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 unsigned long shmem_default_max_inodes(void)
117 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
121 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
122 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
123 struct shmem_inode_info
*info
, pgoff_t index
);
124 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
125 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
);
127 static inline int shmem_getpage(struct inode
*inode
, pgoff_t index
,
128 struct page
**pagep
, enum sgp_type sgp
, int *fault_type
)
130 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
131 mapping_gfp_mask(inode
->i_mapping
), fault_type
);
134 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
136 return sb
->s_fs_info
;
140 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
141 * for shared memory and for shared anonymous (/dev/zero) mappings
142 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
143 * consistent with the pre-accounting of private mappings ...
145 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
147 return (flags
& VM_NORESERVE
) ?
148 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
151 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
153 if (!(flags
& VM_NORESERVE
))
154 vm_unacct_memory(VM_ACCT(size
));
157 static inline int shmem_reacct_size(unsigned long flags
,
158 loff_t oldsize
, loff_t newsize
)
160 if (!(flags
& VM_NORESERVE
)) {
161 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
162 return security_vm_enough_memory_mm(current
->mm
,
163 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
164 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
165 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
171 * ... whereas tmpfs objects are accounted incrementally as
172 * pages are allocated, in order to allow huge sparse files.
173 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
174 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
176 static inline int shmem_acct_block(unsigned long flags
)
178 return (flags
& VM_NORESERVE
) ?
179 security_vm_enough_memory_mm(current
->mm
, VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
182 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
184 if (flags
& VM_NORESERVE
)
185 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
188 static const struct super_operations shmem_ops
;
189 static const struct address_space_operations shmem_aops
;
190 static const struct file_operations shmem_file_operations
;
191 static const struct inode_operations shmem_inode_operations
;
192 static const struct inode_operations shmem_dir_inode_operations
;
193 static const struct inode_operations shmem_special_inode_operations
;
194 static const struct vm_operations_struct shmem_vm_ops
;
196 static LIST_HEAD(shmem_swaplist
);
197 static DEFINE_MUTEX(shmem_swaplist_mutex
);
199 static int shmem_reserve_inode(struct super_block
*sb
)
201 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
202 if (sbinfo
->max_inodes
) {
203 spin_lock(&sbinfo
->stat_lock
);
204 if (!sbinfo
->free_inodes
) {
205 spin_unlock(&sbinfo
->stat_lock
);
208 sbinfo
->free_inodes
--;
209 spin_unlock(&sbinfo
->stat_lock
);
214 static void shmem_free_inode(struct super_block
*sb
)
216 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
217 if (sbinfo
->max_inodes
) {
218 spin_lock(&sbinfo
->stat_lock
);
219 sbinfo
->free_inodes
++;
220 spin_unlock(&sbinfo
->stat_lock
);
225 * shmem_recalc_inode - recalculate the block usage of an inode
226 * @inode: inode to recalc
228 * We have to calculate the free blocks since the mm can drop
229 * undirtied hole pages behind our back.
231 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
232 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
234 * It has to be called with the spinlock held.
236 static void shmem_recalc_inode(struct inode
*inode
)
238 struct shmem_inode_info
*info
= SHMEM_I(inode
);
241 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
243 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
244 if (sbinfo
->max_blocks
)
245 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
246 info
->alloced
-= freed
;
247 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
248 shmem_unacct_blocks(info
->flags
, freed
);
253 * Replace item expected in radix tree by a new item, while holding tree lock.
255 static int shmem_radix_tree_replace(struct address_space
*mapping
,
256 pgoff_t index
, void *expected
, void *replacement
)
261 VM_BUG_ON(!expected
);
262 VM_BUG_ON(!replacement
);
263 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
266 item
= radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
);
267 if (item
!= expected
)
269 radix_tree_replace_slot(pslot
, replacement
);
274 * Sometimes, before we decide whether to proceed or to fail, we must check
275 * that an entry was not already brought back from swap by a racing thread.
277 * Checking page is not enough: by the time a SwapCache page is locked, it
278 * might be reused, and again be SwapCache, using the same swap as before.
280 static bool shmem_confirm_swap(struct address_space
*mapping
,
281 pgoff_t index
, swp_entry_t swap
)
286 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
288 return item
== swp_to_radix_entry(swap
);
292 * Like add_to_page_cache_locked, but error if expected item has gone.
294 static int shmem_add_to_page_cache(struct page
*page
,
295 struct address_space
*mapping
,
296 pgoff_t index
, void *expected
)
300 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
301 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
303 page_cache_get(page
);
304 page
->mapping
= mapping
;
307 spin_lock_irq(&mapping
->tree_lock
);
309 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
311 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
315 __inc_zone_page_state(page
, NR_FILE_PAGES
);
316 __inc_zone_page_state(page
, NR_SHMEM
);
317 spin_unlock_irq(&mapping
->tree_lock
);
319 page
->mapping
= NULL
;
320 spin_unlock_irq(&mapping
->tree_lock
);
321 page_cache_release(page
);
327 * Like delete_from_page_cache, but substitutes swap for page.
329 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
331 struct address_space
*mapping
= page
->mapping
;
334 spin_lock_irq(&mapping
->tree_lock
);
335 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
336 page
->mapping
= NULL
;
338 __dec_zone_page_state(page
, NR_FILE_PAGES
);
339 __dec_zone_page_state(page
, NR_SHMEM
);
340 spin_unlock_irq(&mapping
->tree_lock
);
341 page_cache_release(page
);
346 * Remove swap entry from radix tree, free the swap and its page cache.
348 static int shmem_free_swap(struct address_space
*mapping
,
349 pgoff_t index
, void *radswap
)
353 spin_lock_irq(&mapping
->tree_lock
);
354 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
355 spin_unlock_irq(&mapping
->tree_lock
);
358 free_swap_and_cache(radix_to_swp_entry(radswap
));
363 * Determine (in bytes) how many of the shmem object's pages mapped by the
364 * given offsets are swapped out.
366 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
367 * as long as the inode doesn't go away and racy results are not a problem.
369 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
370 pgoff_t start
, pgoff_t end
)
372 struct radix_tree_iter iter
;
375 unsigned long swapped
= 0;
380 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
381 if (iter
.index
>= end
)
384 page
= radix_tree_deref_slot(slot
);
387 * This should only be possible to happen at index 0, so we
388 * don't need to reset the counter, nor do we risk infinite
391 if (radix_tree_deref_retry(page
))
394 if (radix_tree_exceptional_entry(page
))
397 if (need_resched()) {
399 start
= iter
.index
+ 1;
406 return swapped
<< PAGE_SHIFT
;
410 * Determine (in bytes) how many of the shmem object's pages mapped by the
411 * given vma is swapped out.
413 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
414 * as long as the inode doesn't go away and racy results are not a problem.
416 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
418 struct inode
*inode
= file_inode(vma
->vm_file
);
419 struct shmem_inode_info
*info
= SHMEM_I(inode
);
420 struct address_space
*mapping
= inode
->i_mapping
;
421 unsigned long swapped
;
423 /* Be careful as we don't hold info->lock */
424 swapped
= READ_ONCE(info
->swapped
);
427 * The easier cases are when the shmem object has nothing in swap, or
428 * the vma maps it whole. Then we can simply use the stats that we
434 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
435 return swapped
<< PAGE_SHIFT
;
437 /* Here comes the more involved part */
438 return shmem_partial_swap_usage(mapping
,
439 linear_page_index(vma
, vma
->vm_start
),
440 linear_page_index(vma
, vma
->vm_end
));
444 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
446 void shmem_unlock_mapping(struct address_space
*mapping
)
449 pgoff_t indices
[PAGEVEC_SIZE
];
452 pagevec_init(&pvec
, 0);
454 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
456 while (!mapping_unevictable(mapping
)) {
458 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
459 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
461 pvec
.nr
= find_get_entries(mapping
, index
,
462 PAGEVEC_SIZE
, pvec
.pages
, indices
);
465 index
= indices
[pvec
.nr
- 1] + 1;
466 pagevec_remove_exceptionals(&pvec
);
467 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
468 pagevec_release(&pvec
);
474 * Remove range of pages and swap entries from radix tree, and free them.
