2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/uio.h>
36 static struct vfsmount
*shm_mnt
;
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
45 #include <linux/xattr.h>
46 #include <linux/exportfs.h>
47 #include <linux/posix_acl.h>
48 #include <linux/posix_acl_xattr.h>
49 #include <linux/mman.h>
50 #include <linux/string.h>
51 #include <linux/slab.h>
52 #include <linux/backing-dev.h>
53 #include <linux/shmem_fs.h>
54 #include <linux/writeback.h>
55 #include <linux/blkdev.h>
56 #include <linux/pagevec.h>
57 #include <linux/percpu_counter.h>
58 #include <linux/falloc.h>
59 #include <linux/splice.h>
60 #include <linux/security.h>
61 #include <linux/swapops.h>
62 #include <linux/mempolicy.h>
63 #include <linux/namei.h>
64 #include <linux/ctype.h>
65 #include <linux/migrate.h>
66 #include <linux/highmem.h>
67 #include <linux/seq_file.h>
68 #include <linux/magic.h>
69 #include <linux/syscalls.h>
70 #include <linux/fcntl.h>
71 #include <uapi/linux/memfd.h>
73 #include <asm/uaccess.h>
74 #include <asm/pgtable.h>
78 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
79 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
81 /* Pretend that each entry is of this size in directory's i_size */
82 #define BOGO_DIRENT_SIZE 20
84 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
85 #define SHORT_SYMLINK_LEN 128
88 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
89 * inode->i_private (with i_mutex making sure that it has only one user at
90 * a time): we would prefer not to enlarge the shmem inode just for that.
93 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
94 pgoff_t start
; /* start of range currently being fallocated */
95 pgoff_t next
; /* the next page offset to be fallocated */
96 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
97 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
100 /* Flag allocation requirements to shmem_getpage */
102 SGP_READ
, /* don't exceed i_size, don't allocate page */
103 SGP_CACHE
, /* don't exceed i_size, may allocate page */
104 SGP_DIRTY
, /* like SGP_CACHE, but set new page dirty */
105 SGP_WRITE
, /* may exceed i_size, may allocate !Uptodate page */
106 SGP_FALLOC
, /* like SGP_WRITE, but make existing page Uptodate */
110 static unsigned long shmem_default_max_blocks(void)
112 return totalram_pages
/ 2;
115 static int shmem_default_max_inodes(void)
120 ul
= min3(ul
, totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
125 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
126 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
127 struct shmem_inode_info
*info
, pgoff_t index
);
128 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
129 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
);
131 static inline int shmem_getpage(struct inode
*inode
, pgoff_t index
,
132 struct page
**pagep
, enum sgp_type sgp
, int *fault_type
)
134 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
135 mapping_gfp_mask(inode
->i_mapping
), fault_type
);
138 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
140 return sb
->s_fs_info
;
144 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
145 * for shared memory and for shared anonymous (/dev/zero) mappings
146 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
147 * consistent with the pre-accounting of private mappings ...
149 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
151 return (flags
& VM_NORESERVE
) ?
152 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
155 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
157 if (!(flags
& VM_NORESERVE
))
158 vm_unacct_memory(VM_ACCT(size
));
161 static inline int shmem_reacct_size(unsigned long flags
,
162 loff_t oldsize
, loff_t newsize
)
164 if (!(flags
& VM_NORESERVE
)) {
165 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
166 return security_vm_enough_memory_mm(current
->mm
,
167 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
168 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
169 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
175 * ... whereas tmpfs objects are accounted incrementally as
176 * pages are allocated, in order to allow huge sparse files.
177 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
178 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
180 static inline int shmem_acct_block(unsigned long flags
)
182 return (flags
& VM_NORESERVE
) ?
183 security_vm_enough_memory_mm(current
->mm
, VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
186 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
188 if (flags
& VM_NORESERVE
)
189 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
192 static const struct super_operations shmem_ops
;
193 static const struct address_space_operations shmem_aops
;
194 static const struct file_operations shmem_file_operations
;
195 static const struct inode_operations shmem_inode_operations
;
196 static const struct inode_operations shmem_dir_inode_operations
;
197 static const struct inode_operations shmem_special_inode_operations
;
198 static const struct vm_operations_struct shmem_vm_ops
;
200 static LIST_HEAD(shmem_swaplist
);
201 static DEFINE_MUTEX(shmem_swaplist_mutex
);
203 static int shmem_reserve_inode(struct super_block
*sb
)
205 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
206 if (sbinfo
->max_inodes
) {
207 spin_lock(&sbinfo
->stat_lock
);
208 if (!sbinfo
->free_inodes
) {
209 spin_unlock(&sbinfo
->stat_lock
);
212 sbinfo
->free_inodes
--;
213 spin_unlock(&sbinfo
->stat_lock
);
218 static void shmem_free_inode(struct super_block
*sb
)
220 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
221 if (sbinfo
->max_inodes
) {
222 spin_lock(&sbinfo
->stat_lock
);
223 sbinfo
->free_inodes
++;
224 spin_unlock(&sbinfo
->stat_lock
);
229 * shmem_recalc_inode - recalculate the block usage of an inode
230 * @inode: inode to recalc
232 * We have to calculate the free blocks since the mm can drop
233 * undirtied hole pages behind our back.
235 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
236 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
238 * It has to be called with the spinlock held.
240 static void shmem_recalc_inode(struct inode
*inode
)
242 struct shmem_inode_info
*info
= SHMEM_I(inode
);
245 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
247 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
248 if (sbinfo
->max_blocks
)
249 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
250 info
->alloced
-= freed
;
251 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
252 shmem_unacct_blocks(info
->flags
, freed
);
257 * Replace item expected in radix tree by a new item, while holding tree lock.
259 static int shmem_radix_tree_replace(struct address_space
*mapping
,
260 pgoff_t index
, void *expected
, void *replacement
)
265 VM_BUG_ON(!expected
);
266 VM_BUG_ON(!replacement
);
267 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
270 item
= radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
);
271 if (item
!= expected
)
273 radix_tree_replace_slot(pslot
, replacement
);
278 * Sometimes, before we decide whether to proceed or to fail, we must check
279 * that an entry was not already brought back from swap by a racing thread.
281 * Checking page is not enough: by the time a SwapCache page is locked, it
282 * might be reused, and again be SwapCache, using the same swap as before.
284 static bool shmem_confirm_swap(struct address_space
*mapping
,
285 pgoff_t index
, swp_entry_t swap
)
290 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
292 return item
== swp_to_radix_entry(swap
);
296 * Like add_to_page_cache_locked, but error if expected item has gone.
298 static int shmem_add_to_page_cache(struct page
*page
,
299 struct address_space
*mapping
,
300 pgoff_t index
, void *expected
)
304 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
305 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
307 page_cache_get(page
);
308 page
->mapping
= mapping
;
311 spin_lock_irq(&mapping
->tree_lock
);
313 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
315 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
319 __inc_zone_page_state(page
, NR_FILE_PAGES
);
320 __inc_zone_page_state(page
, NR_SHMEM
);
321 spin_unlock_irq(&mapping
->tree_lock
);
323 page
->mapping
= NULL
;
324 spin_unlock_irq(&mapping
->tree_lock
);
325 page_cache_release(page
);
331 * Like delete_from_page_cache, but substitutes swap for page.
333 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
335 struct address_space
*mapping
= page
->mapping
;
338 spin_lock_irq(&mapping
->tree_lock
);
339 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
340 page
->mapping
= NULL
;
342 __dec_zone_page_state(page
, NR_FILE_PAGES
);
343 __dec_zone_page_state(page
, NR_SHMEM
);
344 spin_unlock_irq(&mapping
->tree_lock
);
345 page_cache_release(page
);
350 * Remove swap entry from radix tree, free the swap and its page cache.
352 static int shmem_free_swap(struct address_space
*mapping
,
353 pgoff_t index
, void *radswap
)
357 spin_lock_irq(&mapping
->tree_lock
);
358 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
359 spin_unlock_irq(&mapping
->tree_lock
);
362 free_swap_and_cache(radix_to_swp_entry(radswap
));
367 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
369 void shmem_unlock_mapping(struct address_space
*mapping
)
372 pgoff_t indices
[PAGEVEC_SIZE
];
375 pagevec_init(&pvec
, 0);
377 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
379 while (!mapping_unevictable(mapping
)) {
381 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
382 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
384 pvec
.nr
= find_get_entries(mapping
, index
,
385 PAGEVEC_SIZE
, pvec
.pages
, indices
);
388 index
= indices
[pvec
.nr
- 1] + 1;
389 pagevec_remove_exceptionals(&pvec
);
390 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
391 pagevec_release(&pvec
);
397 * Remove range of pages and swap entries from radix tree, and free them.
