fs/hugetlbfs/inode.c

   1 /*
   2  * hugetlbpage-backed filesystem.  Based on ramfs.
   3  *
   4  * Nadia Yvette Chambers, 2002
   5  *
   6  * Copyright (C) 2002 Linus Torvalds.
   7  * License: GPL
   8  */
   9
  10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/thread_info.h>
  13 #include <asm/current.h>
  14 #include <linux/sched/signal.h>         /* remove ASAP */
  15 #include <linux/falloc.h>
  16 #include <linux/fs.h>
  17 #include <linux/mount.h>
  18 #include <linux/file.h>
  19 #include <linux/kernel.h>
  20 #include <linux/writeback.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/highmem.h>
  23 #include <linux/init.h>
  24 #include <linux/string.h>
  25 #include <linux/capability.h>
  26 #include <linux/ctype.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/hugetlb.h>
  29 #include <linux/pagevec.h>
  30 #include <linux/fs_parser.h>
  31 #include <linux/mman.h>
  32 #include <linux/slab.h>
  33 #include <linux/dnotify.h>
  34 #include <linux/statfs.h>
  35 #include <linux/security.h>
  36 #include <linux/magic.h>
  37 #include <linux/migrate.h>
  38 #include <linux/uio.h>
  39
  40 #include <linux/uaccess.h>
  41
  42 static const struct super_operations hugetlbfs_ops;
  43 static const struct address_space_operations hugetlbfs_aops;
  44 const struct file_operations hugetlbfs_file_operations;
  45 static const struct inode_operations hugetlbfs_dir_inode_operations;
  46 static const struct inode_operations hugetlbfs_inode_operations;
  47
  48 enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT };
  49
  50 struct hugetlbfs_fs_context {
  51         struct hstate           *hstate;
  52         unsigned long long      max_size_opt;
  53         unsigned long long      min_size_opt;
  54         long                    max_hpages;
  55         long                    nr_inodes;
  56         long                    min_hpages;
  57         enum hugetlbfs_size_type max_val_type;
  58         enum hugetlbfs_size_type min_val_type;
  59         kuid_t                  uid;
  60         kgid_t                  gid;
  61         umode_t                 mode;
  62 };
  63
  64 int sysctl_hugetlb_shm_group;
  65
  66 enum hugetlb_param {
  67         Opt_gid,
  68         Opt_min_size,
  69         Opt_mode,
  70         Opt_nr_inodes,
  71         Opt_pagesize,
  72         Opt_size,
  73         Opt_uid,
  74 };
  75
  76 static const struct fs_parameter_spec hugetlb_fs_parameters[] = {
  77         fsparam_u32   ("gid",           Opt_gid),
  78         fsparam_string("min_size",      Opt_min_size),
  79         fsparam_u32   ("mode",          Opt_mode),
  80         fsparam_string("nr_inodes",     Opt_nr_inodes),
  81         fsparam_string("pagesize",      Opt_pagesize),
  82         fsparam_string("size",          Opt_size),
  83         fsparam_u32   ("uid",           Opt_uid),
  84         {}
  85 };
  86
  87 #ifdef CONFIG_NUMA
  88 static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
  89                                         struct inode *inode, pgoff_t index)
  90 {
  91         vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy,
  92                                                         index);
  93 }
  94
  95 static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
  96 {
  97         mpol_cond_put(vma->vm_policy);
  98 }
  99 #else
 100 static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
 101                                         struct inode *inode, pgoff_t index)
 102 {
 103 }
 104
 105 static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
 106 {
 107 }
 108 #endif
 109
 110 static void huge_pagevec_release(struct pagevec *pvec)
 111 {
 112         int i;
 113
 114         for (i = 0; i < pagevec_count(pvec); ++i)
 115                 put_page(pvec->pages[i]);
 116
 117         pagevec_reinit(pvec);
 118 }
 119
 120 /*
 121  * Mask used when checking the page offset value passed in via system
 122  * calls.  This value will be converted to a loff_t which is signed.
 123  * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the
 124  * value.  The extra bit (- 1 in the shift value) is to take the sign
 125  * bit into account.
 126  */
 127 #define PGOFF_LOFFT_MAX \
 128         (((1UL << (PAGE_SHIFT + 1)) - 1) <<  (BITS_PER_LONG - (PAGE_SHIFT + 1)))
 129
 130 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 131 {
 132         struct inode *inode = file_inode(file);
 133         loff_t len, vma_len;
 134         int ret;
 135         struct hstate *h = hstate_file(file);
 136
 137         /*
 138          * vma address alignment (but not the pgoff alignment) has
 139          * already been checked by prepare_hugepage_range.  If you add
 140          * any error returns here, do so after setting VM_HUGETLB, so
 141          * is_vm_hugetlb_page tests below unmap_region go the right
 142          * way when do_mmap_pgoff unwinds (may be important on powerpc
 143          * and ia64).
 144          */
 145         vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
 146         vma->vm_ops = &hugetlb_vm_ops;
 147
 148         /*
 149          * page based offset in vm_pgoff could be sufficiently large to
 150          * overflow a loff_t when converted to byte offset.  This can
 151          * only happen on architectures where sizeof(loff_t) ==
 152          * sizeof(unsigned long).  So, only check in those instances.
 153          */
 154         if (sizeof(unsigned long) == sizeof(loff_t)) {
 155                 if (vma->vm_pgoff & PGOFF_LOFFT_MAX)
 156                         return -EINVAL;
 157         }
 158
 159         /* must be huge page aligned */
 160         if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
 161                 return -EINVAL;
 162
 163         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 164         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 165         /* check for overflow */
 166         if (len < vma_len)
 167                 return -EINVAL;
 168
 169         inode_lock(inode);
 170         file_accessed(file);
 171
 172         ret = -ENOMEM;
 173         if (hugetlb_reserve_pages(inode,
 174                                 vma->vm_pgoff >> huge_page_order(h),
 175                                 len >> huge_page_shift(h), vma,
 176                                 vma->vm_flags))
 177                 goto out;
 178
 179         ret = 0;
 180         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 181                 i_size_write(inode, len);
 182 out:
 183         inode_unlock(inode);
 184
 185         return ret;
 186 }
 187
 188 /*
 189  * Called under down_write(mmap_sem).
