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/writeback.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/highmem.h>
  19 #include <linux/init.h>
  20 #include <linux/string.h>
  21 #include <linux/capability.h>
  22 #include <linux/backing-dev.h>
  23 #include <linux/hugetlb.h>
  24 #include <linux/pagevec.h>
  25 #include <linux/quotaops.h>
  26 #include <linux/slab.h>
  27 #include <linux/dnotify.h>
  28 #include <linux/statfs.h>
  29 #include <linux/security.h>
  30
  31 #include <asm/uaccess.h>
  32
  33 /* some random number */
  34 #define HUGETLBFS_MAGIC 0x958458f6
  35
  36 static struct super_operations hugetlbfs_ops;
  37 static struct address_space_operations hugetlbfs_aops;
  38 struct file_operations hugetlbfs_file_operations;
  39 static struct inode_operations hugetlbfs_dir_inode_operations;
  40 static struct inode_operations hugetlbfs_inode_operations;
  41
  42 static struct backing_dev_info hugetlbfs_backing_dev_info = {
  43         .ra_pages       = 0,    /* No readahead */
  44         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
  45 };
  46
  47 int sysctl_hugetlb_shm_group;
  48
  49 static void huge_pagevec_release(struct pagevec *pvec)
  50 {
  51         int i;
  52
  53         for (i = 0; i < pagevec_count(pvec); ++i)
  54                 put_page(pvec->pages[i]);
  55
  56         pagevec_reinit(pvec);
  57 }
  58
  59 /*
  60  * huge_pages_needed tries to determine the number of new huge pages that
  61  * will be required to fully populate this VMA.  This will be equal to
  62  * the size of the VMA in huge pages minus the number of huge pages
  63  * (covered by this VMA) that are found in the page cache.
  64  *
  65  * Result is in bytes to be compatible with is_hugepage_mem_enough()
  66  */
  67 static unsigned long
  68 huge_pages_needed(struct address_space *mapping, struct vm_area_struct *vma)
  69 {
  70         int i;
  71         struct pagevec pvec;
  72         unsigned long start = vma->vm_start;
  73         unsigned long end = vma->vm_end;
  74         unsigned long hugepages = (end - start) >> HPAGE_SHIFT;
  75         pgoff_t next = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
  76         pgoff_t endpg = next + hugepages;
  77
  78         pagevec_init(&pvec, 0);
  79         while (next < endpg) {
  80                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE))
  81                         break;
  82                 for (i = 0; i < pagevec_count(&pvec); i++) {
  83                         struct page *page = pvec.pages[i];
  84                         if (page->index > next)
  85                                 next = page->index;
  86                         if (page->index >= endpg)
  87                                 break;
  88                         next++;
  89                         hugepages--;
  90                 }
  91                 huge_pagevec_release(&pvec);
  92         }
  93         return hugepages << HPAGE_SHIFT;
  94 }
  95
  96 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  97 {
  98         struct inode *inode = file->f_dentry->d_inode;
  99         struct address_space *mapping = inode->i_mapping;
 100         unsigned long bytes;
 101         loff_t len, vma_len;
 102         int ret;
 103
 104         if (vma->vm_pgoff & (HPAGE_SIZE / PAGE_SIZE - 1))
 105                 return -EINVAL;
 106
 107         if (vma->vm_start & ~HPAGE_MASK)
 108                 return -EINVAL;
 109
 110         if (vma->vm_end & ~HPAGE_MASK)
 111                 return -EINVAL;
 112
 113         if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
 114                 return -EINVAL;
 115
 116         bytes = huge_pages_needed(mapping, vma);
 117         if (!is_hugepage_mem_enough(bytes))
 118                 return -ENOMEM;
 119
 120         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 121
 122         mutex_lock(&inode->i_mutex);
 123         file_accessed(file);
 124         vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
 125         vma->vm_ops = &hugetlb_vm_ops;
 126
 127         ret = -ENOMEM;
 128         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 129         if (!(vma->vm_flags & VM_WRITE) && len > inode->i_size)
 130                 goto out;
 131
 132         ret = 0;
 133         hugetlb_prefault_arch_hook(vma->vm_mm);
 134         if (inode->i_size < len)
 135                 inode->i_size = len;
 136 out:
 137         mutex_unlock(&inode->i_mutex);
 138
 139         return ret;
 140 }
 141
 142 /*
 143  * Called under down_write(mmap_sem).
