2 * hugetlbpage-backed filesystem. Based on ramfs.
6 * Copyright (C) 2002 Linus Torvalds.
9 #include <linux/module.h>
10 #include <linux/thread_info.h>
11 #include <asm/current.h>
12 #include <linux/sched.h> /* remove ASAP */
14 #include <linux/mount.h>
15 #include <linux/file.h>
16 #include <linux/kernel.h>
17 #include <linux/writeback.h>
18 #include <linux/pagemap.h>
19 #include <linux/highmem.h>
20 #include <linux/init.h>
21 #include <linux/string.h>
22 #include <linux/capability.h>
23 #include <linux/ctype.h>
24 #include <linux/backing-dev.h>
25 #include <linux/hugetlb.h>
26 #include <linux/pagevec.h>
27 #include <linux/parser.h>
28 #include <linux/mman.h>
29 #include <linux/slab.h>
30 #include <linux/dnotify.h>
31 #include <linux/statfs.h>
32 #include <linux/security.h>
33 #include <linux/magic.h>
34 #include <linux/migrate.h>
36 #include <asm/uaccess.h>
38 static const struct super_operations hugetlbfs_ops
;
39 static const struct address_space_operations hugetlbfs_aops
;
40 const struct file_operations hugetlbfs_file_operations
;
41 static const struct inode_operations hugetlbfs_dir_inode_operations
;
42 static const struct inode_operations hugetlbfs_inode_operations
;
44 static struct backing_dev_info hugetlbfs_backing_dev_info
= {
46 .ra_pages
= 0, /* No readahead */
47 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
50 int sysctl_hugetlb_shm_group
;
53 Opt_size
, Opt_nr_inodes
,
54 Opt_mode
, Opt_uid
, Opt_gid
,
59 static const match_table_t tokens
= {
60 {Opt_size
, "size=%s"},
61 {Opt_nr_inodes
, "nr_inodes=%s"},
62 {Opt_mode
, "mode=%o"},
65 {Opt_pagesize
, "pagesize=%s"},
69 static void huge_pagevec_release(struct pagevec
*pvec
)
73 for (i
= 0; i
< pagevec_count(pvec
); ++i
)
74 put_page(pvec
->pages
[i
]);
79 static int hugetlbfs_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
81 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
84 struct hstate
*h
= hstate_file(file
);
87 * vma address alignment (but not the pgoff alignment) has
88 * already been checked by prepare_hugepage_range. If you add
89 * any error returns here, do so after setting VM_HUGETLB, so
90 * is_vm_hugetlb_page tests below unmap_region go the right
91 * way when do_mmap_pgoff unwinds (may be important on powerpc
94 vma
->vm_flags
|= VM_HUGETLB
| VM_RESERVED
;
95 vma
->vm_ops
= &hugetlb_vm_ops
;
97 if (vma
->vm_pgoff
& (~huge_page_mask(h
) >> PAGE_SHIFT
))
100 vma_len
= (loff_t
)(vma
->vm_end
- vma
->vm_start
);
102 mutex_lock(&inode
->i_mutex
);
106 len
= vma_len
+ ((loff_t
)vma
->vm_pgoff
<< PAGE_SHIFT
);
108 if (hugetlb_reserve_pages(inode
,
109 vma
->vm_pgoff
>> huge_page_order(h
),
110 len
>> huge_page_shift(h
), vma
,
115 hugetlb_prefault_arch_hook(vma
->vm_mm
);
116 if (vma
->vm_flags
& VM_WRITE
&& inode
->i_size
< len
)
119 mutex_unlock(&inode
->i_mutex
);
125 * Called under down_write(mmap_sem).