398 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
400 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
403 struct address_space
*mapping
= inode
->i_mapping
;
404 struct shmem_inode_info
*info
= SHMEM_I(inode
);
405 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
406 pgoff_t end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
407 unsigned int partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
408 unsigned int partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
410 pgoff_t indices
[PAGEVEC_SIZE
];
411 long nr_swaps_freed
= 0;
416 end
= -1; /* unsigned, so actually very big */
418 pagevec_init(&pvec
, 0);
420 while (index
< end
) {
421 pvec
.nr
= find_get_entries(mapping
, index
,
422 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
423 pvec
.pages
, indices
);
426 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
427 struct page
*page
= pvec
.pages
[i
];
433 if (radix_tree_exceptional_entry(page
)) {
436 nr_swaps_freed
+= !shmem_free_swap(mapping
,
441 if (!trylock_page(page
))
443 if (!unfalloc
|| !PageUptodate(page
)) {
444 if (page
->mapping
== mapping
) {
445 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
446 truncate_inode_page(mapping
, page
);
451 pagevec_remove_exceptionals(&pvec
);
452 pagevec_release(&pvec
);
458 struct page
*page
= NULL
;
459 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
461 unsigned int top
= PAGE_CACHE_SIZE
;
466 zero_user_segment(page
, partial_start
, top
);
467 set_page_dirty(page
);
469 page_cache_release(page
);
473 struct page
*page
= NULL
;
474 shmem_getpage(inode
, end
, &page
, SGP_READ
, NULL
);
476 zero_user_segment(page
, 0, partial_end
);
477 set_page_dirty(page
);
479 page_cache_release(page
);
486 while (index
< end
) {
489 pvec
.nr
= find_get_entries(mapping
, index
,
490 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
491 pvec
.pages
, indices
);
493 /* If all gone or hole-punch or unfalloc, we're done */
494 if (index
== start
|| end
!= -1)
496 /* But if truncating, restart to make sure all gone */
500 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
501 struct page
*page
= pvec
.pages
[i
];
507 if (radix_tree_exceptional_entry(page
)) {
510 if (shmem_free_swap(mapping
, index
, page
)) {
511 /* Swap was replaced by page: retry */
520 if (!unfalloc
|| !PageUptodate(page
)) {
521 if (page
->mapping
== mapping
) {
522 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
523 truncate_inode_page(mapping
, page
);
525 /* Page was replaced by swap: retry */
533 pagevec_remove_exceptionals(&pvec
);
534 pagevec_release(&pvec
);
538 spin_lock(&info
->lock
);
539 info
->swapped
-= nr_swaps_freed
;
540 shmem_recalc_inode(inode
);
541 spin_unlock(&info
->lock
);
544 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
546 shmem_undo_range(inode
, lstart
, lend
, false);
547 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
549 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
551 static int shmem_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
554 struct inode
*inode
= dentry
->d_inode
;
555 struct shmem_inode_info
*info
= SHMEM_I(inode
);
557 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
558 spin_lock(&info
->lock
);
559 shmem_recalc_inode(inode
);
560 spin_unlock(&info
->lock
);
562 generic_fillattr(inode
, stat
);
566 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
568 struct inode
*inode
= d_inode(dentry
);
569 struct shmem_inode_info
*info
= SHMEM_I(inode
);
572 error
= inode_change_ok(inode
, attr
);
576 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
577 loff_t oldsize
= inode
->i_size
;
578 loff_t newsize
= attr
->ia_size
;
580 /* protected by i_mutex */
581 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
582 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
585 if (newsize
!= oldsize
) {
586 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
590 i_size_write(inode
, newsize
);
591 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
593 if (newsize
<= oldsize
) {
594 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
595 if (oldsize
> holebegin
)
596 unmap_mapping_range(inode
->i_mapping
,
599 shmem_truncate_range(inode
,
600 newsize
, (loff_t
)-1);
601 /* unmap again to remove racily COWed private pages */
602 if (oldsize
> holebegin
)
603 unmap_mapping_range(inode
->i_mapping
,
608 setattr_copy(inode
, attr
);
609 if (attr
->ia_valid
& ATTR_MODE
)
610 error
= posix_acl_chmod(inode
, inode
->i_mode
);
614 static void shmem_evict_inode(struct inode
*inode
)
616 struct shmem_inode_info
*info
= SHMEM_I(inode
);
617 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
619 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
620 shmem_unacct_size(info
->flags
, inode
->i_size
);
622 shmem_truncate_range(inode
, 0, (loff_t
)-1);
623 if (!list_empty(&info
->swaplist
)) {
624 mutex_lock(&shmem_swaplist_mutex
);
625 list_del_init(&info
->swaplist
);
626 mutex_unlock(&shmem_swaplist_mutex
);
630 simple_xattrs_free(&info
->xattrs
);
631 WARN_ON(inode
->i_blocks
);
632 if (!sbinfo
->idr_nouse
&& inode
->i_ino
) {
633 mutex_lock(&sbinfo
->idr_lock
);
634 idr_remove(&sbinfo
->idr
, inode
->i_ino
);
635 mutex_unlock(&sbinfo
->idr_lock
);
637 shmem_free_inode(inode
->i_sb
);
642 * If swap found in inode, free it and move page from swapcache to filecache.
644 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
645 swp_entry_t swap
, struct page
**pagep
)
647 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
653 radswap
= swp_to_radix_entry(swap
);
654 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
656 return -EAGAIN
; /* tell shmem_unuse we found nothing */
659 * Move _head_ to start search for next from here.
660 * But be careful: shmem_evict_inode checks list_empty without taking
661 * mutex, and there's an instant in list_move_tail when info->swaplist
662 * would appear empty, if it were the only one on shmem_swaplist.
664 if (shmem_swaplist
.next
!= &info
->swaplist
)
665 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
667 gfp
= mapping_gfp_mask(mapping
);
668 if (shmem_should_replace_page(*pagep
, gfp
)) {
669 mutex_unlock(&shmem_swaplist_mutex
);
670 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
671 mutex_lock(&shmem_swaplist_mutex
);
673 * We needed to drop mutex to make that restrictive page
674 * allocation, but the inode might have been freed while we
675 * dropped it: although a racing shmem_evict_inode() cannot
676 * complete without emptying the radix_tree, our page lock
677 * on this swapcache page is not enough to prevent that -
678 * free_swap_and_cache() of our swap entry will only
679 * trylock_page(), removing swap from radix_tree whatever.
681 * We must not proceed to shmem_add_to_page_cache() if the
682 * inode has been freed, but of course we cannot rely on
683 * inode or mapping or info to check that. However, we can
684 * safely check if our swap entry is still in use (and here
685 * it can't have got reused for another page): if it's still
686 * in use, then the inode cannot have been freed yet, and we
687 * can safely proceed (if it's no longer in use, that tells
688 * nothing about the inode, but we don't need to unuse swap).
690 if (!page_swapcount(*pagep
))
695 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
696 * but also to hold up shmem_evict_inode(): so inode cannot be freed
697 * beneath us (pagelock doesn't help until the page is in pagecache).
700 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
702 if (error
!= -ENOMEM
) {
704 * Truncation and eviction use free_swap_and_cache(), which
705 * only does trylock page: if we raced, best clean up here.
707 delete_from_swap_cache(*pagep
);
708 set_page_dirty(*pagep
);
710 spin_lock(&info
->lock
);
712 spin_unlock(&info
->lock
);
720 * Search through swapped inodes to find and replace swap by page.
722 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
724 struct list_head
*this, *next
;
725 struct shmem_inode_info
*info
;
726 struct mem_cgroup
*memcg
;
730 * There's a faint possibility that swap page was replaced before
731 * caller locked it: caller will come back later with the right page.
733 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
737 * Charge page using GFP_KERNEL while we can wait, before taking
738 * the shmem_swaplist_mutex which might hold up shmem_writepage().
739 * Charged back to the user (not to caller) when swap account is used.
741 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
);
744 /* No radix_tree_preload: swap entry keeps a place for page in tree */
747 mutex_lock(&shmem_swaplist_mutex
);
748 list_for_each_safe(this, next
, &shmem_swaplist
) {
749 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
751 error
= shmem_unuse_inode(info
, swap
, &page
);
753 list_del_init(&info
->swaplist
);
755 if (error
!= -EAGAIN
)
757 /* found nothing in this: move on to search the next */
759 mutex_unlock(&shmem_swaplist_mutex
);
762 if (error
!= -ENOMEM
)
764 mem_cgroup_cancel_charge(page
, memcg
);
766 mem_cgroup_commit_charge(page
, memcg
, true);
769 page_cache_release(page
);
774 * Move the page from the page cache to the swap cache.
776 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
778 struct shmem_inode_info
*info
;
779 struct address_space
*mapping
;
784 BUG_ON(!PageLocked(page
));
785 mapping
= page
->mapping
;
787 inode
= mapping
->host
;
788 info
= SHMEM_I(inode
);
789 if (info
->flags
& VM_LOCKED
)
791 if (!total_swap_pages
)
795 * Our capabilities prevent regular writeback or sync from ever calling
796 * shmem_writepage; but a stacking filesystem might use ->writepage of
797 * its underlying filesystem, in which case tmpfs should write out to
798 * swap only in response to memory pressure, and not for the writeback
801 if (!wbc
->for_reclaim
) {
802 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
807 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
808 * value into swapfile.c, the only way we can correctly account for a
809 * fallocated page arriving here is now to initialize it and write it.
811 * That's okay for a page already fallocated earlier, but if we have
812 * not yet completed the fallocation, then (a) we want to keep track
813 * of this page in case we have to undo it, and (b) it may not be a
814 * good idea to continue anyway, once we're pushing into swap. So
815 * reactivate the page, and let shmem_fallocate() quit when too many.
817 if (!PageUptodate(page
)) {
818 if (inode
->i_private
) {
819 struct shmem_falloc
*shmem_falloc
;
820 spin_lock(&inode
->i_lock
);
821 shmem_falloc
= inode
->i_private
;
823 !shmem_falloc
->waitq
&&
824 index
>= shmem_falloc
->start
&&
825 index
< shmem_falloc
->next
)
826 shmem_falloc
->nr_unswapped
++;
829 spin_unlock(&inode
->i_lock
);
833 clear_highpage(page
);
834 flush_dcache_page(page
);
835 SetPageUptodate(page
);
838 swap
= get_swap_page();
843 * Add inode to shmem_unuse()'s list of swapped-out inodes,
844 * if it's not already there. Do it now before the page is
845 * moved to swap cache, when its pagelock no longer protects
846 * the inode from eviction. But don't unlock the mutex until
847 * we've incremented swapped, because shmem_unuse_inode() will
848 * prune a !swapped inode from the swaplist under this mutex.
