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/sched.h> /* remove ASAP */
15 #include <linux/falloc.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/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>
40 #include <asm/uaccess.h>
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;
48 struct hugetlbfs_config {
54 struct hstate *hstate;
58 struct hugetlbfs_inode_info {
59 struct shared_policy policy;
60 struct inode vfs_inode;
63 static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
65 return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
68 int sysctl_hugetlb_shm_group;
71 Opt_size, Opt_nr_inodes,
72 Opt_mode, Opt_uid, Opt_gid,
73 Opt_pagesize, Opt_min_size,
77 static const match_table_t tokens = {
78 {Opt_size, "size=%s"},
79 {Opt_nr_inodes, "nr_inodes=%s"},
80 {Opt_mode, "mode=%o"},
83 {Opt_pagesize, "pagesize=%s"},
84 {Opt_min_size, "min_size=%s"},
89 static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
90 struct inode *inode, pgoff_t index)
92 vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy,
96 static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
98 mpol_cond_put(vma->vm_policy);
101 static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
102 struct inode *inode, pgoff_t index)
106 static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
111 static void huge_pagevec_release(struct pagevec *pvec)
115 for (i = 0; i < pagevec_count(pvec); ++i)
116 put_page(pvec->pages[i]);
118 pagevec_reinit(pvec);
121 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
123 struct inode *inode = file_inode(file);
126 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_pgoff unwinds (may be important on powerpc
136 vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
137 vma->vm_ops = &hugetlb_vm_ops;
139 if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
142 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
148 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
150 if (hugetlb_reserve_pages(inode,
151 vma->vm_pgoff >> huge_page_order(h),
152 len >> huge_page_shift(h), vma,
157 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
166 * Called under down_write(mmap_sem).
169 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
171 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
172 unsigned long len, unsigned long pgoff, unsigned long flags)
174 struct mm_struct *mm = current->mm;
175 struct vm_area_struct *vma;
176 struct hstate *h = hstate_file(file);
177 struct vm_unmapped_area_info info;
179 if (len & ~huge_page_mask(h))
184 if (flags & MAP_FIXED) {
185 if (prepare_hugepage_range(file, addr, len))
191 addr = ALIGN(addr, huge_page_size(h));
192 vma = find_vma(mm, addr);
193 if (TASK_SIZE - len >= addr &&
194 (!vma || addr + len <= vma->vm_start))
200 info.low_limit = TASK_UNMAPPED_BASE;
201 info.high_limit = TASK_SIZE;
202 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
203 info.align_offset = 0;
204 return vm_unmapped_area(&info);
209 hugetlbfs_read_actor(struct page *page, unsigned long offset,
210 struct iov_iter *to, unsigned long size)
215 /* Find which 4k chunk and offset with in that chunk */
216 i = offset >> PAGE_SHIFT;
217 offset = offset & ~PAGE_MASK;
221 chunksize = PAGE_SIZE;
224 if (chunksize > size)
226 n = copy_page_to_iter(&page[i], offset, chunksize, to);
238 * Support for read() - Find the page attached to f_mapping and copy out the
239 * data. Its *very* similar to do_generic_mapping_read(), we can't use that
240 * since it has PAGE_SIZE assumptions.
242 static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
244 struct file *file = iocb->ki_filp;
245 struct hstate *h = hstate_file(file);
246 struct address_space *mapping = file->f_mapping;
247 struct inode *inode = mapping->host;
248 unsigned long index = iocb->ki_pos >> huge_page_shift(h);
249 unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
250 unsigned long end_index;
254 while (iov_iter_count(to)) {
258 /* nr is the maximum number of bytes to copy from this page */
259 nr = huge_page_size(h);
260 isize = i_size_read(inode);
263 end_index = (isize - 1) >> huge_page_shift(h);
264 if (index > end_index)
266 if (index == end_index) {
267 nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
274 page = find_lock_page(mapping, index);
275 if (unlikely(page == NULL)) {
277 * We have a HOLE, zero out the user-buffer for the
278 * length of the hole or request.
280 copied = iov_iter_zero(nr, to);
285 * We have the page, copy it to user space buffer.
