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>
38 #include <asm/uaccess.h>
40 static const struct super_operations hugetlbfs_ops
;
41 static const struct address_space_operations hugetlbfs_aops
;
42 const struct file_operations hugetlbfs_file_operations
;
43 static const struct inode_operations hugetlbfs_dir_inode_operations
;
44 static const struct inode_operations hugetlbfs_inode_operations
;
46 struct hugetlbfs_config
{
52 struct hstate
*hstate
;
55 struct hugetlbfs_inode_info
{
56 struct shared_policy policy
;
57 struct inode vfs_inode
;
60 static inline struct hugetlbfs_inode_info
*HUGETLBFS_I(struct inode
*inode
)
62 return container_of(inode
, struct hugetlbfs_inode_info
, vfs_inode
);
65 int sysctl_hugetlb_shm_group
;
68 Opt_size
, Opt_nr_inodes
,
69 Opt_mode
, Opt_uid
, Opt_gid
,
74 static const match_table_t tokens
= {
75 {Opt_size
, "size=%s"},
76 {Opt_nr_inodes
, "nr_inodes=%s"},
77 {Opt_mode
, "mode=%o"},
80 {Opt_pagesize
, "pagesize=%s"},
84 static void huge_pagevec_release(struct pagevec
*pvec
)
88 for (i
= 0; i
< pagevec_count(pvec
); ++i
)
89 put_page(pvec
->pages
[i
]);
94 static int hugetlbfs_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
96 struct inode
*inode
= file_inode(file
);
99 struct hstate
*h
= hstate_file(file
);
102 * vma address alignment (but not the pgoff alignment) has
103 * already been checked by prepare_hugepage_range. If you add
104 * any error returns here, do so after setting VM_HUGETLB, so
105 * is_vm_hugetlb_page tests below unmap_region go the right
106 * way when do_mmap_pgoff unwinds (may be important on powerpc
109 vma
->vm_flags
|= VM_HUGETLB
| VM_DONTEXPAND
;
110 vma
->vm_ops
= &hugetlb_vm_ops
;
112 if (vma
->vm_pgoff
& (~huge_page_mask(h
) >> PAGE_SHIFT
))
115 vma_len
= (loff_t
)(vma
->vm_end
- vma
->vm_start
);
117 mutex_lock(&inode
->i_mutex
);
121 len
= vma_len
+ ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
123 if (hugetlb_reserve_pages(inode
,
124 vma
->vm_pgoff
>> huge_page_order(h
),
125 len
>> huge_page_shift(h
), vma
,
130 hugetlb_prefault_arch_hook(vma
->vm_mm
);
131 if (vma
->vm_flags
& VM_WRITE
&& inode
->i_size
< len
)
134 mutex_unlock(&inode
->i_mutex
);
140 * Called under down_write(mmap_sem).
143 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
145 hugetlb_get_unmapped_area(struct file
*file
, unsigned long addr
,
146 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
148 struct mm_struct
*mm
= current
->mm
;
149 struct vm_area_struct
*vma
;
150 struct hstate
*h
= hstate_file(file
);
151 struct vm_unmapped_area_info info
;
153 if (len
& ~huge_page_mask(h
))
158 if (flags
& MAP_FIXED
) {
159 if (prepare_hugepage_range(file
, addr
, len
))
165 addr
= ALIGN(addr
, huge_page_size(h
));
166 vma
= find_vma(mm
, addr
);
167 if (TASK_SIZE
- len
>= addr
&&
168 (!vma
|| addr
+ len
<= vma
->vm_start
))
174 info
.low_limit
= TASK_UNMAPPED_BASE
;
175 info
.high_limit
= TASK_SIZE
;
176 info
.align_mask
= PAGE_MASK
& ~huge_page_mask(h
);
177 info
.align_offset
= 0;
178 return vm_unmapped_area(&info
);
183 hugetlbfs_read_actor(struct page
*page
, unsigned long offset
,
184 char __user
*buf
, unsigned long count
,
188 unsigned long left
, copied
= 0;
194 /* Find which 4k chunk and offset with in that chunk */
195 i
= offset
>> PAGE_CACHE_SHIFT
;
196 offset
= offset
& ~PAGE_CACHE_MASK
;
199 chunksize
= PAGE_CACHE_SIZE
;
202 if (chunksize
> size
)
204 kaddr
= kmap(&page
[i
]);
205 left
= __copy_to_user(buf
, kaddr
+ offset
, chunksize
);
208 copied
+= (chunksize
- left
);
217 return copied
? copied
: -EFAULT
;
221 * Support for read() - Find the page attached to f_mapping and copy out the
222 * data. Its *very* similar to do_generic_mapping_read(), we can't use that
223 * since it has PAGE_CACHE_SIZE assumptions.
