2 * hugetlbpage-backed filesystem. Based on ramfs.
4 * Nadia Yvette Chambers, 2002
6 * Copyright (C) 2002 Linus Torvalds.
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/module.h>
12 #include <linux/thread_info.h>
13 #include <asm/current.h>
14 #include <linux/sched.h> /* remove ASAP */
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/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 <asm/uaccess.h>
41 static const struct super_operations hugetlbfs_ops
;
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 struct hugetlbfs_config
{
53 struct hstate
*hstate
;
56 struct hugetlbfs_inode_info
{
57 struct shared_policy policy
;
58 struct inode vfs_inode
;
61 static inline struct hugetlbfs_inode_info
*HUGETLBFS_I(struct inode
*inode
)
63 return container_of(inode
, struct hugetlbfs_inode_info
, vfs_inode
);
66 int sysctl_hugetlb_shm_group
;
69 Opt_size
, Opt_nr_inodes
,
70 Opt_mode
, Opt_uid
, Opt_gid
,
75 static const match_table_t tokens
= {
76 {Opt_size
, "size=%s"},
77 {Opt_nr_inodes
, "nr_inodes=%s"},
78 {Opt_mode
, "mode=%o"},
81 {Opt_pagesize
, "pagesize=%s"},
85 static void huge_pagevec_release(struct pagevec
*pvec
)
89 for (i
= 0; i
< pagevec_count(pvec
); ++i
)
90 put_page(pvec
->pages
[i
]);
95 static int hugetlbfs_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
97 struct inode
*inode
= file_inode(file
);
100 struct hstate
*h
= hstate_file(file
);
103 * vma address alignment (but not the pgoff alignment) has
104 * already been checked by prepare_hugepage_range. If you add
105 * any error returns here, do so after setting VM_HUGETLB, so
106 * is_vm_hugetlb_page tests below unmap_region go the right
107 * way when do_mmap_pgoff unwinds (may be important on powerpc
110 vma
->vm_flags
|= VM_HUGETLB
| VM_DONTEXPAND
;
111 vma
->vm_ops
= &hugetlb_vm_ops
;
113 if (vma
->vm_pgoff
& (~huge_page_mask(h
) >> PAGE_SHIFT
))
116 vma_len
= (loff_t
)(vma
->vm_end
- vma
->vm_start
);
118 mutex_lock(&inode
->i_mutex
);
122 len
= vma_len
+ ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
124 if (hugetlb_reserve_pages(inode
,
125 vma
->vm_pgoff
>> huge_page_order(h
),
126 len
>> huge_page_shift(h
), vma
,
131 hugetlb_prefault_arch_hook(vma
->vm_mm
);
132 if (vma
->vm_flags
& VM_WRITE
&& inode
->i_size
< len
)
135 mutex_unlock(&inode
->i_mutex
);
141 * Called under down_write(mmap_sem).
144 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
146 hugetlb_get_unmapped_area(struct file
*file
, unsigned long addr
,
147 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
149 struct mm_struct
*mm
= current
->mm
;
150 struct vm_area_struct
*vma
;
151 struct hstate
*h
= hstate_file(file
);
152 struct vm_unmapped_area_info info
;
154 if (len
& ~huge_page_mask(h
))
159 if (flags
& MAP_FIXED
) {
160 if (prepare_hugepage_range(file
, addr
, len
))
166 addr
= ALIGN(addr
, huge_page_size(h
));
167 vma
= find_vma(mm
, addr
);
168 if (TASK_SIZE
- len
>= addr
&&
169 (!vma
|| addr
+ len
<= vma
->vm_start
))
175 info
.low_limit
= TASK_UNMAPPED_BASE
;
176 info
.high_limit
= TASK_SIZE
;
177 info
.align_mask
= PAGE_MASK
& ~huge_page_mask(h
);
178 info
.align_offset
= 0;
179 return vm_unmapped_area(&info
);
184 hugetlbfs_read_actor(struct page
*page
, unsigned long offset
,
185 struct iov_iter
*to
, unsigned long size
)
190 /* Find which 4k chunk and offset with in that chunk */
191 i
= offset
>> PAGE_CACHE_SHIFT
;
192 offset
= offset
& ~PAGE_CACHE_MASK
;
196 chunksize
= PAGE_CACHE_SIZE
;
199 if (chunksize
> size
)
201 n
= copy_page_to_iter(&page
[i
], offset
, chunksize
, to
);
213 * Support for read() - Find the page attached to f_mapping and copy out the
214 * data. Its *very* similar to do_generic_mapping_read(), we can't use that
215 * since it has PAGE_CACHE_SIZE assumptions.
