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();
916 * Add inode to shmem_unuse()'s list of swapped-out inodes,
917 * if it's not already there. Do it now before the page is
918 * moved to swap cache, when its pagelock no longer protects
919 * the inode from eviction. But don't unlock the mutex until
920 * we've incremented swapped, because shmem_unuse_inode() will
921 * prune a !swapped inode from the swaplist under this mutex.
923 mutex_lock(&shmem_swaplist_mutex
);
924 if (list_empty(&info
->swaplist
))
925 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
927 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
928 spin_lock(&info
->lock
);
929 shmem_recalc_inode(inode
);
931 spin_unlock(&info
->lock
);
933 swap_shmem_alloc(swap
);
934 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
936 mutex_unlock(&shmem_swaplist_mutex
);
937 BUG_ON(page_mapped(page
));
938 swap_writepage(page
, wbc
);
942 mutex_unlock(&shmem_swaplist_mutex
);
943 swapcache_free(swap
);
945 set_page_dirty(page
);
946 if (wbc
->for_reclaim
)
947 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
954 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
958 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
959 return; /* show nothing */
961 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
963 seq_printf(seq
, ",mpol=%s", buffer
);
966 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
968 struct mempolicy
*mpol
= NULL
;
970 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
973 spin_unlock(&sbinfo
->stat_lock
);
977 #endif /* CONFIG_TMPFS */
979 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
980 struct shmem_inode_info
*info
, pgoff_t index
)
982 struct vm_area_struct pvma
;
985 /* Create a pseudo vma that just contains the policy */
987 /* Bias interleave by inode number to distribute better across nodes */
988 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
990 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
992 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
994 /* Drop reference taken by mpol_shared_policy_lookup() */
995 mpol_cond_put(pvma
.vm_policy
);
1000 static struct page
*shmem_alloc_page(gfp_t gfp
,
1001 struct shmem_inode_info
*info
, pgoff_t index
)
1003 struct vm_area_struct pvma
;
1006 /* Create a pseudo vma that just contains the policy */
1008 /* Bias interleave by inode number to distribute better across nodes */
1009 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1011 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1013 page
= alloc_page_vma(gfp
, &pvma
, 0);
1015 /* Drop reference taken by mpol_shared_policy_lookup() */
1016 mpol_cond_put(pvma
.vm_policy
);
1020 #else /* !CONFIG_NUMA */
1022 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1025 #endif /* CONFIG_TMPFS */
1027 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1028 struct shmem_inode_info
*info
, pgoff_t index
)
1030 return swapin_readahead(swap
, gfp
, NULL
, 0);
1033 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
1034 struct shmem_inode_info
*info
, pgoff_t index
)
1036 return alloc_page(gfp
);
1038 #endif /* CONFIG_NUMA */
1040 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1041 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1048 * When a page is moved from swapcache to shmem filecache (either by the
1049 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1050 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1051 * ignorance of the mapping it belongs to. If that mapping has special
1052 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1053 * we may need to copy to a suitable page before moving to filecache.
1055 * In a future release, this may well be extended to respect cpuset and
1056 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1057 * but for now it is a simple matter of zone.
1059 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1061 return page_zonenum(page
) > gfp_zone(gfp
);
1064 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1065 struct shmem_inode_info
*info
, pgoff_t index
)
1067 struct page
*oldpage
, *newpage
;
1068 struct address_space
*swap_mapping
;
1073 swap_index
= page_private(oldpage
);
1074 swap_mapping
= page_mapping(oldpage
);
1077 * We have arrived here because our zones are constrained, so don't
1078 * limit chance of success by further cpuset and node constraints.
1080 gfp
&= ~GFP_CONSTRAINT_MASK
;
1081 newpage
= shmem_alloc_page(gfp
, info
, index
);
1085 page_cache_get(newpage
);
1086 copy_highpage(newpage
, oldpage
);
1087 flush_dcache_page(newpage
);
1089 __SetPageLocked(newpage
);
1090 SetPageUptodate(newpage
);
1091 SetPageSwapBacked(newpage
);
1092 set_page_private(newpage
, swap_index
);
1093 SetPageSwapCache(newpage
);
1096 * Our caller will very soon move newpage out of swapcache, but it's
1097 * a nice clean interface for us to replace oldpage by newpage there.
1099 spin_lock_irq(&swap_mapping
->tree_lock
);
1100 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1103 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
1104 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
1106 spin_unlock_irq(&swap_mapping
->tree_lock
);
1108 if (unlikely(error
)) {
1110 * Is this possible? I think not, now that our callers check
1111 * both PageSwapCache and page_private after getting page lock;
1112 * but be defensive. Reverse old to newpage for clear and free.
1116 mem_cgroup_replace_page(oldpage
, newpage
);
1117 lru_cache_add_anon(newpage
);
1121 ClearPageSwapCache(oldpage
);
1122 set_page_private(oldpage
, 0);
1124 unlock_page(oldpage
);
1125 page_cache_release(oldpage
);
1126 page_cache_release(oldpage
);
1131 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1133 * If we allocate a new one we do not mark it dirty. That's up to the
1134 * vm. If we swap it in we mark it dirty since we also free the swap
1135 * entry since a page cannot live in both the swap and page cache
1137 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1138 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
1140 struct address_space
*mapping
= inode
->i_mapping
;
1141 struct shmem_inode_info
*info
;
1142 struct shmem_sb_info
*sbinfo
;
1143 struct mem_cgroup
*memcg
;
1150 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
1154 page
= find_lock_entry(mapping
, index
);
1155 if (radix_tree_exceptional_entry(page
)) {
1156 swap
= radix_to_swp_entry(page
);
1160 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1161 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1166 if (page
&& sgp
== SGP_WRITE
)
1167 mark_page_accessed(page
);
1169 /* fallocated page? */
1170 if (page
&& !PageUptodate(page
)) {
1171 if (sgp
!= SGP_READ
)
1174 page_cache_release(page
);
1177 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1183 * Fast cache lookup did not find it:
1184 * bring it back from swap or allocate.
1186 info
= SHMEM_I(inode
);
1187 sbinfo
= SHMEM_SB(inode
->i_sb
);
1190 /* Look it up and read it in.. */
1191 page
= lookup_swap_cache(swap
);
1193 /* here we actually do the io */
1195 *fault_type
|= VM_FAULT_MAJOR
;
1196 page
= shmem_swapin(swap
, gfp
, info
, index
);
1203 /* We have to do this with page locked to prevent races */
1205 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1206 !shmem_confirm_swap(mapping
, index
, swap
)) {
1207 error
= -EEXIST
; /* try again */
1210 if (!PageUptodate(page
)) {
1214 wait_on_page_writeback(page
);
1216 if (shmem_should_replace_page(page
, gfp
)) {
1217 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1222 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
,
1225 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1226 swp_to_radix_entry(swap
));
1228 * We already confirmed swap under page lock, and make
1229 * no memory allocation here, so usually no possibility
1230 * of error; but free_swap_and_cache() only trylocks a
1231 * page, so it is just possible that the entry has been
1232 * truncated or holepunched since swap was confirmed.