225 static ssize_t
hugetlbfs_read(struct file
*filp
, char __user
*buf
,
226 size_t len
, loff_t
*ppos
)
228 struct hstate
*h
= hstate_file(filp
);
229 struct address_space
*mapping
= filp
->f_mapping
;
230 struct inode
*inode
= mapping
->host
;
231 unsigned long index
= *ppos
>> huge_page_shift(h
);
232 unsigned long offset
= *ppos
& ~huge_page_mask(h
);
233 unsigned long end_index
;
237 /* validate length */
243 unsigned long nr
, ret
;
246 /* nr is the maximum number of bytes to copy from this page */
247 nr
= huge_page_size(h
);
248 isize
= i_size_read(inode
);
251 end_index
= (isize
- 1) >> huge_page_shift(h
);
252 if (index
>= end_index
) {
253 if (index
> end_index
)
255 nr
= ((isize
- 1) & ~huge_page_mask(h
)) + 1;
262 page
= find_lock_page(mapping
, index
);
263 if (unlikely(page
== NULL
)) {
265 * We have a HOLE, zero out the user-buffer for the
266 * length of the hole or request.
268 ret
= len
< nr
? len
: nr
;
269 if (clear_user(buf
, ret
))
277 * We have the page, copy it to user space buffer.
279 ra
= hugetlbfs_read_actor(page
, offset
, buf
, len
, nr
);
281 page_cache_release(page
);
292 index
+= offset
>> huge_page_shift(h
);
293 offset
&= ~huge_page_mask(h
);
295 /* short read or no more work */
296 if ((ret
!= nr
) || (len
== 0))
300 *ppos
= ((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 truncate_huge_page(struct page
*page
)
322 cancel_dirty_page(page
, /* No IO accounting for huge pages? */0);
323 ClearPageUptodate(page
);
324 delete_from_page_cache(page
);
327 static void truncate_hugepages(struct inode
*inode
, loff_t lstart
)
329 struct hstate
*h
= hstate_inode(inode
);
330 struct address_space
*mapping
= &inode
->i_data
;
331 const pgoff_t start
= lstart
>> huge_page_shift(h
);
336 pagevec_init(&pvec
, 0);
339 if (!pagevec_lookup(&pvec
, mapping
, next
, PAGEVEC_SIZE
)) {
346 for (i
= 0; i
< pagevec_count(&pvec
); ++i
) {
347 struct page
*page
= pvec
.pages
[i
];
350 if (page
->index
> next
)
353 truncate_huge_page(page
);
357 huge_pagevec_release(&pvec
);
359 BUG_ON(!lstart
&& mapping
->nrpages
);
360 hugetlb_unreserve_pages(inode
, start
, freed
);
363 static void hugetlbfs_evict_inode(struct inode
*inode
)
365 struct resv_map
*resv_map
;
367 truncate_hugepages(inode
, 0);
368 resv_map
= (struct resv_map
*)inode
->i_mapping
->private_data
;
369 /* root inode doesn't have the resv_map, so we should check it */
371 resv_map_release(&resv_map
->refs
);
376 hugetlb_vmtruncate_list(struct rb_root
*root
, pgoff_t pgoff
)
378 struct vm_area_struct
*vma
;
380 vma_interval_tree_foreach(vma
, root
, pgoff
, ULONG_MAX
) {
381 unsigned long v_offset
;
384 * Can the expression below overflow on 32-bit arches?
