2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
11 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 * Extended attribute support for tmpfs:
14 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
15 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 * This file is released under the GPL.
21 * This virtual memory filesystem is heavily based on the ramfs. It
22 * extends ramfs by the ability to use swap and honor resource limits
23 * which makes it a completely usable filesystem.
26 #include <linux/module.h>
27 #include <linux/init.h>
29 #include <linux/xattr.h>
30 #include <linux/exportfs.h>
31 #include <linux/generic_acl.h>
33 #include <linux/mman.h>
34 #include <linux/file.h>
35 #include <linux/swap.h>
36 #include <linux/pagemap.h>
37 #include <linux/string.h>
38 #include <linux/slab.h>
39 #include <linux/backing-dev.h>
40 #include <linux/shmem_fs.h>
41 #include <linux/mount.h>
42 #include <linux/writeback.h>
43 #include <linux/vfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/security.h>
46 #include <linux/swapops.h>
47 #include <linux/mempolicy.h>
48 #include <linux/namei.h>
49 #include <linux/ctype.h>
50 #include <linux/migrate.h>
51 #include <linux/highmem.h>
53 #include <asm/uaccess.h>
54 #include <asm/div64.h>
55 #include <asm/pgtable.h>
57 /* This magic number is used in glibc for posix shared memory */
58 #define TMPFS_MAGIC 0x01021994
60 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
61 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
62 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
64 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
65 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
67 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
69 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
70 #define SHMEM_PAGEIN VM_READ
71 #define SHMEM_TRUNCATE VM_WRITE
73 /* Definition to limit shmem_truncate's steps between cond_rescheds */
74 #define LATENCY_LIMIT 64
76 /* Pretend that each entry is of this size in directory's i_size */
77 #define BOGO_DIRENT_SIZE 20
79 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
81 SGP_QUICK
, /* don't try more than file page cache lookup */
82 SGP_READ
, /* don't exceed i_size, don't allocate page */
83 SGP_CACHE
, /* don't exceed i_size, may allocate page */
84 SGP_WRITE
, /* may exceed i_size, may allocate page */
85 SGP_FAULT
, /* same as SGP_CACHE, return with page locked */
88 static int shmem_getpage(struct inode
*inode
, unsigned long idx
,
89 struct page
**pagep
, enum sgp_type sgp
, int *type
);
91 static inline struct page
*shmem_dir_alloc(gfp_t gfp_mask
)
94 * The above definition of ENTRIES_PER_PAGE, and the use of
95 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
96 * might be reconsidered if it ever diverges from PAGE_SIZE.
98 * __GFP_MOVABLE is masked out as swap vectors cannot move
100 return alloc_pages((gfp_mask
& ~__GFP_MOVABLE
) | __GFP_ZERO
,
101 PAGE_CACHE_SHIFT
-PAGE_SHIFT
);
104 static inline void shmem_dir_free(struct page
*page
)
106 __free_pages(page
, PAGE_CACHE_SHIFT
-PAGE_SHIFT
);
109 static struct page
**shmem_dir_map(struct page
*page
)
111 return (struct page
**)kmap_atomic(page
, KM_USER0
);
114 static inline void shmem_dir_unmap(struct page
**dir
)
116 kunmap_atomic(dir
, KM_USER0
);
119 static swp_entry_t
*shmem_swp_map(struct page
*page
)
121 return (swp_entry_t
*)kmap_atomic(page
, KM_USER1
);
124 static inline void shmem_swp_balance_unmap(void)
127 * When passing a pointer to an i_direct entry, to code which
128 * also handles indirect entries and so will shmem_swp_unmap,
129 * we must arrange for the preempt count to remain in balance.
130 * What kmap_atomic of a lowmem page does depends on config
131 * and architecture, so pretend to kmap_atomic some lowmem page.
133 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1
);
136 static inline void shmem_swp_unmap(swp_entry_t
*entry
)
138 kunmap_atomic(entry
, KM_USER1
);
141 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
143 return sb
->s_fs_info
;
147 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
148 * for shared memory and for shared anonymous (/dev/zero) mappings
149 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
150 * consistent with the pre-accounting of private mappings ...
152 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
154 return (flags
& VM_ACCOUNT
)?
155 security_vm_enough_memory(VM_ACCT(size
)): 0;
158 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
160 if (flags
& VM_ACCOUNT
)
161 vm_unacct_memory(VM_ACCT(size
));
165 * ... whereas tmpfs objects are accounted incrementally as
166 * pages are allocated, in order to allow huge sparse files.
167 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
168 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
170 static inline int shmem_acct_block(unsigned long flags
)
172 return (flags
& VM_ACCOUNT
)?
173 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE
));
176 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
178 if (!(flags
& VM_ACCOUNT
))
179 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
182 static const struct super_operations shmem_ops
;
183 static const struct address_space_operations shmem_aops
;
184 static const struct file_operations shmem_file_operations
;
185 static const struct inode_operations shmem_inode_operations
;
186 static const struct inode_operations shmem_dir_inode_operations
;
187 static const struct inode_operations shmem_special_inode_operations
;
188 static struct vm_operations_struct shmem_vm_ops
;
190 static struct backing_dev_info shmem_backing_dev_info __read_mostly
= {
191 .ra_pages
= 0, /* No readahead */
192 .capabilities
= BDI_CAP_NO_ACCT_DIRTY
| BDI_CAP_NO_WRITEBACK
,
193 .unplug_io_fn
= default_unplug_io_fn
,
196 static LIST_HEAD(shmem_swaplist
);
197 static DEFINE_SPINLOCK(shmem_swaplist_lock
);
199 static void shmem_free_blocks(struct inode
*inode
, long pages
)
201 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
202 if (sbinfo
->max_blocks
) {
203 spin_lock(&sbinfo
->stat_lock
);
204 sbinfo
->free_blocks
+= pages
;
205 inode
->i_blocks
-= pages
*BLOCKS_PER_PAGE
;
206 spin_unlock(&sbinfo
->stat_lock
);
211 * shmem_recalc_inode - recalculate the size of an inode
213 * @inode: inode to recalc
215 * We have to calculate the free blocks since the mm can drop
216 * undirtied hole pages behind our back.
218 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
219 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
221 * It has to be called with the spinlock held.
223 static void shmem_recalc_inode(struct inode
*inode
)
225 struct shmem_inode_info
*info
= SHMEM_I(inode
);
228 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
230 info
->alloced
-= freed
;
231 shmem_unacct_blocks(info
->flags
, freed
);
232 shmem_free_blocks(inode
, freed
);
237 * shmem_swp_entry - find the swap vector position in the info structure
239 * @info: info structure for the inode
240 * @index: index of the page to find
241 * @page: optional page to add to the structure. Has to be preset to
244 * If there is no space allocated yet it will return NULL when
245 * page is NULL, else it will use the page for the needed block,
246 * setting it to NULL on return to indicate that it has been used.
248 * The swap vector is organized the following way:
250 * There are SHMEM_NR_DIRECT entries directly stored in the
251 * shmem_inode_info structure. So small files do not need an addional
254 * For pages with index > SHMEM_NR_DIRECT there is the pointer
255 * i_indirect which points to a page which holds in the first half
256 * doubly indirect blocks, in the second half triple indirect blocks:
258 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
259 * following layout (for SHMEM_NR_DIRECT == 16):
261 * i_indirect -> dir --> 16-19
274 static swp_entry_t
*shmem_swp_entry(struct shmem_inode_info
*info
, unsigned long index
, struct page
**page
)
276 unsigned long offset
;
280 if (index
< SHMEM_NR_DIRECT
) {
281 shmem_swp_balance_unmap();
282 return info
->i_direct
+index
;
284 if (!info
->i_indirect
) {
286 info
->i_indirect
= *page
;
289 return NULL
; /* need another page */
292 index
-= SHMEM_NR_DIRECT
;
293 offset
= index
% ENTRIES_PER_PAGE
;
294 index
/= ENTRIES_PER_PAGE
;
295 dir
= shmem_dir_map(info
->i_indirect
);
297 if (index
>= ENTRIES_PER_PAGE
/2) {
298 index
-= ENTRIES_PER_PAGE
/2;
299 dir
+= ENTRIES_PER_PAGE
/2 + index
/ENTRIES_PER_PAGE
;
300 index
%= ENTRIES_PER_PAGE
;
307 shmem_dir_unmap(dir
);
308 return NULL
; /* need another page */
310 shmem_dir_unmap(dir
);
311 dir
= shmem_dir_map(subdir
);
317 if (!page
|| !(subdir
= *page
)) {
318 shmem_dir_unmap(dir
);
319 return NULL
; /* need a page */
324 shmem_dir_unmap(dir
);
325 return shmem_swp_map(subdir
) + offset
;
328 static void shmem_swp_set(struct shmem_inode_info
*info
, swp_entry_t
*entry
, unsigned long value
)
330 long incdec
= value
? 1: -1;
333 info
->swapped
+= incdec
;
334 if ((unsigned long)(entry
- info
->i_direct
) >= SHMEM_NR_DIRECT
) {
335 struct page
*page
= kmap_atomic_to_page(entry
);
336 set_page_private(page
, page_private(page
) + incdec
);
341 * shmem_swp_alloc - get the position of the swap entry for the page.