287 copied = hugetlbfs_read_actor(page, offset, to, nr);
292 if (copied != nr && iov_iter_count(to)) {
297 index += offset >> huge_page_shift(h);
298 offset &= ~huge_page_mask(h);
300 iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
304 static int hugetlbfs_write_begin(struct file *file,
305 struct address_space *mapping,
306 loff_t pos, unsigned len, unsigned flags,
307 struct page **pagep, void **fsdata)
312 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
313 loff_t pos, unsigned len, unsigned copied,
314 struct page *page, void *fsdata)
320 static void remove_huge_page(struct page *page)
322 ClearPageDirty(page);
323 ClearPageUptodate(page);
324 delete_from_page_cache(page);
328 hugetlb_vmdelete_list(struct rb_root *root, pgoff_t start, pgoff_t end)
330 struct vm_area_struct *vma;
333 * end == 0 indicates that the entire range after
334 * start should be unmapped.
336 vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) {
337 unsigned long v_offset;
341 * Can the expression below overflow on 32-bit arches?
342 * No, because the interval tree returns us only those vmas
343 * which overlap the truncated area starting at pgoff,
344 * and no vma on a 32-bit arch can span beyond the 4GB.
346 if (vma->vm_pgoff < start)
347 v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT;
354 v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT)
356 if (v_end > vma->vm_end)
360 unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end,
366 * remove_inode_hugepages handles two distinct cases: truncation and hole
367 * punch. There are subtle differences in operation for each case.
369 * truncation is indicated by end of range being LLONG_MAX
370 * In this case, we first scan the range and release found pages.
371 * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv
372 * maps and global counts. Page faults can not race with truncation
373 * in this routine. hugetlb_no_page() prevents page faults in the
374 * truncated range. It checks i_size before allocation, and again after
375 * with the page table lock for the page held. The same lock must be
376 * acquired to unmap a page.
377 * hole punch is indicated if end is not LLONG_MAX
378 * In the hole punch case we scan the range and release found pages.
379 * Only when releasing a page is the associated region/reserv map
380 * deleted. The region/reserv map for ranges without associated
381 * pages are not modified. Page faults can race with hole punch.
382 * This is indicated if we find a mapped page.
383 * Note: If the passed end of range value is beyond the end of file, but
384 * not LLONG_MAX this routine still performs a hole punch operation.
386 static void remove_inode_hugepages(struct inode *inode, loff_t lstart,
389 struct hstate *h = hstate_inode(inode);
390 struct address_space *mapping = &inode->i_data;
391 const pgoff_t start = lstart >> huge_page_shift(h);
392 const pgoff_t end = lend >> huge_page_shift(h);
393 struct vm_area_struct pseudo_vma;
397 long lookup_nr = PAGEVEC_SIZE;
398 bool truncate_op = (lend == LLONG_MAX);
400 memset(&pseudo_vma, 0, sizeof(struct vm_area_struct));
401 pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
402 pagevec_init(&pvec, 0);
406 * Don't grab more pages than the number left in the range.
408 if (end - next < lookup_nr)
409 lookup_nr = end - next;
412 * When no more pages are found, we are done.
414 if (!pagevec_lookup(&pvec, mapping, next, lookup_nr))
417 for (i = 0; i < pagevec_count(&pvec); ++i) {
418 struct page *page = pvec.pages[i];
423 * The page (index) could be beyond end. This is
424 * only possible in the punch hole case as end is
425 * max page offset in the truncate case.
431 hash = hugetlb_fault_mutex_hash(h, current->mm,
434 mutex_lock(&hugetlb_fault_mutex_table[hash]);
437 * If page is mapped, it was faulted in after being
438 * unmapped in caller. Unmap (again) now after taking
439 * the fault mutex. The mutex will prevent faults
440 * until we finish removing the page.
442 * This race can only happen in the hole punch case.
443 * Getting here in a truncate operation is a bug.