1233 * shmem_undo_range() will have done some of the
1234 * unaccounting, now delete_from_swap_cache() will do
1236 * Reset swap.val? No, leave it so "failed" goes back to
1237 * "repeat": reading a hole and writing should succeed.
1240 mem_cgroup_cancel_charge(page
, memcg
, false);
1241 delete_from_swap_cache(page
);
1247 mem_cgroup_commit_charge(page
, memcg
, true, false);
1249 spin_lock(&info
->lock
);
1251 shmem_recalc_inode(inode
);
1252 spin_unlock(&info
->lock
);
1254 if (sgp
== SGP_WRITE
)
1255 mark_page_accessed(page
);
1257 delete_from_swap_cache(page
);
1258 set_page_dirty(page
);
1262 if (shmem_acct_block(info
->flags
)) {
1266 if (sbinfo
->max_blocks
) {
1267 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1268 sbinfo
->max_blocks
) >= 0) {
1272 percpu_counter_inc(&sbinfo
->used_blocks
);
1275 page
= shmem_alloc_page(gfp
, info
, index
);
1281 __SetPageSwapBacked(page
);
1282 __SetPageLocked(page
);
1283 if (sgp
== SGP_WRITE
)
1284 __SetPageReferenced(page
);
1286 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
,
1290 error
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
1292 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1294 radix_tree_preload_end();
1297 mem_cgroup_cancel_charge(page
, memcg
, false);
1300 mem_cgroup_commit_charge(page
, memcg
, false, false);
1301 lru_cache_add_anon(page
);
1303 spin_lock(&info
->lock
);
1305 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1306 shmem_recalc_inode(inode
);
1307 spin_unlock(&info
->lock
);
1311 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1313 if (sgp
== SGP_FALLOC
)
1317 * Let SGP_WRITE caller clear ends if write does not fill page;
1318 * but SGP_FALLOC on a page fallocated earlier must initialize
1319 * it now, lest undo on failure cancel our earlier guarantee.
1321 if (sgp
!= SGP_WRITE
) {
1322 clear_highpage(page
);
1323 flush_dcache_page(page
);
1324 SetPageUptodate(page
);
1326 if (sgp
== SGP_DIRTY
)
1327 set_page_dirty(page
);
1330 /* Perhaps the file has been truncated since we checked */
1331 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1332 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1334 ClearPageDirty(page
);
1335 delete_from_page_cache(page
);
1336 spin_lock(&info
->lock
);
1337 shmem_recalc_inode(inode
);
1338 spin_unlock(&info
->lock
);
1350 if (sbinfo
->max_blocks
)
1351 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1353 shmem_unacct_blocks(info
->flags
, 1);
1355 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1360 page_cache_release(page
);
1362 if (error
== -ENOSPC
&& !once
++) {
1363 info
= SHMEM_I(inode
);
1364 spin_lock(&info
->lock
);
1365 shmem_recalc_inode(inode
);
1366 spin_unlock(&info
->lock
);
1369 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1374 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1376 struct inode
*inode
= file_inode(vma
->vm_file
);
1378 int ret
= VM_FAULT_LOCKED
;
1381 * Trinity finds that probing a hole which tmpfs is punching can
1382 * prevent the hole-punch from ever completing: which in turn
1383 * locks writers out with its hold on i_mutex. So refrain from
1384 * faulting pages into the hole while it's being punched. Although
1385 * shmem_undo_range() does remove the additions, it may be unable to
1386 * keep up, as each new page needs its own unmap_mapping_range() call,
1387 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1389 * It does not matter if we sometimes reach this check just before the
1390 * hole-punch begins, so that one fault then races with the punch:
1391 * we just need to make racing faults a rare case.
1393 * The implementation below would be much simpler if we just used a
1394 * standard mutex or completion: but we cannot take i_mutex in fault,
1395 * and bloating every shmem inode for this unlikely case would be sad.
1397 if (unlikely(inode
->i_private
)) {
1398 struct shmem_falloc
*shmem_falloc
;
1400 spin_lock(&inode
->i_lock
);
1401 shmem_falloc
= inode
->i_private
;
1403 shmem_falloc
->waitq
&&
1404 vmf
->pgoff
>= shmem_falloc
->start
&&
1405 vmf
->pgoff
< shmem_falloc
->next
) {
1406 wait_queue_head_t
*shmem_falloc_waitq
;
1407 DEFINE_WAIT(shmem_fault_wait
);
1409 ret
= VM_FAULT_NOPAGE
;
1410 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1411 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1412 /* It's polite to up mmap_sem if we can */
1413 up_read(&vma
->vm_mm
->mmap_sem
);
1414 ret
= VM_FAULT_RETRY
;
1417 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1418 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1419 TASK_UNINTERRUPTIBLE
);
1420 spin_unlock(&inode
->i_lock
);
1424 * shmem_falloc_waitq points into the shmem_fallocate()
1425 * stack of the hole-punching task: shmem_falloc_waitq
1426 * is usually invalid by the time we reach here, but
1427 * finish_wait() does not dereference it in that case;
1428 * though i_lock needed lest racing with wake_up_all().
1430 spin_lock(&inode
->i_lock
);
1431 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1432 spin_unlock(&inode
->i_lock
);
1435 spin_unlock(&inode
->i_lock
);
1438 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1440 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1442 if (ret
& VM_FAULT_MAJOR
) {
1443 count_vm_event(PGMAJFAULT
);
1444 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1450 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1452 struct inode
*inode
= file_inode(vma
->vm_file
);
1453 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1456 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1459 struct inode
*inode
= file_inode(vma
->vm_file
);
1462 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1463 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1467 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1469 struct inode
*inode
= file_inode(file
);
1470 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1471 int retval
= -ENOMEM
;
1473 spin_lock(&info
->lock
);
1474 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1475 if (!user_shm_lock(inode
->i_size
, user
))
1477 info
->flags
|= VM_LOCKED
;
1478 mapping_set_unevictable(file
->f_mapping
);
1480 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1481 user_shm_unlock(inode
->i_size
, user
);
1482 info
->flags
&= ~VM_LOCKED
;
1483 mapping_clear_unevictable(file
->f_mapping
);
1488 spin_unlock(&info
->lock
);
1492 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1494 file_accessed(file
);
1495 vma
->vm_ops
= &shmem_vm_ops
;
1499 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1500 umode_t mode
, dev_t dev
, unsigned long flags
)
1502 struct inode
*inode
;
1503 struct shmem_inode_info
*info
;
1504 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1506 if (shmem_reserve_inode(sb
))
1509 inode
= new_inode(sb
);
1511 inode
->i_ino
= get_next_ino();
1512 inode_init_owner(inode
, dir
, mode
);
1513 inode
->i_blocks
= 0;
1514 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1515 inode
->i_generation
= get_seconds();
1516 info
= SHMEM_I(inode
);
1517 memset(info
, 0, (char *)inode
- (char *)info
);
1518 spin_lock_init(&info
->lock
);
1519 info
->seals
= F_SEAL_SEAL
;
1520 info
->flags
= flags
& VM_NORESERVE
;
1521 INIT_LIST_HEAD(&info
->swaplist
);
1522 simple_xattrs_init(&info
->xattrs
);
1523 cache_no_acl(inode
);
1525 switch (mode
& S_IFMT
) {
1527 inode
->i_op
= &shmem_special_inode_operations
;
1528 init_special_inode(inode
, mode
, dev
);
1531 inode
->i_mapping
->a_ops
= &shmem_aops
;
1532 inode
->i_op
= &shmem_inode_operations
;
1533 inode
->i_fop
= &shmem_file_operations
;
1534 mpol_shared_policy_init(&info
->policy
,
1535 shmem_get_sbmpol(sbinfo
));
1539 /* Some things misbehave if size == 0 on a directory */
1540 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1541 inode
->i_op
= &shmem_dir_inode_operations
;
1542 inode
->i_fop
= &simple_dir_operations
;
1546 * Must not load anything in the rbtree,
1547 * mpol_free_shared_policy will not be called.