475 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
477 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
480 struct address_space
*mapping
= inode
->i_mapping
;
481 struct shmem_inode_info
*info
= SHMEM_I(inode
);
482 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
483 pgoff_t end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
484 unsigned int partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
485 unsigned int partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
487 pgoff_t indices
[PAGEVEC_SIZE
];
488 long nr_swaps_freed
= 0;
493 end
= -1; /* unsigned, so actually very big */
495 pagevec_init(&pvec
, 0);
497 while (index
< end
) {
498 pvec
.nr
= find_get_entries(mapping
, index
,
499 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
500 pvec
.pages
, indices
);
503 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
504 struct page
*page
= pvec
.pages
[i
];
510 if (radix_tree_exceptional_entry(page
)) {
513 nr_swaps_freed
+= !shmem_free_swap(mapping
,
518 if (!trylock_page(page
))
520 if (!unfalloc
|| !PageUptodate(page
)) {
521 if (page
->mapping
== mapping
) {
522 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
523 truncate_inode_page(mapping
, page
);
528 pagevec_remove_exceptionals(&pvec
);
529 pagevec_release(&pvec
);
535 struct page
*page
= NULL
;
536 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
538 unsigned int top
= PAGE_CACHE_SIZE
;
543 zero_user_segment(page
, partial_start
, top
);
544 set_page_dirty(page
);
546 page_cache_release(page
);
550 struct page
*page
= NULL
;
551 shmem_getpage(inode
, end
, &page
, SGP_READ
, NULL
);
553 zero_user_segment(page
, 0, partial_end
);
554 set_page_dirty(page
);
556 page_cache_release(page
);
563 while (index
< end
) {
566 pvec
.nr
= find_get_entries(mapping
, index
,
567 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
568 pvec
.pages
, indices
);
570 /* If all gone or hole-punch or unfalloc, we're done */
571 if (index
== start
|| end
!= -1)
573 /* But if truncating, restart to make sure all gone */
577 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
578 struct page
*page
= pvec
.pages
[i
];
584 if (radix_tree_exceptional_entry(page
)) {
587 if (shmem_free_swap(mapping
, index
, page
)) {
588 /* Swap was replaced by page: retry */
597 if (!unfalloc
|| !PageUptodate(page
)) {
598 if (page
->mapping
== mapping
) {
599 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
600 truncate_inode_page(mapping
, page
);
602 /* Page was replaced by swap: retry */
610 pagevec_remove_exceptionals(&pvec
);
611 pagevec_release(&pvec
);
615 spin_lock(&info
->lock
);
616 info
->swapped
-= nr_swaps_freed
;
617 shmem_recalc_inode(inode
);
618 spin_unlock(&info
->lock
);
621 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
623 shmem_undo_range(inode
, lstart
, lend
, false);
624 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
626 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
628 static int shmem_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
631 struct inode
*inode
= dentry
->d_inode
;
632 struct shmem_inode_info
*info
= SHMEM_I(inode
);
634 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
635 spin_lock(&info
->lock
);
636 shmem_recalc_inode(inode
);
637 spin_unlock(&info
->lock
);
639 generic_fillattr(inode
, stat
);
643 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
645 struct inode
*inode
= d_inode(dentry
);
646 struct shmem_inode_info
*info
= SHMEM_I(inode
);
649 error
= inode_change_ok(inode
, attr
);
653 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
654 loff_t oldsize
= inode
->i_size
;
655 loff_t newsize
= attr
->ia_size
;
657 /* protected by i_mutex */
658 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
659 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
662 if (newsize
!= oldsize
) {
663 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
667 i_size_write(inode
, newsize
);
668 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
670 if (newsize
<= oldsize
) {
671 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
672 if (oldsize
> holebegin
)
673 unmap_mapping_range(inode
->i_mapping
,
676 shmem_truncate_range(inode
,
677 newsize
, (loff_t
)-1);
678 /* unmap again to remove racily COWed private pages */
679 if (oldsize
> holebegin
)
680 unmap_mapping_range(inode
->i_mapping
,
685 setattr_copy(inode
, attr
);
686 if (attr
->ia_valid
& ATTR_MODE
)
687 error
= posix_acl_chmod(inode
, inode
->i_mode
);
691 static void shmem_evict_inode(struct inode
*inode
)
693 struct shmem_inode_info
*info
= SHMEM_I(inode
);
695 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
696 shmem_unacct_size(info
->flags
, inode
->i_size
);
698 shmem_truncate_range(inode
, 0, (loff_t
)-1);
699 if (!list_empty(&info
->swaplist
)) {
700 mutex_lock(&shmem_swaplist_mutex
);
701 list_del_init(&info
->swaplist
);
702 mutex_unlock(&shmem_swaplist_mutex
);
705 kfree(info
->symlink
);
707 simple_xattrs_free(&info
->xattrs
);
708 WARN_ON(inode
->i_blocks
);
709 shmem_free_inode(inode
->i_sb
);
714 * If swap found in inode, free it and move page from swapcache to filecache.
716 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
717 swp_entry_t swap
, struct page
**pagep
)
719 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
725 radswap
= swp_to_radix_entry(swap
);
726 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
728 return -EAGAIN
; /* tell shmem_unuse we found nothing */
731 * Move _head_ to start search for next from here.
732 * But be careful: shmem_evict_inode checks list_empty without taking
733 * mutex, and there's an instant in list_move_tail when info->swaplist
734 * would appear empty, if it were the only one on shmem_swaplist.
736 if (shmem_swaplist
.next
!= &info
->swaplist
)
737 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
739 gfp
= mapping_gfp_mask(mapping
);
740 if (shmem_should_replace_page(*pagep
, gfp
)) {
741 mutex_unlock(&shmem_swaplist_mutex
);
742 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
743 mutex_lock(&shmem_swaplist_mutex
);
745 * We needed to drop mutex to make that restrictive page
746 * allocation, but the inode might have been freed while we
747 * dropped it: although a racing shmem_evict_inode() cannot
748 * complete without emptying the radix_tree, our page lock
749 * on this swapcache page is not enough to prevent that -
750 * free_swap_and_cache() of our swap entry will only
751 * trylock_page(), removing swap from radix_tree whatever.
753 * We must not proceed to shmem_add_to_page_cache() if the
754 * inode has been freed, but of course we cannot rely on
755 * inode or mapping or info to check that. However, we can
756 * safely check if our swap entry is still in use (and here
757 * it can't have got reused for another page): if it's still
758 * in use, then the inode cannot have been freed yet, and we
759 * can safely proceed (if it's no longer in use, that tells
760 * nothing about the inode, but we don't need to unuse swap).
762 if (!page_swapcount(*pagep
))
767 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
768 * but also to hold up shmem_evict_inode(): so inode cannot be freed
769 * beneath us (pagelock doesn't help until the page is in pagecache).
772 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
774 if (error
!= -ENOMEM
) {
776 * Truncation and eviction use free_swap_and_cache(), which
777 * only does trylock page: if we raced, best clean up here.
779 delete_from_swap_cache(*pagep
);
780 set_page_dirty(*pagep
);
782 spin_lock(&info
->lock
);
784 spin_unlock(&info
->lock
);
792 * Search through swapped inodes to find and replace swap by page.
794 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
796 struct list_head
*this, *next
;
797 struct shmem_inode_info
*info
;
798 struct mem_cgroup
*memcg
;
802 * There's a faint possibility that swap page was replaced before
803 * caller locked it: caller will come back later with the right page.
805 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
809 * Charge page using GFP_KERNEL while we can wait, before taking
810 * the shmem_swaplist_mutex which might hold up shmem_writepage().
811 * Charged back to the user (not to caller) when swap account is used.
813 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
,
817 /* No radix_tree_preload: swap entry keeps a place for page in tree */
820 mutex_lock(&shmem_swaplist_mutex
);
821 list_for_each_safe(this, next
, &shmem_swaplist
) {
822 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
824 error
= shmem_unuse_inode(info
, swap
, &page
);
826 list_del_init(&info
->swaplist
);
828 if (error
!= -EAGAIN
)
830 /* found nothing in this: move on to search the next */
832 mutex_unlock(&shmem_swaplist_mutex
);
835 if (error
!= -ENOMEM
)
837 mem_cgroup_cancel_charge(page
, memcg
, false);
839 mem_cgroup_commit_charge(page
, memcg
, true, false);
842 page_cache_release(page
);
847 * Move the page from the page cache to the swap cache.
849 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
851 struct shmem_inode_info
*info
;
852 struct address_space
*mapping
;
857 BUG_ON(!PageLocked(page
));
858 mapping
= page
->mapping
;
860 inode
= mapping
->host
;
861 info
= SHMEM_I(inode
);
862 if (info
->flags
& VM_LOCKED
)
864 if (!total_swap_pages
)
868 * Our capabilities prevent regular writeback or sync from ever calling
869 * shmem_writepage; but a stacking filesystem might use ->writepage of
870 * its underlying filesystem, in which case tmpfs should write out to
871 * swap only in response to memory pressure, and not for the writeback
874 if (!wbc
->for_reclaim
) {
875 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
880 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
881 * value into swapfile.c, the only way we can correctly account for a
882 * fallocated page arriving here is now to initialize it and write it.
884 * That's okay for a page already fallocated earlier, but if we have
885 * not yet completed the fallocation, then (a) we want to keep track
886 * of this page in case we have to undo it, and (b) it may not be a
887 * good idea to continue anyway, once we're pushing into swap. So
888 * reactivate the page, and let shmem_fallocate() quit when too many.
890 if (!PageUptodate(page
)) {
891 if (inode
->i_private
) {
892 struct shmem_falloc
*shmem_falloc
;
893 spin_lock(&inode
->i_lock
);
894 shmem_falloc
= inode
->i_private
;
896 !shmem_falloc
->waitq
&&
897 index
>= shmem_falloc
->start
&&
898 index
< shmem_falloc
->next
)
899 shmem_falloc
->nr_unswapped
++;
902 spin_unlock(&inode
->i_lock
);
906 clear_highpage(page
);
907 flush_dcache_page(page
);
908 SetPageUptodate(page
);
911 swap
= get_swap_page();
915 if (mem_cgroup_try_charge_swap(page
, swap
))
919 * Add inode to shmem_unuse()'s list of swapped-out inodes,
920 * if it's not already there. Do it now before the page is
921 * moved to swap cache, when its pagelock no longer protects
922 * the inode from eviction. But don't unlock the mutex until
923 * we've incremented swapped, because shmem_unuse_inode() will
924 * prune a !swapped inode from the swaplist under this mutex.
