2 * hugetlbpage-backed filesystem. Based on ramfs.
4 * Nadia Yvette Chambers, 2002
6 * Copyright (C) 2002 Linus Torvalds.
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/thread_info.h>
13 #include <asm/current.h>
14 #include <linux/falloc.h>
16 #include <linux/mount.h>
17 #include <linux/file.h>
18 #include <linux/kernel.h>
19 #include <linux/writeback.h>
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/init.h>
23 #include <linux/string.h>
24 #include <linux/capability.h>
25 #include <linux/ctype.h>
26 #include <linux/backing-dev.h>
27 #include <linux/hugetlb.h>
28 #include <linux/pagevec.h>
29 #include <linux/fs_parser.h>
30 #include <linux/mman.h>
31 #include <linux/slab.h>
32 #include <linux/dnotify.h>
33 #include <linux/statfs.h>
34 #include <linux/security.h>
35 #include <linux/magic.h>
36 #include <linux/migrate.h>
37 #include <linux/uio.h>
39 #include <linux/uaccess.h>
40 #include <linux/sched/mm.h>
42 static const struct address_space_operations hugetlbfs_aops;
43 const struct file_operations hugetlbfs_file_operations;
44 static const struct inode_operations hugetlbfs_dir_inode_operations;
45 static const struct inode_operations hugetlbfs_inode_operations;
47 enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT };
49 struct hugetlbfs_fs_context {
50 struct hstate *hstate;
51 unsigned long long max_size_opt;
52 unsigned long long min_size_opt;
56 enum hugetlbfs_size_type max_val_type;
57 enum hugetlbfs_size_type min_val_type;
63 int sysctl_hugetlb_shm_group;
75 static const struct fs_parameter_spec hugetlb_fs_parameters[] = {
76 fsparam_u32 ("gid", Opt_gid),
77 fsparam_string("min_size", Opt_min_size),
78 fsparam_u32oct("mode", Opt_mode),
79 fsparam_string("nr_inodes", Opt_nr_inodes),
80 fsparam_string("pagesize", Opt_pagesize),
81 fsparam_string("size", Opt_size),
82 fsparam_u32 ("uid", Opt_uid),
87 static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
88 struct inode *inode, pgoff_t index)
90 vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy,
94 static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
96 mpol_cond_put(vma->vm_policy);
99 static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
100 struct inode *inode, pgoff_t index)
104 static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
110 * Mask used when checking the page offset value passed in via system
111 * calls. This value will be converted to a loff_t which is signed.
112 * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the
113 * value. The extra bit (- 1 in the shift value) is to take the sign
116 #define PGOFF_LOFFT_MAX \
117 (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1)))
119 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
121 struct inode *inode = file_inode(file);
122 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
125 struct hstate *h = hstate_file(file);
129 * vma address alignment (but not the pgoff alignment) has
130 * already been checked by prepare_hugepage_range. If you add
131 * any error returns here, do so after setting VM_HUGETLB, so
132 * is_vm_hugetlb_page tests below unmap_region go the right
133 * way when do_mmap unwinds (may be important on powerpc
136 vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
137 vma->vm_ops = &hugetlb_vm_ops;
139 ret = seal_check_future_write(info->seals, vma);
144 * page based offset in vm_pgoff could be sufficiently large to
145 * overflow a loff_t when converted to byte offset. This can
146 * only happen on architectures where sizeof(loff_t) ==
147 * sizeof(unsigned long). So, only check in those instances.
149 if (sizeof(unsigned long) == sizeof(loff_t)) {
150 if (vma->vm_pgoff & PGOFF_LOFFT_MAX)
154 /* must be huge page aligned */
155 if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
158 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
159 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
160 /* check for overflow */
169 vm_flags = vma->vm_flags;
171 * for SHM_HUGETLB, the pages are reserved in the shmget() call so skip
172 * reserving here. Note: only for SHM hugetlbfs file, the inode
173 * flag S_PRIVATE is set.
175 if (inode->i_flags & S_PRIVATE)
176 vm_flags |= VM_NORESERVE;
178 if (!hugetlb_reserve_pages(inode,
179 vma->vm_pgoff >> huge_page_order(h),
180 len >> huge_page_shift(h), vma,
185 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
186 i_size_write(inode, len);
194 * Called under mmap_write_lock(mm).
198 hugetlb_get_unmapped_area_bottomup(struct file *file, unsigned long addr,
199 unsigned long len, unsigned long pgoff, unsigned long flags)
201 struct hstate *h = hstate_file(file);
202 struct vm_unmapped_area_info info;
206 info.low_limit = current->mm->mmap_base;
207 info.high_limit = arch_get_mmap_end(addr, len, flags);
208 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
209 info.align_offset = 0;
210 return vm_unmapped_area(&info);
214 hugetlb_get_unmapped_area_topdown(struct file *file, unsigned long addr,
215 unsigned long len, unsigned long pgoff, unsigned long flags)
217 struct hstate *h = hstate_file(file);
218 struct vm_unmapped_area_info info;
220 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
222 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
223 info.high_limit = arch_get_mmap_base(addr, current->mm->mmap_base);
224 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
225 info.align_offset = 0;
226 addr = vm_unmapped_area(&info);
229 * A failed mmap() very likely causes application failure,
230 * so fall back to the bottom-up function here. This scenario
231 * can happen with large stack limits and large mmap()
234 if (unlikely(offset_in_page(addr))) {
235 VM_BUG_ON(addr != -ENOMEM);
237 info.low_limit = current->mm->mmap_base;
238 info.high_limit = arch_get_mmap_end(addr, len, flags);
239 addr = vm_unmapped_area(&info);
246 generic_hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
247 unsigned long len, unsigned long pgoff,
250 struct mm_struct *mm = current->mm;
251 struct vm_area_struct *vma;
252 struct hstate *h = hstate_file(file);
253 const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags);
255 if (len & ~huge_page_mask(h))
260 if (flags & MAP_FIXED) {
261 if (prepare_hugepage_range(file, addr, len))
267 addr = ALIGN(addr, huge_page_size(h));
268 vma = find_vma(mm, addr);
269 if (mmap_end - len >= addr &&
270 (!vma || addr + len <= vm_start_gap(vma)))
275 * Use mm->get_unmapped_area value as a hint to use topdown routine.