217 static ssize_t
hugetlbfs_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
219 struct file
*file
= iocb
->ki_filp
;
220 struct hstate
*h
= hstate_file(file
);
221 struct address_space
*mapping
= file
->f_mapping
;
222 struct inode
*inode
= mapping
->host
;
223 unsigned long index
= iocb
->ki_pos
>> huge_page_shift(h
);
224 unsigned long offset
= iocb
->ki_pos
& ~huge_page_mask(h
);
225 unsigned long end_index
;
229 while (iov_iter_count(to
)) {
233 /* nr is the maximum number of bytes to copy from this page */
234 nr
= huge_page_size(h
);
235 isize
= i_size_read(inode
);
238 end_index
= (isize
- 1) >> huge_page_shift(h
);
239 if (index
> end_index
)
241 if (index
== end_index
) {
242 nr
= ((isize
- 1) & ~huge_page_mask(h
)) + 1;
249 page
= find_lock_page(mapping
, index
);
250 if (unlikely(page
== NULL
)) {
252 * We have a HOLE, zero out the user-buffer for the
253 * length of the hole or request.
255 copied
= iov_iter_zero(nr
, to
);
260 * We have the page, copy it to user space buffer.
262 copied
= hugetlbfs_read_actor(page
, offset
, to
, nr
);
263 page_cache_release(page
);
267 if (copied
!= nr
&& iov_iter_count(to
)) {
272 index
+= offset
>> huge_page_shift(h
);
273 offset
&= ~huge_page_mask(h
);
275 iocb
->ki_pos
= ((loff_t
)index
<< huge_page_shift(h
)) + offset
;
279 static int hugetlbfs_write_begin(struct file
*file
,
280 struct address_space
*mapping
,
281 loff_t pos
, unsigned len
, unsigned flags
,
282 struct page
**pagep
, void **fsdata
)
287 static int hugetlbfs_write_end(struct file
*file
, struct address_space
*mapping
,
288 loff_t pos
, unsigned len
, unsigned copied
,
289 struct page
*page
, void *fsdata
)
295 static void truncate_huge_page(struct page
*page
)
297 cancel_dirty_page(page
, /* No IO accounting for huge pages? */0);
298 ClearPageUptodate(page
);
299 delete_from_page_cache(page
);
302 static void truncate_hugepages(struct inode
*inode
, loff_t lstart
)
304 struct hstate
*h
= hstate_inode(inode
);
305 struct address_space
*mapping
= &inode
->i_data
;
306 const pgoff_t start
= lstart
>> huge_page_shift(h
);
311 pagevec_init(&pvec
, 0);
314 if (!pagevec_lookup(&pvec
, mapping
, next
, PAGEVEC_SIZE
)) {
321 for (i
= 0; i
< pagevec_count(&pvec
); ++i
) {
322 struct page
*page
= pvec
.pages
[i
];
325 if (page
->index
> next
)
328 truncate_huge_page(page
);
332 huge_pagevec_release(&pvec
);
334 BUG_ON(!lstart
&& mapping
->nrpages
);
335 hugetlb_unreserve_pages(inode
, start
, freed
);
338 static void hugetlbfs_evict_inode(struct inode
*inode
)
340 struct resv_map
*resv_map
;
342 truncate_hugepages(inode
, 0);
343 resv_map
= (struct resv_map
*)inode
->i_mapping
->private_data
;
344 /* root inode doesn't have the resv_map, so we should check it */
346 resv_map_release(&resv_map
->refs
);
351 hugetlb_vmtruncate_list(struct rb_root
*root
, pgoff_t pgoff
)
353 struct vm_area_struct
*vma
;
355 vma_interval_tree_foreach(vma
, root
, pgoff
, ULONG_MAX
) {
356 unsigned long v_offset
;
359 * Can the expression below overflow on 32-bit arches?
