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/swap.h>
33 static struct vfsmount
*shm_mnt
;
37 * This virtual memory filesystem is heavily based on the ramfs. It
38 * extends ramfs by the ability to use swap and honor resource limits
39 * which makes it a completely usable filesystem.
42 #include <linux/xattr.h>
43 #include <linux/exportfs.h>
44 #include <linux/posix_acl.h>
45 #include <linux/generic_acl.h>
46 #include <linux/mman.h>
47 #include <linux/string.h>
48 #include <linux/slab.h>
49 #include <linux/backing-dev.h>
50 #include <linux/shmem_fs.h>
51 #include <linux/writeback.h>
52 #include <linux/blkdev.h>
53 #include <linux/security.h>
54 #include <linux/swapops.h>
55 #include <linux/mempolicy.h>
56 #include <linux/namei.h>
57 #include <linux/ctype.h>
58 #include <linux/migrate.h>
59 #include <linux/highmem.h>
60 #include <linux/seq_file.h>
61 #include <linux/magic.h>
63 #include <asm/uaccess.h>
64 #include <asm/div64.h>
65 #include <asm/pgtable.h>
68 * The maximum size of a shmem/tmpfs file is limited by the maximum size of
69 * its triple-indirect swap vector - see illustration at shmem_swp_entry().
71 * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
72 * but one eighth of that on a 64-bit kernel. With 8kB page size, maximum
73 * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
74 * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
76 * We use / and * instead of shifts in the definitions below, so that the swap
77 * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
79 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
80 #define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
82 #define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
83 #define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
85 #define SHMEM_MAX_BYTES min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
86 #define SHMEM_MAX_INDEX ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
88 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
89 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
91 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
92 #define SHMEM_PAGEIN VM_READ
93 #define SHMEM_TRUNCATE VM_WRITE
95 /* Definition to limit shmem_truncate's steps between cond_rescheds */
96 #define LATENCY_LIMIT 64
98 /* Pretend that each entry is of this size in directory's i_size */
99 #define BOGO_DIRENT_SIZE 20
101 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
103 SGP_READ
, /* don't exceed i_size, don't allocate page */
104 SGP_CACHE
, /* don't exceed i_size, may allocate page */
105 SGP_DIRTY
, /* like SGP_CACHE, but set new page dirty */
106 SGP_WRITE
, /* may exceed i_size, may allocate page */
110 static unsigned long shmem_default_max_blocks(void)
112 return totalram_pages
/ 2;
115 static unsigned long shmem_default_max_inodes(void)
117 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
121 static int shmem_getpage(struct inode
*inode
, unsigned long idx
,
122 struct page
**pagep
, enum sgp_type sgp
, int *type
);
124 static inline struct page
*shmem_dir_alloc(gfp_t gfp_mask
)
127 * The above definition of ENTRIES_PER_PAGE, and the use of
128 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
129 * might be reconsidered if it ever diverges from PAGE_SIZE.
131 * Mobility flags are masked out as swap vectors cannot move
133 return alloc_pages((gfp_mask
& ~GFP_MOVABLE_MASK
) | __GFP_ZERO
,
134 PAGE_CACHE_SHIFT
-PAGE_SHIFT
);
137 static inline void shmem_dir_free(struct page
*page
)
139 __free_pages(page
, PAGE_CACHE_SHIFT
-PAGE_SHIFT
);
142 static struct page
**shmem_dir_map(struct page
*page
)
144 return (struct page
**)kmap_atomic(page
, KM_USER0
);
147 static inline void shmem_dir_unmap(struct page
**dir
)
149 kunmap_atomic(dir
, KM_USER0
);
152 static swp_entry_t
*shmem_swp_map(struct page
*page
)
154 return (swp_entry_t
*)kmap_atomic(page
, KM_USER1
);
157 static inline void shmem_swp_balance_unmap(void)
160 * When passing a pointer to an i_direct entry, to code which
161 * also handles indirect entries and so will shmem_swp_unmap,
162 * we must arrange for the preempt count to remain in balance.
163 * What kmap_atomic of a lowmem page does depends on config
164 * and architecture, so pretend to kmap_atomic some lowmem page.
166 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1
);
169 static inline void shmem_swp_unmap(swp_entry_t
*entry
)
171 kunmap_atomic(entry
, KM_USER1
);
174 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
176 return sb
->s_fs_info
;
180 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
181 * for shared memory and for shared anonymous (/dev/zero) mappings
182 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
183 * consistent with the pre-accounting of private mappings ...
185 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
187 return (flags
& VM_NORESERVE
) ?
188 0 : security_vm_enough_memory_kern(VM_ACCT(size
));
191 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
193 if (!(flags
& VM_NORESERVE
))
194 vm_unacct_memory(VM_ACCT(size
));
198 * ... whereas tmpfs objects are accounted incrementally as
199 * pages are allocated, in order to allow huge sparse files.
200 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
201 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
203 static inline int shmem_acct_block(unsigned long flags
)
205 return (flags
& VM_NORESERVE
) ?
206 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
209 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
211 if (flags
& VM_NORESERVE
)
212 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
215 static const struct super_operations shmem_ops
;
216 static const struct address_space_operations shmem_aops
;
217 static const struct file_operations shmem_file_operations
;
218 static const struct inode_operations shmem_inode_operations
;
219 static const struct inode_operations shmem_dir_inode_operations
;
220 static const struct inode_operations shmem_special_inode_operations
;
221 static const struct vm_operations_struct shmem_vm_ops
;
223 static struct backing_dev_info shmem_backing_dev_info __read_mostly
= {
224 .ra_pages
= 0, /* No readahead */
225 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
| BDI_CAP_SWAP_BACKED
,
226 .unplug_io_fn
= default_unplug_io_fn
,
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 spin_lock(&sbinfo
->stat_lock
);
237 sbinfo
->free_blocks
+= pages
;
238 inode
->i_blocks
-= pages
*BLOCKS_PER_PAGE
;
239 spin_unlock(&sbinfo
->stat_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 free_blocks against 1 not 0, 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 spin_lock(&sbinfo
->stat_lock
);
425 if (sbinfo
->free_blocks
<= 1) {
426 spin_unlock(&sbinfo
->stat_lock
);
427 return ERR_PTR(-ENOSPC
);
429 sbinfo
->free_blocks
--;
430 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
431 spin_unlock(&sbinfo
->stat_lock
);
434 spin_unlock(&info
->lock
);
435 page
= shmem_dir_alloc(mapping_gfp_mask(inode
->i_mapping
));
436 spin_lock(&info
->lock
);
439 shmem_free_blocks(inode
, 1);
440 return ERR_PTR(-ENOMEM
);
442 if (sgp
!= SGP_WRITE
&&
443 ((loff_t
) index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
444 entry
= ERR_PTR(-EINVAL
);
447 if (info
->next_index
<= index
)
448 info
->next_index
= index
+ 1;
451 /* another task gave its page, or truncated the file */
452 shmem_free_blocks(inode
, 1);
453 shmem_dir_free(page
);
455 if (info
->next_index
<= index
&& !IS_ERR(entry
))
456 info
->next_index
= index
+ 1;
461 * shmem_free_swp - free some swap entries in a directory
462 * @dir: pointer to the directory
463 * @edir: pointer after last entry of the directory
464 * @punch_lock: pointer to spinlock when needed for the holepunch case
466 static int shmem_free_swp(swp_entry_t
*dir
, swp_entry_t
*edir
,
467 spinlock_t
*punch_lock
)
469 spinlock_t
*punch_unlock
= NULL
;
473 for (ptr
= dir
; ptr
< edir
; ptr
++) {
475 if (unlikely(punch_lock
)) {
476 punch_unlock
= punch_lock
;
478 spin_lock(punch_unlock
);
482 free_swap_and_cache(*ptr
);
483 *ptr
= (swp_entry_t
){0};
488 spin_unlock(punch_unlock
);
492 static int shmem_map_and_free_swp(struct page
*subdir
, int offset
,
493 int limit
, struct page
***dir
, spinlock_t
*punch_lock
)
498 ptr
= shmem_swp_map(subdir
);
499 for (; offset
< limit
; offset
+= LATENCY_LIMIT
) {
500 int size
= limit
- offset
;
501 if (size
> LATENCY_LIMIT
)
502 size
= LATENCY_LIMIT
;
503 freed
+= shmem_free_swp(ptr
+offset
, ptr
+offset
+size
,
505 if (need_resched()) {
506 shmem_swp_unmap(ptr
);
508 shmem_dir_unmap(*dir
);
512 ptr
= shmem_swp_map(subdir
);
515 shmem_swp_unmap(ptr
);
519 static void shmem_free_pages(struct list_head
*next
)
525 page
= container_of(next
, struct page
, lru
);
527 shmem_dir_free(page
);
529 if (freed
>= LATENCY_LIMIT
) {
536 static void shmem_truncate_range(struct inode
*inode
, loff_t start
, loff_t end
)
538 struct shmem_inode_info
*info
= SHMEM_I(inode
);
543 unsigned long diroff
;
549 LIST_HEAD(pages_to_free
);
550 long nr_pages_to_free
= 0;
551 long nr_swaps_freed
= 0;
555 spinlock_t
*needs_lock
;
556 spinlock_t
*punch_lock
;
557 unsigned long upper_limit
;
559 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
560 idx
= (start
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
561 if (idx
>= info
->next_index
)
564 spin_lock(&info
->lock
);
565 info
->flags
|= SHMEM_TRUNCATE
;
566 if (likely(end
== (loff_t
) -1)) {
567 limit
= info
->next_index
;
568 upper_limit
= SHMEM_MAX_INDEX
;
569 info
->next_index
= idx
;
573 if (end
+ 1 >= inode
->i_size
) { /* we may free a little more */
574 limit
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >>
576 upper_limit
= SHMEM_MAX_INDEX
;
578 limit
= (end
+ 1) >> PAGE_CACHE_SHIFT
;
581 needs_lock
= &info
->lock
;
585 topdir
= info
->i_indirect
;
586 if (topdir
&& idx
<= SHMEM_NR_DIRECT
&& !punch_hole
) {
587 info
->i_indirect
= NULL
;
589 list_add(&topdir
->lru
, &pages_to_free
);
591 spin_unlock(&info
->lock
);
593 if (info
->swapped
&& idx
< SHMEM_NR_DIRECT
) {
594 ptr
= info
->i_direct
;
596 if (size
> SHMEM_NR_DIRECT
)
597 size
= SHMEM_NR_DIRECT
;
598 nr_swaps_freed
= shmem_free_swp(ptr
+idx
, ptr
+size
, needs_lock
);
602 * If there are no indirect blocks or we are punching a hole
603 * below indirect blocks, nothing to be done.
