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 #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 struct hugetlbfs_config {
  45         kuid_t   uid;
  46         kgid_t   gid;
  47         umode_t mode;
  48         long    nr_blocks;
  49         long    nr_inodes;
  50         struct hstate *hstate;
  51 };
  52
  53 struct hugetlbfs_inode_info {
  54         struct shared_policy policy;
  55         struct inode vfs_inode;
  56 };
  57
  58 static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
  59 {
  60         return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
  61 }
  62
  63 static struct backing_dev_info hugetlbfs_backing_dev_info = {
  64         .name           = "hugetlbfs",
  65         .ra_pages       = 0,    /* No readahead */
  66         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
  67 };
  68
  69 int sysctl_hugetlb_shm_group;
  70
  71 enum {
  72         Opt_size, Opt_nr_inodes,
  73         Opt_mode, Opt_uid, Opt_gid,
  74         Opt_pagesize,
  75         Opt_err,
  76 };
  77
  78 static const match_table_t tokens = {
  79         {Opt_size,      "size=%s"},
  80         {Opt_nr_inodes, "nr_inodes=%s"},
  81         {Opt_mode,      "mode=%o"},
  82         {Opt_uid,       "uid=%u"},
  83         {Opt_gid,       "gid=%u"},
  84         {Opt_pagesize,  "pagesize=%s"},
  85         {Opt_err,       NULL},
  86 };
  87
  88 static void huge_pagevec_release(struct pagevec *pvec)
  89 {
  90         int i;
  91
  92         for (i = 0; i < pagevec_count(pvec); ++i)
  93                 put_page(pvec->pages[i]);
  94
  95         pagevec_reinit(pvec);
  96 }
  97
  98 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  99 {
 100         struct inode *inode = file_inode(file);
 101         loff_t len, vma_len;
 102         int ret;
 103         struct hstate *h = hstate_file(file);
 104
 105         /*
 106          * vma address alignment (but not the pgoff alignment) has
 107          * already been checked by prepare_hugepage_range.  If you add
 108          * any error returns here, do so after setting VM_HUGETLB, so
 109          * is_vm_hugetlb_page tests below unmap_region go the right
 110          * way when do_mmap_pgoff unwinds (may be important on powerpc
 111          * and ia64).
 112          */
 113         vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
 114         vma->vm_ops = &hugetlb_vm_ops;
 115
 116         if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
 117                 return -EINVAL;
 118
 119         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 120
 121         mutex_lock(&inode->i_mutex);
 122         file_accessed(file);
 123
 124         ret = -ENOMEM;
 125         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 126
 127         if (hugetlb_reserve_pages(inode,
 128                                 vma->vm_pgoff >> huge_page_order(h),
 129                                 len >> huge_page_shift(h), vma,
 130                                 vma->vm_flags))
 131                 goto out;
 132
 133         ret = 0;
 134         hugetlb_prefault_arch_hook(vma->vm_mm);
 135         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 136                 inode->i_size = len;
 137 out:
 138         mutex_unlock(&inode->i_mutex);
 139
 140         return ret;
 141 }
 142
 143 /*
 144  * Called under down_write(mmap_sem).