 144  */
 145
 146 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 147 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 148                 unsigned long len, unsigned long pgoff, unsigned long flags);
 149 #else
 150 static unsigned long
 151 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 152                 unsigned long len, unsigned long pgoff, unsigned long flags)
 153 {
 154         struct mm_struct *mm = current->mm;
 155         struct vm_area_struct *vma;
 156         unsigned long start_addr;
 157
 158         if (len & ~HPAGE_MASK)
 159                 return -EINVAL;
 160         if (len > TASK_SIZE)
 161                 return -ENOMEM;
 162
 163         if (addr) {
 164                 addr = ALIGN(addr, HPAGE_SIZE);
 165                 vma = find_vma(mm, addr);
 166                 if (TASK_SIZE - len >= addr &&
 167                     (!vma || addr + len <= vma->vm_start))
 168                         return addr;
 169         }
 170
 171         start_addr = mm->free_area_cache;
 172
 173         if (len <= mm->cached_hole_size)
 174                 start_addr = TASK_UNMAPPED_BASE;
 175
 176 full_search:
 177         addr = ALIGN(start_addr, HPAGE_SIZE);
 178
 179         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 180                 /* At this point:  (!vma || addr < vma->vm_end). */
 181                 if (TASK_SIZE - len < addr) {
 182                         /*
 183                          * Start a new search - just in case we missed
 184                          * some holes.
 185                          */
 186                         if (start_addr != TASK_UNMAPPED_BASE) {
 187                                 start_addr = TASK_UNMAPPED_BASE;
 188                                 goto full_search;
 189                         }
 190                         return -ENOMEM;
 191                 }
 192
 193                 if (!vma || addr + len <= vma->vm_start)
 194                         return addr;
 195                 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
 196         }
 197 }
 198 #endif
 199
 200 /*
 201  * Read a page. Again trivial. If it didn't already exist
 202  * in the page cache, it is zero-filled.
 203  */
 204 static int hugetlbfs_readpage(struct file *file, struct page * page)
 205 {
 206         unlock_page(page);
 207         return -EINVAL;
 208 }
 209
 210 static int hugetlbfs_prepare_write(struct file *file,
 211                         struct page *page, unsigned offset, unsigned to)
 212 {
 213         return -EINVAL;
 214 }
 215
 216 static int hugetlbfs_commit_write(struct file *file,
 217                         struct page *page, unsigned offset, unsigned to)
 218 {
 219         return -EINVAL;
 220 }
 221
 222 static void truncate_huge_page(struct page *page)
 223 {
 224         clear_page_dirty(page);
 225         ClearPageUptodate(page);
 226         remove_from_page_cache(page);
 227         put_page(page);
 228 }
 229
 230 static void truncate_hugepages(struct address_space *mapping, loff_t lstart)
 231 {
 232         const pgoff_t start = lstart >> HPAGE_SHIFT;
 233         struct pagevec pvec;
 234         pgoff_t next;
 235         int i;
 236
 237         pagevec_init(&pvec, 0);
 238         next = start;
 239         while (1) {
 240                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 241                         if (next == start)
 242                                 break;
 243                         next = start;
 244                         continue;
 245                 }
 246
 247                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 248                         struct page *page = pvec.pages[i];
 249
 250                         lock_page(page);
 251                         if (page->index > next)
 252                                 next = page->index;
 253                         ++next;
 254                         truncate_huge_page(page);
 255                         unlock_page(page);
 256                         hugetlb_put_quota(mapping);
 257                 }
 258                 huge_pagevec_release(&pvec);
 259         }
 260         BUG_ON(!lstart && mapping->nrpages);
 261 }
 262
 263 static void hugetlbfs_delete_inode(struct inode *inode)
 264 {
 265         if (inode->i_data.nrpages)
 266                 truncate_hugepages(&inode->i_data, 0);
 267         clear_inode(inode);
 268 }
 269
 270 static void hugetlbfs_forget_inode(struct inode *inode)
 271 {
 272         struct super_block *sb = inode->i_sb;
 273
 274         if (!hlist_unhashed(&inode->i_hash)) {
 275                 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
 276                         list_move(&inode->i_list, &inode_unused);
 277                 inodes_stat.nr_unused++;
 278                 if (!sb || (sb->s_flags & MS_ACTIVE)) {
 279                         spin_unlock(&inode_lock);
 280                         return;
 281                 }
 282                 inode->i_state |= I_WILL_FREE;
 283                 spin_unlock(&inode_lock);
 284                 /*
 285                  * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
 286                  * in our backing_dev_info.