276 * If architectures have special needs, they should define their own
277 * version of hugetlb_get_unmapped_area.
279 if (mm->get_unmapped_area == arch_get_unmapped_area_topdown)
280 return hugetlb_get_unmapped_area_topdown(file, addr, len,
282 return hugetlb_get_unmapped_area_bottomup(file, addr, len,
286 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
288 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
289 unsigned long len, unsigned long pgoff,
292 return generic_hugetlb_get_unmapped_area(file, addr, len, pgoff, flags);
297 * Support for read() - Find the page attached to f_mapping and copy out the
298 * data. This provides functionality similar to filemap_read().
300 static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
302 struct file *file = iocb->ki_filp;
303 struct hstate *h = hstate_file(file);
304 struct address_space *mapping = file->f_mapping;
305 struct inode *inode = mapping->host;
306 unsigned long index = iocb->ki_pos >> huge_page_shift(h);
307 unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
308 unsigned long end_index;
312 while (iov_iter_count(to)) {
316 /* nr is the maximum number of bytes to copy from this page */
317 nr = huge_page_size(h);
318 isize = i_size_read(inode);
321 end_index = (isize - 1) >> huge_page_shift(h);
322 if (index > end_index)
324 if (index == end_index) {
325 nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
332 page = find_lock_page(mapping, index);
333 if (unlikely(page == NULL)) {
335 * We have a HOLE, zero out the user-buffer for the
336 * length of the hole or request.
338 copied = iov_iter_zero(nr, to);
342 if (PageHWPoison(page)) {
349 * We have the page, copy it to user space buffer.
351 copied = copy_page_to_iter(page, offset, nr, to);
356 if (copied != nr && iov_iter_count(to)) {
361 index += offset >> huge_page_shift(h);
362 offset &= ~huge_page_mask(h);
364 iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
368 static int hugetlbfs_write_begin(struct file *file,
369 struct address_space *mapping,
370 loff_t pos, unsigned len,
371 struct page **pagep, void **fsdata)
376 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
377 loff_t pos, unsigned len, unsigned copied,
378 struct page *page, void *fsdata)
384 static void hugetlb_delete_from_page_cache(struct page *page)
386 ClearPageDirty(page);
387 ClearPageUptodate(page);
388 delete_from_page_cache(page);
392 * Called with i_mmap_rwsem held for inode based vma maps. This makes
393 * sure vma (and vm_mm) will not go away. We also hold the hugetlb fault
394 * mutex for the page in the mapping. So, we can not race with page being
395 * faulted into the vma.
397 static bool hugetlb_vma_maps_page(struct vm_area_struct *vma,
398 unsigned long addr, struct page *page)
402 ptep = huge_pte_offset(vma->vm_mm, addr,
403 huge_page_size(hstate_vma(vma)));
408 pte = huge_ptep_get(ptep);
409 if (huge_pte_none(pte) || !pte_present(pte))
412 if (pte_page(pte) == page)
419 * Can vma_offset_start/vma_offset_end overflow on 32-bit arches?
420 * No, because the interval tree returns us only those vmas
421 * which overlap the truncated area starting at pgoff,
422 * and no vma on a 32-bit arch can span beyond the 4GB.
424 static unsigned long vma_offset_start(struct vm_area_struct *vma, pgoff_t start)
426 if (vma->vm_pgoff < start)
427 return (start - vma->vm_pgoff) << PAGE_SHIFT;
432 static unsigned long vma_offset_end(struct vm_area_struct *vma, pgoff_t end)
439 t_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) + vma->vm_start;
440 if (t_end > vma->vm_end)
446 * Called with hugetlb fault mutex held. Therefore, no more mappings to
447 * this folio can be created while executing the routine.
449 static void hugetlb_unmap_file_folio(struct hstate *h,
450 struct address_space *mapping,
451 struct folio *folio, pgoff_t index)
453 struct rb_root_cached *root = &mapping->i_mmap;
454 struct hugetlb_vma_lock *vma_lock;
455 struct page *page = &folio->page;
456 struct vm_area_struct *vma;
457 unsigned long v_start;
461 start = index * pages_per_huge_page(h);
462 end = (index + 1) * pages_per_huge_page(h);
464 i_mmap_lock_write(mapping);
467 vma_interval_tree_foreach(vma, root, start, end - 1) {
468 v_start = vma_offset_start(vma, start);
469 v_end = vma_offset_end(vma, end);
471 if (!hugetlb_vma_maps_page(vma, vma->vm_start + v_start, page))
474 if (!hugetlb_vma_trylock_write(vma)) {
475 vma_lock = vma->vm_private_data;
477 * If we can not get vma lock, we need to drop
478 * immap_sema and take locks in order. First,
479 * take a ref on the vma_lock structure so that
480 * we can be guaranteed it will not go away when
481 * dropping immap_sema.