850 mutex_lock(&shmem_swaplist_mutex
);
851 if (list_empty(&info
->swaplist
))
852 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
854 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
855 spin_lock(&info
->lock
);
856 shmem_recalc_inode(inode
);
858 spin_unlock(&info
->lock
);
860 swap_shmem_alloc(swap
);
861 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
863 mutex_unlock(&shmem_swaplist_mutex
);
864 BUG_ON(page_mapped(page
));
865 swap_writepage(page
, wbc
);
869 mutex_unlock(&shmem_swaplist_mutex
);
870 swapcache_free(swap
);
872 set_page_dirty(page
);
873 if (wbc
->for_reclaim
)
874 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
881 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
885 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
886 return; /* show nothing */
888 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
890 seq_printf(seq
, ",mpol=%s", buffer
);
893 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
895 struct mempolicy
*mpol
= NULL
;
897 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
900 spin_unlock(&sbinfo
->stat_lock
);
904 #endif /* CONFIG_TMPFS */
906 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
907 struct shmem_inode_info
*info
, pgoff_t index
)
909 struct vm_area_struct pvma
;
912 /* Create a pseudo vma that just contains the policy */
914 /* Bias interleave by inode number to distribute better across nodes */
915 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
917 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
919 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
921 /* Drop reference taken by mpol_shared_policy_lookup() */
922 mpol_cond_put(pvma
.vm_policy
);
927 static struct page
*shmem_alloc_page(gfp_t gfp
,
928 struct shmem_inode_info
*info
, pgoff_t index
)
930 struct vm_area_struct pvma
;
933 /* Create a pseudo vma that just contains the policy */
935 /* Bias interleave by inode number to distribute better across nodes */
936 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
938 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
940 page
= alloc_page_vma(gfp
, &pvma
, 0);
942 /* Drop reference taken by mpol_shared_policy_lookup() */
943 mpol_cond_put(pvma
.vm_policy
);
947 #else /* !CONFIG_NUMA */
949 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
952 #endif /* CONFIG_TMPFS */
954 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
955 struct shmem_inode_info
*info
, pgoff_t index
)
957 return swapin_readahead(swap
, gfp
, NULL
, 0);
960 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
961 struct shmem_inode_info
*info
, pgoff_t index
)
963 return alloc_page(gfp
);
965 #endif /* CONFIG_NUMA */
967 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
968 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
975 * When a page is moved from swapcache to shmem filecache (either by the
976 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
977 * shmem_unuse_inode()), it may have been read in earlier from swap, in
978 * ignorance of the mapping it belongs to. If that mapping has special
979 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
980 * we may need to copy to a suitable page before moving to filecache.
982 * In a future release, this may well be extended to respect cpuset and
983 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
984 * but for now it is a simple matter of zone.
986 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
988 return page_zonenum(page
) > gfp_zone(gfp
);
991 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
992 struct shmem_inode_info
*info
, pgoff_t index
)
994 struct page
*oldpage
, *newpage
;
995 struct address_space
*swap_mapping
;
1000 swap_index
= page_private(oldpage
);
1001 swap_mapping
= page_mapping(oldpage
);
1004 * We have arrived here because our zones are constrained, so don't
1005 * limit chance of success by further cpuset and node constraints.
1007 gfp
&= ~GFP_CONSTRAINT_MASK
;
1008 newpage
= shmem_alloc_page(gfp
, info
, index
);
1012 page_cache_get(newpage
);
1013 copy_highpage(newpage
, oldpage
);
1014 flush_dcache_page(newpage
);
1016 __set_page_locked(newpage
);
1017 SetPageUptodate(newpage
);
1018 SetPageSwapBacked(newpage
);
1019 set_page_private(newpage
, swap_index
);
1020 SetPageSwapCache(newpage
);
1023 * Our caller will very soon move newpage out of swapcache, but it's
1024 * a nice clean interface for us to replace oldpage by newpage there.
1026 spin_lock_irq(&swap_mapping
->tree_lock
);
1027 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1030 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
1031 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
1033 spin_unlock_irq(&swap_mapping
->tree_lock
);
1035 if (unlikely(error
)) {
1037 * Is this possible? I think not, now that our callers check
1038 * both PageSwapCache and page_private after getting page lock;
1039 * but be defensive. Reverse old to newpage for clear and free.
1043 mem_cgroup_replace_page(oldpage
, newpage
);
1044 lru_cache_add_anon(newpage
);
1048 ClearPageSwapCache(oldpage
);
1049 set_page_private(oldpage
, 0);
1051 unlock_page(oldpage
);
1052 page_cache_release(oldpage
);
1053 page_cache_release(oldpage
);
1058 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1060 * If we allocate a new one we do not mark it dirty. That's up to the
1061 * vm. If we swap it in we mark it dirty since we also free the swap
1062 * entry since a page cannot live in both the swap and page cache
1064 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1065 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
1067 struct address_space
*mapping
= inode
->i_mapping
;
1068 struct shmem_inode_info
*info
;
1069 struct shmem_sb_info
*sbinfo
;
1070 struct mem_cgroup
*memcg
;
1077 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
1081 page
= find_lock_entry(mapping
, index
);
1082 if (radix_tree_exceptional_entry(page
)) {
1083 swap
= radix_to_swp_entry(page
);
1087 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1088 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1093 if (page
&& sgp
== SGP_WRITE
)
1094 mark_page_accessed(page
);
1096 /* fallocated page? */
1097 if (page
&& !PageUptodate(page
)) {
1098 if (sgp
!= SGP_READ
)
1101 page_cache_release(page
);
1104 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1110 * Fast cache lookup did not find it:
1111 * bring it back from swap or allocate.
1113 info
= SHMEM_I(inode
);
1114 sbinfo
= SHMEM_SB(inode
->i_sb
);
1117 /* Look it up and read it in.. */
1118 page
= lookup_swap_cache(swap
);
1120 /* here we actually do the io */
1122 *fault_type
|= VM_FAULT_MAJOR
;
1123 page
= shmem_swapin(swap
, gfp
, info
, index
);
1130 /* We have to do this with page locked to prevent races */
1132 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1133 !shmem_confirm_swap(mapping
, index
, swap
)) {
1134 error
= -EEXIST
; /* try again */
1137 if (!PageUptodate(page
)) {
1141 wait_on_page_writeback(page
);
1143 if (shmem_should_replace_page(page
, gfp
)) {
1144 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1149 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
);
1151 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1152 swp_to_radix_entry(swap
));
1154 * We already confirmed swap under page lock, and make
1155 * no memory allocation here, so usually no possibility
1156 * of error; but free_swap_and_cache() only trylocks a
1157 * page, so it is just possible that the entry has been
1158 * truncated or holepunched since swap was confirmed.
1159 * shmem_undo_range() will have done some of the
1160 * unaccounting, now delete_from_swap_cache() will do
1162 * Reset swap.val? No, leave it so "failed" goes back to
1163 * "repeat": reading a hole and writing should succeed.
1166 mem_cgroup_cancel_charge(page
, memcg
);
1167 delete_from_swap_cache(page
);
1173 mem_cgroup_commit_charge(page
, memcg
, true);
1175 spin_lock(&info
->lock
);
1177 shmem_recalc_inode(inode
);
1178 spin_unlock(&info
->lock
);
1180 if (sgp
== SGP_WRITE
)
1181 mark_page_accessed(page
);
1183 delete_from_swap_cache(page
);
1184 set_page_dirty(page
);
1188 if (shmem_acct_block(info
->flags
)) {
1192 if (sbinfo
->max_blocks
) {
1193 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1194 sbinfo
->max_blocks
) >= 0) {
1198 percpu_counter_inc(&sbinfo
->used_blocks
);
1201 page
= shmem_alloc_page(gfp
, info
, index
);
1207 __SetPageSwapBacked(page
);
1208 __set_page_locked(page
);
1209 if (sgp
== SGP_WRITE
)
1210 __SetPageReferenced(page
);
1212 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
);
1215 error
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
1217 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1219 radix_tree_preload_end();
1222 mem_cgroup_cancel_charge(page
, memcg
);
1225 mem_cgroup_commit_charge(page
, memcg
, false);
1226 lru_cache_add_anon(page
);
1228 spin_lock(&info
->lock
);
1230 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1231 shmem_recalc_inode(inode
);
1232 spin_unlock(&info
->lock
);
1236 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1238 if (sgp
== SGP_FALLOC
)
1242 * Let SGP_WRITE caller clear ends if write does not fill page;
1243 * but SGP_FALLOC on a page fallocated earlier must initialize
1244 * it now, lest undo on failure cancel our earlier guarantee.
1246 if (sgp
!= SGP_WRITE
) {
1247 clear_highpage(page
);
1248 flush_dcache_page(page
);
1249 SetPageUptodate(page
);
1251 if (sgp
== SGP_DIRTY
)
1252 set_page_dirty(page
);
1255 /* Perhaps the file has been truncated since we checked */
1256 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1257 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1259 ClearPageDirty(page
);
1260 delete_from_page_cache(page
);
1261 spin_lock(&info
->lock
);
1262 shmem_recalc_inode(inode
);
1263 spin_unlock(&info
->lock
);
1275 if (sbinfo
->max_blocks
)
1276 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1278 shmem_unacct_blocks(info
->flags
, 1);
1280 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1285 page_cache_release(page
);
1287 if (error
== -ENOSPC
&& !once
++) {
1288 info
= SHMEM_I(inode
);
1289 spin_lock(&info
->lock
);
1290 shmem_recalc_inode(inode
);
1291 spin_unlock(&info
->lock
);
1294 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1299 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1301 struct inode
*inode
= file_inode(vma
->vm_file
);
1303 int ret
= VM_FAULT_LOCKED
;
1306 * Trinity finds that probing a hole which tmpfs is punching can
1307 * prevent the hole-punch from ever completing: which in turn
1308 * locks writers out with its hold on i_mutex. So refrain from
1309 * faulting pages into the hole while it's being punched. Although
1310 * shmem_undo_range() does remove the additions, it may be unable to
1311 * keep up, as each new page needs its own unmap_mapping_range() call,
1312 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1314 * It does not matter if we sometimes reach this check just before the
1315 * hole-punch begins, so that one fault then races with the punch:
1316 * we just need to make racing faults a rare case.