 145  */
 146
 147 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 148 static unsigned long
 149 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 150                 unsigned long len, unsigned long pgoff, unsigned long flags)
 151 {
 152         struct mm_struct *mm = current->mm;
 153         struct vm_area_struct *vma;
 154         struct hstate *h = hstate_file(file);
 155         struct vm_unmapped_area_info info;
 156
 157         if (len & ~huge_page_mask(h))
 158                 return -EINVAL;
 159         if (len > TASK_SIZE)
 160                 return -ENOMEM;
 161
 162         if (flags & MAP_FIXED) {
 163                 if (prepare_hugepage_range(file, addr, len))
 164                         return -EINVAL;
 165                 return addr;
 166         }
 167
 168         if (addr) {
 169                 addr = ALIGN(addr, huge_page_size(h));
 170                 vma = find_vma(mm, addr);
 171                 if (TASK_SIZE - len >= addr &&
 172                     (!vma || addr + len <= vma->vm_start))
 173                         return addr;
 174         }
 175
 176         info.flags = 0;
 177         info.length = len;
 178         info.low_limit = TASK_UNMAPPED_BASE;
 179         info.high_limit = TASK_SIZE;
 180         info.align_mask = PAGE_MASK & ~huge_page_mask(h);
 181         info.align_offset = 0;
 182         return vm_unmapped_area(&info);
 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         clear_inode(inode);
 371 }
 372
 373 static inline void
 374 hugetlb_vmtruncate_list(struct rb_root *root, pgoff_t pgoff)
 375 {
 376         struct vm_area_struct *vma;
 377
 378         vma_interval_tree_foreach(vma, root, pgoff, ULONG_MAX) {
 379                 unsigned long v_offset;
 380
 381                 /*
 382                  * Can the expression below overflow on 32-bit arches?
 383                  * No, because the interval tree returns us only those vmas
 384                  * which overlap the truncated area starting at pgoff,
 385                  * and no vma on a 32-bit arch can span beyond the 4GB.
 386                  */
 387                 if (vma->vm_pgoff < pgoff)
 388                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 389                 else
 390                         v_offset = 0;
 391
 392                 unmap_hugepage_range(vma, vma->vm_start + v_offset,
 393                                      vma->vm_end, NULL);
 394         }
 395 }
 396
 397 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 398 {
 399         pgoff_t pgoff;
 400         struct address_space *mapping = inode->i_mapping;
 401         struct hstate *h = hstate_inode(inode);
 402
 403         BUG_ON(offset & ~huge_page_mask(h));
 404         pgoff = offset >> PAGE_SHIFT;
 405
 406         i_size_write(inode, offset);
 407         mutex_lock(&mapping->i_mmap_mutex);
 408         if (!RB_EMPTY_ROOT(&mapping->i_mmap))
 409                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 410         mutex_unlock(&mapping->i_mmap_mutex);
 411         truncate_hugepages(inode, offset);
 412         return 0;
 413 }
 414
 415 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 416 {
 417         struct inode *inode = dentry->d_inode;
 418         struct hstate *h = hstate_inode(inode);
 419         int error;
 420         unsigned int ia_valid = attr->ia_valid;
 421
 422         BUG_ON(!inode);
 423
 424         error = inode_change_ok(inode, attr);
 425         if (error)
 426                 return error;
 427
 428         if (ia_valid & ATTR_SIZE) {
 429                 error = -EINVAL;
 430                 if (attr->ia_size & ~huge_page_mask(h))
 431                         return -EINVAL;
 432                 error = hugetlb_vmtruncate(inode, attr->ia_size);
 433                 if (error)
 434                         return error;
 435         }
 436
 437         setattr_copy(inode, attr);
 438         mark_inode_dirty(inode);
 439         return 0;
 440 }
 441
 442 static struct inode *hugetlbfs_get_root(struct super_block *sb,
 443                                         struct hugetlbfs_config *config)
 444 {
 445         struct inode *inode;
 446
 447         inode = new_inode(sb);
 448         if (inode) {
 449                 struct hugetlbfs_inode_info *info;
 450                 inode->i_ino = get_next_ino();
 451                 inode->i_mode = S_IFDIR | config->mode;
 452                 inode->i_uid = config->uid;
 453                 inode->i_gid = config->gid;
 454                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 455                 info = HUGETLBFS_I(inode);
 456                 mpol_shared_policy_init(&info->policy, NULL);
 457                 inode->i_op = &hugetlbfs_dir_inode_operations;
 458                 inode->i_fop = &simple_dir_operations;
 459                 /* directory inodes start off with i_nlink == 2 (for "." entry) */
 460                 inc_nlink(inode);
 461                 lockdep_annotate_inode_mutex_key(inode);
 462         }
 463         return inode;
 464 }
 465
 466 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
 467                                         struct inode *dir,
 468                                         umode_t mode, dev_t dev)
 469 {
 470         struct inode *inode;
 471
 472         inode = new_inode(sb);
 473         if (inode) {
 474                 struct hugetlbfs_inode_info *info;
 475                 inode->i_ino = get_next_ino();
 476                 inode_init_owner(inode, dir, mode);
 477                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 478                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 479                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 480                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
 481                 info = HUGETLBFS_I(inode);
 482                 /*
 483                  * The policy is initialized here even if we are creating a
 484                  * private inode because initialization simply creates an
 485                  * an empty rb tree and calls spin_lock_init(), later when we
 486                  * call mpol_free_shared_policy() it will just return because
 487                  * the rb tree will still be empty.
