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/aio.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>
76 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
77 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
79 /* Pretend that each entry is of this size in directory's i_size */
80 #define BOGO_DIRENT_SIZE 20
82 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
83 #define SHORT_SYMLINK_LEN 128
86 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
87 * inode->i_private (with i_mutex making sure that it has only one user at
88 * a time): we would prefer not to enlarge the shmem inode just for that.
91 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
92 pgoff_t start
; /* start of range currently being fallocated */
93 pgoff_t next
; /* the next page offset to be fallocated */
94 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
95 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
98 /* Flag allocation requirements to shmem_getpage */
100 SGP_READ
, /* don't exceed i_size, don't allocate page */
101 SGP_CACHE
, /* don't exceed i_size, may allocate page */
102 SGP_DIRTY
, /* like SGP_CACHE, but set new page dirty */
103 SGP_WRITE
, /* may exceed i_size, may allocate !Uptodate page */
104 SGP_FALLOC
, /* like SGP_WRITE, but make existing page Uptodate */
108 static unsigned long shmem_default_max_blocks(void)
110 return totalram_pages
/ 2;
113 static unsigned long shmem_default_max_inodes(void)
115 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
119 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
120 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
121 struct shmem_inode_info
*info
, pgoff_t index
);
122 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
123 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
);
125 static inline int shmem_getpage(struct inode
*inode
, pgoff_t index
,
126 struct page
**pagep
, enum sgp_type sgp
, int *fault_type
)
128 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
129 mapping_gfp_mask(inode
->i_mapping
), fault_type
);
132 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
134 return sb
->s_fs_info
;
138 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
139 * for shared memory and for shared anonymous (/dev/zero) mappings
140 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
141 * consistent with the pre-accounting of private mappings ...
143 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
145 return (flags
& VM_NORESERVE
) ?
146 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
149 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
151 if (!(flags
& VM_NORESERVE
))
152 vm_unacct_memory(VM_ACCT(size
));
155 static inline int shmem_reacct_size(unsigned long flags
,
156 loff_t oldsize
, loff_t newsize
)
158 if (!(flags
& VM_NORESERVE
)) {
159 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
160 return security_vm_enough_memory_mm(current
->mm
,
161 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
162 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
163 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
169 * ... whereas tmpfs objects are accounted incrementally as
170 * pages are allocated, in order to allow huge sparse files.
171 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
172 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
174 static inline int shmem_acct_block(unsigned long flags
)
176 return (flags
& VM_NORESERVE
) ?
177 security_vm_enough_memory_mm(current
->mm
, VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
180 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
182 if (flags
& VM_NORESERVE
)
183 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
186 static const struct super_operations shmem_ops
;
187 static const struct address_space_operations shmem_aops
;
188 static const struct file_operations shmem_file_operations
;
189 static const struct inode_operations shmem_inode_operations
;
190 static const struct inode_operations shmem_dir_inode_operations
;
191 static const struct inode_operations shmem_special_inode_operations
;
192 static const struct vm_operations_struct shmem_vm_ops
;
194 static struct backing_dev_info shmem_backing_dev_info __read_mostly
= {
195 .ra_pages
= 0, /* No readahead */
196 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
| BDI_CAP_SWAP_BACKED
,
199 static LIST_HEAD(shmem_swaplist
);
200 static DEFINE_MUTEX(shmem_swaplist_mutex
);
202 static int shmem_reserve_inode(struct super_block
*sb
)
204 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
205 if (sbinfo
->max_inodes
) {
206 spin_lock(&sbinfo
->stat_lock
);
207 if (!sbinfo
->free_inodes
) {
208 spin_unlock(&sbinfo
->stat_lock
);
211 sbinfo
->free_inodes
--;
212 spin_unlock(&sbinfo
->stat_lock
);
217 static void shmem_free_inode(struct super_block
*sb
)
219 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
220 if (sbinfo
->max_inodes
) {
221 spin_lock(&sbinfo
->stat_lock
);
222 sbinfo
->free_inodes
++;
223 spin_unlock(&sbinfo
->stat_lock
);
228 * shmem_recalc_inode - recalculate the block usage of an inode
229 * @inode: inode to recalc
231 * We have to calculate the free blocks since the mm can drop
232 * undirtied hole pages behind our back.
234 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
235 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
237 * It has to be called with the spinlock held.
239 static void shmem_recalc_inode(struct inode
*inode
)
241 struct shmem_inode_info
*info
= SHMEM_I(inode
);
244 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
246 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
247 if (sbinfo
->max_blocks
)
248 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
249 info
->alloced
-= freed
;
250 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
251 shmem_unacct_blocks(info
->flags
, freed
);
256 * Replace item expected in radix tree by a new item, while holding tree lock.
258 static int shmem_radix_tree_replace(struct address_space
*mapping
,
259 pgoff_t index
, void *expected
, void *replacement
)
264 VM_BUG_ON(!expected
);
265 VM_BUG_ON(!replacement
);
266 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
269 item
= radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
);
270 if (item
!= expected
)
272 radix_tree_replace_slot(pslot
, replacement
);
277 * Sometimes, before we decide whether to proceed or to fail, we must check
278 * that an entry was not already brought back from swap by a racing thread.
