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/quotaops.h>
  30 #include <linux/slab.h>
  31 #include <linux/dnotify.h>
  32 #include <linux/statfs.h>
  33 #include <linux/security.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                 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 /*
 316  * Read a page. Again trivial. If it didn't already exist
 317  * in the page cache, it is zero-filled.
 318  */
 319 static int hugetlbfs_readpage(struct file *file, struct page * page)
 320 {
 321         unlock_page(page);
 322         return -EINVAL;
 323 }
 324
 325 static int hugetlbfs_write_begin(struct file *file,
 326                         struct address_space *mapping,
 327                         loff_t pos, unsigned len, unsigned flags,
 328                         struct page **pagep, void **fsdata)
 329 {
 330         return -EINVAL;
 331 }
 332
 333 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 334                         loff_t pos, unsigned len, unsigned copied,
 335                         struct page *page, void *fsdata)
 336 {
 337         BUG();
 338         return -EINVAL;
 339 }
 340
 341 static void truncate_huge_page(struct page *page)
 342 {
 343         cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
 344         ClearPageUptodate(page);
 345         remove_from_page_cache(page);
 346         put_page(page);
 347 }
 348
 349 static void truncate_hugepages(struct inode *inode, loff_t lstart)
 350 {
 351         struct hstate *h = hstate_inode(inode);
 352         struct address_space *mapping = &inode->i_data;
 353         const pgoff_t start = lstart >> huge_page_shift(h);
 354         struct pagevec pvec;
 355         pgoff_t next;
 356         int i, freed = 0;
 357
 358         pagevec_init(&pvec, 0);
 359         next = start;
 360         while (1) {
 361                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 362                         if (next == start)
 363                                 break;
 364                         next = start;
 365                         continue;
 366                 }
 367
 368                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 369                         struct page *page = pvec.pages[i];
 370
 371                         lock_page(page);
 372                         if (page->index > next)
 373                                 next = page->index;
 374                         ++next;
 375                         truncate_huge_page(page);
 376                         unlock_page(page);
 377                         freed++;
 378                 }
 379                 huge_pagevec_release(&pvec);
 380         }
 381         BUG_ON(!lstart && mapping->nrpages);
 382         hugetlb_unreserve_pages(inode, start, freed);
 383 }
 384
 385 static void hugetlbfs_delete_inode(struct inode *inode)
 386 {
 387         truncate_hugepages(inode, 0);
 388         clear_inode(inode);
 389 }
 390
 391 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
 392 {
 393         struct super_block *sb = inode->i_sb;
 394
 395         if (!hlist_unhashed(&inode->i_hash)) {
 396                 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
 397                         list_move(&inode->i_list, &inode_unused);
 398                 inodes_stat.nr_unused++;
 399                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
 400                         spin_unlock(&inode_lock);
 401                         return;
 402                 }
 403                 inode->i_state |= I_WILL_FREE;
 404                 spin_unlock(&inode_lock);
 405                 /*
 406                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
 407                  * in our backing_dev_info.
 408                  */
 409                 write_inode_now(inode, 1);
 410                 spin_lock(&inode_lock);
 411                 inode->i_state &= ~I_WILL_FREE;
 412                 inodes_stat.nr_unused--;
 413                 hlist_del_init(&inode->i_hash);
 414         }
 415         list_del_init(&inode->i_list);
 416         list_del_init(&inode->i_sb_list);
 417         inode->i_state |= I_FREEING;
 418         inodes_stat.nr_inodes--;
 419         spin_unlock(&inode_lock);
 420         truncate_hugepages(inode, 0);
 421         clear_inode(inode);
 422         destroy_inode(inode);
 423 }
 424
 425 static void hugetlbfs_drop_inode(struct inode *inode)
 426 {
 427         if (!inode->i_nlink)
 428                 generic_delete_inode(inode);
 429         else
 430                 hugetlbfs_forget_inode(inode);
 431 }
 432
 433 static inline void
 434 hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
 435 {
 436         struct vm_area_struct *vma;
 437         struct prio_tree_iter iter;
 438
 439         vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
 440                 unsigned long v_offset;
 441
 442                 /*
 443                  * Can the expression below overflow on 32-bit arches?