342 * If it does not exist allocate the entry.
344 * @info: info structure for the inode
345 * @index: index of the page to find
346 * @sgp: check and recheck i_size? skip allocation?
348 static swp_entry_t
*shmem_swp_alloc(struct shmem_inode_info
*info
, unsigned long index
, enum sgp_type sgp
)
350 struct inode
*inode
= &info
->vfs_inode
;
351 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
352 struct page
*page
= NULL
;
355 if (sgp
!= SGP_WRITE
&&
356 ((loff_t
) index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
))
357 return ERR_PTR(-EINVAL
);
359 while (!(entry
= shmem_swp_entry(info
, index
, &page
))) {
361 return shmem_swp_map(ZERO_PAGE(0));
363 * Test free_blocks against 1 not 0, since we have 1 data
364 * page (and perhaps indirect index pages) yet to allocate:
365 * a waste to allocate index if we cannot allocate data.
367 if (sbinfo
->max_blocks
) {
368 spin_lock(&sbinfo
->stat_lock
);
369 if (sbinfo
->free_blocks
<= 1) {
370 spin_unlock(&sbinfo
->stat_lock
);
371 return ERR_PTR(-ENOSPC
);
373 sbinfo
->free_blocks
--;
374 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
375 spin_unlock(&sbinfo
->stat_lock
);
378 spin_unlock(&info
->lock
);
379 page
= shmem_dir_alloc(mapping_gfp_mask(inode
->i_mapping
));
381 set_page_private(page
, 0);
382 spin_lock(&info
->lock
);
385 shmem_free_blocks(inode
, 1);
386 return ERR_PTR(-ENOMEM
);
388 if (sgp
!= SGP_WRITE
&&
389 ((loff_t
) index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
390 entry
= ERR_PTR(-EINVAL
);
393 if (info
->next_index
<= index
)
394 info
->next_index
= index
+ 1;
397 /* another task gave its page, or truncated the file */
398 shmem_free_blocks(inode
, 1);
399 shmem_dir_free(page
);
401 if (info
->next_index
<= index
&& !IS_ERR(entry
))
402 info
->next_index
= index
+ 1;
407 * shmem_free_swp - free some swap entries in a directory
409 * @dir: pointer to the directory
410 * @edir: pointer after last entry of the directory
411 * @punch_lock: pointer to spinlock when needed for the holepunch case
413 static int shmem_free_swp(swp_entry_t
*dir
, swp_entry_t
*edir
,
414 spinlock_t
*punch_lock
)
416 spinlock_t
*punch_unlock
= NULL
;
420 for (ptr
= dir
; ptr
< edir
; ptr
++) {
422 if (unlikely(punch_lock
)) {
423 punch_unlock
= punch_lock
;
425 spin_lock(punch_unlock
);
429 free_swap_and_cache(*ptr
);
430 *ptr
= (swp_entry_t
){0};
435 spin_unlock(punch_unlock
);
439 static int shmem_map_and_free_swp(struct page
*subdir
, int offset
,
440 int limit
, struct page
***dir
, spinlock_t
*punch_lock
)
445 ptr
= shmem_swp_map(subdir
);
446 for (; offset
< limit
; offset
+= LATENCY_LIMIT
) {
447 int size
= limit
- offset
;
448 if (size
> LATENCY_LIMIT
)
449 size
= LATENCY_LIMIT
;
450 freed
+= shmem_free_swp(ptr
+offset
, ptr
+offset
+size
,
452 if (need_resched()) {
453 shmem_swp_unmap(ptr
);
455 shmem_dir_unmap(*dir
);
459 ptr
= shmem_swp_map(subdir
);
462 shmem_swp_unmap(ptr
);
466 static void shmem_free_pages(struct list_head
*next
)
472 page
= container_of(next
, struct page
, lru
);
474 shmem_dir_free(page
);
476 if (freed
>= LATENCY_LIMIT
) {
483 static void shmem_truncate_range(struct inode
*inode
, loff_t start
, loff_t end
)
485 struct shmem_inode_info
*info
= SHMEM_I(inode
);
490 unsigned long diroff
;
496 LIST_HEAD(pages_to_free
);
497 long nr_pages_to_free
= 0;
498 long nr_swaps_freed
= 0;
502 spinlock_t
*needs_lock
;
503 spinlock_t
*punch_lock
;
504 unsigned long upper_limit
;
506 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
507 idx
= (start
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
508 if (idx
>= info
->next_index
)
511 spin_lock(&info
->lock
);
512 info
->flags
|= SHMEM_TRUNCATE
;
513 if (likely(end
== (loff_t
) -1)) {
514 limit
= info
->next_index
;
515 upper_limit
= SHMEM_MAX_INDEX
;
516 info
->next_index
= idx
;
520 if (end
+ 1 >= inode
->i_size
) { /* we may free a little more */
521 limit
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >>
523 upper_limit
= SHMEM_MAX_INDEX
;
525 limit
= (end
+ 1) >> PAGE_CACHE_SHIFT
;
528 needs_lock
= &info
->lock
;
532 topdir
= info
->i_indirect
;
533 if (topdir
&& idx
<= SHMEM_NR_DIRECT
&& !punch_hole
) {
534 info
->i_indirect
= NULL
;
536 list_add(&topdir
->lru
, &pages_to_free
);
538 spin_unlock(&info
->lock
);
540 if (info
->swapped
&& idx
< SHMEM_NR_DIRECT
) {
541 ptr
= info
->i_direct
;
543 if (size
> SHMEM_NR_DIRECT
)
544 size
= SHMEM_NR_DIRECT
;
545 nr_swaps_freed
= shmem_free_swp(ptr
+idx
, ptr
+size
, needs_lock
);
549 * If there are no indirect blocks or we are punching a hole
550 * below indirect blocks, nothing to be done.
552 if (!topdir
|| limit
<= SHMEM_NR_DIRECT
)
556 * The truncation case has already dropped info->lock, and we're safe
557 * because i_size and next_index have already been lowered, preventing
558 * access beyond. But in the punch_hole case, we still need to take
559 * the lock when updating the swap directory, because there might be
560 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
561 * shmem_writepage. However, whenever we find we can remove a whole
562 * directory page (not at the misaligned start or end of the range),
563 * we first NULLify its pointer in the level above, and then have no
564 * need to take the lock when updating its contents: needs_lock and
565 * punch_lock (either pointing to info->lock or NULL) manage this.
568 upper_limit
-= SHMEM_NR_DIRECT
;
569 limit
-= SHMEM_NR_DIRECT
;
570 idx
= (idx
> SHMEM_NR_DIRECT
)? (idx
- SHMEM_NR_DIRECT
): 0;
571 offset
= idx
% ENTRIES_PER_PAGE
;
574 dir
= shmem_dir_map(topdir
);
575 stage
= ENTRIES_PER_PAGEPAGE
/2;
576 if (idx
< ENTRIES_PER_PAGEPAGE
/2) {
578 diroff
= idx
/ENTRIES_PER_PAGE
;
580 dir
+= ENTRIES_PER_PAGE
/2;
581 dir
+= (idx
- ENTRIES_PER_PAGEPAGE
/2)/ENTRIES_PER_PAGEPAGE
;
583 stage
+= ENTRIES_PER_PAGEPAGE
;
586 diroff
= ((idx
- ENTRIES_PER_PAGEPAGE
/2) %
587 ENTRIES_PER_PAGEPAGE
) / ENTRIES_PER_PAGE
;
588 if (!diroff
&& !offset
&& upper_limit
>= stage
) {
590 spin_lock(needs_lock
);
592 spin_unlock(needs_lock
);
597 list_add(&middir
->lru
, &pages_to_free
);
599 shmem_dir_unmap(dir
);
600 dir
= shmem_dir_map(middir
);
608 for (; idx
< limit
; idx
+= ENTRIES_PER_PAGE
, diroff
++) {
609 if (unlikely(idx
== stage
)) {
610 shmem_dir_unmap(dir
);
611 dir
= shmem_dir_map(topdir
) +
612 ENTRIES_PER_PAGE
/2 + idx
/ENTRIES_PER_PAGEPAGE
;
615 idx
+= ENTRIES_PER_PAGEPAGE
;
619 stage
= idx
+ ENTRIES_PER_PAGEPAGE
;
622 needs_lock
= &info
->lock
;
623 if (upper_limit
>= stage
) {
625 spin_lock(needs_lock
);
627 spin_unlock(needs_lock
);
632 list_add(&middir
->lru
, &pages_to_free
);
634 shmem_dir_unmap(dir
);
636 dir
= shmem_dir_map(middir
);
639 punch_lock
= needs_lock
;
640 subdir
= dir
[diroff
];
641 if (subdir
&& !offset
&& upper_limit
-idx
>= ENTRIES_PER_PAGE
) {
643 spin_lock(needs_lock
);
645 spin_unlock(needs_lock
);
650 list_add(&subdir
->lru
, &pages_to_free
);
652 if (subdir
&& page_private(subdir
) /* has swap entries */) {
654 if (size
> ENTRIES_PER_PAGE
)
655 size
= ENTRIES_PER_PAGE
;
656 freed
= shmem_map_and_free_swp(subdir
,
657 offset
, size
, &dir
, punch_lock
);
659 dir
= shmem_dir_map(middir
);
660 nr_swaps_freed
+= freed
;
661 if (offset
|| punch_lock
) {
662 spin_lock(&info
->lock
);
663 set_page_private(subdir
,
664 page_private(subdir
) - freed
);
665 spin_unlock(&info
->lock
);
667 BUG_ON(page_private(subdir
) != freed
);
672 shmem_dir_unmap(dir
);
674 if (inode
->i_mapping
->nrpages
&& (info
->flags
& SHMEM_PAGEIN
)) {
676 * Call truncate_inode_pages again: racing shmem_unuse_inode
677 * may have swizzled a page in from swap since vmtruncate or
678 * generic_delete_inode did it, before we lowered next_index.