360 * No, because the interval tree returns us only those vmas
361 * which overlap the truncated area starting at pgoff,
362 * and no vma on a 32-bit arch can span beyond the 4GB.
364 if (vma
->vm_pgoff
< pgoff
)
365 v_offset
= (pgoff
- vma
->vm_pgoff
) << PAGE_SHIFT
;
369 unmap_hugepage_range(vma
, vma
->vm_start
+ v_offset
,
374 static int hugetlb_vmtruncate(struct inode
*inode
, loff_t offset
)
377 struct address_space
*mapping
= inode
->i_mapping
;
378 struct hstate
*h
= hstate_inode(inode
);
380 BUG_ON(offset
& ~huge_page_mask(h
));
381 pgoff
= offset
>> PAGE_SHIFT
;
383 i_size_write(inode
, offset
);
384 i_mmap_lock_write(mapping
);
385 if (!RB_EMPTY_ROOT(&mapping
->i_mmap
))
386 hugetlb_vmtruncate_list(&mapping
->i_mmap
, pgoff
);
387 i_mmap_unlock_write(mapping
);
388 truncate_hugepages(inode
, offset
);
392 static int hugetlbfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
394 struct inode
*inode
= dentry
->d_inode
;
395 struct hstate
*h
= hstate_inode(inode
);
397 unsigned int ia_valid
= attr
->ia_valid
;
401 error
= inode_change_ok(inode
, attr
);
405 if (ia_valid
& ATTR_SIZE
) {
407 if (attr
->ia_size
& ~huge_page_mask(h
))
409 error
= hugetlb_vmtruncate(inode
, attr
->ia_size
);
414 setattr_copy(inode
, attr
);
415 mark_inode_dirty(inode
);
419 static struct inode
*hugetlbfs_get_root(struct super_block
*sb
,
420 struct hugetlbfs_config
*config
)
424 inode
= new_inode(sb
);
426 struct hugetlbfs_inode_info
*info
;
427 inode
->i_ino
= get_next_ino();
428 inode
->i_mode
= S_IFDIR
| config
->mode
;
429 inode
->i_uid
= config
->uid
;
430 inode
->i_gid
= config
->gid
;
431 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
432 info
= HUGETLBFS_I(inode
);
433 mpol_shared_policy_init(&info
->policy
, NULL
);
434 inode
->i_op
= &hugetlbfs_dir_inode_operations
;
435 inode
->i_fop
= &simple_dir_operations
;
436 /* directory inodes start off with i_nlink == 2 (for "." entry) */
438 lockdep_annotate_inode_mutex_key(inode
);
444 * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
445 * be taken from reclaim -- unlike regular filesystems. This needs an
446 * annotation because huge_pmd_share() does an allocation under
449 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key
;
451 static struct inode
*hugetlbfs_get_inode(struct super_block
*sb
,
453 umode_t mode
, dev_t dev
)
456 struct resv_map
*resv_map
;
458 resv_map
= resv_map_alloc();
462 inode
= new_inode(sb
);
464 struct hugetlbfs_inode_info
*info
;
465 inode
->i_ino
= get_next_ino();
466 inode_init_owner(inode
, dir
, mode
);
467 lockdep_set_class(&inode
->i_mapping
->i_mmap_rwsem
,
468 &hugetlbfs_i_mmap_rwsem_key
);
469 inode
->i_mapping
->a_ops
= &hugetlbfs_aops
;
470 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
471 inode
->i_mapping
->private_data
= resv_map
;
472 info
= HUGETLBFS_I(inode
);
474 * The policy is initialized here even if we are creating a
475 * private inode because initialization simply creates an
476 * an empty rb tree and calls spin_lock_init(), later when we
477 * call mpol_free_shared_policy() it will just return because
478 * the rb tree will still be empty.