445 if (unlikely(page_mapped(page))) {
448 i_mmap_lock_write(mapping);
449 hugetlb_vmdelete_list(&mapping->i_mmap,
450 next * pages_per_huge_page(h),
451 (next + 1) * pages_per_huge_page(h));
452 i_mmap_unlock_write(mapping);
457 * We must free the huge page and remove from page
458 * cache (remove_huge_page) BEFORE removing the
459 * region/reserve map (hugetlb_unreserve_pages). In
460 * rare out of memory conditions, removal of the
461 * region/reserve map could fail. Before free'ing
462 * the page, note PagePrivate which is used in case
465 rsv_on_error = !PagePrivate(page);
466 remove_huge_page(page);
469 if (unlikely(hugetlb_unreserve_pages(inode,
471 hugetlb_fix_reserve_counts(inode,
476 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
479 huge_pagevec_release(&pvec);
484 (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed);
487 static void hugetlbfs_evict_inode(struct inode *inode)
489 struct resv_map *resv_map;
491 remove_inode_hugepages(inode, 0, LLONG_MAX);
492 resv_map = (struct resv_map *)inode->i_mapping->private_data;
493 /* root inode doesn't have the resv_map, so we should check it */
495 resv_map_release(&resv_map->refs);
499 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
502 struct address_space *mapping = inode->i_mapping;
503 struct hstate *h = hstate_inode(inode);
505 BUG_ON(offset & ~huge_page_mask(h));
506 pgoff = offset >> PAGE_SHIFT;
508 i_size_write(inode, offset);
509 i_mmap_lock_write(mapping);
510 if (!RB_EMPTY_ROOT(&mapping->i_mmap))
511 hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0);
512 i_mmap_unlock_write(mapping);
513 remove_inode_hugepages(inode, offset, LLONG_MAX);
517 static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
519 struct hstate *h = hstate_inode(inode);
520 loff_t hpage_size = huge_page_size(h);
521 loff_t hole_start, hole_end;
524 * For hole punch round up the beginning offset of the hole and
525 * round down the end.
527 hole_start = round_up(offset, hpage_size);
528 hole_end = round_down(offset + len, hpage_size);
530 if (hole_end > hole_start) {
531 struct address_space *mapping = inode->i_mapping;
534 i_mmap_lock_write(mapping);
535 if (!RB_EMPTY_ROOT(&mapping->i_mmap))
536 hugetlb_vmdelete_list(&mapping->i_mmap,
537 hole_start >> PAGE_SHIFT,
538 hole_end >> PAGE_SHIFT);
539 i_mmap_unlock_write(mapping);
540 remove_inode_hugepages(inode, hole_start, hole_end);
547 static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
550 struct inode *inode = file_inode(file);
551 struct address_space *mapping = inode->i_mapping;
552 struct hstate *h = hstate_inode(inode);
553 struct vm_area_struct pseudo_vma;
554 struct mm_struct *mm = current->mm;
555 loff_t hpage_size = huge_page_size(h);
556 unsigned long hpage_shift = huge_page_shift(h);
557 pgoff_t start, index, end;
561 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
564 if (mode & FALLOC_FL_PUNCH_HOLE)
565 return hugetlbfs_punch_hole(inode, offset, len);
568 * Default preallocate case.
569 * For this range, start is rounded down and end is rounded up
570 * as well as being converted to page offsets.
572 start = offset >> hpage_shift;
573 end = (offset + len + hpage_size - 1) >> hpage_shift;
577 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
578 error = inode_newsize_ok(inode, offset + len);
583 * Initialize a pseudo vma as this is required by the huge page
584 * allocation routines. If NUMA is configured, use page index
585 * as input to create an allocation policy.
587 memset(&pseudo_vma, 0, sizeof(struct vm_area_struct));
588 pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
589 pseudo_vma.vm_file = file;
591 for (index = start; index < end; index++) {
593 * This is supposed to be the vaddr where the page is being
594 * faulted in, but we have no vaddr here.
598 int avoid_reserve = 0;
603 * fallocate(2) manpage permits EINTR; we may have been
604 * interrupted because we are using up too much memory.
