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>
18 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 * This file is released under the GPL.
24 #include <linux/init.h>
25 #include <linux/vfs.h>
26 #include <linux/mount.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
30 #include <linux/module.h>
31 #include <linux/percpu_counter.h>
32 #include <linux/swap.h>
34 static struct vfsmount
*shm_mnt
;
38 * This virtual memory filesystem is heavily based on the ramfs. It
39 * extends ramfs by the ability to use swap and honor resource limits
40 * which makes it a completely usable filesystem.
43 #include <linux/xattr.h>
44 #include <linux/exportfs.h>
45 #include <linux/posix_acl.h>
46 #include <linux/generic_acl.h>
47 #include <linux/mman.h>
48 #include <linux/string.h>
49 #include <linux/slab.h>
50 #include <linux/backing-dev.h>
51 #include <linux/shmem_fs.h>
52 #include <linux/writeback.h>
53 #include <linux/blkdev.h>
54 #include <linux/security.h>
55 #include <linux/swapops.h>
56 #include <linux/mempolicy.h>
57 #include <linux/namei.h>
58 #include <linux/ctype.h>
59 #include <linux/migrate.h>
60 #include <linux/highmem.h>
61 #include <linux/seq_file.h>
62 #include <linux/magic.h>
64 #include <asm/uaccess.h>
65 #include <asm/div64.h>
66 #include <asm/pgtable.h>
69 * The maximum size of a shmem/tmpfs file is limited by the maximum size of
70 * its triple-indirect swap vector - see illustration at shmem_swp_entry().
72 * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
73 * but one eighth of that on a 64-bit kernel. With 8kB page size, maximum
74 * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
75 * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
77 * We use / and * instead of shifts in the definitions below, so that the swap
78 * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
80 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
81 #define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
83 #define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
84 #define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
86 #define SHMEM_MAX_BYTES min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
87 #define SHMEM_MAX_INDEX ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
89 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
90 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
92 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
93 #define SHMEM_PAGEIN VM_READ
94 #define SHMEM_TRUNCATE VM_WRITE
96 /* Definition to limit shmem_truncate's steps between cond_rescheds */
97 #define LATENCY_LIMIT 64
99 /* Pretend that each entry is of this size in directory's i_size */
100 #define BOGO_DIRENT_SIZE 20
102 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
104 SGP_READ
, /* don't exceed i_size, don't allocate page */
105 SGP_CACHE
, /* don't exceed i_size, may allocate page */
106 SGP_DIRTY
, /* like SGP_CACHE, but set new page dirty */
107 SGP_WRITE
, /* may exceed i_size, may allocate page */
111 static unsigned long shmem_default_max_blocks(void)
113 return totalram_pages
/ 2;
116 static unsigned long shmem_default_max_inodes(void)
118 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
122 static int shmem_getpage(struct inode
*inode
, unsigned long idx
,
123 struct page
**pagep
, enum sgp_type sgp
, int *type
);
125 static inline struct page
*shmem_dir_alloc(gfp_t gfp_mask
)
128 * The above definition of ENTRIES_PER_PAGE, and the use of
129 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
130 * might be reconsidered if it ever diverges from PAGE_SIZE.
132 * Mobility flags are masked out as swap vectors cannot move
134 return alloc_pages((gfp_mask
& ~GFP_MOVABLE_MASK
) | __GFP_ZERO
,
135 PAGE_CACHE_SHIFT
-PAGE_SHIFT
);
138 static inline void shmem_dir_free(struct page
*page
)
140 __free_pages(page
, PAGE_CACHE_SHIFT
-PAGE_SHIFT
);
143 static struct page
**shmem_dir_map(struct page
*page
)
145 return (struct page
**)kmap_atomic(page
, KM_USER0
);
148 static inline void shmem_dir_unmap(struct page
**dir
)
150 kunmap_atomic(dir
, KM_USER0
);
153 static swp_entry_t
*shmem_swp_map(struct page
*page
)
155 return (swp_entry_t
*)kmap_atomic(page
, KM_USER1
);
158 static inline void shmem_swp_balance_unmap(void)
161 * When passing a pointer to an i_direct entry, to code which
162 * also handles indirect entries and so will shmem_swp_unmap,
163 * we must arrange for the preempt count to remain in balance.
164 * What kmap_atomic of a lowmem page does depends on config
165 * and architecture, so pretend to kmap_atomic some lowmem page.
167 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1
);
170 static inline void shmem_swp_unmap(swp_entry_t
*entry
)
172 kunmap_atomic(entry
, KM_USER1
);
175 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
177 return sb
->s_fs_info
;
181 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
182 * for shared memory and for shared anonymous (/dev/zero) mappings
183 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
184 * consistent with the pre-accounting of private mappings ...
186 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
188 return (flags
& VM_NORESERVE
) ?
189 0 : security_vm_enough_memory_kern(VM_ACCT(size
));
192 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
194 if (!(flags
& VM_NORESERVE
))
195 vm_unacct_memory(VM_ACCT(size
));
199 * ... whereas tmpfs objects are accounted incrementally as
200 * pages are allocated, in order to allow huge sparse files.
201 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
202 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
204 static inline int shmem_acct_block(unsigned long flags
)
206 return (flags
& VM_NORESERVE
) ?
207 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
210 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
212 if (flags
& VM_NORESERVE
)
213 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
216 static const struct super_operations shmem_ops
;
217 static const struct address_space_operations shmem_aops
;
218 static const struct file_operations shmem_file_operations
;
219 static const struct inode_operations shmem_inode_operations
;
220 static const struct inode_operations shmem_dir_inode_operations
;
221 static const struct inode_operations shmem_special_inode_operations
;
222 static const struct vm_operations_struct shmem_vm_ops
;
224 static struct backing_dev_info shmem_backing_dev_info __read_mostly
= {
225 .ra_pages
= 0, /* No readahead */
226 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
| BDI_CAP_SWAP_BACKED
,
229 static LIST_HEAD(shmem_swaplist
);
230 static DEFINE_MUTEX(shmem_swaplist_mutex
);
232 static void shmem_free_blocks(struct inode
*inode
, long pages
)
234 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
235 if (sbinfo
->max_blocks
) {
236 percpu_counter_add(&sbinfo
->used_blocks
, -pages
);
237 spin_lock(&inode
->i_lock
);
238 inode
->i_blocks
-= pages
*BLOCKS_PER_PAGE
;
239 spin_unlock(&inode
->i_lock
);
243 static int shmem_reserve_inode(struct super_block
*sb
)
245 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
246 if (sbinfo
->max_inodes
) {
247 spin_lock(&sbinfo
->stat_lock
);
248 if (!sbinfo
->free_inodes
) {
249 spin_unlock(&sbinfo
->stat_lock
);
252 sbinfo
->free_inodes
--;
253 spin_unlock(&sbinfo
->stat_lock
);
258 static void shmem_free_inode(struct super_block
*sb
)
260 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
261 if (sbinfo
->max_inodes
) {
262 spin_lock(&sbinfo
->stat_lock
);
263 sbinfo
->free_inodes
++;
264 spin_unlock(&sbinfo
->stat_lock
);
269 * shmem_recalc_inode - recalculate the size of an inode
270 * @inode: inode to recalc
272 * We have to calculate the free blocks since the mm can drop
273 * undirtied hole pages behind our back.
275 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
276 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
278 * It has to be called with the spinlock held.
280 static void shmem_recalc_inode(struct inode
*inode
)
282 struct shmem_inode_info
*info
= SHMEM_I(inode
);
285 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
287 info
->alloced
-= freed
;
288 shmem_unacct_blocks(info
->flags
, freed
);
289 shmem_free_blocks(inode
, freed
);
294 * shmem_swp_entry - find the swap vector position in the info structure
295 * @info: info structure for the inode
296 * @index: index of the page to find
297 * @page: optional page to add to the structure. Has to be preset to
300 * If there is no space allocated yet it will return NULL when
301 * page is NULL, else it will use the page for the needed block,
302 * setting it to NULL on return to indicate that it has been used.
304 * The swap vector is organized the following way:
306 * There are SHMEM_NR_DIRECT entries directly stored in the
307 * shmem_inode_info structure. So small files do not need an addional
310 * For pages with index > SHMEM_NR_DIRECT there is the pointer
311 * i_indirect which points to a page which holds in the first half
312 * doubly indirect blocks, in the second half triple indirect blocks:
314 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
315 * following layout (for SHMEM_NR_DIRECT == 16):
317 * i_indirect -> dir --> 16-19
330 static swp_entry_t
*shmem_swp_entry(struct shmem_inode_info
*info
, unsigned long index
, struct page
**page
)
332 unsigned long offset
;
336 if (index
< SHMEM_NR_DIRECT
) {
337 shmem_swp_balance_unmap();
338 return info
->i_direct
+index
;
340 if (!info
->i_indirect
) {
342 info
->i_indirect
= *page
;
345 return NULL
; /* need another page */
348 index
-= SHMEM_NR_DIRECT
;
349 offset
= index
% ENTRIES_PER_PAGE
;
350 index
/= ENTRIES_PER_PAGE
;
351 dir
= shmem_dir_map(info
->i_indirect
);
353 if (index
>= ENTRIES_PER_PAGE
/2) {
354 index
-= ENTRIES_PER_PAGE
/2;
355 dir
+= ENTRIES_PER_PAGE
/2 + index
/ENTRIES_PER_PAGE
;
356 index
%= ENTRIES_PER_PAGE
;
363 shmem_dir_unmap(dir
);
364 return NULL
; /* need another page */
366 shmem_dir_unmap(dir
);
367 dir
= shmem_dir_map(subdir
);
373 if (!page
|| !(subdir
= *page
)) {
374 shmem_dir_unmap(dir
);
375 return NULL
; /* need a page */
380 shmem_dir_unmap(dir
);
381 return shmem_swp_map(subdir
) + offset
;
384 static void shmem_swp_set(struct shmem_inode_info
*info
, swp_entry_t
*entry
, unsigned long value
)
386 long incdec
= value
? 1: -1;
389 info
->swapped
+= incdec
;
390 if ((unsigned long)(entry
- info
->i_direct
) >= SHMEM_NR_DIRECT
) {
391 struct page
*page
= kmap_atomic_to_page(entry
);
392 set_page_private(page
, page_private(page
) + incdec
);
397 * shmem_swp_alloc - get the position of the swap entry for the page.