679 * Also, though shmem_getpage checks i_size before adding to
680 * cache, no recheck after: so fix the narrow window there too.
682 * Recalling truncate_inode_pages_range and unmap_mapping_range
683 * every time for punch_hole (which never got a chance to clear
684 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
685 * yet hardly ever necessary: try to optimize them out later.
687 truncate_inode_pages_range(inode
->i_mapping
, start
, end
);
689 unmap_mapping_range(inode
->i_mapping
, start
,
693 spin_lock(&info
->lock
);
694 info
->flags
&= ~SHMEM_TRUNCATE
;
695 info
->swapped
-= nr_swaps_freed
;
696 if (nr_pages_to_free
)
697 shmem_free_blocks(inode
, nr_pages_to_free
);
698 shmem_recalc_inode(inode
);
699 spin_unlock(&info
->lock
);
702 * Empty swap vector directory pages to be freed?
704 if (!list_empty(&pages_to_free
)) {
705 pages_to_free
.prev
->next
= NULL
;
706 shmem_free_pages(pages_to_free
.next
);
710 static void shmem_truncate(struct inode
*inode
)
712 shmem_truncate_range(inode
, inode
->i_size
, (loff_t
)-1);
715 static int shmem_notify_change(struct dentry
*dentry
, struct iattr
*attr
)
717 struct inode
*inode
= dentry
->d_inode
;
718 struct page
*page
= NULL
;
721 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
722 if (attr
->ia_size
< inode
->i_size
) {
724 * If truncating down to a partial page, then
725 * if that page is already allocated, hold it
726 * in memory until the truncation is over, so
727 * truncate_partial_page cannnot miss it were
728 * it assigned to swap.
730 if (attr
->ia_size
& (PAGE_CACHE_SIZE
-1)) {
731 (void) shmem_getpage(inode
,
732 attr
->ia_size
>>PAGE_CACHE_SHIFT
,
733 &page
, SGP_READ
, NULL
);
736 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
737 * detect if any pages might have been added to cache
738 * after truncate_inode_pages. But we needn't bother
739 * if it's being fully truncated to zero-length: the
740 * nrpages check is efficient enough in that case.
743 struct shmem_inode_info
*info
= SHMEM_I(inode
);
744 spin_lock(&info
->lock
);
745 info
->flags
&= ~SHMEM_PAGEIN
;
746 spin_unlock(&info
->lock
);
751 error
= inode_change_ok(inode
, attr
);
753 error
= inode_setattr(inode
, attr
);
754 #ifdef CONFIG_TMPFS_POSIX_ACL
755 if (!error
&& (attr
->ia_valid
& ATTR_MODE
))
756 error
= generic_acl_chmod(inode
, &shmem_acl_ops
);
759 page_cache_release(page
);
763 static void shmem_delete_inode(struct inode
*inode
)
765 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
766 struct shmem_inode_info
*info
= SHMEM_I(inode
);
768 if (inode
->i_op
->truncate
== shmem_truncate
) {
769 truncate_inode_pages(inode
->i_mapping
, 0);
770 shmem_unacct_size(info
->flags
, inode
->i_size
);
772 shmem_truncate(inode
);
773 if (!list_empty(&info
->swaplist
)) {
774 spin_lock(&shmem_swaplist_lock
);
775 list_del_init(&info
->swaplist
);
776 spin_unlock(&shmem_swaplist_lock
);
779 BUG_ON(inode
->i_blocks
);
780 if (sbinfo
->max_inodes
) {
781 spin_lock(&sbinfo
->stat_lock
);
782 sbinfo
->free_inodes
++;
783 spin_unlock(&sbinfo
->stat_lock
);
788 static inline int shmem_find_swp(swp_entry_t entry
, swp_entry_t
*dir
, swp_entry_t
*edir
)
792 for (ptr
= dir
; ptr
< edir
; ptr
++) {
793 if (ptr
->val
== entry
.val
)
799 static int shmem_unuse_inode(struct shmem_inode_info
*info
, swp_entry_t entry
, struct page
*page
)
812 ptr
= info
->i_direct
;
813 spin_lock(&info
->lock
);
814 limit
= info
->next_index
;
816 if (size
> SHMEM_NR_DIRECT
)
817 size
= SHMEM_NR_DIRECT
;
818 offset
= shmem_find_swp(entry
, ptr
, ptr
+size
);
820 shmem_swp_balance_unmap();
823 if (!info
->i_indirect
)
826 dir
= shmem_dir_map(info
->i_indirect
);
827 stage
= SHMEM_NR_DIRECT
+ ENTRIES_PER_PAGEPAGE
/2;
829 for (idx
= SHMEM_NR_DIRECT
; idx
< limit
; idx
+= ENTRIES_PER_PAGE
, dir
++) {
830 if (unlikely(idx
== stage
)) {
831 shmem_dir_unmap(dir
-1);
832 dir
= shmem_dir_map(info
->i_indirect
) +
833 ENTRIES_PER_PAGE
/2 + idx
/ENTRIES_PER_PAGEPAGE
;
836 idx
+= ENTRIES_PER_PAGEPAGE
;
840 stage
= idx
+ ENTRIES_PER_PAGEPAGE
;
842 shmem_dir_unmap(dir
);
843 dir
= shmem_dir_map(subdir
);
846 if (subdir
&& page_private(subdir
)) {
847 ptr
= shmem_swp_map(subdir
);
849 if (size
> ENTRIES_PER_PAGE
)
850 size
= ENTRIES_PER_PAGE
;
851 offset
= shmem_find_swp(entry
, ptr
, ptr
+size
);
853 shmem_dir_unmap(dir
);
856 shmem_swp_unmap(ptr
);
860 shmem_dir_unmap(dir
-1);
862 spin_unlock(&info
->lock
);
866 inode
= &info
->vfs_inode
;
867 if (move_from_swap_cache(page
, idx
, inode
->i_mapping
) == 0) {
868 info
->flags
|= SHMEM_PAGEIN
;
869 shmem_swp_set(info
, ptr
+ offset
, 0);
871 shmem_swp_unmap(ptr
);
872 spin_unlock(&info
->lock
);
874 * Decrement swap count even when the entry is left behind:
875 * try_to_unuse will skip over mms, then reincrement count.
882 * shmem_unuse() search for an eventually swapped out shmem page.
884 int shmem_unuse(swp_entry_t entry
, struct page
*page
)
886 struct list_head
*p
, *next
;
887 struct shmem_inode_info
*info
;
890 spin_lock(&shmem_swaplist_lock
);
891 list_for_each_safe(p
, next
, &shmem_swaplist
) {
892 info
= list_entry(p
, struct shmem_inode_info
, swaplist
);
894 list_del_init(&info
->swaplist
);
895 else if (shmem_unuse_inode(info
, entry
, page
)) {
896 /* move head to start search for next from here */
897 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
902 spin_unlock(&shmem_swaplist_lock
);
907 * Move the page from the page cache to the swap cache.