280 * Checking page is not enough: by the time a SwapCache page is locked, it
281 * might be reused, and again be SwapCache, using the same swap as before.
283 static bool shmem_confirm_swap(struct address_space
*mapping
,
284 pgoff_t index
, swp_entry_t swap
)
289 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
291 return item
== swp_to_radix_entry(swap
);
295 * Like add_to_page_cache_locked, but error if expected item has gone.
297 static int shmem_add_to_page_cache(struct page
*page
,
298 struct address_space
*mapping
,
299 pgoff_t index
, void *expected
)
303 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
304 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
306 page_cache_get(page
);
307 page
->mapping
= mapping
;
310 spin_lock_irq(&mapping
->tree_lock
);
312 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
314 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
318 __inc_zone_page_state(page
, NR_FILE_PAGES
);
319 __inc_zone_page_state(page
, NR_SHMEM
);
320 spin_unlock_irq(&mapping
->tree_lock
);
322 page
->mapping
= NULL
;
323 spin_unlock_irq(&mapping
->tree_lock
);
324 page_cache_release(page
);
330 * Like delete_from_page_cache, but substitutes swap for page.
332 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
334 struct address_space
*mapping
= page
->mapping
;
337 spin_lock_irq(&mapping
->tree_lock
);
338 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
339 page
->mapping
= NULL
;
341 __dec_zone_page_state(page
, NR_FILE_PAGES
);
342 __dec_zone_page_state(page
, NR_SHMEM
);
343 spin_unlock_irq(&mapping
->tree_lock
);
344 page_cache_release(page
);
349 * Remove swap entry from radix tree, free the swap and its page cache.
351 static int shmem_free_swap(struct address_space
*mapping
,
352 pgoff_t index
, void *radswap
)
356 spin_lock_irq(&mapping
->tree_lock
);
357 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
358 spin_unlock_irq(&mapping
->tree_lock
);
361 free_swap_and_cache(radix_to_swp_entry(radswap
));
366 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
368 void shmem_unlock_mapping(struct address_space
*mapping
)
371 pgoff_t indices
[PAGEVEC_SIZE
];
374 pagevec_init(&pvec
, 0);
376 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
378 while (!mapping_unevictable(mapping
)) {
380 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
381 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
383 pvec
.nr
= find_get_entries(mapping
, index
,
384 PAGEVEC_SIZE
, pvec
.pages
, indices
);
387 index
= indices
[pvec
.nr
- 1] + 1;
388 pagevec_remove_exceptionals(&pvec
);
389 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
390 pagevec_release(&pvec
);
396 * Remove range of pages and swap entries from radix tree, and free them.
397 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
399 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
402 struct address_space
*mapping
= inode
->i_mapping
;
403 struct shmem_inode_info
*info
= SHMEM_I(inode
);
404 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
405 pgoff_t end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
406 unsigned int partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
407 unsigned int partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
409 pgoff_t indices
[PAGEVEC_SIZE
];
410 long nr_swaps_freed
= 0;
415 end
= -1; /* unsigned, so actually very big */
417 pagevec_init(&pvec
, 0);
419 while (index
< end
) {
420 pvec
.nr
= find_get_entries(mapping
, index
,
421 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
422 pvec
.pages
, indices
);
425 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
426 struct page
*page
= pvec
.pages
[i
];
432 if (radix_tree_exceptional_entry(page
)) {
435 nr_swaps_freed
+= !shmem_free_swap(mapping
,
440 if (!trylock_page(page
))
442 if (!unfalloc
|| !PageUptodate(page
)) {
443 if (page
->mapping
== mapping
) {
444 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
445 truncate_inode_page(mapping
, page
);
450 pagevec_remove_exceptionals(&pvec
);
451 pagevec_release(&pvec
);
457 struct page
*page
= NULL
;
458 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
460 unsigned int top
= PAGE_CACHE_SIZE
;
465 zero_user_segment(page
, partial_start
, top
);
466 set_page_dirty(page
);
468 page_cache_release(page
);
472 struct page
*page
= NULL
;
473 shmem_getpage(inode
, end
, &page
, SGP_READ
, NULL
);
475 zero_user_segment(page
, 0, partial_end
);
476 set_page_dirty(page
);
478 page_cache_release(page
);
485 while (index
< end
) {
488 pvec
.nr
= find_get_entries(mapping
, index
,
489 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
490 pvec
.pages
, indices
);
492 /* If all gone or hole-punch or unfalloc, we're done */
493 if (index
== start
|| end
!= -1)
495 /* But if truncating, restart to make sure all gone */
499 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
500 struct page
*page
= pvec
.pages
[i
];
506 if (radix_tree_exceptional_entry(page
)) {
509 if (shmem_free_swap(mapping
, index
, page
)) {
510 /* Swap was replaced by page: retry */
519 if (!unfalloc
|| !PageUptodate(page
)) {
520 if (page
->mapping
== mapping
) {
521 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
522 truncate_inode_page(mapping
, page
);
524 /* Page was replaced by swap: retry */
532 pagevec_remove_exceptionals(&pvec
);
533 pagevec_release(&pvec
);
537 spin_lock(&info
->lock
);
538 info
->swapped
-= nr_swaps_freed
;
539 shmem_recalc_inode(inode
);
540 spin_unlock(&info
->lock
);
543 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
545 shmem_undo_range(inode
, lstart
, lend
, false);
546 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