128 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
130 hugetlb_get_unmapped_area(struct file
*file
, unsigned long addr
,
131 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
133 struct mm_struct
*mm
= current
->mm
;
134 struct vm_area_struct
*vma
;
135 unsigned long start_addr
;
136 struct hstate
*h
= hstate_file(file
);
138 if (len
& ~huge_page_mask(h
))
143 if (flags
& MAP_FIXED
) {
144 if (prepare_hugepage_range(file
, addr
, len
))
150 addr
= ALIGN(addr
, huge_page_size(h
));
151 vma
= find_vma(mm
, addr
);
152 if (TASK_SIZE
- len
>= addr
&&
153 (!vma
|| addr
+ len
<= vma
->vm_start
))
157 start_addr
= mm
->free_area_cache
;
159 if (len
<= mm
->cached_hole_size
)
160 start_addr
= TASK_UNMAPPED_BASE
;
163 addr
= ALIGN(start_addr
, huge_page_size(h
));
165 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
166 /* At this point: (!vma || addr < vma->vm_end). */
167 if (TASK_SIZE
- len
< addr
) {
169 * Start a new search - just in case we missed
172 if (start_addr
!= TASK_UNMAPPED_BASE
) {
173 start_addr
= TASK_UNMAPPED_BASE
;
179 if (!vma
|| addr
+ len
<= vma
->vm_start
)
181 addr
= ALIGN(vma
->vm_end
, huge_page_size(h
));
187 hugetlbfs_read_actor(struct page
*page
, unsigned long offset
,
188 char __user
*buf
, unsigned long count
,
192 unsigned long left
, copied
= 0;
198 /* Find which 4k chunk and offset with in that chunk */
199 i
= offset
>> PAGE_CACHE_SHIFT
;
200 offset
= offset
& ~PAGE_CACHE_MASK
;
203 chunksize
= PAGE_CACHE_SIZE
;
206 if (chunksize
> size
)
208 kaddr
= kmap(&page
[i
]);
209 left
= __copy_to_user(buf
, kaddr
+ offset
, chunksize
);
212 copied
+= (chunksize
- left
);
221 return copied
? copied
: -EFAULT
;
225 * Support for read() - Find the page attached to f_mapping and copy out the
226 * data. Its *very* similar to do_generic_mapping_read(), we can't use that
227 * since it has PAGE_CACHE_SIZE assumptions.
229 static ssize_t
hugetlbfs_read(struct file
*filp
, char __user
*buf
,
230 size_t len
, loff_t
*ppos
)
232 struct hstate
*h
= hstate_file(filp
);
233 struct address_space
*mapping
= filp
->f_mapping
;
234 struct inode
*inode
= mapping
->host
;
235 unsigned long index
= *ppos
>> huge_page_shift(h
);
236 unsigned long offset
= *ppos
& ~huge_page_mask(h
);
237 unsigned long end_index
;
241 mutex_lock(&inode
->i_mutex
);
243 /* validate length */
247 isize
= i_size_read(inode
);
251 end_index
= (isize
- 1) >> huge_page_shift(h
);
254 unsigned long nr
, ret
;
257 /* nr is the maximum number of bytes to copy from this page */
258 nr
= huge_page_size(h
);
259 if (index
>= end_index
) {
260 if (index
> end_index
)
262 nr
= ((isize
- 1) & ~huge_page_mask(h
)) + 1;
270 page
= find_get_page(mapping
, index
);
271 if (unlikely(page
== NULL
)) {
273 * We have a HOLE, zero out the user-buffer for the
274 * length of the hole or request.
276 ret
= len
< nr
? len
: nr
;
277 if (clear_user(buf
, ret
))
283 * We have the page, copy it to user space buffer.
285 ra
= hugetlbfs_read_actor(page
, offset
, buf
, len
, nr
);
292 page_cache_release(page
);
299 index
+= offset
>> huge_page_shift(h
);
300 offset
&= ~huge_page_mask(h
);
303 page_cache_release(page
);
305 /* short read or no more work */
306 if ((ret
!= nr
) || (len
== 0))
310 *ppos
= ((loff_t
)index
<< huge_page_shift(h
)) + offset
;
311 mutex_unlock(&inode
->i_mutex
);
315 static int hugetlbfs_write_begin(struct file
*file
,
316 struct address_space
*mapping
,
317 loff_t pos
, unsigned len
, unsigned flags
,
318 struct page
**pagep
, void **fsdata
)
323 static int hugetlbfs_write_end(struct file
*file
, struct address_space
*mapping
,
324 loff_t pos
, unsigned len
, unsigned copied
,
325 struct page
*page
, void *fsdata
)
331 static void truncate_huge_page(struct page
*page
)
333 cancel_dirty_page(page
, /* No IO accounting for huge pages? */0);
334 ClearPageUptodate(page
);
335 delete_from_page_cache(page
);
338 static void truncate_hugepages(struct inode
*inode
, loff_t lstart
)
340 struct hstate
*h
= hstate_inode(inode
);
341 struct address_space
*mapping
= &inode
->i_data
;
342 const pgoff_t start
= lstart
>> huge_page_shift(h
);
347 pagevec_init(&pvec
, 0);
350 if (!pagevec_lookup(&pvec
, mapping
, next
, PAGEVEC_SIZE
)) {
357 for (i
= 0; i
< pagevec_count(&pvec
); ++i
) {
358 struct page
*page
= pvec
.pages
[i
];
361 if (page
->index
> next
)
364 truncate_huge_page(page
);
368 huge_pagevec_release(&pvec
);
370 BUG_ON(!lstart
&& mapping
->nrpages
);
371 hugetlb_unreserve_pages(inode
, start
, freed
);
374 static void hugetlbfs_evict_inode(struct inode
*inode
)
376 truncate_hugepages(inode
, 0);
377 end_writeback(inode
);
381 hugetlb_vmtruncate_list(struct prio_tree_root
*root
, pgoff_t pgoff
)
383 struct vm_area_struct
*vma
;
384 struct prio_tree_iter iter
;
386 vma_prio_tree_foreach(vma
, &iter
, root
, pgoff
, ULONG_MAX
) {
387 unsigned long v_offset
;
390 * Can the expression below overflow on 32-bit arches?