 488                  */
 489                 mpol_shared_policy_init(&info->policy, NULL);
 490                 switch (mode & S_IFMT) {
 491                 default:
 492                         init_special_inode(inode, mode, dev);
 493                         break;
 494                 case S_IFREG:
 495                         inode->i_op = &hugetlbfs_inode_operations;
 496                         inode->i_fop = &hugetlbfs_file_operations;
 497                         break;
 498                 case S_IFDIR:
 499                         inode->i_op = &hugetlbfs_dir_inode_operations;
 500                         inode->i_fop = &simple_dir_operations;
 501
 502                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 503                         inc_nlink(inode);
 504                         break;
 505                 case S_IFLNK:
 506                         inode->i_op = &page_symlink_inode_operations;
 507                         break;
 508                 }
 509                 lockdep_annotate_inode_mutex_key(inode);
 510         }
 511         return inode;
 512 }
 513
 514 /*
 515  * File creation. Allocate an inode, and we're done..
 516  */
 517 static int hugetlbfs_mknod(struct inode *dir,
 518                         struct dentry *dentry, umode_t mode, dev_t dev)
 519 {
 520         struct inode *inode;
 521         int error = -ENOSPC;
 522
 523         inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
 524         if (inode) {
 525                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 526                 d_instantiate(dentry, inode);
 527                 dget(dentry);   /* Extra count - pin the dentry in core */
 528                 error = 0;
 529         }
 530         return error;
 531 }
 532
 533 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 534 {
 535         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 536         if (!retval)
 537                 inc_nlink(dir);
 538         return retval;
 539 }
 540
 541 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
 542 {
 543         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 544 }
 545
 546 static int hugetlbfs_symlink(struct inode *dir,
 547                         struct dentry *dentry, const char *symname)
 548 {
 549         struct inode *inode;
 550         int error = -ENOSPC;
 551
 552         inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
 553         if (inode) {
 554                 int l = strlen(symname)+1;
 555                 error = page_symlink(inode, symname, l);
 556                 if (!error) {
 557                         d_instantiate(dentry, inode);
 558                         dget(dentry);
 559                 } else
 560                         iput(inode);
 561         }
 562         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 563
 564         return error;
 565 }
 566
 567 /*
 568  * mark the head page dirty
 569  */
 570 static int hugetlbfs_set_page_dirty(struct page *page)
 571 {
 572         struct page *head = compound_head(page);
 573
 574         SetPageDirty(head);
 575         return 0;
 576 }
 577
 578 static int hugetlbfs_migrate_page(struct address_space *mapping,
 579                                 struct page *newpage, struct page *page,
 580                                 enum migrate_mode mode)
 581 {
 582         int rc;
 583
 584         rc = migrate_huge_page_move_mapping(mapping, newpage, page);
 585         if (rc != MIGRATEPAGE_SUCCESS)
 586                 return rc;
 587         migrate_page_copy(newpage, page);
 588
 589         return MIGRATEPAGE_SUCCESS;
 590 }
 591
 592 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 593 {
 594         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 595         struct hstate *h = hstate_inode(dentry->d_inode);
 596
 597         buf->f_type = HUGETLBFS_MAGIC;
 598         buf->f_bsize = huge_page_size(h);
 599         if (sbinfo) {
 600                 spin_lock(&sbinfo->stat_lock);
 601                 /* If no limits set, just report 0 for max/free/used
 602                  * blocks, like simple_statfs() */
 603                 if (sbinfo->spool) {
 604                         long free_pages;
 605
 606                         spin_lock(&sbinfo->spool->lock);