1549 mpol_shared_policy_init(&info
->policy
, NULL
);
1553 shmem_free_inode(sb
);
1557 bool shmem_mapping(struct address_space
*mapping
)
1562 return mapping
->host
->i_sb
->s_op
== &shmem_ops
;
1566 static const struct inode_operations shmem_symlink_inode_operations
;
1567 static const struct inode_operations shmem_short_symlink_operations
;
1569 #ifdef CONFIG_TMPFS_XATTR
1570 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1572 #define shmem_initxattrs NULL
1576 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1577 loff_t pos
, unsigned len
, unsigned flags
,
1578 struct page
**pagep
, void **fsdata
)
1580 struct inode
*inode
= mapping
->host
;
1581 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1582 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1584 /* i_mutex is held by caller */
1585 if (unlikely(info
->seals
)) {
1586 if (info
->seals
& F_SEAL_WRITE
)
1588 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
1592 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1596 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1597 loff_t pos
, unsigned len
, unsigned copied
,
1598 struct page
*page
, void *fsdata
)
1600 struct inode
*inode
= mapping
->host
;
1602 if (pos
+ copied
> inode
->i_size
)
1603 i_size_write(inode
, pos
+ copied
);
1605 if (!PageUptodate(page
)) {
1606 if (copied
< PAGE_CACHE_SIZE
) {
1607 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1608 zero_user_segments(page
, 0, from
,
1609 from
+ copied
, PAGE_CACHE_SIZE
);
1611 SetPageUptodate(page
);
1613 set_page_dirty(page
);
1615 page_cache_release(page
);
1620 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1622 struct file
*file
= iocb
->ki_filp
;
1623 struct inode
*inode
= file_inode(file
);
1624 struct address_space
*mapping
= inode
->i_mapping
;
1626 unsigned long offset
;
1627 enum sgp_type sgp
= SGP_READ
;
1630 loff_t
*ppos
= &iocb
->ki_pos
;
1633 * Might this read be for a stacking filesystem? Then when reading
1634 * holes of a sparse file, we actually need to allocate those pages,
1635 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1637 if (!iter_is_iovec(to
))
1640 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1641 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1644 struct page
*page
= NULL
;
1646 unsigned long nr
, ret
;
1647 loff_t i_size
= i_size_read(inode
);
1649 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1650 if (index
> end_index
)
1652 if (index
== end_index
) {
1653 nr
= i_size
& ~PAGE_CACHE_MASK
;
1658 error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1660 if (error
== -EINVAL
)
1668 * We must evaluate after, since reads (unlike writes)
1669 * are called without i_mutex protection against truncate
1671 nr
= PAGE_CACHE_SIZE
;
1672 i_size
= i_size_read(inode
);
1673 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1674 if (index
== end_index
) {
1675 nr
= i_size
& ~PAGE_CACHE_MASK
;
1678 page_cache_release(page
);
1686 * If users can be writing to this page using arbitrary
1687 * virtual addresses, take care about potential aliasing
1688 * before reading the page on the kernel side.
1690 if (mapping_writably_mapped(mapping
))
1691 flush_dcache_page(page
);
1693 * Mark the page accessed if we read the beginning.
1696 mark_page_accessed(page
);
1698 page
= ZERO_PAGE(0);
1699 page_cache_get(page
);
1703 * Ok, we have the page, and it's up-to-date, so
1704 * now we can copy it to user space...
1706 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
1709 index
+= offset
>> PAGE_CACHE_SHIFT
;
1710 offset
&= ~PAGE_CACHE_MASK
;
1712 page_cache_release(page
);
1713 if (!iov_iter_count(to
))
1722 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1723 file_accessed(file
);
1724 return retval
? retval
: error
;
1727 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1728 struct pipe_inode_info
*pipe
, size_t len
,
1731 struct address_space
*mapping
= in
->f_mapping
;
1732 struct inode
*inode
= mapping
->host
;
1733 unsigned int loff
, nr_pages
, req_pages
;
1734 struct page
*pages
[PIPE_DEF_BUFFERS
];
1735 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1737 pgoff_t index
, end_index
;
1740 struct splice_pipe_desc spd
= {
1743 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1745 .ops
= &page_cache_pipe_buf_ops
,
1746 .spd_release
= spd_release_page
,
1749 isize
= i_size_read(inode
);
1750 if (unlikely(*ppos
>= isize
))
1753 left
= isize
- *ppos
;
1754 if (unlikely(left
< len
))
1757 if (splice_grow_spd(pipe
, &spd
))
1760 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1761 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1762 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1763 nr_pages
= min(req_pages
, spd
.nr_pages_max
);
1765 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1766 nr_pages
, spd
.pages
);
1767 index
+= spd
.nr_pages
;
1770 while (spd
.nr_pages
< nr_pages
) {
1771 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1775 spd
.pages
[spd
.nr_pages
++] = page
;
1779 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1780 nr_pages
= spd
.nr_pages
;
1783 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1784 unsigned int this_len
;
1789 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1790 page
= spd
.pages
[page_nr
];
1792 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1793 error
= shmem_getpage(inode
, index
, &page
,
1798 page_cache_release(spd
.pages
[page_nr
]);
1799 spd
.pages
[page_nr
] = page
;
1802 isize
= i_size_read(inode
);
1803 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1804 if (unlikely(!isize
|| index
> end_index
))
1807 if (end_index
== index
) {
1810 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1814 this_len
= min(this_len
, plen
- loff
);
1818 spd
.partial
[page_nr
].offset
= loff
;
1819 spd
.partial
[page_nr
].len
= this_len
;
1826 while (page_nr
< nr_pages
)
1827 page_cache_release(spd
.pages
[page_nr
++]);
1830 error
= splice_to_pipe(pipe
, &spd
);
1832 splice_shrink_spd(&spd
);
1842 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1844 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
1845 pgoff_t index
, pgoff_t end
, int whence
)
1848 struct pagevec pvec
;
1849 pgoff_t indices
[PAGEVEC_SIZE
];
1853 pagevec_init(&pvec
, 0);
1854 pvec
.nr
= 1; /* start small: we may be there already */
1856 pvec
.nr
= find_get_entries(mapping
, index
,
1857 pvec
.nr
, pvec
.pages
, indices
);
1859 if (whence
== SEEK_DATA
)
1863 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
1864 if (index
< indices
[i
]) {
1865 if (whence
== SEEK_HOLE
) {
1871 page
= pvec
.pages
[i
];
1872 if (page
&& !radix_tree_exceptional_entry(page
)) {
1873 if (!PageUptodate(page
))
1877 (page
&& whence
== SEEK_DATA
) ||
1878 (!page
&& whence
== SEEK_HOLE
)) {
1883 pagevec_remove_exceptionals(&pvec
);
1884 pagevec_release(&pvec
);
1885 pvec
.nr
= PAGEVEC_SIZE
;
1891 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
1893 struct address_space
*mapping
= file
->f_mapping
;
1894 struct inode
*inode
= mapping
->host
;
1898 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
1899 return generic_file_llseek_size(file
, offset
, whence
,
1900 MAX_LFS_FILESIZE
, i_size_read(inode
));
1901 mutex_lock(&inode
->i_mutex
);
1902 /* We're holding i_mutex so we can access i_size directly */
1906 else if (offset
>= inode
->i_size
)
1909 start
= offset
>> PAGE_CACHE_SHIFT
;
1910 end
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1911 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