391 * No, because the prio_tree returns us only those vmas
392 * which overlap the truncated area starting at pgoff,
393 * and no vma on a 32-bit arch can span beyond the 4GB.
395 if (vma
->vm_pgoff
< pgoff
)
396 v_offset
= (pgoff
- vma
->vm_pgoff
) << PAGE_SHIFT
;
400 __unmap_hugepage_range(vma
,
401 vma
->vm_start
+ v_offset
, vma
->vm_end
, NULL
);
405 static int hugetlb_vmtruncate(struct inode
*inode
, loff_t offset
)
408 struct address_space
*mapping
= inode
->i_mapping
;
409 struct hstate
*h
= hstate_inode(inode
);
411 BUG_ON(offset
& ~huge_page_mask(h
));
412 pgoff
= offset
>> PAGE_SHIFT
;
414 i_size_write(inode
, offset
);
415 mutex_lock(&mapping
->i_mmap_mutex
);
416 if (!prio_tree_empty(&mapping
->i_mmap
))
417 hugetlb_vmtruncate_list(&mapping
->i_mmap
, pgoff
);
418 mutex_unlock(&mapping
->i_mmap_mutex
);
419 truncate_hugepages(inode
, offset
);
423 static int hugetlbfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
425 struct inode
*inode
= dentry
->d_inode
;
426 struct hstate
*h
= hstate_inode(inode
);
428 unsigned int ia_valid
= attr
->ia_valid
;
432 error
= inode_change_ok(inode
, attr
);
436 if (ia_valid
& ATTR_SIZE
) {
438 if (attr
->ia_size
& ~huge_page_mask(h
))
440 error
= hugetlb_vmtruncate(inode
, attr
->ia_size
);
445 setattr_copy(inode
, attr
);
446 mark_inode_dirty(inode
);
450 static struct inode
*hugetlbfs_get_inode(struct super_block
*sb
, uid_t uid
,
451 gid_t gid
, int mode
, dev_t dev
)
455 inode
= new_inode(sb
);
457 struct hugetlbfs_inode_info
*info
;
458 inode
->i_ino
= get_next_ino();
459 inode
->i_mode
= mode
;
462 inode
->i_mapping
->a_ops
= &hugetlbfs_aops
;
463 inode
->i_mapping
->backing_dev_info
=&hugetlbfs_backing_dev_info
;
464 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
465 INIT_LIST_HEAD(&inode
->i_mapping
->private_list
);
466 info
= HUGETLBFS_I(inode
);
468 * The policy is initialized here even if we are creating a
469 * private inode because initialization simply creates an
470 * an empty rb tree and calls spin_lock_init(), later when we
471 * call mpol_free_shared_policy() it will just return because
472 * the rb tree will still be empty.
474 mpol_shared_policy_init(&info
->policy
, NULL
);
475 switch (mode
& S_IFMT
) {
477 init_special_inode(inode
, mode
, dev
);
480 inode
->i_op
= &hugetlbfs_inode_operations
;
481 inode
->i_fop
= &hugetlbfs_file_operations
;
484 inode
->i_op
= &hugetlbfs_dir_inode_operations
;
485 inode
->i_fop
= &simple_dir_operations
;
487 /* directory inodes start off with i_nlink == 2 (for "." entry) */
491 inode
->i_op
= &page_symlink_inode_operations
;
494 lockdep_annotate_inode_mutex_key(inode
);
500 * File creation. Allocate an inode, and we're done..