 190  */
 191
 192 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 193 static unsigned long
 194 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 195                 unsigned long len, unsigned long pgoff, unsigned long flags)
 196 {
 197         struct mm_struct *mm = current->mm;
 198         struct vm_area_struct *vma;
 199         struct hstate *h = hstate_file(file);
 200         struct vm_unmapped_area_info info;
 201
 202         if (len & ~huge_page_mask(h))
 203                 return -EINVAL;
 204         if (len > TASK_SIZE)
 205                 return -ENOMEM;
 206
 207         if (flags & MAP_FIXED) {
 208                 if (prepare_hugepage_range(file, addr, len))
 209                         return -EINVAL;
 210                 return addr;
 211         }
 212
 213         if (addr) {
 214                 addr = ALIGN(addr, huge_page_size(h));
 215                 vma = find_vma(mm, addr);
 216                 if (TASK_SIZE - len >= addr &&
 217                     (!vma || addr + len <= vm_start_gap(vma)))
 218                         return addr;
 219         }
 220
 221         info.flags = 0;
 222         info.length = len;
 223         info.low_limit = TASK_UNMAPPED_BASE;
 224         info.high_limit = TASK_SIZE;
 225         info.align_mask = PAGE_MASK & ~huge_page_mask(h);
 226         info.align_offset = 0;
 227         return vm_unmapped_area(&info);
 228 }
 229 #endif
 230
 231 static size_t
 232 hugetlbfs_read_actor(struct page *page, unsigned long offset,
 233                         struct iov_iter *to, unsigned long size)
 234 {
 235         size_t copied = 0;
 236         int i, chunksize;
 237
 238         /* Find which 4k chunk and offset with in that chunk */
 239         i = offset >> PAGE_SHIFT;
 240         offset = offset & ~PAGE_MASK;
 241
 242         while (size) {
 243                 size_t n;
 244                 chunksize = PAGE_SIZE;
 245                 if (offset)
 246                         chunksize -= offset;
 247                 if (chunksize > size)
 248                         chunksize = size;
 249                 n = copy_page_to_iter(&page[i], offset, chunksize, to);
 250                 copied += n;
 251                 if (n != chunksize)
 252                         return copied;
 253                 offset = 0;
 254                 size -= chunksize;
 255                 i++;
 256         }
 257         return copied;
 258 }
 259
 260 /*
 261  * Support for read() - Find the page attached to f_mapping and copy out the
 262  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
 263  * since it has PAGE_SIZE assumptions.
 264  */
 265 static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
 266 {
 267         struct file *file = iocb->ki_filp;
 268         struct hstate *h = hstate_file(file);
 269         struct address_space *mapping = file->f_mapping;
 270         struct inode *inode = mapping->host;
 271         unsigned long index = iocb->ki_pos >> huge_page_shift(h);
 272         unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
 273         unsigned long end_index;
 274         loff_t isize;
 275         ssize_t retval = 0;
 276
 277         while (iov_iter_count(to)) {
 278                 struct page *page;
 279                 size_t nr, copied;
 280
 281                 /* nr is the maximum number of bytes to copy from this page */
 282                 nr = huge_page_size(h);
 283                 isize = i_size_read(inode);
 284                 if (!isize)
 285                         break;
 286                 end_index = (isize - 1) >> huge_page_shift(h);
 287                 if (index > end_index)
 288                         break;
 289                 if (index == end_index) {
 290                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
 291                         if (nr <= offset)
 292                                 break;
 293                 }
 294                 nr = nr - offset;
 295
 296                 /* Find the page */
 297                 page = find_lock_page(mapping, index);
 298                 if (unlikely(page == NULL)) {
 299                         /*
 300                          * We have a HOLE, zero out the user-buffer for the
 301                          * length of the hole or request.
 302                          */
 303                         copied = iov_iter_zero(nr, to);
 304                 } else {
 305                         unlock_page(page);
 306
 307                         /*
 308                          * We have the page, copy it to user space buffer.
 309                          */
 310                         copied = hugetlbfs_read_actor(page, offset, to, nr);
 311                         put_page(page);
 312                 }
 313                 offset += copied;
 314                 retval += copied;
 315                 if (copied != nr && iov_iter_count(to)) {
 316                         if (!retval)
 317                                 retval = -EFAULT;
 318                         break;
 319                 }
 320                 index += offset >> huge_page_shift(h);
 321                 offset &= ~huge_page_mask(h);
 322         }
 323         iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
 324         return retval;
 325 }
 326
 327 static int hugetlbfs_write_begin(struct file *file,
 328                         struct address_space *mapping,
 329                         loff_t pos, unsigned len, unsigned flags,
 330                         struct page **pagep, void **fsdata)
 331 {
 332         return -EINVAL;
 333 }
 334
 335 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 336                         loff_t pos, unsigned len, unsigned copied,
 337                         struct page *page, void *fsdata)
 338 {
 339         BUG();
 340         return -EINVAL;
 341 }
 342
 343 static void remove_huge_page(struct page *page)
 344 {
 345         ClearPageDirty(page);
 346         ClearPageUptodate(page);
 347         delete_from_page_cache(page);
 348 }
 349
 350 static void
 351 hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end)
 352 {
 353         struct vm_area_struct *vma;
 354
 355         /*
 356          * end == 0 indicates that the entire range after
 357          * start should be unmapped.
 358          */
 359         vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) {
 360                 unsigned long v_offset;
 361                 unsigned long v_end;
 362
 363                 /*
 364                  * Can the expression below overflow on 32-bit arches?
 365                  * No, because the interval tree returns us only those vmas
 366                  * which overlap the truncated area starting at pgoff,
 367                  * and no vma on a 32-bit arch can span beyond the 4GB.
 368                  */
 369                 if (vma->vm_pgoff < start)
 370                         v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT;
 371                 else
 372                         v_offset = 0;
 373
 374                 if (!end)
 375                         v_end = vma->vm_end;
 376                 else {
 377                         v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT)
 378                                                         + vma->vm_start;
 379                         if (v_end > vma->vm_end)
 380                                 v_end = vma->vm_end;
 381                 }
 382
 383                 unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end,
 384                                                                         NULL);
 385         }
 386 }
 387
 388 /*
 389  * remove_inode_hugepages handles two distinct cases: truncation and hole
 390  * punch.  There are subtle differences in operation for each case.
 391  *
 392  * truncation is indicated by end of range being LLONG_MAX
 393  *      In this case, we first scan the range and release found pages.
