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