909 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
911 struct shmem_inode_info
*info
;
912 swp_entry_t
*entry
, swap
;
913 struct address_space
*mapping
;
917 BUG_ON(!PageLocked(page
));
918 BUG_ON(page_mapped(page
));
920 mapping
= page
->mapping
;
922 inode
= mapping
->host
;
923 info
= SHMEM_I(inode
);
924 if (info
->flags
& VM_LOCKED
)
926 swap
= get_swap_page();
930 spin_lock(&info
->lock
);
931 shmem_recalc_inode(inode
);
932 if (index
>= info
->next_index
) {
933 BUG_ON(!(info
->flags
& SHMEM_TRUNCATE
));
936 entry
= shmem_swp_entry(info
, index
, NULL
);
940 if (move_to_swap_cache(page
, swap
) == 0) {
941 shmem_swp_set(info
, entry
, swap
.val
);
942 shmem_swp_unmap(entry
);
943 spin_unlock(&info
->lock
);
944 if (list_empty(&info
->swaplist
)) {
945 spin_lock(&shmem_swaplist_lock
);
946 /* move instead of add in case we're racing */
947 list_move_tail(&info
->swaplist
, &shmem_swaplist
);
948 spin_unlock(&shmem_swaplist_lock
);
954 shmem_swp_unmap(entry
);
956 spin_unlock(&info
->lock
);
959 set_page_dirty(page
);
960 return AOP_WRITEPAGE_ACTIVATE
; /* Return with the page locked */
964 static inline int shmem_parse_mpol(char *value
, int *policy
, nodemask_t
*policy_nodes
)
966 char *nodelist
= strchr(value
, ':');
970 /* NUL-terminate policy string */
972 if (nodelist_parse(nodelist
, *policy_nodes
))
974 if (!nodes_subset(*policy_nodes
, node_online_map
))
977 if (!strcmp(value
, "default")) {
978 *policy
= MPOL_DEFAULT
;
979 /* Don't allow a nodelist */
982 } else if (!strcmp(value
, "prefer")) {
983 *policy
= MPOL_PREFERRED
;
984 /* Insist on a nodelist of one node only */
986 char *rest
= nodelist
;
987 while (isdigit(*rest
))
992 } else if (!strcmp(value
, "bind")) {
994 /* Insist on a nodelist */
997 } else if (!strcmp(value
, "interleave")) {
998 *policy
= MPOL_INTERLEAVE
;
999 /* Default to nodes online if no nodelist */
1001 *policy_nodes
= node_online_map
;
1005 /* Restore string for error message */
1011 static struct page
*shmem_swapin_async(struct shared_policy
*p
,
1012 swp_entry_t entry
, unsigned long idx
)
1015 struct vm_area_struct pvma
;
1017 /* Create a pseudo vma that just contains the policy */
1018 memset(&pvma
, 0, sizeof(struct vm_area_struct
));
1019 pvma
.vm_end
= PAGE_SIZE
;
1020 pvma
.vm_pgoff
= idx
;
1021 pvma
.vm_policy
= mpol_shared_policy_lookup(p
, idx
);
1022 page
= read_swap_cache_async(entry
, &pvma
, 0);
1023 mpol_free(pvma
.vm_policy
);
1027 struct page
*shmem_swapin(struct shmem_inode_info
*info
, swp_entry_t entry
,
1030 struct shared_policy
*p
= &info
->policy
;
1033 unsigned long offset
;
1035 num
= valid_swaphandles(entry
, &offset
);
1036 for (i
= 0; i
< num
; offset
++, i
++) {
1037 page
= shmem_swapin_async(p
,
1038 swp_entry(swp_type(entry
), offset
), idx
);
1041 page_cache_release(page
);
1043 lru_add_drain(); /* Push any new pages onto the LRU now */
1044 return shmem_swapin_async(p
, entry
, idx
);
1047 static struct page
*
1048 shmem_alloc_page(gfp_t gfp
, struct shmem_inode_info
*info
,
1051 struct vm_area_struct pvma
;
1054 memset(&pvma
, 0, sizeof(struct vm_area_struct
));
1055 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, idx
);
1056 pvma
.vm_pgoff
= idx
;
1057 pvma
.vm_end
= PAGE_SIZE
;
1058 page
= alloc_page_vma(gfp
| __GFP_ZERO
, &pvma
, 0);
1059 mpol_free(pvma
.vm_policy
);
1063 static inline int shmem_parse_mpol(char *value
, int *policy
, nodemask_t
*policy_nodes
)
1068 static inline struct page
*
1069 shmem_swapin(struct shmem_inode_info
*info
,swp_entry_t entry
,unsigned long idx
)
1071 swapin_readahead(entry
, 0, NULL
);
1072 return read_swap_cache_async(entry
, NULL
, 0);
1075 static inline struct page
*
1076 shmem_alloc_page(gfp_t gfp
,struct shmem_inode_info
*info
, unsigned long idx
)
1078 return alloc_page(gfp
| __GFP_ZERO
);
1083 * shmem_getpage - either get the page from swap or allocate a new one
1085 * If we allocate a new one we do not mark it dirty. That's up to the
1086 * vm. If we swap it in we mark it dirty since we also free the swap
1087 * entry since a page cannot live in both the swap and page cache
1089 static int shmem_getpage(struct inode
*inode
, unsigned long idx
,
1090 struct page
**pagep
, enum sgp_type sgp
, int *type
)
1092 struct address_space
*mapping
= inode
->i_mapping
;
1093 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1094 struct shmem_sb_info
*sbinfo
;
1095 struct page
*filepage
= *pagep
;
1096 struct page
*swappage
;
1101 if (idx
>= SHMEM_MAX_INDEX
)
1108 * Normally, filepage is NULL on entry, and either found
1109 * uptodate immediately, or allocated and zeroed, or read
1110 * in under swappage, which is then assigned to filepage.
1111 * But shmem_readpage and shmem_write_begin pass in a locked
1112 * filepage, which may be found not uptodate by other callers
1113 * too, and may need to be copied from the swappage read in.
1117 filepage
= find_lock_page(mapping
, idx
);
1118 if (filepage
&& PageUptodate(filepage
))
1121 if (sgp
== SGP_QUICK
)
1124 spin_lock(&info
->lock
);
1125 shmem_recalc_inode(inode
);
1126 entry
= shmem_swp_alloc(info
, idx
, sgp
);
1127 if (IS_ERR(entry
)) {
1128 spin_unlock(&info
->lock
);
1129 error
= PTR_ERR(entry
);
1135 /* Look it up and read it in.. */
1136 swappage
= lookup_swap_cache(swap
);
1138 shmem_swp_unmap(entry
);
1139 /* here we actually do the io */
1140 if (type
&& !(*type
& VM_FAULT_MAJOR
)) {
1141 __count_vm_event(PGMAJFAULT
);
1142 *type
|= VM_FAULT_MAJOR
;
1144 spin_unlock(&info
->lock
);
1145 swappage
= shmem_swapin(info
, swap
, idx
);
1147 spin_lock(&info
->lock
);
1148 entry
= shmem_swp_alloc(info
, idx
, sgp
);
1150 error
= PTR_ERR(entry
);
1152 if (entry
->val
== swap
.val
)
1154 shmem_swp_unmap(entry
);
1156 spin_unlock(&info
->lock
);
1161 wait_on_page_locked(swappage
);
1162 page_cache_release(swappage
);
1166 /* We have to do this with page locked to prevent races */
1167 if (TestSetPageLocked(swappage
)) {
1168 shmem_swp_unmap(entry
);
1169 spin_unlock(&info
->lock
);
1170 wait_on_page_locked(swappage
);
1171 page_cache_release(swappage
);
1174 if (PageWriteback(swappage
)) {
1175 shmem_swp_unmap(entry
);
1176 spin_unlock(&info
->lock
);
1177 wait_on_page_writeback(swappage
);
1178 unlock_page(swappage
);
1179 page_cache_release(swappage
);
1182 if (!PageUptodate(swappage
)) {
1183 shmem_swp_unmap(entry
);
1184 spin_unlock(&info
->lock
);
1185 unlock_page(swappage
);
1186 page_cache_release(swappage
);
1192 shmem_swp_set(info
, entry
, 0);
1193 shmem_swp_unmap(entry
);
1194 delete_from_swap_cache(swappage
);
1195 spin_unlock(&info
->lock
);
1196 copy_highpage(filepage
, swappage
);
1197 unlock_page(swappage
);
1198 page_cache_release(swappage
);
1199 flush_dcache_page(filepage
);
1200 SetPageUptodate(filepage
);
1201 set_page_dirty(filepage
);
1203 } else if (!(error
= move_from_swap_cache(
1204 swappage
, idx
, mapping
))) {
1205 info
->flags
|= SHMEM_PAGEIN
;
1206 shmem_swp_set(info
, entry
, 0);
1207 shmem_swp_unmap(entry
);
1208 spin_unlock(&info
->lock
);
1209 filepage
= swappage
;
1212 shmem_swp_unmap(entry
);
1213 spin_unlock(&info
->lock
);
1214 unlock_page(swappage
);
1215 page_cache_release(swappage
);
1216 if (error
== -ENOMEM
) {
1217 /* let kswapd refresh zone for GFP_ATOMICs */
1218 congestion_wait(WRITE
, HZ
/50);
1222 } else if (sgp
== SGP_READ
&& !filepage
) {
1223 shmem_swp_unmap(entry
);
1224 filepage
= find_get_page(mapping
, idx
);
1226 (!PageUptodate(filepage
) || TestSetPageLocked(filepage
))) {
1227 spin_unlock(&info
->lock
);
1228 wait_on_page_locked(filepage
);
1229 page_cache_release(filepage
);
1233 spin_unlock(&info
->lock
);