 607                         buf->f_blocks = sbinfo->spool->max_hpages;
 608                         free_pages = sbinfo->spool->max_hpages
 609                                 - sbinfo->spool->used_hpages;
 610                         buf->f_bavail = buf->f_bfree = free_pages;
 611                         spin_unlock(&sbinfo->spool->lock);
 612                         buf->f_files = sbinfo->max_inodes;
 613                         buf->f_ffree = sbinfo->free_inodes;
 614                 }
 615                 spin_unlock(&sbinfo->stat_lock);
 616         }
 617         buf->f_namelen = NAME_MAX;
 618         return 0;
 619 }
 620
 621 static void hugetlbfs_put_super(struct super_block *sb)
 622 {
 623         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 624
 625         if (sbi) {
 626                 sb->s_fs_info = NULL;
 627
 628                 if (sbi->spool)
 629                         hugepage_put_subpool(sbi->spool);
 630
 631                 kfree(sbi);
 632         }
 633 }
 634
 635 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 636 {
 637         if (sbinfo->free_inodes >= 0) {
 638                 spin_lock(&sbinfo->stat_lock);
 639                 if (unlikely(!sbinfo->free_inodes)) {
 640                         spin_unlock(&sbinfo->stat_lock);
 641                         return 0;
 642                 }
 643                 sbinfo->free_inodes--;
 644                 spin_unlock(&sbinfo->stat_lock);
 645         }
 646
 647         return 1;
 648 }
 649
 650 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 651 {
 652         if (sbinfo->free_inodes >= 0) {
 653                 spin_lock(&sbinfo->stat_lock);
 654                 sbinfo->free_inodes++;
 655                 spin_unlock(&sbinfo->stat_lock);
 656         }
 657 }
 658
 659
 660 static struct kmem_cache *hugetlbfs_inode_cachep;
 661
 662 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 663 {
 664         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 665         struct hugetlbfs_inode_info *p;
 666
 667         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 668                 return NULL;
 669         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 670         if (unlikely(!p)) {
 671                 hugetlbfs_inc_free_inodes(sbinfo);
 672                 return NULL;
 673         }
 674         return &p->vfs_inode;
 675 }
 676
 677 static void hugetlbfs_i_callback(struct rcu_head *head)
 678 {
 679         struct inode *inode = container_of(head, struct inode, i_rcu);
 680         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 681 }
 682
 683 static void hugetlbfs_destroy_inode(struct inode *inode)
 684 {
 685         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 686         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 687         call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
 688 }
 689
 690 static const struct address_space_operations hugetlbfs_aops = {
 691         .write_begin    = hugetlbfs_write_begin,
 692         .write_end      = hugetlbfs_write_end,
 693         .set_page_dirty = hugetlbfs_set_page_dirty,
 694         .migratepage    = hugetlbfs_migrate_page,
 695 };
 696
 697
 698 static void init_once(void *foo)
 699 {
 700         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 701
 702         inode_init_once(&ei->vfs_inode);
 703 }
 704
 705 const struct file_operations hugetlbfs_file_operations = {
 706         .read                   = hugetlbfs_read,
 707         .mmap                   = hugetlbfs_file_mmap,
 708         .fsync                  = noop_fsync,
 709         .get_unmapped_area      = hugetlb_get_unmapped_area,
 710         .llseek         = default_llseek,
 711 };
 712
 713 static const struct inode_operations hugetlbfs_dir_inode_operations = {
 714         .create         = hugetlbfs_create,
 715         .lookup         = simple_lookup,
 716         .link           = simple_link,
 717         .unlink         = simple_unlink,
 718         .symlink        = hugetlbfs_symlink,
 719         .mkdir          = hugetlbfs_mkdir,
 720         .rmdir          = simple_rmdir,
 721         .mknod          = hugetlbfs_mknod,
 722         .rename         = simple_rename,
 723         .setattr        = hugetlbfs_setattr,
 724 };