398 * @info: info structure for the inode
399 * @index: index of the page to find
400 * @sgp: check and recheck i_size? skip allocation?
402 * If the entry does not exist, allocate it.
404 static swp_entry_t
*shmem_swp_alloc(struct shmem_inode_info
*info
, unsigned long index
, enum sgp_type sgp
)
406 struct inode
*inode
= &info
->vfs_inode
;
407 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
408 struct page
*page
= NULL
;
411 if (sgp
!= SGP_WRITE
&&
412 ((loff_t
) index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
))
413 return ERR_PTR(-EINVAL
);
415 while (!(entry
= shmem_swp_entry(info
, index
, &page
))) {
417 return shmem_swp_map(ZERO_PAGE(0));
419 * Test used_blocks against 1 less max_blocks, since we have 1 data
420 * page (and perhaps indirect index pages) yet to allocate:
421 * a waste to allocate index if we cannot allocate data.
423 if (sbinfo
->max_blocks
) {
424 if (percpu_counter_compare(&sbinfo
->used_blocks
,
425 sbinfo
->max_blocks
- 1) >= 0)
426 return ERR_PTR(-ENOSPC
);
427 percpu_counter_inc(&sbinfo
->used_blocks
);
428 spin_lock(&inode
->i_lock
);
429 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
430 spin_unlock(&inode
->i_lock
);
433 spin_unlock(&info
->lock
);
434 page
= shmem_dir_alloc(mapping_gfp_mask(inode
->i_mapping
));
435 spin_lock(&info
->lock
);
438 shmem_free_blocks(inode
, 1);
439 return ERR_PTR(-ENOMEM
);
441 if (sgp
!= SGP_WRITE
&&
442 ((loff_t
) index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
443 entry
= ERR_PTR(-EINVAL
);
446 if (info
->next_index
<= index
)
447 info
->next_index
= index
+ 1;
450 /* another task gave its page, or truncated the file */
451 shmem_free_blocks(inode
, 1);
452 shmem_dir_free(page
);
454 if (info
->next_index
<= index
&& !IS_ERR(entry
))
455 info
->next_index
= index
+ 1;
460 * shmem_free_swp - free some swap entries in a directory
461 * @dir: pointer to the directory
462 * @edir: pointer after last entry of the directory
463 * @punch_lock: pointer to spinlock when needed for the holepunch case
465 static int shmem_free_swp(swp_entry_t
*dir
, swp_entry_t
*edir
,
466 spinlock_t
*punch_lock
)
468 spinlock_t
*punch_unlock
= NULL
;
472 for (ptr
= dir
; ptr
< edir
; ptr
++) {
474 if (unlikely(punch_lock
)) {
475 punch_unlock
= punch_lock
;
477 spin_lock(punch_unlock
);
481 free_swap_and_cache(*ptr
);
482 *ptr
= (swp_entry_t
){0};
487 spin_unlock(punch_unlock
);
491 static int shmem_map_and_free_swp(struct page
*subdir
, int offset
,
492 int limit
, struct page
***dir
, spinlock_t
*punch_lock
)
497 ptr
= shmem_swp_map(subdir
);
498 for (; offset
< limit
; offset
+= LATENCY_LIMIT
) {
499 int size
= limit
- offset
;
500 if (size
> LATENCY_LIMIT
)
501 size
= LATENCY_LIMIT
;
502 freed
+= shmem_free_swp(ptr
+offset
, ptr
+offset
+size
,
504 if (need_resched()) {
505 shmem_swp_unmap(ptr
);
507 shmem_dir_unmap(*dir
);
511 ptr
= shmem_swp_map(subdir
);
514 shmem_swp_unmap(ptr
);
518 static void shmem_free_pages(struct list_head
*next
)
524 page
= container_of(next
, struct page
, lru
);
526 shmem_dir_free(page
);
528 if (freed
>= LATENCY_LIMIT
) {
535 static void shmem_truncate_range(struct inode
*inode
, loff_t start
, loff_t end
)
537 struct shmem_inode_info
*info
= SHMEM_I(inode
);
542 unsigned long diroff
;
548 LIST_HEAD(pages_to_free
);
549 long nr_pages_to_free
= 0;
550 long nr_swaps_freed
= 0;
554 spinlock_t
*needs_lock
;
555 spinlock_t
*punch_lock
;
556 unsigned long upper_limit
;
558 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
559 idx
= (start
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
560 if (idx
>= info
->next_index
)
563 spin_lock(&info
->lock
);
564 info
->flags
|= SHMEM_TRUNCATE
;
565 if (likely(end
== (loff_t
) -1)) {
566 limit
= info
->next_index
;
567 upper_limit
= SHMEM_MAX_INDEX
;
568 info
->next_index
= idx
;
572 if (end
+ 1 >= inode
->i_size
) { /* we may free a little more */
573 limit
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >>
575 upper_limit
= SHMEM_MAX_INDEX
;
577 limit
= (end
+ 1) >> PAGE_CACHE_SHIFT
;
580 needs_lock
= &info
->lock
;
584 topdir
= info
->i_indirect
;
585 if (topdir
&& idx
<= SHMEM_NR_DIRECT
&& !punch_hole
) {
586 info
->i_indirect
= NULL
;
588 list_add(&topdir
->lru
, &pages_to_free
);
590 spin_unlock(&info
->lock
);
592 if (info
->swapped
&& idx
< SHMEM_NR_DIRECT
) {
593 ptr
= info
->i_direct
;
595 if (size
> SHMEM_NR_DIRECT
)
596 size
= SHMEM_NR_DIRECT
;
597 nr_swaps_freed
= shmem_free_swp(ptr
+idx
, ptr
+size
, needs_lock
);
601 * If there are no indirect blocks or we are punching a hole
602 * below indirect blocks, nothing to be done.
604 if (!topdir
|| limit
<= SHMEM_NR_DIRECT
)
608 * The truncation case has already dropped info->lock, and we're safe
609 * because i_size and next_index have already been lowered, preventing
610 * access beyond. But in the punch_hole case, we still need to take
611 * the lock when updating the swap directory, because there might be
612 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
613 * shmem_writepage. However, whenever we find we can remove a whole
614 * directory page (not at the misaligned start or end of the range),
615 * we first NULLify its pointer in the level above, and then have no
616 * need to take the lock when updating its contents: needs_lock and
617 * punch_lock (either pointing to info->lock or NULL) manage this.
620 upper_limit
-= SHMEM_NR_DIRECT
;
621 limit
-= SHMEM_NR_DIRECT
;
622 idx
= (idx
> SHMEM_NR_DIRECT
)? (idx
- SHMEM_NR_DIRECT
): 0;
623 offset
= idx
% ENTRIES_PER_PAGE
;
626 dir
= shmem_dir_map(topdir
);
627 stage
= ENTRIES_PER_PAGEPAGE
/2;
628 if (idx
< ENTRIES_PER_PAGEPAGE
/2) {
630 diroff
= idx
/ENTRIES_PER_PAGE
;
632 dir
+= ENTRIES_PER_PAGE
/2;
633 dir
+= (idx
- ENTRIES_PER_PAGEPAGE
/2)/ENTRIES_PER_PAGEPAGE
;
635 stage
+= ENTRIES_PER_PAGEPAGE
;
638 diroff
= ((idx
- ENTRIES_PER_PAGEPAGE
/2) %
639 ENTRIES_PER_PAGEPAGE
) / ENTRIES_PER_PAGE
;
640 if (!diroff
&& !offset
&& upper_limit
>= stage
) {
642 spin_lock(needs_lock
);
644 spin_unlock(needs_lock
);
649 list_add(&middir
->lru
, &pages_to_free
);
651 shmem_dir_unmap(dir
);
652 dir
= shmem_dir_map(middir
);
660 for (; idx
< limit
; idx
+= ENTRIES_PER_PAGE
, diroff
++) {
661 if (unlikely(idx
== stage
)) {
662 shmem_dir_unmap(dir
);
663 dir
= shmem_dir_map(topdir
) +
664 ENTRIES_PER_PAGE
/2 + idx
/ENTRIES_PER_PAGEPAGE
;
667 idx
+= ENTRIES_PER_PAGEPAGE
;
671 stage
= idx
+ ENTRIES_PER_PAGEPAGE
;
674 needs_lock
= &info
->lock
;
675 if (upper_limit
>= stage
) {
677 spin_lock(needs_lock
);
679 spin_unlock(needs_lock
);
684 list_add(&middir
->lru
, &pages_to_free
);
686 shmem_dir_unmap(dir
);
688 dir
= shmem_dir_map(middir
);
691 punch_lock
= needs_lock
;
692 subdir
= dir
[diroff
];
693 if (subdir
&& !offset
&& upper_limit
-idx
>= ENTRIES_PER_PAGE
) {
695 spin_lock(needs_lock
);
697 spin_unlock(needs_lock
);
702 list_add(&subdir
->lru
, &pages_to_free
);
704 if (subdir
&& page_private(subdir
) /* has swap entries */) {
706 if (size
> ENTRIES_PER_PAGE
)
707 size
= ENTRIES_PER_PAGE
;
708 freed
= shmem_map_and_free_swp(subdir
,
709 offset
, size
, &dir
, punch_lock
);
711 dir
= shmem_dir_map(middir
);
712 nr_swaps_freed
+= freed
;
713 if (offset
|| punch_lock
) {
714 spin_lock(&info
->lock
);
715 set_page_private(subdir
,
716 page_private(subdir
) - freed
);
717 spin_unlock(&info
->lock
);
719 BUG_ON(page_private(subdir
) != freed
);
724 shmem_dir_unmap(dir
);
726 if (inode
->i_mapping
->nrpages
&& (info
->flags
& SHMEM_PAGEIN
)) {
728 * Call truncate_inode_pages again: racing shmem_unuse_inode
729 * may have swizzled a page in from swap since
730 * truncate_pagecache or generic_delete_inode did it, before we
731 * lowered next_index. Also, though shmem_getpage checks
732 * i_size before adding to cache, no recheck after: so fix the
733 * narrow window there too.