 394  *      After releasing pages, hugetlb_unreserve_pages cleans up region/reserv
 395  *      maps and global counts.  Page faults can not race with truncation
 396  *      in this routine.  hugetlb_no_page() prevents page faults in the
 397  *      truncated range.  It checks i_size before allocation, and again after
 398  *      with the page table lock for the page held.  The same lock must be
 399  *      acquired to unmap a page.
 400  * hole punch is indicated if end is not LLONG_MAX
 401  *      In the hole punch case we scan the range and release found pages.
 402  *      Only when releasing a page is the associated region/reserv map
 403  *      deleted.  The region/reserv map for ranges without associated
 404  *      pages are not modified.  Page faults can race with hole punch.
 405  *      This is indicated if we find a mapped page.
 406  * Note: If the passed end of range value is beyond the end of file, but
 407  * not LLONG_MAX this routine still performs a hole punch operation.
 408  */
 409 static void remove_inode_hugepages(struct inode *inode, loff_t lstart,
 410                                    loff_t lend)
 411 {
 412         struct hstate *h = hstate_inode(inode);
 413         struct address_space *mapping = &inode->i_data;
 414         const pgoff_t start = lstart >> huge_page_shift(h);
 415         const pgoff_t end = lend >> huge_page_shift(h);
 416         struct vm_area_struct pseudo_vma;
 417         struct pagevec pvec;
 418         pgoff_t next, index;
 419         int i, freed = 0;
 420         bool truncate_op = (lend == LLONG_MAX);
 421
 422         vma_init(&pseudo_vma, current->mm);
 423         pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
 424         pagevec_init(&pvec);
 425         next = start;
 426         while (next < end) {
 427                 /*
 428                  * When no more pages are found, we are done.
 429                  */
 430                 if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1))
 431                         break;
 432
 433                 for (i = 0; i < pagevec_count(&pvec); ++i) {
 434                         struct page *page = pvec.pages[i];
 435                         u32 hash;
 436
 437                         index = page->index;
 438                         hash = hugetlb_fault_mutex_hash(mapping, index);
 439                         mutex_lock(&hugetlb_fault_mutex_table[hash]);
 440
 441                         /*
 442                          * If page is mapped, it was faulted in after being
 443                          * unmapped in caller.  Unmap (again) now after taking
 444                          * the fault mutex.  The mutex will prevent faults
 445                          * until we finish removing the page.
 446                          *
 447                          * This race can only happen in the hole punch case.
 448                          * Getting here in a truncate operation is a bug.
 449                          */
 450                         if (unlikely(page_mapped(page))) {
 451                                 BUG_ON(truncate_op);
 452
 453                                 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 454                                 i_mmap_lock_write(mapping);
 455                                 mutex_lock(&hugetlb_fault_mutex_table[hash]);
 456                                 hugetlb_vmdelete_list(&mapping->i_mmap,
 457                                         index * pages_per_huge_page(h),
 458                                         (index + 1) * pages_per_huge_page(h));
 459                                 i_mmap_unlock_write(mapping);
 460                         }
 461
 462                         lock_page(page);
 463                         /*
 464                          * We must free the huge page and remove from page
 465                          * cache (remove_huge_page) BEFORE removing the
 466                          * region/reserve map (hugetlb_unreserve_pages).  In
 467                          * rare out of memory conditions, removal of the
 468                          * region/reserve map could fail. Correspondingly,
 469                          * the subpool and global reserve usage count can need
 470                          * to be adjusted.
 471                          */
 472                         VM_BUG_ON(PagePrivate(page));
 473                         remove_huge_page(page);
 474                         freed++;
 475                         if (!truncate_op) {
 476                                 if (unlikely(hugetlb_unreserve_pages(inode,
 477                                                         index, index + 1, 1)))
 478                                         hugetlb_fix_reserve_counts(inode);
 479                         }
 480
 481                         unlock_page(page);
 482                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 483                 }
 484                 huge_pagevec_release(&pvec);
 485                 cond_resched();
 486         }
 487
 488         if (truncate_op)
 489                 (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed);
 490 }
 491
 492 static void hugetlbfs_evict_inode(struct inode *inode)
 493 {
 494         struct resv_map *resv_map;
 495
 496         remove_inode_hugepages(inode, 0, LLONG_MAX);
 497
 498         /*
 499          * Get the resv_map from the address space embedded in the inode.
 500          * This is the address space which points to any resv_map allocated
 501          * at inode creation time.  If this is a device special inode,
 502          * i_mapping may not point to the original address space.
 503          */
 504         resv_map = (struct resv_map *)(&inode->i_data)->private_data;
 505         /* Only regular and link inodes have associated reserve maps */
 506         if (resv_map)
 507                 resv_map_release(&resv_map->refs);
 508         clear_inode(inode);
 509 }
 510
 511 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 512 {
 513         pgoff_t pgoff;
 514         struct address_space *mapping = inode->i_mapping;
 515         struct hstate *h = hstate_inode(inode);
 516
 517         BUG_ON(offset & ~huge_page_mask(h));
 518         pgoff = offset >> PAGE_SHIFT;
 519
 520         i_size_write(inode, offset);
 521         i_mmap_lock_write(mapping);
 522         if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
 523                 hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0);
 524         i_mmap_unlock_write(mapping);
 525         remove_inode_hugepages(inode, offset, LLONG_MAX);
 526         return 0;
 527 }
 528
 529 static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
 530 {
 531         struct hstate *h = hstate_inode(inode);
 532         loff_t hpage_size = huge_page_size(h);
 533         loff_t hole_start, hole_end;
 534
 535         /*
 536          * For hole punch round up the beginning offset of the hole and
 537          * round down the end.
 538          */
 539         hole_start = round_up(offset, hpage_size);
 540         hole_end = round_down(offset + len, hpage_size);
 541
 542         if (hole_end > hole_start) {
 543                 struct address_space *mapping = inode->i_mapping;
 544                 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
 545
 546                 inode_lock(inode);
 547
 548                 /* protected by i_mutex */
 549                 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
 550                         inode_unlock(inode);
 551                         return -EPERM;
 552                 }
 553
 554                 i_mmap_lock_write(mapping);
 555                 if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
 556                         hugetlb_vmdelete_list(&mapping->i_mmap,
 557                                                 hole_start >> PAGE_SHIFT,
 558                                                 hole_end  >> PAGE_SHIFT);
 559                 i_mmap_unlock_write(mapping);
 560                 remove_inode_hugepages(inode, hole_start, hole_end);
 561                 inode_unlock(inode);
 562         }
 563
 564         return 0;
 565 }
 566
 567 static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
 568                                 loff_t len)
 569 {
 570         struct inode *inode = file_inode(file);
 571         struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
 572         struct address_space *mapping = inode->i_mapping;
 573         struct hstate *h = hstate_inode(inode);
 574         struct vm_area_struct pseudo_vma;
 575         struct mm_struct *mm = current->mm;
 576         loff_t hpage_size = huge_page_size(h);
 577         unsigned long hpage_shift = huge_page_shift(h);
 578         pgoff_t start, index, end;
 579         int error;
 580         u32 hash;
 581
 582         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 583                 return -EOPNOTSUPP;
 584
 585         if (mode & FALLOC_FL_PUNCH_HOLE)
 586                 return hugetlbfs_punch_hole(inode, offset, len);
 587
 588         /*
 589          * Default preallocate case.