480 mpol_shared_policy_init(&info
->policy
, NULL
);
481 switch (mode
& S_IFMT
) {
483 init_special_inode(inode
, mode
, dev
);
486 inode
->i_op
= &hugetlbfs_inode_operations
;
487 inode
->i_fop
= &hugetlbfs_file_operations
;
490 inode
->i_op
= &hugetlbfs_dir_inode_operations
;
491 inode
->i_fop
= &simple_dir_operations
;
493 /* directory inodes start off with i_nlink == 2 (for "." entry) */
497 inode
->i_op
= &page_symlink_inode_operations
;
500 lockdep_annotate_inode_mutex_key(inode
);
502 kref_put(&resv_map
->refs
, resv_map_release
);
508 * File creation. Allocate an inode, and we're done..
510 static int hugetlbfs_mknod(struct inode
*dir
,
511 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
516 inode
= hugetlbfs_get_inode(dir
->i_sb
, dir
, mode
, dev
);
518 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
519 d_instantiate(dentry
, inode
);
520 dget(dentry
); /* Extra count - pin the dentry in core */
526 static int hugetlbfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
528 int retval
= hugetlbfs_mknod(dir
, dentry
, mode
| S_IFDIR
, 0);
534 static int hugetlbfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, bool excl
)
536 return hugetlbfs_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
539 static int hugetlbfs_symlink(struct inode
*dir
,
540 struct dentry
*dentry
, const char *symname
)
545 inode
= hugetlbfs_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0);
547 int l
= strlen(symname
)+1;
548 error
= page_symlink(inode
, symname
, l
);
550 d_instantiate(dentry
, inode
);
555 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
561 * mark the head page dirty
563 static int hugetlbfs_set_page_dirty(struct page
*page
)
565 struct page
*head
= compound_head(page
);
571 static int hugetlbfs_migrate_page(struct address_space
*mapping
,
572 struct page
*newpage
, struct page
*page
,
573 enum migrate_mode mode
)
577 rc
= migrate_huge_page_move_mapping(mapping
, newpage
, page
);
578 if (rc
!= MIGRATEPAGE_SUCCESS
)
580 migrate_page_copy(newpage
, page
);
582 return MIGRATEPAGE_SUCCESS
;
585 static int hugetlbfs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
587 struct hugetlbfs_sb_info
*sbinfo
= HUGETLBFS_SB(dentry
->d_sb
);
588 struct hstate
*h
= hstate_inode(dentry
->d_inode
);
590 buf
->f_type
= HUGETLBFS_MAGIC
;
591 buf
->f_bsize
= huge_page_size(h
);
593 spin_lock(&sbinfo
->stat_lock
);
594 /* If no limits set, just report 0 for max/free/used
595 * blocks, like simple_statfs() */
599 spin_lock(&sbinfo
->spool
->lock
);
600 buf
->f_blocks
= sbinfo
->spool
->max_hpages
;
601 free_pages
= sbinfo
->spool
->max_hpages
602 - sbinfo
->spool
->used_hpages
;
603 buf
->f_bavail
= buf
->f_bfree
= free_pages
;
604 spin_unlock(&sbinfo
->spool
->lock
);
605 buf
->f_files
= sbinfo
->max_inodes
;
606 buf
->f_ffree
= sbinfo
->free_inodes
;
608 spin_unlock(&sbinfo
->stat_lock
);
610 buf
->f_namelen
= NAME_MAX
;
614 static void hugetlbfs_put_super(struct super_block
*sb
)
616 struct hugetlbfs_sb_info
*sbi
= HUGETLBFS_SB(sb
);
619 sb
->s_fs_info
= NULL
;
622 hugepage_put_subpool(sbi
->spool
);