 444                  * No, because the prio_tree returns us only those vmas
 445                  * which overlap the truncated area starting at pgoff,
 446                  * and no vma on a 32-bit arch can span beyond the 4GB.
 447                  */
 448                 if (vma->vm_pgoff < pgoff)
 449                         v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 450                 else
 451                         v_offset = 0;
 452
 453                 __unmap_hugepage_range(vma,
 454                                 vma->vm_start + v_offset, vma->vm_end, NULL);
 455         }
 456 }
 457
 458 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 459 {
 460         pgoff_t pgoff;
 461         struct address_space *mapping = inode->i_mapping;
 462         struct hstate *h = hstate_inode(inode);
 463
 464         BUG_ON(offset & ~huge_page_mask(h));
 465         pgoff = offset >> PAGE_SHIFT;
 466
 467         i_size_write(inode, offset);
 468         spin_lock(&mapping->i_mmap_lock);
 469         if (!prio_tree_empty(&mapping->i_mmap))
 470                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 471         spin_unlock(&mapping->i_mmap_lock);
 472         truncate_hugepages(inode, offset);
 473         return 0;
 474 }
 475
 476 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 477 {
 478         struct inode *inode = dentry->d_inode;
 479         struct hstate *h = hstate_inode(inode);
 480         int error;
 481         unsigned int ia_valid = attr->ia_valid;
 482
 483         BUG_ON(!inode);
 484
 485         error = inode_change_ok(inode, attr);
 486         if (error)
 487                 goto out;
 488
 489         if (ia_valid & ATTR_SIZE) {
 490                 error = -EINVAL;
 491                 if (!(attr->ia_size & ~huge_page_mask(h)))
 492                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 493                 if (error)
 494                         goto out;
 495                 attr->ia_valid &= ~ATTR_SIZE;
 496         }
 497         error = inode_setattr(inode, attr);
 498 out:
 499         return error;
 500 }
 501
 502 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 503                                         gid_t gid, int mode, dev_t dev)
 504 {
 505         struct inode *inode;
 506
 507         inode = new_inode(sb);
 508         if (inode) {
 509                 struct hugetlbfs_inode_info *info;
 510                 inode->i_mode = mode;
 511                 inode->i_uid = uid;
 512                 inode->i_gid = gid;
 513                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 514                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 515                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 516                 INIT_LIST_HEAD(&inode->i_mapping->private_list);
 517                 info = HUGETLBFS_I(inode);
 518                 mpol_shared_policy_init(&info->policy, NULL);
 519                 switch (mode & S_IFMT) {
 520                 default:
 521                         init_special_inode(inode, mode, dev);
 522                         break;
 523                 case S_IFREG:
 524                         inode->i_op = &hugetlbfs_inode_operations;
 525                         inode->i_fop = &hugetlbfs_file_operations;
 526                         break;
 527                 case S_IFDIR:
 528                         inode->i_op = &hugetlbfs_dir_inode_operations;
 529                         inode->i_fop = &simple_dir_operations;
 530
 531                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 532                         inc_nlink(inode);
 533                         break;
 534                 case S_IFLNK:
 535                         inode->i_op = &page_symlink_inode_operations;
 536                         break;
 537                 }
 538         }
 539         return inode;
 540 }
 541
 542 /*
 543  * File creation. Allocate an inode, and we're done..