483 kref_get(&vma_lock->refs);
487 unmap_hugepage_range(vma, vma->vm_start + v_start, v_end,
488 NULL, ZAP_FLAG_DROP_MARKER);
489 hugetlb_vma_unlock_write(vma);
492 i_mmap_unlock_write(mapping);
496 * Wait on vma_lock. We know it is still valid as we have
497 * a reference. We must 'open code' vma locking as we do
498 * not know if vma_lock is still attached to vma.
500 down_write(&vma_lock->rw_sema);
501 i_mmap_lock_write(mapping);
506 * If lock is no longer attached to vma, then just
507 * unlock, drop our reference and retry looking for
510 up_write(&vma_lock->rw_sema);
511 kref_put(&vma_lock->refs, hugetlb_vma_lock_release);
516 * vma_lock is still attached to vma. Check to see if vma
517 * still maps page and if so, unmap.
519 v_start = vma_offset_start(vma, start);
520 v_end = vma_offset_end(vma, end);
521 if (hugetlb_vma_maps_page(vma, vma->vm_start + v_start, page))
522 unmap_hugepage_range(vma, vma->vm_start + v_start,
524 ZAP_FLAG_DROP_MARKER);
526 kref_put(&vma_lock->refs, hugetlb_vma_lock_release);
527 hugetlb_vma_unlock_write(vma);
534 hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end,
535 zap_flags_t zap_flags)
537 struct vm_area_struct *vma;
540 * end == 0 indicates that the entire range after start should be
541 * unmapped. Note, end is exclusive, whereas the interval tree takes
542 * an inclusive "last".
544 vma_interval_tree_foreach(vma, root, start, end ? end - 1 : ULONG_MAX) {
545 unsigned long v_start;
548 if (!hugetlb_vma_trylock_write(vma))
551 v_start = vma_offset_start(vma, start);
552 v_end = vma_offset_end(vma, end);
554 unmap_hugepage_range(vma, vma->vm_start + v_start, v_end,
558 * Note that vma lock only exists for shared/non-private
559 * vmas. Therefore, lock is not held when calling
560 * unmap_hugepage_range for private vmas.
562 hugetlb_vma_unlock_write(vma);
567 * Called with hugetlb fault mutex held.
568 * Returns true if page was actually removed, false otherwise.
570 static bool remove_inode_single_folio(struct hstate *h, struct inode *inode,
571 struct address_space *mapping,
572 struct folio *folio, pgoff_t index,
578 * If folio is mapped, it was faulted in after being
579 * unmapped in caller. Unmap (again) while holding
580 * the fault mutex. The mutex will prevent faults
581 * until we finish removing the folio.
583 if (unlikely(folio_mapped(folio)))
584 hugetlb_unmap_file_folio(h, mapping, folio, index);
588 * We must remove the folio from page cache before removing
589 * the region/ reserve map (hugetlb_unreserve_pages). In
590 * rare out of memory conditions, removal of the region/reserve
591 * map could fail. Correspondingly, the subpool and global
592 * reserve usage count can need to be adjusted.
594 VM_BUG_ON(HPageRestoreReserve(&folio->page));
595 hugetlb_delete_from_page_cache(&folio->page);
598 if (unlikely(hugetlb_unreserve_pages(inode, index,
600 hugetlb_fix_reserve_counts(inode);
608 * remove_inode_hugepages handles two distinct cases: truncation and hole
609 * punch. There are subtle differences in operation for each case.
611 * truncation is indicated by end of range being LLONG_MAX
612 * In this case, we first scan the range and release found pages.
613 * After releasing pages, hugetlb_unreserve_pages cleans up region/reserve
614 * maps and global counts. Page faults can race with truncation.
615 * During faults, hugetlb_no_page() checks i_size before page allocation,
616 * and again after obtaining page table lock. It will 'back out'
617 * allocations in the truncated range.
618 * hole punch is indicated if end is not LLONG_MAX
619 * In the hole punch case we scan the range and release found pages.
620 * Only when releasing a page is the associated region/reserve map
621 * deleted. The region/reserve map for ranges without associated
622 * pages are not modified. Page faults can race with hole punch.
623 * This is indicated if we find a mapped page.
624 * Note: If the passed end of range value is beyond the end of file, but
625 * not LLONG_MAX this routine still performs a hole punch operation.
627 static void remove_inode_hugepages(struct inode *inode, loff_t lstart,
630 struct hstate *h = hstate_inode(inode);
631 struct address_space *mapping = &inode->i_data;
632 const pgoff_t start = lstart >> huge_page_shift(h);
633 const pgoff_t end = lend >> huge_page_shift(h);
634 struct folio_batch fbatch;
637 bool truncate_op = (lend == LLONG_MAX);
639 folio_batch_init(&fbatch);
641 while (filemap_get_folios(mapping, &next, end - 1, &fbatch)) {
642 for (i = 0; i < folio_batch_count(&fbatch); ++i) {
643 struct folio *folio = fbatch.folios[i];
646 index = folio->index;
647 hash = hugetlb_fault_mutex_hash(mapping, index);
648 mutex_lock(&hugetlb_fault_mutex_table[hash]);
651 * Remove folio that was part of folio_batch.
