);
 384         i_mmap_lock_write(mapping);
 385         if (!RB_EMPTY_ROOT(&mapping->i_mmap))
 386                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 387         i_mmap_unlock_write(mapping);
 388         truncate_hugepages(inode, offset);
 389         return 0;
 390 }
 391
 392 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 393 {
 394         struct inode *inode = dentry->d_inode;
 395         struct hstate *h = hstate_inode(inode);
 396         int error;
 397         unsigned int ia_valid = attr->ia_valid;
 398
 399         BUG_ON(!inode);
 400
 401         error = inode_change_ok(inode, attr);
 402         if (error)
 403                 return error;
 404
 405         if (ia_valid & ATTR_SIZE) {
 406                 error = -EINVAL;
 407                 if (attr->ia_size & ~huge_page_mask(h))
 408                         return -EINVAL;
 409                 error = hugetlb_vmtruncate(inode, attr->ia_size);
 410                 if (error)
 411                         return error;
 412         }
 413
 414         setattr_copy(inode, attr);
 415         mark_inode_dirty(inode);
 416         return 0;
 417 }
 418
 419 static struct inode *hugetlbfs_get_root(struct super_block *sb,
 420                                         struct hugetlbfs_config *config)
 421 {
 422         struct inode *inode;
 423
 424         inode = new_inode(sb);
 425         if (inode) {
 426                 struct hugetlbfs_inode_info *info;
 427                 inode->i_ino = get_next_ino();
 428                 inode->i_mode = S_IFDIR | config->mode;
 429                 inode->i_uid = config->uid;
 430                 inode->i_gid = config->gid;
 431                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 432                 info = HUGETLBFS_I(inode);
 433                 mpol_shared_policy_init(&info->policy, NULL);
 434                 inode->i_op = &hugetlbfs_dir_inode_operations;
 435                 inode->i_fop = &simple_dir_operations;
 436                 /* directory inodes start off with i_nlink == 2 (for "." entry) */
 437                 inc_nlink(inode);
 438                 lockdep_annotate_inode_mutex_key(inode);
 439         }
 440         return inode;
 441 }
 442
 443 /*
 444  * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
 445  * be taken from reclaim -- unlike regular filesystems. This needs an
 446  * annotation because huge_pmd_share() does an allocation under
 447  * i_mmap_rwsem.
 448  */
 449 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
 450
 451 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
 452                                         struct inode *dir,
 453                                         umode_t mode, dev_t dev)
 454 {
 455         struct inode *inode;
 456         struct resv_map *resv_map;
 457
 458         resv_map = resv_map_alloc();
 459         if (!resv_map)
 460                 return NULL;
 461
 462         inode = new_inode(sb);
 463         if (inode) {
 464                 struct hugetlbfs_inode_info *info;
 465                 inode->i_ino = get_next_ino();
 466                 inode_init_owner(inode, dir, mode);
 467                 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
 468                                 &hugetlbfs_i_mmap_rwsem_key);
 469                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 470                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 471                 inode->i_mapping->private_data = resv_map;
 472                 info = HUGETLBFS_I(inode);
 473                 /*
 474                  * The policy is initialized here even if we are creating a
 475                  * private inode because initialization simply creates an
 476                  * an empty rb tree and calls spin_lock_init(), later when we
 477                  * call mpol_free_shared_policy() it will just return because
 478                  * the rb tree will still be empty.
 479                  */
 480                 mpol_shared_policy_init(&info->policy, NULL);
 481                 switch (mode & S_IFMT) {
 482                 default:
 483                         init_special_inode(inode, mode, dev);
 484                         break;
 485                 case S_IFREG:
 486                         inode->i_op = &hugetlbfs_inode_operations;
 487                         inode->i_fop = &hugetlbfs_file_operations;
 488                         break;
 489                 case S_IFDIR:
 490                         inode->i_op = &hugetlbfs_dir_inode_operations;
 491                         inode->i_fop = &simple_dir_operations;
 492
 493                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 494                         inc_nlink(inode);
 495                         break;
 496                 case S_IFLNK:
 497                         inode->i_op = &page_symlink_inode_operations;
 498                         break;
 499                 }
 500                 lockdep_annotate_inode_mutex_key(inode);
 501         } else
 502                 kref_put(&resv_map->refs, resv_map_release);
 503
 504         return inode;
 505 }
 506
 507 /*
 508  * File creation. Allocate an inode, and we're done..
