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);
128 * vma address alignment (but not the pgoff alignment) has
129 * already been checked by prepare_hugepage_range. If you add
130 * any error returns here, do so after setting VM_HUGETLB, so
131 * is_vm_hugetlb_page tests below unmap_region go the right
132 * way when do_mmap unwinds (may be important on powerpc
135 vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
136 vma->vm_ops = &hugetlb_vm_ops;
138 ret = seal_check_future_write(info->seals, vma);
143 * page based offset in vm_pgoff could be sufficiently large to
144 * overflow a loff_t when converted to byte offset. This can
145 * only happen on architectures where sizeof(loff_t) ==
146 * sizeof(unsigned long). So, only check in those instances.
148 if (sizeof(unsigned long) == sizeof(loff_t)) {
149 if (vma->vm_pgoff & PGOFF_LOFFT_MAX)
153 /* must be huge page aligned */
154 if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
157 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
158 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
159 /* check for overflow */
167 if (!hugetlb_reserve_pages(inode,
168 vma->vm_pgoff >> huge_page_order(h),
169 len >> huge_page_shift(h), vma,
174 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
175 i_size_write(inode, len);
183 * Called under mmap_write_lock(mm).
187 hugetlb_get_unmapped_area_bottomup(struct file *file, unsigned long addr,
188 unsigned long len, unsigned long pgoff, unsigned long flags)
190 struct hstate *h = hstate_file(file);
191 struct vm_unmapped_area_info info;
195 info.low_limit = current->mm->mmap_base;
196 info.high_limit = arch_get_mmap_end(addr, len, flags);
197 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
198 info.align_offset = 0;
199 return vm_unmapped_area(&info);
203 hugetlb_get_unmapped_area_topdown(struct file *file, unsigned long addr,
204 unsigned long len, unsigned long pgoff, unsigned long flags)
206 struct hstate *h = hstate_file(file);
207 struct vm_unmapped_area_info info;
209 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
211 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
212 info.high_limit = arch_get_mmap_base(addr, current->mm->mmap_base);
213 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
214 info.align_offset = 0;
215 addr = vm_unmapped_area(&info);
218 * A failed mmap() very likely causes application failure,
219 * so fall back to the bottom-up function here. This scenario
220 * can happen with large stack limits and large mmap()
223 if (unlikely(offset_in_page(addr))) {
224 VM_BUG_ON(addr != -ENOMEM);
226 info.low_limit = current->mm->mmap_base;
227 info.high_limit = arch_get_mmap_end(addr, len, flags);
228 addr = vm_unmapped_area(&info);
235 generic_hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
236 unsigned long len, unsigned long pgoff,
239 struct mm_struct *mm = current->mm;
240 struct vm_area_struct *vma;
241 struct hstate *h = hstate_file(file);
242 const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags);
244 if (len & ~huge_page_mask(h))
249 if (flags & MAP_FIXED) {
250 if (prepare_hugepage_range(file, addr, len))
256 addr = ALIGN(addr, huge_page_size(h));
257 vma = find_vma(mm, addr);
258 if (mmap_end - len >= addr &&
259 (!vma || addr + len <= vm_start_gap(vma)))
264 * Use mm->get_unmapped_area value as a hint to use topdown routine.
265 * If architectures have special needs, they should define their own
266 * version of hugetlb_get_unmapped_area.
268 if (mm->get_unmapped_area == arch_get_unmapped_area_topdown)
269 return hugetlb_get_unmapped_area_topdown(file, addr, len,
271 return hugetlb_get_unmapped_area_bottomup(file, addr, len,
275 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
277 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
278 unsigned long len, unsigned long pgoff,
281 return generic_hugetlb_get_unmapped_area(file, addr, len, pgoff, flags);
286 * Support for read() - Find the page attached to f_mapping and copy out the
287 * data. This provides functionality similar to filemap_read().
289 static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
291 struct file *file = iocb->ki_filp;
292 struct hstate *h = hstate_file(file);
293 struct address_space *mapping = file->f_mapping;
294 struct inode *inode = mapping->host;
295 unsigned long index = iocb->ki_pos >> huge_page_shift(h);
296 unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
297 unsigned long end_index;
301 while (iov_iter_count(to)) {
305 /* nr is the maximum number of bytes to copy from this page */
306 nr = huge_page_size(h);
307 isize = i_size_read(inode);
310 end_index = (isize - 1) >> huge_page_shift(h);
311 if (index > end_index)
313 if (index == end_index) {
314 nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
321 page = find_lock_page(mapping, index);
322 if (unlikely(page == NULL)) {
324 * We have a HOLE, zero out the user-buffer for the
325 * length of the hole or request.
327 copied = iov_iter_zero(nr, to);
332 * We have the page, copy it to user space buffer.