605 if (!topdir
|| limit
<= SHMEM_NR_DIRECT
)
609 * The truncation case has already dropped info->lock, and we're safe
610 * because i_size and next_index have already been lowered, preventing
611 * access beyond. But in the punch_hole case, we still need to take
612 * the lock when updating the swap directory, because there might be
613 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
614 * shmem_writepage. However, whenever we find we can remove a whole
615 * directory page (not at the misaligned start or end of the range),
616 * we first NULLify its pointer in the level above, and then have no
617 * need to take the lock when updating its contents: needs_lock and
618 * punch_lock (either pointing to info->lock or NULL) manage this.
621 upper_limit
-= SHMEM_NR_DIRECT
;
622 limit
-= SHMEM_NR_DIRECT
;
623 idx
= (idx
> SHMEM_NR_DIRECT
)? (idx
- SHMEM_NR_DIRECT
): 0;
624 offset
= idx
% ENTRIES_PER_PAGE
;
627 dir
= shmem_dir_map(topdir
);
628 stage
= ENTRIES_PER_PAGEPAGE
/2;
629 if (idx
< ENTRIES_PER_PAGEPAGE
/2) {
631 diroff
= idx
/ENTRIES_PER_PAGE
;
633 dir
+= ENTRIES_PER_PAGE
/2;
634 dir
+= (idx
- ENTRIES_PER_PAGEPAGE
/2)/ENTRIES_PER_PAGEPAGE
;
636 stage
+= ENTRIES_PER_PAGEPAGE
;
639 diroff
= ((idx
- ENTRIES_PER_PAGEPAGE
/2) %
640 ENTRIES_PER_PAGEPAGE
) / ENTRIES_PER_PAGE
;
641 if (!diroff
&& !offset
&& upper_limit
>= stage
) {
643 spin_lock(needs_lock
);
645 spin_unlock(needs_lock
);
650 list_add(&middir
->lru
, &pages_to_free
);
652 shmem_dir_unmap(dir
);
653 dir
= shmem_dir_map(middir
);
661 for (; idx
< limit
; idx
+= ENTRIES_PER_PAGE
, diroff
++) {
662 if (unlikely(idx
== stage
)) {
663 shmem_dir_unmap(dir
);
664 dir
= shmem_dir_map(topdir
) +
665 ENTRIES_PER_PAGE
/2 + idx
/ENTRIES_PER_PAGEPAGE
;
668 idx
+= ENTRIES_PER_PAGEPAGE
;
672 stage
= idx
+ ENTRIES_PER_PAGEPAGE
;
675 needs_lock
= &info
->lock
;
676 if (upper_limit
>= stage
) {
678 spin_lock(needs_lock
);
680 spin_unlock(needs_lock
);
685 list_add(&middir
->lru
, &pages_to_free
);
687 shmem_dir_unmap(dir
);
689 dir
= shmem_dir_map(middir
);
692 punch_lock
= needs_lock
;
693 subdir
= dir
[diroff
];
694 if (subdir
&& !offset
&& upper_limit
-idx
>= ENTRIES_PER_PAGE
) {
696 spin_lock(needs_lock
);
698 spin_unlock(needs_lock
);
703 list_add(&subdir
->lru
, &pages_to_free
);
705 if (subdir
&& page_private(subdir
) /* has swap entries */) {
707 if (size
> ENTRIES_PER_PAGE
)
708 size
= ENTRIES_PER_PAGE
;
709 freed
= shmem_map_and_free_swp(subdir
,
710 offset
, size
, &dir
, punch_lock
);
712 dir
= shmem_dir_map(middir
);
713 nr_swaps_freed
+= freed
;
714 if (offset
|| punch_lock
) {
715 spin_lock(&info
->lock
);
716 set_page_private(subdir
,
717 page_private(subdir
) - freed
);
718 spin_unlock(&info
->lock
);
720 BUG_ON(page_private(subdir
) != freed
);
725 shmem_dir_unmap(dir
);
727 if (inode
->i_mapping
->nrpages
&& (info
->flags
& SHMEM_PAGEIN
)) {
729 * Call truncate_inode_pages again: racing shmem_unuse_inode
730 * may have swizzled a page in from swap since
731 * truncate_pagecache or generic_delete_inode did it, before we
732 * lowered next_index. Also, though shmem_getpage checks
733 * i_size before adding to cache, no recheck after: so fix the
734 * narrow window there too.
736 * Recalling truncate_inode_pages_range and unmap_mapping_range
737 * every time for punch_hole (which never got a chance to clear
738 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
739 * yet hardly ever necessary: try to optimize them out later.
741 truncate_inode_pages_range(inode
->i_mapping
, start
, end
);
743 unmap_mapping_range(inode
->i_mapping
, start
,
747 spin_lock(&info
->lock
);
748 info
->flags
&= ~SHMEM_TRUNCATE
;
749 info
->swapped
-= nr_swaps_freed
;
750 if (nr_pages_to_free
)
751 shmem_free_blocks(inode
, nr_pages_to_free
);
752 shmem_recalc_inode(inode
);
753 spin_unlock(&info
->lock
);
756 * Empty swap vector directory pages to be freed?
758 if (!list_empty(&pages_to_free
)) {
759 pages_to_free
.prev
->next
= NULL
;
760 shmem_free_pages(pages_to_free
.next
);
764 static int shmem_notify_change(struct dentry
*dentry
, struct iattr
*attr
)
766 struct inode
*inode
= dentry
->d_inode
;
767 loff_t newsize
= attr
->ia_size
;
770 error
= inode_change_ok(inode
, attr
);
774 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)
775 && newsize
!= inode
->i_size
) {
776 struct page
*page
= NULL
;
778 if (newsize
< inode
->i_size
) {
780 * If truncating down to a partial page, then
781 * if that page is already allocated, hold it
782 * in memory until the truncation is over, so
783 * truncate_partial_page cannnot miss it were
784 * it assigned to swap.
786 if (newsize
& (PAGE_CACHE_SIZE
-1)) {
787 (void) shmem_getpage(inode
,
788 newsize
>> PAGE_CACHE_SHIFT
,
789 &page
, SGP_READ
, NULL
);
794 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
795 * detect if any pages might have been added to cache
796 * after truncate_inode_pages. But we needn't bother
797 * if it's being fully truncated to zero-length: the
798 * nrpages check is efficient enough in that case.