385 * No, because the interval tree returns us only those vmas
386 * which overlap the truncated area starting at pgoff,
387 * and no vma on a 32-bit arch can span beyond the 4GB.
389 if (vma
->vm_pgoff
< pgoff
)
390 v_offset
= (pgoff
- vma
->vm_pgoff
) << PAGE_SHIFT
;
394 unmap_hugepage_range(vma
, vma
->vm_start
+ v_offset
,
399 static int hugetlb_vmtruncate(struct inode
*inode
, loff_t offset
)
402 struct address_space
*mapping
= inode
->i_mapping
;
403 struct hstate
*h
= hstate_inode(inode
);
405 BUG_ON(offset
& ~huge_page_mask(h
));
406 pgoff
= offset
>> PAGE_SHIFT
;
408 i_size_write(inode
, offset
);
409 i_mmap_lock_write(mapping
);
410 if (!RB_EMPTY_ROOT(&mapping
->i_mmap
))
411 hugetlb_vmtruncate_list(&mapping
->i_mmap
, pgoff
);
412 i_mmap_unlock_write(mapping
);
413 truncate_hugepages(inode
, offset
);
417 static int hugetlbfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
419 struct inode
*inode
= dentry
->d_inode
;
420 struct hstate
*h
= hstate_inode(inode
);
422 unsigned int ia_valid
= attr
->ia_valid
;
426 error
= inode_change_ok(inode
, attr
);
430 if (ia_valid
& ATTR_SIZE
) {
432 if (attr
->ia_size
& ~huge_page_mask(h
))
434 error
= hugetlb_vmtruncate(inode
, attr
->ia_size
);
439 setattr_copy(inode
, attr
);
440 mark_inode_dirty(inode
);
444 static struct inode
*hugetlbfs_get_root(struct super_block
*sb
,
445 struct hugetlbfs_config
*config
)
449 inode
= new_inode(sb
);
451 struct hugetlbfs_inode_info
*info
;
452 inode
->i_ino
= get_next_ino();
453 inode
->i_mode
= S_IFDIR
| config
->mode
;
454 inode
->i_uid
= config
->uid
;
455 inode
->i_gid
= config
->gid
;
456 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
457 info
= HUGETLBFS_I(inode
);
458 mpol_shared_policy_init(&info
->policy
, NULL
);
459 inode
->i_op
= &hugetlbfs_dir_inode_operations
;
460 inode
->i_fop
= &simple_dir_operations
;
461 /* directory inodes start off with i_nlink == 2 (for "." entry) */
463 lockdep_annotate_inode_mutex_key(inode
);
469 * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
470 * be taken from reclaim -- unlike regular filesystems. This needs an
471 * annotation because huge_pmd_share() does an allocation under
474 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key
;
476 static struct inode
*hugetlbfs_get_inode(struct super_block
*sb
,
478 umode_t mode
, dev_t dev
)
481 struct resv_map
*resv_map
;
483 resv_map
= resv_map_alloc();
487 inode
= new_inode(sb
);
489 struct hugetlbfs_inode_info
*info
;
490 inode
->i_ino
= get_next_ino();
491 inode_init_owner(inode
, dir
, mode
);
492 lockdep_set_class(&inode
->i_mapping
->i_mmap_rwsem
,
493 &hugetlbfs_i_mmap_rwsem_key
);
494 inode
->i_mapping
->a_ops
= &hugetlbfs_aops
;
495 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
496 inode
->i_mapping
->private_data
= resv_map
;
497 info
= HUGETLBFS_I(inode
);
499 * The policy is initialized here even if we are creating a
500 * private inode because initialization simply creates an
501 * an empty rb tree and calls spin_lock_init(), later when we
502 * call mpol_free_shared_policy() it will just return because
503 * the rb tree will still be empty.