334 copied = copy_page_to_iter(page, offset, nr, to);
339 if (copied != nr && iov_iter_count(to)) {
344 index += offset >> huge_page_shift(h);
345 offset &= ~huge_page_mask(h);
347 iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
351 static int hugetlbfs_write_begin(struct file *file,
352 struct address_space *mapping,
353 loff_t pos, unsigned len,
354 struct page **pagep, void **fsdata)
359 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
360 loff_t pos, unsigned len, unsigned copied,
361 struct page *page, void *fsdata)
367 static void hugetlb_delete_from_page_cache(struct page *page)
369 ClearPageDirty(page);
370 ClearPageUptodate(page);
371 delete_from_page_cache(page);
375 * Called with i_mmap_rwsem held for inode based vma maps. This makes
376 * sure vma (and vm_mm) will not go away. We also hold the hugetlb fault
377 * mutex for the page in the mapping. So, we can not race with page being
378 * faulted into the vma.
380 static bool hugetlb_vma_maps_page(struct vm_area_struct *vma,
381 unsigned long addr, struct page *page)
385 ptep = huge_pte_offset(vma->vm_mm, addr,
386 huge_page_size(hstate_vma(vma)));
391 pte = huge_ptep_get(ptep);
392 if (huge_pte_none(pte) || !pte_present(pte))
395 if (pte_page(pte) == page)
402 * Can vma_offset_start/vma_offset_end overflow on 32-bit arches?
403 * No, because the interval tree returns us only those vmas
404 * which overlap the truncated area starting at pgoff,
405 * and no vma on a 32-bit arch can span beyond the 4GB.
407 static unsigned long vma_offset_start(struct vm_area_struct *vma, pgoff_t start)
409 if (vma->vm_pgoff < start)
410 return (start - vma->vm_pgoff) << PAGE_SHIFT;
415 static unsigned long vma_offset_end(struct vm_area_struct *vma, pgoff_t end)
422 t_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) + vma->vm_start;
423 if (t_end > vma->vm_end)
429 * Called with hugetlb fault mutex held. Therefore, no more mappings to
430 * this folio can be created while executing the routine.
432 static void hugetlb_unmap_file_folio(struct hstate *h,
433 struct address_space *mapping,
434 struct folio *folio, pgoff_t index)
436 struct rb_root_cached *root = &mapping->i_mmap;
437 struct hugetlb_vma_lock *vma_lock;
438 struct page *page = &folio->page;
439 struct vm_area_struct *vma;
440 unsigned long v_start;
444 start = index * pages_per_huge_page(h);
445 end = (index + 1) * pages_per_huge_page(h);
447 i_mmap_lock_write(mapping);
450 vma_interval_tree_foreach(vma, root, start, end - 1) {
451 v_start = vma_offset_start(vma, start);
452 v_end = vma_offset_end(vma, end);
454 if (!hugetlb_vma_maps_page(vma, vma->vm_start + v_start, page))
457 if (!hugetlb_vma_trylock_write(vma)) {
458 vma_lock = vma->vm_private_data;
460 * If we can not get vma lock, we need to drop
461 * immap_sema and take locks in order. First,
462 * take a ref on the vma_lock structure so that
463 * we can be guaranteed it will not go away when
464 * dropping immap_sema.
466 kref_get(&vma_lock->refs);
470 unmap_hugepage_range(vma, vma->vm_start + v_start, v_end,
471 NULL, ZAP_FLAG_DROP_MARKER);
472 hugetlb_vma_unlock_write(vma);
475 i_mmap_unlock_write(mapping);
479 * Wait on vma_lock. We know it is still valid as we have
480 * a reference. We must 'open code' vma locking as we do
481 * not know if vma_lock is still attached to vma.
483 down_write(&vma_lock->rw_sema);
484 i_mmap_lock_write(mapping);
489 * If lock is no longer attached to vma, then just
490 * unlock, drop our reference and retry looking for
493 up_write(&vma_lock->rw_sema);
494 kref_put(&vma_lock->refs, hugetlb_vma_lock_release);
499 * vma_lock is still attached to vma. Check to see if vma
500 * still maps page and if so, unmap.
502 v_start = vma_offset_start(vma, start);
503 v_end = vma_offset_end(vma, end);
504 if (hugetlb_vma_maps_page(vma, vma->vm_start + v_start, page))
505 unmap_hugepage_range(vma, vma->vm_start + v_start,
507 ZAP_FLAG_DROP_MARKER);
509 kref_put(&vma_lock->refs, hugetlb_vma_lock_release);
510 hugetlb_vma_unlock_write(vma);
517 hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end,
518 zap_flags_t zap_flags)
520 struct vm_area_struct *vma;
523 * end == 0 indicates that the entire range after start should be
524 * unmapped. Note, end is exclusive, whereas the interval tree takes
525 * an inclusive "last".
527 vma_interval_tree_foreach(vma, root, start, end ? end - 1 : ULONG_MAX) {
528 unsigned long v_start;
531 if (!hugetlb_vma_trylock_write(vma))
534 v_start = vma_offset_start(vma, start);
535 v_end = vma_offset_end(vma, end);
537 unmap_hugepage_range(vma, vma->vm_start + v_start, v_end,
541 * Note that vma lock only exists for shared/non-private
542 * vmas. Therefore, lock is not held when calling
543 * unmap_hugepage_range for private vmas.