 287                  */
 288                 write_inode_now(inode, 1);
 289                 spin_lock(&inode_lock);
 290                 inode->i_state &= ~I_WILL_FREE;
 291                 inodes_stat.nr_unused--;
 292                 hlist_del_init(&inode->i_hash);
 293         }
 294         list_del_init(&inode->i_list);
 295         list_del_init(&inode->i_sb_list);
 296         inode->i_state |= I_FREEING;
 297         inodes_stat.nr_inodes--;
 298         spin_unlock(&inode_lock);
 299         if (inode->i_data.nrpages)
 300                 truncate_hugepages(&inode->i_data, 0);
 301         clear_inode(inode);
 302         destroy_inode(inode);
 303 }
 304
 305 static void hugetlbfs_drop_inode(struct inode *inode)
 306 {
 307         if (!inode->i_nlink)
 308                 generic_delete_inode(inode);
 309         else
 310                 hugetlbfs_forget_inode(inode);
 311 }
 312
 313 /*
 314  * h_pgoff is in HPAGE_SIZE units.
 315  * vma->vm_pgoff is in PAGE_SIZE units.
 316  */
 317 static inline void
 318 hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
 319 {
 320         struct vm_area_struct *vma;
 321         struct prio_tree_iter iter;
 322
 323         vma_prio_tree_foreach(vma, &iter, root, h_pgoff, ULONG_MAX) {
 324                 unsigned long h_vm_pgoff;
 325                 unsigned long v_offset;
 326
 327                 h_vm_pgoff = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
 328                 v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
 329                 /*
 330                  * Is this VMA fully outside the truncation point?
 331                  */
 332                 if (h_vm_pgoff >= h_pgoff)
 333                         v_offset = 0;
 334
 335                 unmap_hugepage_range(vma,
 336                                 vma->vm_start + v_offset, vma->vm_end);
 337         }
 338 }
 339
 340 /*
 341  * Expanding truncates are not allowed.
 342  */
 343 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 344 {
 345         unsigned long pgoff;
 346         struct address_space *mapping = inode->i_mapping;
 347
 348         if (offset > inode->i_size)
 349                 return -EINVAL;
 350
 351         BUG_ON(offset & ~HPAGE_MASK);
 352         pgoff = offset >> HPAGE_SHIFT;
 353
 354         inode->i_size = offset;
 355         spin_lock(&mapping->i_mmap_lock);
 356         if (!prio_tree_empty(&mapping->i_mmap))
 357                 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 358         spin_unlock(&mapping->i_mmap_lock);
 359         truncate_hugepages(mapping, offset);
 360         return 0;
 361 }
 362
 363 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 364 {
 365         struct inode *inode = dentry->d_inode;
 366         int error;
 367         unsigned int ia_valid = attr->ia_valid;
 368
 369         BUG_ON(!inode);
 370
 371         error = inode_change_ok(inode, attr);
 372         if (error)
 373                 goto out;
 374
 375         if (ia_valid & ATTR_SIZE) {
 376                 error = -EINVAL;
 377                 if (!(attr->ia_size & ~HPAGE_MASK))
 378                         error = hugetlb_vmtruncate(inode, attr->ia_size);
 379                 if (error)
 380                         goto out;
 381                 attr->ia_valid &= ~ATTR_SIZE;
 382         }
 383         error = inode_setattr(inode, attr);
 384 out:
 385         return error;
 386 }
 387
 388 static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
 389                                         gid_t gid, int mode, dev_t dev)
 390 {
 391         struct inode *inode;
 392
 393         inode = new_inode(sb);
 394         if (inode) {
 395                 struct hugetlbfs_inode_info *info;
 396                 inode->i_mode = mode;
 397                 inode->i_uid = uid;
 398                 inode->i_gid = gid;
 399                 inode->i_blksize = HPAGE_SIZE;
 400                 inode->i_blocks = 0;
 401                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 402                 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 403                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 404                 info = HUGETLBFS_I(inode);
 405                 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