502 static int hugetlbfs_mknod(struct inode
*dir
,
503 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
509 if (dir
->i_mode
& S_ISGID
) {
514 gid
= current_fsgid();
516 inode
= hugetlbfs_get_inode(dir
->i_sb
, current_fsuid(), gid
, 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
, struct nameidata
*nd
)
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
)
546 if (dir
->i_mode
& S_ISGID
)
549 gid
= current_fsgid();
551 inode
= hugetlbfs_get_inode(dir
->i_sb
, current_fsuid(),
552 gid
, S_IFLNK
|S_IRWXUGO
, 0);
554 int l
= strlen(symname
)+1;
555 error
= page_symlink(inode
, symname
, l
);
557 d_instantiate(dentry
, inode
);
562 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
568 * mark the head page dirty
570 static int hugetlbfs_set_page_dirty(struct page
*page
)
572 struct page
*head
= compound_head(page
);
578 static int hugetlbfs_migrate_page(struct address_space
*mapping
,
579 struct page
*newpage
, struct page
*page
)
583 rc
= migrate_huge_page_move_mapping(mapping
, newpage
, page
);
586 migrate_page_copy(newpage
, page
);
591 static int hugetlbfs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
593 struct hugetlbfs_sb_info
*sbinfo
= HUGETLBFS_SB(dentry
->d_sb
);
594 struct hstate
*h
= hstate_inode(dentry
->d_inode
);
596 buf
->f_type
= HUGETLBFS_MAGIC
;
597 buf
->f_bsize
= huge_page_size(h
);
599 spin_lock(&sbinfo
->stat_lock
);
600 /* If no limits set, just report 0 for max/free/used
601 * blocks, like simple_statfs() */
602 if (sbinfo
->max_blocks
>= 0) {
603 buf
->f_blocks
= sbinfo
->max_blocks
;
604 buf
->f_bavail
= buf
->f_bfree
= sbinfo
->free_blocks
;
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
;
624 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info
*sbinfo
)
626 if (sbinfo
->free_inodes
>= 0) {
627 spin_lock(&sbinfo
->stat_lock
);
628 if (unlikely(!sbinfo
->free_inodes
)) {
629 spin_unlock(&sbinfo
->stat_lock
);
632 sbinfo
->free_inodes
--;
633 spin_unlock(&sbinfo
->stat_lock
);
639 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info
*sbinfo
)
641 if (sbinfo
->free_inodes
>= 0) {
642 spin_lock(&sbinfo
->stat_lock
);
643 sbinfo
->free_inodes
++;
644 spin_unlock(&sbinfo
->stat_lock
);
649 static struct kmem_cache
*hugetlbfs_inode_cachep
;
651 static struct inode
*hugetlbfs_alloc_inode(struct super_block
*sb
)
653 struct hugetlbfs_sb_info
*sbinfo
= HUGETLBFS_SB(sb
);
654 struct hugetlbfs_inode_info
*p
;
656 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo
)))
658 p
= kmem_cache_alloc(hugetlbfs_inode_cachep
, GFP_KERNEL
);
660 hugetlbfs_inc_free_inodes(sbinfo
);
663 return &p
->vfs_inode
;
666 static void hugetlbfs_i_callback(struct rcu_head
*head
)
668 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
669 kmem_cache_free(hugetlbfs_inode_cachep
, HUGETLBFS_I(inode
));
672 static void hugetlbfs_destroy_inode(struct inode
*inode
)
674 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode
->i_sb
));
675 mpol_free_shared_policy(&HUGETLBFS_I(inode
)->policy
);
676 call_rcu(&inode
->i_rcu
, hugetlbfs_i_callback
);
679 static const struct address_space_operations hugetlbfs_aops
= {
680 .write_begin
= hugetlbfs_write_begin
,
681 .write_end
= hugetlbfs_write_end
,
682 .set_page_dirty
= hugetlbfs_set_page_dirty
,
683 .migratepage
= hugetlbfs_migrate_page
,
687 static void init_once(void *foo
)
689 struct hugetlbfs_inode_info
*ei
= (struct hugetlbfs_inode_info
*)foo
;
691 inode_init_once(&ei
->vfs_inode
