fs/hugetlbfs/inode.c

   1 /*
   2  * hugetlbpage-backed filesystem.  Based on ramfs.
   3  *
   4  * William Irwin, 2002
   5  *
   6  * Copyright (C) 2002 Linus Torvalds.
   7  */
   8
   9 #include <linux/module.h>
  10 #include <linux/thread_info.h>
  11 #include <asm/current.h>
  12 #include <linux/sched.h>                /* remove ASAP */
  13 #include <linux/fs.h>
  14 #include <linux/mount.h>
  15 #include <linux/file.h>
  16 #include <linux/kernel.h>
  17 #include <linux/writeback.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/highmem.h>
  20 #include <linux/init.h>
  21 #include <linux/string.h>
  22 #include <linux/capability.h>
  23 #include <linux/ctype.h>
  24 #include <linux/backing-dev.h>
  25 #include <linux/hugetlb.h>
  26 #include <linux/pagevec.h>
  27 #include <linux/parser.h>
  28 #include <linux/mman.h>
  29 #include <linux/slab.h>
  30 #include <linux/dnotify.h>
  31 #include <linux/statfs.h>
  32 #include <linux/security.h>
  33 #include <linux/ima.h>
  34
  35 #include <asm/uaccess.h>
  36
  37 /* some random number */
  38 #define HUGETLBFS_MAGIC 0x958458f6
  39
  40 static const struct super_operations hugetlbfs_ops;
  41 static const struct address_space_operations hugetlbfs_aops;
  42 const struct file_operations hugetlbfs_file_operations;
  43 static const struct inode_operations hugetlbfs_dir_inode_operations;
  44 static const struct inode_operations hugetlbfs_inode_operations;
  45
  46 static struct backing_dev_info hugetlbfs_backing_dev_info = {
  47         .ra_pages       = 0,    /* No readahead */
  48         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
  49 };
  50
  51 int sysctl_hugetlb_shm_group;
  52
  53 enum {
  54         Opt_size, Opt_nr_inodes,
  55         Opt_mode, Opt_uid, Opt_gid,
  56         Opt_pagesize,
  57         Opt_err,
  58 };
  59
  60 static const match_table_t tokens = {
  61         {Opt_size,      "size=%s"},
  62         {Opt_nr_inodes, "nr_inodes=%s"},
  63         {Opt_mode,      "mode=%o"},
  64         {Opt_uid,       "uid=%u"},
  65         {Opt_gid,       "gid=%u"},
  66         {Opt_pagesize,  "pagesize=%s"},
  67         {Opt_err,       NULL},
  68 };
  69
  70 static void huge_pagevec_release(struct pagevec *pvec)
  71 {
  72         int i;
  73
  74         for (i = 0; i < pagevec_count(pvec); ++i)
  75                 put_page(pvec->pages[i]);
  76
  77         pagevec_reinit(pvec);
  78 }
  79
  80 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  81 {
  82         struct inode *inode = file->f_path.dentry->d_inode;
  83         loff_t len, vma_len;
  84         int ret;
  85         struct hstate *h = hstate_file(file);
  86
  87         /*
  88          * vma address alignment (but not the pgoff alignment) has
  89          * already been checked by prepare_hugepage_range.  If you add
  90          * any error returns here, do so after setting VM_HUGETLB, so
  91          * is_vm_hugetlb_page tests below unmap_region go the right
  92          * way when do_mmap_pgoff unwinds (may be important on powerpc
  93          * and ia64).
  94          */
  95         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
  96         vma->vm_ops = &hugetlb_vm_ops;
  97
  98         if (vma->vm_pgoff & ~(huge_page_mask(h) >> PAGE_SHIFT))
  99                 return -EINVAL;
 100
 101         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 102
 103         mutex_lock(&inode->i_mutex);
 104         file_accessed(file);
 105
 106         ret = -ENOMEM;
 107         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 108
 109         if (hugetlb_reserve_pages(inode,
 110                                 vma->vm_pgoff >> huge_page_order(h),
 111                                 len >> huge_page_shift(h), vma,
 112                                 vma->vm_flags))
 113                 goto out;
 114
 115         ret = 0;
 116         hugetlb_prefault_arch_hook(vma->vm_mm);
 117         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 118                 inode->i_size = len;
 119 out:
 120         mutex_unlock(&inode->i_mutex);
 121
 122         return ret;
 123 }
 124
 125 /*
 126  * Called under down_write(mmap_sem).