628 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info
*sbinfo
)
630 if (sbinfo
->free_inodes
>= 0) {
631 spin_lock(&sbinfo
->stat_lock
);
632 if (unlikely(!sbinfo
->free_inodes
)) {
633 spin_unlock(&sbinfo
->stat_lock
);
636 sbinfo
->free_inodes
--;
637 spin_unlock(&sbinfo
->stat_lock
);
643 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info
*sbinfo
)
645 if (sbinfo
->free_inodes
>= 0) {
646 spin_lock(&sbinfo
->stat_lock
);
647 sbinfo
->free_inodes
++;
648 spin_unlock(&sbinfo
->stat_lock
);
653 static struct kmem_cache
*hugetlbfs_inode_cachep
;
655 static struct inode
*hugetlbfs_alloc_inode(struct super_block
*sb
)
657 struct hugetlbfs_sb_info
*sbinfo
= HUGETLBFS_SB(sb
);
658 struct hugetlbfs_inode_info
*p
;
660 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo
)))
662 p
= kmem_cache_alloc(hugetlbfs_inode_cachep
, GFP_KERNEL
);
664 hugetlbfs_inc_free_inodes(sbinfo
);
667 return &p
->vfs_inode
;
670 static void hugetlbfs_i_callback(struct rcu_head
*head
)
672 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
673 kmem_cache_free(hugetlbfs_inode_cachep
, HUGETLBFS_I(inode
));
676 static void hugetlbfs_destroy_inode(struct inode
*inode
)
678 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode
->i_sb
));
679 mpol_free_shared_policy(&HUGETLBFS_I(inode
)->policy
);
680 call_rcu(&inode
->i_rcu
, hugetlbfs_i_callback
);
683 static const struct address_space_operations hugetlbfs_aops
= {
684 .write_begin
= hugetlbfs_write_begin
,
685 .write_end
= hugetlbfs_write_end
,
686 .set_page_dirty
= hugetlbfs_set_page_dirty
,
687 .migratepage
= hugetlbfs_migrate_page
,
691 static void init_once(void *foo
)
693 struct hugetlbfs_inode_info
*ei
= (struct hugetlbfs_inode_info
*)foo
;
695 inode_init_once(&ei
->vfs_inode
);
698 const struct file_operations hugetlbfs_file_operations
= {
699 .read
= new_sync_read
,
700 .read_iter
= hugetlbfs_read_iter
,
701 .mmap
= hugetlbfs_file_mmap
,
703 .get_unmapped_area
= hugetlb_get_unmapped_area
,
704 .llseek
= default_llseek
,
707 static const struct inode_operations hugetlbfs_dir_inode_operations
= {
708 .create
= hugetlbfs_create
,
709 .lookup
= simple_lookup
,
711 .unlink
= simple_unlink
,
712 .symlink
= hugetlbfs_symlink
,
713 .mkdir
= hugetlbfs_mkdir
,
714 .rmdir
= simple_rmdir
,
715 .mknod
= hugetlbfs_mknod
,
716 .rename
= simple_rename
,
717 .setattr
= hugetlbfs_setattr
,
720 static const struct inode_operations hugetlbfs_inode_operations
= {
721 .setattr
= hugetlbfs_setattr
,
724 static const struct super_operations hugetlbfs_ops
= {
725 .alloc_inode
= hugetlbfs_alloc_inode
,
726 .destroy_inode
= hugetlbfs_destroy_inode
,
727 .evict_inode
= hugetlbfs_evict_inode
,
728 .statfs
= hugetlbfs_statfs
,
729 .put_super
= hugetlbfs_put_super
,
730 .show_options
= generic_show_options
,
734 hugetlbfs_parse_options(char *options
, struct hugetlbfs_config
*pconfig
)
737 substring_t args
[MAX_OPT_ARGS
];
739 unsigned long long size
= 0;
740 enum { NO_SIZE
, SIZE_STD
, SIZE_PERCENT
} setsize
= NO_SIZE
;
745 while ((p
= strsep(&options
, ",")) != NULL
) {
750 token
= match_token(p
, tokens
, args
);
753 if (match_int(&args
[0], &option