735 * Recalling truncate_inode_pages_range and unmap_mapping_range
736 * every time for punch_hole (which never got a chance to clear
737 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
738 * yet hardly ever necessary: try to optimize them out later.
740 truncate_inode_pages_range(inode
->i_mapping
, start
, end
);
742 unmap_mapping_range(inode
->i_mapping
, start
,
746 spin_lock(&info
->lock
);
747 info
->flags
&= ~SHMEM_TRUNCATE
;
748 info
->swapped
-= nr_swaps_freed
;
749 if (nr_pages_to_free
)
750 shmem_free_blocks(inode
, nr_pages_to_free
);
751 shmem_recalc_inode(inode
);
752 spin_unlock(&info
->lock
);
755 * Empty swap vector directory pages to be freed?
757 if (!list_empty(&pages_to_free
)) {
758 pages_to_free
.prev
->next
= NULL
;
759 shmem_free_pages(pages_to_free
.next
);
763 static int shmem_notify_change(struct dentry
*dentry
, struct iattr
*attr
)
765 struct inode
*inode
= dentry
->d_inode
;
766 loff_t newsize
= attr
->ia_size
;
769 error
= inode_change_ok(inode
, attr
);
773 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)
774 && newsize
!= inode
->i_size
) {
775 struct page
*page
= NULL
;
777 if (newsize
< inode
->i_size
) {
779 * If truncating down to a partial page, then
780 * if that page is already allocated, hold it
781 * in memory until the truncation is over, so
782 * truncate_partial_page cannot miss it were
783 * it assigned to swap.
785 if (newsize
& (PAGE_CACHE_SIZE
-1)) {
786 (void) shmem_getpage(inode
,
787 newsize
>> PAGE_CACHE_SHIFT
,
788 &page
, SGP_READ
, NULL
);
793 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
794 * detect if any pages might have been added to cache
795 * after truncate_inode_pages. But we needn't bother
796 * if it's being fully truncated to zero-length: the
797 * nrpages check is efficient enough in that case.
800 struct shmem_inode_info
*info
= SHMEM_I(inode
);
801 spin_lock(&info
->lock
);
802 info
->flags
&= ~SHMEM_PAGEIN
;
803 spin_unlock(&info
->lock
);
807 /* XXX(truncate): truncate_setsize should be called last */
808 truncate_setsize(inode
, newsize
);
810 page_cache_release(page
);
811 shmem_truncate_range(inode
, newsize
, (loff_t
)-1);
814 setattr_copy(inode
, attr
);
815 #ifdef CONFIG_TMPFS_POSIX_ACL
816 if (attr
->ia_valid
& ATTR_MODE
)
817 error
= generic_acl_chmod(inode
);
822 static void shmem_evict_inode(struct inode
*inode
)
824 struct shmem_inode_info
*info
= SHMEM_I(inode
);
826 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
827 truncate_inode_pages(inode
->i_mapping
, 0);
828 shmem_unacct_size(info
->flags
, inode
->i_size
);
830 shmem_truncate_range(inode
, 0, (loff_t
)-1);
831 if (!list_empty(&info
->swaplist
)) {
832 mutex_lock(&shmem_swaplist_mutex
);
833 list_del_init(&info
->swaplist
);
834 mutex_unlock(&shmem_swaplist_mutex
);
837 BUG_ON(inode
->i_blocks
);
838 shmem_free_inode(inode
->i_sb
);
839 end_writeback(inode
);
842 static inline int shmem_find_swp(swp_entry_t entry
, swp_entry_t
*dir
, swp_entry_t
*edir
)
846 for (ptr
= dir
; ptr
< edir
; ptr
++) {
847 if (ptr
->val
== entry
.val
)
853 static int shmem_unuse_inode(struct shmem_inode_info
*info
, swp_entry_t entry
, struct page
*page
)
855 struct address_space
*mapping
;
867 ptr
= info
->i_direct
;
868 spin_lock(&info
->lock
);
869 if (!info
->swapped
) {
870 list_del_init(&info
->swaplist
);
873 limit
= info
->next_index
;
875 if (size
> SHMEM_NR_DIRECT
)
876 size
= SHMEM_NR_DIRECT
;
877 offset
= shmem_find_swp(entry
, ptr
, ptr
+size
);
879 shmem_swp_balance_unmap();
882 if (!info
->i_indirect
)
885 dir
= shmem_dir_map(info
->i_indirect
);
886 stage
= SHMEM_NR_DIRECT
+ ENTRIES_PER_PAGEPAGE
/2;
888 for (idx
= SHMEM_NR_DIRECT
; idx
< limit
; idx
+= ENTRIES_PER_PAGE
, dir
++) {
889 if (unlikely(idx
== stage
)) {
890 shmem_dir_unmap(dir
-1);
891 if (cond_resched_lock(&info
->lock
)) {
892 /* check it has not been truncated */
893 if (limit
> info
->next_index
) {
894 limit
= info
->next_index
;
899 dir
= shmem_dir_map(info
->i_indirect
) +
900 ENTRIES_PER_PAGE
/2 + idx
/ENTRIES_PER_PAGEPAGE
;
903 idx
+= ENTRIES_PER_PAGEPAGE
;
907 stage
= idx
+ ENTRIES_PER_PAGEPAGE
;
909 shmem_dir_unmap(dir
);
910 dir
= shmem_dir_map(subdir
);
913 if (subdir
&& page_private(subdir
)) {
914 ptr
= shmem_swp_map(subdir
);
916 if (size
> ENTRIES_PER_PAGE
)
917 size
= ENTRIES_PER_PAGE
;
918 offset
= shmem_find_swp(entry
, ptr
, ptr
+size
);
920 shmem_dir_unmap(dir
);
923 shmem_swp_unmap(ptr
);
927 shmem_dir_unmap(dir
-1);
929 spin_unlock(&info
->lock
);
936 * Move _head_ to start search for next from here.
937 * But be careful: shmem_evict_inode checks list_empty without taking
938 * mutex, and there's an instant in list_move_tail when info->swaplist
939 * would appear empty, if it were the only one on shmem_swaplist. We
940 * could avoid doing it if inode NULL; or use this minor optimization.
942 if (shmem_swaplist
.next
!= &info
->swaplist
)
943 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
946 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
947 * but also to hold up shmem_evict_inode(): so inode cannot be freed
948 * beneath us (pagelock doesn't help until the page is in pagecache).
950 mapping
= info
->vfs_inode
.i_mapping
;
951 error
= add_to_page_cache_locked(page
, mapping
, idx
, GFP_NOWAIT
);
952 /* which does mem_cgroup_uncharge_cache_page on error */
954 if (error
== -EEXIST
) {
955 struct page
*filepage
= find_get_page(mapping
, idx
);
959 * There might be a more uptodate page coming down
960 * from a stacked writepage: forget our swappage if so.
962 if (PageUptodate(filepage
))
964 page_cache_release(filepage
);
968 delete_from_swap_cache(page
);
969 set_page_dirty(page
);
970 info
->flags
|= SHMEM_PAGEIN
;
971 shmem_swp_set(info
, ptr
, 0);
973 error
= 1; /* not an error, but entry was found */
975 shmem_swp_unmap(ptr
);
976 spin_unlock(&info
->lock
);
981 * shmem_unuse() search for an eventually swapped out shmem page.
983 int shmem_unuse(swp_entry_t entry
, struct page
*page
)
985 struct list_head
*p
, *next
;
986 struct shmem_inode_info
*info
;
991 * Charge page using GFP_KERNEL while we can wait, before taking
992 * the shmem_swaplist_mutex which might hold up shmem_writepage().