1912 new_offset
<<= PAGE_CACHE_SHIFT
;
1913 if (new_offset
> offset
) {
1914 if (new_offset
< inode
->i_size
)
1915 offset
= new_offset
;
1916 else if (whence
== SEEK_DATA
)
1919 offset
= inode
->i_size
;
1924 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
1925 mutex_unlock(&inode
->i_mutex
);
1930 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1931 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1933 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1934 #define LAST_SCAN 4 /* about 150ms max */
1936 static void shmem_tag_pins(struct address_space
*mapping
)
1938 struct radix_tree_iter iter
;
1948 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1949 page
= radix_tree_deref_slot(slot
);
1950 if (!page
|| radix_tree_exception(page
)) {
1951 if (radix_tree_deref_retry(page
))
1953 } else if (page_count(page
) - page_mapcount(page
) > 1) {
1954 spin_lock_irq(&mapping
->tree_lock
);
1955 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
1957 spin_unlock_irq(&mapping
->tree_lock
);
1960 if (need_resched()) {
1962 start
= iter
.index
+ 1;
1970 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1971 * via get_user_pages(), drivers might have some pending I/O without any active
1972 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1973 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1974 * them to be dropped.
1975 * The caller must guarantee that no new user will acquire writable references
1976 * to those pages to avoid races.
1978 static int shmem_wait_for_pins(struct address_space
*mapping
)
1980 struct radix_tree_iter iter
;
1986 shmem_tag_pins(mapping
);
1989 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
1990 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
1994 lru_add_drain_all();
1995 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2001 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2002 start
, SHMEM_TAG_PINNED
) {
2004 page
= radix_tree_deref_slot(slot
);
2005 if (radix_tree_exception(page
)) {
2006 if (radix_tree_deref_retry(page
))
2013 page_count(page
) - page_mapcount(page
) != 1) {
2014 if (scan
< LAST_SCAN
)
2015 goto continue_resched
;
2018 * On the last scan, we clean up all those tags
2019 * we inserted; but make a note that we still
2020 * found pages pinned.
2025 spin_lock_irq(&mapping
->tree_lock
);
2026 radix_tree_tag_clear(&mapping
->page_tree
,
2027 iter
.index
, SHMEM_TAG_PINNED
);
2028 spin_unlock_irq(&mapping
->tree_lock
);
2030 if (need_resched()) {
2032 start
= iter
.index
+ 1;
2042 #define F_ALL_SEALS (F_SEAL_SEAL | \
2047 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2049 struct inode
*inode
= file_inode(file
);
2050 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2055 * Sealing allows multiple parties to share a shmem-file but restrict
2056 * access to a specific subset of file operations. Seals can only be
2057 * added, but never removed. This way, mutually untrusted parties can
2058 * share common memory regions with a well-defined policy. A malicious
2059 * peer can thus never perform unwanted operations on a shared object.
2061 * Seals are only supported on special shmem-files and always affect
2062 * the whole underlying inode. Once a seal is set, it may prevent some
2063 * kinds of access to the file. Currently, the following seals are
2065 * SEAL_SEAL: Prevent further seals from being set on this file
2066 * SEAL_SHRINK: Prevent the file from shrinking
2067 * SEAL_GROW: Prevent the file from growing
2068 * SEAL_WRITE: Prevent write access to the file
2070 * As we don't require any trust relationship between two parties, we
2071 * must prevent seals from being removed. Therefore, sealing a file
2072 * only adds a given set of seals to the file, it never touches
2073 * existing seals. Furthermore, the "setting seals"-operation can be
2074 * sealed itself, which basically prevents any further seal from being
2077 * Semantics of sealing are only defined on volatile files. Only
2078 * anonymous shmem files support sealing. More importantly, seals are
2079 * never written to disk. Therefore, there's no plan to support it on
2083 if (file
->f_op
!= &shmem_file_operations
)
2085 if (!(file
->f_mode
& FMODE_WRITE
))
2087 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2090 mutex_lock(&inode
->i_mutex
);
2092 if (info
->seals
& F_SEAL_SEAL
) {
2097 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2098 error
= mapping_deny_writable(file
->f_mapping
);
2102 error
= shmem_wait_for_pins(file
->f_mapping
);
2104 mapping_allow_writable(file
->f_mapping
);
2109 info
->seals
|= seals
;
2113 mutex_unlock(&inode
->i_mutex
);
2116 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2118 int shmem_get_seals(struct file
*file
)
2120 if (file
->f_op
!= &shmem_file_operations
)
2123 return SHMEM_I(file_inode(file
))->seals
;
2125 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2127 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2133 /* disallow upper 32bit */
2137 error
= shmem_add_seals(file
, arg
);
2140 error
= shmem_get_seals(file
);
2150 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2153 struct inode
*inode
= file_inode(file
);
2154 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2155 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2156 struct shmem_falloc shmem_falloc
;
2157 pgoff_t start
, index
, end
;
2160 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2163 mutex_lock(&inode
->i_mutex
);
2165 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2166 struct address_space
*mapping
= file
->f_mapping
;
2167 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2168 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2169 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2171 /* protected by i_mutex */
2172 if (info
->seals
& F_SEAL_WRITE
) {
2177 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2178 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2179 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2180 spin_lock(&inode
->i_lock
);
2181 inode
->i_private
= &shmem_falloc
;
2182 spin_unlock(&inode
->i_lock
);
2184 if ((u64
)unmap_end
> (u64
)unmap_start
)
2185 unmap_mapping_range(mapping
, unmap_start
,
2186 1 + unmap_end
- unmap_start
, 0);
2187 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2188 /* No need to unmap again: hole-punching leaves COWed pages */
2190 spin_lock(&inode
->i_lock
);
2191 inode
->i_private
= NULL
;
2192 wake_up_all(&shmem_falloc_waitq
);
2193 spin_unlock(&inode
->i_lock
);
2198 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2199 error
= inode_newsize_ok(inode
, offset
+ len
);
2203 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2208 start
= offset
>> PAGE_CACHE_SHIFT
;
2209 end
= (offset
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
2210 /* Try to avoid a swapstorm if len is impossible to satisfy */
2211 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2216 shmem_falloc
.waitq
= NULL
;
2217 shmem_falloc
.start
= start
;
2218 shmem_falloc
.next
= start
;
2219 shmem_falloc
.nr_falloced
= 0;
2220 shmem_falloc
.nr_unswapped
= 0;
2221 spin_lock(&inode
->i_lock
);
2222 inode
->i_private
= &shmem_falloc
;
2223 spin_unlock(&inode
->i_lock
);
2225 for (index
= start
; index
< end
; index
++) {
2229 * Good, the fallocate(2) manpage permits EINTR: we may have
2230 * been interrupted because we are using up too much memory.