 590          * For this range, start is rounded down and end is rounded up
 591          * as well as being converted to page offsets.
 592          */
 593         start = offset >> hpage_shift;
 594         end = (offset + len + hpage_size - 1) >> hpage_shift;
 595
 596         inode_lock(inode);
 597
 598         /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
 599         error = inode_newsize_ok(inode, offset + len);
 600         if (error)
 601                 goto out;
 602
 603         if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
 604                 error = -EPERM;
 605                 goto out;
 606         }
 607
 608         /*
 609          * Initialize a pseudo vma as this is required by the huge page
 610          * allocation routines.  If NUMA is configured, use page index
 611          * as input to create an allocation policy.
 612          */
 613         vma_init(&pseudo_vma, mm);
 614         pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
 615         pseudo_vma.vm_file = file;
 616
 617         for (index = start; index < end; index++) {
 618                 /*
 619                  * This is supposed to be the vaddr where the page is being
 620                  * faulted in, but we have no vaddr here.
 621                  */
 622                 struct page *page;
 623                 unsigned long addr;
 624                 int avoid_reserve = 0;
 625
 626                 cond_resched();
 627
 628                 /*
 629                  * fallocate(2) manpage permits EINTR; we may have been
 630                  * interrupted because we are using up too much memory.
 631                  */
 632                 if (signal_pending(current)) {
 633                         error = -EINTR;
 634                         break;
 635                 }
 636
 637                 /* Set numa allocation policy based on index */
 638                 hugetlb_set_vma_policy(&pseudo_vma, inode, index);
 639
 640                 /* addr is the offset within the file (zero based) */
 641                 addr = index * hpage_size;
 642
 643                 /* mutex taken here, fault path and hole punch */
 644                 hash = hugetlb_fault_mutex_hash(mapping, index);
 645                 mutex_lock(&hugetlb_fault_mutex_table[hash]);
 646
 647                 /* See if already present in mapping to avoid alloc/free */
 648                 page = find_get_page(mapping, index);
 649                 if (page) {
 650                         put_page(page);
 651                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 652                         hugetlb_drop_vma_policy(&pseudo_vma);
 653                         continue;
 654                 }
 655
 656                 /* Allocate page and add to page cache */
 657                 page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve);
 658                 hugetlb_drop_vma_policy(&pseudo_vma);
 659                 if (IS_ERR(page)) {
 660                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 661                         error = PTR_ERR(page);
 662                         goto out;
 663                 }
 664                 clear_huge_page(page, addr, pages_per_huge_page(h));
 665                 __SetPageUptodate(page);
 666                 error = huge_add_to_page_cache(page, mapping, index);
 667                 if (unlikely(error)) {
 668                         put_page(page);
 669                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 670                         goto out;
 671                 }
 672
 673                 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 674
 675                 /*
 676                  * unlock_page because locked by add_to_page_cache()
 677                  * page_put due to reference from alloc_huge_page()
 678                  */
 679                 unlock_page(page);
 680                 put_page(page);
 681         }
 682
 683         if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
 684                 i_size_write(inode, offset + len);
 685         inode->i_ctime = current_time(inode);
 686 out:
 687         inode_unlock(inode);
 688         return error;
 689 }
 690
 691 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 692 {
 693         struct inode *inode = d_inode(dentry);
 694         struct hstate *h = hstate_inode(inode);
 695         int error;
 696         unsigned int ia_valid = attr->ia_valid;
 697         struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
 698
 699         BUG_ON(!inode);
 700
 701         error = setattr_prepare(dentry, attr);
 702         if (error)
 703                 return error;
 704
 705         if (ia_valid & ATTR_SIZE) {
 706                 loff_t oldsize = inode->i_size;
 707                 loff_t newsize = attr->ia_size;
 708
 709                 if (newsize & ~huge_page_mask(h))
 710                         return -EINVAL;
 711                 /* protected by i_mutex */
 712                 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
 713                     (newsize > oldsize && (info->seals & F_SEAL_GROW)))
 714                         return -EPERM;
 715                 error = hugetlb_vmtruncate(inode, newsize);
 716                 if (error)
 717                         return error;
 718         }
 719
 720         setattr_copy(inode, attr);
 721         mark_inode_dirty(inode);
 722         return 0;
 723 }
 724
 725 static struct inode *hugetlbfs_get_root(struct super_block *sb,
 726                                         struct hugetlbfs_fs_context *ctx)
 727 {
 728         struct inode *inode;
 729
 730         inode = new_inode(sb);
 731         if (inode) {
 732                 inode->i_ino = get_next_ino();
 733                 inode->i_mode = S_IFDIR | ctx->mode;
 734                 inode->i_uid = ctx->uid;
 735                 inode->i_gid = ctx->gid;
 736                 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 737                 inode->i_op = &hugetlbfs_dir_inode_operations;
 738                 inode->i_fop = &simple_dir_operations;
 739                 /* directory inodes start off with i_nlink == 2 (for "." entry) */
 740                 inc_nlink(inode);
 741                 lockdep_annotate_inode_mutex_key(inode);
 742         }
 743         return inode;
 744 }
 745
 746 /*
 747  * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
 748  * be taken from reclaim -- unlike regular filesystems. This needs an
 749  * annotation because huge_pmd_share() does an allocation under hugetlb's
 750  * i_mmap_rwsem.
 751  */
 752 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
 753
 754 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
 755                                         struct inode *dir,
 756                                         umode_t mode, dev_t dev)
 757 {
 758         struct inode *inode;
 759         struct resv_map *resv_map = NULL;
 760
 761         /*
 762          * Reserve maps are only needed for inodes that can have associated
 763          * page allocations.