548 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
550 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
552 struct inode
*inode
= dentry
->d_inode
;
553 struct shmem_inode_info
*info
= SHMEM_I(inode
);
556 error
= inode_change_ok(inode
, attr
);
560 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
561 loff_t oldsize
= inode
->i_size
;
562 loff_t newsize
= attr
->ia_size
;
564 /* protected by i_mutex */
565 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
566 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
569 if (newsize
!= oldsize
) {
570 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
574 i_size_write(inode
, newsize
);
575 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
577 if (newsize
< oldsize
) {
578 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
579 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
580 shmem_truncate_range(inode
, newsize
, (loff_t
)-1);
581 /* unmap again to remove racily COWed private pages */
582 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
586 setattr_copy(inode
, attr
);
587 if (attr
->ia_valid
& ATTR_MODE
)
588 error
= posix_acl_chmod(inode
, inode
->i_mode
);
592 static void shmem_evict_inode(struct inode
*inode
)
594 struct shmem_inode_info
*info
= SHMEM_I(inode
);
596 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
597 shmem_unacct_size(info
->flags
, inode
->i_size
);
599 shmem_truncate_range(inode
, 0, (loff_t
)-1);
600 if (!list_empty(&info
->swaplist
)) {
601 mutex_lock(&shmem_swaplist_mutex
);
602 list_del_init(&info
->swaplist
);
603 mutex_unlock(&shmem_swaplist_mutex
);
606 kfree(info
->symlink
);
608 simple_xattrs_free(&info
->xattrs
);
609 WARN_ON(inode
->i_blocks
);
610 shmem_free_inode(inode
->i_sb
);
615 * If swap found in inode, free it and move page from swapcache to filecache.
617 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
618 swp_entry_t swap
, struct page
**pagep
)
620 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
626 radswap
= swp_to_radix_entry(swap
);
627 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
629 return -EAGAIN
; /* tell shmem_unuse we found nothing */
632 * Move _head_ to start search for next from here.
633 * But be careful: shmem_evict_inode checks list_empty without taking
634 * mutex, and there's an instant in list_move_tail when info->swaplist
635 * would appear empty, if it were the only one on shmem_swaplist.
637 if (shmem_swaplist
.next
!= &info
->swaplist
)
638 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
640 gfp
= mapping_gfp_mask(mapping
);
641 if (shmem_should_replace_page(*pagep
, gfp
)) {
642 mutex_unlock(&shmem_swaplist_mutex
);
643 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
644 mutex_lock(&shmem_swaplist_mutex
);
646 * We needed to drop mutex to make that restrictive page
647 * allocation, but the inode might have been freed while we
648 * dropped it: although a racing shmem_evict_inode() cannot
649 * complete without emptying the radix_tree, our page lock
650 * on this swapcache page is not enough to prevent that -
651 * free_swap_and_cache() of our swap entry will only
652 * trylock_page(), removing swap from radix_tree whatever.
654 * We must not proceed to shmem_add_to_page_cache() if the
655 * inode has been freed, but of course we cannot rely on
656 * inode or mapping or info to check that. However, we can
657 * safely check if our swap entry is still in use (and here
658 * it can't have got reused for another page): if it's still
659 * in use, then the inode cannot have been freed yet, and we
660 * can safely proceed (if it's no longer in use, that tells
661 * nothing about the inode, but we don't need to unuse swap).
663 if (!page_swapcount(*pagep
))
668 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
669 * but also to hold up shmem_evict_inode(): so inode cannot be freed
670 * beneath us (pagelock doesn't help until the page is in pagecache).
673 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
675 if (error
!= -ENOMEM
) {
677 * Truncation and eviction use free_swap_and_cache(), which
678 * only does trylock page: if we raced, best clean up here.
680 delete_from_swap_cache(*pagep
);
681 set_page_dirty(*pagep
);
683 spin_lock(&info
->lock
);
685 spin_unlock(&info
->lock
);
693 * Search through swapped inodes to find and replace swap by page.
695 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
697 struct list_head
*this, *next
;
698 struct shmem_inode_info
*info
;
699 struct mem_cgroup
*memcg
;
703 * There's a faint possibility that swap page was replaced before
704 * caller locked it: caller will come back later with the right page.
706 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
710 * Charge page using GFP_KERNEL while we can wait, before taking
711 * the shmem_swaplist_mutex which might hold up shmem_writepage().
712 * Charged back to the user (not to caller) when swap account is used.
714 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
);
717 /* No radix_tree_preload: swap entry keeps a place for page in tree */
720 mutex_lock(&shmem_swaplist_mutex
);
721 list_for_each_safe(this, next
, &shmem_swaplist
) {
722 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
724 error
= shmem_unuse_inode(info
, swap
, &page
);
726 list_del_init(&info
->swaplist
);
728 if (error
!= -EAGAIN
)
730 /* found nothing in this: move on to search the next */
732 mutex_unlock(&shmem_swaplist_mutex
);
735 if (error
!= -ENOMEM
)
737 mem_cgroup_cancel_charge(page
, memcg
);
739 mem_cgroup_commit_charge(page
, memcg
, true);
742 page_cache_release(page
);
747 * Move the page from the page cache to the swap cache.