505 mpol_shared_policy_init(&info
->policy
, NULL
);
506 switch (mode
& S_IFMT
) {
508 init_special_inode(inode
, mode
, dev
);
511 inode
->i_op
= &hugetlbfs_inode_operations
;
512 inode
->i_fop
= &hugetlbfs_file_operations
;
515 inode
->i_op
= &hugetlbfs_dir_inode_operations
;
516 inode
->i_fop
= &simple_dir_operations
;
518 /* directory inodes start off with i_nlink == 2 (for "." entry) */
522 inode
->i_op
= &page_symlink_inode_operations
;
525 lockdep_annotate_inode_mutex_key(inode
);
527 kref_put(&resv_map
->refs
, resv_map_release
);
533 * File creation. Allocate an inode, and we're done..
535 static int hugetlbfs_mknod(struct inode
*dir
,
536 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
541 inode
= hugetlbfs_get_inode(dir
->i_sb
, dir
, mode
, dev
);
543 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
544 d_instantiate(dentry
, inode
);
545 dget(dentry
); /* Extra count - pin the dentry in core */
551 static int hugetlbfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
553 int retval
= hugetlbfs_mknod(dir
, dentry
, mode
| S_IFDIR
, 0);
559 static int hugetlbfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, bool excl
)
561 return hugetlbfs_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
564 static int hugetlbfs_symlink(struct inode
*dir
,
565 struct dentry
*dentry
, const char *symname
)
570 inode
= hugetlbfs_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0);
572 int l
= strlen(symname
)+1;
573 error
= page_symlink(inode
, symname
, l
);
575 d_instantiate(dentry
, inode
);
580 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
586 * mark the head page dirty
588 static int hugetlbfs_set_page_dirty(struct page
*page
)
590 struct page
*head
= compound_head(page
);
596 static int hugetlbfs_migrate_page(struct address_space
*mapping
,
597 struct page
*newpage
, struct page
*page
,
598 enum migrate_mode mode
)
602 rc
= migrate_huge_page_move_mapping(mapping
, newpage
, page
);
603 if (rc
!= MIGRATEPAGE_SUCCESS
)
605 migrate_page_copy(newpage
, page
);
607 return MIGRATEPAGE_SUCCESS
;
610 static int hugetlbfs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
612 struct hugetlbfs_sb_info
*sbinfo
= HUGETLBFS_SB(dentry
->d_sb
);
613 struct hstate
*h
= hstate_inode(dentry
->d_inode
);
615 buf
->f_type
= HUGETLBFS_MAGIC
;
616 buf
->f_bsize
= huge_page_size(h
);
618 spin_lock(&sbinfo
->stat_lock
);
619 /* If no limits set, just report 0 for max/free/used
620 * blocks, like simple_statfs() */
624 spin_lock(&sbinfo
->spool
->lock
);
625 buf
->f_blocks
= sbinfo
->spool
->max_hpages
;
626 free_pages
= sbinfo
->spool
->max_hpages
627 - sbinfo
->spool
->used_hpages
;
628 buf
->f_bavail
= buf
->f_bfree
= free_pages
;
629 spin_unlock(&sbinfo
->spool
->lock
);
630 buf
->f_files
= sbinfo
->max_inodes
;
631 buf
->f_ffree
= sbinfo
->free_inodes
;
633 spin_unlock(&sbinfo
->stat_lock
);
635 buf
->f_namelen
= NAME_MAX
;
639 static void hugetlbfs_put_super(struct super_block
*sb
)
641 struct hugetlbfs_sb_info
*sbi
= HUGETLBFS_SB(sb
);
644 sb
->s_fs_info
= NULL
;
647 hugepage_put_subpool(sbi
->spool
);