2232 if (signal_pending(current
))
2234 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2237 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
,
2240 /* Remove the !PageUptodate pages we added */
2241 shmem_undo_range(inode
,
2242 (loff_t
)start
<< PAGE_CACHE_SHIFT
,
2243 (loff_t
)index
<< PAGE_CACHE_SHIFT
, true);
2248 * Inform shmem_writepage() how far we have reached.
2249 * No need for lock or barrier: we have the page lock.
2251 shmem_falloc
.next
++;
2252 if (!PageUptodate(page
))
2253 shmem_falloc
.nr_falloced
++;
2256 * If !PageUptodate, leave it that way so that freeable pages
2257 * can be recognized if we need to rollback on error later.
2258 * But set_page_dirty so that memory pressure will swap rather
2259 * than free the pages we are allocating (and SGP_CACHE pages
2260 * might still be clean: we now need to mark those dirty too).
2262 set_page_dirty(page
);
2264 page_cache_release(page
);
2268 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2269 i_size_write(inode
, offset
+ len
);
2270 inode
->i_ctime
= CURRENT_TIME
;
2272 spin_lock(&inode
->i_lock
);
2273 inode
->i_private
= NULL
;
2274 spin_unlock(&inode
->i_lock
);
2276 mutex_unlock(&inode
->i_mutex
);
2280 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2282 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2284 buf
->f_type
= TMPFS_MAGIC
;
2285 buf
->f_bsize
= PAGE_CACHE_SIZE
;
2286 buf
->f_namelen
= NAME_MAX
;
2287 if (sbinfo
->max_blocks
) {
2288 buf
->f_blocks
= sbinfo
->max_blocks
;
2290 buf
->f_bfree
= sbinfo
->max_blocks
-
2291 percpu_counter_sum(&sbinfo
->used_blocks
);
2293 if (sbinfo
->max_inodes
) {
2294 buf
->f_files
= sbinfo
->max_inodes
;
2295 buf
->f_ffree
= sbinfo
->free_inodes
;
2297 /* else leave those fields 0 like simple_statfs */
2302 * File creation. Allocate an inode, and we're done..
2305 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2307 struct inode
*inode
;
2308 int error
= -ENOSPC
;
2310 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2312 error
= simple_acl_create(dir
, inode
);
2315 error
= security_inode_init_security(inode
, dir
,
2317 shmem_initxattrs
, NULL
);
2318 if (error
&& error
!= -EOPNOTSUPP
)
2322 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2323 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2324 d_instantiate(dentry
, inode
);
2325 dget(dentry
); /* Extra count - pin the dentry in core */
2334 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2336 struct inode
*inode
;
2337 int error
= -ENOSPC
;
2339 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2341 error
= security_inode_init_security(inode
, dir
,
2343 shmem_initxattrs
, NULL
);
2344 if (error
&& error
!= -EOPNOTSUPP
)
2346 error
= simple_acl_create(dir
, inode
);
2349 d_tmpfile(dentry
, inode
);
2357 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2361 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2367 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2370 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2376 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2378 struct inode
*inode
= d_inode(old_dentry
);
2382 * No ordinary (disk based) filesystem counts links as inodes;
2383 * but each new link needs a new dentry, pinning lowmem, and
2384 * tmpfs dentries cannot be pruned until they are unlinked.
2386 ret
= shmem_reserve_inode(inode
->i_sb
);
2390 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2391 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2393 ihold(inode
); /* New dentry reference */
2394 dget(dentry
); /* Extra pinning count for the created dentry */
2395 d_instantiate(dentry
, inode
);
2400 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2402 struct inode
*inode
= d_inode(dentry
);
2404 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2405 shmem_free_inode(inode
->i_sb
);
2407 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2408 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2410 dput(dentry
); /* Undo the count from "create" - this does all the work */
2414 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2416 if (!simple_empty(dentry
))
2419 drop_nlink(d_inode(dentry
));
2421 return shmem_unlink(dir
, dentry
);
2424 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2426 bool old_is_dir
= d_is_dir(old_dentry
);
2427 bool new_is_dir
= d_is_dir(new_dentry
);
2429 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2431 drop_nlink(old_dir
);
2434 drop_nlink(new_dir
);
2438 old_dir
->i_ctime
= old_dir
->i_mtime
=
2439 new_dir
->i_ctime
= new_dir
->i_mtime
=
2440 d_inode(old_dentry
)->i_ctime
=
2441 d_inode(new_dentry
)->i_ctime
= CURRENT_TIME
;
2446 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
2448 struct dentry
*whiteout
;
2451 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2455 error
= shmem_mknod(old_dir
, whiteout
,
2456 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2462 * Cheat and hash the whiteout while the old dentry is still in
2463 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2465 * d_lookup() will consistently find one of them at this point,
2466 * not sure which one, but that isn't even important.
2473 * The VFS layer already does all the dentry stuff for rename,
2474 * we just have to decrement the usage count for the target if
2475 * it exists so that the VFS layer correctly free's it when it
2478 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
2480 struct inode
*inode
= d_inode(old_dentry
);
2481 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2483 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
2486 if (flags
& RENAME_EXCHANGE
)
2487 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
2489 if (!simple_empty(new_dentry
))
2492 if (flags
& RENAME_WHITEOUT
) {
2495 error
= shmem_whiteout(old_dir
, old_dentry
);
2500 if (d_really_is_positive(new_dentry
)) {
2501 (void) shmem_unlink(new_dir
, new_dentry
);
2502 if (they_are_dirs
) {
2503 drop_nlink(d_inode(new_dentry
));
2504 drop_nlink(old_dir
);
2506 } else if (they_are_dirs
) {
2507 drop_nlink(old_dir
);
2511 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2512 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2513 old_dir
->i_ctime
= old_dir
->i_mtime
=
2514 new_dir
->i_ctime
= new_dir
->i_mtime
=
2515 inode
->i_ctime
= CURRENT_TIME
;
2519 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2523 struct inode
*inode
;
2525 struct shmem_inode_info
*info
;
2527 len
= strlen(symname
) + 1;
2528 if (len
> PAGE_CACHE_SIZE
)
2529 return -ENAMETOOLONG
;
2531 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2535 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2536 shmem_initxattrs
, NULL
);
2538 if (error
!= -EOPNOTSUPP
) {
2545 info
= SHMEM_I(inode
);
2546 inode
->i_size
= len
-1;
2547 if (len
<= SHORT_SYMLINK_LEN
) {
2548 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
2549 if (!info
->symlink
) {
2553 inode
->i_op
= &shmem_short_symlink_operations
;
2554 inode
->i_link
= info
->symlink
;
2556 inode_nohighmem(inode
);
2557 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
2562 inode
->i_mapping
->a_ops
= &shmem_aops
;
2563 inode
->i_op
= &shmem_symlink_inode_operations
;
2564 memcpy(page_address(page
), symname
, len
);
2565 SetPageUptodate(page
);
2566 set_page_dirty(page
);
2568 page_cache_release(page
);
2570 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2571 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2572 d_instantiate(dentry
, inode
);
2577 static void shmem_put_link(void *arg
)
2579 mark_page_accessed(arg
);
2583 static const char *shmem_get_link(struct dentry
*dentry
,
2584 struct inode
*inode
,
2585 struct delayed_call
*done
)
2587 struct page
*page
= NULL
;
2590 page
= find_get_page(inode
->i_mapping
, 0);
2592 return ERR_PTR(-ECHILD
);
2593 if (!PageUptodate(page
)) {
2595 return ERR_PTR(-ECHILD
);
2598 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
, NULL
);
2600 return ERR_PTR(error
);
2603 set_delayed_call(done
, shmem_put_link
, page
);
2604 return page_address(page
);
2607 #ifdef CONFIG_TMPFS_XATTR
2609 * Superblocks without xattr inode operations may get some security.* xattr
2610 * support from the LSM "for free". As soon as we have any other xattrs
2611 * like ACLs, we also need to implement the security.* handlers at
2612 * filesystem level, though.