 544  */
 545 static int hugetlbfs_mknod(struct inode *dir,
 546                         struct dentry *dentry, int mode, dev_t dev)
 547 {
 548         struct inode *inode;
 549         int error = -ENOSPC;
 550         gid_t gid;
 551
 552         if (dir->i_mode & S_ISGID) {
 553                 gid = dir->i_gid;
 554                 if (S_ISDIR(mode))
 555                         mode |= S_ISGID;
 556         } else {
 557                 gid = current_fsgid();
 558         }
 559         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
 560         if (inode) {
 561                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 562                 d_instantiate(dentry, inode);
 563                 dget(dentry);   /* Extra count - pin the dentry in core */
 564                 error = 0;
 565         }
 566         return error;
 567 }
 568
 569 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 570 {
 571         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 572         if (!retval)
 573                 inc_nlink(dir);
 574         return retval;
 575 }
 576
 577 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 578 {
 579         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 580 }
 581
 582 static int hugetlbfs_symlink(struct inode *dir,
 583                         struct dentry *dentry, const char *symname)
 584 {
 585         struct inode *inode;
 586         int error = -ENOSPC;
 587         gid_t gid;
 588
 589         if (dir->i_mode & S_ISGID)
 590                 gid = dir->i_gid;
 591         else
 592                 gid = current_fsgid();
 593
 594         inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
 595                                         gid, S_IFLNK|S_IRWXUGO, 0);
 596         if (inode) {
 597                 int l = strlen(symname)+1;
 598                 error = page_symlink(inode, symname, l);
 599                 if (!error) {
 600                         d_instantiate(dentry, inode);
 601                         dget(dentry);
 602                 } else
 603                         iput(inode);
 604         }
 605         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 606
 607         return error;
 608 }
 609
 610 /*
 611  * mark the head page dirty
 612  */
 613 static int hugetlbfs_set_page_dirty(struct page *page)
 614 {
 615         struct page *head = compound_head(page);
 616
 617         SetPageDirty(head);
 618         return 0;
 619 }
 620
 621 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 622 {
 623         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 624         struct hstate *h = hstate_inode(dentry->d_inode);
 625
 626         buf->f_type = HUGETLBFS_MAGIC;
 627         buf->f_bsize = huge_page_size(h);
 628         if (sbinfo) {
 629                 spin_lock(&sbinfo->stat_lock);
 630                 /* If no limits set, just report 0 for max/free/used
 631                  * blocks, like simple_statfs() */
 632                 if (sbinfo->max_blocks >= 0) {
 633                         buf->f_blocks = sbinfo->max_blocks;
 634                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 635                         buf->f_files = sbinfo->max_inodes;
 636                         buf->f_ffree = sbinfo->free_inodes;
 637                 }
 638                 spin_unlock(&sbinfo->stat_lock);
 639         }
 640         buf->f_namelen = NAME_MAX;
 641         return 0;
 642 }
 643
 644 static void hugetlbfs_put_super(struct super_block *sb)
 645 {
 646         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 647
 648         if (sbi) {
 649                 sb->s_fs_info = NULL;
 650                 kfree(sbi);
 651         }
 652 }
 653
 654 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 655 {
 656         if (sbinfo->free_inodes >= 0) {
 657                 spin_lock(&sbinfo->stat_lock);
 658                 if (unlikely(!sbinfo->free_inodes)) {
 659                         spin_unlock(&sbinfo->stat_lock);
 660                         return 0;
 661                 }
 662                 sbinfo->free_inodes--;
 663                 spin_unlock(&sbinfo->stat_lock);
 664         }
 665
 666         return 1;
 667 }
 668
 669 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 670 {
 671         if (sbinfo->free_inodes >= 0) {
 672                 spin_lock(&sbinfo->stat_lock);
 673                 sbinfo->free_inodes++;
 674                 spin_unlock(&sbinfo->stat_lock);
 675         }
 676 }
 677
 678
 679 static struct kmem_cache *hugetlbfs_inode_cachep;