606 if (signal_pending(current)) {
611 /* Set numa allocation policy based on index */
612 hugetlb_set_vma_policy(&pseudo_vma, inode, index);
614 /* addr is the offset within the file (zero based) */
615 addr = index * hpage_size;
617 /* mutex taken here, fault path and hole punch */
618 hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping,
620 mutex_lock(&hugetlb_fault_mutex_table[hash]);
622 /* See if already present in mapping to avoid alloc/free */
623 page = find_get_page(mapping, index);
626 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
627 hugetlb_drop_vma_policy(&pseudo_vma);
631 /* Allocate page and add to page cache */
632 page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve);
633 hugetlb_drop_vma_policy(&pseudo_vma);
635 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
636 error = PTR_ERR(page);
639 clear_huge_page(page, addr, pages_per_huge_page(h));
640 __SetPageUptodate(page);
641 error = huge_add_to_page_cache(page, mapping, index);
642 if (unlikely(error)) {
644 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
648 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
651 * page_put due to reference from alloc_huge_page()
652 * unlock_page because locked by add_to_page_cache()
658 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
659 i_size_write(inode, offset + len);
660 inode->i_ctime = CURRENT_TIME;
666 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
668 struct inode *inode = d_inode(dentry);
669 struct hstate *h = hstate_inode(inode);
671 unsigned int ia_valid = attr->ia_valid;
675 error = inode_change_ok(inode, attr);
679 if (ia_valid & ATTR_SIZE) {
681 if (attr->ia_size & ~huge_page_mask(h))
683 error = hugetlb_vmtruncate(inode, attr->ia_size);
688 setattr_copy(inode, attr);
689 mark_inode_dirty(inode);
693 static struct inode *hugetlbfs_get_root(struct super_block *sb,
694 struct hugetlbfs_config *config)
698 inode = new_inode(sb);
700 struct hugetlbfs_inode_info *info;
701 inode->i_ino = get_next_ino();
702 inode->i_mode = S_IFDIR | config->mode;
703 inode->i_uid = config->uid;
704 inode->i_gid = config->gid;
705 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
706 info = HUGETLBFS_I(inode);
707 mpol_shared_policy_init(&info->policy, NULL);
708 inode->i_op = &hugetlbfs_dir_inode_operations;
709 inode->i_fop = &simple_dir_operations;
710 /* directory inodes start off with i_nlink == 2 (for "." entry) */
712 lockdep_annotate_inode_mutex_key(inode);
718 * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
719 * be taken from reclaim -- unlike regular filesystems. This needs an
720 * annotation because huge_pmd_share() does an allocation under hugetlb's
723 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
725 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
727 umode_t mode, dev_t dev)
730 struct resv_map *resv_map;
732 resv_map = resv_map_alloc();
736 inode = new_inode(sb);
738 struct hugetlbfs_inode_info *info;
739 inode->i_ino = get_next_ino();
740 inode_init_owner(inode, dir, mode);
741 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
742 &hugetlbfs_i_mmap_rwsem_key);
743 inode->i_mapping->a_ops = &hugetlbfs_aops;
744 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
745 inode->i_mapping->private_data = resv_map;
746 info = HUGETLBFS_I(inode);
748 * The policy is initialized here even if we are creating a
749 * private inode because initialization simply creates an
750 * an empty rb tree and calls rwlock_init(), later when we
751 * call mpol_free_shared_policy() it will just return because
752 * the rb tree will still be empty.
754 mpol_shared_policy_init(&info->policy, NULL);
755 switch (mode & S_IFMT) {
757 init_special_inode(inode, mode, dev);
760 inode->i_op = &hugetlbfs_inode_operations;
761 inode->i_fop = &hugetlbfs_file_operations;
764 inode->i_op = &hugetlbfs_dir_inode_operations;
765 inode->i_fop = &simple_dir_operations;
767 /* directory inodes start off with i_nlink == 2 (for "." entry) */
771 inode->i_op = &page_symlink_inode_operations;
772 inode_nohighmem(inode);
775 lockdep_annotate_inode_mutex_key(inode);
777 kref_put(&resv_map->refs, resv_map_release);
783 * File creation. Allocate an inode, and we're done..