1318 * The implementation below would be much simpler if we just used a
1319 * standard mutex or completion: but we cannot take i_mutex in fault,
1320 * and bloating every shmem inode for this unlikely case would be sad.
1322 if (unlikely(inode
->i_private
)) {
1323 struct shmem_falloc
*shmem_falloc
;
1325 spin_lock(&inode
->i_lock
);
1326 shmem_falloc
= inode
->i_private
;
1328 shmem_falloc
->waitq
&&
1329 vmf
->pgoff
>= shmem_falloc
->start
&&
1330 vmf
->pgoff
< shmem_falloc
->next
) {
1331 wait_queue_head_t
*shmem_falloc_waitq
;
1332 DEFINE_WAIT(shmem_fault_wait
);
1334 ret
= VM_FAULT_NOPAGE
;
1335 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1336 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1337 /* It's polite to up mmap_sem if we can */
1338 up_read(&vma
->vm_mm
->mmap_sem
);
1339 ret
= VM_FAULT_RETRY
;
1342 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1343 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1344 TASK_UNINTERRUPTIBLE
);
1345 spin_unlock(&inode
->i_lock
);
1349 * shmem_falloc_waitq points into the shmem_fallocate()
1350 * stack of the hole-punching task: shmem_falloc_waitq
1351 * is usually invalid by the time we reach here, but
1352 * finish_wait() does not dereference it in that case;
1353 * though i_lock needed lest racing with wake_up_all().
1355 spin_lock(&inode
->i_lock
);
1356 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1357 spin_unlock(&inode
->i_lock
);
1360 spin_unlock(&inode
->i_lock
);
1363 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1365 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1367 if (ret
& VM_FAULT_MAJOR
) {
1368 count_vm_event(PGMAJFAULT
);
1369 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1375 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1377 struct inode
*inode
= file_inode(vma
->vm_file
);
1378 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1381 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1384 struct inode
*inode
= file_inode(vma
->vm_file
);
1387 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1388 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1392 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1394 struct inode
*inode
= file_inode(file
);
1395 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1396 int retval
= -ENOMEM
;
1398 spin_lock(&info
->lock
);
1399 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1400 if (!user_shm_lock(inode
->i_size
, user
))
1402 info
->flags
|= VM_LOCKED
;
1403 mapping_set_unevictable(file
->f_mapping
);
1405 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1406 user_shm_unlock(inode
->i_size
, user
);
1407 info
->flags
&= ~VM_LOCKED
;
1408 mapping_clear_unevictable(file
->f_mapping
);
1413 spin_unlock(&info
->lock
);
1417 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1419 file_accessed(file
);
1420 vma
->vm_ops
= &shmem_vm_ops
;
1424 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1425 umode_t mode
, dev_t dev
, unsigned long flags
)
1427 struct inode
*inode
;
1428 struct shmem_inode_info
*info
;
1429 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1432 if (shmem_reserve_inode(sb
))
1435 inode
= new_inode(sb
);
1437 inode_init_owner(inode
, dir
, mode
);
1438 inode
->i_blocks
= 0;
1439 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1440 inode
->i_generation
= get_seconds();
1441 info
= SHMEM_I(inode
);
1442 memset(info
, 0, (char *)inode
- (char *)info
);
1443 spin_lock_init(&info
->lock
);
1444 info
->seals
= F_SEAL_SEAL
;
1445 info
->flags
= flags
& VM_NORESERVE
;
1446 INIT_LIST_HEAD(&info
->swaplist
);
1447 simple_xattrs_init(&info
->xattrs
);
1448 cache_no_acl(inode
);
1450 switch (mode
& S_IFMT
) {
1452 inode
->i_op
= &shmem_special_inode_operations
;
1453 init_special_inode(inode
, mode
, dev
);
1456 inode
->i_mapping
->a_ops
= &shmem_aops
;
1457 inode
->i_op
= &shmem_inode_operations
;
1458 inode
->i_fop
= &shmem_file_operations
;
1459 mpol_shared_policy_init(&info
->policy
,
1460 shmem_get_sbmpol(sbinfo
));
1464 /* Some things misbehave if size == 0 on a directory */
1465 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1466 inode
->i_op
= &shmem_dir_inode_operations
;
1467 inode
->i_fop
= &simple_dir_operations
;
1471 * Must not load anything in the rbtree,
1472 * mpol_free_shared_policy will not be called.
1474 mpol_shared_policy_init(&info
->policy
, NULL
);
1478 if (!sbinfo
->idr_nouse
) {
1479 /* inum 0 and 1 are unused */
1480 mutex_lock(&sbinfo
->idr_lock
);
1481 ino
= idr_alloc(&sbinfo
->idr
, inode
, 2, INT_MAX
,
1485 mutex_unlock(&sbinfo
->idr_lock
);
1486 __insert_inode_hash(inode
, inode
->i_ino
);
1489 mutex_unlock(&sbinfo
->idr_lock
);
1491 /* shmem_free_inode() will be called */
1495 inode
->i_ino
= get_next_ino();
1497 shmem_free_inode(sb
);
1501 bool shmem_mapping(struct address_space
*mapping
)
1506 return mapping
->host
->i_sb
->s_op
== &shmem_ops
;
1510 static const struct inode_operations shmem_symlink_inode_operations
;
1511 static const struct inode_operations shmem_short_symlink_operations
;
1513 #ifdef CONFIG_TMPFS_XATTR
1514 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1516 #define shmem_initxattrs NULL
1520 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1521 loff_t pos
, unsigned len
, unsigned flags
,
1522 struct page
**pagep
, void **fsdata
)
1524 struct inode
*inode
= mapping
->host
;
1525 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1526 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1528 /* i_mutex is held by caller */
1529 if (unlikely(info
->seals
)) {
1530 if (info
->seals
& F_SEAL_WRITE
)
1532 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
1536 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1540 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1541 loff_t pos
, unsigned len
, unsigned copied
,
1542 struct page
*page
, void *fsdata
)
1544 struct inode
*inode
= mapping
->host
;
1546 if (pos
+ copied
> inode
->i_size
)
1547 i_size_write(inode
, pos
+ copied
);
1549 if (!PageUptodate(page
)) {
1550 if (copied
< PAGE_CACHE_SIZE
) {
1551 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1552 zero_user_segments(page
, 0, from
,
1553 from
+ copied
, PAGE_CACHE_SIZE
);
1555 SetPageUptodate(page
);
1557 set_page_dirty(page
);
1559 page_cache_release(page
);
1564 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1566 struct file
*file
= iocb
->ki_filp
;
1567 struct inode
*inode
= file_inode(file
);
1568 struct address_space
*mapping
= inode
->i_mapping
;
1570 unsigned long offset
;
1571 enum sgp_type sgp
= SGP_READ
;
1574 loff_t
*ppos
= &iocb
->ki_pos
;
1577 * Might this read be for a stacking filesystem? Then when reading
1578 * holes of a sparse file, we actually need to allocate those pages,
1579 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1581 if (!iter_is_iovec(to
))
1584 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1585 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1588 struct page
*page
= NULL
;
1590 unsigned long nr
, ret
;
1591 loff_t i_size
= i_size_read(inode
);
1593 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1594 if (index
> end_index
)
1596 if (index
== end_index
) {
1597 nr
= i_size
& ~PAGE_CACHE_MASK
;
1602 error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1604 if (error
== -EINVAL
)
1612 * We must evaluate after, since reads (unlike writes)
1613 * are called without i_mutex protection against truncate
1615 nr
= PAGE_CACHE_SIZE
;
1616 i_size
= i_size_read(inode
);
1617 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1618 if (index
== end_index
) {
1619 nr
= i_size
& ~PAGE_CACHE_MASK
;
1622 page_cache_release(page
);
1630 * If users can be writing to this page using arbitrary
1631 * virtual addresses, take care about potential aliasing
1632 * before reading the page on the kernel side.
1634 if (mapping_writably_mapped(mapping
))
1635 flush_dcache_page(page
);
1637 * Mark the page accessed if we read the beginning.
1640 mark_page_accessed(page
);
1642 page
= ZERO_PAGE(0);
1643 page_cache_get(page
);
1647 * Ok, we have the page, and it's up-to-date, so
1648 * now we can copy it to user space...