545 hugetlb_vma_unlock_write(vma);
550 * Called with hugetlb fault mutex held.
551 * Returns true if page was actually removed, false otherwise.
553 static bool remove_inode_single_folio(struct hstate *h, struct inode *inode,
554 struct address_space *mapping,
555 struct folio *folio, pgoff_t index,
561 * If folio is mapped, it was faulted in after being
562 * unmapped in caller. Unmap (again) while holding
563 * the fault mutex. The mutex will prevent faults
564 * until we finish removing the folio.
566 if (unlikely(folio_mapped(folio)))
567 hugetlb_unmap_file_folio(h, mapping, folio, index);
571 * We must remove the folio from page cache before removing
572 * the region/ reserve map (hugetlb_unreserve_pages). In
573 * rare out of memory conditions, removal of the region/reserve
574 * map could fail. Correspondingly, the subpool and global
575 * reserve usage count can need to be adjusted.
577 VM_BUG_ON(HPageRestoreReserve(&folio->page));
578 hugetlb_delete_from_page_cache(&folio->page);
581 if (unlikely(hugetlb_unreserve_pages(inode, index,
583 hugetlb_fix_reserve_counts(inode);
591 * remove_inode_hugepages handles two distinct cases: truncation and hole
592 * punch. There are subtle differences in operation for each case.
594 * truncation is indicated by end of range being LLONG_MAX
595 * In this case, we first scan the range and release found pages.
596 * After releasing pages, hugetlb_unreserve_pages cleans up region/reserve
597 * maps and global counts. Page faults can race with truncation.
598 * During faults, hugetlb_no_page() checks i_size before page allocation,
599 * and again after obtaining page table lock. It will 'back out'
600 * allocations in the truncated range.
601 * hole punch is indicated if end is not LLONG_MAX
602 * In the hole punch case we scan the range and release found pages.
603 * Only when releasing a page is the associated region/reserve map
604 * deleted. The region/reserve map for ranges without associated
605 * pages are not modified. Page faults can race with hole punch.
606 * This is indicated if we find a mapped page.
607 * Note: If the passed end of range value is beyond the end of file, but
608 * not LLONG_MAX this routine still performs a hole punch operation.
610 static void remove_inode_hugepages(struct inode *inode, loff_t lstart,
613 struct hstate *h = hstate_inode(inode);
614 struct address_space *mapping = &inode->i_data;
615 const pgoff_t start = lstart >> huge_page_shift(h);
616 const pgoff_t end = lend >> huge_page_shift(h);
617 struct folio_batch fbatch;
620 bool truncate_op = (lend == LLONG_MAX);
622 folio_batch_init(&fbatch);
624 while (filemap_get_folios(mapping, &next, end - 1, &fbatch)) {
625 for (i = 0; i < folio_batch_count(&fbatch); ++i) {
626 struct folio *folio = fbatch.folios[i];
629 index = folio->index;
630 hash = hugetlb_fault_mutex_hash(mapping, index);
631 mutex_lock(&hugetlb_fault_mutex_table[hash]);
634 * Remove folio that was part of folio_batch.
636 if (remove_inode_single_folio(h, inode, mapping, folio,
640 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
642 folio_batch_release(&fbatch);
647 (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed);
650 static void hugetlbfs_evict_inode(struct inode *inode)
652 struct resv_map *resv_map;
654 remove_inode_hugepages(inode, 0, LLONG_MAX);
657 * Get the resv_map from the address space embedded in the inode.
658 * This is the address space which points to any resv_map allocated
659 * at inode creation time. If this is a device special inode,
660 * i_mapping may not point to the original address space.
662 resv_map = (struct resv_map *)(&inode->i_data)->private_data;
663 /* Only regular and link inodes have associated reserve maps */
665 resv_map_release(&resv_map->refs);
669 static void hugetlb_vmtruncate(struct inode *inode, loff_t offset)
672 struct address_space *mapping = inode->i_mapping;
673 struct hstate *h = hstate_inode(inode);
675 BUG_ON(offset & ~huge_page_mask(h));
676 pgoff = offset >> PAGE_SHIFT;
678 i_size_write(inode, offset);
679 i_mmap_lock_write(mapping);
680 if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
681 hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0,
682 ZAP_FLAG_DROP_MARKER);
683 i_mmap_unlock_write(mapping);
684 remove_inode_hugepages(inode, offset, LLONG_MAX);
687 static void hugetlbfs_zero_partial_page(struct hstate *h,
688 struct address_space *mapping,
692 pgoff_t idx = start >> huge_page_shift(h);
695 folio = filemap_lock_folio(mapping, idx);
699 start = start & ~huge_page_mask(h);
700 end = end & ~huge_page_mask(h);
702 end = huge_page_size(h);
704 folio_zero_segment(folio, (size_t)start, (size_t)end);
710 static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
712 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
713 struct address_space *mapping = inode->i_mapping;
714 struct hstate *h = hstate_inode(inode);
715 loff_t hpage_size = huge_page_size(h);
716 loff_t hole_start, hole_end;
719 * hole_start and hole_end indicate the full pages within the hole.