993 * Charged back to the user (not to caller) when swap account is used.
994 * add_to_page_cache() will be called with GFP_NOWAIT.
996 error
= mem_cgroup_cache_charge(page
, current
->mm
, GFP_KERNEL
);
1000 * Try to preload while we can wait, to not make a habit of
1001 * draining atomic reserves; but don't latch on to this cpu,
1002 * it's okay if sometimes we get rescheduled after this.
1004 error
= radix_tree_preload(GFP_KERNEL
);
1007 radix_tree_preload_end();
1009 mutex_lock(&shmem_swaplist_mutex
);
1010 list_for_each_safe(p
, next
, &shmem_swaplist
) {
1011 info
= list_entry(p
, struct shmem_inode_info
, swaplist
);
1012 found
= shmem_unuse_inode(info
, entry
, page
);
1017 mutex_unlock(&shmem_swaplist_mutex
);
1021 mem_cgroup_uncharge_cache_page(page
);
1026 page_cache_release(page
);
1031 * Move the page from the page cache to the swap cache.
1033 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1035 struct shmem_inode_info
*info
;
1036 swp_entry_t
*entry
, swap
;
1037 struct address_space
*mapping
;
1038 unsigned long index
;
1039 struct inode
*inode
;
1040 bool unlock_mutex
= false;
1042 BUG_ON(!PageLocked(page
));
1043 mapping
= page
->mapping
;
1044 index
= page
->index
;
1045 inode
= mapping
->host
;
1046 info
= SHMEM_I(inode
);
1047 if (info
->flags
& VM_LOCKED
)
1049 if (!total_swap_pages
)
1053 * shmem_backing_dev_info's capabilities prevent regular writeback or
1054 * sync from ever calling shmem_writepage; but a stacking filesystem
1055 * may use the ->writepage of its underlying filesystem, in which case
1056 * tmpfs should write out to swap only in response to memory pressure,
1057 * and not for the writeback threads or sync. However, in those cases,
1058 * we do still want to check if there's a redundant swappage to be
1061 if (wbc
->for_reclaim
)
1062 swap
= get_swap_page();
1067 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1068 * if it's not already there. Do it now because we cannot take
1069 * mutex while holding spinlock, and must do so before the page
1070 * is moved to swap cache, when its pagelock no longer protects
1071 * the inode from eviction. But don't unlock the mutex until
1072 * we've taken the spinlock, because shmem_unuse_inode() will
1073 * prune a !swapped inode from the swaplist under both locks.
1075 if (swap
.val
&& list_empty(&info
->swaplist
)) {
1076 mutex_lock(&shmem_swaplist_mutex
);
1077 /* move instead of add in case we're racing */
1078 list_move_tail(&info
->swaplist
, &shmem_swaplist
);
1079 unlock_mutex
= true;
1082 spin_lock(&info
->lock
);
1084 mutex_unlock(&shmem_swaplist_mutex
);
1086 if (index
>= info
->next_index
) {
1087 BUG_ON(!(info
->flags
& SHMEM_TRUNCATE
));
1090 entry
= shmem_swp_entry(info
, index
, NULL
);
1093 * The more uptodate page coming down from a stacked
1094 * writepage should replace our old swappage.
1096 free_swap_and_cache(*entry
);
1097 shmem_swp_set(info
, entry
, 0);
1099 shmem_recalc_inode(inode
);
1101 if (swap
.val
&& add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1102 delete_from_page_cache(page
);
1103 shmem_swp_set(info
, entry
, swap
.val
);
1104 shmem_swp_unmap(entry
);
1105 spin_unlock(&info
->lock
);
1106 swap_shmem_alloc(swap
);
1107 BUG_ON(page_mapped(page
));
1108 swap_writepage(page
, wbc
);
1112 shmem_swp_unmap(entry
);
1114 spin_unlock(&info
->lock
);
1116 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1117 * clear SWAP_HAS_CACHE flag.
1119 swapcache_free(swap
, NULL
);
1121 set_page_dirty(page
);
1122 if (wbc
->for_reclaim
)
1123 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1130 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1134 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1135 return; /* show nothing */
1137 mpol_to_str(buffer
, sizeof(buffer
), mpol
, 1);
1139 seq_printf(seq
, ",mpol=%s", buffer
);
1142 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1144 struct mempolicy
*mpol
= NULL
;
1146 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1147 mpol
= sbinfo
->mpol
;
1149 spin_unlock(&sbinfo
->stat_lock
);
1153 #endif /* CONFIG_TMPFS */
1155 static struct page
*shmem_swapin(swp_entry_t entry
, gfp_t gfp
,
1156 struct shmem_inode_info
*info
, unsigned long idx
)
1158 struct mempolicy mpol
, *spol
;
1159 struct vm_area_struct pvma
;
1162 spol
= mpol_cond_copy(&mpol
,
1163 mpol_shared_policy_lookup(&info
->policy
, idx
));
1165 /* Create a pseudo vma that just contains the policy */
1167 pvma
.vm_pgoff
= idx
;
1169 pvma
.vm_policy
= spol
;
1170 page
= swapin_readahead(entry
, gfp
, &pvma
, 0);
1174 static struct page
*shmem_alloc_page(gfp_t gfp
,
1175 struct shmem_inode_info
*info
, unsigned long idx
)
1177 struct vm_area_struct pvma
;
1179 /* Create a pseudo vma that just contains the policy */
1181 pvma
.vm_pgoff
= idx
;
1183 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, idx
);
1186 * alloc_page_vma() will drop the shared policy reference
1188 return alloc_page_vma(gfp
, &pvma
, 0);
1190 #else /* !CONFIG_NUMA */
1192 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*p
)
1195 #endif /* CONFIG_TMPFS */
1197 static inline struct page
*shmem_swapin(swp_entry_t entry
, gfp_t gfp
,
1198 struct shmem_inode_info
*info
, unsigned long idx
)
1200 return swapin_readahead(entry
, gfp
, NULL
, 0);
1203 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
1204 struct shmem_inode_info
*info
, unsigned long idx
)
1206 return alloc_page(gfp
);
1208 #endif /* CONFIG_NUMA */
1210 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1211 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1218 * shmem_getpage - either get the page from swap or allocate a new one
1220 * If we allocate a new one we do not mark it dirty. That's up to the
1221 * vm. If we swap it in we mark it dirty since we also free the swap
1222 * entry since a page cannot live in both the swap and page cache
1224 static int shmem_getpage(struct inode
*inode
, unsigned long idx
,
1225 struct page
**pagep
, enum sgp_type sgp
, int *type
)
1227 struct address_space
*mapping
= inode
->i_mapping
;
1228 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1229 struct shmem_sb_info
*sbinfo
;
1230 struct page
*filepage
= *pagep
;
1231 struct page
*swappage
;
1232 struct page
*prealloc_page
= NULL
;
1238 if (idx
>= SHMEM_MAX_INDEX
)
1245 * Normally, filepage is NULL on entry, and either found
1246 * uptodate immediately, or allocated and zeroed, or read
1247 * in under swappage, which is then assigned to filepage.
1248 * But shmem_readpage (required for splice) passes in a locked
1249 * filepage, which may be found not uptodate by other callers
1250 * too, and may need to be copied from the swappage read in.
1254 filepage
= find_lock_page(mapping
, idx
);
1255 if (filepage
&& PageUptodate(filepage
))
1257 gfp
= mapping_gfp_mask(mapping
);
1260 * Try to preload while we can wait, to not make a habit of
1261 * draining atomic reserves; but don't latch on to this cpu.