1650 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
1653 index
+= offset
>> PAGE_CACHE_SHIFT
;
1654 offset
&= ~PAGE_CACHE_MASK
;
1656 page_cache_release(page
);
1657 if (!iov_iter_count(to
))
1666 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1667 file_accessed(file
);
1668 return retval
? retval
: error
;
1671 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1672 struct pipe_inode_info
*pipe
, size_t len
,
1675 struct address_space
*mapping
= in
->f_mapping
;
1676 struct inode
*inode
= mapping
->host
;
1677 unsigned int loff
, nr_pages
, req_pages
;
1678 struct page
*pages
[PIPE_DEF_BUFFERS
];
1679 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1681 pgoff_t index
, end_index
;
1684 struct splice_pipe_desc spd
= {
1687 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1689 .ops
= &page_cache_pipe_buf_ops
,
1690 .spd_release
= spd_release_page
,
1693 isize
= i_size_read(inode
);
1694 if (unlikely(*ppos
>= isize
))
1697 left
= isize
- *ppos
;
1698 if (unlikely(left
< len
))
1701 if (splice_grow_spd(pipe
, &spd
))
1704 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1705 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1706 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1707 nr_pages
= min(req_pages
, spd
.nr_pages_max
);
1709 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1710 nr_pages
, spd
.pages
);
1711 index
+= spd
.nr_pages
;
1714 while (spd
.nr_pages
< nr_pages
) {
1715 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1719 spd
.pages
[spd
.nr_pages
++] = page
;
1723 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1724 nr_pages
= spd
.nr_pages
;
1727 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1728 unsigned int this_len
;
1733 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1734 page
= spd
.pages
[page_nr
];
1736 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1737 error
= shmem_getpage(inode
, index
, &page
,
1742 page_cache_release(spd
.pages
[page_nr
]);
1743 spd
.pages
[page_nr
] = page
;
1746 isize
= i_size_read(inode
);
1747 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1748 if (unlikely(!isize
|| index
> end_index
))
1751 if (end_index
== index
) {
1754 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1758 this_len
= min(this_len
, plen
- loff
);
1762 spd
.partial
[page_nr
].offset
= loff
;
1763 spd
.partial
[page_nr
].len
= this_len
;
1770 while (page_nr
< nr_pages
)
1771 page_cache_release(spd
.pages
[page_nr
++]);
1774 error
= splice_to_pipe(pipe
, &spd
);
1776 splice_shrink_spd(&spd
);
1786 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1788 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
1789 pgoff_t index
, pgoff_t end
, int whence
)
1792 struct pagevec pvec
;
1793 pgoff_t indices
[PAGEVEC_SIZE
];
1797 pagevec_init(&pvec
, 0);
1798 pvec
.nr
= 1; /* start small: we may be there already */
1800 pvec
.nr
= find_get_entries(mapping
, index
,
1801 pvec
.nr
, pvec
.pages
, indices
);
1803 if (whence
== SEEK_DATA
)
1807 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
1808 if (index
< indices
[i
]) {
1809 if (whence
== SEEK_HOLE
) {
1815 page
= pvec
.pages
[i
];
1816 if (page
&& !radix_tree_exceptional_entry(page
)) {
1817 if (!PageUptodate(page
))
1821 (page
&& whence
== SEEK_DATA
) ||
1822 (!page
&& whence
== SEEK_HOLE
)) {
1827 pagevec_remove_exceptionals(&pvec
);
1828 pagevec_release(&pvec
);
1829 pvec
.nr
= PAGEVEC_SIZE
;
1835 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
1837 struct address_space
*mapping
= file
->f_mapping
;
1838 struct inode
*inode
= mapping
->host
;
1842 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
1843 return generic_file_llseek_size(file
, offset
, whence
,
1844 MAX_LFS_FILESIZE
, i_size_read(inode
));
1845 mutex_lock(&inode
->i_mutex
);
1846 /* We're holding i_mutex so we can access i_size directly */
1850 else if (offset
>= inode
->i_size
)
1853 start
= offset
>> PAGE_CACHE_SHIFT
;
1854 end
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1855 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
1856 new_offset
<<= PAGE_CACHE_SHIFT
;
1857 if (new_offset
> offset
) {
1858 if (new_offset
< inode
->i_size
)
1859 offset
= new_offset
;
1860 else if (whence
== SEEK_DATA
)
1863 offset
= inode
->i_size
;
1868 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
1869 mutex_unlock(&inode
->i_mutex
);
1874 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1875 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1877 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1878 #define LAST_SCAN 4 /* about 150ms max */
1880 static void shmem_tag_pins(struct address_space
*mapping
)
1882 struct radix_tree_iter iter
;
1892 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1893 page
= radix_tree_deref_slot(slot
);
1894 if (!page
|| radix_tree_exception(page
)) {
1895 if (radix_tree_deref_retry(page
))
1897 } else if (page_count(page
) - page_mapcount(page
) > 1) {
1898 spin_lock_irq(&mapping
->tree_lock
);
1899 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
1901 spin_unlock_irq(&mapping
->tree_lock
);
1904 if (need_resched()) {
1906 start
= iter
.index
+ 1;
1914 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1915 * via get_user_pages(), drivers might have some pending I/O without any active
1916 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1917 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1918 * them to be dropped.
1919 * The caller must guarantee that no new user will acquire writable references
1920 * to those pages to avoid races.
1922 static int shmem_wait_for_pins(struct address_space
*mapping
)
1924 struct radix_tree_iter iter
;
1930 shmem_tag_pins(mapping
);
1933 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
1934 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
1938 lru_add_drain_all();
1939 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
1945 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
1946 start
, SHMEM_TAG_PINNED
) {
1948 page
= radix_tree_deref_slot(slot
);
1949 if (radix_tree_exception(page
)) {
1950 if (radix_tree_deref_retry(page
))
1957 page_count(page
) - page_mapcount(page
) != 1) {
1958 if (scan
< LAST_SCAN
)
1959 goto continue_resched
;
1962 * On the last scan, we clean up all those tags
1963 * we inserted; but make a note that we still
1964 * found pages pinned.
1969 spin_lock_irq(&mapping
->tree_lock
);
1970 radix_tree_tag_clear(&mapping
->page_tree
,
1971 iter
.index
, SHMEM_TAG_PINNED
);
1972 spin_unlock_irq(&mapping
->tree_lock
);
1974 if (need_resched()) {
1976 start
= iter
.index
+ 1;
1986 #define F_ALL_SEALS (F_SEAL_SEAL | \
1991 int shmem_add_seals(struct file
*file
, unsigned int seals
)
1993 struct inode
*inode
= file_inode(file
);
1994 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1999 * Sealing allows multiple parties to share a shmem-file but restrict
2000 * access to a specific subset of file operations. Seals can only be
2001 * added, but never removed. This way, mutually untrusted parties can
2002 * share common memory regions with a well-defined policy. A malicious
2003 * peer can thus never perform unwanted operations on a shared object.
2005 * Seals are only supported on special shmem-files and always affect
2006 * the whole underlying inode. Once a seal is set, it may prevent some
2007 * kinds of access to the file. Currently, the following seals are
2009 * SEAL_SEAL: Prevent further seals from being set on this file
2010 * SEAL_SHRINK: Prevent the file from shrinking
2011 * SEAL_GROW: Prevent the file from growing
2012 * SEAL_WRITE: Prevent write access to the file
2014 * As we don't require any trust relationship between two parties, we
2015 * must prevent seals from being removed. Therefore, sealing a file
2016 * only adds a given set of seals to the file, it never touches
2017 * existing seals. Furthermore, the "setting seals"-operation can be
2018 * sealed itself, which basically prevents any further seal from being
2021 * Semantics of sealing are only defined on volatile files. Only
2022 * anonymous shmem files support sealing. More importantly, seals are
2023 * never written to disk. Therefore, there's no plan to support it on
2027 if (file
->f_op
!= &shmem_file_operations
)
2029 if (!(file
->f_mode
& FMODE_WRITE
))
2031 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2034 mutex_lock(&inode
->i_mutex
);
2036 if (info
->seals
& F_SEAL_SEAL
) {
2041 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2042 error
= mapping_deny_writable(file
->f_mapping
);
2046 error
= shmem_wait_for_pins(file
->f_mapping
);
2048 mapping_allow_writable(file
->f_mapping
);
2053 info
->seals
|= seals
;
2057 mutex_unlock(&inode
->i_mutex
);
2060 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2062 int shmem_get_seals(struct file
*file
)
2064 if (file
->f_op
!= &shmem_file_operations
)
2067 return SHMEM_I(file_inode(file
))->seals
;
2069 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2071 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2077 /* disallow upper 32bit */
2081 error
= shmem_add_seals(file
, arg
);
2084 error
= shmem_get_seals(file
);
2094 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2097 struct inode
*inode
= file_inode(file
);
2098 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2099 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2100 struct shmem_falloc shmem_falloc
;
2101 pgoff_t start
, index
, end
;
2104 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2107 mutex_lock(&inode
->i_mutex
);
2109 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2110 struct address_space
*mapping
= file
->f_mapping
;
2111 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2112 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2113 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2115 /* protected by i_mutex */
2116 if (info
->seals
& F_SEAL_WRITE
) {
2121 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2122 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2123 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2124 spin_lock(&inode
->i_lock
);
2125 inode
->i_private
= &shmem_falloc
;
2126 spin_unlock(&inode
->i_lock
);
2128 if ((u64
)unmap_end
> (u64
)unmap_start
)
2129 unmap_mapping_range(mapping
, unmap_start
,
2130 1 + unmap_end
- unmap_start
, 0);
2131 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2132 /* No need to unmap again: hole-punching leaves COWed pages */
2134 spin_lock(&inode
->i_lock
);
2135 inode
->i_private
= NULL
;
2136 wake_up_all(&shmem_falloc_waitq
);
2137 spin_unlock(&inode
->i_lock
);
2142 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2143 error
= inode_newsize_ok(inode
, offset
+ len
);
2147 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2152 start
= offset
>> PAGE_CACHE_SHIFT
;
2153 end
= (offset
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
2154 /* Try to avoid a swapstorm if len is impossible to satisfy */
2155 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2160 shmem_falloc
.waitq
= NULL
;
2161 shmem_falloc
.start
= start
;
2162 shmem_falloc
.next
= start
;
2163 shmem_falloc
.nr_falloced
= 0;
2164 shmem_falloc
.nr_unswapped
= 0;
2165 spin_lock(&inode
->i_lock
);
2166 inode
->i_private
= &shmem_falloc
;
2167 spin_unlock(&inode
->i_lock
);
2169 for (index
= start
; index
< end
; index
++) {
2173 * Good, the fallocate(2) manpage permits EINTR: we may have
2174 * been interrupted because we are using up too much memory.