785 static int hugetlbfs_mknod(struct inode *dir,
786 struct dentry *dentry, umode_t mode, dev_t dev)
791 inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
793 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
794 d_instantiate(dentry, inode);
795 dget(dentry); /* Extra count - pin the dentry in core */
801 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
803 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
809 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
811 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
814 static int hugetlbfs_symlink(struct inode *dir,
815 struct dentry *dentry, const char *symname)
820 inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
822 int l = strlen(symname)+1;
823 error = page_symlink(inode, symname, l);
825 d_instantiate(dentry, inode);
830 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
836 * mark the head page dirty
838 static int hugetlbfs_set_page_dirty(struct page *page)
840 struct page *head = compound_head(page);
846 static int hugetlbfs_migrate_page(struct address_space *mapping,
847 struct page *newpage, struct page *page,
848 enum migrate_mode mode)
852 rc = migrate_huge_page_move_mapping(mapping, newpage, page);
853 if (rc != MIGRATEPAGE_SUCCESS)
855 migrate_page_copy(newpage, page);
857 return MIGRATEPAGE_SUCCESS;
860 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
862 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
863 struct hstate *h = hstate_inode(d_inode(dentry));
865 buf->f_type = HUGETLBFS_MAGIC;
866 buf->f_bsize = huge_page_size(h);
868 spin_lock(&sbinfo->stat_lock);
869 /* If no limits set, just report 0 for max/free/used
870 * blocks, like simple_statfs() */
874 spin_lock(&sbinfo->spool->lock);
875 buf->f_blocks = sbinfo->spool->max_hpages;
876 free_pages = sbinfo->spool->max_hpages
877 - sbinfo->spool->used_hpages;
878 buf->f_bavail = buf->f_bfree = free_pages;
879 spin_unlock(&sbinfo->spool->lock);
880 buf->f_files = sbinfo->max_inodes;
881 buf->f_ffree = sbinfo->free_inodes;
883 spin_unlock(&sbinfo->stat_lock);
885 buf->f_namelen = NAME_MAX;
889 static void hugetlbfs_put_super(struct super_block *sb)
891 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
894 sb->s_fs_info = NULL;
897 hugepage_put_subpool(sbi->spool);
903 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
905 if (sbinfo->free_inodes >= 0) {
906 spin_lock(&sbinfo->stat_lock);
907 if (unlikely(!sbinfo->free_inodes)) {
908 spin_unlock(&sbinfo->stat_lock);
911 sbinfo->free_inodes--;
912 spin_unlock(&sbinfo->stat_lock);
918 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
920 if (sbinfo->free_inodes >= 0) {
921 spin_lock(&sbinfo->stat_lock);
922 sbinfo->free_inodes++;
923 spin_unlock(&sbinfo->stat_lock);
928 static struct kmem_cache *hugetlbfs_inode_cachep;
930 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
932 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
933 struct hugetlbfs_inode_info *p;
935 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
937 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
939 hugetlbfs_inc_free_inodes(sbinfo);
942 return &p->vfs_inode;
945 static void hugetlbfs_i_callback(struct rcu_head *head)
947 struct inode *inode = container_of(head, struct inode, i_rcu);
948 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
951 static void hugetlbfs_destroy_inode(struct inode *inode)
953 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
954 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
955 call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
958 static const struct address_space_operations hugetlbfs_aops = {
959 .write_begin = hugetlbfs_write_begin,
960 .write_end = hugetlbfs_write_end,
961 .set_page_dirty = hugetlbfs_set_page_dirty,
962 .migratepage = hugetlbfs_migrate_page,
966 static void init_once(void *foo)
968 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
970 inode_init_once(&ei->vfs_inode);
973 const struct file_operations hugetlbfs_file_operations = {
974 .read_iter = hugetlbfs_read_iter,
975 .mmap = hugetlbfs_file_mmap,
977 .get_unmapped_area = hugetlb_get_unmapped_area,
978 .llseek = default_llseek,
979 .fallocate = hugetlbfs_fallocate,
982 static const struct inode_operations hugetlbfs_dir_inode_operations = {