926 mutex_lock(&shmem_swaplist_mutex
);
927 if (list_empty(&info
->swaplist
))
928 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
930 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
931 spin_lock(&info
->lock
);
932 shmem_recalc_inode(inode
);
934 spin_unlock(&info
->lock
);
936 swap_shmem_alloc(swap
);
937 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
939 mutex_unlock(&shmem_swaplist_mutex
);
940 BUG_ON(page_mapped(page
));
941 swap_writepage(page
, wbc
);
945 mutex_unlock(&shmem_swaplist_mutex
);
947 swapcache_free(swap
);
949 set_page_dirty(page
);
950 if (wbc
->for_reclaim
)
951 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
958 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
962 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
963 return; /* show nothing */
965 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
967 seq_printf(seq
, ",mpol=%s", buffer
);
970 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
972 struct mempolicy
*mpol
= NULL
;
974 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
977 spin_unlock(&sbinfo
->stat_lock
);
981 #endif /* CONFIG_TMPFS */
983 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
984 struct shmem_inode_info
*info
, pgoff_t index
)
986 struct vm_area_struct pvma
;
989 /* Create a pseudo vma that just contains the policy */
991 /* Bias interleave by inode number to distribute better across nodes */
992 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
994 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
996 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
998 /* Drop reference taken by mpol_shared_policy_lookup() */
999 mpol_cond_put(pvma
.vm_policy
);
1004 static struct page
*shmem_alloc_page(gfp_t gfp
,
1005 struct shmem_inode_info
*info
, pgoff_t index
)
1007 struct vm_area_struct pvma
;
1010 /* Create a pseudo vma that just contains the policy */
1012 /* Bias interleave by inode number to distribute better across nodes */
1013 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1015 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1017 page
= alloc_page_vma(gfp
, &pvma
, 0);
1019 /* Drop reference taken by mpol_shared_policy_lookup() */
1020 mpol_cond_put(pvma
.vm_policy
);
1024 #else /* !CONFIG_NUMA */
1026 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1029 #endif /* CONFIG_TMPFS */
1031 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1032 struct shmem_inode_info
*info
, pgoff_t index
)
1034 return swapin_readahead(swap
, gfp
, NULL
, 0);
1037 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
1038 struct shmem_inode_info
*info
, pgoff_t index
)
1040 return alloc_page(gfp
);
1042 #endif /* CONFIG_NUMA */
1044 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1045 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1052 * When a page is moved from swapcache to shmem filecache (either by the
1053 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1054 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1055 * ignorance of the mapping it belongs to. If that mapping has special
1056 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1057 * we may need to copy to a suitable page before moving to filecache.
1059 * In a future release, this may well be extended to respect cpuset and
1060 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1061 * but for now it is a simple matter of zone.
1063 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1065 return page_zonenum(page
) > gfp_zone(gfp
);
1068 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1069 struct shmem_inode_info
*info
, pgoff_t index
)
1071 struct page
*oldpage
, *newpage
;
1072 struct address_space
*swap_mapping
;
1077 swap_index
= page_private(oldpage
);
1078 swap_mapping
= page_mapping(oldpage
);
1081 * We have arrived here because our zones are constrained, so don't
1082 * limit chance of success by further cpuset and node constraints.
1084 gfp
&= ~GFP_CONSTRAINT_MASK
;
1085 newpage
= shmem_alloc_page(gfp
, info
, index
);
1089 page_cache_get(newpage
);
1090 copy_highpage(newpage
, oldpage
);
1091 flush_dcache_page(newpage
);
1093 __SetPageLocked(newpage
);
1094 SetPageUptodate(newpage
);
1095 SetPageSwapBacked(newpage
);
1096 set_page_private(newpage
, swap_index
);
1097 SetPageSwapCache(newpage
);
1100 * Our caller will very soon move newpage out of swapcache, but it's
1101 * a nice clean interface for us to replace oldpage by newpage there.
1103 spin_lock_irq(&swap_mapping
->tree_lock
);
1104 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1107 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
1108 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
1110 spin_unlock_irq(&swap_mapping
->tree_lock
);
1112 if (unlikely(error
)) {
1114 * Is this possible? I think not, now that our callers check
1115 * both PageSwapCache and page_private after getting page lock;
1116 * but be defensive. Reverse old to newpage for clear and free.
1120 mem_cgroup_replace_page(oldpage
, newpage
);
1121 lru_cache_add_anon(newpage
);
1125 ClearPageSwapCache(oldpage
);
1126 set_page_private(oldpage
, 0);
1128 unlock_page(oldpage
);
1129 page_cache_release(oldpage
);
1130 page_cache_release(oldpage
);
1135 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1137 * If we allocate a new one we do not mark it dirty. That's up to the
1138 * vm. If we swap it in we mark it dirty since we also free the swap
1139 * entry since a page cannot live in both the swap and page cache
1141 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1142 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
1144 struct address_space
*mapping
= inode
->i_mapping
;
1145 struct shmem_inode_info
*info
;
1146 struct shmem_sb_info
*sbinfo
;
1147 struct mem_cgroup
*memcg
;
1154 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
1158 page
= find_lock_entry(mapping
, index
);
1159 if (radix_tree_exceptional_entry(page
)) {
1160 swap
= radix_to_swp_entry(page
);
1164 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1165 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1170 if (page
&& sgp
== SGP_WRITE
)
1171 mark_page_accessed(page
);
1173 /* fallocated page? */
1174 if (page
&& !PageUptodate(page
)) {
1175 if (sgp
!= SGP_READ
)
1178 page_cache_release(page
);
1181 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1187 * Fast cache lookup did not find it:
1188 * bring it back from swap or allocate.
1190 info
= SHMEM_I(inode
);
1191 sbinfo
= SHMEM_SB(inode
->i_sb
);
1194 /* Look it up and read it in.. */
1195 page
= lookup_swap_cache(swap
);
1197 /* here we actually do the io */
1199 *fault_type
|= VM_FAULT_MAJOR
;
1200 page
= shmem_swapin(swap
, gfp
, info
, index
);
1207 /* We have to do this with page locked to prevent races */
1209 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1210 !shmem_confirm_swap(mapping
, index
, swap
)) {
1211 error
= -EEXIST
; /* try again */
1214 if (!PageUptodate(page
)) {
1218 wait_on_page_writeback(page
);
1220 if (shmem_should_replace_page(page
, gfp
)) {
1221 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1226 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
,
1229 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1230 swp_to_radix_entry(swap
));
1232 * We already confirmed swap under page lock, and make
1233 * no memory allocation here, so usually no possibility
1234 * of error; but free_swap_and_cache() only trylocks a
1235 * page, so it is just possible that the entry has been
1236 * truncated or holepunched since swap was confirmed.
1237 * shmem_undo_range() will have done some of the
1238 * unaccounting, now delete_from_swap_cache() will do
1240 * Reset swap.val? No, leave it so "failed" goes back to
1241 * "repeat": reading a hole and writing should succeed.
1244 mem_cgroup_cancel_charge(page
, memcg
, false);
1245 delete_from_swap_cache(page
);
1251 mem_cgroup_commit_charge(page
, memcg
, true, false);
1253 spin_lock(&info
->lock
);
1255 shmem_recalc_inode(inode
);
1256 spin_unlock(&info
->lock
);
1258 if (sgp
== SGP_WRITE
)
1259 mark_page_accessed(page
);
1261 delete_from_swap_cache(page
);
1262 set_page_dirty(page
);
1266 if (shmem_acct_block(info
->flags
)) {
1270 if (sbinfo
->max_blocks
) {
1271 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1272 sbinfo
->max_blocks
) >= 0) {
1276 percpu_counter_inc(&sbinfo
->used_blocks
);
1279 page
= shmem_alloc_page(gfp
, info
, index
);
1285 __SetPageSwapBacked(page
);
1286 __SetPageLocked(page
);
1287 if (sgp
== SGP_WRITE
)
1288 __SetPageReferenced(page
);
1290 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
,
1294 error
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
1296 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1298 radix_tree_preload_end();
1301 mem_cgroup_cancel_charge(page
, memcg
, false);
1304 mem_cgroup_commit_charge(page
, memcg
, false, false);
1305 lru_cache_add_anon(page
);
1307 spin_lock(&info
->lock
);
1309 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1310 shmem_recalc_inode(inode
);
1311 spin_unlock(&info
->lock
);
1315 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1317 if (sgp
== SGP_FALLOC
)
1321 * Let SGP_WRITE caller clear ends if write does not fill page;
1322 * but SGP_FALLOC on a page fallocated earlier must initialize
1323 * it now, lest undo on failure cancel our earlier guarantee.
1325 if (sgp
!= SGP_WRITE
) {
1326 clear_highpage(page
);
1327 flush_dcache_page(page
);
1328 SetPageUptodate(page
);
1330 if (sgp
== SGP_DIRTY
)
1331 set_page_dirty(page
);
1334 /* Perhaps the file has been truncated since we checked */
1335 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1336 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1338 ClearPageDirty(page
);
1339 delete_from_page_cache(page
);
1340 spin_lock(&info
->lock
);
1341 shmem_recalc_inode(inode
);
1342 spin_unlock(&info
->lock
);
1354 if (sbinfo
->max_blocks
)
1355 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1357 shmem_unacct_blocks(info
->flags
, 1);
1359 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1364 page_cache_release(page
);
1366 if (error
== -ENOSPC
&& !once
++) {
1367 info
= SHMEM_I(inode
);
1368 spin_lock(&info
->lock
);
1369 shmem_recalc_inode(inode
);
1370 spin_unlock(&info
->lock
);
1373 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1378 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1380 struct inode
*inode
= file_inode(vma
->vm_file
);
1382 int ret
= VM_FAULT_LOCKED
;
1385 * Trinity finds that probing a hole which tmpfs is punching can
1386 * prevent the hole-punch from ever completing: which in turn
1387 * locks writers out with its hold on i_mutex. So refrain from
1388 * faulting pages into the hole while it's being punched. Although
1389 * shmem_undo_range() does remove the additions, it may be unable to
1390 * keep up, as each new page needs its own unmap_mapping_range() call,
1391 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1393 * It does not matter if we sometimes reach this check just before the
1394 * hole-punch begins, so that one fault then races with the punch:
1395 * we just need to make racing faults a rare case.