 127  */
 128
 129 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 130 static unsigned long
 131 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 132                 unsigned long len, unsigned long pgoff, unsigned long flags)
 133 {
 134         struct mm_struct *mm = current->mm;
 135         struct vm_area_struct *vma;
 136         unsigned long start_addr;
 137         struct hstate *h = hstate_file(file);
 138
 139         if (len & ~huge_page_mask(h))
 140                 return -EINVAL;
 141         if (len > TASK_SIZE)
 142                 return -ENOMEM;
 143
 144         if (flags & MAP_FIXED) {
 145                 if (prepare_hugepage_range(file, addr, len))
 146                         return -EINVAL;
 147                 return addr;
 148         }
 149
 150         if (addr) {
 151                 addr = ALIGN(addr, huge_page_size(h));
 152                 vma = find_vma(mm, addr);
 153                 if (TASK_SIZE - len >= addr &&
 154                     (!vma || addr + len <= vma->vm_start))
 155                         return addr;
 156         }
 157
 158         start_addr = mm->free_area_cache;
 159
 160         if (len <= mm->cached_hole_size)
 161                 start_addr = TASK_UNMAPPED_BASE;
 162
 163 full_search:
 164         addr = ALIGN(start_addr, huge_page_size(h));
 165
 166         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 167                 /* At this point:  (!vma || addr < vma->vm_end). */
 168                 if (TASK_SIZE - len < addr) {
 169                         /*
 170                          * Start a new search - just in case we missed
 171                          * some holes.
 172                          */
 173                         if (start_addr != TASK_UNMAPPED_BASE) {
 174                                 start_addr = TASK_UNMAPPED_BASE;
 175                                 goto full_search;
 176                         }
 177                         return -ENOMEM;
 178                 }
 179
 180                 if (!vma || addr + len <= vma->vm_start)
 181                         return addr;
 182                 addr = ALIGN(vma->vm_end, huge_page_size(h));
 183         }
 184 }
 185 #endif
 186
 187 static int
 188 hugetlbfs_read_actor(struct page *page, unsigned long offset,
 189                         char __user *buf, unsigned long count,
 190                         unsigned long size)
 191 {
 192         char *kaddr;
 193         unsigned long left, copied = 0;
 194         int i, chunksize;
 195
 196         if (size > count)
 197                 size = count;
 198
 199         /* Find which 4k chunk and offset with in that chunk */
 200         i = offset >> PAGE_CACHE_SHIFT;
 201         offset = offset & ~PAGE_CACHE_MASK;
 202
 203         while (size) {
 204                 chunksize = PAGE_CACHE_SIZE;
 205                 if (offset)
 206                         chunksize -= offset;
 207                 if (chunksize > size)
 208                         chunksize = size;
 209                 kaddr = kmap(&page[i]);
 210                 left = __copy_to_user(buf, kaddr + offset, chunksize);
 211                 kunmap(&page[i]);
 212                 if (left) {
 213                         copied += (chunksize - left);
 214                         break;
 215                 }
 216                 offset = 0;
 217                 size -= chunksize;
 218                 buf += chunksize;
 219                 copied += chunksize;
 220                 i++;
 221         }
 222         return copied ? copied : -EFAULT;
 223 }
 224
 225 /*
 226  * Support for read() - Find the page attached to f_mapping and copy out the
 227  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
 228  * since it has PAGE_CACHE_SIZE assumptions.