653 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info
*sbinfo
)
655 if (sbinfo
->free_inodes
>= 0) {
656 spin_lock(&sbinfo
->stat_lock
);
657 if (unlikely(!sbinfo
->free_inodes
)) {
658 spin_unlock(&sbinfo
->stat_lock
);
661 sbinfo
->free_inodes
--;
662 spin_unlock(&sbinfo
->stat_lock
);
668 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info
*sbinfo
)
670 if (sbinfo
->free_inodes
>= 0) {
671 spin_lock(&sbinfo
->stat_lock
);
672 sbinfo
->free_inodes
++;
673 spin_unlock(&sbinfo
->stat_lock
);
678 static struct kmem_cache
*hugetlbfs_inode_cachep
;
680 static struct inode
*hugetlbfs_alloc_inode(struct super_block
*sb
)
682 struct hugetlbfs_sb_info
*sbinfo
= HUGETLBFS_SB(sb
);
683 struct hugetlbfs_inode_info
*p
;
685 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo
)))
687 p
= kmem_cache_alloc(hugetlbfs_inode_cachep
, GFP_KERNEL
);
689 hugetlbfs_inc_free_inodes(sbinfo
);
692 return &p
->vfs_inode
;
695 static void hugetlbfs_i_callback(struct rcu_head
*head
)
697 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
698 kmem_cache_free(hugetlbfs_inode_cachep
, HUGETLBFS_I(inode
));
701 static void hugetlbfs_destroy_inode(struct inode
*inode
)
703 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode
->i_sb
));
704 mpol_free_shared_policy(&HUGETLBFS_I(inode
)->policy
);
705 call_rcu(&inode
->i_rcu
, hugetlbfs_i_callback
);
708 static const struct address_space_operations hugetlbfs_aops
= {
709 .write_begin
= hugetlbfs_write_begin
,
710 .write_end
= hugetlbfs_write_end
,
711 .set_page_dirty
= hugetlbfs_set_page_dirty
,
712 .migratepage
= hugetlbfs_migrate_page
,
716 static void init_once(void *foo
)
718 struct hugetlbfs_inode_info
*ei
= (struct hugetlbfs_inode_info
*)foo
;
720 inode_init_once(&ei
->vfs_inode
);
723 const struct file_operations hugetlbfs_file_operations
= {
724 .read
= hugetlbfs_read
,
725 .mmap
= hugetlbfs_file_mmap
,
727 .get_unmapped_area
= hugetlb_get_unmapped_area
,
728 .llseek
= default_llseek
,
731 static const struct inode_operations hugetlbfs_dir_inode_operations
= {
732 .create
= hugetlbfs_create
,
733 .lookup
= simple_lookup
,
735 .unlink
= simple_unlink
,
736 .symlink
= hugetlbfs_symlink
,
737 .mkdir
= hugetlbfs_mkdir
,
738 .rmdir
= simple_rmdir
,
739 .mknod
= hugetlbfs_mknod
,
740 .rename
= simple_rename
,
741 .setattr
= hugetlbfs_setattr
,
744 static const struct inode_operations hugetlbfs_inode_operations
= {
745 .setattr
= hugetlbfs_setattr
,
748 static const struct super_operations hugetlbfs_ops
= {
749 .alloc_inode
= hugetlbfs_alloc_inode
,
750 .destroy_inode
= hugetlbfs_destroy_inode
,
751 .evict_inode
= hugetlbfs_evict_inode
,
752 .statfs
= hugetlbfs_statfs
,
753 .put_super
= hugetlbfs_put_super
,
754 .show_options
= generic_show_options
,
758 hugetlbfs_parse_options(char *options
, struct hugetlbfs_config
*pconfig
)
761 substring_t args
[MAX_OPT_ARGS
];
763 unsigned long long size
= 0;
764 enum { NO_SIZE
, SIZE_STD
, SIZE_PERCENT
} setsize
= NO_SIZE
;
769 while ((p
= strsep(&options
, ",")) != NULL
) {
774 token
= match_token(p
, tokens
, args
);
777 if (match_int(&args
[0], &option
))
779 pconfig