1397 * The implementation below would be much simpler if we just used a
1398 * standard mutex or completion: but we cannot take i_mutex in fault,
1399 * and bloating every shmem inode for this unlikely case would be sad.
1401 if (unlikely(inode
->i_private
)) {
1402 struct shmem_falloc
*shmem_falloc
;
1404 spin_lock(&inode
->i_lock
);
1405 shmem_falloc
= inode
->i_private
;
1407 shmem_falloc
->waitq
&&
1408 vmf
->pgoff
>= shmem_falloc
->start
&&
1409 vmf
->pgoff
< shmem_falloc
->next
) {
1410 wait_queue_head_t
*shmem_falloc_waitq
;
1411 DEFINE_WAIT(shmem_fault_wait
);
1413 ret
= VM_FAULT_NOPAGE
;
1414 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1415 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1416 /* It's polite to up mmap_sem if we can */
1417 up_read(&vma
->vm_mm
->mmap_sem
);
1418 ret
= VM_FAULT_RETRY
;
1421 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1422 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1423 TASK_UNINTERRUPTIBLE
);
1424 spin_unlock(&inode
->i_lock
);
1428 * shmem_falloc_waitq points into the shmem_fallocate()
1429 * stack of the hole-punching task: shmem_falloc_waitq
1430 * is usually invalid by the time we reach here, but
1431 * finish_wait() does not dereference it in that case;
1432 * though i_lock needed lest racing with wake_up_all().
1434 spin_lock(&inode
->i_lock
);
1435 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1436 spin_unlock(&inode
->i_lock
);
1439 spin_unlock(&inode
->i_lock
);
1442 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1444 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1446 if (ret
& VM_FAULT_MAJOR
) {
1447 count_vm_event(PGMAJFAULT
);
1448 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1454 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1456 struct inode
*inode
= file_inode(vma
->vm_file
);
1457 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1460 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1463 struct inode
*inode
= file_inode(vma
->vm_file
);
1466 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1467 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1471 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1473 struct inode
*inode
= file_inode(file
);
1474 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1475 int retval
= -ENOMEM
;
1477 spin_lock(&info
->lock
);
1478 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1479 if (!user_shm_lock(inode
->i_size
, user
))
1481 info
->flags
|= VM_LOCKED
;
1482 mapping_set_unevictable(file
->f_mapping
);
1484 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1485 user_shm_unlock(inode
->i_size
, user
);
1486 info
->flags
&= ~VM_LOCKED
;
1487 mapping_clear_unevictable(file
->f_mapping
);
1492 spin_unlock(&info
->lock
);
1496 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1498 file_accessed(file
);
1499 vma
->vm_ops
= &shmem_vm_ops
;
1503 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1504 umode_t mode
, dev_t dev
, unsigned long flags
)
1506 struct inode
*inode
;
1507 struct shmem_inode_info
*info
;
1508 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1510 if (shmem_reserve_inode(sb
))
1513 inode
= new_inode(sb
);
1515 inode
->i_ino
= get_next_ino();
1516 inode_init_owner(inode
, dir
, mode
);
1517 inode
->i_blocks
= 0;
1518 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1519 inode
->i_generation
= get_seconds();
1520 info
= SHMEM_I(inode
);
1521 memset(info
, 0, (char *)inode
- (char *)info
);
1522 spin_lock_init(&info
->lock
);
1523 info
->seals
= F_SEAL_SEAL
;
1524 info
->flags
= flags
& VM_NORESERVE
;
1525 INIT_LIST_HEAD(&info
->swaplist
);
1526 simple_xattrs_init(&info
->xattrs
);
1527 cache_no_acl(inode
);
1529 switch (mode
& S_IFMT
) {
1531 inode
->i_op
= &shmem_special_inode_operations
;
1532 init_special_inode(inode
, mode
, dev
);
1535 inode
->i_mapping
->a_ops
= &shmem_aops
;
1536 inode
->i_op
= &shmem_inode_operations
;
1537 inode
->i_fop
= &shmem_file_operations
;
1538 mpol_shared_policy_init(&info
->policy
,
1539 shmem_get_sbmpol(sbinfo
));
1543 /* Some things misbehave if size == 0 on a directory */
1544 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1545 inode
->i_op
= &shmem_dir_inode_operations
;
1546 inode
->i_fop
= &simple_dir_operations
;
1550 * Must not load anything in the rbtree,
1551 * mpol_free_shared_policy will not be called.
1553 mpol_shared_policy_init(&info
->policy
, NULL
);
1557 shmem_free_inode(sb
);
1561 bool shmem_mapping(struct address_space
*mapping
)
1566 return mapping
->host
->i_sb
->s_op
== &shmem_ops
;
1570 static const struct inode_operations shmem_symlink_inode_operations
;
1571 static const struct inode_operations shmem_short_symlink_operations
;
1573 #ifdef CONFIG_TMPFS_XATTR
1574 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1576 #define shmem_initxattrs NULL
1580 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1581 loff_t pos
, unsigned len
, unsigned flags
,
1582 struct page
**pagep
, void **fsdata
)
1584 struct inode
*inode
= mapping
->host
;
1585 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1586 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1588 /* i_mutex is held by caller */
1589 if (unlikely(info
->seals
)) {
1590 if (info
->seals
& F_SEAL_WRITE
)
1592 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
1596 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1600 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1601 loff_t pos
, unsigned len
, unsigned copied
,
1602 struct page
*page
, void *fsdata
)
1604 struct inode
*inode
= mapping
->host
;
1606 if (pos
+ copied
> inode
->i_size
)
1607 i_size_write(inode
, pos
+ copied
);
1609 if (!PageUptodate(page
)) {
1610 if (copied
< PAGE_CACHE_SIZE
) {
1611 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1612 zero_user_segments(page
, 0, from
,
1613 from
+ copied
, PAGE_CACHE_SIZE
);
1615 SetPageUptodate(page
);
1617 set_page_dirty(page
);
1619 page_cache_release(page
);
1624 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1626 struct file
*file
= iocb
->ki_filp
;
1627 struct inode
*inode
= file_inode(file
);
1628 struct address_space
*mapping
= inode
->i_mapping
;
1630 unsigned long offset
;
1631 enum sgp_type sgp
= SGP_READ
;
1634 loff_t
*ppos
= &iocb
->ki_pos
;
1637 * Might this read be for a stacking filesystem? Then when reading
1638 * holes of a sparse file, we actually need to allocate those pages,
1639 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1641 if (!iter_is_iovec(to
))
1644 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1645 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1648 struct page
*page
= NULL
;
1650 unsigned long nr
, ret
;
1651 loff_t i_size
= i_size_read(inode
);
1653 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1654 if (index
> end_index
)
1656 if (index
== end_index
) {
1657 nr
= i_size
& ~PAGE_CACHE_MASK
;
1662 error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1664 if (error
== -EINVAL
)
1672 * We must evaluate after, since reads (unlike writes)
1673 * are called without i_mutex protection against truncate
1675 nr
= PAGE_CACHE_SIZE
;
1676 i_size
= i_size_read(inode
);
1677 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1678 if (index
== end_index
) {
1679 nr
= i_size
& ~PAGE_CACHE_MASK
;
1682 page_cache_release(page
);
1690 * If users can be writing to this page using arbitrary
1691 * virtual addresses, take care about potential aliasing
1692 * before reading the page on the kernel side.
1694 if (mapping_writably_mapped(mapping
))
1695 flush_dcache_page(page
);
1697 * Mark the page accessed if we read the beginning.
1700 mark_page_accessed(page
);
1702 page
= ZERO_PAGE(0);
1703 page_cache_get(page
);
1707 * Ok, we have the page, and it's up-to-date, so
1708 * now we can copy it to user space...