 680
 681 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 682 {
 683         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 684         struct hugetlbfs_inode_info *p;
 685
 686         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 687                 return NULL;
 688         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 689         if (unlikely(!p)) {
 690                 hugetlbfs_inc_free_inodes(sbinfo);
 691                 return NULL;
 692         }
 693         return &p->vfs_inode;
 694 }
 695
 696 static void hugetlbfs_destroy_inode(struct inode *inode)
 697 {
 698         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 699         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 700         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 701 }
 702
 703 static const struct address_space_operations hugetlbfs_aops = {
 704         .readpage       = hugetlbfs_readpage,
 705         .write_begin    = hugetlbfs_write_begin,
 706         .write_end      = hugetlbfs_write_end,
 707         .set_page_dirty = hugetlbfs_set_page_dirty,
 708 };
 709
 710
 711 static void init_once(void *foo)
 712 {
 713         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 714
 715         inode_init_once(&ei->vfs_inode);
 716 }
 717
 718 const struct file_operations hugetlbfs_file_operations = {
 719         .read                   = hugetlbfs_read,
 720         .mmap                   = hugetlbfs_file_mmap,
 721         .fsync                  = simple_sync_file,
 722         .get_unmapped_area      = hugetlb_get_unmapped_area,
 723 };
 724
 725 static const struct inode_operations hugetlbfs_dir_inode_operations = {
 726         .create         = hugetlbfs_create,
 727         .lookup         = simple_lookup,
 728         .link           = simple_link,
 729         .unlink         = simple_unlink,
 730         .symlink        = hugetlbfs_symlink,
 731         .mkdir          = hugetlbfs_mkdir,
 732         .rmdir          = simple_rmdir,
 733         .mknod          = hugetlbfs_mknod,
 734         .rename         = simple_rename,
 735         .setattr        = hugetlbfs_setattr,
 736 };
 737
 738 static const struct inode_operations hugetlbfs_inode_operations = {
 739         .setattr        = hugetlbfs_setattr,
 740 };
 741
 742 static const struct super_operations hugetlbfs_ops = {
 743         .alloc_inode    = hugetlbfs_alloc_inode,
 744         .destroy_inode  = hugetlbfs_destroy_inode,
 745         .statfs         = hugetlbfs_statfs,
 746         .delete_inode   = hugetlbfs_delete_inode,
 747         .drop_inode     = hugetlbfs_drop_inode,
 748         .put_super      = hugetlbfs_put_super,
 749         .show_options   = generic_show_options,
 750 };
 751
 752 static int
 753 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 754 {
 755         char *p, *rest;
 756         substring_t args[MAX_OPT_ARGS];
 757         int option;
 758         unsigned long long size = 0;
 759         enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
 760
 761         if (!options)
 762                 return 0;
 763
 764         while ((p = strsep(&options, ",")) != NULL) {
 765                 int token;
 766                 if (!*p)
 767                         continue;
 768
 769                 token = match_token(p, tokens, args);
 770                 switch (token) {
 771                 case Opt_uid:
 772                         if (match_int(&args[0], &option))
 773                                 goto bad_val;
 774                         pconfig->uid = option;
 775                         break;
 776
 777                 case Opt_gid:
 778                         if (match_int(&args[0], &option))
 779                                 goto bad_val;
 780                         pconfig->gid = option;
 781                         break;
 782
 783                 case Opt_mode:
 784                         if (match_octal(&args[0], &option))
 785                                 goto bad_val;
 786                         pconfig->mode = option & 01777U;
 787                         break;
 788
 789                 case Opt_size: {
 790                         /* memparse() will accept a K/M/G without a digit */
 791                         if (!isdigit(*args[0].from))
 792                                 goto bad_val;
 793                         size = memparse(args[0].from, &rest);
 794                         setsize = SIZE_STD;
 795                         if (*rest == '%')
 796                                 setsize = SIZE_PERCENT;