 725
 726 static const struct inode_operations hugetlbfs_inode_operations = {
 727         .setattr        = hugetlbfs_setattr,
 728 };
 729
 730 static const struct super_operations hugetlbfs_ops = {
 731         .alloc_inode    = hugetlbfs_alloc_inode,
 732         .destroy_inode  = hugetlbfs_destroy_inode,
 733         .evict_inode    = hugetlbfs_evict_inode,
 734         .statfs         = hugetlbfs_statfs,
 735         .put_super      = hugetlbfs_put_super,
 736         .show_options   = generic_show_options,
 737 };
 738
 739 static int
 740 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 741 {
 742         char *p, *rest;
 743         substring_t args[MAX_OPT_ARGS];
 744         int option;
 745         unsigned long long size = 0;
 746         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
 747
 748         if (!options)
 749                 return 0;
 750
 751         while ((p = strsep(&options, ",")) != NULL) {
 752                 int token;
 753                 if (!*p)
 754                         continue;
 755
 756                 token = match_token(p, tokens, args);
 757                 switch (token) {
 758                 case Opt_uid:
 759                         if (match_int(&args[0], &option))
 760                                 goto bad_val;
 761                         pconfig->uid = make_kuid(current_user_ns(), option);
 762                         if (!uid_valid(pconfig->uid))
 763                                 goto bad_val;
 764                         break;
 765
 766                 case Opt_gid:
 767                         if (match_int(&args[0], &option))
 768                                 goto bad_val;
 769                         pconfig->gid = make_kgid(current_user_ns(), option);
 770                         if (!gid_valid(pconfig->gid))
 771                                 goto bad_val;
 772                         break;
 773
 774                 case Opt_mode:
 775                         if (match_octal(&args[0], &option))
 776                                 goto bad_val;
 777                         pconfig->mode = option & 01777U;
 778                         break;
 779
 780                 case Opt_size: {
 781                         /* memparse() will accept a K/M/G without a digit */
 782                         if (!isdigit(*args[0].from))
 783                                 goto bad_val;
 784                         size = memparse(args[0].from, &rest);
 785                         setsize = SIZE_STD;
 786                         if (*rest == '%')
 787                                 setsize = SIZE_PERCENT;
 788                         break;
 789                 }
 790
 791                 case Opt_nr_inodes:
 792                         /* memparse() will accept a K/M/G without a digit */
 793                         if (!isdigit(*args[0].from))
 794                                 goto bad_val;
 795                         pconfig->nr_inodes = memparse(args[0].from, &rest);
 796                         break;
 797
 798                 case Opt_pagesize: {
 799                         unsigned long ps;
 800                         ps = memparse(args[0].from, &rest);
 801                         pconfig->hstate = size_to_hstate(ps);
 802                         if (!pconfig->hstate) {
 803                                 printk(KERN_ERR
 804                                 "hugetlbfs: Unsupported page size %lu MB\n",
 805                                         ps >> 20);
 806                                 return -EINVAL;
 807                         }
 808                         break;
 809                 }
 810
 811                 default:
 812                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
 813                                  p);
 814                         return -EINVAL;
 815                         break;
 816                 }
 817         }
 818
 819         /* Do size after hstate is set up */
 820         if (setsize > NO_SIZE) {
 821                 struct hstate *h = pconfig->hstate;
 822                 if (setsize == SIZE_PERCENT) {
 823                         size <<= huge_page_shift(h);
 824                         size *= h->max_huge_pages;
 825                         do_div(size, 100);