 229  */
 230 static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
 231                               size_t len, loff_t *ppos)
 232 {
 233         struct hstate *h = hstate_file(filp);
 234         struct address_space *mapping = filp->f_mapping;
 235         struct inode *inode = mapping->host;
 236         unsigned long index = *ppos >> huge_page_shift(h);
 237         unsigned long offset = *ppos & ~huge_page_mask(h);
 238         unsigned long end_index;
 239         loff_t isize;
 240         ssize_t retval = 0;
 241
 242         mutex_lock(&inode->i_mutex);
 243
 244         /* validate length */
 245         if (len == 0)
 246                 goto out;
 247
 248         isize = i_size_read(inode);
 249         if (!isize)
 250                 goto out;
 251
 252         end_index = (isize - 1) >> huge_page_shift(h);
 253         for (;;) {
 254                 struct page *page;
 255                 unsigned long nr, ret;
 256                 int ra;
 257
 258                 /* nr is the maximum number of bytes to copy from this page */
 259                 nr = huge_page_size(h);
 260                 if (index >= end_index) {
 261                         if (index > end_index)
 262                                 goto out;
 263                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
 264                         if (nr <= offset) {
 265                                 goto out;
 266                         }
 267                 }
 268                 nr = nr - offset;
 269
 270                 /* Find the page */
 271                 page = find_get_page(mapping, index);
 272                 if (unlikely(page == NULL)) {
 273                         /*
 274                          * We have a HOLE, zero out the user-buffer for the
 275                          * length of the hole or request.
 276                          */
 277                         ret = len < nr ? len : nr;
 278                         if (clear_user(buf, ret))
 279                                 ra = -EFAULT;
 280                         else
 281                                 ra = 0;
 282                 } else {
 283                         /*
 284                          * We have the page, copy it to user space buffer.
 285                          */
 286                         ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
 287                         ret = ra;
 288                 }
 289                 if (ra < 0) {
 290                         if (retval == 0)
 291                                 retval = ra;
 292                         if (page)
 293                                 page_cache_release(page);
 294                         goto out;
 295                 }
 296
 297                 offset += ret;
 298                 retval += ret;
 299                 len -= ret;
 300                 index += offset >> huge_page_shift(h);
 301                 offset &= ~huge_page_mask(h);
 302
 303                 if (page)
 304                         page_cache_release(page);
 305
 306                 /* short read or no more work */
 307                 if ((ret != nr) || (len == 0))
 308                         break;
 309         }
 310 out:
 311         *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
 312         mutex_unlock(&inode->i_mutex);
 313         return retval;
 314 }
 315
 316 static int hugetlbfs_write_begin(struct file *file,
 317                         struct address_space *mapping,
 318                         loff_t pos, unsigned len, unsigned flags,
 319                         struct page **pagep, void **fsdata)
 320 {
 321         return -EINVAL;
 322 }
 323
 324 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 325                         loff_t pos, unsigned len, unsigned copied,
 326                         struct page *page, void *fsdata)
 327 {
 328         BUG();
 329         return -EINVAL;
 330 }
 331
 332 static void truncate_huge_page(struct page *page)
 333 {
 334         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
 335         ClearPageUptodate(page);
 336         remove_from_page_cache(page);
 337         put_page(page);
 338 }
 339
 340 static void truncate_hugepages(struct inode *inode, loff_t lstart)
 341 {
 342         struct hstate *h = hstate_inode(inode);
 343         struct address_space *mapping = &inode->i_data;
 344         const pgoff_t start = lstart >> huge_page_shift(h);
 345         struct pagevec pvec;
 346         pgoff_t next;
 347         int i, freed = 0;
 348
 349         pagevec_init(&pvec, 0);
 350         next = start;
 351         while (1) {
 352                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 353                         if (next == start)
 354                                 break;
 355                         next = start;
 356                         continue;
 357                 }
 358
 359                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 360                         struct page *page = pvec.pages[i];
 361
 362                         lock_page(page);
 363                         if (page->index > next)
 364                                 next = page->index;
 365                         ++next;
 366                         truncate_huge_page(page);
 367                         unlock_page(page);
 368                         freed++;
 369                 }
 370                 huge_pagevec_release(&pvec);
 371         }
 372         BUG_ON(!lstart && mapping->nrpages);
 373         hugetlb_unreserve_pages(inode, start, freed);
 374 }
 375
 376 static void hugetlbfs_delete_inode(struct inode *inode)
 377 {
 378         truncate_hugepages(inode, 0);
 379         clear_inode(inode);
 380 }
 381
 382 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
 383 {
 384         struct super_block *sb = inode->i_sb;
 385
 386         if (!hlist_unhashed(&inode->i_hash)) {
 387                 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
 388                         list_move(&inode->i_list, &inode_unused);
 389                 inodes_stat.nr_unused++;
 390                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
 391                         spin_unlock(&inode_lock);
 392                         return;
 393                 }
 394                 inode->i_state |= I_WILL_FREE;
 395                 spin_unlock(&inode_lock);
 396                 /*
 397                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
 398                  * in our backing_dev_info.