 764          */
 765         if (S_ISREG(mode) || S_ISLNK(mode)) {
 766                 resv_map = resv_map_alloc();
 767                 if (!resv_map)
 768                         return NULL;
 769         }
 770
 771         inode = new_inode(sb);
 772         if (inode) {
 773                 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
 774
 775                 inode->i_ino = get_next_ino();
 776                 inode_init_owner(inode, dir, mode);
 777                 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
 778                                 &hugetlbfs_i_mmap_rwsem_key);
 779                 inode->i_mapping->a_ops = &hugetlbfs_aops;
 780                 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 781                 inode->i_mapping->private_data = resv_map;
 782                 info->seals = F_SEAL_SEAL;
 783                 switch (mode & S_IFMT) {
 784                 default:
 785                         init_special_inode(inode, mode, dev);
 786                         break;
 787                 case S_IFREG:
 788                         inode->i_op = &hugetlbfs_inode_operations;
 789                         inode->i_fop = &hugetlbfs_file_operations;
 790                         break;
 791                 case S_IFDIR:
 792                         inode->i_op = &hugetlbfs_dir_inode_operations;
 793                         inode->i_fop = &simple_dir_operations;
 794
 795                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
 796                         inc_nlink(inode);
 797                         break;
 798                 case S_IFLNK:
 799                         inode->i_op = &page_symlink_inode_operations;
 800                         inode_nohighmem(inode);
 801                         break;
 802                 }
 803                 lockdep_annotate_inode_mutex_key(inode);
 804         } else {
 805                 if (resv_map)
 806                         kref_put(&resv_map->refs, resv_map_release);
 807         }
 808
 809         return inode;
 810 }
 811
 812 /*
 813  * File creation. Allocate an inode, and we're done..
 814  */
 815 static int do_hugetlbfs_mknod(struct inode *dir,
 816                         struct dentry *dentry,
 817                         umode_t mode,
 818                         dev_t dev,
 819                         bool tmpfile)
 820 {
 821         struct inode *inode;
 822         int error = -ENOSPC;
 823
 824         inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
 825         if (inode) {
 826                 dir->i_ctime = dir->i_mtime = current_time(dir);
 827                 if (tmpfile) {
 828                         d_tmpfile(dentry, inode);
 829                 } else {
 830                         d_instantiate(dentry, inode);
 831                         dget(dentry);/* Extra count - pin the dentry in core */
 832                 }
 833                 error = 0;
 834         }
 835         return error;
 836 }
 837
 838 static int hugetlbfs_mknod(struct inode *dir,
 839                         struct dentry *dentry, umode_t mode, dev_t dev)
 840 {
 841         return do_hugetlbfs_mknod(dir, dentry, mode, dev, false);
 842 }
 843
 844 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 845 {
 846         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 847         if (!retval)
 848                 inc_nlink(dir);
 849         return retval;
 850 }
 851
 852 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
 853 {
 854         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 855 }
 856
 857 static int hugetlbfs_tmpfile(struct inode *dir,
 858                         struct dentry *dentry, umode_t mode)
 859 {
 860         return do_hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0, true);
 861 }
 862
 863 static int hugetlbfs_symlink(struct inode *dir,
 864                         struct dentry *dentry, const char *symname)
 865 {
 866         struct inode *inode;
 867         int error = -ENOSPC;
 868
 869         inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
 870         if (inode) {
 871                 int l = strlen(symname)+1;
 872                 error = page_symlink(inode, symname, l);
 873                 if (!error) {
 874                         d_instantiate(dentry, inode);
 875                         dget(dentry);
 876                 } else
 877                         iput(inode);
 878         }
 879         dir->i_ctime = dir->i_mtime = current_time(dir);
 880
 881         return error;
 882 }
 883
 884 /*
 885  * mark the head page dirty
 886  */
 887 static int hugetlbfs_set_page_dirty(struct page *page)
 888 {
 889         struct page *head = compound_head(page);
 890
 891         SetPageDirty(head);
 892         return 0;
 893 }
 894
 895 static int hugetlbfs_migrate_page(struct address_space *mapping,
 896                                 struct page *newpage, struct page *page,
 897                                 enum migrate_mode mode)
 898 {
 899         int rc;
 900
 901         rc = migrate_huge_page_move_mapping(mapping, newpage, page);
 902         if (rc != MIGRATEPAGE_SUCCESS)
 903                 return rc;
 904
 905         /*
 906          * page_private is subpool pointer in hugetlb pages.  Transfer to
 907          * new page.  PagePrivate is not associated with page_private for
 908          * hugetlb pages and can not be set here as only page_huge_active
 909          * pages can be migrated.
 910          */
 911         if (page_private(page)) {
 912                 set_page_private(newpage, page_private(page));
 913                 set_page_private(page, 0);
 914         }
 915
 916         if (mode != MIGRATE_SYNC_NO_COPY)
 917                 migrate_page_copy(newpage, page);
 918         else
 919                 migrate_page_states(newpage, page);
 920
 921         return MIGRATEPAGE_SUCCESS;
 922 }
 923
 924 static int hugetlbfs_error_remove_page(struct address_space *mapping,
 925                                 struct page *page)
 926 {
 927         struct inode *inode = mapping->host;
 928         pgoff_t index = page->index;
 929
 930         remove_huge_page(page);
 931         if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1)))
 932                 hugetlb_fix_reserve_counts(inode);
 933
 934         return 0;
 935 }
 936
 937 /*
 938  * Display the mount options in /proc/mounts.