1263 error
= radix_tree_preload(gfp
& ~__GFP_HIGHMEM
);
1266 radix_tree_preload_end();
1267 if (sgp
!= SGP_READ
&& !prealloc_page
) {
1268 /* We don't care if this fails */
1269 prealloc_page
= shmem_alloc_page(gfp
, info
, idx
);
1270 if (prealloc_page
) {
1271 if (mem_cgroup_cache_charge(prealloc_page
,
1272 current
->mm
, GFP_KERNEL
)) {
1273 page_cache_release(prealloc_page
);
1274 prealloc_page
= NULL
;
1281 spin_lock(&info
->lock
);
1282 shmem_recalc_inode(inode
);
1283 entry
= shmem_swp_alloc(info
, idx
, sgp
);
1284 if (IS_ERR(entry
)) {
1285 spin_unlock(&info
->lock
);
1286 error
= PTR_ERR(entry
);
1292 /* Look it up and read it in.. */
1293 swappage
= lookup_swap_cache(swap
);
1295 shmem_swp_unmap(entry
);
1296 /* here we actually do the io */
1297 if (type
&& !(*type
& VM_FAULT_MAJOR
)) {
1298 __count_vm_event(PGMAJFAULT
);
1299 *type
|= VM_FAULT_MAJOR
;
1301 spin_unlock(&info
->lock
);
1302 swappage
= shmem_swapin(swap
, gfp
, info
, idx
);
1304 spin_lock(&info
->lock
);
1305 entry
= shmem_swp_alloc(info
, idx
, sgp
);
1307 error
= PTR_ERR(entry
);
1309 if (entry
->val
== swap
.val
)
1311 shmem_swp_unmap(entry
);
1313 spin_unlock(&info
->lock
);
1318 wait_on_page_locked(swappage
);
1319 page_cache_release(swappage
);
1323 /* We have to do this with page locked to prevent races */
1324 if (!trylock_page(swappage
)) {
1325 shmem_swp_unmap(entry
);
1326 spin_unlock(&info
->lock
);
1327 wait_on_page_locked(swappage
);
1328 page_cache_release(swappage
);
1331 if (PageWriteback(swappage
)) {
1332 shmem_swp_unmap(entry
);
1333 spin_unlock(&info
->lock
);
1334 wait_on_page_writeback(swappage
);
1335 unlock_page(swappage
);
1336 page_cache_release(swappage
);
1339 if (!PageUptodate(swappage
)) {
1340 shmem_swp_unmap(entry
);
1341 spin_unlock(&info
->lock
);
1342 unlock_page(swappage
);
1343 page_cache_release(swappage
);
1349 shmem_swp_set(info
, entry
, 0);
1350 shmem_swp_unmap(entry
);
1351 delete_from_swap_cache(swappage
);
1352 spin_unlock(&info
->lock
);
1353 copy_highpage(filepage
, swappage
);
1354 unlock_page(swappage
);
1355 page_cache_release(swappage
);
1356 flush_dcache_page(filepage
);
1357 SetPageUptodate(filepage
);
1358 set_page_dirty(filepage
);
1360 } else if (!(error
= add_to_page_cache_locked(swappage
, mapping
,
1361 idx
, GFP_NOWAIT
))) {
1362 info
->flags
|= SHMEM_PAGEIN
;
1363 shmem_swp_set(info
, entry
, 0);
1364 shmem_swp_unmap(entry
);
1365 delete_from_swap_cache(swappage
);
1366 spin_unlock(&info
->lock
);
1367 filepage
= swappage
;
1368 set_page_dirty(filepage
);
1371 shmem_swp_unmap(entry
);
1372 spin_unlock(&info
->lock
);
1373 if (error
== -ENOMEM
) {
1375 * reclaim from proper memory cgroup and
1376 * call memcg's OOM if needed.
1378 error
= mem_cgroup_shmem_charge_fallback(
1383 unlock_page(swappage
);
1384 page_cache_release(swappage
);
1388 unlock_page(swappage
);
1389 page_cache_release(swappage
);
1392 } else if (sgp
== SGP_READ
&& !filepage
) {
1393 shmem_swp_unmap(entry
);
1394 filepage
= find_get_page(mapping
, idx
);
1396 (!PageUptodate(filepage
) || !trylock_page(filepage
))) {
1397 spin_unlock(&info
->lock
);
1398 wait_on_page_locked(filepage
);
1399 page_cache_release(filepage
);
1403 spin_unlock(&info
->lock
);
1405 shmem_swp_unmap(entry
);
1406 sbinfo
= SHMEM_SB(inode
->i_sb
);
1407 if (sbinfo
->max_blocks
) {
1408 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1409 sbinfo
->max_blocks
) >= 0 ||
1410 shmem_acct_block(info
->flags
))
1412 percpu_counter_inc(&sbinfo
->used_blocks
);
1413 spin_lock(&inode
->i_lock
);
1414 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1415 spin_unlock(&inode
->i_lock
);
1416 } else if (shmem_acct_block(info
->flags
))
1422 if (!prealloc_page
) {
1423 spin_unlock(&info
->lock
);
1424 filepage
= shmem_alloc_page(gfp
, info
, idx
);
1426 shmem_unacct_blocks(info
->flags
, 1);
1427 shmem_free_blocks(inode
, 1);
1431 SetPageSwapBacked(filepage
);
1434 * Precharge page while we can wait, compensate
1437 error
= mem_cgroup_cache_charge(filepage
,
1438 current
->mm
, GFP_KERNEL
);
1440 page_cache_release(filepage
);
1441 shmem_unacct_blocks(info
->flags
, 1);
1442 shmem_free_blocks(inode
, 1);
1447 spin_lock(&info
->lock
);
1449 filepage
= prealloc_page
;
1450 prealloc_page
= NULL
;
1451 SetPageSwapBacked(filepage
);
1454 entry
= shmem_swp_alloc(info
, idx
, sgp
);
1456 error
= PTR_ERR(entry
);
1459 shmem_swp_unmap(entry
);
1461 ret
= error
|| swap
.val
;
1463 mem_cgroup_uncharge_cache_page(filepage
);
1465 ret
= add_to_page_cache_lru(filepage
, mapping
,
1468 * At add_to_page_cache_lru() failure, uncharge will
1469 * be done automatically.
1472 spin_unlock(&info
->lock
);
1473 page_cache_release(filepage
);
1474 shmem_unacct_blocks(info
->flags
, 1);
1475 shmem_free_blocks(inode
, 1);
1481 info
->flags
|= SHMEM_PAGEIN
;
1485 spin_unlock(&info
->lock
);
1486 clear_highpage(filepage
);
1487 flush_dcache_page(filepage
);
1488 SetPageUptodate(filepage
);
1489 if (sgp
== SGP_DIRTY
)
1490 set_page_dirty(filepage
);
1499 * Perhaps the page was brought in from swap between find_lock_page
1500 * and taking info->lock? We allow for that at add_to_page_cache_lru,
1501 * but must also avoid reporting a spurious ENOSPC while working on a
1502 * full tmpfs. (When filepage has been passed in to shmem_getpage, it
1503 * is already in page cache, which prevents this race from occurring.)
1506 struct page
*page
= find_get_page(mapping
, idx
);
1508 spin_unlock(&info
->lock
);
1509 page_cache_release(page
);
1513 spin_unlock(&info
->lock
);
1516 if (*pagep
!= filepage
) {
1517 unlock_page(filepage
);
1518 page_cache_release(filepage
);
1521 if (prealloc_page
) {
1522 mem_cgroup_uncharge_cache_page(prealloc_page
);
1523 page_cache_release(prealloc_page
);
1528 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1530 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1534 if (((loff_t
)vmf
->pgoff
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
))
1535 return VM_FAULT_SIGBUS
;
1537 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1539 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1541 return ret
| VM_FAULT_LOCKED
;
1545 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
1547 struct inode
*i
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1548 return mpol_set_shared_policy(&SHMEM_I(i
)->policy
, vma
, new);
1551 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1554 struct inode
*i
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1557 idx
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1558 return mpol_shared_policy_lookup(&SHMEM_I(i
)->policy
, idx
);
1562 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1564 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1565 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1566 int retval
= -ENOMEM
;
1568 spin_lock(&info
->lock
);
1569 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1570 if (!user_shm_lock(inode
->i_size
, user
))
1572 info
->flags
|= VM_LOCKED
;
1573 mapping_set_unevictable(file
->f_mapping
);
1575 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1576 user_shm_unlock(inode
->i_size
, user
);
1577 info
->flags
&= ~VM_LOCKED
;
1578 mapping_clear_unevictable(file
->f_mapping
);
1579 scan_mapping_unevictable_pages(file
->f_mapping
);
1584 spin_unlock(&info
->lock
);
1588 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1590 file_accessed(file
);
1591 vma
->vm_ops
= &shmem_vm_ops
;
1592 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
1596 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1597 int mode
, dev_t dev
, unsigned long flags
)
1599 struct inode
*inode
;
1600 struct shmem_inode_info
*info
;
1601 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1603 if (shmem_reserve_inode(sb
))
1606 inode
= new_inode(sb
);
1608 inode
->i_ino
= get_next_ino();
1609 inode_init_owner(inode
, dir
, mode
);
1610 inode
->i_blocks
= 0;
1611 inode
->i_mapping
->backing_dev_info
= &shmem_backing_dev_info
;
1612 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1613 inode
->i_generation
= get_seconds();
1614 info
= SHMEM_I(inode
);
1615 memset(info
, 0, (char *)inode
- (char *)info
);
1616 spin_lock_init(&info
->lock
);
1617 info
->flags
= flags
& VM_NORESERVE
;
1618 INIT_LIST_HEAD(&info
->swaplist
);
1619 cache_no_acl(inode
);
1621 switch (mode
& S_IFMT
) {
1623 inode
->i_op
= &shmem_special_inode_operations
;
1624 init_special_inode(inode
, mode
, dev
);
1627 inode
->i_mapping
->a_ops
= &shmem_aops
;
1628 inode
->i_op
= &shmem_inode_operations
;
1629 inode
->i_fop
= &shmem_file_operations
;
1630 mpol_shared_policy_init(&info
->policy
,
1631 shmem_get_sbmpol(sbinfo
));
1635 /* Some things misbehave if size == 0 on a directory */
1636 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1637 inode
->i_op
= &shmem_dir_inode_operations
;
1638 inode
->i_fop
= &simple_dir_operations
;
1642 * Must not load anything in the rbtree,
1643 * mpol_free_shared_policy will not be called.
1645 mpol_shared_policy_init(&info
->policy
, NULL
);
1649 shmem_free_inode(sb
);
1654 static const struct inode_operations shmem_symlink_inode_operations
;
1655 static const struct inode_operations shmem_symlink_inline_operations
;
1658 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1659 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1660 * below the loop driver, in the generic fashion that many filesystems support.