1235 shmem_swp_unmap(entry
);
1236 sbinfo
= SHMEM_SB(inode
->i_sb
);
1237 if (sbinfo
->max_blocks
) {
1238 spin_lock(&sbinfo
->stat_lock
);
1239 if (sbinfo
->free_blocks
== 0 ||
1240 shmem_acct_block(info
->flags
)) {
1241 spin_unlock(&sbinfo
->stat_lock
);
1242 spin_unlock(&info
->lock
);
1246 sbinfo
->free_blocks
--;
1247 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1248 spin_unlock(&sbinfo
->stat_lock
);
1249 } else if (shmem_acct_block(info
->flags
)) {
1250 spin_unlock(&info
->lock
);
1256 spin_unlock(&info
->lock
);
1257 filepage
= shmem_alloc_page(mapping_gfp_mask(mapping
),
1261 shmem_unacct_blocks(info
->flags
, 1);
1262 shmem_free_blocks(inode
, 1);
1267 spin_lock(&info
->lock
);
1268 entry
= shmem_swp_alloc(info
, idx
, sgp
);
1270 error
= PTR_ERR(entry
);
1273 shmem_swp_unmap(entry
);
1275 if (error
|| swap
.val
|| 0 != add_to_page_cache_lru(
1276 filepage
, mapping
, idx
, GFP_ATOMIC
)) {
1277 spin_unlock(&info
->lock
);
1278 page_cache_release(filepage
);
1279 shmem_unacct_blocks(info
->flags
, 1);
1280 shmem_free_blocks(inode
, 1);
1286 info
->flags
|= SHMEM_PAGEIN
;
1290 spin_unlock(&info
->lock
);
1291 flush_dcache_page(filepage
);
1292 SetPageUptodate(filepage
);
1295 if (*pagep
!= filepage
) {
1297 if (sgp
!= SGP_FAULT
)
1298 unlock_page(filepage
);
1304 if (*pagep
!= filepage
) {
1305 unlock_page(filepage
);
1306 page_cache_release(filepage
);
1311 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1313 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1317 if (((loff_t
)vmf
->pgoff
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
))
1318 return VM_FAULT_SIGBUS
;
1320 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_FAULT
, &ret
);
1322 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1324 mark_page_accessed(vmf
->page
);
1325 return ret
| VM_FAULT_LOCKED
;
1329 int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
1331 struct inode
*i
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1332 return mpol_set_shared_policy(&SHMEM_I(i
)->policy
, vma
, new);
1336 shmem_get_policy(struct vm_area_struct
*vma
, unsigned long addr
)
1338 struct inode
*i
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1341 idx
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1342 return mpol_shared_policy_lookup(&SHMEM_I(i
)->policy
, idx
);
1346 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1348 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1349 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1350 int retval
= -ENOMEM
;
1352 spin_lock(&info
->lock
);
1353 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1354 if (!user_shm_lock(inode
->i_size
, user
))
1356 info
->flags
|= VM_LOCKED
;
1358 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1359 user_shm_unlock(inode
->i_size
, user
);
1360 info
->flags
&= ~VM_LOCKED
;
1364 spin_unlock(&info
->lock
);
1368 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1370 file_accessed(file
);
1371 vma
->vm_ops
= &shmem_vm_ops
;
1372 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
1376 static struct inode
*
1377 shmem_get_inode(struct super_block
*sb
, int mode
, dev_t dev
)
1379 struct inode
*inode
;
1380 struct shmem_inode_info
*info
;
1381 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1383 if (sbinfo
->max_inodes
) {
1384 spin_lock(&sbinfo
->stat_lock
);
1385 if (!sbinfo
->free_inodes
) {
1386 spin_unlock(&sbinfo
->stat_lock
);
1389 sbinfo
->free_inodes
--;
1390 spin_unlock(&sbinfo
->stat_lock
);
1393 inode
= new_inode(sb
);
1395 inode
->i_mode
= mode
;
1396 inode
->i_uid
= current
->fsuid
;
1397 inode
->i_gid
= current
->fsgid
;
1398 inode
->i_blocks
= 0;
1399 inode
->i_mapping
->a_ops
= &shmem_aops
;
1400 inode
->i_mapping
->backing_dev_info
= &shmem_backing_dev_info
;
1401 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1402 inode
->i_generation
= get_seconds();
1403 info
= SHMEM_I(inode
);
1404 memset(info
, 0, (char *)inode
- (char *)info
);
1405 spin_lock_init(&info
->lock
);
1406 INIT_LIST_HEAD(&info
->swaplist
);
1408 switch (mode
& S_IFMT
) {
1410 inode
->i_op
= &shmem_special_inode_operations
;
1411 init_special_inode(inode
, mode
, dev
);
1414 inode
->i_op
= &shmem_inode_operations
;
1415 inode
->i_fop
= &shmem_file_operations
;
1416 mpol_shared_policy_init(&info
->policy
, sbinfo
->policy
,
1417 &sbinfo
->policy_nodes
);
1421 /* Some things misbehave if size == 0 on a directory */
1422 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1423 inode
->i_op
= &shmem_dir_inode_operations
;
1424 inode
->i_fop
= &simple_dir_operations
;
1428 * Must not load anything in the rbtree,
1429 * mpol_free_shared_policy will not be called.
1431 mpol_shared_policy_init(&info
->policy
, MPOL_DEFAULT
,
1435 } else if (sbinfo
->max_inodes
) {
1436 spin_lock(&sbinfo
->stat_lock
);
1437 sbinfo
->free_inodes
++;
1438 spin_unlock(&sbinfo
->stat_lock
);
1444 static const struct inode_operations shmem_symlink_inode_operations
;
1445 static const struct inode_operations shmem_symlink_inline_operations
;
1448 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1449 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1450 * below the loop driver, in the generic fashion that many filesystems support.
1452 static int shmem_readpage(struct file
*file
, struct page
*page
)
1454 struct inode
*inode
= page
->mapping
->host
;
1455 int error
= shmem_getpage(inode
, page
->index
, &page
, SGP_CACHE
, NULL
);
1461 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1462 loff_t pos
, unsigned len
, unsigned flags
,
1463 struct page
**pagep
, void **fsdata
)
1465 struct inode
*inode
= mapping
->host
;
1466 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1468 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1472 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1473 loff_t pos
, unsigned len
, unsigned copied
,
1474 struct page
*page
, void *fsdata
)
1476 struct inode
*inode
= mapping
->host
;
1478 set_page_dirty(page
);
1479 page_cache_release(page
);
1481 if (pos
+copied
> inode
->i_size
)
1482 i_size_write(inode
, pos
+copied
);
1488 shmem_file_write(struct file
*file
, const char __user
*buf
, size_t count
, loff_t
*ppos
)
1490 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1492 unsigned long written
;
1495 if ((ssize_t
) count
< 0)
1498 if (!access_ok(VERIFY_READ
, buf
, count
))
1501 mutex_lock(&inode
->i_mutex
);
1506 err
= generic_write_checks(file
, &pos
, &count
, 0);
1510 err
= remove_suid(file
->f_path
.dentry
);
1514 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
1517 struct page
*page
= NULL
;
1518 unsigned long bytes
, index
, offset
;
1522 offset
= (pos
& (PAGE_CACHE_SIZE
-1)); /* Within page */
1523 index
= pos
>> PAGE_CACHE_SHIFT
;
1524 bytes
= PAGE_CACHE_SIZE
- offset
;
1529 * We don't hold page lock across copy from user -
1530 * what would it guard against? - so no deadlock here.
1531 * But it still may be a good idea to prefault below.
1534 err
= shmem_getpage(inode
, index
, &page
, SGP_WRITE
, NULL
);
1539 if (PageHighMem(page
)) {
1540 volatile unsigned char dummy
;
1541 __get_user(dummy
, buf
);
1542 __get_user(dummy
, buf
+ bytes
- 1);
1544 kaddr
= kmap_atomic(page
, KM_USER0
);
1545 left
= __copy_from_user_inatomic(kaddr
+ offset
,
1547 kunmap_atomic(kaddr
, KM_USER0
);
1551 left
= __copy_from_user(kaddr
+ offset
, buf
, bytes
);
1559 if (pos
> inode
->i_size
)
1560 i_size_write(inode
, pos
);
1562 flush_dcache_page(page
);
1563 set_page_dirty(page
);
1564 mark_page_accessed(page
);
1565 page_cache_release(page
);
1575 * Our dirty pages are not counted in nr_dirty,
1576 * and we do not attempt to balance dirty pages.