801 struct shmem_inode_info
*info
= SHMEM_I(inode
);
802 spin_lock(&info
->lock
);
803 info
->flags
&= ~SHMEM_PAGEIN
;
804 spin_unlock(&info
->lock
);
808 /* XXX(truncate): truncate_setsize should be called last */
809 truncate_setsize(inode
, newsize
);
811 page_cache_release(page
);
812 shmem_truncate_range(inode
, newsize
, (loff_t
)-1);
815 setattr_copy(inode
, attr
);
816 #ifdef CONFIG_TMPFS_POSIX_ACL
817 if (attr
->ia_valid
& ATTR_MODE
)
818 error
= generic_acl_chmod(inode
);
823 static void shmem_delete_inode(struct inode
*inode
)
825 struct shmem_inode_info
*info
= SHMEM_I(inode
);
827 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
828 truncate_inode_pages(inode
->i_mapping
, 0);
829 shmem_unacct_size(info
->flags
, inode
->i_size
);
831 shmem_truncate_range(inode
, 0, (loff_t
)-1);
832 if (!list_empty(&info
->swaplist
)) {
833 mutex_lock(&shmem_swaplist_mutex
);
834 list_del_init(&info
->swaplist
);
835 mutex_unlock(&shmem_swaplist_mutex
);
838 BUG_ON(inode
->i_blocks
);
839 shmem_free_inode(inode
->i_sb
);
843 static inline int shmem_find_swp(swp_entry_t entry
, swp_entry_t
*dir
, swp_entry_t
*edir
)
847 for (ptr
= dir
; ptr
< edir
; ptr
++) {
848 if (ptr
->val
== entry
.val
)
854 static int shmem_unuse_inode(struct shmem_inode_info
*info
, swp_entry_t entry
, struct page
*page
)
868 ptr
= info
->i_direct
;
869 spin_lock(&info
->lock
);
870 if (!info
->swapped
) {
871 list_del_init(&info
->swaplist
);
874 limit
= info
->next_index
;
876 if (size
> SHMEM_NR_DIRECT
)
877 size
= SHMEM_NR_DIRECT
;
878 offset
= shmem_find_swp(entry
, ptr
, ptr
+size
);
881 if (!info
->i_indirect
)
884 dir
= shmem_dir_map(info
->i_indirect
);
885 stage
= SHMEM_NR_DIRECT
+ ENTRIES_PER_PAGEPAGE
/2;
887 for (idx
= SHMEM_NR_DIRECT
; idx
< limit
; idx
+= ENTRIES_PER_PAGE
, dir
++) {
888 if (unlikely(idx
== stage
)) {
889 shmem_dir_unmap(dir
-1);
890 if (cond_resched_lock(&info
->lock
)) {
891 /* check it has not been truncated */
892 if (limit
> info
->next_index
) {
893 limit
= info
->next_index
;
898 dir
= shmem_dir_map(info
->i_indirect
) +
899 ENTRIES_PER_PAGE
/2 + idx
/ENTRIES_PER_PAGEPAGE
;
902 idx
+= ENTRIES_PER_PAGEPAGE
;
906 stage
= idx
+ ENTRIES_PER_PAGEPAGE
;
908 shmem_dir_unmap(dir
);
909 dir
= shmem_dir_map(subdir
);
912 if (subdir
&& page_private(subdir
)) {
913 ptr
= shmem_swp_map(subdir
);
915 if (size
> ENTRIES_PER_PAGE
)
916 size
= ENTRIES_PER_PAGE
;
917 offset
= shmem_find_swp(entry
, ptr
, ptr
+size
);
918 shmem_swp_unmap(ptr
);
920 shmem_dir_unmap(dir
);
926 shmem_dir_unmap(dir
-1);
928 spin_unlock(&info
->lock
);
932 inode
= igrab(&info
->vfs_inode
);
933 spin_unlock(&info
->lock
);
936 * Move _head_ to start search for next from here.
937 * But be careful: shmem_delete_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
);
944 mutex_unlock(&shmem_swaplist_mutex
);
950 * Charge page using GFP_KERNEL while we can wait.
951 * Charged back to the user(not to caller) when swap account is used.
952 * add_to_page_cache() will be called with GFP_NOWAIT.
954 error
= mem_cgroup_cache_charge(page
, current
->mm
, GFP_KERNEL
);
957 error
= radix_tree_preload(GFP_KERNEL
);
959 mem_cgroup_uncharge_cache_page(page
);
964 spin_lock(&info
->lock
);
965 ptr
= shmem_swp_entry(info
, idx
, NULL
);
966 if (ptr
&& ptr
->val
== entry
.val
) {
967 error
= add_to_page_cache_locked(page
, inode
->i_mapping
,
969 /* does mem_cgroup_uncharge_cache_page on error */
970 } else /* we must compensate for our precharge above */
971 mem_cgroup_uncharge_cache_page(page
);
973 if (error
== -EEXIST
) {
974 struct page
*filepage
= find_get_page(inode
->i_mapping
, idx
);
978 * There might be a more uptodate page coming down
979 * from a stacked writepage: forget our swappage if so.
981 if (PageUptodate(filepage
))
983 page_cache_release(filepage
);
987 delete_from_swap_cache(page
);
988 set_page_dirty(page
);
989 info
->flags
|= SHMEM_PAGEIN
;
990 shmem_swp_set(info
, ptr
, 0);
992 error
= 1; /* not an error, but entry was found */
995 shmem_swp_unmap(ptr
);
996 spin_unlock(&info
->lock
);
997 radix_tree_preload_end();
1000 page_cache_release(page
);
1001 iput(inode
); /* allows for NULL */
1006 * shmem_unuse() search for an eventually swapped out shmem page.
1008 int shmem_unuse(swp_entry_t entry
, struct page
*page
)
1010 struct list_head
*p
, *next
;
1011 struct shmem_inode_info
*info
;
1014 mutex_lock(&shmem_swaplist_mutex
);
1015 list_for_each_safe(p
, next
, &shmem_swaplist
) {
1016 info
= list_entry(p
, struct shmem_inode_info
, swaplist
);
1017 found
= shmem_unuse_inode(info
, entry
, page
);
1022 mutex_unlock(&shmem_swaplist_mutex
);
1024 * Can some race bring us here? We've been holding page lock,
1025 * so I think not; but would rather try again later than BUG()
1028 page_cache_release(page
);
1030 return (found
< 0) ? found
: 0;
1034 * Move the page from the page cache to the swap cache.
1036 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1038 struct shmem_inode_info
*info
;
1039 swp_entry_t
*entry
, swap
;
1040 struct address_space
*mapping
;
1041 unsigned long index
;
1042 struct inode
*inode
;
1044 BUG_ON(!PageLocked(page
));
1045 mapping
= page
->mapping
;
1046 index
= page
->index
;
1047 inode
= mapping
->host
;
1048 info
= SHMEM_I(inode
);
1049 if (info
->flags
& VM_LOCKED
)
1051 if (!total_swap_pages
)
1055 * shmem_backing_dev_info's capabilities prevent regular writeback or
1056 * sync from ever calling shmem_writepage; but a stacking filesystem
1057 * may use the ->writepage of its underlying filesystem, in which case
1058 * tmpfs should write out to swap only in response to memory pressure,
1059 * and not for the writeback threads or sync. However, in those cases,
1060 * we do still want to check if there's a redundant swappage to be
1063 if (wbc
->for_reclaim
)
1064 swap
= get_swap_page();
1068 spin_lock(&info
->lock
);
1069 if (index
>= info
->next_index
) {
1070 BUG_ON(!(info
->flags
& SHMEM_TRUNCATE
));
1073 entry
= shmem_swp_entry(info
, index
, NULL
);
1076 * The more uptodate page coming down from a stacked
1077 * writepage should replace our old swappage.
1079 free_swap_and_cache(*entry
);
1080 shmem_swp_set(info
, entry
, 0);
1082 shmem_recalc_inode(inode
);
1084 if (swap
.val
&& add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1085 remove_from_page_cache(page
);
1086 shmem_swp_set(info
, entry
, swap
.val
);
1087 shmem_swp_unmap(entry
);
1088 if (list_empty(&info
->swaplist
))
1089 inode
= igrab(inode
);
1092 spin_unlock(&info
->lock
);
1093 swap_shmem_alloc(swap
);
1094 BUG_ON(page_mapped(page
));
1095 page_cache_release(page
); /* pagecache ref */
1096 swap_writepage(page
, wbc
);
1098 mutex_lock(&shmem_swaplist_mutex
);
1099 /* move instead of add in case we're racing */
1100 list_move_tail(&info
->swaplist
, &shmem_swaplist
);
1101 mutex_unlock(&shmem_swaplist_mutex
);
1107 shmem_swp_unmap(entry
);
1109 spin_unlock(&info
->lock
);
1111 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1112 * clear SWAP_HAS_CACHE flag.