 399                  */
 400                 write_inode_now(inode, 1);
 401                 spin_lock(&inode_lock);
 402                 inode->i_state &= ~I_WILL_FREE;
 403                 inodes_stat.nr_unused--;
 404                 hlist_del_init(&inode->i_hash);
 405         }
 406         list_del_init(&inode->i_list);
 407         list_del_init(&inode->i_sb_list);
 408         inode->i_state |= I_FREEING;
 409         inodes_stat.nr_inodes--;
 410         spin_unlock(&inode_lock);
 411         truncate_hugepages(inode, 0);
 412         clear_inode(inode);
 413         destroy_inode(inode);
 414 }
 415
 416 static void hugetlbfs_drop_inode(struct inode *inode)
 417 {
 418         if (!inode->i_nlink)
 419                 generic_delete_inode(inode);
 420         else
 421                 hugetlbfs_forget_inode(inode);
 422 }
 423
 424 static inline void
 425 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
 426 {
 427         struct vm_area_struct *vma;
 428         struct prio_tree_iter iter;
 429
 430         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
 431                 unsigned long v_offset;
 432
 433                 /*
 434                  * Can the expression below overflow on 32-bit arches?
 435                  * No, because the prio_tree returns us only those vmas
 436                  * which overlap the truncated area starting at pgoff,
 437                  * and no vma on a 32-bit arch can span beyond the 4GB.
 438                  */
 439                 if (vma->vm_pgoff < pgoff)
 440                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 441                 else
 442                         v_offset = 0;
 443
 444                 __unmap_hugepage_range(vma,
 445                                 vma->vm_start + v_offset, vma->vm_end, NULL);
 446         }
 447 }
 448
 449 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 450 {
 451         pgoff_t pgoff;
 452         struct address_space *mapping = inode->i_mapping;
 453         struct hstate *h = hstate_inode(inode);
 454
 455         BUG_ON(offset & ~huge_page_mask(h));
 456         pgoff = offset >> PAGE_SHIFT;
 457
 458         i_size_write(inode, offset);
 459         spin_lock(&mapping->i_mmap_lock);
 460         if (!prio_tree_empty(&mapping->i_mmap))
 461                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 462         spin_unlock(&mapping->i_mmap_lock);
 463         truncate_hugepages(inode, offset);
 464         return 0;
 465 }
 466
 467 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 468 {
 469         struct inode *inode = dentry->d_inode;
 470         struct hstate *h = hstate_inode(inode);
 471         int error;
 472         unsigned int ia_valid = attr->ia_valid;
 473
 474         BUG_ON(!inode);
 475
 476         error = inode_change_ok(inode, attr);
 477         if (error)
 478                 goto out;
 479
 480         if (ia_valid & ATTR_SIZE) {
 481                 error = -EINVAL;
 482                 if (!(attr->ia_size & ~huge_page_mask(h)))
 483                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 484                 if (error)
 485                         goto out;
 486                 attr->ia_valid &= ~ATTR_SIZE;
 487         }
 488         error = inode_setattr(inode, attr);
 489 out:
 490         return error;
 491 }
 492
 493 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 494                                         gid_t gid, int mode, dev_t dev)
 495 {
 496         struct inode *inode;
 497
 498         inode = new_inode(sb);
 499         if (inode) {
 500                 struct hugetlbfs_inode_info *info;
 501                 inode->i_mode = mode;
 502                 inode->i_uid = uid;
 503                 inode->i_gid = gid;
 504                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 505                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 506                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 507                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
 508                 info = HUGETLBFS_I(inode);
 509                 mpol_shared_policy_init(&info->policy, NULL);
 510                 switch (mode & S_IFMT) {
 511                 default:
 512                         init_special_inode(inode, mode, dev);
 513                         break;
 514                 case S_IFREG:
 515                         inode->i_op = &hugetlbfs_inode_operations;
 516                         inode->i_fop = &hugetlbfs_file_operations;
 517                         break;
 518                 case S_IFDIR:
 519                         inode->i_op = &hugetlbfs_dir_inode_operations;
 520                         inode->i_fop = &simple_dir_operations;
 521
 522                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 523                         inc_nlink(inode);
 524                         break;
 525                 case S_IFLNK:
 526                         inode->i_op = &page_symlink_inode_operations;
 527                         break;
 528                 }
 529         }
 530         return inode;
 531 }
 532
 533 /*
 534  * File creation. Allocate an inode, and we're done..