 406                 switch (mode & S_IFMT) {
 407                 default:
 408                         init_special_inode(inode, mode, dev);
 409                         break;
 410                 case S_IFREG:
 411                         inode->i_op = &hugetlbfs_inode_operations;
 412                         inode->i_fop = &hugetlbfs_file_operations;
 413                         break;
 414                 case S_IFDIR:
 415                         inode->i_op = &hugetlbfs_dir_inode_operations;
 416                         inode->i_fop = &simple_dir_operations;
 417
 418                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 419                         inode->i_nlink++;
 420                         break;
 421                 case S_IFLNK:
 422                         inode->i_op = &page_symlink_inode_operations;
 423                         break;
 424                 }
 425         }
 426         return inode;
 427 }
 428
 429 /*
 430  * File creation. Allocate an inode, and we're done..
 431  */
 432 static int hugetlbfs_mknod(struct inode *dir,
 433                         struct dentry *dentry, int mode, dev_t dev)
 434 {
 435         struct inode *inode;
 436         int error = -ENOSPC;
 437         gid_t gid;
 438
 439         if (dir->i_mode & S_ISGID) {
 440                 gid = dir->i_gid;
 441                 if (S_ISDIR(mode))
 442                         mode |= S_ISGID;
 443         } else {
 444                 gid = current->fsgid;
 445         }
 446         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
 447         if (inode) {
 448                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 449                 d_instantiate(dentry, inode);
 450                 dget(dentry);   /* Extra count - pin the dentry in core */
 451                 error = 0;
 452         }
 453         return error;
 454 }
 455
 456 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 457 {
 458         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 459         if (!retval)
 460                 dir->i_nlink++;
 461         return retval;
 462 }
 463
 464 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
 465 {
 466         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 467 }
 468
 469 static int hugetlbfs_symlink(struct inode *dir,
 470                         struct dentry *dentry, const char *symname)
 471 {
 472         struct inode *inode;
 473         int error = -ENOSPC;
 474         gid_t gid;
 475
 476         if (dir->i_mode & S_ISGID)
 477                 gid = dir->i_gid;
 478         else
 479                 gid = current->fsgid;
 480
 481         inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
 482                                         gid, S_IFLNK|S_IRWXUGO, 0);
 483         if (inode) {
 484                 int l = strlen(symname)+1;
 485                 error = page_symlink(inode, symname, l);
 486                 if (!error) {
 487                         d_instantiate(dentry, inode);
 488                         dget(dentry);
 489                 } else
 490                         iput(inode);
 491         }
 492         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 493
 494         return error;
 495 }
 496
 497 /*
 498  * For direct-IO reads into hugetlb pages
 499  */
 500 static int hugetlbfs_set_page_dirty(struct page *page)
 501 {
 502         return 0;
 503 }
 504
 505 static int hugetlbfs_statfs(struct super_block *sb, struct kstatfs *buf)
 506 {
 507         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 508
 509         buf->f_type = HUGETLBFS_MAGIC;
 510         buf->f_bsize = HPAGE_SIZE;
 511         if (sbinfo) {
 512                 spin_lock(&sbinfo->stat_lock);
 513                 /* If no limits set, just report 0 for max/free/used
 514                  * blocks, like simple_statfs() */
 515                 if (sbinfo->max_blocks >= 0) {
 516                         buf->f_blocks = sbinfo->max_blocks;
 517                         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 518                         buf->f_files = sbinfo->max_inodes;
 519                         buf->f_ffree = sbinfo->free_inodes;
 520                 }
 521                 spin_unlock(&sbinfo->stat_lock);