1114 swapcache_free(swap
, NULL
);
1116 set_page_dirty(page
);
1117 if (wbc
->for_reclaim
)
1118 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1125 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1129 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1130 return; /* show nothing */
1132 mpol_to_str(buffer
, sizeof(buffer
), mpol
, 1);
1134 seq_printf(seq
, ",mpol=%s", buffer
);
1137 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1139 struct mempolicy
*mpol
= NULL
;
1141 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1142 mpol
= sbinfo
->mpol
;
1144 spin_unlock(&sbinfo
->stat_lock
);
1148 #endif /* CONFIG_TMPFS */
1150 static struct page
*shmem_swapin(swp_entry_t entry
, gfp_t gfp
,
1151 struct shmem_inode_info
*info
, unsigned long idx
)
1153 struct mempolicy mpol
, *spol
;
1154 struct vm_area_struct pvma
;
1157 spol
= mpol_cond_copy(&mpol
,
1158 mpol_shared_policy_lookup(&info
->policy
, idx
));
1160 /* Create a pseudo vma that just contains the policy */
1162 pvma
.vm_pgoff
= idx
;
1164 pvma
.vm_policy
= spol
;
1165 page
= swapin_readahead(entry
, gfp
, &pvma
, 0);
1169 static struct page
*shmem_alloc_page(gfp_t gfp
,
1170 struct shmem_inode_info
*info
, unsigned long idx
)
1172 struct vm_area_struct pvma
;
1174 /* Create a pseudo vma that just contains the policy */
1176 pvma
.vm_pgoff
= idx
;
1178 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, idx
);
1181 * alloc_page_vma() will drop the shared policy reference
1183 return alloc_page_vma(gfp
, &pvma
, 0);
1185 #else /* !CONFIG_NUMA */
1187 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*p
)
1190 #endif /* CONFIG_TMPFS */
1192 static inline struct page
*shmem_swapin(swp_entry_t entry
, gfp_t gfp
,
1193 struct shmem_inode_info
*info
, unsigned long idx
)
1195 return swapin_readahead(entry
, gfp
, NULL
, 0);
1198 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
1199 struct shmem_inode_info
*info
, unsigned long idx
)
1201 return alloc_page(gfp
);
1203 #endif /* CONFIG_NUMA */
1205 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1206 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1213 * shmem_getpage - either get the page from swap or allocate a new one
1215 * If we allocate a new one we do not mark it dirty. That's up to the
1216 * vm. If we swap it in we mark it dirty since we also free the swap
1217 * entry since a page cannot live in both the swap and page cache
1219 static int shmem_getpage(struct inode
*inode
, unsigned long idx
,
1220 struct page
**pagep
, enum sgp_type sgp
, int *type
)
1222 struct address_space
*mapping
= inode
->i_mapping
;
1223 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1224 struct shmem_sb_info
*sbinfo
;
1225 struct page
*filepage
= *pagep
;
1226 struct page
*swappage
;
1232 if (idx
>= SHMEM_MAX_INDEX
)
1239 * Normally, filepage is NULL on entry, and either found
1240 * uptodate immediately, or allocated and zeroed, or read
1241 * in under swappage, which is then assigned to filepage.
1242 * But shmem_readpage (required for splice) passes in a locked
1243 * filepage, which may be found not uptodate by other callers
1244 * too, and may need to be copied from the swappage read in.
1248 filepage
= find_lock_page(mapping
, idx
);
1249 if (filepage
&& PageUptodate(filepage
))
1252 gfp
= mapping_gfp_mask(mapping
);
1255 * Try to preload while we can wait, to not make a habit of
1256 * draining atomic reserves; but don't latch on to this cpu.
1258 error
= radix_tree_preload(gfp
& ~__GFP_HIGHMEM
);
1261 radix_tree_preload_end();
1264 spin_lock(&info
->lock
);
1265 shmem_recalc_inode(inode
);
1266 entry
= shmem_swp_alloc(info
, idx
, sgp
);
1267 if (IS_ERR(entry
)) {
1268 spin_unlock(&info
->lock
);
1269 error
= PTR_ERR(entry
);
1275 /* Look it up and read it in.. */
1276 swappage
= lookup_swap_cache(swap
);
1278 shmem_swp_unmap(entry
);
1279 /* here we actually do the io */
1280 if (type
&& !(*type
& VM_FAULT_MAJOR
)) {
1281 __count_vm_event(PGMAJFAULT
);
1282 *type
|= VM_FAULT_MAJOR
;
1284 spin_unlock(&info
->lock
);
1285 swappage
= shmem_swapin(swap
, gfp
, info
, idx
);
1287 spin_lock(&info
->lock
);
1288 entry
= shmem_swp_alloc(info
, idx
, sgp
);
1290 error
= PTR_ERR(entry
);
1292 if (entry
->val
== swap
.val
)
1294 shmem_swp_unmap(entry
);
1296 spin_unlock(&info
->lock
);
1301 wait_on_page_locked(swappage
);
1302 page_cache_release(swappage
);
1306 /* We have to do this with page locked to prevent races */
1307 if (!trylock_page(swappage
)) {
1308 shmem_swp_unmap(entry
);
1309 spin_unlock(&info
->lock
);
1310 wait_on_page_locked(swappage
);
1311 page_cache_release(swappage
);
1314 if (PageWriteback(swappage
)) {
1315 shmem_swp_unmap(entry
);
1316 spin_unlock(&info
->lock
);
1317 wait_on_page_writeback(swappage
);
1318 unlock_page(swappage
);
1319 page_cache_release(swappage
);
1322 if (!PageUptodate(swappage
)) {
1323 shmem_swp_unmap(entry
);
1324 spin_unlock(&info
->lock
);
1325 unlock_page(swappage
);
1326 page_cache_release(swappage
);
1332 shmem_swp_set(info
, entry
, 0);
1333 shmem_swp_unmap(entry
);
1334 delete_from_swap_cache(swappage
);
1335 spin_unlock(&info
->lock
);
1336 copy_highpage(filepage
, swappage
);
1337 unlock_page(swappage
);
1338 page_cache_release(swappage
);
1339 flush_dcache_page(filepage
);
1340 SetPageUptodate(filepage
);
1341 set_page_dirty(filepage
);
1343 } else if (!(error
= add_to_page_cache_locked(swappage
, mapping
,
1344 idx
, GFP_NOWAIT
))) {
1345 info
->flags
|= SHMEM_PAGEIN
;
1346 shmem_swp_set(info
, entry
, 0);
1347 shmem_swp_unmap(entry
);
1348 delete_from_swap_cache(swappage
);
1349 spin_unlock(&info
->lock
);
1350 filepage
= swappage
;
1351 set_page_dirty(filepage
);
1354 shmem_swp_unmap(entry
);
1355 spin_unlock(&info
->lock
);
1356 if (error
== -ENOMEM
) {
1358 * reclaim from proper memory cgroup and
1359 * call memcg's OOM if needed.