2616 * Callback for security_inode_init_security() for acquiring xattrs.
2618 static int shmem_initxattrs(struct inode
*inode
,
2619 const struct xattr
*xattr_array
,
2622 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2623 const struct xattr
*xattr
;
2624 struct simple_xattr
*new_xattr
;
2627 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2628 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
2632 len
= strlen(xattr
->name
) + 1;
2633 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2635 if (!new_xattr
->name
) {
2640 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2641 XATTR_SECURITY_PREFIX_LEN
);
2642 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2645 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
2651 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
2652 struct dentry
*dentry
, const char *name
,
2653 void *buffer
, size_t size
)
2655 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2657 name
= xattr_full_name(handler
, name
);
2658 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
2661 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
2662 struct dentry
*dentry
, const char *name
,
2663 const void *value
, size_t size
, int flags
)
2665 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2667 name
= xattr_full_name(handler
, name
);
2668 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
2671 static const struct xattr_handler shmem_security_xattr_handler
= {
2672 .prefix
= XATTR_SECURITY_PREFIX
,
2673 .get
= shmem_xattr_handler_get
,
2674 .set
= shmem_xattr_handler_set
,
2677 static const struct xattr_handler shmem_trusted_xattr_handler
= {
2678 .prefix
= XATTR_TRUSTED_PREFIX
,
2679 .get
= shmem_xattr_handler_get
,
2680 .set
= shmem_xattr_handler_set
,
2683 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2684 #ifdef CONFIG_TMPFS_POSIX_ACL
2685 &posix_acl_access_xattr_handler
,
2686 &posix_acl_default_xattr_handler
,
2688 &shmem_security_xattr_handler
,
2689 &shmem_trusted_xattr_handler
,
2693 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2695 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2696 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
2698 #endif /* CONFIG_TMPFS_XATTR */
2700 static const struct inode_operations shmem_short_symlink_operations
= {
2701 .readlink
= generic_readlink
,
2702 .get_link
= simple_get_link
,
2703 #ifdef CONFIG_TMPFS_XATTR
2704 .setxattr
= generic_setxattr
,
2705 .getxattr
= generic_getxattr
,
2706 .listxattr
= shmem_listxattr
,
2707 .removexattr
= generic_removexattr
,
2711 static const struct inode_operations shmem_symlink_inode_operations
= {
2712 .readlink
= generic_readlink
,
2713 .get_link
= shmem_get_link
,
2714 #ifdef CONFIG_TMPFS_XATTR
2715 .setxattr
= generic_setxattr
,
2716 .getxattr
= generic_getxattr
,
2717 .listxattr
= shmem_listxattr
,
2718 .removexattr
= generic_removexattr
,
2722 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2724 return ERR_PTR(-ESTALE
);
2727 static int shmem_match(struct inode
*ino
, void *vfh
)
2731 inum
= (inum
<< 32) | fh
[1];
2732 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2735 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2736 struct fid
*fid
, int fh_len
, int fh_type
)
2738 struct inode
*inode
;
2739 struct dentry
*dentry
= NULL
;
2746 inum
= (inum
<< 32) | fid
->raw
[1];
2748 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2749 shmem_match
, fid
->raw
);
2751 dentry
= d_find_alias(inode
);
2758 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
2759 struct inode
*parent
)
2763 return FILEID_INVALID
;
2766 if (inode_unhashed(inode
)) {
2767 /* Unfortunately insert_inode_hash is not idempotent,
2768 * so as we hash inodes here rather than at creation
2769 * time, we need a lock to ensure we only try
2772 static DEFINE_SPINLOCK(lock
);
2774 if (inode_unhashed(inode
))
2775 __insert_inode_hash(inode
,
2776 inode
->i_ino
+ inode
->i_generation
);
2780 fh
[0] = inode
->i_generation
;
2781 fh
[1] = inode
->i_ino
;
2782 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2788 static const struct export_operations shmem_export_ops
= {
2789 .get_parent
= shmem_get_parent
,
2790 .encode_fh
= shmem_encode_fh
,
2791 .fh_to_dentry
= shmem_fh_to_dentry
,
2794 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2797 char *this_char
, *value
, *rest
;
2798 struct mempolicy
*mpol
= NULL
;
2802 while (options
!= NULL
) {
2803 this_char
= options
;
2806 * NUL-terminate this option: unfortunately,
2807 * mount options form a comma-separated list,
2808 * but mpol's nodelist may also contain commas.
2810 options
= strchr(options
, ',');
2811 if (options
== NULL
)
2814 if (!isdigit(*options
)) {
2821 if ((value
= strchr(this_char
,'=')) != NULL
) {
2825 "tmpfs: No value for mount option '%s'\n",
2830 if (!strcmp(this_char
,"size")) {
2831 unsigned long long size
;
2832 size
= memparse(value
,&rest
);
2834 size
<<= PAGE_SHIFT
;
2835 size
*= totalram_pages
;
2841 sbinfo
->max_blocks
=
2842 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2843 } else if (!strcmp(this_char
,"nr_blocks")) {
2844 sbinfo
->max_blocks
= memparse(value
, &rest
);
2847 } else if (!strcmp(this_char
,"nr_inodes")) {
2848 sbinfo
->max_inodes
= memparse(value
, &rest
);
2851 } else if (!strcmp(this_char
,"mode")) {
2854 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2857 } else if (!strcmp(this_char
,"uid")) {
2860 uid
= simple_strtoul(value
, &rest
, 0);
2863 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
2864 if (!uid_valid(sbinfo
->uid
))
2866 } else if (!strcmp(this_char
,"gid")) {
2869 gid
= simple_strtoul(value
, &rest
, 0);
2872 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
2873 if (!gid_valid(sbinfo
->gid
))
2875 } else if (!strcmp(this_char
,"mpol")) {
2878 if (mpol_parse_str(value
, &mpol
))
2881 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2886 sbinfo
->mpol
= mpol
;
2890 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2898 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2900 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2901 struct shmem_sb_info config
= *sbinfo
;
2902 unsigned long inodes
;
2903 int error
= -EINVAL
;
2906 if (shmem_parse_options(data
, &config
, true))
2909 spin_lock(&sbinfo
->stat_lock
);
2910 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2911 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2913 if (config
.max_inodes
< inodes
)
2916 * Those tests disallow limited->unlimited while any are in use;
2917 * but we must separately disallow unlimited->limited, because
2918 * in that case we have no record of how much is already in use.