1662 static int shmem_readpage(struct file
*file
, struct page
*page
)
1664 struct inode
*inode
= page
->mapping
->host
;
1665 int error
= shmem_getpage(inode
, page
->index
, &page
, SGP_CACHE
, NULL
);
1671 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1672 loff_t pos
, unsigned len
, unsigned flags
,
1673 struct page
**pagep
, void **fsdata
)
1675 struct inode
*inode
= mapping
->host
;
1676 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1678 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1682 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1683 loff_t pos
, unsigned len
, unsigned copied
,
1684 struct page
*page
, void *fsdata
)
1686 struct inode
*inode
= mapping
->host
;
1688 if (pos
+ copied
> inode
->i_size
)
1689 i_size_write(inode
, pos
+ copied
);
1691 set_page_dirty(page
);
1693 page_cache_release(page
);
1698 static void do_shmem_file_read(struct file
*filp
, loff_t
*ppos
, read_descriptor_t
*desc
, read_actor_t actor
)
1700 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
1701 struct address_space
*mapping
= inode
->i_mapping
;
1702 unsigned long index
, offset
;
1703 enum sgp_type sgp
= SGP_READ
;
1706 * Might this read be for a stacking filesystem? Then when reading
1707 * holes of a sparse file, we actually need to allocate those pages,
1708 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1710 if (segment_eq(get_fs(), KERNEL_DS
))
1713 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1714 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1717 struct page
*page
= NULL
;
1718 unsigned long end_index
, nr
, ret
;
1719 loff_t i_size
= i_size_read(inode
);
1721 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1722 if (index
> end_index
)
1724 if (index
== end_index
) {
1725 nr
= i_size
& ~PAGE_CACHE_MASK
;
1730 desc
->error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1732 if (desc
->error
== -EINVAL
)
1740 * We must evaluate after, since reads (unlike writes)
1741 * are called without i_mutex protection against truncate
1743 nr
= PAGE_CACHE_SIZE
;
1744 i_size
= i_size_read(inode
);
1745 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1746 if (index
== end_index
) {
1747 nr
= i_size
& ~PAGE_CACHE_MASK
;
1750 page_cache_release(page
);
1758 * If users can be writing to this page using arbitrary
1759 * virtual addresses, take care about potential aliasing
1760 * before reading the page on the kernel side.
1762 if (mapping_writably_mapped(mapping
))
1763 flush_dcache_page(page
);
1765 * Mark the page accessed if we read the beginning.
1768 mark_page_accessed(page
);
1770 page
= ZERO_PAGE(0);
1771 page_cache_get(page
);
1775 * Ok, we have the page, and it's up-to-date, so
1776 * now we can copy it to user space...
1778 * The actor routine returns how many bytes were actually used..
1779 * NOTE! This may not be the same as how much of a user buffer
1780 * we filled up (we may be padding etc), so we can only update
1781 * "pos" here (the actor routine has to update the user buffer
1782 * pointers and the remaining count).
1784 ret
= actor(desc
, page
, offset
, nr
);
1786 index
+= offset
>> PAGE_CACHE_SHIFT
;
1787 offset
&= ~PAGE_CACHE_MASK
;
1789 page_cache_release(page
);
1790 if (ret
!= nr
|| !desc
->count
)
1796 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1797 file_accessed(filp
);
1800 static ssize_t
shmem_file_aio_read(struct kiocb
*iocb
,
1801 const struct iovec
*iov
, unsigned long nr_segs
, loff_t pos
)
1803 struct file
*filp
= iocb
->ki_filp
;
1807 loff_t
*ppos
= &iocb
->ki_pos
;
1809 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1813 for (seg
= 0; seg
< nr_segs
; seg
++) {
1814 read_descriptor_t desc
;
1817 desc
.arg
.buf
= iov
[seg
].iov_base
;
1818 desc
.count
= iov
[seg
].iov_len
;
1819 if (desc
.count
== 0)
1822 do_shmem_file_read(filp
, ppos
, &desc
, file_read_actor
);
1823 retval
+= desc
.written
;
1825 retval
= retval
?: desc
.error
;
1834 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
1836 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
1838 buf
->f_type
= TMPFS_MAGIC
;
1839 buf
->f_bsize
= PAGE_CACHE_SIZE
;
1840 buf
->f_namelen
= NAME_MAX
;
1841 if (sbinfo
->max_blocks
) {
1842 buf
->f_blocks
= sbinfo
->max_blocks
;
1843 buf
->f_bavail
= buf
->f_bfree
=
1844 sbinfo
->max_blocks
- percpu_counter_sum(&sbinfo
->used_blocks
);
1846 if (sbinfo
->max_inodes
) {
1847 buf
->f_files
= sbinfo
->max_inodes
;
1848 buf
->f_ffree
= sbinfo
->free_inodes
;
1850 /* else leave those fields 0 like simple_statfs */
1855 * File creation. Allocate an inode, and we're done..
1858 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1860 struct inode
*inode
;
1861 int error
= -ENOSPC
;
1863 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
1865 error
= security_inode_init_security(inode
, dir
,
1866 &dentry
->d_name
, NULL
,
1869 if (error
!= -EOPNOTSUPP
) {
1874 #ifdef CONFIG_TMPFS_POSIX_ACL
1875 error
= generic_acl_init(inode
, dir
);
1883 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1884 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1885 d_instantiate(dentry
, inode
);
1886 dget(dentry
); /* Extra count - pin the dentry in core */
1891 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1895 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
1901 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1902 struct nameidata
*nd
)
1904 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
1910 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1912 struct inode
*inode
= old_dentry
->d_inode
;
1916 * No ordinary (disk based) filesystem counts links as inodes;
1917 * but each new link needs a new dentry, pinning lowmem, and
1918 * tmpfs dentries cannot be pruned until they are unlinked.
1920 ret
= shmem_reserve_inode(inode
->i_sb
);
1924 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1925 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1927 ihold(inode
); /* New dentry reference */
1928 dget(dentry
); /* Extra pinning count for the created dentry */
1929 d_instantiate(dentry
, inode
);
1934 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
1936 struct inode
*inode
= dentry
->d_inode
;
1938 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
1939 shmem_free_inode(inode
->i_sb
);
1941 dir
->i_size
-= BOGO_DIRENT_SIZE
;
1942 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1944 dput(dentry
); /* Undo the count from "create" - this does all the work */
1948 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1950 if (!simple_empty(dentry
))
1953 drop_nlink(dentry
->d_inode
);
1955 return shmem_unlink(dir
, dentry
);
1959 * The VFS layer already does all the dentry stuff for rename,
1960 * we just have to decrement the usage count for the target if
1961 * it exists so that the VFS layer correctly free's it when it
1964 static int shmem_rename(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
1966 struct inode
*inode
= old_dentry
->d_inode
;
1967 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
1969 if (!simple_empty(new_dentry
))
1972 if (new_dentry
->d_inode
) {
1973 (void) shmem_unlink(new_dir
, new_dentry
);
1975 drop_nlink(old_dir
);
1976 } else if (they_are_dirs
) {
1977 drop_nlink(old_dir
);
1981 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
1982 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
1983 old_dir
->i_ctime
= old_dir
->i_mtime
=
1984 new_dir
->i_ctime
= new_dir
->i_mtime
=
1985 inode
->i_ctime
= CURRENT_TIME
;
1989 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1993 struct inode
*inode
;
1994 struct page
*page
= NULL
;
1996 struct shmem_inode_info
*info
;
1998 len
= strlen(symname
) + 1;
1999 if (len
> PAGE_CACHE_SIZE
)
2000 return -ENAMETOOLONG
;
2002 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2006 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
, NULL
,
2009 if (error
!= -EOPNOTSUPP
) {
2016 info
= SHMEM_I(inode
);
2017 inode
->i_size
= len
-1;
2018 if (len
<= (char *)inode
- (char *)info
) {
2020 memcpy(info
, symname
, len
);
2021 inode
->i_op
= &shmem_symlink_inline_operations
;
2023 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
2028 inode
->i_mapping
->a_ops
= &shmem_aops
;
2029 inode
->i_op
= &shmem_symlink_inode_operations
;
2030 kaddr
= kmap_atomic(page
, KM_USER0
);
2031 memcpy(kaddr
, symname
, len
);
2032 kunmap_atomic(kaddr
, KM_USER0
);
2033 set_page_dirty(page
);
2035 page_cache_release(page
);
2037 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2038 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2039 d_instantiate(dentry
, inode
);
2044 static void *shmem_follow_link_inline(struct dentry
*dentry
, struct nameidata
*nd
)
2046 nd_set_link(nd
, (char *)SHMEM_I(dentry
->d_inode
));
2050 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
2052 struct page
*page
= NULL
;
2053 int res
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
2054 nd_set_link(nd
, res
? ERR_PTR(res
) : kmap(page
));
2060 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2062 if (!IS_ERR(nd_get_link(nd
))) {
2063 struct page
*page
= cookie
;
2065 mark_page_accessed(page
);
2066 page_cache_release(page
);
2070 static const struct inode_operations shmem_symlink_inline_operations
= {
2071 .readlink
= generic_readlink
,
2072 .follow_link
= shmem_follow_link_inline
,
2075 static const struct inode_operations shmem_symlink_inode_operations
= {
2076 .readlink
= generic_readlink
,
2077 .follow_link
= shmem_follow_link
,
2078 .put_link
= shmem_put_link
,
2081 #ifdef CONFIG_TMPFS_POSIX_ACL
2083 * Superblocks without xattr inode operations will get security.* xattr
2084 * support from the VFS "for free". As soon as we have any other xattrs
2085 * like ACLs, we also need to implement the security.* handlers at
2086 * filesystem level, though.