 522         }
 523         buf->f_namelen = NAME_MAX;
 524         return 0;
 525 }
 526
 527 static void hugetlbfs_put_super(struct super_block *sb)
 528 {
 529         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 530
 531         if (sbi) {
 532                 sb->s_fs_info = NULL;
 533                 kfree(sbi);
 534         }
 535 }
 536
 537 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 538 {
 539         if (sbinfo->free_inodes >= 0) {
 540                 spin_lock(&sbinfo->stat_lock);
 541                 if (unlikely(!sbinfo->free_inodes)) {
 542                         spin_unlock(&sbinfo->stat_lock);
 543                         return 0;
 544                 }
 545                 sbinfo->free_inodes--;
 546                 spin_unlock(&sbinfo->stat_lock);
 547         }
 548
 549         return 1;
 550 }
 551
 552 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 553 {
 554         if (sbinfo->free_inodes >= 0) {
 555                 spin_lock(&sbinfo->stat_lock);
 556                 sbinfo->free_inodes++;
 557                 spin_unlock(&sbinfo->stat_lock);
 558         }
 559 }
 560
 561
 562 static kmem_cache_t *hugetlbfs_inode_cachep;
 563
 564 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 565 {
 566         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 567         struct hugetlbfs_inode_info *p;
 568
 569         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 570                 return NULL;
 571         p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
 572         if (unlikely(!p)) {
 573                 hugetlbfs_inc_free_inodes(sbinfo);
 574                 return NULL;
 575         }
 576         return &p->vfs_inode;
 577 }
 578
 579 static void hugetlbfs_destroy_inode(struct inode *inode)
 580 {
 581         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 582         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 583         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 584 }
 585
 586 static struct address_space_operations hugetlbfs_aops = {
 587         .readpage       = hugetlbfs_readpage,
 588         .prepare_write  = hugetlbfs_prepare_write,
 589         .commit_write   = hugetlbfs_commit_write,
 590         .set_page_dirty = hugetlbfs_set_page_dirty,
 591 };
 592
 593
 594 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
 595 {
 596         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 597
 598         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 599             SLAB_CTOR_CONSTRUCTOR)
 600                 inode_init_once(&ei->vfs_inode);
 601 }
 602
 603 struct file_operations hugetlbfs_file_operations = {
 604         .mmap                   = hugetlbfs_file_mmap,
 605         .fsync                  = simple_sync_file,
 606         .get_unmapped_area      = hugetlb_get_unmapped_area,
 607 };
 608
 609 static struct inode_operations hugetlbfs_dir_inode_operations = {
 610         .create         = hugetlbfs_create,
 611         .lookup         = simple_lookup,
 612         .link           = simple_link,
 613         .unlink         = simple_unlink,
 614         .symlink        = hugetlbfs_symlink,
 615         .mkdir          = hugetlbfs_mkdir,
 616         .rmdir          = simple_rmdir,
 617         .mknod          = hugetlbfs_mknod,
 618         .rename         = simple_rename,
 619         .setattr        = hugetlbfs_setattr,
 620 };
 621
 622 static struct inode_operations hugetlbfs_inode_operations = {
 623         .setattr        = hugetlbfs_setattr,
 624 };
 625
 626 static struct super_operations hugetlbfs_ops = {
 627         .alloc_inode    = hugetlbfs_alloc_inode,
 628         .destroy_inode  = hugetlbfs_destroy_inode,
 629         .statfs         = hugetlbfs_statfs,
 630         .delete_inode   = hugetlbfs_delete_inode,
 631         .drop_inode     = hugetlbfs_drop_inode,
 632         .put_super      = hugetlbfs_put_super,
 633 };
 634
 635 static int
 636 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 637 {
 638         char *opt, *value, *rest;
 639
 640         if (!options)
 641                 return 0;
 642         while ((opt = strsep(&options, ",")) != NULL) {
 643                 if (!*opt)
 644                         continue;
 645