1710 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
1713 index
+= offset
>> PAGE_CACHE_SHIFT
;
1714 offset
&= ~PAGE_CACHE_MASK
;
1716 page_cache_release(page
);
1717 if (!iov_iter_count(to
))
1726 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1727 file_accessed(file
);
1728 return retval
? retval
: error
;
1731 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1732 struct pipe_inode_info
*pipe
, size_t len
,
1735 struct address_space
*mapping
= in
->f_mapping
;
1736 struct inode
*inode
= mapping
->host
;
1737 unsigned int loff
, nr_pages
, req_pages
;
1738 struct page
*pages
[PIPE_DEF_BUFFERS
];
1739 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1741 pgoff_t index
, end_index
;
1744 struct splice_pipe_desc spd
= {
1747 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1749 .ops
= &page_cache_pipe_buf_ops
,
1750 .spd_release
= spd_release_page
,
1753 isize
= i_size_read(inode
);
1754 if (unlikely(*ppos
>= isize
))
1757 left
= isize
- *ppos
;
1758 if (unlikely(left
< len
))
1761 if (splice_grow_spd(pipe
, &spd
))
1764 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1765 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1766 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1767 nr_pages
= min(req_pages
, spd
.nr_pages_max
);
1769 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1770 nr_pages
, spd
.pages
);
1771 index
+= spd
.nr_pages
;
1774 while (spd
.nr_pages
< nr_pages
) {
1775 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1779 spd
.pages
[spd
.nr_pages
++] = page
;
1783 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1784 nr_pages
= spd
.nr_pages
;
1787 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1788 unsigned int this_len
;
1793 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1794 page
= spd
.pages
[page_nr
];
1796 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1797 error
= shmem_getpage(inode
, index
, &page
,
1802 page_cache_release(spd
.pages
[page_nr
]);
1803 spd
.pages
[page_nr
] = page
;
1806 isize
= i_size_read(inode
);
1807 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1808 if (unlikely(!isize
|| index
> end_index
))
1811 if (end_index
== index
) {
1814 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1818 this_len
= min(this_len
, plen
- loff
);
1822 spd
.partial
[page_nr
].offset
= loff
;
1823 spd
.partial
[page_nr
].len
= this_len
;
1830 while (page_nr
< nr_pages
)
1831 page_cache_release(spd
.pages
[page_nr
++]);
1834 error
= splice_to_pipe(pipe
, &spd
);
1836 splice_shrink_spd(&spd
);
1846 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1848 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
1849 pgoff_t index
, pgoff_t end
, int whence
)
1852 struct pagevec pvec
;
1853 pgoff_t indices
[PAGEVEC_SIZE
];
1857 pagevec_init(&pvec
, 0);
1858 pvec
.nr
= 1; /* start small: we may be there already */
1860 pvec
.nr
= find_get_entries(mapping
, index
,
1861 pvec
.nr
, pvec
.pages
, indices
);
1863 if (whence
== SEEK_DATA
)
1867 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
1868 if (index
< indices
[i
]) {
1869 if (whence
== SEEK_HOLE
) {
1875 page
= pvec
.pages
[i
];
1876 if (page
&& !radix_tree_exceptional_entry(page
)) {
1877 if (!PageUptodate(page
))
1881 (page
&& whence
== SEEK_DATA
) ||
1882 (!page
&& whence
== SEEK_HOLE
)) {
1887 pagevec_remove_exceptionals(&pvec
);
1888 pagevec_release(&pvec
);
1889 pvec
.nr
= PAGEVEC_SIZE
;
1895 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
1897 struct address_space
*mapping
= file
->f_mapping
;
1898 struct inode
*inode
= mapping
->host
;
1902 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
1903 return generic_file_llseek_size(file
, offset
, whence
,
1904 MAX_LFS_FILESIZE
, i_size_read(inode
));
1905 mutex_lock(&inode
->i_mutex
);
1906 /* We're holding i_mutex so we can access i_size directly */
1910 else if (offset
>= inode
->i_size
)
1913 start
= offset
>> PAGE_CACHE_SHIFT
;
1914 end
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1915 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
1916 new_offset
<<= PAGE_CACHE_SHIFT
;
1917 if (new_offset
> offset
) {
1918 if (new_offset
< inode
->i_size
)
1919 offset
= new_offset
;
1920 else if (whence
== SEEK_DATA
)
1923 offset
= inode
->i_size
;
1928 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
1929 mutex_unlock(&inode
->i_mutex
);
1934 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1935 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1937 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1938 #define LAST_SCAN 4 /* about 150ms max */
1940 static void shmem_tag_pins(struct address_space
*mapping
)
1942 struct radix_tree_iter iter
;
1952 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1953 page
= radix_tree_deref_slot(slot
);
1954 if (!page
|| radix_tree_exception(page
)) {
1955 if (radix_tree_deref_retry(page
))
1957 } else if (page_count(page
) - page_mapcount(page
) > 1) {
1958 spin_lock_irq(&mapping
->tree_lock
);
1959 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
1961 spin_unlock_irq(&mapping
->tree_lock
);
1964 if (need_resched()) {
1966 start
= iter
.index
+ 1;
1974 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1975 * via get_user_pages(), drivers might have some pending I/O without any active
1976 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1977 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1978 * them to be dropped.
1979 * The caller must guarantee that no new user will acquire writable references
1980 * to those pages to avoid races.
1982 static int shmem_wait_for_pins(struct address_space
*mapping
)
1984 struct radix_tree_iter iter
;
1990 shmem_tag_pins(mapping
);
1993 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
1994 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
1998 lru_add_drain_all();
1999 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2005 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2006 start
, SHMEM_TAG_PINNED
) {
2008 page
= radix_tree_deref_slot(slot
);
2009 if (radix_tree_exception(page
)) {
2010 if (radix_tree_deref_retry(page
))
2017 page_count(page
) - page_mapcount(page
) != 1) {
2018 if (scan
< LAST_SCAN
)
2019 goto continue_resched
;
2022 * On the last scan, we clean up all those tags
2023 * we inserted; but make a note that we still
2024 * found pages pinned.
2029 spin_lock_irq(&mapping
->tree_lock
);
2030 radix_tree_tag_clear(&mapping
->page_tree
,
2031 iter
.index
, SHMEM_TAG_PINNED
);
2032 spin_unlock_irq(&mapping
->tree_lock
);
2034 if (need_resched()) {
2036 start
= iter
.index
+ 1;
2046 #define F_ALL_SEALS (F_SEAL_SEAL | \
2051 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2053 struct inode
*inode
= file_inode(file
);
2054 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2059 * Sealing allows multiple parties to share a shmem-file but restrict
2060 * access to a specific subset of file operations. Seals can only be
2061 * added, but never removed. This way, mutually untrusted parties can
2062 * share common memory regions with a well-defined policy. A malicious
2063 * peer can thus never perform unwanted operations on a shared object.
2065 * Seals are only supported on special shmem-files and always affect
2066 * the whole underlying inode. Once a seal is set, it may prevent some
2067 * kinds of access to the file. Currently, the following seals are
2069 * SEAL_SEAL: Prevent further seals from being set on this file
2070 * SEAL_SHRINK: Prevent the file from shrinking
2071 * SEAL_GROW: Prevent the file from growing
2072 * SEAL_WRITE: Prevent write access to the file
2074 * As we don't require any trust relationship between two parties, we
2075 * must prevent seals from being removed. Therefore, sealing a file
2076 * only adds a given set of seals to the file, it never touches
2077 * existing seals. Furthermore, the "setting seals"-operation can be
2078 * sealed itself, which basically prevents any further seal from being
2081 * Semantics of sealing are only defined on volatile files. Only
2082 * anonymous shmem files support sealing. More importantly, seals are
2083 * never written to disk. Therefore, there's no plan to support it on
2087 if (file
->f_op
!= &shmem_file_operations
)
2089 if (!(file
->f_mode
& FMODE_WRITE
))
2091 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2094 mutex_lock(&inode
->i_mutex
);
2096 if (info
->seals
& F_SEAL_SEAL
) {
2101 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2102 error
= mapping_deny_writable(file
->f_mapping
);
2106 error
= shmem_wait_for_pins(file
->f_mapping
);
2108 mapping_allow_writable(file
->f_mapping
);
2113 info
->seals
|= seals
;
2117 mutex_unlock(&inode
->i_mutex
);
2120 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2122 int shmem_get_seals(struct file
*file
)
2124 if (file
->f_op
!= &shmem_file_operations
)
2127 return SHMEM_I(file_inode(file
))->seals
;
2129 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2131 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2137 /* disallow upper 32bit */
2141 error
= shmem_add_seals(file
, arg
);
2144 error
= shmem_get_seals(file
);
2154 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2157 struct inode
*inode
= file_inode(file
);
2158 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2159 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2160 struct shmem_falloc shmem_falloc
;
2161 pgoff_t start
, index
, end
;
2164 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2167 mutex_lock(&inode
->i_mutex
);
2169 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2170 struct address_space
*mapping
= file
->f_mapping
;
2171 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2172 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2173 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2175 /* protected by i_mutex */
2176 if (info
->seals
& F_SEAL_WRITE
) {
2181 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2182 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2183 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2184 spin_lock(&inode
->i_lock
);
2185 inode
->i_private
= &shmem_falloc
;
2186 spin_unlock(&inode
->i_lock
);
2188 if ((u64
)unmap_end
> (u64
)unmap_start
)
2189 unmap_mapping_range(mapping
, unmap_start
,
2190 1 + unmap_end
- unmap_start
, 0);
2191 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2192 /* No need to unmap again: hole-punching leaves COWed pages */
2194 spin_lock(&inode
->i_lock
);
2195 inode
->i_private
= NULL
;
2196 wake_up_all(&shmem_falloc_waitq
);
2197 spin_unlock(&inode
->i_lock
);
2202 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2203 error
= inode_newsize_ok(inode
, offset
+ len
);
2207 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2212 start
= offset
>> PAGE_CACHE_SHIFT
;
2213 end
= (offset
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
2214 /* Try to avoid a swapstorm if len is impossible to satisfy */
2215 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2220 shmem_falloc
.waitq
= NULL
;
2221 shmem_falloc
.start
= start
;
2222 shmem_falloc
.next
= start
;
2223 shmem_falloc
.nr_falloced
= 0;
2224 shmem_falloc
.nr_unswapped
= 0;
2225 spin_lock(&inode
->i_lock
);
2226 inode
->i_private
= &shmem_falloc
;
2227 spin_unlock(&inode
->i_lock
);
2229 for (index
= start
; index
< end
; index
++) {
2233 * Good, the fallocate(2) manpage permits EINTR: we may have
2234 * been interrupted because we are using up too much memory.