);
694 const struct file_operations hugetlbfs_file_operations
= {
695 .read
= hugetlbfs_read
,
696 .mmap
= hugetlbfs_file_mmap
,
698 .get_unmapped_area
= hugetlb_get_unmapped_area
,
699 .llseek
= default_llseek
,
702 static const struct inode_operations hugetlbfs_dir_inode_operations
= {
703 .create
= hugetlbfs_create
,
704 .lookup
= simple_lookup
,
706 .unlink
= simple_unlink
,
707 .symlink
= hugetlbfs_symlink
,
708 .mkdir
= hugetlbfs_mkdir
,
709 .rmdir
= simple_rmdir
,
710 .mknod
= hugetlbfs_mknod
,
711 .rename
= simple_rename
,
712 .setattr
= hugetlbfs_setattr
,
715 static const struct inode_operations hugetlbfs_inode_operations
= {
716 .setattr
= hugetlbfs_setattr
,
719 static const struct super_operations hugetlbfs_ops
= {
720 .alloc_inode
= hugetlbfs_alloc_inode
,
721 .destroy_inode
= hugetlbfs_destroy_inode
,
722 .evict_inode
= hugetlbfs_evict_inode
,
723 .statfs
= hugetlbfs_statfs
,
724 .put_super
= hugetlbfs_put_super
,
725 .show_options
= generic_show_options
,
729 hugetlbfs_parse_options(char *options
, struct hugetlbfs_config
*pconfig
)
732 substring_t args
[MAX_OPT_ARGS
];
734 unsigned long long size
= 0;
735 enum { NO_SIZE
, SIZE_STD
, SIZE_PERCENT
} setsize
= NO_SIZE
;
740 while ((p
= strsep(&options
, ",")) != NULL
) {
745 token
= match_token(p
, tokens
, args
);
748 if (match_int(&args
[0], &option
))
750 pconfig
->uid
= option
;
754 if (match_int(&args
[0], &option
))
756 pconfig
->gid
= option
;
760 if (match_octal(&args
[0], &option
))
762 pconfig
->mode
= option
& 01777U;
766 /* memparse() will accept a K/M/G without a digit */
767 if (!isdigit(*args
[0].from
))
769 size
= memparse(args
[0].from
, &rest
);
772 setsize
= SIZE_PERCENT
;
777 /* memparse() will accept a K/M/G without a digit */
778 if (!isdigit(*args
[0].from
))
780 pconfig
->nr_inodes
= memparse(args
[0].from
, &rest
);
785 ps
= memparse(args
[0].from
, &rest
);
786 pconfig
->hstate
= size_to_hstate(ps
);
787 if (!pconfig
->hstate
) {
789 "hugetlbfs: Unsupported page size %lu MB\n",
797 printk(KERN_ERR
"hugetlbfs: Bad mount option: \"%s\"\n",
804 /* Do size after hstate is set up */
805 if (setsize
> NO_SIZE
) {
806 struct hstate
*h
= pconfig
->hstate
;
807 if (setsize
== SIZE_PERCENT
) {
808 size
<<= huge_page_shift(h
);
809 size
*= h
->max_huge_pages
;
812 pconfig
->nr_blocks
= (size
>> huge_page_shift(h
));
818 printk(KERN_ERR
"hugetlbfs: Bad value '%s' for mount option '%s'\n",
824 hugetlbfs_fill_super(struct super_block
*sb
, void *data
, int silent
)
826 struct inode
* inode
;
827 struct dentry
* root
;
829 struct hugetlbfs_config config
;
830 struct hugetlbfs_sb_info
*sbinfo
;
832 save_mount_options(sb
, data
);
834 config
.nr_blocks
= -1; /* No limit on size by default */
835 config
.nr_inodes
= -1; /* No limit on number of inodes by default */
836 config
.uid
= current_fsuid();
837 config
.gid
= current_fsgid();
839 config
.hstate
= &default_hstate
;
840 ret
= hugetlbfs_parse_options(data
, &config
);
844 sbinfo
= kmalloc(sizeof(struct hugetlbfs_sb_info
), GFP_KERNEL
);
847 sb
->s_fs_info
= sbinfo
;
848 sbinfo
->hstate
= config
.hstate
;
849 spin_lock_init(&sbinfo
->stat_lock
);
850 sbinfo
->max_blocks
= config
.nr_blocks
;
851 sbinfo
->free_blocks
= config
.nr_blocks
;
852 sbinfo
->max_inodes
= config
.nr_inodes
;
853 sbinfo
->free_inodes
= config
.nr_inodes
;
854 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
855 sb
->s_blocksize
= huge_page_size(config