 826                 }
 827                 pconfig->nr_blocks = (size >> huge_page_shift(h));
 828         }
 829
 830         return 0;
 831
 832 bad_val:
 833         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
 834                args[0].from, p);
 835         return -EINVAL;
 836 }
 837
 838 static int
 839 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 840 {
 841         int ret;
 842         struct hugetlbfs_config config;
 843         struct hugetlbfs_sb_info *sbinfo;
 844
 845         save_mount_options(sb, data);
 846
 847         config.nr_blocks = -1; /* No limit on size by default */
 848         config.nr_inodes = -1; /* No limit on number of inodes by default */
 849         config.uid = current_fsuid();
 850         config.gid = current_fsgid();
 851         config.mode = 0755;
 852         config.hstate = &default_hstate;
 853         ret = hugetlbfs_parse_options(data, &config);
 854         if (ret)
 855                 return ret;
 856
 857         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 858         if (!sbinfo)
 859                 return -ENOMEM;
 860         sb->s_fs_info = sbinfo;
 861         sbinfo->hstate = config.hstate;
 862         spin_lock_init(&sbinfo->stat_lock);
 863         sbinfo->max_inodes = config.nr_inodes;
 864         sbinfo->free_inodes = config.nr_inodes;
 865         sbinfo->spool = NULL;
 866         if (config.nr_blocks != -1) {
 867                 sbinfo->spool = hugepage_new_subpool(config.nr_blocks);
 868                 if (!sbinfo->spool)
 869                         goto out_free;
 870         }
 871         sb->s_maxbytes = MAX_LFS_FILESIZE;
 872         sb->s_blocksize = huge_page_size(config.hstate);
 873         sb->s_blocksize_bits = huge_page_shift(config.hstate);
 874         sb->s_magic = HUGETLBFS_MAGIC;
 875         sb->s_op = &hugetlbfs_ops;
 876         sb->s_time_gran = 1;
 877         sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
 878         if (!sb->s_root)
 879                 goto out_free;
 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 MODULE_ALIAS_FS("hugetlbfs");
 900
 901 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
 902
 903 static int can_do_hugetlb_shm(void)
 904 {
 905         kgid_t shm_group;
 906         shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
 907         return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
 908 }
 909
 910 static int get_hstate_idx(int page_size_log)
 911 {
 912         struct hstate *h = hstate_sizelog(page_size_log);
 913
 914         if (!h)
 915                 return -1;
 916         return h - hstates;
 917 }
 918
 919 static char *hugetlb_dname(struct dentry *dentry, char *buffer, int buflen)
 920 {
 921         return dynamic_dname(dentry, buffer, buflen, "/%s (deleted)",
 922                                 dentry->d_name.name);
 923 }
 924
 925 static struct dentry_operations anon_ops = {
 926         .d_dname = hugetlb_dname
 927 };
 928
 929 /*
 930  * Note that size should be aligned to proper hugepage size in caller side,
 931  * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
 932  */
 933 struct file *hugetlb_file_setup(const char *name, size_t size,
 934                                 vm_flags_t acctflag, struct user_struct **user,
 935                                 int creat_flags, int page_size_log)
 936 {
 937         struct file *file = ERR_PTR(-ENOMEM);
 938         struct inode *inode;
 939         struct path path;
 940         struct super_block *sb;
 941         struct qstr quick_string;
 942         int hstate_idx;
 943
 944         hstate_idx = get_hstate_idx(page_size_log);
 945         if (hstate_idx < 0)
 946                 return ERR_PTR(-ENODEV);
 947
 948         *user = NULL;
 949         if (!hugetlbfs_vfsmount[hstate_idx])
 950                 return ERR_PTR(-ENOENT);
 951
 952         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
 953                 *user = current_user();
 954                 if (user_shm_lock(size, *user)) {
 955                         task_lock(current);
 956                         printk_once(KERN_WARNING
 957                                 "%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
 958                                 current->comm, current->pid);