2236 if (signal_pending(current
))
2238 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2241 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
,
2244 /* Remove the !PageUptodate pages we added */
2245 shmem_undo_range(inode
,
2246 (loff_t
)start
<< PAGE_CACHE_SHIFT
,
2247 (loff_t
)index
<< PAGE_CACHE_SHIFT
, true);
2252 * Inform shmem_writepage() how far we have reached.
2253 * No need for lock or barrier: we have the page lock.
2255 shmem_falloc
.next
++;
2256 if (!PageUptodate(page
))
2257 shmem_falloc
.nr_falloced
++;
2260 * If !PageUptodate, leave it that way so that freeable pages
2261 * can be recognized if we need to rollback on error later.
2262 * But set_page_dirty so that memory pressure will swap rather
2263 * than free the pages we are allocating (and SGP_CACHE pages
2264 * might still be clean: we now need to mark those dirty too).
2266 set_page_dirty(page
);
2268 page_cache_release(page
);
2272 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2273 i_size_write(inode
, offset
+ len
);
2274 inode
->i_ctime
= CURRENT_TIME
;
2276 spin_lock(&inode
->i_lock
);
2277 inode
->i_private
= NULL
;
2278 spin_unlock(&inode
->i_lock
);
2280 mutex_unlock(&inode
->i_mutex
);
2284 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2286 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2288 buf
->f_type
= TMPFS_MAGIC
;
2289 buf
->f_bsize
= PAGE_CACHE_SIZE
;
2290 buf
->f_namelen
= NAME_MAX
;
2291 if (sbinfo
->max_blocks
) {
2292 buf
->f_blocks
= sbinfo
->max_blocks
;
2294 buf
->f_bfree
= sbinfo
->max_blocks
-
2295 percpu_counter_sum(&sbinfo
->used_blocks
);
2297 if (sbinfo
->max_inodes
) {
2298 buf
->f_files
= sbinfo
->max_inodes
;
2299 buf
->f_ffree
= sbinfo
->free_inodes
;
2301 /* else leave those fields 0 like simple_statfs */
2306 * File creation. Allocate an inode, and we're done..
2309 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2311 struct inode
*inode
;
2312 int error
= -ENOSPC
;
2314 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2316 error
= simple_acl_create(dir
, inode
);
2319 error
= security_inode_init_security(inode
, dir
,
2321 shmem_initxattrs
, NULL
);
2322 if (error
&& error
!= -EOPNOTSUPP
)
2326 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2327 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2328 d_instantiate(dentry
, inode
);
2329 dget(dentry
); /* Extra count - pin the dentry in core */
2338 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2340 struct inode
*inode
;
2341 int error
= -ENOSPC
;
2343 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2345 error
= security_inode_init_security(inode
, dir
,
2347 shmem_initxattrs
, NULL
);
2348 if (error
&& error
!= -EOPNOTSUPP
)
2350 error
= simple_acl_create(dir
, inode
);
2353 d_tmpfile(dentry
, inode
);
2361 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2365 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2371 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2374 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2380 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2382 struct inode
*inode
= d_inode(old_dentry
);
2386 * No ordinary (disk based) filesystem counts links as inodes;
2387 * but each new link needs a new dentry, pinning lowmem, and
2388 * tmpfs dentries cannot be pruned until they are unlinked.
2390 ret
= shmem_reserve_inode(inode
->i_sb
);
2394 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2395 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2397 ihold(inode
); /* New dentry reference */
2398 dget(dentry
); /* Extra pinning count for the created dentry */
2399 d_instantiate(dentry
, inode
);
2404 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2406 struct inode
*inode
= d_inode(dentry
);
2408 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2409 shmem_free_inode(inode
->i_sb
);
2411 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2412 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2414 dput(dentry
); /* Undo the count from "create" - this does all the work */
2418 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2420 if (!simple_empty(dentry
))
2423 drop_nlink(d_inode(dentry
));
2425 return shmem_unlink(dir
, dentry
);
2428 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2430 bool old_is_dir
= d_is_dir(old_dentry
);
2431 bool new_is_dir
= d_is_dir(new_dentry
);
2433 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2435 drop_nlink(old_dir
);
2438 drop_nlink(new_dir
);
2442 old_dir
->i_ctime
= old_dir
->i_mtime
=
2443 new_dir
->i_ctime
= new_dir
->i_mtime
=
2444 d_inode(old_dentry
)->i_ctime
=
2445 d_inode(new_dentry
)->i_ctime
= CURRENT_TIME
;
2450 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
2452 struct dentry
*whiteout
;
2455 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2459 error
= shmem_mknod(old_dir
, whiteout
,
2460 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2466 * Cheat and hash the whiteout while the old dentry is still in
2467 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2469 * d_lookup() will consistently find one of them at this point,
2470 * not sure which one, but that isn't even important.
2477 * The VFS layer already does all the dentry stuff for rename,
2478 * we just have to decrement the usage count for the target if
2479 * it exists so that the VFS layer correctly free's it when it
2482 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
2484 struct inode
*inode
= d_inode(old_dentry
);
2485 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2487 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
2490 if (flags
& RENAME_EXCHANGE
)
2491 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
2493 if (!simple_empty(new_dentry
))
2496 if (flags
& RENAME_WHITEOUT
) {
2499 error
= shmem_whiteout(old_dir
, old_dentry
);
2504 if (d_really_is_positive(new_dentry
)) {
2505 (void) shmem_unlink(new_dir
, new_dentry
);
2506 if (they_are_dirs
) {
2507 drop_nlink(d_inode(new_dentry
));
2508 drop_nlink(old_dir
);
2510 } else if (they_are_dirs
) {
2511 drop_nlink(old_dir
);
2515 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2516 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2517 old_dir
->i_ctime
= old_dir
->i_mtime
=
2518 new_dir
->i_ctime
= new_dir
->i_mtime
=
2519 inode
->i_ctime
= CURRENT_TIME
;
2523 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2527 struct inode
*inode
;
2529 struct shmem_inode_info
*info
;
2531 len
= strlen(symname
) + 1;
2532 if (len
> PAGE_CACHE_SIZE
)
2533 return -ENAMETOOLONG
;
2535 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2539 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2540 shmem_initxattrs
, NULL
);
2542 if (error
!= -EOPNOTSUPP
) {
2549 info
= SHMEM_I(inode
);
2550 inode
->i_size
= len
-1;
2551 if (len
<= SHORT_SYMLINK_LEN
) {
2552 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
2553 if (!info
->symlink
) {
2557 inode
->i_op
= &shmem_short_symlink_operations
;
2558 inode
->i_link
= info
->symlink
;
2560 inode_nohighmem(inode
);
2561 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
2566 inode
->i_mapping
->a_ops
= &shmem_aops
;
2567 inode
->i_op
= &shmem_symlink_inode_operations
;
2568 memcpy(page_address(page
), symname
, len
);
2569 SetPageUptodate(page
);
2570 set_page_dirty(page
);
2572 page_cache_release(page
);
2574 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2575 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2576 d_instantiate(dentry
, inode
);
2581 static void shmem_put_link(void *arg
)
2583 mark_page_accessed(arg
);
2587 static const char *shmem_get_link(struct dentry
*dentry
,
2588 struct inode
*inode
,
2589 struct delayed_call
*done
)
2591 struct page
*page
= NULL
;
2594 page
= find_get_page(inode
->i_mapping
, 0);
2596 return ERR_PTR(-ECHILD
);
2597 if (!PageUptodate(page
)) {
2599 return ERR_PTR(-ECHILD
);
2602 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
, NULL
);
2604 return ERR_PTR(error
);
2607 set_delayed_call(done
, shmem_put_link
, page
);
2608 return page_address(page
);
2611 #ifdef CONFIG_TMPFS_XATTR
2613 * Superblocks without xattr inode operations may get some security.* xattr
2614 * support from the LSM "for free". As soon as we have any other xattrs
2615 * like ACLs, we also need to implement the security.* handlers at
2616 * filesystem level, though.
2620 * Callback for security_inode_init_security() for acquiring xattrs.