749 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
751 struct shmem_inode_info
*info
;
752 struct address_space
*mapping
;
757 BUG_ON(!PageLocked(page
));
758 mapping
= page
->mapping
;
760 inode
= mapping
->host
;
761 info
= SHMEM_I(inode
);
762 if (info
->flags
& VM_LOCKED
)
764 if (!total_swap_pages
)
768 * shmem_backing_dev_info's capabilities prevent regular writeback or
769 * sync from ever calling shmem_writepage; but a stacking filesystem
770 * might use ->writepage of its underlying filesystem, in which case
771 * tmpfs should write out to swap only in response to memory pressure,
772 * and not for the writeback threads or sync.
774 if (!wbc
->for_reclaim
) {
775 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
780 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
781 * value into swapfile.c, the only way we can correctly account for a
782 * fallocated page arriving here is now to initialize it and write it.
784 * That's okay for a page already fallocated earlier, but if we have
785 * not yet completed the fallocation, then (a) we want to keep track
786 * of this page in case we have to undo it, and (b) it may not be a
787 * good idea to continue anyway, once we're pushing into swap. So
788 * reactivate the page, and let shmem_fallocate() quit when too many.
790 if (!PageUptodate(page
)) {
791 if (inode
->i_private
) {
792 struct shmem_falloc
*shmem_falloc
;
793 spin_lock(&inode
->i_lock
);
794 shmem_falloc
= inode
->i_private
;
796 !shmem_falloc
->waitq
&&
797 index
>= shmem_falloc
->start
&&
798 index
< shmem_falloc
->next
)
799 shmem_falloc
->nr_unswapped
++;
802 spin_unlock(&inode
->i_lock
);
806 clear_highpage(page
);
807 flush_dcache_page(page
);
808 SetPageUptodate(page
);
811 swap
= get_swap_page();
816 * Add inode to shmem_unuse()'s list of swapped-out inodes,
817 * if it's not already there. Do it now before the page is
818 * moved to swap cache, when its pagelock no longer protects
819 * the inode from eviction. But don't unlock the mutex until
820 * we've incremented swapped, because shmem_unuse_inode() will
821 * prune a !swapped inode from the swaplist under this mutex.
823 mutex_lock(&shmem_swaplist_mutex
);
824 if (list_empty(&info
->swaplist
))
825 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
827 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
828 swap_shmem_alloc(swap
);
829 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
831 spin_lock(&info
->lock
);
833 shmem_recalc_inode(inode
);
834 spin_unlock(&info
->lock
);
836 mutex_unlock(&shmem_swaplist_mutex
);
837 BUG_ON(page_mapped(page
));
838 swap_writepage(page
, wbc
);
842 mutex_unlock(&shmem_swaplist_mutex
);
843 swapcache_free(swap
);
845 set_page_dirty(page
);
846 if (wbc
->for_reclaim
)
847 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
854 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
858 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
859 return; /* show nothing */
861 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
863 seq_printf(seq
, ",mpol=%s", buffer
);
866 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
868 struct mempolicy
*mpol
= NULL
;
870 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
873 spin_unlock(&sbinfo
->stat_lock
);
877 #endif /* CONFIG_TMPFS */
879 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
880 struct shmem_inode_info
*info
, pgoff_t index
)
882 struct vm_area_struct pvma
;
885 /* Create a pseudo vma that just contains the policy */
887 /* Bias interleave by inode number to distribute better across nodes */
888 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
890 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
892 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
894 /* Drop reference taken by mpol_shared_policy_lookup() */
895 mpol_cond_put(pvma
.vm_policy
);
900 static struct page
*shmem_alloc_page(gfp_t gfp
,
901 struct shmem_inode_info
*info
, pgoff_t index
)
903 struct vm_area_struct pvma
;
906 /* Create a pseudo vma that just contains the policy */
908 /* Bias interleave by inode number to distribute better across nodes */
909 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
911 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
913 page
= alloc_page_vma(gfp
, &pvma
, 0);
915 /* Drop reference taken by mpol_shared_policy_lookup() */
916 mpol_cond_put(pvma
.vm_policy
);
920 #else /* !CONFIG_NUMA */
922 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
925 #endif /* CONFIG_TMPFS */
927 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
928 struct shmem_inode_info
*info
, pgoff_t index
)
930 return swapin_readahead(swap
, gfp
, NULL
, 0);
933 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
934 struct shmem_inode_info
*info
, pgoff_t index
)
936 return alloc_page(gfp
);
938 #endif /* CONFIG_NUMA */
940 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
941 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
948 * When a page is moved from swapcache to shmem filecache (either by the
949 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
950 * shmem_unuse_inode()), it may have been read in earlier from swap, in
951 * ignorance of the mapping it belongs to. If that mapping has special
952 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
953 * we may need to copy to a suitable page before moving to filecache.
955 * In a future release, this may well be extended to respect cpuset and
956 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
957 * but for now it is a simple matter of zone.
959 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
961 return page_zonenum(page
) > gfp_zone(gfp
);
964 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
965 struct shmem_inode_info
*info
, pgoff_t index
)
967 struct page
*oldpage
, *newpage
;
968 struct address_space
*swap_mapping
;
973 swap_index
= page_private(oldpage
);
974 swap_mapping
= page_mapping(oldpage
);
977 * We have arrived here because our zones are constrained, so don't
978 * limit chance of success by further cpuset and node constraints.