 535  */
 536 static int hugetlbfs_mknod(struct inode *dir,
 537                         struct dentry *dentry, int mode, dev_t dev)
 538 {
 539         struct inode *inode;
 540         int error = -ENOSPC;
 541         gid_t gid;
 542
 543         if (dir->i_mode & S_ISGID) {
 544                 gid = dir->i_gid;
 545                 if (S_ISDIR(mode))
 546                         mode |= S_ISGID;
 547         } else {
 548                 gid = current_fsgid();
 549         }
 550         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
 551         if (inode) {
 552                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 553                 d_instantiate(dentry, inode);
 554                 dget(dentry);   /* Extra count - pin the dentry in core */
 555                 error = 0;
 556         }
 557         return error;
 558 }
 559
 560 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 561 {
 562         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 563         if (!retval)
 564                 inc_nlink(dir);
 565         return retval;
 566 }
 567
 568 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 569 {
 570         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 571 }
 572
 573 static int hugetlbfs_symlink(struct inode *dir,
 574                         struct dentry *dentry, const char *symname)
 575 {
 576         struct inode *inode;
 577         int error = -ENOSPC;
 578         gid_t gid;
 579
 580         if (dir->i_mode & S_ISGID)
 581                 gid = dir->i_gid;
 582         else
 583                 gid = current_fsgid();
 584
 585         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
 586                                         gid, S_IFLNK|S_IRWXUGO, 0);
 587         if (inode) {
 588                 int l = strlen(symname)+1;
 589                 error = page_symlink(inode, symname, l);
 590                 if (!error) {
 591                         d_instantiate(dentry, inode);
 592                         dget(dentry);
 593                 } else
 594                         iput(inode);
 595         }
 596         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 597
 598         return error;
 599 }
 600
 601 /*
 602  * mark the head page dirty
 603  */
 604 static int hugetlbfs_set_page_dirty(struct page *page)
 605 {
 606         struct page *head = compound_head(page);
 607
 608         SetPageDirty(head);
 609         return 0;
 610 }
 611
 612 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 613 {
 614         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 615         struct hstate *h = hstate_inode(dentry->d_inode);
 616
 617         buf->f_type = HUGETLBFS_MAGIC;
 618         buf->f_bsize = huge_page_size(h);
 619         if (sbinfo) {
 620                 spin_lock(&sbinfo->stat_lock);
 621                 /* If no limits set, just report 0 for max/free/used
 622                  * blocks, like simple_statfs() */
 623                 if (sbinfo->max_blocks >= 0) {
 624                         buf->f_blocks = sbinfo->max_blocks;
 625                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 626                         buf->f_files = sbinfo->max_inodes;
 627                         buf->f_ffree = sbinfo->free_inodes;
 628                 }
 629                 spin_unlock(&sbinfo->stat_lock);
 630         }
 631         buf->f_namelen = NAME_MAX;
 632         return 0;
 633 }
 634
 635 static void hugetlbfs_put_super(struct super_block *sb)
 636 {
 637         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 638
 639         if (sbi) {
 640                 sb->s_fs_info = NULL;
 641                 kfree(sbi);
 642         }
 643 }
 644
 645 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 646 {
 647         if (sbinfo->free_inodes >= 0) {
 648                 spin_lock(&sbinfo->stat_lock);
 649                 if (unlikely(!sbinfo->free_inodes)) {
 650                         spin_unlock(&sbinfo->stat_lock);
 651                         return 0;
 652                 }
 653                 sbinfo->free_inodes--;
 654                 spin_unlock(&sbinfo->stat_lock);
 655         }
 656
 657         return 1;
 658 }
 659
 660 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 661 {
 662         if (sbinfo->free_inodes >= 0) {
 663                 spin_lock(&sbinfo->stat_lock);
 664                 sbinfo->free_inodes++;
 665                 spin_unlock(&sbinfo->stat_lock);
 666         }
 667 }
 668
 669
 670 static struct kmem_cache *hugetlbfs_inode_cachep;
 671
 672 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 673 {