2176 if (signal_pending(current
))
2178 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2181 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
,
2184 /* Remove the !PageUptodate pages we added */
2185 if (index
> start
) {
2186 shmem_undo_range(inode
,
2187 (loff_t
)start
<< PAGE_CACHE_SHIFT
,
2188 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) - 1, true);
2194 * Inform shmem_writepage() how far we have reached.
2195 * No need for lock or barrier: we have the page lock.
2197 shmem_falloc
.next
++;
2198 if (!PageUptodate(page
))
2199 shmem_falloc
.nr_falloced
++;
2202 * If !PageUptodate, leave it that way so that freeable pages
2203 * can be recognized if we need to rollback on error later.
2204 * But set_page_dirty so that memory pressure will swap rather
2205 * than free the pages we are allocating (and SGP_CACHE pages
2206 * might still be clean: we now need to mark those dirty too).
2208 set_page_dirty(page
);
2210 page_cache_release(page
);
2214 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2215 i_size_write(inode
, offset
+ len
);
2216 inode
->i_ctime
= CURRENT_TIME
;
2218 spin_lock(&inode
->i_lock
);
2219 inode
->i_private
= NULL
;
2220 spin_unlock(&inode
->i_lock
);
2222 mutex_unlock(&inode
->i_mutex
);
2226 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2228 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2230 buf
->f_type
= TMPFS_MAGIC
;
2231 buf
->f_bsize
= PAGE_CACHE_SIZE
;
2232 buf
->f_namelen
= NAME_MAX
;
2233 if (sbinfo
->max_blocks
) {
2234 buf
->f_blocks
= sbinfo
->max_blocks
;
2236 buf
->f_bfree
= sbinfo
->max_blocks
-
2237 percpu_counter_sum(&sbinfo
->used_blocks
);
2239 if (sbinfo
->max_inodes
) {
2240 buf
->f_files
= sbinfo
->max_inodes
;
2241 buf
->f_ffree
= sbinfo
->free_inodes
;
2243 /* else leave those fields 0 like simple_statfs */
2248 * File creation. Allocate an inode, and we're done..
2251 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2253 struct inode
*inode
;
2254 int error
= -ENOSPC
;
2256 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2258 error
= simple_acl_create(dir
, inode
);
2261 error
= security_inode_init_security(inode
, dir
,
2263 shmem_initxattrs
, NULL
);
2264 if (error
&& error
!= -EOPNOTSUPP
)
2268 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2269 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2270 d_instantiate(dentry
, inode
);
2271 dget(dentry
); /* Extra count - pin the dentry in core */
2280 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2282 struct inode
*inode
;
2283 int error
= -ENOSPC
;
2285 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2287 error
= security_inode_init_security(inode
, dir
,
2289 shmem_initxattrs
, NULL
);
2290 if (error
&& error
!= -EOPNOTSUPP
)
2292 error
= simple_acl_create(dir
, inode
);
2295 d_tmpfile(dentry
, inode
);
2303 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2307 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2313 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2316 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2322 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2324 struct inode
*inode
= d_inode(old_dentry
);
2328 * No ordinary (disk based) filesystem counts links as inodes;
2329 * but each new link needs a new dentry, pinning lowmem, and
2330 * tmpfs dentries cannot be pruned until they are unlinked.
2332 ret
= shmem_reserve_inode(inode
->i_sb
);
2336 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2337 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2339 ihold(inode
); /* New dentry reference */
2340 dget(dentry
); /* Extra pinning count for the created dentry */
2341 d_instantiate(dentry
, inode
);
2346 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2348 struct inode
*inode
= d_inode(dentry
);
2350 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2351 shmem_free_inode(inode
->i_sb
);
2353 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2354 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2356 dput(dentry
); /* Undo the count from "create" - this does all the work */
2360 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2362 if (!simple_empty(dentry
))
2365 drop_nlink(d_inode(dentry
));
2367 return shmem_unlink(dir
, dentry
);
2370 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2372 bool old_is_dir
= d_is_dir(old_dentry
);
2373 bool new_is_dir
= d_is_dir(new_dentry
);
2375 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2377 drop_nlink(old_dir
);
2380 drop_nlink(new_dir
);
2384 old_dir
->i_ctime
= old_dir
->i_mtime
=
2385 new_dir
->i_ctime
= new_dir
->i_mtime
=
2386 d_inode(old_dentry
)->i_ctime
=
2387 d_inode(new_dentry
)->i_ctime
= CURRENT_TIME
;
2392 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
2394 struct dentry
*whiteout
;
2397 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2401 error
= shmem_mknod(old_dir
, whiteout
,
2402 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2408 * Cheat and hash the whiteout while the old dentry is still in
2409 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2411 * d_lookup() will consistently find one of them at this point,
2412 * not sure which one, but that isn't even important.
2419 * The VFS layer already does all the dentry stuff for rename,
2420 * we just have to decrement the usage count for the target if
2421 * it exists so that the VFS layer correctly free's it when it
2424 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
2426 struct inode
*inode
= d_inode(old_dentry
);
2427 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2429 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
2432 if (flags
& RENAME_EXCHANGE
)
2433 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
2435 if (!simple_empty(new_dentry
))
2438 if (flags
& RENAME_WHITEOUT
) {
2441 error
= shmem_whiteout(old_dir
, old_dentry
);
2446 if (d_really_is_positive(new_dentry
)) {
2447 (void) shmem_unlink(new_dir
, new_dentry
);
2448 if (they_are_dirs
) {
2449 drop_nlink(d_inode(new_dentry
));
2450 drop_nlink(old_dir
);
2452 } else if (they_are_dirs
) {
2453 drop_nlink(old_dir
);
2457 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2458 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2459 old_dir
->i_ctime
= old_dir
->i_mtime
=
2460 new_dir
->i_ctime
= new_dir
->i_mtime
=
2461 inode
->i_ctime
= CURRENT_TIME
;
2465 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2469 struct inode
*inode
;
2472 struct shmem_inode_info
*info
;
2474 len
= strlen(symname
) + 1;
2475 if (len
> PAGE_CACHE_SIZE
)
2476 return -ENAMETOOLONG
;
2478 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2482 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2483 shmem_initxattrs
, NULL
);
2485 if (error
!= -EOPNOTSUPP
) {
2492 info
= SHMEM_I(inode
);
2493 inode
->i_size
= len
-1;
2494 if (len
<= SHORT_SYMLINK_LEN
) {
2495 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
2496 if (!inode
->i_link
) {
2500 inode
->i_op
= &shmem_short_symlink_operations
;
2502 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
2507 inode
->i_mapping
->a_ops
= &shmem_aops
;
2508 inode
->i_op
= &shmem_symlink_inode_operations
;
2509 kaddr
= kmap_atomic(page
);
2510 memcpy(kaddr
, symname
, len
);
2511 kunmap_atomic(kaddr
);
2512 SetPageUptodate(page
);
2513 set_page_dirty(page
);
2515 page_cache_release(page
);
2517 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2518 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2519 d_instantiate(dentry
, inode
);
2524 static const char *shmem_follow_link(struct dentry
*dentry
, void **cookie
)
2526 struct page
*page
= NULL
;
2527 int error
= shmem_getpage(d_inode(dentry
), 0, &page
, SGP_READ
, NULL
);
2529 return ERR_PTR(error
);
2535 static void shmem_put_link(struct inode
*unused
, void *cookie
)
2537 struct page
*page
= cookie
;
2539 mark_page_accessed(page
);
2540 page_cache_release(page
);
2543 #ifdef CONFIG_TMPFS_XATTR
2545 * Superblocks without xattr inode operations may get some security.* xattr
2546 * support from the LSM "for free". As soon as we have any other xattrs
2547 * like ACLs, we also need to implement the security.* handlers at
2548 * filesystem level, though.
2552 * Callback for security_inode_init_security() for acquiring xattrs.
2554 static int shmem_initxattrs(struct inode
*inode
,
2555 const struct xattr
*xattr_array
,
2558 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2559 const struct xattr
*xattr
;
2560 struct simple_xattr
*new_xattr
;
2563 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2564 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
2568 len
= strlen(xattr
->name
) + 1;
2569 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2571 if (!new_xattr
->name
) {
2576 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2577 XATTR_SECURITY_PREFIX_LEN
);
2578 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2581 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
2587 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2588 #ifdef CONFIG_TMPFS_POSIX_ACL
2589 &posix_acl_access_xattr_handler
,
2590 &posix_acl_default_xattr_handler
,
2595 static int shmem_xattr_validate(const char *name
)
2597 struct { const char *prefix
; size_t len
; } arr
[] = {
2598 { XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
},
2599 { XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
}
2603 for (i
= 0; i
< ARRAY_SIZE(arr
); i
++) {
2604 size_t preflen
= arr
[i
].len
;
2605 if (strncmp(name
, arr
[i
].prefix
, preflen
) == 0) {
2614 static ssize_t
shmem_getxattr(struct dentry
*dentry
, const char *name
,
2615 void *buffer
, size_t size
)
2617 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2621 * If this is a request for a synthetic attribute in the system.*
2622 * namespace use the generic infrastructure to resolve a handler
2623 * for it via sb->s_xattr.
2625 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2626 return generic_getxattr(dentry
, name
, buffer
, size
);
2628 err
= shmem_xattr_validate(name
);
2632 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
2635 static int shmem_setxattr(struct dentry
*dentry
, const char *name
,
2636 const void *value
, size_t size
, int flags
)
2638 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2642 * If this is a request for a synthetic attribute in the system.*
2643 * namespace use the generic infrastructure to resolve a handler
2644 * for it via sb->s_xattr.