1361 error
= mem_cgroup_shmem_charge_fallback(
1366 unlock_page(swappage
);
1367 page_cache_release(swappage
);
1371 unlock_page(swappage
);
1372 page_cache_release(swappage
);
1375 } else if (sgp
== SGP_READ
&& !filepage
) {
1376 shmem_swp_unmap(entry
);
1377 filepage
= find_get_page(mapping
, idx
);
1379 (!PageUptodate(filepage
) || !trylock_page(filepage
))) {
1380 spin_unlock(&info
->lock
);
1381 wait_on_page_locked(filepage
);
1382 page_cache_release(filepage
);
1386 spin_unlock(&info
->lock
);
1388 shmem_swp_unmap(entry
);
1389 sbinfo
= SHMEM_SB(inode
->i_sb
);
1390 if (sbinfo
->max_blocks
) {
1391 spin_lock(&sbinfo
->stat_lock
);
1392 if (sbinfo
->free_blocks
== 0 ||
1393 shmem_acct_block(info
->flags
)) {
1394 spin_unlock(&sbinfo
->stat_lock
);
1395 spin_unlock(&info
->lock
);
1399 sbinfo
->free_blocks
--;
1400 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1401 spin_unlock(&sbinfo
->stat_lock
);
1402 } else if (shmem_acct_block(info
->flags
)) {
1403 spin_unlock(&info
->lock
);
1411 spin_unlock(&info
->lock
);
1412 filepage
= shmem_alloc_page(gfp
, info
, idx
);
1414 shmem_unacct_blocks(info
->flags
, 1);
1415 shmem_free_blocks(inode
, 1);
1419 SetPageSwapBacked(filepage
);
1421 /* Precharge page while we can wait, compensate after */
1422 error
= mem_cgroup_cache_charge(filepage
, current
->mm
,
1425 page_cache_release(filepage
);
1426 shmem_unacct_blocks(info
->flags
, 1);
1427 shmem_free_blocks(inode
, 1);
1432 spin_lock(&info
->lock
);
1433 entry
= shmem_swp_alloc(info
, idx
, sgp
);
1435 error
= PTR_ERR(entry
);
1438 shmem_swp_unmap(entry
);
1440 ret
= error
|| swap
.val
;
1442 mem_cgroup_uncharge_cache_page(filepage
);
1444 ret
= add_to_page_cache_lru(filepage
, mapping
,
1447 * At add_to_page_cache_lru() failure, uncharge will
1448 * be done automatically.
1451 spin_unlock(&info
->lock
);
1452 page_cache_release(filepage
);
1453 shmem_unacct_blocks(info
->flags
, 1);
1454 shmem_free_blocks(inode
, 1);
1460 info
->flags
|= SHMEM_PAGEIN
;
1464 spin_unlock(&info
->lock
);
1465 clear_highpage(filepage
);
1466 flush_dcache_page(filepage
);
1467 SetPageUptodate(filepage
);
1468 if (sgp
== SGP_DIRTY
)
1469 set_page_dirty(filepage
);
1476 if (*pagep
!= filepage
) {
1477 unlock_page(filepage
);
1478 page_cache_release(filepage
);
1483 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1485 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1489 if (((loff_t
)vmf
->pgoff
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
))
1490 return VM_FAULT_SIGBUS
;
1492 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1494 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1496 return ret
| VM_FAULT_LOCKED
;
1500 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*new)
1502 struct inode
*i
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1503 return mpol_set_shared_policy(&SHMEM_I(i
)->policy
, vma
, new);
1506 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1509 struct inode
*i
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1512 idx
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1513 return mpol_shared_policy_lookup(&SHMEM_I(i
)->policy
, idx
);
1517 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1519 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1520 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1521 int retval
= -ENOMEM
;
1523 spin_lock(&info
->lock
);
1524 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1525 if (!user_shm_lock(inode
->i_size
, user
))
1527 info
->flags
|= VM_LOCKED
;
1528 mapping_set_unevictable(file
->f_mapping
);
1530 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1531 user_shm_unlock(inode
->i_size
, user
);
1532 info
->flags
&= ~VM_LOCKED
;
1533 mapping_clear_unevictable(file
->f_mapping
);
1534 scan_mapping_unevictable_pages(file
->f_mapping
);
1539 spin_unlock(&info
->lock
);
1543 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1545 file_accessed(file
);
1546 vma
->vm_ops
= &shmem_vm_ops
;
1547 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
1551 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1552 int mode
, dev_t dev
, unsigned long flags
)
1554 struct inode
*inode
;
1555 struct shmem_inode_info
*info
;
1556 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1558 if (shmem_reserve_inode(sb
))
1561 inode
= new_inode(sb
);
1563 inode_init_owner(inode
, dir
, mode
);
1564 inode
->i_blocks
= 0;
1565 inode
->i_mapping
->backing_dev_info
= &shmem_backing_dev_info
;
1566 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1567 inode
->i_generation
= get_seconds();
1568 info
= SHMEM_I(inode
);
1569 memset(info
, 0, (char *)inode
- (char *)info
);
1570 spin_lock_init(&info
->lock
);
1571 info
->flags
= flags
& VM_NORESERVE
;
1572 INIT_LIST_HEAD(&info
->swaplist
);
1573 cache_no_acl(inode
);
1575 switch (mode
& S_IFMT
) {
1577 inode
->i_op
= &shmem_special_inode_operations
;
1578 init_special_inode(inode
, mode
, dev
);
1581 inode
->i_mapping
->a_ops
= &shmem_aops
;
1582 inode
->i_op
= &shmem_inode_operations
;
1583 inode
->i_fop
= &shmem_file_operations
;
1584 mpol_shared_policy_init(&info
->policy
,
1585 shmem_get_sbmpol(sbinfo
));
1589 /* Some things misbehave if size == 0 on a directory */
1590 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1591 inode
->i_op
= &shmem_dir_inode_operations
;
1592 inode
->i_fop
= &simple_dir_operations
;
1596 * Must not load anything in the rbtree,
1597 * mpol_free_shared_policy will not be called.
1599 mpol_shared_policy_init(&info
->policy
, NULL
);
1603 shmem_free_inode(sb
);
1608 static const struct inode_operations shmem_symlink_inode_operations
;
1609 static const struct inode_operations shmem_symlink_inline_operations
;
1612 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1613 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1614 * below the loop driver, in the generic fashion that many filesystems support.
1616 static int shmem_readpage(struct file
*file
, struct page
*page
)
1618 struct inode
*inode
= page
->mapping
->host
;
1619 int error
= shmem_getpage(inode
, page
->index
, &page
, SGP_CACHE
, NULL
);
1625 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1626 loff_t pos
, unsigned len
, unsigned flags
,
1627 struct page
**pagep
, void **fsdata
)
1629 struct inode
*inode
= mapping
->host
;
1630 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1632 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1636 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1637 loff_t pos
, unsigned len
, unsigned copied
,
1638 struct page
*page
, void *fsdata
)
1640 struct inode
*inode
= mapping
->host
;
1642 if (pos
+ copied
> inode
->i_size
)
1643 i_size_write(inode
, pos
+ copied
);
1645 set_page_dirty(page
);
1647 page_cache_release(page
);
1652 static void do_shmem_file_read(struct file
*filp
, loff_t
*ppos
, read_descriptor_t
*desc
, read_actor_t actor
)
1654 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
1655 struct address_space
*mapping
= inode
->i_mapping
;
1656 unsigned long index
, offset
;
1657 enum sgp_type sgp
= SGP_READ
;
1660 * Might this read be for a stacking filesystem? Then when reading
1661 * holes of a sparse file, we actually need to allocate those pages,
1662 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1664 if (segment_eq(get_fs(), KERNEL_DS
))
1667 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1668 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1671 struct page
*page
= NULL
;
1672 unsigned long end_index
, nr
, ret
;
1673 loff_t i_size
= i_size_read(inode
);
1675 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1676 if (index
> end_index
)
1678 if (index
== end_index
) {
1679 nr
= i_size
& ~PAGE_CACHE_MASK
;
1684 desc
->error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1686 if (desc
->error
== -EINVAL
)
1694 * We must evaluate after, since reads (unlike writes)
1695 * are called without i_mutex protection against truncate
1697 nr
= PAGE_CACHE_SIZE
;
1698 i_size
= i_size_read(inode
);
1699 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1700 if (index
== end_index
) {
1701 nr
= i_size
& ~PAGE_CACHE_MASK
;
1704 page_cache_release(page
);
1712 * If users can be writing to this page using arbitrary
1713 * virtual addresses, take care about potential aliasing
1714 * before reading the page on the kernel side.
1716 if (mapping_writably_mapped(mapping
))
1717 flush_dcache_page(page
);
1719 * Mark the page accessed if we read the beginning.
1722 mark_page_accessed(page
);
1724 page
= ZERO_PAGE(0);
1725 page_cache_get(page
);
1729 * Ok, we have the page, and it's up-to-date, so
1730 * now we can copy it to user space...
1732 * The actor routine returns how many bytes were actually used..
1733 * NOTE! This may not be the same as how much of a user buffer
1734 * we filled up (we may be padding etc), so we can only update
1735 * "pos" here (the actor routine has to update the user buffer
1736 * pointers and the remaining count).