721 hole_start = round_up(offset, hpage_size);
722 hole_end = round_down(offset + len, hpage_size);
726 /* protected by i_rwsem */
727 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
732 i_mmap_lock_write(mapping);
734 /* If range starts before first full page, zero partial page. */
735 if (offset < hole_start)
736 hugetlbfs_zero_partial_page(h, mapping,
737 offset, min(offset + len, hole_start));
739 /* Unmap users of full pages in the hole. */
740 if (hole_end > hole_start) {
741 if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
742 hugetlb_vmdelete_list(&mapping->i_mmap,
743 hole_start >> PAGE_SHIFT,
744 hole_end >> PAGE_SHIFT, 0);
747 /* If range extends beyond last full page, zero partial page. */
748 if ((offset + len) > hole_end && (offset + len) > hole_start)
749 hugetlbfs_zero_partial_page(h, mapping,
750 hole_end, offset + len);
752 i_mmap_unlock_write(mapping);
754 /* Remove full pages from the file. */
755 if (hole_end > hole_start)
756 remove_inode_hugepages(inode, hole_start, hole_end);
763 static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
766 struct inode *inode = file_inode(file);
767 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
768 struct address_space *mapping = inode->i_mapping;
769 struct hstate *h = hstate_inode(inode);
770 struct vm_area_struct pseudo_vma;
771 struct mm_struct *mm = current->mm;
772 loff_t hpage_size = huge_page_size(h);
773 unsigned long hpage_shift = huge_page_shift(h);
774 pgoff_t start, index, end;
778 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
781 if (mode & FALLOC_FL_PUNCH_HOLE)
782 return hugetlbfs_punch_hole(inode, offset, len);
785 * Default preallocate case.
786 * For this range, start is rounded down and end is rounded up
787 * as well as being converted to page offsets.
789 start = offset >> hpage_shift;
790 end = (offset + len + hpage_size - 1) >> hpage_shift;
794 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
795 error = inode_newsize_ok(inode, offset + len);
799 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
805 * Initialize a pseudo vma as this is required by the huge page
806 * allocation routines. If NUMA is configured, use page index
807 * as input to create an allocation policy.
809 vma_init(&pseudo_vma, mm);
810 pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
811 pseudo_vma.vm_file = file;
813 for (index = start; index < end; index++) {
815 * This is supposed to be the vaddr where the page is being
816 * faulted in, but we have no vaddr here.
824 * fallocate(2) manpage permits EINTR; we may have been
825 * interrupted because we are using up too much memory.
827 if (signal_pending(current)) {
832 /* Set numa allocation policy based on index */
833 hugetlb_set_vma_policy(&pseudo_vma, inode, index);
835 /* addr is the offset within the file (zero based) */
836 addr = index * hpage_size;
838 /* mutex taken here, fault path and hole punch */
839 hash = hugetlb_fault_mutex_hash(mapping, index);
840 mutex_lock(&hugetlb_fault_mutex_table[hash]);
842 /* See if already present in mapping to avoid alloc/free */
843 page = find_get_page(mapping, index);
846 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
847 hugetlb_drop_vma_policy(&pseudo_vma);
852 * Allocate page without setting the avoid_reserve argument.
853 * There certainly are no reserves associated with the
854 * pseudo_vma. However, there could be shared mappings with
855 * reserves for the file at the inode level. If we fallocate
856 * pages in these areas, we need to consume the reserves
857 * to keep reservation accounting consistent.