2622 static int shmem_initxattrs(struct inode
*inode
,
2623 const struct xattr
*xattr_array
,
2626 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2627 const struct xattr
*xattr
;
2628 struct simple_xattr
*new_xattr
;
2631 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2632 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
2636 len
= strlen(xattr
->name
) + 1;
2637 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2639 if (!new_xattr
->name
) {
2644 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2645 XATTR_SECURITY_PREFIX_LEN
);
2646 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2649 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
2655 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
2656 struct dentry
*dentry
, const char *name
,
2657 void *buffer
, size_t size
)
2659 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2661 name
= xattr_full_name(handler
, name
);
2662 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
2665 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
2666 struct dentry
*dentry
, const char *name
,
2667 const void *value
, size_t size
, int flags
)
2669 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2671 name
= xattr_full_name(handler
, name
);
2672 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
2675 static const struct xattr_handler shmem_security_xattr_handler
= {
2676 .prefix
= XATTR_SECURITY_PREFIX
,
2677 .get
= shmem_xattr_handler_get
,
2678 .set
= shmem_xattr_handler_set
,
2681 static const struct xattr_handler shmem_trusted_xattr_handler
= {
2682 .prefix
= XATTR_TRUSTED_PREFIX
,
2683 .get
= shmem_xattr_handler_get
,
2684 .set
= shmem_xattr_handler_set
,
2687 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2688 #ifdef CONFIG_TMPFS_POSIX_ACL
2689 &posix_acl_access_xattr_handler
,
2690 &posix_acl_default_xattr_handler
,
2692 &shmem_security_xattr_handler
,
2693 &shmem_trusted_xattr_handler
,
2697 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2699 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2700 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
2702 #endif /* CONFIG_TMPFS_XATTR */
2704 static const struct inode_operations shmem_short_symlink_operations
= {
2705 .readlink
= generic_readlink
,
2706 .get_link
= simple_get_link
,
2707 #ifdef CONFIG_TMPFS_XATTR
2708 .setxattr
= generic_setxattr
,
2709 .getxattr
= generic_getxattr
,
2710 .listxattr
= shmem_listxattr
,
2711 .removexattr
= generic_removexattr
,
2715 static const struct inode_operations shmem_symlink_inode_operations
= {
2716 .readlink
= generic_readlink
,
2717 .get_link
= shmem_get_link
,
2718 #ifdef CONFIG_TMPFS_XATTR
2719 .setxattr
= generic_setxattr
,
2720 .getxattr
= generic_getxattr
,
2721 .listxattr
= shmem_listxattr
,
2722 .removexattr
= generic_removexattr
,
2726 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2728 return ERR_PTR(-ESTALE
);
2731 static int shmem_match(struct inode
*ino
, void *vfh
)
2735 inum
= (inum
<< 32) | fh
[1];
2736 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2739 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2740 struct fid
*fid
, int fh_len
, int fh_type
)
2742 struct inode
*inode
;
2743 struct dentry
*dentry
= NULL
;
2750 inum
= (inum
<< 32) | fid
->raw
[1];
2752 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2753 shmem_match
, fid
->raw
);
2755 dentry
= d_find_alias(inode
);
2762 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
2763 struct inode
*parent
)
2767 return FILEID_INVALID
;
2770 if (inode_unhashed(inode
)) {
2771 /* Unfortunately insert_inode_hash is not idempotent,
2772 * so as we hash inodes here rather than at creation
2773 * time, we need a lock to ensure we only try
2776 static DEFINE_SPINLOCK(lock
);
2778 if (inode_unhashed(inode
))
2779 __insert_inode_hash(inode
,
2780 inode
->i_ino
+ inode
->i_generation
);
2784 fh
[0] = inode
->i_generation
;
2785 fh
[1] = inode
->i_ino
;
2786 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2792 static const struct export_operations shmem_export_ops
= {
2793 .get_parent
= shmem_get_parent
,
2794 .encode_fh
= shmem_encode_fh
,
2795 .fh_to_dentry
= shmem_fh_to_dentry
,
2798 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2801 char *this_char
, *value
, *rest
;
2802 struct mempolicy
*mpol
= NULL
;
2806 while (options
!= NULL
) {
2807 this_char
= options
;
2810 * NUL-terminate this option: unfortunately,
2811 * mount options form a comma-separated list,
2812 * but mpol's nodelist may also contain commas.
2814 options
= strchr(options
, ',');
2815 if (options
== NULL
)
2818 if (!isdigit(*options
)) {
2825 if ((value
= strchr(this_char
,'=')) != NULL
) {
2829 "tmpfs: No value for mount option '%s'\n",
2834 if (!strcmp(this_char
,"size")) {
2835 unsigned long long size
;
2836 size
= memparse(value
,&rest
);
2838 size
<<= PAGE_SHIFT
;
2839 size
*= totalram_pages
;
2845 sbinfo
->max_blocks
=
2846 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2847 } else if (!strcmp(this_char
,"nr_blocks")) {
2848 sbinfo
->max_blocks
= memparse(value
, &rest
);
2851 } else if (!strcmp(this_char
,"nr_inodes")) {
2852 sbinfo
->max_inodes
= memparse(value
, &rest
);
2855 } else if (!strcmp(this_char
,"mode")) {
2858 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2861 } else if (!strcmp(this_char
,"uid")) {
2864 uid
= simple_strtoul(value
, &rest
, 0);
2867 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
2868 if (!uid_valid(sbinfo
->uid
))
2870 } else if (!strcmp(this_char
,"gid")) {
2873 gid
= simple_strtoul(value
, &rest
, 0);
2876 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
2877 if (!gid_valid(sbinfo
->gid
))
2879 } else if (!strcmp(this_char
,"mpol")) {
2882 if (mpol_parse_str(value
, &mpol
))
2885 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2890 sbinfo
->mpol
= mpol
;
2894 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2902 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2904 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2905 struct shmem_sb_info config
= *sbinfo
;
2906 unsigned long inodes
;
2907 int error
= -EINVAL
;
2910 if (shmem_parse_options(data
, &config
, true))
2913 spin_lock(&sbinfo
->stat_lock
);
2914 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2915 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2917 if (config
.max_inodes
< inodes
)
2920 * Those tests disallow limited->unlimited while any are in use;
2921 * but we must separately disallow unlimited->limited, because
2922 * in that case we have no record of how much is already in use.
2924 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2926 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2930 sbinfo
->max_blocks
= config
.max_blocks
;
2931 sbinfo
->max_inodes
= config
.max_inodes
;
2932 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2935 * Preserve previous mempolicy unless mpol remount option was specified.
2938 mpol_put(sbinfo
->mpol
);
2939 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2942 spin_unlock(&sbinfo
->stat_lock
);
2946 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2948 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2950 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2951 seq_printf(seq
, ",size=%luk",
2952 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2953 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2954 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2955 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2956 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2957 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
2958 seq_printf(seq
, ",uid=%u",
2959 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
2960 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
2961 seq_printf(seq
, ",gid=%u",
2962 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
2963 shmem_show_mpol(seq
, sbinfo
->mpol
);
2967 #define MFD_NAME_PREFIX "memfd:"
2968 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2969 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2971 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2973 SYSCALL_DEFINE2(memfd_create
,
2974 const char __user
*, uname
,
2975 unsigned int, flags
)
2977 struct shmem_inode_info
*info
;
2983 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
2986 /* length includes terminating zero */
2987 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
2990 if (len
> MFD_NAME_MAX_LEN
+ 1)
2993 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
2997 strcpy(name
, MFD_NAME_PREFIX
);
2998 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
3003 /* terminating-zero may have changed after strnlen_user() returned */
3004 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3009 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3015 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3017 error
= PTR_ERR(file
);
3020 info
= SHMEM_I(file_inode(file
));
3021 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3022 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3023 if (flags
& MFD_ALLOW_SEALING
)
3024 info
->seals
&= ~F_SEAL_SEAL
;
3026 fd_install(fd
, file
);
3037 #endif /* CONFIG_TMPFS */
3039 static void shmem_put_super(struct super_block
*sb
)
3041 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3043 percpu_counter_destroy(&sbinfo
->used_blocks
);
3044 mpol_put(sbinfo
->mpol
);
3046 sb
->s_fs_info
= NULL
;
3049 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3051 struct inode
*inode
;
3052 struct shmem_sb_info
*sbinfo
;
3055 /* Round up to L1_CACHE_BYTES to resist false sharing */
3056 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3057 L1_CACHE_BYTES
), GFP_KERNEL
);
3061 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3062 sbinfo
->uid
= current_fsuid();
3063 sbinfo
->gid
= current_fsgid();
3064 sb
->s_fs_info
= sbinfo
;
3068 * Per default we only allow half of the physical ram per
3069 * tmpfs instance, limiting inodes to one per page of lowmem;
3070 * but the internal instance is left unlimited.