1738 ret
= actor(desc
, page
, offset
, nr
);
1740 index
+= offset
>> PAGE_CACHE_SHIFT
;
1741 offset
&= ~PAGE_CACHE_MASK
;
1743 page_cache_release(page
);
1744 if (ret
!= nr
|| !desc
->count
)
1750 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1751 file_accessed(filp
);
1754 static ssize_t
shmem_file_aio_read(struct kiocb
*iocb
,
1755 const struct iovec
*iov
, unsigned long nr_segs
, loff_t pos
)
1757 struct file
*filp
= iocb
->ki_filp
;
1761 loff_t
*ppos
= &iocb
->ki_pos
;
1763 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1767 for (seg
= 0; seg
< nr_segs
; seg
++) {
1768 read_descriptor_t desc
;
1771 desc
.arg
.buf
= iov
[seg
].iov_base
;
1772 desc
.count
= iov
[seg
].iov_len
;
1773 if (desc
.count
== 0)
1776 do_shmem_file_read(filp
, ppos
, &desc
, file_read_actor
);
1777 retval
+= desc
.written
;
1779 retval
= retval
?: desc
.error
;
1788 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
1790 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
1792 buf
->f_type
= TMPFS_MAGIC
;
1793 buf
->f_bsize
= PAGE_CACHE_SIZE
;
1794 buf
->f_namelen
= NAME_MAX
;
1795 spin_lock(&sbinfo
->stat_lock
);
1796 if (sbinfo
->max_blocks
) {
1797 buf
->f_blocks
= sbinfo
->max_blocks
;
1798 buf
->f_bavail
= buf
->f_bfree
= sbinfo
->free_blocks
;
1800 if (sbinfo
->max_inodes
) {
1801 buf
->f_files
= sbinfo
->max_inodes
;
1802 buf
->f_ffree
= sbinfo
->free_inodes
;
1804 /* else leave those fields 0 like simple_statfs */
1805 spin_unlock(&sbinfo
->stat_lock
);
1810 * File creation. Allocate an inode, and we're done..
1813 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1815 struct inode
*inode
;
1816 int error
= -ENOSPC
;
1818 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
1820 error
= security_inode_init_security(inode
, dir
, NULL
, NULL
,
1823 if (error
!= -EOPNOTSUPP
) {
1828 #ifdef CONFIG_TMPFS_POSIX_ACL
1829 error
= generic_acl_init(inode
, dir
);
1837 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1838 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1839 d_instantiate(dentry
, inode
);
1840 dget(dentry
); /* Extra count - pin the dentry in core */
1845 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1849 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
1855 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1856 struct nameidata
*nd
)
1858 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
1864 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1866 struct inode
*inode
= old_dentry
->d_inode
;
1870 * No ordinary (disk based) filesystem counts links as inodes;
1871 * but each new link needs a new dentry, pinning lowmem, and
1872 * tmpfs dentries cannot be pruned until they are unlinked.
1874 ret
= shmem_reserve_inode(inode
->i_sb
);
1878 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1879 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1881 atomic_inc(&inode
->i_count
); /* New dentry reference */
1882 dget(dentry
); /* Extra pinning count for the created dentry */
1883 d_instantiate(dentry
, inode
);
1888 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
1890 struct inode
*inode
= dentry
->d_inode
;
1892 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
1893 shmem_free_inode(inode
->i_sb
);
1895 dir
->i_size
-= BOGO_DIRENT_SIZE
;
1896 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1898 dput(dentry
); /* Undo the count from "create" - this does all the work */
1902 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1904 if (!simple_empty(dentry
))
1907 drop_nlink(dentry
->d_inode
);
1909 return shmem_unlink(dir
, dentry
);
1913 * The VFS layer already does all the dentry stuff for rename,
1914 * we just have to decrement the usage count for the target if
1915 * it exists so that the VFS layer correctly free's it when it
1918 static int shmem_rename(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
1920 struct inode
*inode
= old_dentry
->d_inode
;
1921 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
1923 if (!simple_empty(new_dentry
))
1926 if (new_dentry
->d_inode
) {
1927 (void) shmem_unlink(new_dir
, new_dentry
);
1929 drop_nlink(old_dir
);
1930 } else if (they_are_dirs
) {
1931 drop_nlink(old_dir
);
1935 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
1936 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
1937 old_dir
->i_ctime
= old_dir
->i_mtime
=
1938 new_dir
->i_ctime
= new_dir
->i_mtime
=
1939 inode
->i_ctime
= CURRENT_TIME
;
1943 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1947 struct inode
*inode
;
1948 struct page
*page
= NULL
;
1950 struct shmem_inode_info
*info
;
1952 len
= strlen(symname
) + 1;
1953 if (len
> PAGE_CACHE_SIZE
)
1954 return -ENAMETOOLONG
;
1956 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
1960 error
= security_inode_init_security(inode
, dir
, NULL
, NULL
,
1963 if (error
!= -EOPNOTSUPP
) {
1970 info
= SHMEM_I(inode
);
1971 inode
->i_size
= len
-1;
1972 if (len
<= (char *)inode
- (char *)info
) {
1974 memcpy(info
, symname
, len
);
1975 inode
->i_op
= &shmem_symlink_inline_operations
;
1977 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
1982 inode
->i_mapping
->a_ops
= &shmem_aops
;
1983 inode
->i_op
= &shmem_symlink_inode_operations
;
1984 kaddr
= kmap_atomic(page
, KM_USER0
);
1985 memcpy(kaddr
, symname
, len
);
1986 kunmap_atomic(kaddr
, KM_USER0
);
1987 set_page_dirty(page
);
1989 page_cache_release(page
);
1991 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1992 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1993 d_instantiate(dentry
, inode
);
1998 static void *shmem_follow_link_inline(struct dentry
*dentry
, struct nameidata
*nd
)
2000 nd_set_link(nd
, (char *)SHMEM_I(dentry
->d_inode
));
2004 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
2006 struct page
*page
= NULL
;
2007 int res
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
2008 nd_set_link(nd
, res
? ERR_PTR(res
) : kmap(page
));
2014 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2016 if (!IS_ERR(nd_get_link(nd
))) {
2017 struct page
*page
= cookie
;
2019 mark_page_accessed(page
);
2020 page_cache_release(page
);
2024 static const struct inode_operations shmem_symlink_inline_operations
= {
2025 .readlink
= generic_readlink
,
2026 .follow_link
= shmem_follow_link_inline
,
2029 static const struct inode_operations shmem_symlink_inode_operations
= {
2030 .readlink
= generic_readlink
,
2031 .follow_link
= shmem_follow_link
,
2032 .put_link
= shmem_put_link
,
2035 #ifdef CONFIG_TMPFS_POSIX_ACL
2037 * Superblocks without xattr inode operations will get security.* xattr
2038 * support from the VFS "for free". As soon as we have any other xattrs
2039 * like ACLs, we also need to implement the security.* handlers at
2040 * filesystem level, though.
2043 static size_t shmem_xattr_security_list(struct dentry
*dentry
, char *list
,
2044 size_t list_len
, const char *name
,
2045 size_t name_len
, int handler_flags
)
2047 return security_inode_listsecurity(dentry
->d_inode
, list
, list_len
);
2050 static int shmem_xattr_security_get(struct dentry
*dentry
, const char *name
,
2051 void *buffer
, size_t size
, int handler_flags
)
2053 if (strcmp(name
, "") == 0)
2055 return xattr_getsecurity(dentry
->d_inode
, name
, buffer
, size
);
2058 static int shmem_xattr_security_set(struct dentry
*dentry
, const char *name
,
2059 const void *value
, size_t size
, int flags
, int handler_flags
)
2061 if (strcmp(name
, "") == 0)
2063 return security_inode_setsecurity(dentry
->d_inode
, name
, value
,
2067 static const struct xattr_handler shmem_xattr_security_handler
= {
2068 .prefix
= XATTR_SECURITY_PREFIX
,
2069 .list
= shmem_xattr_security_list
,
2070 .get
= shmem_xattr_security_get
,
2071 .set
= shmem_xattr_security_set
,
2074 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2075 &generic_acl_access_handler
,
2076 &generic_acl_default_handler
,
2077 &shmem_xattr_security_handler
,
2082 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2084 return ERR_PTR(-ESTALE
);
2087 static int shmem_match(struct inode
*ino
, void *vfh
)
2091 inum
= (inum
<< 32) | fh
[1];
2092 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2095 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2096 struct fid
*fid
, int fh_len
, int fh_type
)
2098 struct inode
*inode
;
2099 struct dentry
*dentry
= NULL
;
2100 u64 inum
= fid
->raw
[2];
2101 inum
= (inum
<< 32) | fid
->raw
[1];
2106 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2107 shmem_match
, fid
->raw
);
2109 dentry
= d_find_alias(inode
);
2116 static int shmem_encode_fh(struct dentry
*dentry
, __u32
*fh
, int *len
,
2119 struct inode
*inode
= dentry
->d_inode
;
2124 if (hlist_unhashed(&inode
->i_hash
)) {
2125 /* Unfortunately insert_inode_hash is not idempotent,
2126 * so as we hash inodes here rather than at creation
2127 * time, we need a lock to ensure we only try
2130 static DEFINE_SPINLOCK(lock
);
2132 if (hlist_unhashed(&inode
->i_hash
))
2133 __insert_inode_hash(inode
,
2134 inode
->i_ino
+ inode
->i_generation
);
2138 fh
[0] = inode
->i_generation
;
2139 fh
[1] = inode
->i_ino
;
2140 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2146 static const struct export_operations shmem_export_ops
= {
2147 .get_parent
= shmem_get_parent
,
2148 .encode_fh
= shmem_encode_fh
,
2149 .fh_to_dentry
= shmem_fh_to_dentry
,
2152 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2155 char *this_char
, *value
, *rest
;
2157 while (options
!= NULL
) {
2158 this_char
= options
;
2161 * NUL-terminate this option: unfortunately,
2162 * mount options form a comma-separated list,
2163 * but mpol's nodelist may also contain commas.