 674         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 675         struct hugetlbfs_inode_info *p;
 676
 677         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 678                 return NULL;
 679         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 680         if (unlikely(!p)) {
 681                 hugetlbfs_inc_free_inodes(sbinfo);
 682                 return NULL;
 683         }
 684         return &p->vfs_inode;
 685 }
 686
 687 static void hugetlbfs_destroy_inode(struct inode *inode)
 688 {
 689         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 690         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 691         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 692 }
 693
 694 static const struct address_space_operations hugetlbfs_aops = {
 695         .write_begin    = hugetlbfs_write_begin,
 696         .write_end      = hugetlbfs_write_end,
 697         .set_page_dirty = hugetlbfs_set_page_dirty,
 698 };
 699
 700
 701 static void init_once(void *foo)
 702 {
 703         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 704
 705         inode_init_once(&ei->vfs_inode);
 706 }
 707
 708 const struct file_operations hugetlbfs_file_operations = {
 709         .read                   = hugetlbfs_read,
 710         .mmap                   = hugetlbfs_file_mmap,
 711         .fsync                  = simple_sync_file,
 712         .get_unmapped_area      = hugetlb_get_unmapped_area,
 713 };
 714
 715 static const struct inode_operations hugetlbfs_dir_inode_operations = {
 716         .create         = hugetlbfs_create,
 717         .lookup         = simple_lookup,
 718         .link           = simple_link,
 719         .unlink         = simple_unlink,
 720         .symlink        = hugetlbfs_symlink,
 721         .mkdir          = hugetlbfs_mkdir,
 722         .rmdir          = simple_rmdir,
 723         .mknod          = hugetlbfs_mknod,
 724         .rename         = simple_rename,
 725         .setattr        = hugetlbfs_setattr,
 726 };
 727
 728 static const struct inode_operations hugetlbfs_inode_operations = {
 729         .setattr        = hugetlbfs_setattr,
 730 };
 731
 732 static const struct super_operations hugetlbfs_ops = {
 733         .alloc_inode    = hugetlbfs_alloc_inode,
 734         .destroy_inode  = hugetlbfs_destroy_inode,
 735         .statfs         = hugetlbfs_statfs,
 736         .delete_inode   = hugetlbfs_delete_inode,
 737         .drop_inode     = hugetlbfs_drop_inode,
 738         .put_super      = hugetlbfs_put_super,
 739         .show_options   = generic_show_options,
 740 };
 741
 742 static int
 743 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 744 {
 745         char *p, *rest;
 746         substring_t args[MAX_OPT_ARGS];
 747         int option;
 748         unsigned long long size = 0;
 749         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
 750
 751         if (!options)
 752                 return 0;
 753
 754         while ((p = strsep(&options, ",")) != NULL) {
 755                 int token;
 756                 if (!*p)
 757                         continue;
 758
 759                 token = match_token(p, tokens, args);
 760                 switch (token) {
 761                 case Opt_uid:
 762                         if (match_int(&args[0], &option))
 763                                 goto bad_val;
 764                         pconfig->uid = option;
 765                         break;
 766
 767                 case Opt_gid:
 768                         if (match_int(&args[0], &option))
 769                                 goto bad_val;
 770                         pconfig->gid = option;
 771                         break;
 772
 773                 case Opt_mode:
 774                         if (match_octal(&args[0], &option))
 775                                 goto bad_val;
 776                         pconfig->mode = option & 01777U;
 777                         break;
 778
 779                 case Opt_size: {
 780                         /* memparse() will accept a K/M/G without a digit */
 781                         if (!isdigit(*args[0].from))
 782                                 goto bad_val;
 783                         size = memparse(args[0].from, &rest);
 784                         setsize = SIZE_STD;
 785                         if (*rest == '%')
 786                                 setsize = SIZE_PERCENT;
 787                         break;
 788                 }
 789
 790                 case Opt_nr_inodes:
 791                         /* memparse() will accept a K/M/G without a digit */
