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