 646                 value = strchr(opt, '=');
 647                 if (!value || !*value)
 648                         return -EINVAL;
 649                 else
 650                         *value++ = '\0';
 651
 652                 if (!strcmp(opt, "uid"))
 653                         pconfig->uid = simple_strtoul(value, &value, 0);
 654                 else if (!strcmp(opt, "gid"))
 655                         pconfig->gid = simple_strtoul(value, &value, 0);
 656                 else if (!strcmp(opt, "mode"))
 657                         pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
 658                 else if (!strcmp(opt, "size")) {
 659                         unsigned long long size = memparse(value, &rest);
 660                         if (*rest == '%') {
 661                                 size <<= HPAGE_SHIFT;
 662                                 size *= max_huge_pages;
 663                                 do_div(size, 100);
 664                                 rest++;
 665                         }
 666                         size &= HPAGE_MASK;
 667                         pconfig->nr_blocks = (size >> HPAGE_SHIFT);
 668                         value = rest;
 669                 } else if (!strcmp(opt,"nr_inodes")) {
 670                         pconfig->nr_inodes = memparse(value, &rest);
 671                         value = rest;
 672                 } else
 673                         return -EINVAL;
 674
 675                 if (*value)
 676                         return -EINVAL;
 677         }
 678         return 0;
 679 }
 680
 681 static int
 682 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 683 {
 684         struct inode * inode;
 685         struct dentry * root;
 686         int ret;
 687         struct hugetlbfs_config config;
 688         struct hugetlbfs_sb_info *sbinfo;
 689
 690         config.nr_blocks = -1; /* No limit on size by default */
 691         config.nr_inodes = -1; /* No limit on number of inodes by default */
 692         config.uid = current->fsuid;
 693         config.gid = current->fsgid;
 694         config.mode = 0755;
 695         ret = hugetlbfs_parse_options(data, &config);
 696
 697         if (ret)
 698                 return ret;
 699
 700         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 701         if (!sbinfo)
 702                 return -ENOMEM;
 703         sb->s_fs_info = sbinfo;
 704         spin_lock_init(&sbinfo->stat_lock);
 705         sbinfo->max_blocks = config.nr_blocks;
 706         sbinfo->free_blocks = config.nr_blocks;
 707         sbinfo->max_inodes = config.nr_inodes;
 708         sbinfo->free_inodes = config.nr_inodes;
 709         sb->s_maxbytes = MAX_LFS_FILESIZE;
 710         sb->s_blocksize = HPAGE_SIZE;
 711         sb->s_blocksize_bits = HPAGE_SHIFT;
 712         sb->s_magic = HUGETLBFS_MAGIC;
 713         sb->s_op = &hugetlbfs_ops;
 714         sb->s_time_gran = 1;
 715         inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 716                                         S_IFDIR | config.mode, 0);
 717         if (!inode)
 718                 goto out_free;
 719
 720         root = d_alloc_root(inode);
 721         if (!root) {
 722                 iput(inode);
 723                 goto out_free;
 724         }
 725         sb->s_root = root;
 726         return 0;
 727 out_free:
 728         kfree(sbinfo);
 729         return -ENOMEM;
 730 }
 731
 732 int hugetlb_get_quota(struct address_space *mapping)
 733 {
 734         int ret = 0;
 735         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 736
 737         if (sbinfo->free_blocks > -1) {
 738                 spin_lock(&sbinfo->stat_lock);
 739                 if (sbinfo->free_blocks > 0)
 740                         sbinfo->free_blocks--;
 741                 else
 742                         ret = -ENOMEM;
 743                 spin_unlock(&sbinfo->stat_lock);
 744         }
 745
 746         return ret;
 747 }
 748
 749 void hugetlb_put_quota(struct address_space *mapping)
 750 {
 751         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 752
 753         if (sbinfo->free_blocks > -1) {
 754                 spin_lock(&sbinfo->stat_lock);
 755                 sbinfo->free_blocks++;
 756                 spin_unlock(&sbinfo->stat_lock);