859 page = alloc_huge_page(&pseudo_vma, addr, 0);
860 hugetlb_drop_vma_policy(&pseudo_vma);
862 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
863 error = PTR_ERR(page);
866 clear_huge_page(page, addr, pages_per_huge_page(h));
867 __SetPageUptodate(page);
868 error = hugetlb_add_to_page_cache(page, mapping, index);
869 if (unlikely(error)) {
870 restore_reserve_on_error(h, &pseudo_vma, addr, page);
872 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
876 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
878 SetHPageMigratable(page);
880 * unlock_page because locked by hugetlb_add_to_page_cache()
881 * put_page() due to reference from alloc_huge_page()
887 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
888 i_size_write(inode, offset + len);
889 inode->i_ctime = current_time(inode);
895 static int hugetlbfs_setattr(struct user_namespace *mnt_userns,
896 struct dentry *dentry, struct iattr *attr)
898 struct inode *inode = d_inode(dentry);
899 struct hstate *h = hstate_inode(inode);
901 unsigned int ia_valid = attr->ia_valid;
902 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
904 error = setattr_prepare(&init_user_ns, dentry, attr);
908 if (ia_valid & ATTR_SIZE) {
909 loff_t oldsize = inode->i_size;
910 loff_t newsize = attr->ia_size;
912 if (newsize & ~huge_page_mask(h))
914 /* protected by i_rwsem */
915 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
916 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
918 hugetlb_vmtruncate(inode, newsize);
921 setattr_copy(&init_user_ns, inode, attr);
922 mark_inode_dirty(inode);
926 static struct inode *hugetlbfs_get_root(struct super_block *sb,
927 struct hugetlbfs_fs_context *ctx)
931 inode = new_inode(sb);
933 inode->i_ino = get_next_ino();
934 inode->i_mode = S_IFDIR | ctx->mode;
935 inode->i_uid = ctx->uid;
936 inode->i_gid = ctx->gid;
937 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
938 inode->i_op = &hugetlbfs_dir_inode_operations;
939 inode->i_fop = &simple_dir_operations;
940 /* directory inodes start off with i_nlink == 2 (for "." entry) */
942 lockdep_annotate_inode_mutex_key(inode);
948 * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
949 * be taken from reclaim -- unlike regular filesystems. This needs an
950 * annotation because huge_pmd_share() does an allocation under hugetlb's
953 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
955 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
957 umode_t mode, dev_t dev)
960 struct resv_map *resv_map = NULL;
963 * Reserve maps are only needed for inodes that can have associated
966 if (S_ISREG(mode) || S_ISLNK(mode)) {
967 resv_map = resv_map_alloc();
972 inode = new_inode(sb);
974 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
976 inode->i_ino = get_next_ino();
977 inode_init_owner(&init_user_ns, inode, dir, mode);
978 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
979 &hugetlbfs_i_mmap_rwsem_key);
980 inode->i_mapping->a_ops = &hugetlbfs_aops;
981 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
982 inode->i_mapping->private_data = resv_map;
983 info->seals = F_SEAL_SEAL;
984 switch (mode & S_IFMT) {
986 init_special_inode(inode, mode, dev);
989 inode->i_op = &hugetlbfs_inode_operations;
990 inode->i_fop = &hugetlbfs_file_operations;
993 inode->i_op = &hugetlbfs_dir_inode_operations;
994 inode->i_fop = &simple_dir_operations;
996 /* directory inodes start off with i_nlink == 2 (for "." entry) */
1000 inode->i_op = &page_symlink_inode_operations;
1001 inode_nohighmem(inode);
1004 lockdep_annotate_inode_mutex_key(inode);
1007 kref_put(&resv_map->refs, resv_map_release);
1014 * File creation. Allocate an inode, and we're done..
1016 static int hugetlbfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1017 struct dentry *dentry, umode_t mode, dev_t dev)
1019 struct inode *inode;
1021 inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
1024 dir->i_ctime = dir->i_mtime = current_time(dir);
1025 d_instantiate(dentry, inode);
1026 dget(dentry);/* Extra count - pin the dentry in core */
1030 static int hugetlbfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1031 struct dentry *dentry, umode_t mode)
1033 int retval = hugetlbfs_mknod(&init_user_ns, dir, dentry,
1040 static int hugetlbfs_create(struct user_namespace *mnt_userns,
1041 struct inode *dir, struct dentry *dentry,
1042 umode_t mode, bool excl)
1044 return hugetlbfs_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
1047 static int hugetlbfs_tmpfile(struct user_namespace *mnt_userns,
1048 struct inode *dir, struct file *file,
1051 struct inode *inode;
1053 inode = hugetlbfs_get_inode(dir->i_sb, dir, mode | S_IFREG, 0);
1056 dir->i_ctime = dir->i_mtime = current_time(dir);
1057 d_tmpfile(file, inode);
1058 return finish_open_simple(file, 0);
1061 static int hugetlbfs_symlink(struct user_namespace *mnt_userns,
1062 struct inode *dir, struct dentry *dentry,
1063 const char *symname)
1065 struct inode *inode;
1066 int error = -ENOSPC;
1068 inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
1070 int l = strlen(symname)+1;
1071 error = page_symlink(inode, symname, l);
1073 d_instantiate(dentry, inode);
1078 dir->i_ctime = dir->i_mtime = current_time(dir);
1083 #ifdef CONFIG_MIGRATION
1084 static int hugetlbfs_migrate_folio(struct address_space *mapping,
1085 struct folio *dst, struct folio *src,
1086 enum migrate_mode mode)
1090 rc = migrate_huge_page_move_mapping(mapping, dst, src);
1091 if (rc != MIGRATEPAGE_SUCCESS)
1094 if (hugetlb_page_subpool(&src->page)) {
1095 hugetlb_set_page_subpool(&dst->page,
1096 hugetlb_page_subpool(&src->page));
1097 hugetlb_set_page_subpool(&src->page, NULL);
1100 if (mode != MIGRATE_SYNC_NO_COPY)
1101 folio_migrate_copy(dst, src);
1103 folio_migrate_flags(dst, src);
1105 return MIGRATEPAGE_SUCCESS;
1108 #define hugetlbfs_migrate_folio NULL
1111 static int hugetlbfs_error_remove_page(struct address_space *mapping,
1114 struct inode *inode = mapping->host;
1115 pgoff_t index = page->index;
1117 hugetlb_delete_from_page_cache(page);
1118 if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1)))
1119 hugetlb_fix_reserve_counts(inode);
1125 * Display the mount options in /proc/mounts.