653 if (remove_inode_single_folio(h, inode, mapping, folio,
657 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
659 folio_batch_release(&fbatch);
664 (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed);
667 static void hugetlbfs_evict_inode(struct inode *inode)
669 struct resv_map *resv_map;
671 remove_inode_hugepages(inode, 0, LLONG_MAX);
674 * Get the resv_map from the address space embedded in the inode.
675 * This is the address space which points to any resv_map allocated
676 * at inode creation time. If this is a device special inode,
677 * i_mapping may not point to the original address space.
679 resv_map = (struct resv_map *)(&inode->i_data)->private_data;
680 /* Only regular and link inodes have associated reserve maps */
682 resv_map_release(&resv_map->refs);
686 static void hugetlb_vmtruncate(struct inode *inode, loff_t offset)
689 struct address_space *mapping = inode->i_mapping;
690 struct hstate *h = hstate_inode(inode);
692 BUG_ON(offset & ~huge_page_mask(h));
693 pgoff = offset >> PAGE_SHIFT;
695 i_size_write(inode, offset);
696 i_mmap_lock_write(mapping);
697 if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
698 hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0,
699 ZAP_FLAG_DROP_MARKER);
700 i_mmap_unlock_write(mapping);
701 remove_inode_hugepages(inode, offset, LLONG_MAX);
704 static void hugetlbfs_zero_partial_page(struct hstate *h,
705 struct address_space *mapping,
709 pgoff_t idx = start >> huge_page_shift(h);
712 folio = filemap_lock_folio(mapping, idx);
716 start = start & ~huge_page_mask(h);
717 end = end & ~huge_page_mask(h);
719 end = huge_page_size(h);
721 folio_zero_segment(folio, (size_t)start, (size_t)end);
727 static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
729 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
730 struct address_space *mapping = inode->i_mapping;
731 struct hstate *h = hstate_inode(inode);
732 loff_t hpage_size = huge_page_size(h);
733 loff_t hole_start, hole_end;
736 * hole_start and hole_end indicate the full pages within the hole.
738 hole_start = round_up(offset, hpage_size);
739 hole_end = round_down(offset + len, hpage_size);
743 /* protected by i_rwsem */
744 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
749 i_mmap_lock_write(mapping);
751 /* If range starts before first full page, zero partial page. */
752 if (offset < hole_start)
753 hugetlbfs_zero_partial_page(h, mapping,
754 offset, min(offset + len, hole_start));
756 /* Unmap users of full pages in the hole. */
757 if (hole_end > hole_start) {
758 if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
759 hugetlb_vmdelete_list(&mapping->i_mmap,
760 hole_start >> PAGE_SHIFT,
761 hole_end >> PAGE_SHIFT, 0);
764 /* If range extends beyond last full page, zero partial page. */
765 if ((offset + len) > hole_end && (offset + len) > hole_start)
766 hugetlbfs_zero_partial_page(h, mapping,
767 hole_end, offset + len);
769 i_mmap_unlock_write(mapping);
771 /* Remove full pages from the file. */
772 if (hole_end > hole_start)
773 remove_inode_hugepages(inode, hole_start, hole_end);
780 static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
783 struct inode *inode = file_inode(file);
784 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
785 struct address_space *mapping = inode->i_mapping;
786 struct hstate *h = hstate_inode(inode);
787 struct vm_area_struct pseudo_vma;
788 struct mm_struct *mm = current->mm;
789 loff_t hpage_size = huge_page_size(h);
790 unsigned long hpage_shift = huge_page_shift(h);
791 pgoff_t start, index, end;
795 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
798 if (mode & FALLOC_FL_PUNCH_HOLE)
799 return hugetlbfs_punch_hole(inode, offset, len);
802 * Default preallocate case.
803 * For this range, start is rounded down and end is rounded up
804 * as well as being converted to page offsets.
806 start = offset >> hpage_shift;
807 end = (offset + len + hpage_size - 1) >> hpage_shift;
811 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
812 error = inode_newsize_ok(inode, offset + len);
816 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
822 * Initialize a pseudo vma as this is required by the huge page
823 * allocation routines. If NUMA is configured, use page index
824 * as input to create an allocation policy.
826 vma_init(&pseudo_vma, mm);
827 pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
828 pseudo_vma.vm_file = file;
830 for (index = start; index < end; index++) {
832 * This is supposed to be the vaddr where the page is being
833 * faulted in, but we have no vaddr here.
841 * fallocate(2) manpage permits EINTR; we may have been
842 * interrupted because we are using up too much memory.
844 if (signal_pending(current)) {
849 /* Set numa allocation policy based on index */
850 hugetlb_set_vma_policy(&pseudo_vma, inode, index);
852 /* addr is the offset within the file (zero based) */
853 addr = index * hpage_size;
855 /* mutex taken here, fault path and hole punch */
856 hash = hugetlb_fault_mutex_hash(mapping, index);
857 mutex_lock(&hugetlb_fault_mutex_table[hash]);
859 /* See if already present in mapping to avoid alloc/free */
860 page = find_get_page(mapping, index);
863 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
864 hugetlb_drop_vma_policy(&pseudo_vma);
869 * Allocate page without setting the avoid_reserve argument.