1586 mutex_unlock(&inode
->i_mutex
);
1590 static void do_shmem_file_read(struct file
*filp
, loff_t
*ppos
, read_descriptor_t
*desc
, read_actor_t actor
)
1592 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
1593 struct address_space
*mapping
= inode
->i_mapping
;
1594 unsigned long index
, offset
;
1596 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1597 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1600 struct page
*page
= NULL
;
1601 unsigned long end_index
, nr
, ret
;
1602 loff_t i_size
= i_size_read(inode
);
1604 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1605 if (index
> end_index
)
1607 if (index
== end_index
) {
1608 nr
= i_size
& ~PAGE_CACHE_MASK
;
1613 desc
->error
= shmem_getpage(inode
, index
, &page
, SGP_READ
, NULL
);
1615 if (desc
->error
== -EINVAL
)
1621 * We must evaluate after, since reads (unlike writes)
1622 * are called without i_mutex protection against truncate
1624 nr
= PAGE_CACHE_SIZE
;
1625 i_size
= i_size_read(inode
);
1626 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1627 if (index
== end_index
) {
1628 nr
= i_size
& ~PAGE_CACHE_MASK
;
1631 page_cache_release(page
);
1639 * If users can be writing to this page using arbitrary
1640 * virtual addresses, take care about potential aliasing
1641 * before reading the page on the kernel side.
1643 if (mapping_writably_mapped(mapping
))
1644 flush_dcache_page(page
);
1646 * Mark the page accessed if we read the beginning.
1649 mark_page_accessed(page
);
1651 page
= ZERO_PAGE(0);
1652 page_cache_get(page
);
1656 * Ok, we have the page, and it's up-to-date, so
1657 * now we can copy it to user space...
1659 * The actor routine returns how many bytes were actually used..
1660 * NOTE! This may not be the same as how much of a user buffer
1661 * we filled up (we may be padding etc), so we can only update
1662 * "pos" here (the actor routine has to update the user buffer
1663 * pointers and the remaining count).
1665 ret
= actor(desc
, page
, offset
, nr
);
1667 index
+= offset
>> PAGE_CACHE_SHIFT
;
1668 offset
&= ~PAGE_CACHE_MASK
;
1670 page_cache_release(page
);
1671 if (ret
!= nr
|| !desc
->count
)
1677 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1678 file_accessed(filp
);
1681 static ssize_t
shmem_file_read(struct file
*filp
, char __user
*buf
, size_t count
, loff_t
*ppos
)
1683 read_descriptor_t desc
;
1685 if ((ssize_t
) count
< 0)
1687 if (!access_ok(VERIFY_WRITE
, buf
, count
))
1697 do_shmem_file_read(filp
, ppos
, &desc
, file_read_actor
);
1699 return desc
.written
;
1703 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
1705 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
1707 buf
->f_type
= TMPFS_MAGIC
;
1708 buf
->f_bsize
= PAGE_CACHE_SIZE
;
1709 buf
->f_namelen
= NAME_MAX
;
1710 spin_lock(&sbinfo
->stat_lock
);
1711 if (sbinfo
->max_blocks
) {
1712 buf
->f_blocks
= sbinfo
->max_blocks
;
1713 buf
->f_bavail
= buf
->f_bfree
= sbinfo
->free_blocks
;
1715 if (sbinfo
->max_inodes
) {
1716 buf
->f_files
= sbinfo
->max_inodes
;
1717 buf
->f_ffree
= sbinfo
->free_inodes
;
1719 /* else leave those fields 0 like simple_statfs */
1720 spin_unlock(&sbinfo
->stat_lock
);
1725 * File creation. Allocate an inode, and we're done..
1728 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1730 struct inode
*inode
= shmem_get_inode(dir
->i_sb
, mode
, dev
);
1731 int error
= -ENOSPC
;
1734 error
= security_inode_init_security(inode
, dir
, NULL
, NULL
,
1737 if (error
!= -EOPNOTSUPP
) {
1742 error
= shmem_acl_init(inode
, dir
);
1747 if (dir
->i_mode
& S_ISGID
) {
1748 inode
->i_gid
= dir
->i_gid
;
1750 inode
->i_mode
|= S_ISGID
;
1752 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1753 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1754 d_instantiate(dentry
, inode
);
1755 dget(dentry
); /* Extra count - pin the dentry in core */
1760 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1764 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
1770 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1771 struct nameidata
*nd
)
1773 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
1779 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1781 struct inode
*inode
= old_dentry
->d_inode
;
1782 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1785 * No ordinary (disk based) filesystem counts links as inodes;
1786 * but each new link needs a new dentry, pinning lowmem, and
1787 * tmpfs dentries cannot be pruned until they are unlinked.
1789 if (sbinfo
->max_inodes
) {
1790 spin_lock(&sbinfo
->stat_lock
);
1791 if (!sbinfo
->free_inodes
) {
1792 spin_unlock(&sbinfo
->stat_lock
);
1795 sbinfo
->free_inodes
--;
1796 spin_unlock(&sbinfo
->stat_lock
);
1799 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1800 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1802 atomic_inc(&inode
->i_count
); /* New dentry reference */
1803 dget(dentry
); /* Extra pinning count for the created dentry */
1804 d_instantiate(dentry
, inode
);
1808 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
1810 struct inode
*inode
= dentry
->d_inode
;
1812 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
)) {
1813 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1814 if (sbinfo
->max_inodes
) {
1815 spin_lock(&sbinfo
->stat_lock
);
1816 sbinfo
->free_inodes
++;
1817 spin_unlock(&sbinfo
->stat_lock
);
1821 dir
->i_size
-= BOGO_DIRENT_SIZE
;
1822 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1824 dput(dentry
); /* Undo the count from "create" - this does all the work */
1828 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1830 if (!simple_empty(dentry
))
1833 drop_nlink(dentry
->d_inode
);
1835 return shmem_unlink(dir
, dentry
);
1839 * The VFS layer already does all the dentry stuff for rename,
1840 * we just have to decrement the usage count for the target if
1841 * it exists so that the VFS layer correctly free's it when it
1844 static int shmem_rename(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
1846 struct inode
*inode
= old_dentry
->d_inode
;
1847 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
1849 if (!simple_empty(new_dentry
))
1852 if (new_dentry
->d_inode
) {
1853 (void) shmem_unlink(new_dir
, new_dentry
);
1855 drop_nlink(old_dir
);
1856 } else if (they_are_dirs
) {
1857 drop_nlink(old_dir
);
1861 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
1862 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
1863 old_dir
->i_ctime
= old_dir
->i_mtime
=
1864 new_dir
->i_ctime
= new_dir
->i_mtime
=
1865 inode
->i_ctime
= CURRENT_TIME
;
1869 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1873 struct inode
*inode
;
1874 struct page
*page
= NULL
;
1876 struct shmem_inode_info
*info
;
1878 len
= strlen(symname
) + 1;
1879 if (len
> PAGE_CACHE_SIZE
)
1880 return -ENAMETOOLONG
;
1882 inode
= shmem_get_inode(dir
->i_sb
, S_IFLNK
|S_IRWXUGO
, 0);
1886 error
= security_inode_init_security(inode
, dir
, NULL
, NULL
,
1889 if (error
!= -EOPNOTSUPP
) {
1896 info
= SHMEM_I(inode
);
1897 inode
->i_size
= len
-1;
1898 if (len
<= (char *)inode
- (char *)info
) {
1900 memcpy(info
, symname
, len
);
1901 inode
->i_op
= &shmem_symlink_inline_operations
;
1903 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
1908 inode
->i_op
= &shmem_symlink_inode_operations
;
1909 kaddr
= kmap_atomic(page
, KM_USER0
);
1910 memcpy(kaddr
, symname
, len
);
1911 kunmap_atomic(kaddr
, KM_USER0
);
1912 set_page_dirty(page
);
1913 page_cache_release(page
);
1915 if (dir
->i_mode
& S_ISGID
)
1916 inode
->i_gid
= dir
->i_gid
;
1917 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1918 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1919 d_instantiate(dentry
, inode
);
1924 static void *shmem_follow_link_inline(struct dentry
*dentry
, struct nameidata
*nd
)
1926 nd_set_link(nd
, (char *)SHMEM_I(dentry
->d_inode
));
1930 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
1932 struct page
*page
= NULL
;
1933 int res
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
1934 nd_set_link(nd
, res
? ERR_PTR(res
) : kmap(page
));
1938 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
1940 if (!IS_ERR(nd_get_link(nd
))) {
1941 struct page
*page
= cookie
;
1943 mark_page_accessed(page
);
1944 page_cache_release(page
);
1948 static const struct inode_operations shmem_symlink_inline_operations
= {
1949 .readlink
= generic_readlink
,
1950 .follow_link
= shmem_follow_link_inline
,
1953 static const struct inode_operations shmem_symlink_inode_operations
= {
1954 .truncate
= shmem_truncate
,
1955 .readlink
= generic_readlink
,
1956 .follow_link
= shmem_follow_link
,
1957 .put_link
= shmem_put_link
,
1960 #ifdef CONFIG_TMPFS_POSIX_ACL
1962 * Superblocks without xattr inode operations will get security.* xattr
1963 * support from the VFS "for free". As soon as we have any other xattrs
1964 * like ACLs, we also need to implement the security.* handlers at
1965 * filesystem level, though.