 939  */
 940 static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
 941 {
 942         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb);
 943         struct hugepage_subpool *spool = sbinfo->spool;
 944         unsigned long hpage_size = huge_page_size(sbinfo->hstate);
 945         unsigned hpage_shift = huge_page_shift(sbinfo->hstate);
 946         char mod;
 947
 948         if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
 949                 seq_printf(m, ",uid=%u",
 950                            from_kuid_munged(&init_user_ns, sbinfo->uid));
 951         if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
 952                 seq_printf(m, ",gid=%u",
 953                            from_kgid_munged(&init_user_ns, sbinfo->gid));
 954         if (sbinfo->mode != 0755)
 955                 seq_printf(m, ",mode=%o", sbinfo->mode);
 956         if (sbinfo->max_inodes != -1)
 957                 seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
 958
 959         hpage_size /= 1024;
 960         mod = 'K';
 961         if (hpage_size >= 1024) {
 962                 hpage_size /= 1024;
 963                 mod = 'M';
 964         }
 965         seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
 966         if (spool) {
 967                 if (spool->max_hpages != -1)
 968                         seq_printf(m, ",size=%llu",
 969                                    (unsigned long long)spool->max_hpages << hpage_shift);
 970                 if (spool->min_hpages != -1)
 971                         seq_printf(m, ",min_size=%llu",
 972                                    (unsigned long long)spool->min_hpages << hpage_shift);
 973         }
 974         return 0;
 975 }
 976
 977 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 978 {
 979         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 980         struct hstate *h = hstate_inode(d_inode(dentry));
 981
 982         buf->f_type = HUGETLBFS_MAGIC;
 983         buf->f_bsize = huge_page_size(h);
 984         if (sbinfo) {
 985                 spin_lock(&sbinfo->stat_lock);
 986                 /* If no limits set, just report 0 for max/free/used
 987                  * blocks, like simple_statfs() */
 988                 if (sbinfo->spool) {
 989                         long free_pages;
 990
 991                         spin_lock(&sbinfo->spool->lock);
 992                         buf->f_blocks = sbinfo->spool->max_hpages;
 993                         free_pages = sbinfo->spool->max_hpages
 994                                 - sbinfo->spool->used_hpages;
 995                         buf->f_bavail = buf->f_bfree = free_pages;
 996                         spin_unlock(&sbinfo->spool->lock);
 997                         buf->f_files = sbinfo->max_inodes;
 998                         buf->f_ffree = sbinfo->free_inodes;
 999                 }
1000                 spin_unlock(&sbinfo->stat_lock);
1001         }
1002         buf->f_namelen = NAME_MAX;
1003         return 0;
1004 }
1005
1006 static void hugetlbfs_put_super(struct super_block *sb)
1007 {
1008         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
1009
1010         if (sbi) {
1011                 sb->s_fs_info = NULL;
1012
1013                 if (sbi->spool)
1014                         hugepage_put_subpool(sbi->spool);
1015
1016                 kfree(sbi);
1017         }
1018 }
1019
1020 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
1021 {
1022         if (sbinfo->free_inodes >= 0) {
1023                 spin_lock(&sbinfo->stat_lock);
1024                 if (unlikely(!sbinfo->free_inodes)) {
1025                         spin_unlock(&sbinfo->stat_lock);
1026                         return 0;
1027                 }
1028                 sbinfo->free_inodes--;
1029                 spin_unlock(&sbinfo->stat_lock);
1030         }
1031
1032         return 1;
1033 }
1034
1035 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
1036 {
1037         if (sbinfo->free_inodes >= 0) {
1038                 spin_lock(&sbinfo->stat_lock);
1039                 sbinfo->free_inodes++;
1040                 spin_unlock(&sbinfo->stat_lock);
1041         }
1042 }
1043
1044
1045 static struct kmem_cache *hugetlbfs_inode_cachep;
1046
1047 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
1048 {
1049         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
1050         struct hugetlbfs_inode_info *p;
1051
1052         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
1053                 return NULL;
1054         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
1055         if (unlikely(!p)) {
1056                 hugetlbfs_inc_free_inodes(sbinfo);
1057                 return NULL;
1058         }
1059
1060         /*
1061          * Any time after allocation, hugetlbfs_destroy_inode can be called
1062          * for the inode.  mpol_free_shared_policy is unconditionally called
1063          * as part of hugetlbfs_destroy_inode.  So, initialize policy here
1064          * in case of a quick call to destroy.
1065          *
1066          * Note that the policy is initialized even if we are creating a
1067          * private inode.  This simplifies hugetlbfs_destroy_inode.
1068          */
1069         mpol_shared_policy_init(&p->policy, NULL);
1070
1071         return &p->vfs_inode;
1072 }
1073
1074 static void hugetlbfs_free_inode(struct inode *inode)
1075 {
1076         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
1077 }
1078
1079 static void hugetlbfs_destroy_inode(struct inode *inode)
1080 {
1081         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
1082         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
1083 }
1084
1085 static const struct address_space_operations hugetlbfs_aops = {
1086         .write_begin    = hugetlbfs_write_begin,
1087         .write_end      = hugetlbfs_write_end,
1088         .set_page_dirty = hugetlbfs_set_page_dirty,
1089         .migratepage    = hugetlbfs_migrate_page,
1090         .error_remove_page      = hugetlbfs_error_remove_page,
1091 };
1092
1093
1094 static void init_once(void *foo)
1095 {
1096         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
1097
1098         inode_init_once(&ei->vfs_inode);
1099 }
1100
1101 const struct file_operations hugetlbfs_file_operations = {
1102         .read_iter              = hugetlbfs_read_iter,
1103         .mmap                   = hugetlbfs_file_mmap,
1104         .fsync                  = noop_fsync,
1105         .get_unmapped_area      = hugetlb_get_unmapped_area,
1106         .llseek                 = default_llseek,
1107         .fallocate              = hugetlbfs_fallocate,
1108 };
1109
1110 static const struct inode_operations hugetlbfs_dir_inode_operations = {
1111         .create         = hugetlbfs_create,
1112         .lookup         = simple_lookup,
1113         .link           = simple_link,
1114         .unlink         = simple_unlink,
1115         .symlink        = hugetlbfs_symlink,
1116         .mkdir          = hugetlbfs_mkdir,
1117         .rmdir          = simple_rmdir,
1118         .mknod          = hugetlbfs_mknod,
1119         .rename         = simple_rename,
1120         .setattr        = hugetlbfs_setattr,
1121         .tmpfile        = hugetlbfs_tmpfile,
1122 };
1123
1124 static const struct inode_operations hugetlbfs_inode_operations = {
1125         .setattr        = hugetlbfs_setattr,
1126 };
1127
1128 static const struct super_operations hugetlbfs_ops = {
1129         .alloc_inode    = hugetlbfs_alloc_inode,
1130         .free_inode     = hugetlbfs_free_inode,
1131         .destroy_inode  = hugetlbfs_destroy_inode,
1132         .evict_inode    = hugetlbfs_evict_inode,
1133         .statfs         = hugetlbfs_statfs,
1134         .put_super      = hugetlbfs_put_super,
1135         .show_options   = hugetlbfs_show_options,
1136 };
1137
1138 /*
1139  * Convert size option passed from command line to number of huge pages
1140  * in the pool specified by hstate.  Size option could be in bytes
1141  * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
1142  */
1143 static long
1144 hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
1145                          enum hugetlbfs_size_type val_type)
1146 {
1147         if (val_type == NO_SIZE)
1148                 return -1;
1149
1150         if (val_type == SIZE_PERCENT) {
1151                 size_opt <<= huge_page_shift(h);
1152                 size_opt *= h->max_huge_pages;
1153                 do_div(size_opt, 100);
1154         }
1155
1156         size_opt >>= huge_page_shift(h);
1157         return size_opt;
1158 }
1159
1160 /*
1161  * Parse one mount parameter.