870 * There certainly are no reserves associated with the
871 * pseudo_vma. However, there could be shared mappings with
872 * reserves for the file at the inode level. If we fallocate
873 * pages in these areas, we need to consume the reserves
874 * to keep reservation accounting consistent.
876 page = alloc_huge_page(&pseudo_vma, addr, 0);
877 hugetlb_drop_vma_policy(&pseudo_vma);
879 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
880 error = PTR_ERR(page);
883 clear_huge_page(page, addr, pages_per_huge_page(h));
884 __SetPageUptodate(page);
885 error = hugetlb_add_to_page_cache(page, mapping, index);
886 if (unlikely(error)) {
887 restore_reserve_on_error(h, &pseudo_vma, addr, page);
889 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
893 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
895 SetHPageMigratable(page);
897 * unlock_page because locked by hugetlb_add_to_page_cache()
898 * put_page() due to reference from alloc_huge_page()
904 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
905 i_size_write(inode, offset + len);
906 inode->i_ctime = current_time(inode);
912 static int hugetlbfs_setattr(struct user_namespace *mnt_userns,
913 struct dentry *dentry, struct iattr *attr)
915 struct inode *inode = d_inode(dentry);
916 struct hstate *h = hstate_inode(inode);
918 unsigned int ia_valid = attr->ia_valid;
919 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
921 error = setattr_prepare(&init_user_ns, dentry, attr);
925 if (ia_valid & ATTR_SIZE) {
926 loff_t oldsize = inode->i_size;
927 loff_t newsize = attr->ia_size;
929 if (newsize & ~huge_page_mask(h))
931 /* protected by i_rwsem */
932 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
933 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
935 hugetlb_vmtruncate(inode, newsize);
938 setattr_copy(&init_user_ns, inode, attr);
939 mark_inode_dirty(inode);
943 static struct inode *hugetlbfs_get_root(struct super_block *sb,
944 struct hugetlbfs_fs_context *ctx)
948 inode = new_inode(sb);
950 inode->i_ino = get_next_ino();
951 inode->i_mode = S_IFDIR | ctx->mode;
952 inode->i_uid = ctx->uid;
953 inode->i_gid = ctx->gid;
954 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
955 inode->i_op = &hugetlbfs_dir_inode_operations;
956 inode->i_fop = &simple_dir_operations;
957 /* directory inodes start off with i_nlink == 2 (for "." entry) */
959 lockdep_annotate_inode_mutex_key(inode);
965 * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
966 * be taken from reclaim -- unlike regular filesystems. This needs an
967 * annotation because huge_pmd_share() does an allocation under hugetlb's
970 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
972 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
974 umode_t mode, dev_t dev)
977 struct resv_map *resv_map = NULL;
980 * Reserve maps are only needed for inodes that can have associated
983 if (S_ISREG(mode) || S_ISLNK(mode)) {
984 resv_map = resv_map_alloc();
989 inode = new_inode(sb);
991 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
993 inode->i_ino = get_next_ino();
994 inode_init_owner(&init_user_ns, inode, dir, mode);
995 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
996 &hugetlbfs_i_mmap_rwsem_key);
997 inode->i_mapping->a_ops = &hugetlbfs_aops;
998 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
999 inode->i_mapping->private_data = resv_map;
1000 info->seals = F_SEAL_SEAL;
1001 switch (mode & S_IFMT) {
1003 init_special_inode(inode, mode, dev);
1006 inode->i_op = &hugetlbfs_inode_operations;
1007 inode->i_fop = &hugetlbfs_file_operations;
1010 inode->i_op = &hugetlbfs_dir_inode_operations;
1011 inode->i_fop = &simple_dir_operations;
1013 /* directory inodes start off with i_nlink == 2 (for "." entry) */
1017 inode->i_op = &page_symlink_inode_operations;
1018 inode_nohighmem(inode);
1021 lockdep_annotate_inode_mutex_key(inode);
1024 kref_put(&resv_map->refs, resv_map_release);
1031 * File creation. Allocate an inode, and we're done..
