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