 757         }
 758 }
 759
 760 static struct super_block *hugetlbfs_get_sb(struct file_system_type *fs_type,
 761         int flags, const char *dev_name, void *data)
 762 {
 763         return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 764 }
 765
 766 static struct file_system_type hugetlbfs_fs_type = {
 767         .name           = "hugetlbfs",
 768         .get_sb         = hugetlbfs_get_sb,
 769         .kill_sb        = kill_litter_super,
 770 };
 771
 772 static struct vfsmount *hugetlbfs_vfsmount;
 773
 774 /*
 775  * Return the next identifier for a shm file
 776  */
 777 static unsigned long hugetlbfs_counter(void)
 778 {
 779         static DEFINE_SPINLOCK(lock);
 780         static unsigned long counter;
 781         unsigned long ret;
 782
 783         spin_lock(&lock);
 784         ret = ++counter;
 785         spin_unlock(&lock);
 786         return ret;
 787 }
 788
 789 static int can_do_hugetlb_shm(void)
 790 {
 791         return likely(capable(CAP_IPC_LOCK) ||
 792                         in_group_p(sysctl_hugetlb_shm_group) ||
 793                         can_do_mlock());
 794 }
 795
 796 struct file *hugetlb_zero_setup(size_t size)
 797 {
 798         int error = -ENOMEM;
 799         struct file *file;
 800         struct inode *inode;
 801         struct dentry *dentry, *root;
 802         struct qstr quick_string;
 803         char buf[16];
 804
 805         if (!can_do_hugetlb_shm())
 806                 return ERR_PTR(-EPERM);
 807
 808         if (!is_hugepage_mem_enough(size))
 809                 return ERR_PTR(-ENOMEM);
 810
 811         if (!user_shm_lock(size, current->user))
 812                 return ERR_PTR(-ENOMEM);
 813
 814         root = hugetlbfs_vfsmount->mnt_root;
 815         snprintf(buf, 16, "%lu", hugetlbfs_counter());
 816         quick_string.name = buf;
 817         quick_string.len = strlen(quick_string.name);
 818         quick_string.hash = 0;
 819         dentry = d_alloc(root, &quick_string);
 820         if (!dentry)
 821                 goto out_shm_unlock;
 822
 823         error = -ENFILE;
 824         file = get_empty_filp();
 825         if (!file)
 826                 goto out_dentry;
 827
 828         error = -ENOSPC;
 829         inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
 830                                 current->fsgid, S_IFREG | S_IRWXUGO, 0);
 831         if (!inode)
 832                 goto out_file;
 833
 834         d_instantiate(dentry, inode);
 835         inode->i_size = size;
 836         inode->i_nlink = 0;
 837         file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
 838         file->f_dentry = dentry;
 839         file->f_mapping = inode->i_mapping;
 840         file->f_op = &hugetlbfs_file_operations;
 841         file->f_mode = FMODE_WRITE | FMODE_READ;
 842         return file;
 843
 844 out_file:
 845         put_filp(file);
 846 out_dentry:
 847         dput(dentry);
 848 out_shm_unlock:
 849         user_shm_unlock(size, current->user);
 850         return ERR_PTR(error);
 851 }
 852
 853 static int __init init_hugetlbfs_fs(void)
 854 {
 855         int error;
 856         struct vfsmount *vfsmount;
 857
 858         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
 859                                         sizeof(struct hugetlbfs_inode_info),
 860                                         0, 0, init_once, NULL);
 861         if (hugetlbfs_inode_cachep == NULL)
 862                 return -ENOMEM;
 863
 864         error = register_filesystem(&hugetlbfs_fs_type);
 865         if (error)
 866                 goto out;
 867
 868         vfsmount = kern_mount(&hugetlbfs_fs_type);
 869
 870         if (!IS_ERR(vfsmount)) {
 871                 hugetlbfs_vfsmount = vfsmount;
 872                 return 0;
 873         }
 874
 875         error = PTR_ERR(vfsmount);
 876
 877  out:
 878         if (error)
 879                 kmem_cache_destroy(hugetlbfs_inode_cachep);
 880         return error;
 881 }
 882
 883 static void __exit exit_hugetlbfs_fs(void)
 884 {
 885         kmem_cache_destroy(hugetlbfs_inode_cachep);
 886         unregister_filesystem(&hugetlbfs_fs_type);
 887 }
 888
 889 module_init(init_hugetlbfs_fs)
 890 module_exit(exit_hugetlbfs_fs)
 891
 892 MODULE_LICENSE("GPL");