983 .create = hugetlbfs_create,
984 .lookup = simple_lookup,
986 .unlink = simple_unlink,
987 .symlink = hugetlbfs_symlink,
988 .mkdir = hugetlbfs_mkdir,
989 .rmdir = simple_rmdir,
990 .mknod = hugetlbfs_mknod,
991 .rename = simple_rename,
992 .setattr = hugetlbfs_setattr,
995 static const struct inode_operations hugetlbfs_inode_operations = {
996 .setattr = hugetlbfs_setattr,
999 static const struct super_operations hugetlbfs_ops = {
1000 .alloc_inode = hugetlbfs_alloc_inode,
1001 .destroy_inode = hugetlbfs_destroy_inode,
1002 .evict_inode = hugetlbfs_evict_inode,
1003 .statfs = hugetlbfs_statfs,
1004 .put_super = hugetlbfs_put_super,
1005 .show_options = generic_show_options,
1008 enum { NO_SIZE, SIZE_STD, SIZE_PERCENT };
1011 * Convert size option passed from command line to number of huge pages
1012 * in the pool specified by hstate. Size option could be in bytes
1013 * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
1016 hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
1019 if (val_type == NO_SIZE)
1022 if (val_type == SIZE_PERCENT) {
1023 size_opt <<= huge_page_shift(h);
1024 size_opt *= h->max_huge_pages;
1025 do_div(size_opt, 100);
1028 size_opt >>= huge_page_shift(h);
1033 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
1036 substring_t args[MAX_OPT_ARGS];
1038 unsigned long long max_size_opt = 0, min_size_opt = 0;
1039 int max_val_type = NO_SIZE, min_val_type = NO_SIZE;
1044 while ((p = strsep(&options, ",")) != NULL) {
1049 token = match_token(p, tokens, args);
1052 if (match_int(&args[0], &option))
1054 pconfig->uid = make_kuid(current_user_ns(), option);
1055 if (!uid_valid(pconfig->uid))
1060 if (match_int(&args[0], &option))
1062 pconfig->gid = make_kgid(current_user_ns(), option);
1063 if (!gid_valid(pconfig->gid))
1068 if (match_octal(&args[0], &option))
1070 pconfig->mode = option & 01777U;
1074 /* memparse() will accept a K/M/G without a digit */
1075 if (!isdigit(*args[0].from))
1077 max_size_opt = memparse(args[0].from, &rest);
1078 max_val_type = SIZE_STD;
1080 max_val_type = SIZE_PERCENT;
1085 /* memparse() will accept a K/M/G without a digit */
1086 if (!isdigit(*args[0].from))
1088 pconfig->nr_inodes = memparse(args[0].from, &rest);
1091 case Opt_pagesize: {
1093 ps = memparse(args[0].from, &rest);
1094 pconfig->hstate = size_to_hstate(ps);
1095 if (!pconfig->hstate) {
1096 pr_err("Unsupported page size %lu MB\n",
1103 case Opt_min_size: {
1104 /* memparse() will accept a K/M/G without a digit */
1105 if (!isdigit(*args[0].from))
1107 min_size_opt = memparse(args[0].from, &rest);
1108 min_val_type = SIZE_STD;
1110 min_val_type = SIZE_PERCENT;
1115 pr_err("Bad mount option: \"%s\"\n", p);
1122 * Use huge page pool size (in hstate) to convert the size
1123 * options to number of huge pages. If NO_SIZE, -1 is returned.
1125 pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
1126 max_size_opt, max_val_type);
1127 pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
1128 min_size_opt, min_val_type);
1131 * If max_size was specified, then min_size must be smaller
1133 if (max_val_type > NO_SIZE &&
1134 pconfig->min_hpages > pconfig->max_hpages) {
1135 pr_err("minimum size can not be greater than maximum size\n");
1142 pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p);
1147 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
1150 struct hugetlbfs_config config;
1151 struct hugetlbfs_sb_info *sbinfo;
1153 save_mount_options(sb, data);
1155 config.max_hpages = -1; /* No limit on size by default */
1156 config.nr_inodes = -1; /* No limit on number of inodes by default */
1157 config.uid = current_fsuid();
1158 config.gid = current_fsgid();
1160 config.hstate = &default_hstate;
1161 config.min_hpages = -1; /* No default minimum size */
1162 ret = hugetlbfs_parse_options(data, &config);
1166 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
1169 sb->s_fs_info = sbinfo;
1170 sbinfo->hstate = config.hstate;
1171 spin_lock_init(&sbinfo->stat_lock);
1172 sbinfo->max_inodes = config.nr_inodes;
1173 sbinfo->free_inodes = config.nr_inodes;
1174 sbinfo->spool = NULL;
1176 * Allocate and initialize subpool if maximum or minimum size is
1177 * specified. Any needed reservations (for minimim size) are taken
1178 * taken when the subpool is created.