))
755 pconfig
->uid
= make_kuid(current_user_ns(), option
);
756 if (!uid_valid(pconfig
->uid
))
761 if (match_int(&args
[0], &option
))
763 pconfig
->gid
= make_kgid(current_user_ns(), option
);
764 if (!gid_valid(pconfig
->gid
))
769 if (match_octal(&args
[0], &option
))
771 pconfig
->mode
= option
& 01777U;
775 /* memparse() will accept a K/M/G without a digit */
776 if (!isdigit(*args
[0].from
))
778 size
= memparse(args
[0].from
, &rest
);
781 setsize
= SIZE_PERCENT
;
786 /* memparse() will accept a K/M/G without a digit */
787 if (!isdigit(*args
[0].from
))
789 pconfig
->nr_inodes
= memparse(args
[0].from
, &rest
);
794 ps
= memparse(args
[0].from
, &rest
);
795 pconfig
->hstate
= size_to_hstate(ps
);
796 if (!pconfig
->hstate
) {
797 pr_err("Unsupported page size %lu MB\n",
805 pr_err("Bad mount option: \"%s\"\n", p
);
811 /* Do size after hstate is set up */
812 if (setsize
> NO_SIZE
) {
813 struct hstate
*h
= pconfig
->hstate
;
814 if (setsize
== SIZE_PERCENT
) {
815 size
<<= huge_page_shift(h
);
816 size
*= h
->max_huge_pages
;
819 pconfig
->nr_blocks
= (size
>> huge_page_shift(h
));
825 pr_err("Bad value '%s' for mount option '%s'\n", args
[0].from
, p
);
830 hugetlbfs_fill_super(struct super_block
*sb
, void *data
, int silent
)
833 struct hugetlbfs_config config
;
834 struct hugetlbfs_sb_info
*sbinfo
;
836 save_mount_options(sb
, data
);
838 config
.nr_blocks
= -1; /* No limit on size by default */
839 config
.nr_inodes
= -1; /* No limit on number of inodes by default */
840 config
.uid
= current_fsuid();
841 config
.gid
= current_fsgid();
843 config
.hstate
= &default_hstate
;
844 ret
= hugetlbfs_parse_options(data
, &config
);
848 sbinfo
= kmalloc(sizeof(struct hugetlbfs_sb_info
), GFP_KERNEL
);
851 sb
->s_fs_info
= sbinfo
;
852 sbinfo
->hstate
= config
.hstate
;
853 spin_lock_init(&sbinfo
->stat_lock
);
854 sbinfo
->max_inodes
= config
.nr_inodes
;
855 sbinfo
->free_inodes
= config
.nr_inodes
;
856 sbinfo
->spool
= NULL
;
857 if (config
.nr_blocks
!= -1) {
858 sbinfo
->spool
= hugepage_new_subpool(config
.nr_blocks
);
862 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
863 sb
->s_blocksize
= huge_page_size(config
.hstate
);
864 sb
->s_blocksize_bits
= huge_page_shift(config
.hstate
);
865 sb
->s_magic
= HUGETLBFS_MAGIC
;
866 sb
->s_op
= &hugetlbfs_ops
;
868 sb
->s_root
= d_make_root(hugetlbfs_get_root(sb
, &config
));
873 kfree(sbinfo
->spool
);
878 static struct dentry
*hugetlbfs_mount(struct file_system_type
*fs_type
,
879 int flags
, const char *dev_name
, void *data
)
881 return mount_nodev(fs_type
, flags
, data
, hugetlbfs_fill_super
);
884 static struct file_system_type hugetlbfs_fs_type
= {
886 .mount
= hugetlbfs_mount
,
887 .kill_sb
= kill_litter_super
,
889 MODULE_ALIAS_FS("hugetlbfs");
891 static struct vfsmount
*hugetlbfs_vfsmount
[HUGE_MAX_HSTATE
];
893 static int can_do_hugetlb_shm(void)
896 shm_group
= make_kgid(&init_user_ns
, sysctl_hugetlb_shm_group
);
897 return capable(CAP_IPC_LOCK
) || in_group_p(shm_group
);
900 static int get_hstate_idx(int page_size_log
)
902 struct hstate
*h
= hstate_sizelog(page_size_log