1033 static int hugetlbfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1034 struct dentry *dentry, umode_t mode, dev_t dev)
1036 struct inode *inode;
1038 inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
1041 dir->i_ctime = dir->i_mtime = current_time(dir);
1042 d_instantiate(dentry, inode);
1043 dget(dentry);/* Extra count - pin the dentry in core */
1047 static int hugetlbfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1048 struct dentry *dentry, umode_t mode)
1050 int retval = hugetlbfs_mknod(&init_user_ns, dir, dentry,
1057 static int hugetlbfs_create(struct user_namespace *mnt_userns,
1058 struct inode *dir, struct dentry *dentry,
1059 umode_t mode, bool excl)
1061 return hugetlbfs_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
1064 static int hugetlbfs_tmpfile(struct user_namespace *mnt_userns,
1065 struct inode *dir, struct file *file,
1068 struct inode *inode;
1070 inode = hugetlbfs_get_inode(dir->i_sb, dir, mode | S_IFREG, 0);
1073 dir->i_ctime = dir->i_mtime = current_time(dir);
1074 d_tmpfile(file, inode);
1075 return finish_open_simple(file, 0);
1078 static int hugetlbfs_symlink(struct user_namespace *mnt_userns,
1079 struct inode *dir, struct dentry *dentry,
1080 const char *symname)
1082 struct inode *inode;
1083 int error = -ENOSPC;
1085 inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
1087 int l = strlen(symname)+1;
1088 error = page_symlink(inode, symname, l);
1090 d_instantiate(dentry, inode);
1095 dir->i_ctime = dir->i_mtime = current_time(dir);
1100 #ifdef CONFIG_MIGRATION
1101 static int hugetlbfs_migrate_folio(struct address_space *mapping,
1102 struct folio *dst, struct folio *src,
1103 enum migrate_mode mode)
1107 rc = migrate_huge_page_move_mapping(mapping, dst, src);
1108 if (rc != MIGRATEPAGE_SUCCESS)
1111 if (hugetlb_page_subpool(&src->page)) {
1112 hugetlb_set_page_subpool(&dst->page,
1113 hugetlb_page_subpool(&src->page));
1114 hugetlb_set_page_subpool(&src->page, NULL);
1117 if (mode != MIGRATE_SYNC_NO_COPY)
1118 folio_migrate_copy(dst, src);
1120 folio_migrate_flags(dst, src);
1122 return MIGRATEPAGE_SUCCESS;
1125 #define hugetlbfs_migrate_folio NULL
1128 static int hugetlbfs_error_remove_page(struct address_space *mapping,
1135 * Display the mount options in /proc/mounts.
1137 static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
1139 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb);
1140 struct hugepage_subpool *spool = sbinfo->spool;
1141 unsigned long hpage_size = huge_page_size(sbinfo->hstate);
1142 unsigned hpage_shift = huge_page_shift(sbinfo->hstate);
1145 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
1146 seq_printf(m, ",uid=%u",
1147 from_kuid_munged(&init_user_ns, sbinfo->uid));
1148 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
1149 seq_printf(m, ",gid=%u",
1150 from_kgid_munged(&init_user_ns, sbinfo->gid));
1151 if (sbinfo->mode != 0755)
1152 seq_printf(m, ",mode=%o", sbinfo->mode);
1153 if (sbinfo->max_inodes != -1)
1154 seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
1158 if (hpage_size >= 1024) {
1162 seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
1164 if (spool->max_hpages != -1)
1165 seq_printf(m, ",size=%llu",
1166 (unsigned long long)spool->max_hpages << hpage_shift);
1167 if (spool->min_hpages != -1)
1168 seq_printf(m, ",min_size=%llu",
1169 (unsigned long long)spool->min_hpages << hpage_shift);
1174 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1176 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
1177 struct hstate *h = hstate_inode(d_inode(dentry));
1179 buf->f_type = HUGETLBFS_MAGIC;
1180 buf->f_bsize = huge_page_size(h);
1182 spin_lock(&sbinfo->stat_lock);
1183 /* If no limits set, just report 0 or -1 for max/free/used
1184 * blocks, like simple_statfs() */
1185 if (sbinfo->spool) {
1188 spin_lock_irq(&sbinfo->spool->lock);
1189 buf->f_blocks = sbinfo->spool->max_hpages;
1190 free_pages = sbinfo->spool->max_hpages
1191 - sbinfo->spool->used_hpages;
1192 buf->f_bavail = buf->f_bfree = free_pages;
1193 spin_unlock_irq(&sbinfo->spool->lock);
1194 buf->f_files = sbinfo->max_inodes;
1195 buf->f_ffree = sbinfo->free_inodes;
1197 spin_unlock(&sbinfo->stat_lock);
1199 buf->f_namelen = NAME_MAX;
1203 static void hugetlbfs_put_super(struct super_block *sb)
1205 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
1208 sb->s_fs_info = NULL;
1211 hugepage_put_subpool(sbi->spool);
1217 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
1219 if (sbinfo->free_inodes >= 0) {
1220 spin_lock(&sbinfo->stat_lock);
1221 if (unlikely(!sbinfo->free_inodes)) {
1222 spin_unlock(&sbinfo->stat_lock);
1225 sbinfo->free_inodes--;
1226 spin_unlock(&sbinfo->stat_lock);
1232 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
1234 if (sbinfo->free_inodes >= 0) {
1235 spin_lock(&sbinfo->stat_lock);
1236 sbinfo->free_inodes++;
1237 spin_unlock(&sbinfo->stat_lock);
1242 static struct kmem_cache *hugetlbfs_inode_cachep;
1244 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
1246 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
1247 struct hugetlbfs_inode_info *p;
1249 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
1251 p = alloc_inode_sb(sb, hugetlbfs_inode_cachep, GFP_KERNEL);
1253 hugetlbfs_inc_free_inodes(sbinfo);
1258 * Any time after allocation, hugetlbfs_destroy_inode can be called
1259 * for the inode. mpol_free_shared_policy is unconditionally called
1260 * as part of hugetlbfs_destroy_inode. So, initialize policy here
1261 * in case of a quick call to destroy.
1263 * Note that the policy is initialized even if we are creating a
1264 * private inode. This simplifies hugetlbfs_destroy_inode.