2646 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2647 return generic_setxattr(dentry
, name
, value
, size
, flags
);
2649 err
= shmem_xattr_validate(name
);
2653 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
2656 static int shmem_removexattr(struct dentry
*dentry
, const char *name
)
2658 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2662 * If this is a request for a synthetic attribute in the system.*
2663 * namespace use the generic infrastructure to resolve a handler
2664 * for it via sb->s_xattr.
2666 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2667 return generic_removexattr(dentry
, name
);
2669 err
= shmem_xattr_validate(name
);
2673 return simple_xattr_remove(&info
->xattrs
, name
);
2676 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2678 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2679 return simple_xattr_list(&info
->xattrs
, buffer
, size
);
2681 #endif /* CONFIG_TMPFS_XATTR */
2683 static const struct inode_operations shmem_short_symlink_operations
= {
2684 .readlink
= generic_readlink
,
2685 .follow_link
= simple_follow_link
,
2686 #ifdef CONFIG_TMPFS_XATTR
2687 .setxattr
= shmem_setxattr
,
2688 .getxattr
= shmem_getxattr
,
2689 .listxattr
= shmem_listxattr
,
2690 .removexattr
= shmem_removexattr
,
2694 static const struct inode_operations shmem_symlink_inode_operations
= {
2695 .readlink
= generic_readlink
,
2696 .follow_link
= shmem_follow_link
,
2697 .put_link
= shmem_put_link
,
2698 #ifdef CONFIG_TMPFS_XATTR
2699 .setxattr
= shmem_setxattr
,
2700 .getxattr
= shmem_getxattr
,
2701 .listxattr
= shmem_listxattr
,
2702 .removexattr
= shmem_removexattr
,
2706 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2708 return ERR_PTR(-ESTALE
);
2711 static int shmem_match(struct inode
*ino
, void *vfh
)
2715 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2718 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2719 struct fid
*fid
, int fh_len
, int fh_type
)
2721 struct inode
*inode
;
2722 struct dentry
*dentry
= NULL
;
2729 inode
= ilookup5(sb
, inum
, shmem_match
, fid
->raw
);
2731 dentry
= d_find_alias(inode
);
2738 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
2739 struct inode
*parent
)
2743 return FILEID_INVALID
;
2746 fh
[0] = inode
->i_generation
;
2747 fh
[1] = inode
->i_ino
;
2753 static const struct export_operations shmem_export_ops
= {
2754 .get_parent
= shmem_get_parent
,
2755 .encode_fh
= shmem_encode_fh
,
2756 .fh_to_dentry
= shmem_fh_to_dentry
,
2759 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2762 char *this_char
, *value
, *rest
;
2763 struct mempolicy
*mpol
= NULL
;
2767 while (options
!= NULL
) {
2768 this_char
= options
;
2771 * NUL-terminate this option: unfortunately,
2772 * mount options form a comma-separated list,
2773 * but mpol's nodelist may also contain commas.
2775 options
= strchr(options
, ',');
2776 if (options
== NULL
)
2779 if (!isdigit(*options
)) {
2786 if ((value
= strchr(this_char
,'=')) != NULL
) {
2790 "tmpfs: No value for mount option '%s'\n",
2795 if (!strcmp(this_char
,"size")) {
2796 unsigned long long size
;
2797 size
= memparse(value
,&rest
);
2799 size
<<= PAGE_SHIFT
;
2800 size
*= totalram_pages
;
2806 sbinfo
->max_blocks
=
2807 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2808 } else if (!strcmp(this_char
,"nr_blocks")) {
2809 sbinfo
->max_blocks
= memparse(value
, &rest
);
2812 } else if (!strcmp(this_char
,"nr_inodes")) {
2813 sbinfo
->max_inodes
= memparse(value
, &rest
);
2814 if (*rest
|| sbinfo
->max_inodes
< 2)
2816 } else if (!strcmp(this_char
,"mode")) {
2819 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2822 } else if (!strcmp(this_char
,"uid")) {
2825 uid
= simple_strtoul(value
, &rest
, 0);
2828 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
2829 if (!uid_valid(sbinfo
->uid
))
2831 } else if (!strcmp(this_char
,"gid")) {
2834 gid
= simple_strtoul(value
, &rest
, 0);
2837 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
2838 if (!gid_valid(sbinfo
->gid
))
2840 } else if (!strcmp(this_char
,"mpol")) {
2843 if (mpol_parse_str(value
, &mpol
))
2846 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2851 sbinfo
->mpol
= mpol
;
2855 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2863 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2865 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2866 struct shmem_sb_info config
= *sbinfo
;
2868 int error
= -EINVAL
;
2871 if (shmem_parse_options(data
, &config
, true))
2874 spin_lock(&sbinfo
->stat_lock
);
2875 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2876 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2878 if (config
.max_inodes
< inodes
)
2881 * Those tests disallow limited->unlimited while any are in use;
2882 * but we must separately disallow unlimited->limited, because
2883 * in that case we have no record of how much is already in use.
2885 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2887 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2891 sbinfo
->max_blocks
= config
.max_blocks
;
2892 sbinfo
->max_inodes
= config
.max_inodes
;
2893 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2896 * Preserve previous mempolicy unless mpol remount option was specified.
2899 mpol_put(sbinfo
->mpol
);
2900 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2903 spin_unlock(&sbinfo
->stat_lock
);
2907 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2909 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2911 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2912 seq_printf(seq
, ",size=%luk",
2913 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2914 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2915 seq_printf(seq
, ",nr_inodes=%d", sbinfo
->max_inodes
);
2916 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2917 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2918 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
2919 seq_printf(seq
, ",uid=%u",
2920 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
2921 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
2922 seq_printf(seq
, ",gid=%u",
2923 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
2924 shmem_show_mpol(seq
, sbinfo
->mpol
);
2928 #define MFD_NAME_PREFIX "memfd:"
2929 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2930 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2932 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2934 SYSCALL_DEFINE2(memfd_create
,
2935 const char __user
*, uname
,
2936 unsigned int, flags
)
2938 struct shmem_inode_info
*info
;
2944 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
2947 /* length includes terminating zero */
2948 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
2951 if (len
> MFD_NAME_MAX_LEN
+ 1)
2954 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
2958 strcpy(name
, MFD_NAME_PREFIX
);
2959 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
2964 /* terminating-zero may have changed after strnlen_user() returned */
2965 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
2970 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
2976 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
2978 error
= PTR_ERR(file
);
2981 info
= SHMEM_I(file_inode(file
));
2982 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
2983 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
2984 if (flags
& MFD_ALLOW_SEALING
)
2985 info
->seals
&= ~F_SEAL_SEAL
;
2987 fd_install(fd
, file
);
2998 #endif /* CONFIG_TMPFS */
3000 static void shmem_put_super(struct super_block
*sb
)
3002 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3004 if (!sbinfo
->idr_nouse
)
3005 idr_destroy(&sbinfo
->idr
);
3006 percpu_counter_destroy(&sbinfo
->used_blocks
);
3007 mpol_put(sbinfo
->mpol
);
3009 sb
->s_fs_info
= NULL
;
3012 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3014 struct inode
*inode
;
3015 struct shmem_sb_info
*sbinfo
;
3018 /* Round up to L1_CACHE_BYTES to resist false sharing */
3019 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3020 L1_CACHE_BYTES
), GFP_KERNEL
);
3024 mutex_init(&sbinfo
->idr_lock
);
3025 idr_init(&sbinfo
->idr
);
3026 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3027 sbinfo
->uid
= current_fsuid();
3028 sbinfo
->gid
= current_fsgid();
3029 sb
->s_fs_info
= sbinfo
;
3033 * Per default we only allow half of the physical ram per
3034 * tmpfs instance, limiting inodes to one per page of lowmem;
3035 * but the internal instance is left unlimited.