 509  */
 510 static int hugetlbfs_mknod(struct inode *dir,
 511                         struct dentry *dentry, umode_t mode, dev_t dev)
 512 {
 513         struct inode *inode;
 514         int error = -ENOSPC;
 515
 516         inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
 517         if (inode) {
 518                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 519                 d_instantiate(dentry, inode);
 520                 dget(dentry);   /* Extra count - pin the dentry in core */
 521                 error = 0;
 522         }
 523         return error;
 524 }
 525
 526 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 527 {
 528         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 529         if (!retval)
 530                 inc_nlink(dir);
 531         return retval;
 532 }
 533
 534 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
 535 {
 536         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 537 }
 538
 539 static int hugetlbfs_symlink(struct inode *dir,
 540                         struct dentry *dentry, const char *symname)
 541 {
 542         struct inode *inode;
 543         int error = -ENOSPC;
 544
 545         inode = hugetlbfs_get_inode(dir->i_sb, dir, 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                                 enum migrate_mode mode)
 574 {
 575         int rc;
 576
 577         rc = migrate_huge_page_move_mapping(mapping, newpage, page);
 578         if (rc != MIGRATEPAGE_SUCCESS)
 579                 return rc;
 580         migrate_page_copy(newpage, page);
 581
 582         return MIGRATEPAGE_SUCCESS;
 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         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 674 }
 675
 676 static void hugetlbfs_destroy_inode(struct inode *inode)
 677 {
 678         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 679         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 680         call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
 681 }
 682
 683 static const struct address_space_operations hugetlbfs_aops = {
 684         .write_begin    = hugetlbfs_write_begin,
 685         .write_end      = hugetlbfs_write_end,
 686         .set_page_dirty = hugetlbfs_set_page_dirty,
 687         .migratepage    = hugetlbfs_migrate_page,
 688 };
 689
 690
 691 static void init_once(void *foo)
 692 {
 693         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 694
 695         inode_init_once(&ei->vfs_inode);
 696 }
 697
 698 const struct file_operations hugetlbfs_file_operations = {
 699         .read_iter              = hugetlbfs_read_iter,
 700         .mmap                   = hugetlbfs_file_mmap,
 701         .fsync                  = noop_fsync,
 702         .get_unmapped_area      = hugetlb_get_unmapped_area,
 703         .llseek         = default_llseek,
 704 };
 705
 706 static const struct inode_operations hugetlbfs_dir_inode_operations = {
 707         .create         = hugetlbfs_create,
 708         .lookup         = simple_lookup,
 709         .link           = simple_link,
 710         .unlink         = simple_unlink,
 711         .symlink        = hugetlbfs_symlink,
 712         .mkdir          = hugetlbfs_mkdir,
 713         .rmdir          = simple_rmdir,
 714         .mknod          = hugetlbfs_mknod,
 715         .rename         = simple_rename,
 716         .setattr        = hugetlbfs_setattr,
 717 };
 718
 719 static const struct inode_operations hugetlbfs_inode_operations = {
 720         .setattr        = hugetlbfs_setattr,
 721 };
 722
 723 static const struct super_operations hugetlbfs_ops = {
 724         .alloc_inode    = hugetlbfs_alloc_inode,
 725         .destroy_inode  = hugetlbfs_destroy_inode,
 726         .evict_inode    = hugetlbfs_evict_inode,
 727         .statfs         = hugetlbfs_statfs,
 728         .put_super      = hugetlbfs_put_super,
 729         .show_options   = generic_show_options,
 730 };
 731
 732 static int
 733 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 734 {
 735         char *p, *rest;
 736         substring_t args[MAX_OPT_ARGS];
 737         int option;
 738         unsigned long long size = 0;
 739         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
 740
 741         if (!options)
 742                 return 0;
 743
 744         while ((p = strsep(&options, ",")) != NULL) {
 745                 int token;
 746                 if (!*p)
 747                         continue;
 748
 749                 token = match_token(p, tokens, args);
 750                 switch (token) {
 751                 case Opt_uid:
 752                         if (match_int(&args[0], &option))
 753                                 goto bad_val;
 754                         pconfig->uid = make_kuid(current_user_ns(), option);
 755                         if (!uid_valid(pconfig->uid))
 756                                 goto bad_val;
 757                         break;
 758
 759                 case Opt_gid:
 760                         if (match_int(&args[0], &option))
 761                                 goto bad_val;
 762                         pconfig->gid = make_kgid(current_user_ns(), option);
 763                         if (!gid_valid(pconfig->gid))
 764                                 goto bad_val;
 765                         break;
 766
 767                 case Opt_mode:
 768                         if (match_octal(&args[0], &option))