 792                         if (!isdigit(*args[0].from))
 793                                 goto bad_val;
 794                         pconfig->nr_inodes = memparse(args[0].from, &rest);
 795                         break;
 796
 797                 case Opt_pagesize: {
 798                         unsigned long ps;
 799                         ps = memparse(args[0].from, &rest);
 800                         pconfig->hstate = size_to_hstate(ps);
 801                         if (!pconfig->hstate) {
 802                                 printk(KERN_ERR
 803                                 "hugetlbfs: Unsupported page size %lu MB\n",
 804                                         ps >> 20);
 805                                 return -EINVAL;
 806                         }
 807                         break;
 808                 }
 809
 810                 default:
 811                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
 812                                  p);
 813                         return -EINVAL;
 814                         break;
 815                 }
 816         }
 817
 818         /* Do size after hstate is set up */
 819         if (setsize > NO_SIZE) {
 820                 struct hstate *h = pconfig->hstate;
 821                 if (setsize == SIZE_PERCENT) {
 822                         size <<= huge_page_shift(h);
 823                         size *= h->max_huge_pages;
 824                         do_div(size, 100);
 825                 }
 826                 pconfig->nr_blocks = (size >> huge_page_shift(h));
 827         }
 828
 829         return 0;
 830
 831 bad_val:
 832         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
 833                args[0].from, p);
 834         return -EINVAL;
 835 }
 836
 837 static int
 838 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 839 {
 840         struct inode * inode;
 841         struct dentry * root;
 842         int ret;
 843         struct hugetlbfs_config config;
 844         struct hugetlbfs_sb_info *sbinfo;
 845
 846         save_mount_options(sb, data);
 847
 848         config.nr_blocks = -1; /* No limit on size by default */
 849         config.nr_inodes = -1; /* No limit on number of inodes by default */
 850         config.uid = current_fsuid();
 851         config.gid = current_fsgid();
 852         config.mode = 0755;
 853         config.hstate = &default_hstate;
 854         ret = hugetlbfs_parse_options(data, &config);
 855         if (ret)
 856                 return ret;
 857
 858         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 859         if (!sbinfo)
 860                 return -ENOMEM;
 861         sb->s_fs_info = sbinfo;
 862         sbinfo->hstate = config.hstate;
 863         spin_lock_init(&sbinfo->stat_lock);
 864         sbinfo->max_blocks = config.nr_blocks;
 865         sbinfo->free_blocks = config.nr_blocks;
 866         sbinfo->max_inodes = config.nr_inodes;
 867         sbinfo->free_inodes = config.nr_inodes;
 868         sb->s_maxbytes = MAX_LFS_FILESIZE;
 869         sb->s_blocksize = huge_page_size(config.hstate);
 870         sb->s_blocksize_bits = huge_page_shift(config.hstate);
 871         sb->s_magic = HUGETLBFS_MAGIC;
 872         sb->s_op = &hugetlbfs_ops;
 873         sb->s_time_gran = 1;
 874         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 875                                         S_IFDIR | config.mode, 0);
 876         if (!inode)
 877                 goto out_free;
 878
 879         root = d_alloc_root(inode);
 880         if (!root) {
 881                 iput(inode);
 882                 goto out_free;
 883         }
 884         sb->s_root = root;
 885         return 0;
 886 out_free:
 887         kfree(sbinfo);
 888         return -ENOMEM;
 889 }
 890
 891 int hugetlb_get_quota(struct address_space *mapping, long delta)
 892 {
 893         int ret = 0;
 894         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 895
 896         if (sbinfo->free_blocks > -1) {
 897                 spin_lock(&sbinfo->stat_lock);
 898                 if (sbinfo->free_blocks - delta >= 0)
 899                         sbinfo->free_blocks -= delta;
 900                 else
 901                         ret = -ENOMEM;
 902                 spin_unlock(&sbinfo->stat_lock);
 903         }
 904
 905         return ret;
 906 }
 907
 908 void hugetlb_put_quota(struct address_space *mapping, long delta)
 909 {
 910         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 911
 912         if (sbinfo->free_blocks > -1) {
 913                 spin_lock(&sbinfo->stat_lock);
 914                 sbinfo->free_blocks += delta;