2920 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2922 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2926 sbinfo
->max_blocks
= config
.max_blocks
;
2927 sbinfo
->max_inodes
= config
.max_inodes
;
2928 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2931 * Preserve previous mempolicy unless mpol remount option was specified.
2934 mpol_put(sbinfo
->mpol
);
2935 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2938 spin_unlock(&sbinfo
->stat_lock
);
2942 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2944 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2946 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2947 seq_printf(seq
, ",size=%luk",
2948 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2949 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2950 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2951 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2952 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2953 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
2954 seq_printf(seq
, ",uid=%u",
2955 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
2956 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
2957 seq_printf(seq
, ",gid=%u",
2958 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
2959 shmem_show_mpol(seq
, sbinfo
->mpol
);
2963 #define MFD_NAME_PREFIX "memfd:"
2964 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2965 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2967 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2969 SYSCALL_DEFINE2(memfd_create
,
2970 const char __user
*, uname
,
2971 unsigned int, flags
)
2973 struct shmem_inode_info
*info
;
2979 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
2982 /* length includes terminating zero */
2983 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
2986 if (len
> MFD_NAME_MAX_LEN
+ 1)
2989 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
2993 strcpy(name
, MFD_NAME_PREFIX
);
2994 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
2999 /* terminating-zero may have changed after strnlen_user() returned */
3000 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3005 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3011 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3013 error
= PTR_ERR(file
);
3016 info
= SHMEM_I(file_inode(file
));
3017 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3018 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3019 if (flags
& MFD_ALLOW_SEALING
)
3020 info
->seals
&= ~F_SEAL_SEAL
;
3022 fd_install(fd
, file
);
3033 #endif /* CONFIG_TMPFS */
3035 static void shmem_put_super(struct super_block
*sb
)
3037 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3039 percpu_counter_destroy(&sbinfo
->used_blocks
);
3040 mpol_put(sbinfo
->mpol
);
3042 sb
->s_fs_info
= NULL
;
3045 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3047 struct inode
*inode
;
3048 struct shmem_sb_info
*sbinfo
;
3051 /* Round up to L1_CACHE_BYTES to resist false sharing */
3052 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3053 L1_CACHE_BYTES
), GFP_KERNEL
);
3057 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3058 sbinfo
->uid
= current_fsuid();
3059 sbinfo
->gid
= current_fsgid();
3060 sb
->s_fs_info
= sbinfo
;
3064 * Per default we only allow half of the physical ram per
3065 * tmpfs instance, limiting inodes to one per page of lowmem;
3066 * but the internal instance is left unlimited.
3068 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3069 sbinfo
->max_blocks
= shmem_default_max_blocks();
3070 sbinfo
->max_inodes
= shmem_default_max_inodes();
3071 if (shmem_parse_options(data
, sbinfo
, false)) {
3076 sb
->s_flags
|= MS_NOUSER
;
3078 sb
->s_export_op
= &shmem_export_ops
;
3079 sb
->s_flags
|= MS_NOSEC
;
3081 sb
->s_flags
|= MS_NOUSER
;
3084 spin_lock_init(&sbinfo
->stat_lock
);
3085 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3087 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3089 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3090 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
3091 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
3092 sb
->s_magic
= TMPFS_MAGIC
;
3093 sb
->s_op
= &shmem_ops
;
3094 sb
->s_time_gran
= 1;
3095 #ifdef CONFIG_TMPFS_XATTR
3096 sb
->s_xattr
= shmem_xattr_handlers
;
3098 #ifdef CONFIG_TMPFS_POSIX_ACL
3099 sb
->s_flags
|= MS_POSIXACL
;
3102 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3105 inode
->i_uid
= sbinfo
->uid
;
3106 inode
->i_gid
= sbinfo
->gid
;
3107 sb
->s_root
= d_make_root(inode
);
3113 shmem_put_super(sb
);
3117 static struct kmem_cache
*shmem_inode_cachep
;
3119 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3121 struct shmem_inode_info
*info
;
3122 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3125 return &info
->vfs_inode
;
3128 static void shmem_destroy_callback(struct rcu_head
*head
)
3130 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3131 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3134 static void shmem_destroy_inode(struct inode
*inode
)
3136 if (S_ISREG(inode
->i_mode
))
3137 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3138 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3141 static void shmem_init_inode(void *foo
)
3143 struct shmem_inode_info
*info
= foo
;
3144 inode_init_once(&info
->vfs_inode
);
3147 static int shmem_init_inodecache(void)
3149 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3150 sizeof(struct shmem_inode_info
),
3151 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3155 static void shmem_destroy_inodecache(void)
3157 kmem_cache_destroy(shmem_inode_cachep
);
3160 static const struct address_space_operations shmem_aops
= {
3161 .writepage
= shmem_writepage
,
3162 .set_page_dirty
= __set_page_dirty_no_writeback
,
3164 .write_begin
= shmem_write_begin
,
3165 .write_end
= shmem_write_end
,
3167 #ifdef CONFIG_MIGRATION
3168 .migratepage
= migrate_page
,
3170 .error_remove_page
= generic_error_remove_page
,
3173 static const struct file_operations shmem_file_operations
= {
3176 .llseek
= shmem_file_llseek
,
3177 .read_iter
= shmem_file_read_iter
,
3178 .write_iter
= generic_file_write_iter
,
3179 .fsync
= noop_fsync
,
3180 .splice_read
= shmem_file_splice_read
,
3181 .splice_write
= iter_file_splice_write
,
3182 .fallocate
= shmem_fallocate
,
3186 static const struct inode_operations shmem_inode_operations
= {
3187 .getattr
= shmem_getattr
,
3188 .setattr
= shmem_setattr
,
3189 #ifdef CONFIG_TMPFS_XATTR
3190 .setxattr
= generic_setxattr
,
3191 .getxattr
= generic_getxattr
,
3192 .listxattr
= shmem_listxattr
,
3193 .removexattr
= generic_removexattr
,
3194 .set_acl
= simple_set_acl
,
3198 static const struct inode_operations shmem_dir_inode_operations
= {
3200 .create
= shmem_create
,
3201 .lookup
= simple_lookup
,
3203 .unlink
= shmem_unlink
,
3204 .symlink
= shmem_symlink
,
3205 .mkdir
= shmem_mkdir
,
3206 .rmdir
= shmem_rmdir
,
3207 .mknod
= shmem_mknod
,
3208 .rename2
= shmem_rename2
,
3209 .tmpfile
= shmem_tmpfile
,
3211 #ifdef CONFIG_TMPFS_XATTR
3212 .setxattr
= generic_setxattr
,
3213 .getxattr
= generic_getxattr
,
3214 .listxattr
= shmem_listxattr
,
3215 .removexattr