 769                                 goto bad_val;
 770                         pconfig->mode = option & 01777U;
 771                         break;
 772
 773                 case Opt_size: {
 774                         /* memparse() will accept a K/M/G without a digit */
 775                         if (!isdigit(*args[0].from))
 776                                 goto bad_val;
 777                         size = memparse(args[0].from, &rest);
 778                         setsize = SIZE_STD;
 779                         if (*rest == '%')
 780                                 setsize = SIZE_PERCENT;
 781                         break;
 782                 }
 783
 784                 case Opt_nr_inodes:
 785                         /* memparse() will accept a K/M/G without a digit */
 786                         if (!isdigit(*args[0].from))
 787                                 goto bad_val;
 788                         pconfig->nr_inodes = memparse(args[0].from, &rest);
 789                         break;
 790
 791                 case Opt_pagesize: {
 792                         unsigned long ps;
 793                         ps = memparse(args[0].from, &rest);
 794                         pconfig->hstate = size_to_hstate(ps);
 795                         if (!pconfig->hstate) {
 796                                 pr_err("Unsupported page size %lu MB\n",
 797                                         ps >> 20);
 798                                 return -EINVAL;
 799                         }
 800                         break;
 801                 }
 802
 803                 default:
 804                         pr_err("Bad mount option: \"%s\"\n", p);
 805                         return -EINVAL;
 806                         break;
 807                 }
 808         }
 809
 810         /* Do size after hstate is set up */
 811         if (setsize > NO_SIZE) {
 812                 struct hstate *h = pconfig->hstate;
 813                 if (setsize == SIZE_PERCENT) {
 814                         size <<= huge_page_shift(h);
 815                         size *= h->max_huge_pages;
 816                         do_div(size, 100);
 817                 }
 818                 pconfig->nr_blocks = (size >> huge_page_shift(h));
 819         }
 820
 821         return 0;
 822
 823 bad_val:
 824         pr_err("Bad value '%s' for mount option '%s'\n", 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         int ret;
 832         struct hugetlbfs_config config;
 833         struct hugetlbfs_sb_info *sbinfo;
 834
 835         save_mount_options(sb, data);
 836
 837         config.nr_blocks = -1; /* No limit on size by default */
 838         config.nr_inodes = -1; /* No limit on number of inodes by default */
 839         config.uid = current_fsuid();
 840         config.gid = current_fsgid();
 841         config.mode = 0755;
 842         config.hstate = &default_hstate;
 843         ret = hugetlbfs_parse_options(data, &config);
 844         if (ret)
 845                 return ret;
 846
 847         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 848         if (!sbinfo)
 849                 return -ENOMEM;
 850         sb->s_fs_info = sbinfo;
 851         sbinfo->hstate = config.hstate;
 852         spin_lock_init(&sbinfo->stat_lock);
 853         sbinfo->max_inodes = config.nr_inodes;
 854         sbinfo->free_inodes = config.nr_inodes;
 855         sbinfo->spool = NULL;
 856         if (config.nr_blocks != -1) {
 857                 sbinfo->spool = hugepage_new_subpool(config.nr_blocks);
 858                 if (!sbinfo->spool)
 859                         goto out_free;
 860         }
 861         sb->s_maxbytes = MAX_LFS_FILESIZE;
 862         sb->s_blocksize = huge_page_size(config.hstate);
 863         sb->s_blocksize_bits = huge_page_shift(config.hstate);
 864         sb->s_magic = HUGETLBFS_MAGIC;
 865         sb->s_op = &hugetlbfs_ops;
 866         sb->s_time_gran = 1;
 867         sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
 868         if (!sb->s_root)
 869                 goto out_free;
 870         return 0;
 871 out_free:
 872         kfree(sbinfo->spool);
 873         kfree(sbinfo);
 874         return -ENOMEM;
 875 }
 876
 877 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
 878         int flags, const char *dev_name, void *data)
 879 {
 880         return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 881 }
 882
 883 static struct file_system_type hugetlbfs_fs_type = {
 884         .name           = "hugetlbfs",
 885         .mount          = hugetlbfs_mount,
 886         .kill_sb        = kill_litter_super,
 887 };
 888 MODULE_ALIAS_FS("hugetlbfs");
 889
 890 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
 891
 892 static int can_do_hugetlb_shm(void)
 893 {
 894         kgid_t shm_group;
 895         shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
 896         return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
 897 }
 898
 899 static int get_hstate_idx(int page_size_log)
 900 {
 901         struct hstate *h = hstate_sizelog(page_size_log);
 902
 903         if (!h)
 904                 return -1;
 905         return h - hstates;
 906 }
 907
 908 static const struct dentry_operations anon_ops = {
 909         .d_dname = simple_dname
 910 };
 911
 912 /*
 913  * Note that size should be aligned to proper hugepage size in caller side,