->uid
= make_kuid(current_user_ns(), option
);
780 if (!uid_valid(pconfig
->uid
))
785 if (match_int(&args
[0], &option
))
787 pconfig
->gid
= make_kgid(current_user_ns(), option
);
788 if (!gid_valid(pconfig
->gid
))
793 if (match_octal(&args
[0], &option
))
795 pconfig
->mode
= option
& 01777U;
799 /* memparse() will accept a K/M/G without a digit */
800 if (!isdigit(*args
[0].from
))
802 size
= memparse(args
[0].from
, &rest
);
805 setsize
= SIZE_PERCENT
;
810 /* memparse() will accept a K/M/G without a digit */
811 if (!isdigit(*args
[0].from
))
813 pconfig
->nr_inodes
= memparse(args
[0].from
, &rest
);
818 ps
= memparse(args
[0].from
, &rest
);
819 pconfig
->hstate
= size_to_hstate(ps
);
820 if (!pconfig
->hstate
) {
821 pr_err("Unsupported page size %lu MB\n",
829 pr_err("Bad mount option: \"%s\"\n", p
);
835 /* Do size after hstate is set up */
836 if (setsize
> NO_SIZE
) {
837 struct hstate
*h
= pconfig
->hstate
;
838 if (setsize
== SIZE_PERCENT
) {
839 size
<<= huge_page_shift(h
);
840 size
*= h
->max_huge_pages
;
843 pconfig
->nr_blocks
= (size
>> huge_page_shift(h
));
849 pr_err("Bad value '%s' for mount option '%s'\n", args
[0].from
, p
);
854 hugetlbfs_fill_super(struct super_block
*sb
, void *data
, int silent
)
857 struct hugetlbfs_config config
;
858 struct hugetlbfs_sb_info
*sbinfo
;
860 save_mount_options(sb
, data
);
862 config
.nr_blocks
= -1; /* No limit on size by default */
863 config
.nr_inodes
= -1; /* No limit on number of inodes by default */
864 config
.uid
= current_fsuid();
865 config
.gid
= current_fsgid();
867 config
.hstate
= &default_hstate
;
868 ret
= hugetlbfs_parse_options(data
, &config
);
872 sbinfo
= kmalloc(sizeof(struct hugetlbfs_sb_info
), GFP_KERNEL
);
875 sb
->s_fs_info
= sbinfo
;
876 sbinfo
->hstate
= config
.hstate
;
877 spin_lock_init(&sbinfo
->stat_lock
);
878 sbinfo
->max_inodes
= config
.nr_inodes
;
879 sbinfo
->free_inodes
= config
.nr_inodes
;
880 sbinfo
->spool
= NULL
;
881 if (config
.nr_blocks
!= -1) {
882 sbinfo
->spool
= hugepage_new_subpool(config
.nr_blocks
);
886 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
887 sb
->s_blocksize
= huge_page_size(config
.hstate
);
888 sb
->s_blocksize_bits
= huge_page_shift(config
.hstate
);
889 sb
->s_magic
= HUGETLBFS_MAGIC
;
890 sb
->s_op
= &hugetlbfs_ops
;
892 sb
->s_root
= d_make_root(hugetlbfs_get_root(sb
, &config
));
897 kfree(sbinfo
->spool
);
902 static struct dentry
*hugetlbfs_mount(struct file_system_type
*fs_type
,
903 int flags
, const char *dev_name
, void *data
)
905 return mount_nodev(fs_type
, flags
, data
, hugetlbfs_fill_super
);
908 static struct file_system_type hugetlbfs_fs_type
= {
910 .mount
= hugetlbfs_mount
,
911 .kill_sb
= kill_litter_super
,
913 MODULE_ALIAS_FS("hugetlbfs");
915 static struct vfsmount
*hugetlbfs_vfsmount
[HUGE_MAX_HSTATE
];
917 static int can_do_hugetlb_shm(void)
920 shm_group
= make_kgid(&init_user_ns
, sysctl_hugetlb_shm_group
);
921 return capable(CAP_IPC_LOCK
) || in_group_p(shm_group
);
924 static int get_hstate_idx(int page_size_log
)
926 struct hstate
*h
= hstate_sizelog(page_size_log