1968 static size_t shmem_xattr_security_list(struct inode
*inode
, char *list
,
1969 size_t list_len
, const char *name
,
1972 return security_inode_listsecurity(inode
, list
, list_len
);
1975 static int shmem_xattr_security_get(struct inode
*inode
, const char *name
,
1976 void *buffer
, size_t size
)
1978 if (strcmp(name
, "") == 0)
1980 return security_inode_getsecurity(inode
, name
, buffer
, size
,
1984 static int shmem_xattr_security_set(struct inode
*inode
, const char *name
,
1985 const void *value
, size_t size
, int flags
)
1987 if (strcmp(name
, "") == 0)
1989 return security_inode_setsecurity(inode
, name
, value
, size
, flags
);
1992 static struct xattr_handler shmem_xattr_security_handler
= {
1993 .prefix
= XATTR_SECURITY_PREFIX
,
1994 .list
= shmem_xattr_security_list
,
1995 .get
= shmem_xattr_security_get
,
1996 .set
= shmem_xattr_security_set
,
1999 static struct xattr_handler
*shmem_xattr_handlers
[] = {
2000 &shmem_xattr_acl_access_handler
,
2001 &shmem_xattr_acl_default_handler
,
2002 &shmem_xattr_security_handler
,
2007 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2009 return ERR_PTR(-ESTALE
);
2012 static int shmem_match(struct inode
*ino
, void *vfh
)
2016 inum
= (inum
<< 32) | fh
[1];
2017 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2020 static struct dentry
*shmem_get_dentry(struct super_block
*sb
, void *vfh
)
2022 struct dentry
*de
= NULL
;
2023 struct inode
*inode
;
2026 inum
= (inum
<< 32) | fh
[1];
2028 inode
= ilookup5(sb
, (unsigned long)(inum
+fh
[0]), shmem_match
, vfh
);
2030 de
= d_find_alias(inode
);
2034 return de
? de
: ERR_PTR(-ESTALE
);
2037 static struct dentry
*shmem_decode_fh(struct super_block
*sb
, __u32
*fh
,
2039 int (*acceptable
)(void *context
, struct dentry
*de
),
2043 return ERR_PTR(-ESTALE
);
2045 return sb
->s_export_op
->find_exported_dentry(sb
, fh
, NULL
, acceptable
,
2049 static int shmem_encode_fh(struct dentry
*dentry
, __u32
*fh
, int *len
,
2052 struct inode
*inode
= dentry
->d_inode
;
2057 if (hlist_unhashed(&inode
->i_hash
)) {
2058 /* Unfortunately insert_inode_hash is not idempotent,
2059 * so as we hash inodes here rather than at creation
2060 * time, we need a lock to ensure we only try
2063 static DEFINE_SPINLOCK(lock
);
2065 if (hlist_unhashed(&inode
->i_hash
))
2066 __insert_inode_hash(inode
,
2067 inode
->i_ino
+ inode
->i_generation
);
2071 fh
[0] = inode
->i_generation
;
2072 fh
[1] = inode
->i_ino
;
2073 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2079 static struct export_operations shmem_export_ops
= {
2080 .get_parent
= shmem_get_parent
,
2081 .get_dentry
= shmem_get_dentry
,
2082 .encode_fh
= shmem_encode_fh
,
2083 .decode_fh
= shmem_decode_fh
,
2086 static int shmem_parse_options(char *options
, int *mode
, uid_t
*uid
,
2087 gid_t
*gid
, unsigned long *blocks
, unsigned long *inodes
,
2088 int *policy
, nodemask_t
*policy_nodes
)
2090 char *this_char
, *value
, *rest
;
2092 while (options
!= NULL
) {
2093 this_char
= options
;
2096 * NUL-terminate this option: unfortunately,
2097 * mount options form a comma-separated list,
2098 * but mpol's nodelist may also contain commas.
2100 options
= strchr(options
, ',');
2101 if (options
== NULL
)
2104 if (!isdigit(*options
)) {
2111 if ((value
= strchr(this_char
,'=')) != NULL
) {
2115 "tmpfs: No value for mount option '%s'\n",
2120 if (!strcmp(this_char
,"size")) {
2121 unsigned long long size
;
2122 size
= memparse(value
,&rest
);
2124 size
<<= PAGE_SHIFT
;
2125 size
*= totalram_pages
;
2131 *blocks
= size
>> PAGE_CACHE_SHIFT
;
2132 } else if (!strcmp(this_char
,"nr_blocks")) {
2133 *blocks
= memparse(value
,&rest
);
2136 } else if (!strcmp(this_char
,"nr_inodes")) {
2137 *inodes
= memparse(value
,&rest
);
2140 } else if (!strcmp(this_char
,"mode")) {
2143 *mode
= simple_strtoul(value
,&rest
,8);
2146 } else if (!strcmp(this_char
,"uid")) {
2149 *uid
= simple_strtoul(value
,&rest
,0);
2152 } else if (!strcmp(this_char
,"gid")) {
2155 *gid
= simple_strtoul(value
,&rest
,0);
2158 } else if (!strcmp(this_char
,"mpol")) {
2159 if (shmem_parse_mpol(value
,policy
,policy_nodes
))
2162 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2170 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2176 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2178 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2179 unsigned long max_blocks
= sbinfo
->max_blocks
;
2180 unsigned long max_inodes
= sbinfo
->max_inodes
;
2181 int policy
= sbinfo
->policy
;
2182 nodemask_t policy_nodes
= sbinfo
->policy_nodes
;
2183 unsigned long blocks
;
2184 unsigned long inodes
;
2185 int error
= -EINVAL
;
2187 if (shmem_parse_options(data
, NULL
, NULL
, NULL
, &max_blocks
,
2188 &max_inodes
, &policy
, &policy_nodes
))
2191 spin_lock(&sbinfo
->stat_lock
);
2192 blocks
= sbinfo
->max_blocks
- sbinfo
->free_blocks
;
2193 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2194 if (max_blocks
< blocks
)
2196 if (max_inodes
< inodes
)
2199 * Those tests also disallow limited->unlimited while any are in
2200 * use, so i_blocks will always be zero when max_blocks is zero;
2201 * but we must separately disallow unlimited->limited, because
2202 * in that case we have no record of how much is already in use.
2204 if (max_blocks
&& !sbinfo
->max_blocks
)
2206 if (max_inodes
&& !sbinfo
->max_inodes
)
2210 sbinfo
->max_blocks
= max_blocks
;
2211 sbinfo
->free_blocks
= max_blocks
- blocks
;
2212 sbinfo
->max_inodes
= max_inodes
;
2213 sbinfo
->free_inodes
= max_inodes
- inodes
;
2214 sbinfo
->policy
= policy
;
2215 sbinfo
->policy_nodes
= policy_nodes
;
2217 spin_unlock(&sbinfo
->stat_lock
);
2222 static void shmem_put_super(struct super_block
*sb
)
2224 kfree(sb
->s_fs_info
);
2225 sb
->s_fs_info
= NULL
;
2228 static int shmem_fill_super(struct super_block
*sb
,
2229 void *data
, int silent
)
2231 struct inode
*inode
;
2232 struct dentry
*root
;
2233 int mode
= S_IRWXUGO
| S_ISVTX
;
2234 uid_t uid
= current
->fsuid
;
2235 gid_t gid
= current
->fsgid
;
2237 struct shmem_sb_info
*sbinfo
;
2238 unsigned long blocks
= 0;
2239 unsigned long inodes
= 0;
2240 int policy
= MPOL_DEFAULT
;
2241 nodemask_t policy_nodes
= node_online_map
;
2245 * Per default we only allow half of the physical ram per
2246 * tmpfs instance, limiting inodes to one per page of lowmem;
2247 * but the internal instance is left unlimited.