980 gfp
&= ~GFP_CONSTRAINT_MASK
;
981 newpage
= shmem_alloc_page(gfp
, info
, index
);
985 page_cache_get(newpage
);
986 copy_highpage(newpage
, oldpage
);
987 flush_dcache_page(newpage
);
989 __set_page_locked(newpage
);
990 SetPageUptodate(newpage
);
991 SetPageSwapBacked(newpage
);
992 set_page_private(newpage
, swap_index
);
993 SetPageSwapCache(newpage
);
996 * Our caller will very soon move newpage out of swapcache, but it's
997 * a nice clean interface for us to replace oldpage by newpage there.
999 spin_lock_irq(&swap_mapping
->tree_lock
);
1000 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1003 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
1004 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
1006 spin_unlock_irq(&swap_mapping
->tree_lock
);
1008 if (unlikely(error
)) {
1010 * Is this possible? I think not, now that our callers check
1011 * both PageSwapCache and page_private after getting page lock;
1012 * but be defensive. Reverse old to newpage for clear and free.
1016 mem_cgroup_migrate(oldpage
, newpage
, false);
1017 lru_cache_add_anon(newpage
);
1021 ClearPageSwapCache(oldpage
);
1022 set_page_private(oldpage
, 0);
1024 unlock_page(oldpage
);
1025 page_cache_release(oldpage
);
1026 page_cache_release(oldpage
);
1031 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1033 * If we allocate a new one we do not mark it dirty. That's up to the
1034 * vm. If we swap it in we mark it dirty since we also free the swap
1035 * entry since a page cannot live in both the swap and page cache
1037 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1038 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
1040 struct address_space
*mapping
= inode
->i_mapping
;
1041 struct shmem_inode_info
*info
;
1042 struct shmem_sb_info
*sbinfo
;
1043 struct mem_cgroup
*memcg
;
1050 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
1054 page
= find_lock_entry(mapping
, index
);
1055 if (radix_tree_exceptional_entry(page
)) {
1056 swap
= radix_to_swp_entry(page
);
1060 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1061 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1066 if (page
&& sgp
== SGP_WRITE
)
1067 mark_page_accessed(page
);
1069 /* fallocated page? */
1070 if (page
&& !PageUptodate(page
)) {
1071 if (sgp
!= SGP_READ
)
1074 page_cache_release(page
);
1077 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1083 * Fast cache lookup did not find it:
1084 * bring it back from swap or allocate.
1086 info
= SHMEM_I(inode
);
1087 sbinfo
= SHMEM_SB(inode
->i_sb
);
1090 /* Look it up and read it in.. */
1091 page
= lookup_swap_cache(swap
);
1093 /* here we actually do the io */
1095 *fault_type
|= VM_FAULT_MAJOR
;
1096 page
= shmem_swapin(swap
, gfp
, info
, index
);
1103 /* We have to do this with page locked to prevent races */
1105 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1106 !shmem_confirm_swap(mapping
, index
, swap
)) {
1107 error
= -EEXIST
; /* try again */
1110 if (!PageUptodate(page
)) {
1114 wait_on_page_writeback(page
);
1116 if (shmem_should_replace_page(page
, gfp
)) {
1117 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1122 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
);
1124 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1125 swp_to_radix_entry(swap
));
1127 * We already confirmed swap under page lock, and make
1128 * no memory allocation here, so usually no possibility
1129 * of error; but free_swap_and_cache() only trylocks a
1130 * page, so it is just possible that the entry has been
1131 * truncated or holepunched since swap was confirmed.
1132 * shmem_undo_range() will have done some of the
1133 * unaccounting, now delete_from_swap_cache() will do
1134 * the rest (including mem_cgroup_uncharge_swapcache).
1135 * Reset swap.val? No, leave it so "failed" goes back to
1136 * "repeat": reading a hole and writing should succeed.
1139 mem_cgroup_cancel_charge(page
, memcg
);
1140 delete_from_swap_cache(page
);
1146 mem_cgroup_commit_charge(page
, memcg
, true);
1148 spin_lock(&info
->lock
);
1150 shmem_recalc_inode(inode
);
1151 spin_unlock(&info
->lock
);
1153 if (sgp
== SGP_WRITE
)
1154 mark_page_accessed(page
);
1156 delete_from_swap_cache(page
);
1157 set_page_dirty(page
);
1161 if (shmem_acct_block(info
->flags
)) {
1165 if (sbinfo
->max_blocks
) {
1166 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1167 sbinfo
->max_blocks
) >= 0) {
1171 percpu_counter_inc(&sbinfo
->used_blocks
);
1174 page
= shmem_alloc_page(gfp
, info
, index
);
1180 __SetPageSwapBacked(page
);
1181 __set_page_locked(page
);
1182 if (sgp
== SGP_WRITE
)
1183 __SetPageReferenced(page
);
1185 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
);
1188 error
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
1190 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1192 radix_tree_preload_end();
1195 mem_cgroup_cancel_charge(page
, memcg
);
1198 mem_cgroup_commit_charge(page
, memcg
, false);
1199 lru_cache_add_anon(page
);
1201 spin_lock(&info
->lock
);
1203 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1204 shmem_recalc_inode(inode
);
1205 spin_unlock(&info
->lock
);
1209 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1211 if (sgp
== SGP_FALLOC
)
1215 * Let SGP_WRITE caller clear ends if write does not fill page;
1216 * but SGP_FALLOC on a page fallocated earlier must initialize
1217 * it now, lest undo on failure cancel our earlier guarantee.