3037 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3038 sbinfo
->max_blocks
= shmem_default_max_blocks();
3039 sbinfo
->max_inodes
= shmem_default_max_inodes();
3040 if (shmem_parse_options(data
, sbinfo
, false)) {
3045 sb
->s_flags
|= MS_NOUSER
;
3047 sb
->s_export_op
= &shmem_export_ops
;
3048 sb
->s_flags
|= MS_NOSEC
;
3050 sb
->s_flags
|= MS_NOUSER
;
3053 spin_lock_init(&sbinfo
->stat_lock
);
3054 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3056 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3058 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3059 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
3060 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
3061 sb
->s_magic
= TMPFS_MAGIC
;
3062 sb
->s_op
= &shmem_ops
;
3063 sb
->s_time_gran
= 1;
3064 #ifdef CONFIG_TMPFS_XATTR
3065 sb
->s_xattr
= shmem_xattr_handlers
;
3067 #ifdef CONFIG_TMPFS_POSIX_ACL
3068 sb
->s_flags
|= MS_POSIXACL
;
3071 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3074 inode
->i_uid
= sbinfo
->uid
;
3075 inode
->i_gid
= sbinfo
->gid
;
3076 sb
->s_root
= d_make_root(inode
);
3082 shmem_put_super(sb
);
3086 static struct kmem_cache
*shmem_inode_cachep
;
3088 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3090 struct shmem_inode_info
*info
;
3091 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3094 return &info
->vfs_inode
;
3097 static void shmem_destroy_callback(struct rcu_head
*head
)
3099 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3100 kfree(inode
->i_link
);
3101 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3104 static void shmem_destroy_inode(struct inode
*inode
)
3106 if (S_ISREG(inode
->i_mode
))
3107 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3108 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3111 static void shmem_init_inode(void *foo
)
3113 struct shmem_inode_info
*info
= foo
;
3114 inode_init_once(&info
->vfs_inode
);
3117 static int shmem_init_inodecache(void)
3119 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3120 sizeof(struct shmem_inode_info
),
3121 0, SLAB_PANIC
, shmem_init_inode
);
3125 static void shmem_destroy_inodecache(void)
3127 kmem_cache_destroy(shmem_inode_cachep
);
3130 static __init
void shmem_no_idr(struct super_block
*sb
)
3132 struct shmem_sb_info
*sbinfo
;
3134 sbinfo
= SHMEM_SB(sb
);
3135 sbinfo
->idr_nouse
= true;
3136 idr_destroy(&sbinfo
->idr
);
3139 static const struct address_space_operations shmem_aops
= {
3140 .writepage
= shmem_writepage
,
3141 .set_page_dirty
= __set_page_dirty_no_writeback
,
3143 .write_begin
= shmem_write_begin
,
3144 .write_end
= shmem_write_end
,
3146 #ifdef CONFIG_MIGRATION
3147 .migratepage
= migrate_page
,
3149 .error_remove_page
= generic_error_remove_page
,
3152 static const struct file_operations shmem_file_operations
= {
3155 .llseek
= shmem_file_llseek
,
3156 .read_iter
= shmem_file_read_iter
,
3157 .write_iter
= generic_file_write_iter
,
3158 .fsync
= noop_fsync
,
3159 .splice_read
= shmem_file_splice_read
,
3160 .splice_write
= iter_file_splice_write
,
3161 .fallocate
= shmem_fallocate
,
3165 static const struct inode_operations shmem_inode_operations
= {
3166 .getattr
= shmem_getattr
,
3167 .setattr
= shmem_setattr
,
3168 #ifdef CONFIG_TMPFS_XATTR
3169 .setxattr
= shmem_setxattr
,
3170 .getxattr
= shmem_getxattr
,
3171 .listxattr
= shmem_listxattr
,
3172 .removexattr
= shmem_removexattr
,
3173 .set_acl
= simple_set_acl
,
3177 static const struct inode_operations shmem_dir_inode_operations
= {
3179 .create
= shmem_create
,
3180 .lookup
= simple_lookup
,
3182 .unlink
= shmem_unlink
,
3183 .symlink
= shmem_symlink
,
3184 .mkdir
= shmem_mkdir
,
3185 .rmdir
= shmem_rmdir
,
3186 .mknod
= shmem_mknod
,
3187 .rename2
= shmem_rename2
,
3188 .tmpfile
= shmem_tmpfile
,
3190 #ifdef CONFIG_TMPFS_XATTR
3191 .setxattr
= shmem_setxattr
,
3192 .getxattr
= shmem_getxattr
,
3193 .listxattr
= shmem_listxattr
,
3194 .removexattr
= shmem_removexattr
,
3196 #ifdef CONFIG_TMPFS_POSIX_ACL
3197 .setattr
= shmem_setattr
,
3198 .set_acl
= simple_set_acl
,
3202 static const struct inode_operations shmem_special_inode_operations
= {
3203 #ifdef CONFIG_TMPFS_XATTR
3204 .setxattr
= shmem_setxattr
,
3205 .getxattr
= shmem_getxattr
,
3206 .listxattr
= shmem_listxattr
,
3207 .removexattr
= shmem_removexattr
,
3209 #ifdef CONFIG_TMPFS_POSIX_ACL
3210 .setattr
= shmem_setattr
,
3211 .set_acl
= simple_set_acl
,
3215 static const struct super_operations shmem_ops
= {
3216 .alloc_inode
= shmem_alloc_inode
,
3217 .destroy_inode
= shmem_destroy_inode
,
3219 .statfs
= shmem_statfs
,
3220 .remount_fs
= shmem_remount_fs
,
3221 .show_options
= shmem_show_options
,
3223 .evict_inode
= shmem_evict_inode
,
3224 .drop_inode
= generic_delete_inode
,
3225 .put_super
= shmem_put_super
,
3228 static const struct vm_operations_struct shmem_vm_ops
= {
3229 .fault
= shmem_fault
,
3230 .map_pages
= filemap_map_pages
,
3232 .set_policy
= shmem_set_policy
,
3233 .get_policy
= shmem_get_policy
,
3237 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3238 int flags
, const char *dev_name
, void *data
)
3240 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3243 static struct file_system_type shmem_fs_type
= {
3244 .owner
= THIS_MODULE
,
3246 .mount
= shmem_mount
,
3247 .kill_sb
= kill_litter_super
,
3248 .fs_flags
= FS_USERNS_MOUNT
,
3251 int __init
shmem_init(void)
3255 /* If rootfs called this, don't re-init */
3256 if (shmem_inode_cachep
)
3259 error
= shmem_init_inodecache();
3263 error
= register_filesystem(&shmem_fs_type
);
3265 printk(KERN_ERR
"Could not register tmpfs\n");
3269 shm_mnt
= kern_mount(&shmem_fs_type
);
3270 if (IS_ERR(shm_mnt
)) {
3271 error
= PTR_ERR(shm_mnt
);
3272 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
3275 shmem_no_idr(shm_mnt
->mnt_sb
);
3279 unregister_filesystem(&shmem_fs_type
);
3281 shmem_destroy_inodecache();
3283 shm_mnt
= ERR_PTR(error
);
3287 #else /* !CONFIG_SHMEM */
3290 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3292 * This is intended for small system where the benefits of the full
3293 * shmem code (swap-backed and resource-limited) are outweighed by
3294 * their complexity. On systems without swap this code should be
3295 * effectively equivalent, but much lighter weight.
3298 static struct file_system_type shmem_fs_type
= {
3300 .mount
= ramfs_mount
,
3301 .kill_sb
= kill_litter_super
,
3302 .fs_flags
= FS_USERNS_MOUNT
,
3305 int __init
shmem_init(void)
3307 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
3309 shm_mnt
= kern_mount(&shmem_fs_type
);
3310 BUG_ON(IS_ERR(shm_mnt
));
3315 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
3320 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
3325 void shmem_unlock_mapping(struct address_space
*mapping
)
3329 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
3331 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
3333 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
3335 #define shmem_vm_ops generic_file_vm_ops
3336 #define shmem_file_operations ramfs_file_operations
3337 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3338 #define shmem_acct_size(flags, size) 0
3339 #define shmem_unacct_size(flags, size) do {} while (0)
3341 #endif /* CONFIG_SHMEM */
3345 static struct dentry_operations anon_ops
= {
3346 .d_dname
= simple_dname
3349 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
3350 unsigned long flags
, unsigned int i_flags
)
3353 struct inode
*inode
;
3355 struct super_block
*sb
;
3358 if (IS_ERR(shm_mnt
))
3359 return ERR_CAST(shm_mnt
);
3361 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
3362 return ERR_PTR(-EINVAL
);
3364 if (shmem_acct_size(flags
, size
))
3365 return ERR_PTR(-ENOMEM
);
3367 res
= ERR_PTR(-ENOMEM
);
3369 this.len
= strlen(name
);
3370 this.hash
= 0; /* will go */
3371 sb
= shm_mnt
->mnt_sb
;
3372 path
.mnt
= mntget(shm_mnt
);
3373 path
.dentry
= d_alloc_pseudo(sb
, &this);
3376 d_set_d_op(path
.dentry
, &anon_ops
);
3378 res
= ERR_PTR(-ENOSPC
);
3379 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
3383 inode
->i_flags
|= i_flags
;
3384 d_instantiate(path
.dentry
, inode
);
3385 inode
->i_size
= size
;
3386 clear_nlink(inode
); /* It is unlinked */
3387 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
3391 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
3392 &shmem_file_operations
);
3399 shmem_unacct_size(flags
, size
);
3406 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3407 * kernel internal. There will be NO LSM permission checks against the
3408 * underlying inode. So users of this interface must do LSM checks at a
3409 * higher layer. The users are the big_key and shm implementations. LSM
3410 * checks are provided at the key or shm level rather than the inode.
3411 * @name: name for dentry (to be seen in /proc/<pid>/maps
3412 * @size: size to be set for the file
3413 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3415 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3417 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
3421 * shmem_file_setup - get an unlinked file living in tmpfs
3422 * @name: name for dentry (to be seen in /proc/<pid>/maps
3423 * @size: size to be set for the file
3424 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3426 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3428 return __shmem_file_setup(name
, size
, flags
, 0);
3430 EXPORT_SYMBOL_GPL(shmem_file_setup
);
3433 * shmem_zero_setup - setup a shared anonymous mapping
3434 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3436 int shmem_zero_setup(struct vm_area_struct
*vma
)
3439 loff_t size
= vma
->vm_end
- vma
->vm_start
;
3442 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3443 * between XFS directory reading and selinux: since this file is only
3444 * accessible to the user through its mapping, use S_PRIVATE flag to
3445 * bypass file security, in the same way as shmem_kernel_file_setup().
3447 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
3449 return PTR_ERR(file
);
3453 vma
->vm_file
= file
;
3454 vma
->vm_ops
= &shmem_vm_ops
;
3459 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3460 * @mapping: the page's address_space
3461 * @index: the page index
3462 * @gfp: the page allocator flags to use if allocating
3464 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3465 * with any new page allocations done using the specified allocation flags.
3466 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3467 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3468 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3470 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3471 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3473 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
3474 pgoff_t index
, gfp_t gfp
)
3477 struct inode
*inode
= mapping
->host
;
3481 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
3482 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
3484 page
= ERR_PTR(error
);
3490 * The tiny !SHMEM case uses ramfs without swap
3492 return read_cache_page_gfp(mapping
, index
, gfp
);
3495 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);