2249 if (!(sb
->s_flags
& MS_NOUSER
)) {
2250 blocks
= totalram_pages
/ 2;
2251 inodes
= totalram_pages
- totalhigh_pages
;
2252 if (inodes
> blocks
)
2254 if (shmem_parse_options(data
, &mode
, &uid
, &gid
, &blocks
,
2255 &inodes
, &policy
, &policy_nodes
))
2258 sb
->s_export_op
= &shmem_export_ops
;
2260 sb
->s_flags
|= MS_NOUSER
;
2263 /* Round up to L1_CACHE_BYTES to resist false sharing */
2264 sbinfo
= kmalloc(max((int)sizeof(struct shmem_sb_info
),
2265 L1_CACHE_BYTES
), GFP_KERNEL
);
2269 spin_lock_init(&sbinfo
->stat_lock
);
2270 sbinfo
->max_blocks
= blocks
;
2271 sbinfo
->free_blocks
= blocks
;
2272 sbinfo
->max_inodes
= inodes
;
2273 sbinfo
->free_inodes
= inodes
;
2274 sbinfo
->policy
= policy
;
2275 sbinfo
->policy_nodes
= policy_nodes
;
2277 sb
->s_fs_info
= sbinfo
;
2278 sb
->s_maxbytes
= SHMEM_MAX_BYTES
;
2279 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
2280 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
2281 sb
->s_magic
= TMPFS_MAGIC
;
2282 sb
->s_op
= &shmem_ops
;
2283 sb
->s_time_gran
= 1;
2284 #ifdef CONFIG_TMPFS_POSIX_ACL
2285 sb
->s_xattr
= shmem_xattr_handlers
;
2286 sb
->s_flags
|= MS_POSIXACL
;
2289 inode
= shmem_get_inode(sb
, S_IFDIR
| mode
, 0);
2294 root
= d_alloc_root(inode
);
2303 shmem_put_super(sb
);
2307 static struct kmem_cache
*shmem_inode_cachep
;
2309 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
2311 struct shmem_inode_info
*p
;
2312 p
= (struct shmem_inode_info
*)kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
2315 return &p
->vfs_inode
;
2318 static void shmem_destroy_inode(struct inode
*inode
)
2320 if ((inode
->i_mode
& S_IFMT
) == S_IFREG
) {
2321 /* only struct inode is valid if it's an inline symlink */
2322 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
2324 shmem_acl_destroy_inode(inode
);
2325 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
2328 static void init_once(void *foo
, struct kmem_cache
*cachep
,
2329 unsigned long flags
)
2331 struct shmem_inode_info
*p
= (struct shmem_inode_info
*) foo
;
2333 inode_init_once(&p
->vfs_inode
);
2334 #ifdef CONFIG_TMPFS_POSIX_ACL
2336 p
->i_default_acl
= NULL
;
2340 static int init_inodecache(void)
2342 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
2343 sizeof(struct shmem_inode_info
),
2345 if (shmem_inode_cachep
== NULL
)
2350 static void destroy_inodecache(void)
2352 kmem_cache_destroy(shmem_inode_cachep
);
2355 static const struct address_space_operations shmem_aops
= {
2356 .writepage
= shmem_writepage
,
2357 .set_page_dirty
= __set_page_dirty_no_writeback
,
2359 .readpage
= shmem_readpage
,
2360 .write_begin
= shmem_write_begin
,
2361 .write_end
= shmem_write_end
,
2363 .migratepage
= migrate_page
,
2366 static const struct file_operations shmem_file_operations
= {
2369 .llseek
= generic_file_llseek
,
2370 .read
= shmem_file_read
,
2371 .write
= shmem_file_write
,
2372 .fsync
= simple_sync_file
,
2373 .splice_read
= generic_file_splice_read
,
2374 .splice_write
= generic_file_splice_write
,
2378 static const struct inode_operations shmem_inode_operations
= {
2379 .truncate
= shmem_truncate
,
2380 .setattr
= shmem_notify_change
,
2381 .truncate_range
= shmem_truncate_range
,
2382 #ifdef CONFIG_TMPFS_POSIX_ACL
2383 .setxattr
= generic_setxattr
,
2384 .getxattr
= generic_getxattr
,
2385 .listxattr
= generic_listxattr
,
2386 .removexattr
= generic_removexattr
,
2387 .permission
= shmem_permission
,
2392 static const struct inode_operations shmem_dir_inode_operations
= {
2394 .create
= shmem_create
,
2395 .lookup
= simple_lookup
,
2397 .unlink
= shmem_unlink
,
2398 .symlink
= shmem_symlink
,
2399 .mkdir
= shmem_mkdir
,
2400 .rmdir
= shmem_rmdir
,
2401 .mknod
= shmem_mknod
,
2402 .rename
= shmem_rename
,
2404 #ifdef CONFIG_TMPFS_POSIX_ACL
2405 .setattr
= shmem_notify_change
,
2406 .setxattr
= generic_setxattr
,
2407 .getxattr
= generic_getxattr
,
2408 .listxattr
= generic_listxattr
,
2409 .removexattr
= generic_removexattr
,
2410 .permission
= shmem_permission
,
2414 static const struct inode_operations shmem_special_inode_operations
= {
2415 #ifdef CONFIG_TMPFS_POSIX_ACL
2416 .setattr
= shmem_notify_change
,
2417 .setxattr
= generic_setxattr
,
2418 .getxattr
= generic_getxattr
,
2419 .listxattr
= generic_listxattr
,
2420 .removexattr
= generic_removexattr
,
2421 .permission
= shmem_permission
,
2425 static const struct super_operations shmem_ops
= {
2426 .alloc_inode
= shmem_alloc_inode
,
2427 .destroy_inode
= shmem_destroy_inode
,
2429 .statfs
= shmem_statfs
,
2430 .remount_fs
= shmem_remount_fs
,
2432 .delete_inode
= shmem_delete_inode
,
2433 .drop_inode
= generic_delete_inode
,
2434 .put_super
= shmem_put_super
,
2437 static struct vm_operations_struct shmem_vm_ops
= {
2438 .fault
= shmem_fault
,
2440 .set_policy
= shmem_set_policy
,
2441 .get_policy
= shmem_get_policy
,
2446 static int shmem_get_sb(struct file_system_type
*fs_type
,
2447 int flags
, const char *dev_name
, void *data
, struct vfsmount
*mnt
)
2449 return get_sb_nodev(fs_type
, flags
, data
, shmem_fill_super
, mnt
);
2452 static struct file_system_type tmpfs_fs_type
= {
2453 .owner
= THIS_MODULE
,
2455 .get_sb
= shmem_get_sb
,
2456 .kill_sb
= kill_litter_super
,
2458 static struct vfsmount
*shm_mnt
;
2460 static int __init
init_tmpfs(void)
2464 error
= init_inodecache();
2468 error
= register_filesystem(&tmpfs_fs_type
);
2470 printk(KERN_ERR
"Could not register tmpfs\n");
2474 shm_mnt
= vfs_kern_mount(&tmpfs_fs_type
, MS_NOUSER
,
2475 tmpfs_fs_type
.name
, NULL
);
2476 if (IS_ERR(shm_mnt
)) {
2477 error
= PTR_ERR(shm_mnt
);
2478 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
2484 unregister_filesystem(&tmpfs_fs_type
);
2486 destroy_inodecache();
2488 shm_mnt
= ERR_PTR(error
);
2491 module_init(init_tmpfs
)
2494 * shmem_file_setup - get an unlinked file living in tmpfs
2496 * @name: name for dentry (to be seen in /proc/<pid>/maps
2497 * @size: size to be set for the file
2500 struct file
*shmem_file_setup(char *name
, loff_t size
, unsigned long flags
)
2504 struct inode
*inode
;
2505 struct dentry
*dentry
, *root
;
2508 if (IS_ERR(shm_mnt
))
2509 return (void *)shm_mnt
;
2511 if (size
< 0 || size
> SHMEM_MAX_BYTES
)
2512 return ERR_PTR(-EINVAL
);
2514 if (shmem_acct_size(flags
, size
))
2515 return ERR_PTR(-ENOMEM
);
2519 this.len
= strlen(name
);
2520 this.hash
= 0; /* will go */
2521 root
= shm_mnt
->mnt_root
;
2522 dentry
= d_alloc(root
, &this);
2527 file
= get_empty_filp();
2532 inode
= shmem_get_inode(root
->d_sb
, S_IFREG
| S_IRWXUGO
, 0);
2536 SHMEM_I(inode
)->flags
= flags
& VM_ACCOUNT
;
2537 d_instantiate(dentry
, inode
);
2538 inode
->i_size
= size
;
2539 inode
->i_nlink
= 0; /* It is unlinked */
2540 file
->f_path
.mnt
= mntget(shm_mnt
);
2541 file
->f_path
.dentry
= dentry
;
2542 file
->f_mapping
= inode
->i_mapping
;
2543 file
->f_op
= &shmem_file_operations
;
2544 file
->f_mode
= FMODE_WRITE
| FMODE_READ
;
2552 shmem_unacct_size(flags
, size
);
2553 return ERR_PTR(error
);
2557 * shmem_zero_setup - setup a shared anonymous mapping
2559 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2561 int shmem_zero_setup(struct vm_area_struct
*vma
)
2564 loff_t size
= vma
->vm_end
- vma
->vm_start
;
2566 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
);
2568 return PTR_ERR(file
);
2572 vma
->vm_file
= file
;
2573 vma
->vm_ops
= &shmem_vm_ops
;