1180 if (config.max_hpages != -1 || config.min_hpages != -1) {
1181 sbinfo->spool = hugepage_new_subpool(config.hstate,
1187 sb->s_maxbytes = MAX_LFS_FILESIZE;
1188 sb->s_blocksize = huge_page_size(config.hstate);
1189 sb->s_blocksize_bits = huge_page_shift(config.hstate);
1190 sb->s_magic = HUGETLBFS_MAGIC;
1191 sb->s_op = &hugetlbfs_ops;
1192 sb->s_time_gran = 1;
1193 sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
1198 kfree(sbinfo->spool);
1203 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
1204 int flags, const char *dev_name, void *data)
1206 return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
1209 static struct file_system_type hugetlbfs_fs_type = {
1210 .name = "hugetlbfs",
1211 .mount = hugetlbfs_mount,
1212 .kill_sb = kill_litter_super,
1215 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
1217 static int can_do_hugetlb_shm(void)
1220 shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
1221 return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
1224 static int get_hstate_idx(int page_size_log)
1226 struct hstate *h = hstate_sizelog(page_size_log);
1233 static const struct dentry_operations anon_ops = {
1234 .d_dname = simple_dname
1238 * Note that size should be aligned to proper hugepage size in caller side,
1239 * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
1241 struct file *hugetlb_file_setup(const char *name, size_t size,
1242 vm_flags_t acctflag, struct user_struct **user,
1243 int creat_flags, int page_size_log)
1245 struct file *file = ERR_PTR(-ENOMEM);
1246 struct inode *inode;
1248 struct super_block *sb;
1249 struct qstr quick_string;
1252 hstate_idx = get_hstate_idx(page_size_log);
1254 return ERR_PTR(-ENODEV);
1257 if (!hugetlbfs_vfsmount[hstate_idx])
1258 return ERR_PTR(-ENOENT);
1260 if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
1261 *user = current_user();
1262 if (user_shm_lock(size, *user)) {
1264 pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
1265 current->comm, current->pid);
1266 task_unlock(current);
1269 return ERR_PTR(-EPERM);
1273 sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb;
1274 quick_string.name = name;
1275 quick_string.len = strlen(quick_string.name);
1276 quick_string.hash = 0;
1277 path.dentry = d_alloc_pseudo(sb, &quick_string);
1279 goto out_shm_unlock;
1281 d_set_d_op(path.dentry, &anon_ops);
1282 path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]);
1283 file = ERR_PTR(-ENOSPC);
1284 inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0);
1287 if (creat_flags == HUGETLB_SHMFS_INODE)
1288 inode->i_flags |= S_PRIVATE;
1290 file = ERR_PTR(-ENOMEM);
1291 if (hugetlb_reserve_pages(inode, 0,
1292 size >> huge_page_shift(hstate_inode(inode)), NULL,
1296 d_instantiate(path.dentry, inode);
1297 inode->i_size = size;
1300 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
1301 &hugetlbfs_file_operations);
1303 goto out_dentry; /* inode is already attached */
1313 user_shm_unlock(size, *user);
1319 static int __init init_hugetlbfs_fs(void)
1325 if (!hugepages_supported()) {
1326 pr_info("disabling because there are no supported hugepage sizes\n");
1331 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1332 sizeof(struct hugetlbfs_inode_info),
1333 0, SLAB_ACCOUNT, init_once);
1334 if (hugetlbfs_inode_cachep == NULL)
1337 error = register_filesystem(&hugetlbfs_fs_type);
1342 for_each_hstate(h) {
1344 unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10);
1346 snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb);
1347 hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type,
1350 if (IS_ERR(hugetlbfs_vfsmount[i])) {
1351 pr_err("Cannot mount internal hugetlbfs for "
1352 "page size %uK", ps_kb);
1353 error = PTR_ERR(hugetlbfs_vfsmount[i]);
1354 hugetlbfs_vfsmount[i] = NULL;
1358 /* Non default hstates are optional */
1359 if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx]))
1363 kmem_cache_destroy(hugetlbfs_inode_cachep);
1367 fs_initcall(init_hugetlbfs_fs)