);
909 static const struct dentry_operations anon_ops
= {
910 .d_dname
= simple_dname
914 * Note that size should be aligned to proper hugepage size in caller side,
915 * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
917 struct file
*hugetlb_file_setup(const char *name
, size_t size
,
918 vm_flags_t acctflag
, struct user_struct
**user
,
919 int creat_flags
, int page_size_log
)
921 struct file
*file
= ERR_PTR(-ENOMEM
);
924 struct super_block
*sb
;
925 struct qstr quick_string
;
928 hstate_idx
= get_hstate_idx(page_size_log
);
930 return ERR_PTR(-ENODEV
);
933 if (!hugetlbfs_vfsmount
[hstate_idx
])
934 return ERR_PTR(-ENOENT
);
936 if (creat_flags
== HUGETLB_SHMFS_INODE
&& !can_do_hugetlb_shm()) {
937 *user
= current_user();
938 if (user_shm_lock(size
, *user
)) {
940 pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
941 current
->comm
, current
->pid
);
942 task_unlock(current
);
945 return ERR_PTR(-EPERM
);
949 sb
= hugetlbfs_vfsmount
[hstate_idx
]->mnt_sb
;
950 quick_string
.name
= name
;
951 quick_string
.len
= strlen(quick_string
.name
);
952 quick_string
.hash
= 0;
953 path
.dentry
= d_alloc_pseudo(sb
, &quick_string
);
957 d_set_d_op(path
.dentry
, &anon_ops
);
958 path
.mnt
= mntget(hugetlbfs_vfsmount
[hstate_idx
]);
959 file
= ERR_PTR(-ENOSPC
);
960 inode
= hugetlbfs_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0);
964 file
= ERR_PTR(-ENOMEM
);
965 if (hugetlb_reserve_pages(inode
, 0,
966 size
>> huge_page_shift(hstate_inode(inode
)), NULL
,
970 d_instantiate(path
.dentry
, inode
);
971 inode
->i_size
= size
;
974 file
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
975 &hugetlbfs_file_operations
);
977 goto out_dentry
; /* inode is already attached */
987 user_shm_unlock(size
, *user
);
993 static int __init
init_hugetlbfs_fs(void)
999 if (!hugepages_supported()) {
1000 pr_info("disabling because there are no supported hugepage sizes\n");
1005 hugetlbfs_inode_cachep
= kmem_cache_create("hugetlbfs_inode_cache",
1006 sizeof(struct hugetlbfs_inode_info
),
1008 if (hugetlbfs_inode_cachep
== NULL
)
1011 error
= register_filesystem(&hugetlbfs_fs_type
);
1016 for_each_hstate(h
) {
1018 unsigned ps_kb
= 1U << (h
->order
+ PAGE_SHIFT
- 10);
1020 snprintf(buf
, sizeof(buf
), "pagesize=%uK", ps_kb
);
1021 hugetlbfs_vfsmount
[i
] = kern_mount_data(&hugetlbfs_fs_type
,
1024 if (IS_ERR(hugetlbfs_vfsmount
[i
])) {
1025 pr_err("Cannot mount internal hugetlbfs for "
1026 "page size %uK", ps_kb
);
1027 error
= PTR_ERR(hugetlbfs_vfsmount
[i
]);
1028 hugetlbfs_vfsmount
[i
] = NULL
;
1032 /* Non default hstates are optional */
1033 if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount
[default_hstate_idx
]))
1037 kmem_cache_destroy(hugetlbfs_inode_cachep
);
1042 static void __exit
exit_hugetlbfs_fs(void)
1049 * Make sure all delayed rcu free inodes are flushed before we
1053 kmem_cache_destroy(hugetlbfs_inode_cachep
);
1056 kern_unmount(hugetlbfs_vfsmount
[i
++]);
1057 unregister_filesystem(&hugetlbfs_fs_type
);
1060 module_init(init_hugetlbfs_fs
)
1061 module_exit(exit_hugetlbfs_fs
)
1063 MODULE_LICENSE("GPL");