1162  */
1163 static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param)
1164 {
1165         struct hugetlbfs_fs_context *ctx = fc->fs_private;
1166         struct fs_parse_result result;
1167         char *rest;
1168         unsigned long ps;
1169         int opt;
1170
1171         opt = fs_parse(fc, hugetlb_fs_parameters, param, &result);
1172         if (opt < 0)
1173                 return opt;
1174
1175         switch (opt) {
1176         case Opt_uid:
1177                 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
1178                 if (!uid_valid(ctx->uid))
1179                         goto bad_val;
1180                 return 0;
1181
1182         case Opt_gid:
1183                 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
1184                 if (!gid_valid(ctx->gid))
1185                         goto bad_val;
1186                 return 0;
1187
1188         case Opt_mode:
1189                 ctx->mode = result.uint_32 & 01777U;
1190                 return 0;
1191
1192         case Opt_size:
1193                 /* memparse() will accept a K/M/G without a digit */
1194                 if (!isdigit(param->string[0]))
1195                         goto bad_val;
1196                 ctx->max_size_opt = memparse(param->string, &rest);
1197                 ctx->max_val_type = SIZE_STD;
1198                 if (*rest == '%')
1199                         ctx->max_val_type = SIZE_PERCENT;
1200                 return 0;
1201
1202         case Opt_nr_inodes:
1203                 /* memparse() will accept a K/M/G without a digit */
1204                 if (!isdigit(param->string[0]))
1205                         goto bad_val;
1206                 ctx->nr_inodes = memparse(param->string, &rest);
1207                 return 0;
1208
1209         case Opt_pagesize:
1210                 ps = memparse(param->string, &rest);
1211                 ctx->hstate = size_to_hstate(ps);
1212                 if (!ctx->hstate) {
1213                         pr_err("Unsupported page size %lu MB\n", ps >> 20);
1214                         return -EINVAL;
1215                 }
1216                 return 0;
1217
1218         case Opt_min_size:
1219                 /* memparse() will accept a K/M/G without a digit */
1220                 if (!isdigit(param->string[0]))
1221                         goto bad_val;
1222                 ctx->min_size_opt = memparse(param->string, &rest);
1223                 ctx->min_val_type = SIZE_STD;
1224                 if (*rest == '%')
1225                         ctx->min_val_type = SIZE_PERCENT;
1226                 return 0;
1227
1228         default:
1229                 return -EINVAL;
1230         }
1231
1232 bad_val:
1233         return invalfc(fc, "Bad value '%s' for mount option '%s'\n",
1234                       param->string, param->key);
1235 }
1236
1237 /*
1238  * Validate the parsed options.
1239  */
1240 static int hugetlbfs_validate(struct fs_context *fc)
1241 {
1242         struct hugetlbfs_fs_context *ctx = fc->fs_private;
1243
1244         /*
1245          * Use huge page pool size (in hstate) to convert the size
1246          * options to number of huge pages.  If NO_SIZE, -1 is returned.
1247          */
1248         ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate,
1249                                                    ctx->max_size_opt,
1250                                                    ctx->max_val_type);
1251         ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate,
1252                                                    ctx->min_size_opt,
1253                                                    ctx->min_val_type);
1254
1255         /*
1256          * If max_size was specified, then min_size must be smaller
1257          */
1258         if (ctx->max_val_type > NO_SIZE &&
1259             ctx->min_hpages > ctx->max_hpages) {
1260                 pr_err("Minimum size can not be greater than maximum size\n");
1261                 return -EINVAL;
1262         }
1263
1264         return 0;
1265 }
1266
1267 static int
1268 hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc)
1269 {
1270         struct hugetlbfs_fs_context *ctx = fc->fs_private;
1271         struct hugetlbfs_sb_info *sbinfo;
1272
1273         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
1274         if (!sbinfo)
1275                 return -ENOMEM;
1276         sb->s_fs_info = sbinfo;
1277         spin_lock_init(&sbinfo->stat_lock);
1278         sbinfo->hstate          = ctx->hstate;
1279         sbinfo->max_inodes      = ctx->nr_inodes;
1280         sbinfo->free_inodes     = ctx->nr_inodes;
1281         sbinfo->spool           = NULL;
1282         sbinfo->uid             = ctx->uid;
1283         sbinfo->gid             = ctx->gid;
1284         sbinfo->mode            = ctx->mode;
1285
1286         /*
1287          * Allocate and initialize subpool if maximum or minimum size is
1288          * specified.  Any needed reservations (for minimim size) are taken
1289          * taken when the subpool is created.