 914  * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
 915  */
 916 struct file *hugetlb_file_setup(const char *name, size_t size,
 917                                 vm_flags_t acctflag, struct user_struct **user,
 918                                 int creat_flags, int page_size_log)
 919 {
 920         struct file *file = ERR_PTR(-ENOMEM);
 921         struct inode *inode;
 922         struct path path;
 923         struct super_block *sb;
 924         struct qstr quick_string;
 925         int hstate_idx;
 926
 927         hstate_idx = get_hstate_idx(page_size_log);
 928         if (hstate_idx < 0)
 929                 return ERR_PTR(-ENODEV);
 930
 931         *user = NULL;
 932         if (!hugetlbfs_vfsmount[hstate_idx])
 933                 return ERR_PTR(-ENOENT);
 934
 935         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
 936                 *user = current_user();
 937                 if (user_shm_lock(size, *user)) {
 938                         task_lock(current);
 939                         pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
 940                                 current->comm, current->pid);
 941                         task_unlock(current);
 942                 } else {
 943                         *user = NULL;
 944                         return ERR_PTR(-EPERM);
 945                 }
 946         }
 947
 948         sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb;
 949         quick_string.name = name;
 950         quick_string.len = strlen(quick_string.name);
 951         quick_string.hash = 0;
 952         path.dentry = d_alloc_pseudo(sb, &quick_string);
 953         if (!path.dentry)
 954                 goto out_shm_unlock;
 955
 956         d_set_d_op(path.dentry, &anon_ops);
 957         path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]);
 958         file = ERR_PTR(-ENOSPC);
 959         inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0);
 960         if (!inode)
 961                 goto out_dentry;
 962
 963         file = ERR_PTR(-ENOMEM);
 964         if (hugetlb_reserve_pages(inode, 0,
 965                         size >> huge_page_shift(hstate_inode(inode)), NULL,
 966                         acctflag))
 967                 goto out_inode;
 968
 969         d_instantiate(path.dentry, inode);
 970         inode->i_size = size;
 971         clear_nlink(inode);
 972
 973         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
 974                         &hugetlbfs_file_operations);
 975         if (IS_ERR(file))
 976                 goto out_dentry; /* inode is already attached */
 977
 978         return file;
 979
 980 out_inode:
 981         iput(inode);
 982 out_dentry:
 983         path_put(&path);
 984 out_shm_unlock:
 985         if (*user) {
 986                 user_shm_unlock(size, *user);
 987                 *user = NULL;
 988         }
 989         return file;
 990 }
 991
 992 static int __init init_hugetlbfs_fs(void)
 993 {
 994         struct hstate *h;
 995         int error;
 996         int i;
 997
 998         if (!hugepages_supported()) {
 999                 pr_info("disabling because there are no supported hugepage sizes\n");
1000                 return -ENOTSUPP;
1001         }
1002
1003         error = -ENOMEM;
1004         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1005                                         sizeof(struct hugetlbfs_inode_info),
1006                                         0, 0, init_once);
1007         if (hugetlbfs_inode_cachep == NULL)
1008                 goto out2;
1009
1010         error = register_filesystem(&hugetlbfs_fs_type);
1011         if (error)
1012                 goto out;
1013
1014         i = 0;
1015         for_each_hstate(h) {
1016                 char buf[50];
1017                 unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10);
1018
1019                 snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb);
1020                 hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type,
1021                                                         buf);
1022
1023                 if (IS_ERR(hugetlbfs_vfsmount[i])) {
1024                         pr_err("Cannot mount internal hugetlbfs for "
1025                                 "page size %uK", ps_kb);
1026                         error = PTR_ERR(hugetlbfs_vfsmount[i]);
1027                         hugetlbfs_vfsmount[i] = NULL;
1028                 }
1029                 i++;
1030         }
1031         /* Non default hstates are optional */
1032         if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx]))
1033                 return 0;
1034
1035  out:
1036         kmem_cache_destroy(hugetlbfs_inode_cachep);
1037  out2:
1038         return error;
1039 }
1040
1041 static void __exit exit_hugetlbfs_fs(void)
1042 {
1043         struct hstate *h;
1044         int i;
1045
1046
1047         /*
1048          * Make sure all delayed rcu free inodes are flushed before we
1049          * destroy cache.
1050          */
1051         rcu_barrier();
1052         kmem_cache_destroy(hugetlbfs_inode_cachep);
1053         i = 0;
1054         for_each_hstate(h)
1055                 kern_unmount(hugetlbfs_vfsmount[i++]);
1056         unregister_filesystem(&hugetlbfs_fs_type);
1057 }
1058
1059 module_init(init_hugetlbfs_fs)
1060 module_exit(exit_hugetlbfs_fs)
1061
1062 MODULE_LICENSE("GPL");