2089 static size_t shmem_xattr_security_list(struct dentry
*dentry
, char *list
,
2090 size_t list_len
, const char *name
,
2091 size_t name_len
, int handler_flags
)
2093 return security_inode_listsecurity(dentry
->d_inode
, list
, list_len
);
2096 static int shmem_xattr_security_get(struct dentry
*dentry
, const char *name
,
2097 void *buffer
, size_t size
, int handler_flags
)
2099 if (strcmp(name
, "") == 0)
2101 return xattr_getsecurity(dentry
->d_inode
, name
, buffer
, size
);
2104 static int shmem_xattr_security_set(struct dentry
*dentry
, const char *name
,
2105 const void *value
, size_t size
, int flags
, int handler_flags
)
2107 if (strcmp(name
, "") == 0)
2109 return security_inode_setsecurity(dentry
->d_inode
, name
, value
,
2113 static const struct xattr_handler shmem_xattr_security_handler
= {
2114 .prefix
= XATTR_SECURITY_PREFIX
,
2115 .list
= shmem_xattr_security_list
,
2116 .get
= shmem_xattr_security_get
,
2117 .set
= shmem_xattr_security_set
,
2120 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2121 &generic_acl_access_handler
,
2122 &generic_acl_default_handler
,
2123 &shmem_xattr_security_handler
,
2128 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2130 return ERR_PTR(-ESTALE
);
2133 static int shmem_match(struct inode
*ino
, void *vfh
)
2137 inum
= (inum
<< 32) | fh
[1];
2138 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2141 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2142 struct fid
*fid
, int fh_len
, int fh_type
)
2144 struct inode
*inode
;
2145 struct dentry
*dentry
= NULL
;
2146 u64 inum
= fid
->raw
[2];
2147 inum
= (inum
<< 32) | fid
->raw
[1];
2152 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2153 shmem_match
, fid
->raw
);
2155 dentry
= d_find_alias(inode
);
2162 static int shmem_encode_fh(struct dentry
*dentry
, __u32
*fh
, int *len
,
2165 struct inode
*inode
= dentry
->d_inode
;
2172 if (inode_unhashed(inode
)) {
2173 /* Unfortunately insert_inode_hash is not idempotent,
2174 * so as we hash inodes here rather than at creation
2175 * time, we need a lock to ensure we only try
2178 static DEFINE_SPINLOCK(lock
);
2180 if (inode_unhashed(inode
))
2181 __insert_inode_hash(inode
,
2182 inode
->i_ino
+ inode
->i_generation
);
2186 fh
[0] = inode
->i_generation
;
2187 fh
[1] = inode
->i_ino
;
2188 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2194 static const struct export_operations shmem_export_ops
= {
2195 .get_parent
= shmem_get_parent
,
2196 .encode_fh
= shmem_encode_fh
,
2197 .fh_to_dentry
= shmem_fh_to_dentry
,
2200 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2203 char *this_char
, *value
, *rest
;
2205 while (options
!= NULL
) {
2206 this_char
= options
;
2209 * NUL-terminate this option: unfortunately,
2210 * mount options form a comma-separated list,
2211 * but mpol's nodelist may also contain commas.
2213 options
= strchr(options
, ',');
2214 if (options
== NULL
)
2217 if (!isdigit(*options
)) {
2224 if ((value
= strchr(this_char
,'=')) != NULL
) {
2228 "tmpfs: No value for mount option '%s'\n",
2233 if (!strcmp(this_char
,"size")) {
2234 unsigned long long size
;
2235 size
= memparse(value
,&rest
);
2237 size
<<= PAGE_SHIFT
;
2238 size
*= totalram_pages
;
2244 sbinfo
->max_blocks
=
2245 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2246 } else if (!strcmp(this_char
,"nr_blocks")) {
2247 sbinfo
->max_blocks
= memparse(value
, &rest
);
2250 } else if (!strcmp(this_char
,"nr_inodes")) {
2251 sbinfo
->max_inodes
= memparse(value
, &rest
);
2254 } else if (!strcmp(this_char
,"mode")) {
2257 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2260 } else if (!strcmp(this_char
,"uid")) {
2263 sbinfo
->uid
= simple_strtoul(value
, &rest
, 0);
2266 } else if (!strcmp(this_char
,"gid")) {
2269 sbinfo
->gid
= simple_strtoul(value
, &rest
, 0);
2272 } else if (!strcmp(this_char
,"mpol")) {
2273 if (mpol_parse_str(value
, &sbinfo
->mpol
, 1))
2276 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2284 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2290 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2292 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2293 struct shmem_sb_info config
= *sbinfo
;
2294 unsigned long inodes
;
2295 int error
= -EINVAL
;
2297 if (shmem_parse_options(data
, &config
, true))
2300 spin_lock(&sbinfo
->stat_lock
);
2301 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2302 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2304 if (config
.max_inodes
< inodes
)
2307 * Those tests also disallow limited->unlimited while any are in
2308 * use, so i_blocks will always be zero when max_blocks is zero;
2309 * but we must separately disallow unlimited->limited, because
2310 * in that case we have no record of how much is already in use.
2312 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2314 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2318 sbinfo
->max_blocks
= config
.max_blocks
;
2319 sbinfo
->max_inodes
= config
.max_inodes
;
2320 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2322 mpol_put(sbinfo
->mpol
);
2323 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2325 spin_unlock(&sbinfo
->stat_lock
);
2329 static int shmem_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
2331 struct shmem_sb_info
*sbinfo
= SHMEM_SB(vfs
->mnt_sb
);
2333 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2334 seq_printf(seq
, ",size=%luk",
2335 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2336 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2337 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2338 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2339 seq_printf(seq
, ",mode=%03o", sbinfo
->mode
);
2340 if (sbinfo
->uid
!= 0)
2341 seq_printf(seq
, ",uid=%u", sbinfo
->uid
);
2342 if (sbinfo
->gid
!= 0)
2343 seq_printf(seq
, ",gid=%u", sbinfo
->gid
);
2344 shmem_show_mpol(seq
, sbinfo
->mpol
);
2347 #endif /* CONFIG_TMPFS */
2349 static void shmem_put_super(struct super_block
*sb
)
2351 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2353 percpu_counter_destroy(&sbinfo
->used_blocks
);
2355 sb
->s_fs_info
= NULL
;
2358 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
2360 struct inode
*inode
;
2361 struct dentry
*root
;
2362 struct shmem_sb_info
*sbinfo
;
2365 /* Round up to L1_CACHE_BYTES to resist false sharing */
2366 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
2367 L1_CACHE_BYTES
), GFP_KERNEL
);
2371 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
2372 sbinfo
->uid
= current_fsuid();
2373 sbinfo
->gid
= current_fsgid();
2374 sb
->s_fs_info
= sbinfo
;
2378 * Per default we only allow half of the physical ram per
2379 * tmpfs instance, limiting inodes to one per page of lowmem;
2380 * but the internal instance is left unlimited.