1127 static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
1129 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb);
1130 struct hugepage_subpool *spool = sbinfo->spool;
1131 unsigned long hpage_size = huge_page_size(sbinfo->hstate);
1132 unsigned hpage_shift = huge_page_shift(sbinfo->hstate);
1135 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
1136 seq_printf(m, ",uid=%u",
1137 from_kuid_munged(&init_user_ns, sbinfo->uid));
1138 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
1139 seq_printf(m, ",gid=%u",
1140 from_kgid_munged(&init_user_ns, sbinfo->gid));
1141 if (sbinfo->mode != 0755)
1142 seq_printf(m, ",mode=%o", sbinfo->mode);
1143 if (sbinfo->max_inodes != -1)
1144 seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
1148 if (hpage_size >= 1024) {
1152 seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
1154 if (spool->max_hpages != -1)
1155 seq_printf(m, ",size=%llu",
1156 (unsigned long long)spool->max_hpages << hpage_shift);
1157 if (spool->min_hpages != -1)
1158 seq_printf(m, ",min_size=%llu",
1159 (unsigned long long)spool->min_hpages << hpage_shift);
1164 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1166 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
1167 struct hstate *h = hstate_inode(d_inode(dentry));
1169 buf->f_type = HUGETLBFS_MAGIC;
1170 buf->f_bsize = huge_page_size(h);
1172 spin_lock(&sbinfo->stat_lock);
1173 /* If no limits set, just report 0 or -1 for max/free/used
1174 * blocks, like simple_statfs() */
1175 if (sbinfo->spool) {
1178 spin_lock_irq(&sbinfo->spool->lock);
1179 buf->f_blocks = sbinfo->spool->max_hpages;
1180 free_pages = sbinfo->spool->max_hpages
1181 - sbinfo->spool->used_hpages;
1182 buf->f_bavail = buf->f_bfree = free_pages;
1183 spin_unlock_irq(&sbinfo->spool->lock);
1184 buf->f_files = sbinfo->max_inodes;
1185 buf->f_ffree = sbinfo->free_inodes;
1187 spin_unlock(&sbinfo->stat_lock);
1189 buf->f_namelen = NAME_MAX;
1193 static void hugetlbfs_put_super(struct super_block *sb)
1195 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
1198 sb->s_fs_info = NULL;
1201 hugepage_put_subpool(sbi->spool);
1207 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
1209 if (sbinfo->free_inodes >= 0) {
1210 spin_lock(&sbinfo->stat_lock);
1211 if (unlikely(!sbinfo->free_inodes)) {
1212 spin_unlock(&sbinfo->stat_lock);
1215 sbinfo->free_inodes--;
1216 spin_unlock(&sbinfo->stat_lock);
1222 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
1224 if (sbinfo->free_inodes >= 0) {
1225 spin_lock(&sbinfo->stat_lock);
1226 sbinfo->free_inodes++;
1227 spin_unlock(&sbinfo->stat_lock);
1232 static struct kmem_cache *hugetlbfs_inode_cachep;
1234 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
1236 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
1237 struct hugetlbfs_inode_info *p;
1239 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
1241 p = alloc_inode_sb(sb, hugetlbfs_inode_cachep, GFP_KERNEL);
1243 hugetlbfs_inc_free_inodes(sbinfo);
1248 * Any time after allocation, hugetlbfs_destroy_inode can be called
1249 * for the inode. mpol_free_shared_policy is unconditionally called
1250 * as part of hugetlbfs_destroy_inode. So, initialize policy here
1251 * in case of a quick call to destroy.
1253 * Note that the policy is initialized even if we are creating a
1254 * private inode. This simplifies hugetlbfs_destroy_inode.