 797                         break;
 798                 }
 799
 800                 case Opt_nr_inodes:
 801                         /* memparse() will accept a K/M/G without a digit */
 802                         if (!isdigit(*args[0].from))
 803                                 goto bad_val;
 804                         pconfig->nr_inodes = memparse(args[0].from, &rest);
 805                         break;
 806
 807                 case Opt_pagesize: {
 808                         unsigned long ps;
 809                         ps = memparse(args[0].from, &rest);
 810                         pconfig->hstate = size_to_hstate(ps);
 811                         if (!pconfig->hstate) {
 812                                 printk(KERN_ERR
 813                                 "hugetlbfs: Unsupported page size %lu MB\n",
 814                                         ps >> 20);
 815                                 return -EINVAL;
 816                         }
 817                         break;
 818                 }
 819
 820                 default:
 821                         printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
 822                                  p);
 823                         return -EINVAL;
 824                         break;
 825                 }
 826         }
 827
 828         /* Do size after hstate is set up */
 829         if (setsize > NO_SIZE) {
 830                 struct hstate *h = pconfig->hstate;
 831                 if (setsize == SIZE_PERCENT) {
 832                         size <<= huge_page_shift(h);
 833                         size *= h->max_huge_pages;
 834                         do_div(size, 100);
 835                 }
 836                 pconfig->nr_blocks = (size >> huge_page_shift(h));
 837         }
 838
 839         return 0;
 840
 841 bad_val:
 842         printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
 843                args[0].from, p);
 844         return 1;
 845 }
 846
 847 static int
 848 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 849 {
 850         struct inode * inode;
 851         struct dentry * root;
 852         int ret;
 853         struct hugetlbfs_config config;
 854         struct hugetlbfs_sb_info *sbinfo;
 855
 856         save_mount_options(sb, data);
 857
 858         config.nr_blocks = -1; /* No limit on size by default */
 859         config.nr_inodes = -1; /* No limit on number of inodes by default */
 860         config.uid = current_fsuid();
 861         config.gid = current_fsgid();
 862         config.mode = 0755;
 863         config.hstate = &default_hstate;
 864         ret = hugetlbfs_parse_options(data, &config);
 865         if (ret)
 866                 return ret;
 867
 868         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 869         if (!sbinfo)
 870                 return -ENOMEM;
 871         sb->s_fs_info = sbinfo;
 872         sbinfo->hstate = config.hstate;
 873         spin_lock_init(&sbinfo->stat_lock);
 874         sbinfo->max_blocks = config.nr_blocks;
 875         sbinfo->free_blocks = config.nr_blocks;
 876         sbinfo->max_inodes = config.nr_inodes;
 877         sbinfo->free_inodes = config.nr_inodes;
 878         sb->s_maxbytes = MAX_LFS_FILESIZE;
 879         sb->s_blocksize = huge_page_size(config.hstate);
 880         sb->s_blocksize_bits = huge_page_shift(config.hstate);
 881         sb->s_magic = HUGETLBFS_MAGIC;
 882         sb->s_op = &hugetlbfs_ops;
 883         sb->s_time_gran = 1;
 884         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 885                                         S_IFDIR | config.mode, 0);
 886         if (!inode)
 887                 goto out_free;
 888
 889         root = d_alloc_root(inode);
 890         if (!root) {
 891                 iput(inode);
 892                 goto out_free;
 893         }
 894         sb->s_root = root;
 895         return 0;
 896 out_free:
 897         kfree(sbinfo);
 898         return -ENOMEM;
 899 }
 900
 901 int hugetlb_get_quota(struct address_space *mapping, long delta)
 902 {
 903         int ret = 0;
 904         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 905
 906         if (sbinfo->free_blocks > -1) {
 907                 spin_lock(&sbinfo->stat_lock);
 908                 if (sbinfo->free_blocks - delta >= 0)
 909                         sbinfo->free_blocks -= delta;
 910                 else
 911                         ret = -ENOMEM;
 912                 spin_unlock(&sbinfo->stat_lock);
 913         }
 914
 915         return ret;