3072 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3073 sbinfo
->max_blocks
= shmem_default_max_blocks();
3074 sbinfo
->max_inodes
= shmem_default_max_inodes();
3075 if (shmem_parse_options(data
, sbinfo
, false)) {
3080 sb
->s_flags
|= MS_NOUSER
;
3082 sb
->s_export_op
= &shmem_export_ops
;
3083 sb
->s_flags
|= MS_NOSEC
;
3085 sb
->s_flags
|= MS_NOUSER
;
3088 spin_lock_init(&sbinfo
->stat_lock
);
3089 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3091 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3093 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3094 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
3095 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
3096 sb
->s_magic
= TMPFS_MAGIC
;
3097 sb
->s_op
= &shmem_ops
;
3098 sb
->s_time_gran
= 1;
3099 #ifdef CONFIG_TMPFS_XATTR
3100 sb
->s_xattr
= shmem_xattr_handlers
;
3102 #ifdef CONFIG_TMPFS_POSIX_ACL
3103 sb
->s_flags
|= MS_POSIXACL
;
3106 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3109 inode
->i_uid
= sbinfo
->uid
;
3110 inode
->i_gid
= sbinfo
->gid
;
3111 sb
->s_root
= d_make_root(inode
);
3117 shmem_put_super(sb
);
3121 static struct kmem_cache
*shmem_inode_cachep
;
3123 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3125 struct shmem_inode_info
*info
;
3126 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3129 return &info
->vfs_inode
;
3132 static void shmem_destroy_callback(struct rcu_head
*head
)
3134 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3135 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3138 static void shmem_destroy_inode(struct inode
*inode
)
3140 if (S_ISREG(inode
->i_mode
))
3141 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3142 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3145 static void shmem_init_inode(void *foo
)
3147 struct shmem_inode_info
*info
= foo
;
3148 inode_init_once(&info
->vfs_inode
);
3151 static int shmem_init_inodecache(void)
3153 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3154 sizeof(struct shmem_inode_info
),
3155 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3159 static void shmem_destroy_inodecache(void)
3161 kmem_cache_destroy(shmem_inode_cachep
);
3164 static const struct address_space_operations shmem_aops
= {
3165 .writepage
= shmem_writepage
,
3166 .set_page_dirty
= __set_page_dirty_no_writeback
,
3168 .write_begin
= shmem_write_begin
,
3169 .write_end
= shmem_write_end
,
3171 #ifdef CONFIG_MIGRATION
3172 .migratepage
= migrate_page
,
3174 .error_remove_page
= generic_error_remove_page
,
3177 static const struct file_operations shmem_file_operations
= {
3180 .llseek
= shmem_file_llseek
,
3181 .read_iter
= shmem_file_read_iter
,
3182 .write_iter
= generic_file_write_iter
,
3183 .fsync
= noop_fsync
,
3184 .splice_read
= shmem_file_splice_read
,
3185 .splice_write
= iter_file_splice_write
,
3186 .fallocate
= shmem_fallocate
,
3190 static const struct inode_operations shmem_inode_operations
= {
3191 .getattr
= shmem_getattr
,
3192 .setattr
= shmem_setattr
,
3193 #ifdef CONFIG_TMPFS_XATTR
3194 .setxattr
= generic_setxattr
,
3195 .getxattr
= generic_getxattr
,
3196 .listxattr
= shmem_listxattr
,
3197 .removexattr
= generic_removexattr
,
3198 .set_acl
= simple_set_acl
,
3202 static const struct inode_operations shmem_dir_inode_operations
= {
3204 .create
= shmem_create
,
3205 .lookup
= simple_lookup
,
3207 .unlink
= shmem_unlink
,
3208 .symlink
= shmem_symlink
,
3209 .mkdir
= shmem_mkdir
,
3210 .rmdir
= shmem_rmdir
,
3211 .mknod
= shmem_mknod
,
3212 .rename2
= shmem_rename2
,
3213 .tmpfile
= shmem_tmpfile
,
3215 #ifdef CONFIG_TMPFS_XATTR
3216 .setxattr
= generic_setxattr
,
3217 .getxattr
= generic_getxattr
,
3218 .listxattr
= shmem_listxattr
,
3219 .removexattr
= generic_removexattr
,
3221 #ifdef CONFIG_TMPFS_POSIX_ACL
3222 .setattr
= shmem_setattr
,
3223 .set_acl
= simple_set_acl
,
3227 static const struct inode_operations shmem_special_inode_operations
= {
3228 #ifdef CONFIG_TMPFS_XATTR
3229 .setxattr
= generic_setxattr
,
3230 .getxattr
= generic_getxattr
,
3231 .listxattr
= shmem_listxattr
,
3232 .removexattr
= generic_removexattr
,
3234 #ifdef CONFIG_TMPFS_POSIX_ACL
3235 .setattr
= shmem_setattr
,
3236 .set_acl
= simple_set_acl
,
3240 static const struct super_operations shmem_ops
= {
3241 .alloc_inode
= shmem_alloc_inode
,
3242 .destroy_inode
= shmem_destroy_inode
,
3244 .statfs
= shmem_statfs
,
3245 .remount_fs
= shmem_remount_fs
,
3246 .show_options
= shmem_show_options
,
3248 .evict_inode
= shmem_evict_inode
,
3249 .drop_inode
= generic_delete_inode
,
3250 .put_super
= shmem_put_super
,
3253 static const struct vm_operations_struct shmem_vm_ops
= {
3254 .fault
= shmem_fault
,
3255 .map_pages
= filemap_map_pages
,
3257 .set_policy
= shmem_set_policy
,
3258 .get_policy
= shmem_get_policy
,
3262 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3263 int flags
, const char *dev_name
, void *data
)
3265 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3268 static struct file_system_type shmem_fs_type
= {
3269 .owner
= THIS_MODULE
,
3271 .mount
= shmem_mount
,
3272 .kill_sb
= kill_litter_super
,
3273 .fs_flags
= FS_USERNS_MOUNT
,
3276 int __init
shmem_init(void)
3280 /* If rootfs called this, don't re-init */
3281 if (shmem_inode_cachep
)
3284 error
= shmem_init_inodecache();
3288 error
= register_filesystem(&shmem_fs_type
);
3290 printk(KERN_ERR
"Could not register tmpfs\n");
3294 shm_mnt
= kern_mount(&shmem_fs_type
);
3295 if (IS_ERR(shm_mnt
)) {
3296 error
= PTR_ERR(shm_mnt
);
3297 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
3303 unregister_filesystem(&shmem_fs_type
);
3305 shmem_destroy_inodecache();
3307 shm_mnt
= ERR_PTR(error
);
3311 #else /* !CONFIG_SHMEM */
3314 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3316 * This is intended for small system where the benefits of the full
3317 * shmem code (swap-backed and resource-limited) are outweighed by
3318 * their complexity. On systems without swap this code should be
3319 * effectively equivalent, but much lighter weight.
3322 static struct file_system_type shmem_fs_type
= {
3324 .mount
= ramfs_mount
,
3325 .kill_sb
= kill_litter_super
,
3326 .fs_flags
= FS_USERNS_MOUNT
,
3329 int __init
shmem_init(void)
3331 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
3333 shm_mnt
= kern_mount(&shmem_fs_type
);
3334 BUG_ON(IS_ERR(shm_mnt
));
3339 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
3344 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
3349 void shmem_unlock_mapping(struct address_space
*mapping
)
3353 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
3355 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
3357 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
3359 #define shmem_vm_ops generic_file_vm_ops
3360 #define shmem_file_operations ramfs_file_operations
3361 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3362 #define shmem_acct_size(flags, size) 0
3363 #define shmem_unacct_size(flags, size) do {} while (0)
3365 #endif /* CONFIG_SHMEM */
3369 static struct dentry_operations anon_ops
= {
3370 .d_dname
= simple_dname
3373 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
3374 unsigned long flags
, unsigned int i_flags
)
3377 struct inode
*inode
;
3379 struct super_block
*sb
;
3382 if (IS_ERR(shm_mnt
))
3383 return ERR_CAST(shm_mnt
);
3385 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
3386 return ERR_PTR(-EINVAL
);
3388 if (shmem_acct_size(flags
, size
))
3389 return ERR_PTR(-ENOMEM
);
3391 res
= ERR_PTR(-ENOMEM
);
3393 this.len
= strlen(name
);
3394 this.hash
= 0; /* will go */
3395 sb
= shm_mnt
->mnt_sb
;
3396 path
.mnt
= mntget(shm_mnt
);
3397 path
.dentry
= d_alloc_pseudo(sb
, &this);
3400 d_set_d_op(path
.dentry
, &anon_ops
);
3402 res
= ERR_PTR(-ENOSPC
);
3403 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
3407 inode
->i_flags
|= i_flags
;
3408 d_instantiate(path
.dentry
, inode
);
3409 inode
->i_size
= size
;
3410 clear_nlink(inode
); /* It is unlinked */
3411 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
3415 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
3416 &shmem_file_operations
);
3423 shmem_unacct_size(flags
, size
);
3430 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3431 * kernel internal. There will be NO LSM permission checks against the
3432 * underlying inode. So users of this interface must do LSM checks at a
3433 * higher layer. The users are the big_key and shm implementations. LSM
3434 * checks are provided at the key or shm level rather than the inode.
3435 * @name: name for dentry (to be seen in /proc/<pid>/maps
3436 * @size: size to be set for the file
3437 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3439 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3441 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
3445 * shmem_file_setup - get an unlinked file living in tmpfs
3446 * @name: name for dentry (to be seen in /proc/<pid>/maps
3447 * @size: size to be set for the file
3448 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3450 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3452 return __shmem_file_setup(name
, size
, flags
, 0);
3454 EXPORT_SYMBOL_GPL(shmem_file_setup
);
3457 * shmem_zero_setup - setup a shared anonymous mapping
3458 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3460 int shmem_zero_setup(struct vm_area_struct
*vma
)
3463 loff_t size
= vma
->vm_end
- vma
->vm_start
;
3466 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3467 * between XFS directory reading and selinux: since this file is only
3468 * accessible to the user through its mapping, use S_PRIVATE flag to
3469 * bypass file security, in the same way as shmem_kernel_file_setup().
3471 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
3473 return PTR_ERR(file
);
3477 vma
->vm_file
= file
;
3478 vma
->vm_ops
= &shmem_vm_ops
;
3483 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3484 * @mapping: the page's address_space
3485 * @index: the page index
3486 * @gfp: the page allocator flags to use if allocating
3488 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3489 * with any new page allocations done using the specified allocation flags.
3490 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3491 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3492 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3494 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3495 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3497 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
3498 pgoff_t index
, gfp_t gfp
)
3501 struct inode
*inode
= mapping
->host
;
3505 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
3506 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
3508 page
= ERR_PTR(error
);
3514 * The tiny !SHMEM case uses ramfs without swap
3516 return read_cache_page_gfp(mapping
, index
, gfp
);
3519 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);