1266 mpol_shared_policy_init(&p->policy, NULL);
1268 return &p->vfs_inode;
1271 static void hugetlbfs_free_inode(struct inode *inode)
1273 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
1276 static void hugetlbfs_destroy_inode(struct inode *inode)
1278 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
1279 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
1282 static const struct address_space_operations hugetlbfs_aops = {
1283 .write_begin = hugetlbfs_write_begin,
1284 .write_end = hugetlbfs_write_end,
1285 .dirty_folio = noop_dirty_folio,
1286 .migrate_folio = hugetlbfs_migrate_folio,
1287 .error_remove_page = hugetlbfs_error_remove_page,
1291 static void init_once(void *foo)
1293 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
1295 inode_init_once(&ei->vfs_inode);
1298 const struct file_operations hugetlbfs_file_operations = {
1299 .read_iter = hugetlbfs_read_iter,
1300 .mmap = hugetlbfs_file_mmap,
1301 .fsync = noop_fsync,
1302 .get_unmapped_area = hugetlb_get_unmapped_area,
1303 .llseek = default_llseek,
1304 .fallocate = hugetlbfs_fallocate,
1307 static const struct inode_operations hugetlbfs_dir_inode_operations = {
1308 .create = hugetlbfs_create,
1309 .lookup = simple_lookup,
1310 .link = simple_link,
1311 .unlink = simple_unlink,
1312 .symlink = hugetlbfs_symlink,
1313 .mkdir = hugetlbfs_mkdir,
1314 .rmdir = simple_rmdir,
1315 .mknod = hugetlbfs_mknod,
1316 .rename = simple_rename,
1317 .setattr = hugetlbfs_setattr,
1318 .tmpfile = hugetlbfs_tmpfile,
1321 static const struct inode_operations hugetlbfs_inode_operations = {
1322 .setattr = hugetlbfs_setattr,
1325 static const struct super_operations hugetlbfs_ops = {
1326 .alloc_inode = hugetlbfs_alloc_inode,
1327 .free_inode = hugetlbfs_free_inode,
1328 .destroy_inode = hugetlbfs_destroy_inode,
1329 .evict_inode = hugetlbfs_evict_inode,
1330 .statfs = hugetlbfs_statfs,
1331 .put_super = hugetlbfs_put_super,
1332 .show_options = hugetlbfs_show_options,
1336 * Convert size option passed from command line to number of huge pages
1337 * in the pool specified by hstate. Size option could be in bytes
1338 * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
1341 hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
1342 enum hugetlbfs_size_type val_type)
1344 if (val_type == NO_SIZE)
1347 if (val_type == SIZE_PERCENT) {
1348 size_opt <<= huge_page_shift(h);
1349 size_opt *= h->max_huge_pages;
1350 do_div(size_opt, 100);
1353 size_opt >>= huge_page_shift(h);
1358 * Parse one mount parameter.
1360 static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param)
1362 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1363 struct fs_parse_result result;
1369 opt = fs_parse(fc, hugetlb_fs_parameters, param, &result);
1375 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
1376 if (!uid_valid(ctx->uid))
1381 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
1382 if (!gid_valid(ctx->gid))
1387 ctx->mode = result.uint_32 & 01777U;
1391 /* memparse() will accept a K/M/G without a digit */
1392 if (!param->string || !isdigit(param->string[0]))
1394 ctx->max_size_opt = memparse(param->string, &rest);
1395 ctx->max_val_type = SIZE_STD;
1397 ctx->max_val_type = SIZE_PERCENT;
1401 /* memparse() will accept a K/M/G without a digit */
1402 if (!param->string || !isdigit(param->string[0]))
1404 ctx->nr_inodes = memparse(param->string, &rest);
1408 ps = memparse(param->string, &rest);
1409 h = size_to_hstate(ps);
1411 pr_err("Unsupported page size %lu MB\n", ps / SZ_1M);
1418 /* memparse() will accept a K/M/G without a digit */
1419 if (!param->string || !isdigit(param->string[0]))
1421 ctx->min_size_opt = memparse(param->string, &rest);
1422 ctx->min_val_type = SIZE_STD;
1424 ctx->min_val_type = SIZE_PERCENT;
1432 return invalfc(fc, "Bad value '%s' for mount option '%s'\n",
1433 param->string, param->key);
1437 * Validate the parsed options.
1439 static int hugetlbfs_validate(struct fs_context *fc)
1441 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1444 * Use huge page pool size (in hstate) to convert the size
1445 * options to number of huge pages. If NO_SIZE, -1 is returned.
1447 ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate,
1450 ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate,
1455 * If max_size was specified, then min_size must be smaller
1457 if (ctx->max_val_type > NO_SIZE &&
1458 ctx->min_hpages > ctx->max_hpages) {
1459 pr_err("Minimum size can not be greater than maximum size\n");
1467 hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc)
1469 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1470 struct hugetlbfs_sb_info *sbinfo;
1472 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
1475 sb->s_fs_info = sbinfo;
1476 spin_lock_init(&sbinfo->stat_lock);
1477 sbinfo->hstate = ctx->hstate;
1478 sbinfo->max_inodes = ctx->nr_inodes;
1479 sbinfo->free_inodes = ctx->nr_inodes;
1480 sbinfo->spool = NULL;
1481 sbinfo->uid = ctx->uid;
1482 sbinfo->gid = ctx->gid;
1483 sbinfo->mode = ctx->mode;
1486 * Allocate and initialize subpool if maximum or minimum size is
1487 * specified. Any needed reservations (for minimum size) are taken
1488 * when the subpool is created.