1290          */
1291         if (ctx->max_hpages != -1 || ctx->min_hpages != -1) {
1292                 sbinfo->spool = hugepage_new_subpool(ctx->hstate,
1293                                                      ctx->max_hpages,
1294                                                      ctx->min_hpages);
1295                 if (!sbinfo->spool)
1296                         goto out_free;
1297         }
1298         sb->s_maxbytes = MAX_LFS_FILESIZE;
1299         sb->s_blocksize = huge_page_size(ctx->hstate);
1300         sb->s_blocksize_bits = huge_page_shift(ctx->hstate);
1301         sb->s_magic = HUGETLBFS_MAGIC;
1302         sb->s_op = &hugetlbfs_ops;
1303         sb->s_time_gran = 1;
1304         sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx));
1305         if (!sb->s_root)
1306                 goto out_free;
1307         return 0;
1308 out_free:
1309         kfree(sbinfo->spool);
1310         kfree(sbinfo);
1311         return -ENOMEM;
1312 }
1313
1314 static int hugetlbfs_get_tree(struct fs_context *fc)
1315 {
1316         int err = hugetlbfs_validate(fc);
1317         if (err)
1318                 return err;
1319         return get_tree_nodev(fc, hugetlbfs_fill_super);
1320 }
1321
1322 static void hugetlbfs_fs_context_free(struct fs_context *fc)
1323 {
1324         kfree(fc->fs_private);
1325 }
1326
1327 static const struct fs_context_operations hugetlbfs_fs_context_ops = {
1328         .free           = hugetlbfs_fs_context_free,
1329         .parse_param    = hugetlbfs_parse_param,
1330         .get_tree       = hugetlbfs_get_tree,
1331 };
1332
1333 static int hugetlbfs_init_fs_context(struct fs_context *fc)
1334 {
1335         struct hugetlbfs_fs_context *ctx;
1336
1337         ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL);
1338         if (!ctx)
1339                 return -ENOMEM;
1340
1341         ctx->max_hpages = -1; /* No limit on size by default */
1342         ctx->nr_inodes  = -1; /* No limit on number of inodes by default */
1343         ctx->uid        = current_fsuid();
1344         ctx->gid        = current_fsgid();
1345         ctx->mode       = 0755;
1346         ctx->hstate     = &default_hstate;
1347         ctx->min_hpages = -1; /* No default minimum size */
1348         ctx->max_val_type = NO_SIZE;
1349         ctx->min_val_type = NO_SIZE;
1350         fc->fs_private = ctx;
1351         fc->ops = &hugetlbfs_fs_context_ops;
1352         return 0;
1353 }
1354
1355 static struct file_system_type hugetlbfs_fs_type = {
1356         .name                   = "hugetlbfs",
1357         .init_fs_context        = hugetlbfs_init_fs_context,
1358         .parameters             = hugetlb_fs_parameters,
1359         .kill_sb                = kill_litter_super,
1360 };
1361
1362 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
1363
1364 static int can_do_hugetlb_shm(void)
1365 {
1366         kgid_t shm_group;
1367         shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
1368         return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
1369 }
1370
1371 static int get_hstate_idx(int page_size_log)
1372 {
1373         struct hstate *h = hstate_sizelog(page_size_log);
1374
1375         if (!h)
1376                 return -1;
1377         return h - hstates;
1378 }
1379
1380 /*
1381  * Note that size should be aligned to proper hugepage size in caller side,
1382  * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
1383  */
1384 struct file *hugetlb_file_setup(const char *name, size_t size,
1385                                 vm_flags_t acctflag, struct user_struct **user,
1386                                 int creat_flags, int page_size_log)
1387 {
1388         struct inode *inode;
1389         struct vfsmount *mnt;
1390         int hstate_idx;
1391         struct file *file;
1392
1393         hstate_idx = get_hstate_idx(page_size_log);
1394         if (hstate_idx < 0)
1395                 return ERR_PTR(-ENODEV);
1396
1397         *user = NULL;
1398         mnt = hugetlbfs_vfsmount[hstate_idx];
1399         if (!mnt)
1400                 return ERR_PTR(-ENOENT);
1401
1402         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
1403                 *user = current_user();
1404                 if (user_shm_lock(size, *user)) {
1405                         task_lock(current);
1406                         pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
1407                                 current->comm, current->pid);
1408                         task_unlock(current);
1409                 } else {
1410                         *user = NULL;
1411                         return ERR_PTR(-EPERM);
1412                 }
1413         }
1414
1415         file = ERR_PTR(-ENOSPC);
1416         inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0);
1417         if (!inode)
1418                 goto out;
1419         if (creat_flags == HUGETLB_SHMFS_INODE)
1420                 inode->i_flags |= S_PRIVATE;
1421
1422         inode->i_size = size;
1423         clear_nlink(inode);
1424
1425         if (hugetlb_reserve_pages(inode, 0,
1426                         size >> huge_page_shift(hstate_inode(inode)), NULL,
1427                         acctflag))
1428                 file = ERR_PTR(-ENOMEM);
1429         else
1430                 file = alloc_file_pseudo(inode, mnt, name, O_RDWR,
1431                                         &hugetlbfs_file_operations);
1432         if (!IS_ERR(file))
1433                 return file;
1434
1435         iput(inode);
1436 out:
1437         if (*user) {
1438                 user_shm_unlock(size, *user);
1439                 *user = NULL;
1440         }
1441         return file;
1442 }
1443
1444 static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h)
1445 {
1446         struct fs_context *fc;
1447         struct vfsmount *mnt;
1448
1449         fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT);
1450         if (IS_ERR(fc)) {
1451                 mnt = ERR_CAST(fc);
1452         } else {
1453                 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1454                 ctx->hstate = h;
1455                 mnt = fc_mount(fc);
1456                 put_fs_context(fc);
1457         }
1458         if (IS_ERR(mnt))
1459                 pr_err("Cannot mount internal hugetlbfs for page size %uK",
1460                        1U << (h->order + PAGE_SHIFT - 10));
1461         return mnt;
1462 }
1463
1464 static int __init init_hugetlbfs_fs(void)
1465 {
1466         struct vfsmount *mnt;
1467         struct hstate *h;
1468         int error;
1469         int i;
1470
1471         if (!hugepages_supported()) {
1472                 pr_info("disabling because there are no supported hugepage sizes\n");
1473                 return -ENOTSUPP;
1474         }
1475
1476         error = -ENOMEM;
1477         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1478                                         sizeof(struct hugetlbfs_inode_info),
1479                                         0, SLAB_ACCOUNT, init_once);
1480         if (hugetlbfs_inode_cachep == NULL)
1481                 goto out;
1482
1483         error = register_filesystem(&hugetlbfs_fs_type);
1484         if (error)
1485                 goto out_free;
1486
1487         /* default hstate mount is required */
1488         mnt = mount_one_hugetlbfs(&hstates[default_hstate_idx]);
1489         if (IS_ERR(mnt)) {
1490                 error = PTR_ERR(mnt);
1491                 goto out_unreg;
1492         }
1493         hugetlbfs_vfsmount[default_hstate_idx] = mnt;
1494
1495         /* other hstates are optional */
1496         i = 0;
1497         for_each_hstate(h) {
1498                 if (i == default_hstate_idx) {
1499                         i++;
1500                         continue;
1501                 }
1502
1503                 mnt = mount_one_hugetlbfs(h);
1504                 if (IS_ERR(mnt))
1505                         hugetlbfs_vfsmount[i] = NULL;
1506                 else
1507                         hugetlbfs_vfsmount[i] = mnt;
1508                 i++;
1509         }
1510
1511         return 0;
1512
1513  out_unreg:
1514         (void)unregister_filesystem(&hugetlbfs_fs_type);
1515  out_free:
1516         kmem_cache_destroy(hugetlbfs_inode_cachep);
1517  out:
1518         return error;
1519 }
1520 fs_initcall(init_hugetlbfs_fs)