1256 mpol_shared_policy_init(&p->policy, NULL);
1258 return &p->vfs_inode;
1261 static void hugetlbfs_free_inode(struct inode *inode)
1263 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
1266 static void hugetlbfs_destroy_inode(struct inode *inode)
1268 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
1269 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
1272 static const struct address_space_operations hugetlbfs_aops = {
1273 .write_begin = hugetlbfs_write_begin,
1274 .write_end = hugetlbfs_write_end,
1275 .dirty_folio = noop_dirty_folio,
1276 .migrate_folio = hugetlbfs_migrate_folio,
1277 .error_remove_page = hugetlbfs_error_remove_page,
1281 static void init_once(void *foo)
1283 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
1285 inode_init_once(&ei->vfs_inode);
1288 const struct file_operations hugetlbfs_file_operations = {
1289 .read_iter = hugetlbfs_read_iter,
1290 .mmap = hugetlbfs_file_mmap,
1291 .fsync = noop_fsync,
1292 .get_unmapped_area = hugetlb_get_unmapped_area,
1293 .llseek = default_llseek,
1294 .fallocate = hugetlbfs_fallocate,
1297 static const struct inode_operations hugetlbfs_dir_inode_operations = {
1298 .create = hugetlbfs_create,
1299 .lookup = simple_lookup,
1300 .link = simple_link,
1301 .unlink = simple_unlink,
1302 .symlink = hugetlbfs_symlink,
1303 .mkdir = hugetlbfs_mkdir,
1304 .rmdir = simple_rmdir,
1305 .mknod = hugetlbfs_mknod,
1306 .rename = simple_rename,
1307 .setattr = hugetlbfs_setattr,
1308 .tmpfile = hugetlbfs_tmpfile,
1311 static const struct inode_operations hugetlbfs_inode_operations = {
1312 .setattr = hugetlbfs_setattr,
1315 static const struct super_operations hugetlbfs_ops = {
1316 .alloc_inode = hugetlbfs_alloc_inode,
1317 .free_inode = hugetlbfs_free_inode,
1318 .destroy_inode = hugetlbfs_destroy_inode,
1319 .evict_inode = hugetlbfs_evict_inode,
1320 .statfs = hugetlbfs_statfs,
1321 .put_super = hugetlbfs_put_super,
1322 .show_options = hugetlbfs_show_options,
1326 * Convert size option passed from command line to number of huge pages
1327 * in the pool specified by hstate. Size option could be in bytes
1328 * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
1331 hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
1332 enum hugetlbfs_size_type val_type)
1334 if (val_type == NO_SIZE)
1337 if (val_type == SIZE_PERCENT) {
1338 size_opt <<= huge_page_shift(h);
1339 size_opt *= h->max_huge_pages;
1340 do_div(size_opt, 100);
1343 size_opt >>= huge_page_shift(h);
1348 * Parse one mount parameter.
1350 static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param)
1352 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1353 struct fs_parse_result result;
1358 opt = fs_parse(fc, hugetlb_fs_parameters, param, &result);
1364 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
1365 if (!uid_valid(ctx->uid))
1370 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
1371 if (!gid_valid(ctx->gid))
1376 ctx->mode = result.uint_32 & 01777U;
1380 /* memparse() will accept a K/M/G without a digit */
1381 if (!isdigit(param->string[0]))
1383 ctx->max_size_opt = memparse(param->string, &rest);
1384 ctx->max_val_type = SIZE_STD;
1386 ctx->max_val_type = SIZE_PERCENT;
1390 /* memparse() will accept a K/M/G without a digit */
1391 if (!isdigit(param->string[0]))
1393 ctx->nr_inodes = memparse(param->string, &rest);
1397 ps = memparse(param->string, &rest);
1398 ctx->hstate = size_to_hstate(ps);
1400 pr_err("Unsupported page size %lu MB\n", ps / SZ_1M);
1406 /* memparse() will accept a K/M/G without a digit */
1407 if (!isdigit(param->string[0]))
1409 ctx->min_size_opt = memparse(param->string, &rest);
1410 ctx->min_val_type = SIZE_STD;
1412 ctx->min_val_type = SIZE_PERCENT;
1420 return invalfc(fc, "Bad value '%s' for mount option '%s'\n",
1421 param->string, param->key);
1425 * Validate the parsed options.
1427 static int hugetlbfs_validate(struct fs_context *fc)
1429 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1432 * Use huge page pool size (in hstate) to convert the size
1433 * options to number of huge pages. If NO_SIZE, -1 is returned.
1435 ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate,
1438 ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate,
1443 * If max_size was specified, then min_size must be smaller
1445 if (ctx->max_val_type > NO_SIZE &&
1446 ctx->min_hpages > ctx->max_hpages) {
1447 pr_err("Minimum size can not be greater than maximum size\n");
1455 hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc)
1457 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1458 struct hugetlbfs_sb_info *sbinfo;
1460 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
1463 sb->s_fs_info = sbinfo;
1464 spin_lock_init(&sbinfo->stat_lock);
1465 sbinfo->hstate = ctx->hstate;
1466 sbinfo->max_inodes = ctx->nr_inodes;
1467 sbinfo->free_inodes = ctx->nr_inodes;
1468 sbinfo->spool = NULL;
1469 sbinfo->uid = ctx->uid;
1470 sbinfo->gid = ctx->gid;
1471 sbinfo->mode = ctx->mode;
1474 * Allocate and initialize subpool if maximum or minimum size is
1475 * specified. Any needed reservations (for minimum size) are taken
1476 * when the subpool is created.
1478 if (ctx->max_hpages != -1 || ctx->min_hpages != -1) {
1479 sbinfo->spool = hugepage_new_subpool(ctx->hstate,
1485 sb->s_maxbytes = MAX_LFS_FILESIZE;
1486 sb->s_blocksize = huge_page_size(ctx->hstate);
1487 sb->s_blocksize_bits = huge_page_shift(ctx->hstate);
1488 sb->s_magic = HUGETLBFS_MAGIC;
1489 sb->s_op = &hugetlbfs_ops;
1490 sb->s_time_gran = 1;
1493 * Due to the special and limited functionality of hugetlbfs, it does
1494 * not work well as a stacking filesystem.