1219 if (sgp
!= SGP_WRITE
) {
1220 clear_highpage(page
);
1221 flush_dcache_page(page
);
1222 SetPageUptodate(page
);
1224 if (sgp
== SGP_DIRTY
)
1225 set_page_dirty(page
);
1228 /* Perhaps the file has been truncated since we checked */
1229 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1230 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1244 info
= SHMEM_I(inode
);
1245 ClearPageDirty(page
);
1246 delete_from_page_cache(page
);
1247 spin_lock(&info
->lock
);
1249 inode
->i_blocks
-= BLOCKS_PER_PAGE
;
1250 spin_unlock(&info
->lock
);
1252 sbinfo
= SHMEM_SB(inode
->i_sb
);
1253 if (sbinfo
->max_blocks
)
1254 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1256 shmem_unacct_blocks(info
->flags
, 1);
1258 if (swap
.val
&& error
!= -EINVAL
&&
1259 !shmem_confirm_swap(mapping
, index
, swap
))
1264 page_cache_release(page
);
1266 if (error
== -ENOSPC
&& !once
++) {
1267 info
= SHMEM_I(inode
);
1268 spin_lock(&info
->lock
);
1269 shmem_recalc_inode(inode
);
1270 spin_unlock(&info
->lock
);
1273 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1278 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1280 struct inode
*inode
= file_inode(vma
->vm_file
);
1282 int ret
= VM_FAULT_LOCKED
;
1285 * Trinity finds that probing a hole which tmpfs is punching can
1286 * prevent the hole-punch from ever completing: which in turn
1287 * locks writers out with its hold on i_mutex. So refrain from
1288 * faulting pages into the hole while it's being punched. Although
1289 * shmem_undo_range() does remove the additions, it may be unable to
1290 * keep up, as each new page needs its own unmap_mapping_range() call,
1291 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1293 * It does not matter if we sometimes reach this check just before the
1294 * hole-punch begins, so that one fault then races with the punch:
1295 * we just need to make racing faults a rare case.
1297 * The implementation below would be much simpler if we just used a
1298 * standard mutex or completion: but we cannot take i_mutex in fault,
1299 * and bloating every shmem inode for this unlikely case would be sad.
1301 if (unlikely(inode
->i_private
)) {
1302 struct shmem_falloc
*shmem_falloc
;
1304 spin_lock(&inode
->i_lock
);
1305 shmem_falloc
= inode
->i_private
;
1307 shmem_falloc
->waitq
&&
1308 vmf
->pgoff
>= shmem_falloc
->start
&&
1309 vmf
->pgoff
< shmem_falloc
->next
) {
1310 wait_queue_head_t
*shmem_falloc_waitq
;
1311 DEFINE_WAIT(shmem_fault_wait
);
1313 ret
= VM_FAULT_NOPAGE
;
1314 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1315 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1316 /* It's polite to up mmap_sem if we can */
1317 up_read(&vma
->vm_mm
->mmap_sem
);
1318 ret
= VM_FAULT_RETRY
;
1321 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1322 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1323 TASK_UNINTERRUPTIBLE
);
1324 spin_unlock(&inode
->i_lock
);
1328 * shmem_falloc_waitq points into the shmem_fallocate()
1329 * stack of the hole-punching task: shmem_falloc_waitq
1330 * is usually invalid by the time we reach here, but
1331 * finish_wait() does not dereference it in that case;
1332 * though i_lock needed lest racing with wake_up_all().