 959                         task_unlock(current);
 960                 } else {
 961                         *user = NULL;
 962                         return ERR_PTR(-EPERM);
 963                 }
 964         }
 965
 966         sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb;
 967         quick_string.name = name;
 968         quick_string.len = strlen(quick_string.name);
 969         quick_string.hash = 0;
 970         path.dentry = d_alloc_pseudo(sb, &quick_string);
 971         if (!path.dentry)
 972                 goto out_shm_unlock;
 973
 974         d_set_d_op(path.dentry, &anon_ops);
 975         path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]);
 976         file = ERR_PTR(-ENOSPC);
 977         inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0);
 978         if (!inode)
 979                 goto out_dentry;
 980
 981         file = ERR_PTR(-ENOMEM);
 982         if (hugetlb_reserve_pages(inode, 0,
 983                         size >> huge_page_shift(hstate_inode(inode)), NULL,
 984                         acctflag))
 985                 goto out_inode;
 986
 987         d_instantiate(path.dentry, inode);
 988         inode->i_size = size;
 989         clear_nlink(inode);
 990
 991         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
 992                         &hugetlbfs_file_operations);
 993         if (IS_ERR(file))
 994                 goto out_dentry; /* inode is already attached */
 995
 996         return file;
 997
 998 out_inode:
 999         iput(inode);
1000 out_dentry:
1001         path_put(&path);
1002 out_shm_unlock:
1003         if (*user) {
1004                 user_shm_unlock(size, *user);
1005                 *user = NULL;
1006         }
1007         return file;
1008 }
1009
1010 static int __init init_hugetlbfs_fs(void)
1011 {
1012         struct hstate *h;
1013         int error;
1014         int i;
1015
1016         error = bdi_init(&hugetlbfs_backing_dev_info);
1017         if (error)
1018                 return error;
1019
1020         error = -ENOMEM;
1021         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1022                                         sizeof(struct hugetlbfs_inode_info),
1023                                         0, 0, init_once);
1024         if (hugetlbfs_inode_cachep == NULL)
1025                 goto out2;
1026
1027         error = register_filesystem(&hugetlbfs_fs_type);
1028         if (error)
1029                 goto out;
1030
1031         i = 0;
1032         for_each_hstate(h) {
1033                 char buf[50];
1034                 unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10);
1035
1036                 snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb);
1037                 hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type,
1038                                                         buf);
1039
1040                 if (IS_ERR(hugetlbfs_vfsmount[i])) {
1041                         pr_err("hugetlb: Cannot mount internal hugetlbfs for "
1042                                 "page size %uK", ps_kb);
1043                         error = PTR_ERR(hugetlbfs_vfsmount[i]);
1044                         hugetlbfs_vfsmount[i] = NULL;
1045                 }
1046                 i++;
1047         }
1048         /* Non default hstates are optional */
1049         if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx]))
1050                 return 0;
1051
1052  out:
1053         kmem_cache_destroy(hugetlbfs_inode_cachep);
1054  out2:
1055         bdi_destroy(&hugetlbfs_backing_dev_info);
1056         return error;
1057 }
1058
1059 static void __exit exit_hugetlbfs_fs(void)
1060 {
1061         struct hstate *h;
1062         int i;
1063
1064
1065         /*
1066          * Make sure all delayed rcu free inodes are flushed before we
1067          * destroy cache.
1068          */
1069         rcu_barrier();
1070         kmem_cache_destroy(hugetlbfs_inode_cachep);
1071         i = 0;
1072         for_each_hstate(h)
1073                 kern_unmount(hugetlbfs_vfsmount[i++]);
1074         unregister_filesystem(&hugetlbfs_fs_type);
1075         bdi_destroy(&hugetlbfs_backing_dev_info);
1076 }
1077
1078 module_init(init_hugetlbfs_fs)
1079 module_exit(exit_hugetlbfs_fs)
1080
1081 MODULE_LICENSE("GPL");