2165 options
= strchr(options
, ',');
2166 if (options
== NULL
)
2169 if (!isdigit(*options
)) {
2176 if ((value
= strchr(this_char
,'=')) != NULL
) {
2180 "tmpfs: No value for mount option '%s'\n",
2185 if (!strcmp(this_char
,"size")) {
2186 unsigned long long size
;
2187 size
= memparse(value
,&rest
);
2189 size
<<= PAGE_SHIFT
;
2190 size
*= totalram_pages
;
2196 sbinfo
->max_blocks
=
2197 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2198 } else if (!strcmp(this_char
,"nr_blocks")) {
2199 sbinfo
->max_blocks
= memparse(value
, &rest
);
2202 } else if (!strcmp(this_char
,"nr_inodes")) {
2203 sbinfo
->max_inodes
= memparse(value
, &rest
);
2206 } else if (!strcmp(this_char
,"mode")) {
2209 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2212 } else if (!strcmp(this_char
,"uid")) {
2215 sbinfo
->uid
= simple_strtoul(value
, &rest
, 0);
2218 } else if (!strcmp(this_char
,"gid")) {
2221 sbinfo
->gid
= simple_strtoul(value
, &rest
, 0);
2224 } else if (!strcmp(this_char
,"mpol")) {
2225 if (mpol_parse_str(value
, &sbinfo
->mpol
, 1))
2228 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2236 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2242 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2244 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2245 struct shmem_sb_info config
= *sbinfo
;
2246 unsigned long blocks
;
2247 unsigned long inodes
;
2248 int error
= -EINVAL
;
2250 if (shmem_parse_options(data
, &config
, true))
2253 spin_lock(&sbinfo
->stat_lock
);
2254 blocks
= sbinfo
->max_blocks
- sbinfo
->free_blocks
;
2255 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2256 if (config
.max_blocks
< blocks
)
2258 if (config
.max_inodes
< inodes
)
2261 * Those tests also disallow limited->unlimited while any are in
2262 * use, so i_blocks will always be zero when max_blocks is zero;
2263 * but we must separately disallow unlimited->limited, because
2264 * in that case we have no record of how much is already in use.
2266 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2268 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2272 sbinfo
->max_blocks
= config
.max_blocks
;
2273 sbinfo
->free_blocks
= config
.max_blocks
- blocks
;
2274 sbinfo
->max_inodes
= config
.max_inodes
;
2275 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2277 mpol_put(sbinfo
->mpol
);
2278 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2280 spin_unlock(&sbinfo
->stat_lock
);
2284 static int shmem_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
2286 struct shmem_sb_info
*sbinfo
= SHMEM_SB(vfs
->mnt_sb
);
2288 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2289 seq_printf(seq
, ",size=%luk",
2290 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2291 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2292 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2293 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2294 seq_printf(seq
, ",mode=%03o", sbinfo
->mode
);
2295 if (sbinfo
->uid
!= 0)
2296 seq_printf(seq
, ",uid=%u", sbinfo
->uid
);
2297 if (sbinfo
->gid
!= 0)
2298 seq_printf(seq
, ",gid=%u", sbinfo
->gid
);
2299 shmem_show_mpol(seq
, sbinfo
->mpol
);
2302 #endif /* CONFIG_TMPFS */
2304 static void shmem_put_super(struct super_block
*sb
)
2306 kfree(sb
->s_fs_info
);
2307 sb
->s_fs_info
= NULL
;
2310 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
2312 struct inode
*inode
;
2313 struct dentry
*root
;
2314 struct shmem_sb_info
*sbinfo
;
2317 /* Round up to L1_CACHE_BYTES to resist false sharing */
2318 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
2319 L1_CACHE_BYTES
), GFP_KERNEL
);
2323 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
2324 sbinfo
->uid
= current_fsuid();
2325 sbinfo
->gid
= current_fsgid();
2326 sb
->s_fs_info
= sbinfo
;
2330 * Per default we only allow half of the physical ram per
2331 * tmpfs instance, limiting inodes to one per page of lowmem;
2332 * but the internal instance is left unlimited.
2334 if (!(sb
->s_flags
& MS_NOUSER
)) {
2335 sbinfo
->max_blocks
= shmem_default_max_blocks();
2336 sbinfo
->max_inodes
= shmem_default_max_inodes();
2337 if (shmem_parse_options(data
, sbinfo
, false)) {
2342 sb
->s_export_op
= &shmem_export_ops
;
2344 sb
->s_flags
|= MS_NOUSER
;
2347 spin_lock_init(&sbinfo
->stat_lock
);
2348 sbinfo
->free_blocks
= sbinfo
->max_blocks
;
2349 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
2351 sb
->s_maxbytes
= SHMEM_MAX_BYTES
;
2352 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
2353 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
2354 sb
->s_magic
= TMPFS_MAGIC
;
2355 sb
->s_op
= &shmem_ops
;
2356 sb
->s_time_gran
= 1;
2357 #ifdef CONFIG_TMPFS_POSIX_ACL
2358 sb
->s_xattr
= shmem_xattr_handlers
;
2359 sb
->s_flags
|= MS_POSIXACL
;
2362 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
2365 inode
->i_uid
= sbinfo
->uid
;
2366 inode
->i_gid
= sbinfo
->gid
;
2367 root
= d_alloc_root(inode
);
2376 shmem_put_super(sb
);
2380 static struct kmem_cache
*shmem_inode_cachep
;
2382 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
2384 struct shmem_inode_info
*p
;
2385 p
= (struct shmem_inode_info
*)kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
2388 return &p
->vfs_inode
;
2391 static void shmem_destroy_inode(struct inode
*inode
)
2393 if ((inode
->i_mode
& S_IFMT
) == S_IFREG
) {
2394 /* only struct inode is valid if it's an inline symlink */
2395 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
2397 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
2400 static void init_once(void *foo
)
2402 struct shmem_inode_info
*p
= (struct shmem_inode_info
*) foo
;
2404 inode_init_once(&p
->vfs_inode
);
2407 static int init_inodecache(void)
2409 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
2410 sizeof(struct shmem_inode_info
),
2411 0, SLAB_PANIC
, init_once
);
2415 static void destroy_inodecache(void)
2417 kmem_cache_destroy(shmem_inode_cachep
);
2420 static const struct address_space_operations shmem_aops
= {
2421 .writepage
= shmem_writepage
,
2422 .set_page_dirty
= __set_page_dirty_no_writeback
,
2424 .readpage
= shmem_readpage
,
2425 .write_begin
= shmem_write_begin
,
2426 .write_end
= shmem_write_end
,
2428 .migratepage
= migrate_page
,
2429 .error_remove_page
= generic_error_remove_page
,
2432 static const struct file_operations shmem_file_operations
= {
2435 .llseek
= generic_file_llseek
,
2436 .read
= do_sync_read
,
2437 .write
= do_sync_write
,
2438 .aio_read
= shmem_file_aio_read
,
2439 .aio_write
= generic_file_aio_write
,
2440 .fsync
= noop_fsync
,
2441 .splice_read
= generic_file_splice_read
,
2442 .splice_write
= generic_file_splice_write
,
2446 static const struct inode_operations shmem_inode_operations
= {
2447 .setattr
= shmem_notify_change
,
2448 .truncate_range
= shmem_truncate_range
,
2449 #ifdef CONFIG_TMPFS_POSIX_ACL
2450 .setxattr
= generic_setxattr
,
2451 .getxattr
= generic_getxattr
,
2452 .listxattr
= generic_listxattr
,
2453 .removexattr