);
933 static const struct dentry_operations anon_ops
= {
934 .d_dname
= simple_dname
938 * Note that size should be aligned to proper hugepage size in caller side,
939 * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
941 struct file
*hugetlb_file_setup(const char *name
, size_t size
,
942 vm_flags_t acctflag
, struct user_struct
**user
,
943 int creat_flags
, int page_size_log
)
945 struct file
*file
= ERR_PTR(-ENOMEM
);
948 struct super_block
*sb
;
949 struct qstr quick_string
;
952 hstate_idx
= get_hstate_idx(page_size_log
);
954 return ERR_PTR(-ENODEV
);
957 if (!hugetlbfs_vfsmount
[hstate_idx
])
958 return ERR_PTR(-ENOENT
);
960 if (creat_flags
== HUGETLB_SHMFS_INODE
&& !can_do_hugetlb_shm()) {
961 *user
= current_user();
962 if (user_shm_lock(size
, *user
)) {
964 pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
965 current
->comm
, current
->pid
);
966 task_unlock(current
);
969 return ERR_PTR(-EPERM
);
973 sb
= hugetlbfs_vfsmount
[hstate_idx
]->mnt_sb
;
974 quick_string
.name
= name
;
975 quick_string
.len
= strlen(quick_string
.name
);
976 quick_string
.hash
= 0;
977 path
.dentry
= d_alloc_pseudo(sb
, &quick_string
);
981 d_set_d_op(path
.dentry
, &anon_ops
);
982 path
.mnt
= mntget(hugetlbfs_vfsmount
[hstate_idx
]);
983 file
= ERR_PTR(-ENOSPC
);
984 inode
= hugetlbfs_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0);
988 file
= ERR_PTR(-ENOMEM
);
989 if (hugetlb_reserve_pages(inode
, 0,
990 size
>> huge_page_shift(hstate_inode(inode
)), NULL
,
994 d_instantiate(path
.dentry
, inode
);
995 inode
->i_size
= size
;
998 file
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
999 &hugetlbfs_file_operations
);
1001 goto out_dentry
; /* inode is already attached */
1011 user_shm_unlock(size
, *user
);
1017 static int __init
init_hugetlbfs_fs(void)
1023 if (!hugepages_supported()) {
1024 pr_info("disabling because there are no supported hugepage sizes\n");
1029 hugetlbfs_inode_cachep
= kmem_cache_create("hugetlbfs_inode_cache",
1030 sizeof(struct hugetlbfs_inode_info
),
1032 if (hugetlbfs_inode_cachep
== NULL
)
1035 error
= register_filesystem(&hugetlbfs_fs_type
);
1040 for_each_hstate(h
) {
1042 unsigned ps_kb
= 1U << (h
->order
+ PAGE_SHIFT
- 10);
1044 snprintf(buf
, sizeof(buf
), "pagesize=%uK", ps_kb
);
1045 hugetlbfs_vfsmount
[i
] = kern_mount_data(&hugetlbfs_fs_type
,
1048 if (IS_ERR(hugetlbfs_vfsmount
[i
])) {
1049 pr_err("Cannot mount internal hugetlbfs for "
1050 "page size %uK", ps_kb
);
1051 error
= PTR_ERR(hugetlbfs_vfsmount
[i
]);
1052 hugetlbfs_vfsmount
[i
] = NULL
;
1056 /* Non default hstates are optional */
1057 if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount
[default_hstate_idx
]))
1061 kmem_cache_destroy(hugetlbfs_inode_cachep
);
1066 static void __exit
exit_hugetlbfs_fs(void)
1073 * Make sure all delayed rcu free inodes are flushed before we
1077 kmem_cache_destroy(hugetlbfs_inode_cachep
);
1080 kern_unmount(hugetlbfs_vfsmount
[i
++]);
1081 unregister_filesystem(&hugetlbfs_fs_type
);
1084 module_init(init_hugetlbfs_fs
)
1085 module_exit(exit_hugetlbfs_fs
)
1087 MODULE_LICENSE("GPL");