.hstate
);
856 sb
->s_blocksize_bits
= huge_page_shift(config
.hstate
);
857 sb
->s_magic
= HUGETLBFS_MAGIC
;
858 sb
->s_op
= &hugetlbfs_ops
;
860 inode
= hugetlbfs_get_inode(sb
, config
.uid
, config
.gid
,
861 S_IFDIR
| config
.mode
, 0);
865 root
= d_alloc_root(inode
);
877 int hugetlb_get_quota(struct address_space
*mapping
, long delta
)
880 struct hugetlbfs_sb_info
*sbinfo
= HUGETLBFS_SB(mapping
->host
->i_sb
);
882 if (sbinfo
->free_blocks
> -1) {
883 spin_lock(&sbinfo
->stat_lock
);
884 if (sbinfo
->free_blocks
- delta
>= 0)
885 sbinfo
->free_blocks
-= delta
;
888 spin_unlock(&sbinfo
->stat_lock
);
894 void hugetlb_put_quota(struct address_space
*mapping
, long delta
)
896 struct hugetlbfs_sb_info
*sbinfo
= HUGETLBFS_SB(mapping
->host
->i_sb
);
898 if (sbinfo
->free_blocks
> -1) {
899 spin_lock(&sbinfo
->stat_lock
);
900 sbinfo
->free_blocks
+= delta
;
901 spin_unlock(&sbinfo
->stat_lock
);
905 static struct dentry
*hugetlbfs_mount(struct file_system_type
*fs_type
,
906 int flags
, const char *dev_name
, void *data
)
908 return mount_nodev(fs_type
, flags
, data
, hugetlbfs_fill_super
);
911 static struct file_system_type hugetlbfs_fs_type
= {
913 .mount
= hugetlbfs_mount
,
914 .kill_sb
= kill_litter_super
,
917 static struct vfsmount
*hugetlbfs_vfsmount
;
919 static int can_do_hugetlb_shm(void)
921 return capable(CAP_IPC_LOCK
) || in_group_p(sysctl_hugetlb_shm_group
);
924 struct file
*hugetlb_file_setup(const char *name
, size_t size
,
926 struct user_struct
**user
, int creat_flags
)
933 struct qstr quick_string
;
936 if (!hugetlbfs_vfsmount
)
937 return ERR_PTR(-ENOENT
);
939 if (creat_flags
== HUGETLB_SHMFS_INODE
&& !can_do_hugetlb_shm()) {
940 *user
= current_user();
941 if (user_shm_lock(size
, *user
)) {
942 printk_once(KERN_WARNING
"Using mlock ulimits for SHM_HUGETLB is deprecated\n");
945 return ERR_PTR(-EPERM
);
949 root
= hugetlbfs_vfsmount
->mnt_root
;
950 quick_string
.name
= name
;
951 quick_string
.len
= strlen(quick_string
.name
);
952 quick_string
.hash
= 0;
953 path
.dentry
= d_alloc(root
, &quick_string
);
957 path
.mnt
= mntget(hugetlbfs_vfsmount
);
959 inode
= hugetlbfs_get_inode(root
->d_sb
, current_fsuid(),
960 current_fsgid(), S_IFREG
| S_IRWXUGO
, 0);
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
;
975 file
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
976 &hugetlbfs_file_operations
);
978 goto out_dentry
; /* inode is already attached */
988 user_shm_unlock(size
, *user
);
991 return ERR_PTR(error
);
994 static int __init
init_hugetlbfs_fs(void)
997 struct vfsmount
*vfsmount
;
999 error
= bdi_init(&hugetlbfs_backing_dev_info
);
1003 hugetlbfs_inode_cachep
= kmem_cache_create("hugetlbfs_inode_cache",
1004 sizeof(struct hugetlbfs_inode_info
),
1006 if (hugetlbfs_inode_cachep
== NULL
)
1009 error
= register_filesystem(&hugetlbfs_fs_type
);
1013 vfsmount
= kern_mount(&hugetlbfs_fs_type
);
1015 if (!IS_ERR(vfsmount
)) {
1016 hugetlbfs_vfsmount
= vfsmount
;
1020 error
= PTR_ERR(vfsmount
);
1024 kmem_cache_destroy(hugetlbfs_inode_cachep
);
1026 bdi_destroy(&hugetlbfs_backing_dev_info
);
1030 static void __exit
exit_hugetlbfs_fs(void)
1032 kmem_cache_destroy(hugetlbfs_inode_cachep
);
1033 kern_unmount(hugetlbfs_vfsmount
);
1034 unregister_filesystem(&hugetlbfs_fs_type
);
1035 bdi_destroy(&hugetlbfs_backing_dev_info
);
1038 module_init(init_hugetlbfs_fs
)
1039 module_exit(exit_hugetlbfs_fs
)
1041 MODULE_LICENSE("GPL");