1334 spin_lock(&inode
->i_lock
);
1335 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1336 spin_unlock(&inode
->i_lock
);
1339 spin_unlock(&inode
->i_lock
);
1342 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1344 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1346 if (ret
& VM_FAULT_MAJOR
) {
1347 count_vm_event(PGMAJFAULT
);
1348 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1354 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1356 struct inode
*inode
= file_inode(vma
->vm_file
);
1357 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1360 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1363 struct inode
*inode
= file_inode(vma
->vm_file
);
1366 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1367 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1371 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1373 struct inode
*inode
= file_inode(file
);
1374 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1375 int retval
= -ENOMEM
;
1377 spin_lock(&info
->lock
);
1378 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1379 if (!user_shm_lock(inode
->i_size
, user
))
1381 info
->flags
|= VM_LOCKED
;
1382 mapping_set_unevictable(file
->f_mapping
);
1384 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1385 user_shm_unlock(inode
->i_size
, user
);
1386 info
->flags
&= ~VM_LOCKED
;
1387 mapping_clear_unevictable(file
->f_mapping
);
1392 spin_unlock(&info
->lock
);
1396 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1398 file_accessed(file
);
1399 vma
->vm_ops
= &shmem_vm_ops
;
1403 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1404 umode_t mode
, dev_t dev
, unsigned long flags
)
1406 struct inode
*inode
;
1407 struct shmem_inode_info
*info
;
1408 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1410 if (shmem_reserve_inode(sb
))
1413 inode
= new_inode(sb
);
1415 inode
->i_ino
= get_next_ino();
1416 inode_init_owner(inode
, dir
, mode
);
1417 inode
->i_blocks
= 0;
1418 inode
->i_mapping
->backing_dev_info
= &shmem_backing_dev_info
;
1419 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1420 inode
->i_generation
= get_seconds();
1421 info
= SHMEM_I(inode
);
1422 memset(info
, 0, (char *)inode
- (char *)info
);
1423 spin_lock_init(&info
->lock
);
1424 info
->seals
= F_SEAL_SEAL
;
1425 info
->flags
= flags
& VM_NORESERVE
;
1426 INIT_LIST_HEAD(&info
->swaplist
);
1427 simple_xattrs_init(&info
->xattrs
);
1428 cache_no_acl(inode
);
1430 switch (mode
& S_IFMT
) {
1432 inode
->i_op
= &shmem_special_inode_operations
;
1433 init_special_inode(inode
, mode
, dev
);
1436 inode
->i_mapping
->a_ops
= &shmem_aops
;
1437 inode
->i_op
= &shmem_inode_operations
;
1438 inode
->i_fop
= &shmem_file_operations
;
1439 mpol_shared_policy_init(&info
->policy
,
1440 shmem_get_sbmpol(sbinfo
));
1444 /* Some things misbehave if size == 0 on a directory */
1445 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1446 inode
->i_op
= &shmem_dir_inode_operations
;
1447 inode
->i_fop
= &simple_dir_operations
;
1451 * Must not load anything in the rbtree,
1452 * mpol_free_shared_policy will not be called.
1454 mpol_shared_policy_init(&info
->policy
, NULL
);
1458 shmem_free_inode(sb
);
1462 bool shmem_mapping(struct address_space
*mapping
)
1464 return mapping
->backing_dev_info
== &shmem_backing_dev_info
;
1468 static const struct inode_operations shmem_symlink_inode_operations
;
1469 static const struct inode_operations shmem_short_symlink_operations
;
1471 #ifdef CONFIG_TMPFS_XATTR
1472 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1474 #define shmem_initxattrs NULL
1478 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1479 loff_t pos
, unsigned len
, unsigned flags
,
1480 struct page
**pagep
, void **fsdata
)
1482 struct inode
*inode
= mapping
->host
;
1483 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1484 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1486 /* i_mutex is held by caller */
1487 if (unlikely(info
->seals
)) {
1488 if (info
->seals
& F_SEAL_WRITE
)
1490 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
1494 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1498 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1499 loff_t pos
, unsigned len
, unsigned copied
,
1500 struct page
*page
, void *fsdata
)
1502 struct inode
*inode
= mapping
->host
;
1504 if (pos
+ copied
> inode
->i_size
)
1505 i_size_write(inode
, pos
+ copied
);
1507 if (!PageUptodate(page
)) {
1508 if (copied
< PAGE_CACHE_SIZE
) {
1509 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1510 zero_user_segments(page
, 0, from
,
1511 from
+ copied
, PAGE_CACHE_SIZE
);
1513 SetPageUptodate(page
);
1515 set_page_dirty(page
);
1517 page_cache_release(page
);
1522 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1524 struct file
*file
= iocb
->ki_filp
;
1525 struct inode
*inode
= file_inode(file
);
1526 struct address_space
*mapping
= inode
->i_mapping
;
1528 unsigned long offset
;
1529 enum sgp_type sgp
= SGP_READ
;
1532 loff_t
*ppos
= &iocb
->ki_pos
;
1535 * Might this read be for a stacking filesystem? Then when reading
1536 * holes of a sparse file, we actually need to allocate those pages,
1537 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1539 if (segment_eq(get_fs(), KERNEL_DS
))
1542 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1543 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1546 struct page
*page
= NULL
;
1548 unsigned long nr
, ret
;
1549 loff_t i_size
= i_size_read(inode
);
1551 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1552 if (index
> end_index
)
1554 if (index
== end_index
) {
1555 nr
= i_size
& ~PAGE_CACHE_MASK
;
1560 error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1562 if (error
== -EINVAL
)
1570 * We must evaluate after, since reads (unlike writes)
1571 * are called without i_mutex protection against truncate
1573 nr
= PAGE_CACHE_SIZE
;
1574 i_size
= i_size_read(inode
);
1575 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1576 if (index
== end_index
) {
1577 nr
= i_size
& ~PAGE_CACHE_MASK
;
1580 page_cache_release(page
);
1588 * If users can be writing to this page using arbitrary
1589 * virtual addresses, take care about potential aliasing
1590 * before reading the page on the kernel side.
1592 if (mapping_writably_mapped(mapping
))
1593 flush_dcache_page(page
);
1595 * Mark the page accessed if we read the beginning.
1598 mark_page_accessed(page
);
1600 page
= ZERO_PAGE(0);
1601 page_cache_get(page
);
1605 * Ok, we have the page, and it's up-to-date, so
1606 * now we can copy it to user space...