1490 if (ctx->max_hpages != -1 || ctx->min_hpages != -1) {
1491 sbinfo->spool = hugepage_new_subpool(ctx->hstate,
1497 sb->s_maxbytes = MAX_LFS_FILESIZE;
1498 sb->s_blocksize = huge_page_size(ctx->hstate);
1499 sb->s_blocksize_bits = huge_page_shift(ctx->hstate);
1500 sb->s_magic = HUGETLBFS_MAGIC;
1501 sb->s_op = &hugetlbfs_ops;
1502 sb->s_time_gran = 1;
1505 * Due to the special and limited functionality of hugetlbfs, it does
1506 * not work well as a stacking filesystem.
1508 sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
1509 sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx));
1514 kfree(sbinfo->spool);
1519 static int hugetlbfs_get_tree(struct fs_context *fc)
1521 int err = hugetlbfs_validate(fc);
1524 return get_tree_nodev(fc, hugetlbfs_fill_super);
1527 static void hugetlbfs_fs_context_free(struct fs_context *fc)
1529 kfree(fc->fs_private);
1532 static const struct fs_context_operations hugetlbfs_fs_context_ops = {
1533 .free = hugetlbfs_fs_context_free,
1534 .parse_param = hugetlbfs_parse_param,
1535 .get_tree = hugetlbfs_get_tree,
1538 static int hugetlbfs_init_fs_context(struct fs_context *fc)
1540 struct hugetlbfs_fs_context *ctx;
1542 ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL);
1546 ctx->max_hpages = -1; /* No limit on size by default */
1547 ctx->nr_inodes = -1; /* No limit on number of inodes by default */
1548 ctx->uid = current_fsuid();
1549 ctx->gid = current_fsgid();
1551 ctx->hstate = &default_hstate;
1552 ctx->min_hpages = -1; /* No default minimum size */
1553 ctx->max_val_type = NO_SIZE;
1554 ctx->min_val_type = NO_SIZE;
1555 fc->fs_private = ctx;
1556 fc->ops = &hugetlbfs_fs_context_ops;
1560 static struct file_system_type hugetlbfs_fs_type = {
1561 .name = "hugetlbfs",
1562 .init_fs_context = hugetlbfs_init_fs_context,
1563 .parameters = hugetlb_fs_parameters,
1564 .kill_sb = kill_litter_super,
1567 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
1569 static int can_do_hugetlb_shm(void)
1572 shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
1573 return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
1576 static int get_hstate_idx(int page_size_log)
1578 struct hstate *h = hstate_sizelog(page_size_log);
1582 return hstate_index(h);
1586 * Note that size should be aligned to proper hugepage size in caller side,
1587 * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
1589 struct file *hugetlb_file_setup(const char *name, size_t size,
1590 vm_flags_t acctflag, int creat_flags,
1593 struct inode *inode;
1594 struct vfsmount *mnt;
1598 hstate_idx = get_hstate_idx(page_size_log);
1600 return ERR_PTR(-ENODEV);
1602 mnt = hugetlbfs_vfsmount[hstate_idx];
1604 return ERR_PTR(-ENOENT);
1606 if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
1607 struct ucounts *ucounts = current_ucounts();
1609 if (user_shm_lock(size, ucounts)) {
1610 pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is obsolete\n",
1611 current->comm, current->pid);
1612 user_shm_unlock(size, ucounts);
1614 return ERR_PTR(-EPERM);
1617 file = ERR_PTR(-ENOSPC);
1618 inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0);
1621 if (creat_flags == HUGETLB_SHMFS_INODE)
1622 inode->i_flags |= S_PRIVATE;
1624 inode->i_size = size;
1627 if (!hugetlb_reserve_pages(inode, 0,
1628 size >> huge_page_shift(hstate_inode(inode)), NULL,
1630 file = ERR_PTR(-ENOMEM);
1632 file = alloc_file_pseudo(inode, mnt, name, O_RDWR,
1633 &hugetlbfs_file_operations);
1642 static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h)
1644 struct fs_context *fc;
1645 struct vfsmount *mnt;
1647 fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT);
1651 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1657 pr_err("Cannot mount internal hugetlbfs for page size %luK",
1658 huge_page_size(h) / SZ_1K);
1662 static int __init init_hugetlbfs_fs(void)
1664 struct vfsmount *mnt;
1669 if (!hugepages_supported()) {
1670 pr_info("disabling because there are no supported hugepage sizes\n");
1675 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1676 sizeof(struct hugetlbfs_inode_info),
1677 0, SLAB_ACCOUNT, init_once);
1678 if (hugetlbfs_inode_cachep == NULL)
1681 error = register_filesystem(&hugetlbfs_fs_type);
1685 /* default hstate mount is required */
1686 mnt = mount_one_hugetlbfs(&default_hstate);
1688 error = PTR_ERR(mnt);
1691 hugetlbfs_vfsmount[default_hstate_idx] = mnt;
1693 /* other hstates are optional */
1695 for_each_hstate(h) {
1696 if (i == default_hstate_idx) {
1701 mnt = mount_one_hugetlbfs(h);
1703 hugetlbfs_vfsmount[i] = NULL;
1705 hugetlbfs_vfsmount[i] = mnt;
1712 (void)unregister_filesystem(&hugetlbfs_fs_type);
1714 kmem_cache_destroy(hugetlbfs_inode_cachep);
1718 fs_initcall(init_hugetlbfs_fs)