2382 if (!(sb
->s_flags
& MS_NOUSER
)) {
2383 sbinfo
->max_blocks
= shmem_default_max_blocks();
2384 sbinfo
->max_inodes
= shmem_default_max_inodes();
2385 if (shmem_parse_options(data
, sbinfo
, false)) {
2390 sb
->s_export_op
= &shmem_export_ops
;
2392 sb
->s_flags
|= MS_NOUSER
;
2395 spin_lock_init(&sbinfo
->stat_lock
);
2396 if (percpu_counter_init(&sbinfo
->used_blocks
, 0))
2398 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
2400 sb
->s_maxbytes
= SHMEM_MAX_BYTES
;
2401 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
2402 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
2403 sb
->s_magic
= TMPFS_MAGIC
;
2404 sb
->s_op
= &shmem_ops
;
2405 sb
->s_time_gran
= 1;
2406 #ifdef CONFIG_TMPFS_POSIX_ACL
2407 sb
->s_xattr
= shmem_xattr_handlers
;
2408 sb
->s_flags
|= MS_POSIXACL
;
2411 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
2414 inode
->i_uid
= sbinfo
->uid
;
2415 inode
->i_gid
= sbinfo
->gid
;
2416 root
= d_alloc_root(inode
);
2425 shmem_put_super(sb
);
2429 static struct kmem_cache
*shmem_inode_cachep
;
2431 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
2433 struct shmem_inode_info
*p
;
2434 p
= (struct shmem_inode_info
*)kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
2437 return &p
->vfs_inode
;
2440 static void shmem_i_callback(struct rcu_head
*head
)
2442 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
2443 INIT_LIST_HEAD(&inode
->i_dentry
);
2444 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
2447 static void shmem_destroy_inode(struct inode
*inode
)
2449 if ((inode
->i_mode
& S_IFMT
) == S_IFREG
) {
2450 /* only struct inode is valid if it's an inline symlink */
2451 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
2453 call_rcu(&inode
->i_rcu
, shmem_i_callback
);
2456 static void init_once(void *foo
)
2458 struct shmem_inode_info
*p
= (struct shmem_inode_info
*) foo
;
2460 inode_init_once(&p
->vfs_inode
);
2463 static int init_inodecache(void)
2465 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
2466 sizeof(struct shmem_inode_info
),
2467 0, SLAB_PANIC
, init_once
);
2471 static void destroy_inodecache(void)
2473 kmem_cache_destroy(shmem_inode_cachep
);
2476 static const struct address_space_operations shmem_aops
= {
2477 .writepage
= shmem_writepage
,
2478 .set_page_dirty
= __set_page_dirty_no_writeback
,
2480 .readpage
= shmem_readpage
,
2481 .write_begin
= shmem_write_begin
,
2482 .write_end
= shmem_write_end
,
2484 .migratepage
= migrate_page
,
2485 .error_remove_page
= generic_error_remove_page
,
2488 static const struct file_operations shmem_file_operations
= {
2491 .llseek
= generic_file_llseek
,
2492 .read
= do_sync_read
,
2493 .write
= do_sync_write
,
2494 .aio_read
= shmem_file_aio_read
,
2495 .aio_write
= generic_file_aio_write
,
2496 .fsync
= noop_fsync
,
2497 .splice_read
= generic_file_splice_read
,
2498 .splice_write
= generic_file_splice_write
,
2502 static const struct inode_operations shmem_inode_operations
= {
2503 .setattr
= shmem_notify_change
,
2504 .truncate_range
= shmem_truncate_range
,
2505 #ifdef CONFIG_TMPFS_POSIX_ACL
2506 .setxattr
= generic_setxattr
,
2507 .getxattr
= generic_getxattr
,
2508 .listxattr
= generic_listxattr
,
2509 .removexattr
= generic_removexattr
,
2510 .check_acl
= generic_check_acl
,
2515 static const struct inode_operations shmem_dir_inode_operations
= {
2517 .create
= shmem_create
,
2518 .lookup
= simple_lookup
,
2520 .unlink
= shmem_unlink
,
2521 .symlink
= shmem_symlink
,
2522 .mkdir
= shmem_mkdir
,
2523 .rmdir
= shmem_rmdir
,
2524 .mknod
= shmem_mknod
,
2525 .rename
= shmem_rename
,
2527 #ifdef CONFIG_TMPFS_POSIX_ACL
2528 .setattr
= shmem_notify_change
,
2529 .setxattr
= generic_setxattr
,
2530 .getxattr
= generic_getxattr
,
2531 .listxattr
= generic_listxattr
,
2532 .removexattr
= generic_removexattr
,
2533 .check_acl
= generic_check_acl
,
2537 static const struct inode_operations shmem_special_inode_operations
= {
2538 #ifdef CONFIG_TMPFS_POSIX_ACL
2539 .setattr
= shmem_notify_change
,
2540 .setxattr
= generic_setxattr
,
2541 .getxattr
= generic_getxattr
,
2542 .listxattr
= generic_listxattr
,
2543 .removexattr
= generic_removexattr
,
2544 .check_acl
= generic_check_acl
,
2548 static const struct super_operations shmem_ops
= {
2549 .alloc_inode
= shmem_alloc_inode
,
2550 .destroy_inode
= shmem_destroy_inode
,
2552 .statfs
= shmem_statfs
,
2553 .remount_fs
= shmem_remount_fs
,
2554 .show_options
= shmem_show_options
,
2556 .evict_inode
= shmem_evict_inode
,
2557 .drop_inode
= generic_delete_inode
,
2558 .put_super
= shmem_put_super
,
2561 static const struct vm_operations_struct shmem_vm_ops
= {
2562 .fault
= shmem_fault
,
2564 .set_policy
= shmem_set_policy
,
2565 .get_policy
= shmem_get_policy
,
2570 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
2571 int flags
, const char *dev_name
, void *data
)
2573 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
2576 static struct file_system_type tmpfs_fs_type
= {
2577 .owner
= THIS_MODULE
,
2579 .mount
= shmem_mount
,
2580 .kill_sb
= kill_litter_super
,
2583 int __init
init_tmpfs(void)
2587 error
= bdi_init(&shmem_backing_dev_info
);
2591 error
= init_inodecache();
2595 error
= register_filesystem(&tmpfs_fs_type
);
2597 printk(KERN_ERR
"Could not register tmpfs\n");
2601 shm_mnt
= vfs_kern_mount(&tmpfs_fs_type
, MS_NOUSER
,
2602 tmpfs_fs_type
.name
, NULL
);
2603 if (IS_ERR(shm_mnt
)) {
2604 error
= PTR_ERR(shm_mnt
);
2605 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
2611 unregister_filesystem(&tmpfs_fs_type
);
2613 destroy_inodecache();
2615 bdi_destroy(&shmem_backing_dev_info
);
2617 shm_mnt
= ERR_PTR(error
);
2621 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2623 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2624 * @inode: the inode to be searched
2625 * @pgoff: the offset to be searched
2626 * @pagep: the pointer for the found page to be stored
2627 * @ent: the pointer for the found swap entry to be stored
2629 * If a page is found, refcount of it is incremented. Callers should handle
2632 void mem_cgroup_get_shmem_target(struct inode
*inode
, pgoff_t pgoff
,
2633 struct page
**pagep
, swp_entry_t
*ent
)
2635 swp_entry_t entry
= { .val
= 0 }, *ptr
;
2636 struct page
*page
= NULL
;
2637 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2639 if ((pgoff
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
))
2642 spin_lock(&info
->lock
);
2643 ptr
= shmem_swp_entry(info
, pgoff
, NULL
);
2645 if (ptr
&& ptr
->val
) {
2646 entry
.val
= ptr
->val
;
2647 page
= find_get_page(&swapper_space
, entry
.val
);
2650 page
= find_get_page(inode
->i_mapping
, pgoff
);
2652 shmem_swp_unmap(ptr
);
2653 spin_unlock(&info
->lock
);
2660 #else /* !CONFIG_SHMEM */
2663 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2665 * This is intended for small system where the benefits of the full
2666 * shmem code (swap-backed and resource-limited) are outweighed by
2667 * their complexity. On systems without swap this code should be
2668 * effectively equivalent, but much lighter weight.
2671 #include <linux/ramfs.h>
2673 static struct file_system_type tmpfs_fs_type
= {
2675 .mount
= ramfs_mount
,
2676 .kill_sb
= kill_litter_super
,
2679 int __init
init_tmpfs(void)
2681 BUG_ON(register_filesystem(&tmpfs_fs_type
) != 0);
2683 shm_mnt
= kern_mount(&tmpfs_fs_type
);
2684 BUG_ON(IS_ERR(shm_mnt
));
2689 int shmem_unuse(swp_entry_t entry
, struct page
*page
)
2694 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2699 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2701 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2702 * @inode: the inode to be searched
2703 * @pgoff: the offset to be searched
2704 * @pagep: the pointer for the found page to be stored
2705 * @ent: the pointer for the found swap entry to be stored
2707 * If a page is found, refcount of it is incremented. Callers should handle
2710 void mem_cgroup_get_shmem_target(struct inode
*inode
, pgoff_t pgoff
,
2711 struct page
**pagep
, swp_entry_t
*ent
)
2713 struct page
*page
= NULL
;
2715 if ((pgoff
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
))
2717 page
= find_get_page(inode
->i_mapping
, pgoff
);
2720 *ent
= (swp_entry_t
){ .val
= 0 };
2724 #define shmem_vm_ops generic_file_vm_ops
2725 #define shmem_file_operations ramfs_file_operations
2726 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2727 #define shmem_acct_size(flags, size) 0
2728 #define shmem_unacct_size(flags, size) do {} while (0)
2729 #define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
2731 #endif /* CONFIG_SHMEM */
2736 * shmem_file_setup - get an unlinked file living in tmpfs
2737 * @name: name for dentry (to be seen in /proc/<pid>/maps
2738 * @size: size to be set for the file
2739 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2741 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
2745 struct inode
*inode
;
2747 struct dentry
*root
;
2750 if (IS_ERR(shm_mnt
))
2751 return (void *)shm_mnt
;
2753 if (size
< 0 || size
> SHMEM_MAX_BYTES
)
2754 return ERR_PTR(-EINVAL
);
2756 if (shmem_acct_size(flags
, size
))
2757 return ERR_PTR(-ENOMEM
);
2761 this.len
= strlen(name
);
2762 this.hash
= 0; /* will go */
2763 root
= shm_mnt
->mnt_root
;
2764 path
.dentry
= d_alloc(root
, &this);
2767 path
.mnt
= mntget(shm_mnt
);
2770 inode
= shmem_get_inode(root
->d_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
2774 d_instantiate(path
.dentry
, inode
);
2775 inode
->i_size
= size
;
2776 inode
->i_nlink
= 0; /* It is unlinked */
2778 error
= ramfs_nommu_expand_for_mapping(inode
, size
);
2784 file
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
2785 &shmem_file_operations
);
2794 shmem_unacct_size(flags
, size
);
2795 return ERR_PTR(error
);
2797 EXPORT_SYMBOL_GPL(shmem_file_setup
);
2800 * shmem_zero_setup - setup a shared anonymous mapping
2801 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2803 int shmem_zero_setup(struct vm_area_struct
*vma
)
2806 loff_t size
= vma
->vm_end
- vma
->vm_start
;
2808 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
);
2810 return PTR_ERR(file
);
2814 vma
->vm_file
= file
;
2815 vma
->vm_ops
= &shmem_vm_ops
;
2816 vma
->vm_flags
|= VM_CAN_NONLINEAR
;