1496 sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
1497 sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx));
1502 kfree(sbinfo->spool);
1507 static int hugetlbfs_get_tree(struct fs_context *fc)
1509 int err = hugetlbfs_validate(fc);
1512 return get_tree_nodev(fc, hugetlbfs_fill_super);
1515 static void hugetlbfs_fs_context_free(struct fs_context *fc)
1517 kfree(fc->fs_private);
1520 static const struct fs_context_operations hugetlbfs_fs_context_ops = {
1521 .free = hugetlbfs_fs_context_free,
1522 .parse_param = hugetlbfs_parse_param,
1523 .get_tree = hugetlbfs_get_tree,
1526 static int hugetlbfs_init_fs_context(struct fs_context *fc)
1528 struct hugetlbfs_fs_context *ctx;
1530 ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL);
1534 ctx->max_hpages = -1; /* No limit on size by default */
1535 ctx->nr_inodes = -1; /* No limit on number of inodes by default */
1536 ctx->uid = current_fsuid();
1537 ctx->gid = current_fsgid();
1539 ctx->hstate = &default_hstate;
1540 ctx->min_hpages = -1; /* No default minimum size */
1541 ctx->max_val_type = NO_SIZE;
1542 ctx->min_val_type = NO_SIZE;
1543 fc->fs_private = ctx;
1544 fc->ops = &hugetlbfs_fs_context_ops;
1548 static struct file_system_type hugetlbfs_fs_type = {
1549 .name = "hugetlbfs",
1550 .init_fs_context = hugetlbfs_init_fs_context,
1551 .parameters = hugetlb_fs_parameters,
1552 .kill_sb = kill_litter_super,
1555 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
1557 static int can_do_hugetlb_shm(void)
1560 shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
1561 return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
1564 static int get_hstate_idx(int page_size_log)
1566 struct hstate *h = hstate_sizelog(page_size_log);
1570 return hstate_index(h);
1574 * Note that size should be aligned to proper hugepage size in caller side,
1575 * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
1577 struct file *hugetlb_file_setup(const char *name, size_t size,
1578 vm_flags_t acctflag, int creat_flags,
1581 struct inode *inode;
1582 struct vfsmount *mnt;
1586 hstate_idx = get_hstate_idx(page_size_log);
1588 return ERR_PTR(-ENODEV);
1590 mnt = hugetlbfs_vfsmount[hstate_idx];
1592 return ERR_PTR(-ENOENT);
1594 if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
1595 struct ucounts *ucounts = current_ucounts();
1597 if (user_shm_lock(size, ucounts)) {
1598 pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is obsolete\n",
1599 current->comm, current->pid);
1600 user_shm_unlock(size, ucounts);
1602 return ERR_PTR(-EPERM);
1605 file = ERR_PTR(-ENOSPC);
1606 inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0);
1609 if (creat_flags == HUGETLB_SHMFS_INODE)
1610 inode->i_flags |= S_PRIVATE;
1612 inode->i_size = size;
1615 if (!hugetlb_reserve_pages(inode, 0,
1616 size >> huge_page_shift(hstate_inode(inode)), NULL,
1618 file = ERR_PTR(-ENOMEM);
1620 file = alloc_file_pseudo(inode, mnt, name, O_RDWR,
1621 &hugetlbfs_file_operations);
1630 static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h)
1632 struct fs_context *fc;
1633 struct vfsmount *mnt;
1635 fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT);
1639 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1645 pr_err("Cannot mount internal hugetlbfs for page size %luK",
1646 huge_page_size(h) / SZ_1K);
1650 static int __init init_hugetlbfs_fs(void)
1652 struct vfsmount *mnt;
1657 if (!hugepages_supported()) {
1658 pr_info("disabling because there are no supported hugepage sizes\n");
1663 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1664 sizeof(struct hugetlbfs_inode_info),
1665 0, SLAB_ACCOUNT, init_once);
1666 if (hugetlbfs_inode_cachep == NULL)
1669 error = register_filesystem(&hugetlbfs_fs_type);
1673 /* default hstate mount is required */
1674 mnt = mount_one_hugetlbfs(&default_hstate);
1676 error = PTR_ERR(mnt);
1679 hugetlbfs_vfsmount[default_hstate_idx] = mnt;
1681 /* other hstates are optional */
1683 for_each_hstate(h) {
1684 if (i == default_hstate_idx) {
1689 mnt = mount_one_hugetlbfs(h);
1691 hugetlbfs_vfsmount[i] = NULL;
1693 hugetlbfs_vfsmount[i] = mnt;
1700 (void)unregister_filesystem(&hugetlbfs_fs_type);
1702 kmem_cache_destroy(hugetlbfs_inode_cachep);
1706 fs_initcall(init_hugetlbfs_fs)