= generic_removexattr
,
3217 #ifdef CONFIG_TMPFS_POSIX_ACL
3218 .setattr
= shmem_setattr
,
3219 .set_acl
= simple_set_acl
,
3223 static const struct inode_operations shmem_special_inode_operations
= {
3224 #ifdef CONFIG_TMPFS_XATTR
3225 .setxattr
= generic_setxattr
,
3226 .getxattr
= generic_getxattr
,
3227 .listxattr
= shmem_listxattr
,
3228 .removexattr
= generic_removexattr
,
3230 #ifdef CONFIG_TMPFS_POSIX_ACL
3231 .setattr
= shmem_setattr
,
3232 .set_acl
= simple_set_acl
,
3236 static const struct super_operations shmem_ops
= {
3237 .alloc_inode
= shmem_alloc_inode
,
3238 .destroy_inode
= shmem_destroy_inode
,
3240 .statfs
= shmem_statfs
,
3241 .remount_fs
= shmem_remount_fs
,
3242 .show_options
= shmem_show_options
,
3244 .evict_inode
= shmem_evict_inode
,
3245 .drop_inode
= generic_delete_inode
,
3246 .put_super
= shmem_put_super
,
3249 static const struct vm_operations_struct shmem_vm_ops
= {
3250 .fault
= shmem_fault
,
3251 .map_pages
= filemap_map_pages
,
3253 .set_policy
= shmem_set_policy
,
3254 .get_policy
= shmem_get_policy
,
3258 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3259 int flags
, const char *dev_name
, void *data
)
3261 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3264 static struct file_system_type shmem_fs_type
= {
3265 .owner
= THIS_MODULE
,
3267 .mount
= shmem_mount
,
3268 .kill_sb
= kill_litter_super
,
3269 .fs_flags
= FS_USERNS_MOUNT
,
3272 int __init
shmem_init(void)
3276 /* If rootfs called this, don't re-init */
3277 if (shmem_inode_cachep
)
3280 error
= shmem_init_inodecache();
3284 error
= register_filesystem(&shmem_fs_type
);
3286 printk(KERN_ERR
"Could not register tmpfs\n");
3290 shm_mnt
= kern_mount(&shmem_fs_type
);
3291 if (IS_ERR(shm_mnt
)) {
3292 error
= PTR_ERR(shm_mnt
);
3293 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
3299 unregister_filesystem(&shmem_fs_type
);
3301 shmem_destroy_inodecache();
3303 shm_mnt
= ERR_PTR(error
);
3307 #else /* !CONFIG_SHMEM */
3310 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3312 * This is intended for small system where the benefits of the full
3313 * shmem code (swap-backed and resource-limited) are outweighed by
3314 * their complexity. On systems without swap this code should be
3315 * effectively equivalent, but much lighter weight.
3318 static struct file_system_type shmem_fs_type
= {
3320 .mount
= ramfs_mount
,
3321 .kill_sb
= kill_litter_super
,
3322 .fs_flags
= FS_USERNS_MOUNT
,
3325 int __init
shmem_init(void)
3327 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
3329 shm_mnt
= kern_mount(&shmem_fs_type
);
3330 BUG_ON(IS_ERR(shm_mnt
));
3335 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
3340 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
3345 void shmem_unlock_mapping(struct address_space
*mapping
)
3349 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
3351 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
3353 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
3355 #define shmem_vm_ops generic_file_vm_ops
3356 #define shmem_file_operations ramfs_file_operations
3357 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3358 #define shmem_acct_size(flags, size) 0
3359 #define shmem_unacct_size(flags, size) do {} while (0)
3361 #endif /* CONFIG_SHMEM */
3365 static struct dentry_operations anon_ops
= {
3366 .d_dname
= simple_dname
3369 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
3370 unsigned long flags
, unsigned int i_flags
)
3373 struct inode
*inode
;
3375 struct super_block
*sb
;
3378 if (IS_ERR(shm_mnt
))
3379 return ERR_CAST(shm_mnt
);
3381 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
3382 return ERR_PTR(-EINVAL
);
3384 if (shmem_acct_size(flags
, size
))
3385 return ERR_PTR(-ENOMEM
);
3387 res
= ERR_PTR(-ENOMEM
);
3389 this.len
= strlen(name
);
3390 this.hash
= 0; /* will go */
3391 sb
= shm_mnt
->mnt_sb
;
3392 path
.mnt
= mntget(shm_mnt
);
3393 path
.dentry
= d_alloc_pseudo(sb
, &this);
3396 d_set_d_op(path
.dentry
, &anon_ops
);
3398 res
= ERR_PTR(-ENOSPC
);
3399 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
3403 inode
->i_flags
|= i_flags
;
3404 d_instantiate(path
.dentry
, inode
);
3405 inode
->i_size
= size
;
3406 clear_nlink(inode
); /* It is unlinked */
3407 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
3411 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
3412 &shmem_file_operations
);
3419 shmem_unacct_size(flags
, size
);
3426 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3427 * kernel internal. There will be NO LSM permission checks against the
3428 * underlying inode. So users of this interface must do LSM checks at a
3429 * higher layer. The users are the big_key and shm implementations. LSM
3430 * checks are provided at the key or shm level rather than the inode.
3431 * @name: name for dentry (to be seen in /proc/<pid>/maps
3432 * @size: size to be set for the file
3433 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3435 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3437 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
3441 * shmem_file_setup - get an unlinked file living in tmpfs
3442 * @name: name for dentry (to be seen in /proc/<pid>/maps
3443 * @size: size to be set for the file
3444 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3446 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3448 return __shmem_file_setup(name
, size
, flags
, 0);
3450 EXPORT_SYMBOL_GPL(shmem_file_setup
);
3453 * shmem_zero_setup - setup a shared anonymous mapping
3454 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3456 int shmem_zero_setup(struct vm_area_struct
*vma
)
3459 loff_t size
= vma
->vm_end
- vma
->vm_start
;
3462 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3463 * between XFS directory reading and selinux: since this file is only
3464 * accessible to the user through its mapping, use S_PRIVATE flag to
3465 * bypass file security, in the same way as shmem_kernel_file_setup().
3467 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
3469 return PTR_ERR(file
);
3473 vma
->vm_file
= file
;
3474 vma
->vm_ops
= &shmem_vm_ops
;
3479 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3480 * @mapping: the page's address_space
3481 * @index: the page index
3482 * @gfp: the page allocator flags to use if allocating
3484 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3485 * with any new page allocations done using the specified allocation flags.
3486 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3487 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3488 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3490 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3491 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3493 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
3494 pgoff_t index
, gfp_t gfp
)
3497 struct inode
*inode
= mapping
->host
;
3501 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
3502 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
3504 page
= ERR_PTR(error
);
3510 * The tiny !SHMEM case uses ramfs without swap
3512 return read_cache_page_gfp(mapping
, index
, gfp
);
3515 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);