= generic_removexattr
,
2454 .check_acl
= generic_check_acl
,
2459 static const struct inode_operations shmem_dir_inode_operations
= {
2461 .create
= shmem_create
,
2462 .lookup
= simple_lookup
,
2464 .unlink
= shmem_unlink
,
2465 .symlink
= shmem_symlink
,
2466 .mkdir
= shmem_mkdir
,
2467 .rmdir
= shmem_rmdir
,
2468 .mknod
= shmem_mknod
,
2469 .rename
= shmem_rename
,
2471 #ifdef CONFIG_TMPFS_POSIX_ACL
2472 .setattr
= shmem_notify_change
,
2473 .setxattr
= generic_setxattr
,
2474 .getxattr
= generic_getxattr
,
2475 .listxattr
= generic_listxattr
,
2476 .removexattr
= generic_removexattr
,
2477 .check_acl
= generic_check_acl
,
2481 static const struct inode_operations shmem_special_inode_operations
= {
2482 #ifdef CONFIG_TMPFS_POSIX_ACL
2483 .setattr
= shmem_notify_change
,
2484 .setxattr
= generic_setxattr
,
2485 .getxattr
= generic_getxattr
,
2486 .listxattr
= generic_listxattr
,
2487 .removexattr
= generic_removexattr
,
2488 .check_acl
= generic_check_acl
,
2492 static const struct super_operations shmem_ops
= {
2493 .alloc_inode
= shmem_alloc_inode
,
2494 .destroy_inode
= shmem_destroy_inode
,
2496 .statfs
= shmem_statfs
,
2497 .remount_fs
= shmem_remount_fs
,
2498 .show_options
= shmem_show_options
,
2500 .delete_inode
= shmem_delete_inode
,
2501 .drop_inode
= generic_delete_inode
,
2502 .put_super
= shmem_put_super
,
2505 static const struct vm_operations_struct shmem_vm_ops
= {
2506 .fault
= shmem_fault
,
2508 .set_policy
= shmem_set_policy
,
2509 .get_policy
= shmem_get_policy
,
2514 static int shmem_get_sb(struct file_system_type
*fs_type
,
2515 int flags
, const char *dev_name
, void *data
, struct vfsmount
*mnt
)
2517 return get_sb_nodev(fs_type
, flags
, data
, shmem_fill_super
, mnt
);
2520 static struct file_system_type tmpfs_fs_type
= {
2521 .owner
= THIS_MODULE
,
2523 .get_sb
= shmem_get_sb
,
2524 .kill_sb
= kill_litter_super
,
2527 int __init
init_tmpfs(void)
2531 error
= bdi_init(&shmem_backing_dev_info
);
2535 error
= init_inodecache();
2539 error
= register_filesystem(&tmpfs_fs_type
);
2541 printk(KERN_ERR
"Could not register tmpfs\n");
2545 shm_mnt
= vfs_kern_mount(&tmpfs_fs_type
, MS_NOUSER
,
2546 tmpfs_fs_type
.name
, NULL
);
2547 if (IS_ERR(shm_mnt
)) {
2548 error
= PTR_ERR(shm_mnt
);
2549 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
2555 unregister_filesystem(&tmpfs_fs_type
);
2557 destroy_inodecache();
2559 bdi_destroy(&shmem_backing_dev_info
);
2561 shm_mnt
= ERR_PTR(error
);
2565 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2567 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2568 * @inode: the inode to be searched
2569 * @pgoff: the offset to be searched
2570 * @pagep: the pointer for the found page to be stored
2571 * @ent: the pointer for the found swap entry to be stored
2573 * If a page is found, refcount of it is incremented. Callers should handle
2576 void mem_cgroup_get_shmem_target(struct inode
*inode
, pgoff_t pgoff
,
2577 struct page
**pagep
, swp_entry_t
*ent
)
2579 swp_entry_t entry
= { .val
= 0 }, *ptr
;
2580 struct page
*page
= NULL
;
2581 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2583 if ((pgoff
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
))
2586 spin_lock(&info
->lock
);
2587 ptr
= shmem_swp_entry(info
, pgoff
, NULL
);
2589 if (ptr
&& ptr
->val
) {
2590 entry
.val
= ptr
->val
;
2591 page
= find_get_page(&swapper_space
, entry
.val
);
2594 page
= find_get_page(inode
->i_mapping
, pgoff
);
2596 shmem_swp_unmap(ptr
);
2597 spin_unlock(&info
->lock
);
2604 #else /* !CONFIG_SHMEM */
2607 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2609 * This is intended for small system where the benefits of the full
2610 * shmem code (swap-backed and resource-limited) are outweighed by
2611 * their complexity. On systems without swap this code should be
2612 * effectively equivalent, but much lighter weight.
2615 #include <linux/ramfs.h>
2617 static struct file_system_type tmpfs_fs_type
= {
2619 .get_sb
= ramfs_get_sb
,
2620 .kill_sb
= kill_litter_super
,
2623 int __init
init_tmpfs(void)
2625 BUG_ON(register_filesystem(&tmpfs_fs_type
) != 0);
2627 shm_mnt
= kern_mount(&tmpfs_fs_type
);
2628 BUG_ON(IS_ERR(shm_mnt
));
2633 int shmem_unuse(swp_entry_t entry
, struct page
*page
)
2638 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2643 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2645 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2646 * @inode: the inode to be searched
2647 * @pgoff: the offset to be searched
2648 * @pagep: the pointer for the found page to be stored
2649 * @ent: the pointer for the found swap entry to be stored
2651 * If a page is found, refcount of it is incremented. Callers should handle
2654 void mem_cgroup_get_shmem_target(struct inode
*inode
, pgoff_t pgoff
,
2655 struct page
**pagep
, swp_entry_t
*ent
)
2657 struct page
*page
= NULL
;
2659 if ((pgoff
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
))
2661 page
= find_get_page(inode
->i_mapping
, pgoff
);
2664 *ent
= (swp_entry_t
){ .val
= 0 };
2668 #define shmem_vm_ops generic_file_vm_ops
2669 #define shmem_file_operations ramfs_file_operations
2670 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2671 #define shmem_acct_size(flags, size) 0
2672 #define shmem_unacct_size(flags, size) do {} while (0)
2673 #define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
2675 #endif /* CONFIG_SHMEM */
2680 * shmem_file_setup - get an unlinked file living in tmpfs
2681 * @name: name for dentry (to be seen in /proc/<pid>/maps
2682 * @size: size to be set for the file
2683 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2685 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
2689 struct inode
*inode
;
2691 struct dentry
*root
;
2694 if (IS_ERR(shm_mnt
))
2695 return (void *)shm_mnt
;
2697 if (size
< 0 || size
> SHMEM_MAX_BYTES
)
2698 return ERR_PTR(-EINVAL
);
2700 if (shmem_acct_size(flags
, size
))
2701 return ERR_PTR(-ENOMEM
);
2705 this.len
= strlen(name
);
2706 this.hash
= 0; /* will go */
2707 root
= shm_mnt
->mnt_root
;
2708 path
.dentry
= d_alloc(root
, &this);
2711 path
.mnt
= mntget(shm_mnt
);
2714 inode
= shmem_get_inode(root
->d_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
2718 d_instantiate(path
.dentry
, inode
);
2719 inode
->i_size
= size
;
2720 inode
->i_nlink
= 0; /* It is unlinked */
2722 error
= ramfs_nommu_expand_for_mapping(inode
, size
);
2728 file
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
2729 &shmem_file_operations
);
2738 shmem_unacct_size(flags
, size
);
2739 return ERR_PTR(error
);
2741 EXPORT_SYMBOL_GPL(shmem_file_setup
);
2744 * shmem_zero_setup - setup a shared anonymous mapping
2745 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2747 int shmem_zero_setup(struct vm_area_struct
*vma
)
2750 loff_t size
= vma
->vm_end
- vma
->vm_start
;
2752 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
);
2754 return PTR_ERR(file
);
2758 vma
->vm_file
= file
;
2759 vma
->vm_ops
= &shmem_vm_ops
;