 915                 spin_unlock(&sbinfo->stat_lock);
 916         }
 917 }
 918
 919 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
 920         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 921 {
 922         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
 923 }
 924
 925 static struct file_system_type hugetlbfs_fs_type = {
 926         .name           = "hugetlbfs",
 927         .get_sb         = hugetlbfs_get_sb,
 928         .kill_sb        = kill_litter_super,
 929 };
 930
 931 static struct vfsmount *hugetlbfs_vfsmount;
 932
 933 static int can_do_hugetlb_shm(void)
 934 {
 935         return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
 936 }
 937
 938 struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag,
 939                                                 struct user_struct **user)
 940 {
 941         int error = -ENOMEM;
 942         struct file *file;
 943         struct inode *inode;
 944         struct dentry *dentry, *root;
 945         struct qstr quick_string;
 946
 947         *user = NULL;
 948         if (!hugetlbfs_vfsmount)
 949                 return ERR_PTR(-ENOENT);
 950
 951         if (!can_do_hugetlb_shm()) {
 952                 *user = current_user();
 953                 if (user_shm_lock(size, *user)) {
 954                         WARN_ONCE(1,
 955                           "Using mlock ulimits for SHM_HUGETLB deprecated\n");
 956                 } else {
 957                         *user = NULL;
 958                         return ERR_PTR(-EPERM);
 959                 }
 960         }
 961
 962         root = hugetlbfs_vfsmount->mnt_root;
 963         quick_string.name = name;
 964         quick_string.len = strlen(quick_string.name);
 965         quick_string.hash = 0;
 966         dentry = d_alloc(root, &quick_string);
 967         if (!dentry)
 968                 goto out_shm_unlock;
 969
 970         error = -ENOSPC;
 971         inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
 972                                 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
 973         if (!inode)
 974                 goto out_dentry;
 975
 976         error = -ENOMEM;
 977         if (hugetlb_reserve_pages(inode, 0,
 978                         size >> huge_page_shift(hstate_inode(inode)), NULL,
 979                         acctflag))
 980                 goto out_inode;
 981
 982         d_instantiate(dentry, inode);
 983         inode->i_size = size;
 984         inode->i_nlink = 0;
 985
 986         error = -ENFILE;
 987         file = alloc_file(hugetlbfs_vfsmount, dentry,
 988                         FMODE_WRITE | FMODE_READ,
 989                         &hugetlbfs_file_operations);
 990         if (!file)
 991                 goto out_dentry; /* inode is already attached */
 992         ima_counts_get(file);
 993
 994         return file;
 995
 996 out_inode:
 997         iput(inode);
 998 out_dentry:
 999         dput(dentry);
1000 out_shm_unlock:
1001         if (*user) {
1002                 user_shm_unlock(size, *user);
1003                 *user = NULL;
1004         }
1005         return ERR_PTR(error);
1006 }
1007
1008 static int __init init_hugetlbfs_fs(void)
1009 {
1010         int error;
1011         struct vfsmount *vfsmount;
1012
1013         error = bdi_init(&hugetlbfs_backing_dev_info);
1014         if (error)
1015                 return error;
1016
1017         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1018                                         sizeof(struct hugetlbfs_inode_info),
1019                                         0, 0, init_once);
1020         if (hugetlbfs_inode_cachep == NULL)
1021                 goto out2;
1022
1023         error = register_filesystem(&hugetlbfs_fs_type);
1024         if (error)
1025                 goto out;
1026
1027         vfsmount = kern_mount(&hugetlbfs_fs_type);
1028
1029         if (!IS_ERR(vfsmount)) {
1030                 hugetlbfs_vfsmount = vfsmount;
1031                 return 0;
1032         }
1033
1034         error = PTR_ERR(vfsmount);
1035
1036  out:
1037         if (error)
1038                 kmem_cache_destroy(hugetlbfs_inode_cachep);
1039  out2:
1040         bdi_destroy(&hugetlbfs_backing_dev_info);
1041         return error;
1042 }
1043
1044 static void __exit exit_hugetlbfs_fs(void)
1045 {
1046         kmem_cache_destroy(hugetlbfs_inode_cachep);
1047         unregister_filesystem(&hugetlbfs_fs_type);
1048         bdi_destroy(&hugetlbfs_backing_dev_info);
1049 }
1050
1051 module_init(init_hugetlbfs_fs)
1052 module_exit(exit_hugetlbfs_fs)
1053
1054 MODULE_LICENSE("GPL");