 916 }
 917
 918 void hugetlb_put_quota(struct address_space *mapping, long delta)
 919 {
 920         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 921
 922         if (sbinfo->free_blocks > -1) {
 923                 spin_lock(&sbinfo->stat_lock);
 924                 sbinfo->free_blocks += delta;
 925                 spin_unlock(&sbinfo->stat_lock);
 926         }
 927 }
 928
 929 static int hugetlbfs_get_sb(struct file_system_type *fs_type,
 930         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 931 {
 932         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
 933 }
 934
 935 static struct file_system_type hugetlbfs_fs_type = {
 936         .name           = "hugetlbfs",
 937         .get_sb         = hugetlbfs_get_sb,
 938         .kill_sb        = kill_litter_super,
 939 };
 940
 941 static struct vfsmount *hugetlbfs_vfsmount;
 942
 943 static int can_do_hugetlb_shm(void)
 944 {
 945         return likely(capable(CAP_IPC_LOCK) ||
 946                         in_group_p(sysctl_hugetlb_shm_group) ||
 947                         can_do_mlock());
 948 }
 949
 950 struct file *hugetlb_file_setup(const char *name, size_t size)
 951 {
 952         int error = -ENOMEM;
 953         struct file *file;
 954         struct inode *inode;
 955         struct dentry *dentry, *root;
 956         struct qstr quick_string;
 957         struct user_struct *user = current_user();
 958
 959         if (!hugetlbfs_vfsmount)
 960                 return ERR_PTR(-ENOENT);
 961
 962         if (!can_do_hugetlb_shm())
 963                 return ERR_PTR(-EPERM);
 964
 965         if (!user_shm_lock(size, user))
 966                 return ERR_PTR(-ENOMEM);
 967
 968         root = hugetlbfs_vfsmount->mnt_root;
 969         quick_string.name = name;
 970         quick_string.len = strlen(quick_string.name);
 971         quick_string.hash = 0;
 972         dentry = d_alloc(root, &quick_string);
 973         if (!dentry)
 974                 goto out_shm_unlock;
 975
 976         error = -ENOSPC;
 977         inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
 978                                 current_fsgid(), S_IFREG | S_IRWXUGO, 0);
 979         if (!inode)
 980                 goto out_dentry;
 981
 982         error = -ENOMEM;
 983         if (hugetlb_reserve_pages(inode, 0,
 984                         size >> huge_page_shift(hstate_inode(inode)), NULL))
 985                 goto out_inode;
 986
 987         d_instantiate(dentry, inode);
 988         inode->i_size = size;
 989         inode->i_nlink = 0;
 990
 991         error = -ENFILE;
 992         file = alloc_file(hugetlbfs_vfsmount, dentry,
 993                         FMODE_WRITE | FMODE_READ,
 994                         &hugetlbfs_file_operations);
 995         if (!file)
 996                 goto out_dentry; /* inode is already attached */
 997
 998         return file;
 999
1000 out_inode:
1001         iput(inode);
1002 out_dentry:
1003         dput(dentry);
1004 out_shm_unlock:
1005         user_shm_unlock(size, user);
1006         return ERR_PTR(error);
1007 }
1008
1009 static int __init init_hugetlbfs_fs(void)
1010 {
1011         int error;
1012         struct vfsmount *vfsmount;
1013
1014         error = bdi_init(&hugetlbfs_backing_dev_info);
1015         if (error)
1016                 return error;
1017
1018         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1019                                         sizeof(struct hugetlbfs_inode_info),
1020                                         0, 0, init_once);
1021         if (hugetlbfs_inode_cachep == NULL)
1022                 goto out2;
1023
1024         error = register_filesystem(&hugetlbfs_fs_type);
1025         if (error)
1026                 goto out;
1027
1028         vfsmount = kern_mount(&hugetlbfs_fs_type);
1029
1030         if (!IS_ERR(vfsmount)) {
1031                 hugetlbfs_vfsmount = vfsmount;
1032                 return 0;
1033         }
1034
1035         error = PTR_ERR(vfsmount);
1036
1037  out:
1038         if (error)
1039                 kmem_cache_destroy(hugetlbfs_inode_cachep);
1040  out2:
1041         bdi_destroy(&hugetlbfs_backing_dev_info);
1042         return error;
1043 }
1044
1045 static void __exit exit_hugetlbfs_fs(void)
1046 {
1047         kmem_cache_destroy(hugetlbfs_inode_cachep);
1048         unregister_filesystem(&hugetlbfs_fs_type);
1049         bdi_destroy(&hugetlbfs_backing_dev_info);
1050 }
1051
1052 module_init(init_hugetlbfs_fs)
1053 module_exit(exit_hugetlbfs_fs)
1054
1055 MODULE_LICENSE("GPL");