1608 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
1611 index
+= offset
>> PAGE_CACHE_SHIFT
;
1612 offset
&= ~PAGE_CACHE_MASK
;
1614 page_cache_release(page
);
1615 if (!iov_iter_count(to
))
1624 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1625 file_accessed(file
);
1626 return retval
? retval
: error
;
1629 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1630 struct pipe_inode_info
*pipe
, size_t len
,
1633 struct address_space
*mapping
= in
->f_mapping
;
1634 struct inode
*inode
= mapping
->host
;
1635 unsigned int loff
, nr_pages
, req_pages
;
1636 struct page
*pages
[PIPE_DEF_BUFFERS
];
1637 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1639 pgoff_t index
, end_index
;
1642 struct splice_pipe_desc spd
= {
1645 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1647 .ops
= &page_cache_pipe_buf_ops
,
1648 .spd_release
= spd_release_page
,
1651 isize
= i_size_read(inode
);
1652 if (unlikely(*ppos
>= isize
))
1655 left
= isize
- *ppos
;
1656 if (unlikely(left
< len
))
1659 if (splice_grow_spd(pipe
, &spd
))
1662 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1663 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1664 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1665 nr_pages
= min(req_pages
, spd
.nr_pages_max
);
1667 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1668 nr_pages
, spd
.pages
);
1669 index
+= spd
.nr_pages
;
1672 while (spd
.nr_pages
< nr_pages
) {
1673 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1677 spd
.pages
[spd
.nr_pages
++] = page
;
1681 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1682 nr_pages
= spd
.nr_pages
;
1685 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1686 unsigned int this_len
;
1691 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1692 page
= spd
.pages
[page_nr
];
1694 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1695 error
= shmem_getpage(inode
, index
, &page
,
1700 page_cache_release(spd
.pages
[page_nr
]);
1701 spd
.pages
[page_nr
] = page
;
1704 isize
= i_size_read(inode
);
1705 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1706 if (unlikely(!isize
|| index
> end_index
))
1709 if (end_index
== index
) {
1712 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1716 this_len
= min(this_len
, plen
- loff
);
1720 spd
.partial
[page_nr
].offset
= loff
;
1721 spd
.partial
[page_nr
].len
= this_len
;
1728 while (page_nr
< nr_pages
)
1729 page_cache_release(spd
.pages
[page_nr
++]);
1732 error
= splice_to_pipe(pipe
, &spd
);
1734 splice_shrink_spd(&spd
);
1744 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1746 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
1747 pgoff_t index
, pgoff_t end
, int whence
)
1750 struct pagevec pvec
;
1751 pgoff_t indices
[PAGEVEC_SIZE
];
1755 pagevec_init(&pvec
, 0);
1756 pvec
.nr
= 1; /* start small: we may be there already */
1758 pvec
.nr
= find_get_entries(mapping
, index
,
1759 pvec
.nr
, pvec
.pages
, indices
);
1761 if (whence
== SEEK_DATA
)
1765 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
1766 if (index
< indices
[i
]) {
1767 if (whence
== SEEK_HOLE
) {
1773 page
= pvec
.pages
[i
];
1774 if (page
&& !radix_tree_exceptional_entry(page
)) {
1775 if (!PageUptodate(page
))
1779 (page
&& whence
== SEEK_DATA
) ||
1780 (!page
&& whence
== SEEK_HOLE
)) {
1785 pagevec_remove_exceptionals(&pvec
);
1786 pagevec_release(&pvec
);
1787 pvec
.nr
= PAGEVEC_SIZE
;
1793 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
1795 struct address_space
*mapping
= file
->f_mapping
;
1796 struct inode
*inode
= mapping
->host
;
1800 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
1801 return generic_file_llseek_size(file
, offset
, whence
,
1802 MAX_LFS_FILESIZE
, i_size_read(inode
));
1803 mutex_lock(&inode
->i_mutex
);
1804 /* We're holding i_mutex so we can access i_size directly */
1808 else if (offset
>= inode
->i_size
)
1811 start
= offset
>> PAGE_CACHE_SHIFT
;
1812 end
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1813 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
1814 new_offset
<<= PAGE_CACHE_SHIFT
;
1815 if (new_offset
> offset
) {
1816 if (new_offset
< inode
->i_size
)
1817 offset
= new_offset
;
1818 else if (whence
== SEEK_DATA
)
1821 offset
= inode
->i_size
;
1826 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
1827 mutex_unlock(&inode
->i_mutex
);
1832 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1833 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1835 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1836 #define LAST_SCAN 4 /* about 150ms max */
1838 static void shmem_tag_pins(struct address_space
*mapping
)
1840 struct radix_tree_iter iter
;
1850 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1851 page
= radix_tree_deref_slot(slot
);
1852 if (!page
|| radix_tree_exception(page
)) {
1853 if (radix_tree_deref_retry(page
))
1855 } else if (page_count(page
) - page_mapcount(page
) > 1) {
1856 spin_lock_irq(&mapping
->tree_lock
);
1857 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
1859 spin_unlock_irq(&mapping
->tree_lock
);
1862 if (need_resched()) {
1864 start
= iter
.index
+ 1;
1872 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1873 * via get_user_pages(), drivers might have some pending I/O without any active
1874 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1875 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1876 * them to be dropped.
1877 * The caller must guarantee that no new user will acquire writable references
1878 * to those pages to avoid races.
1880 static int shmem_wait_for_pins(struct address_space
*mapping
)
1882 struct radix_tree_iter iter
;
1888 shmem_tag_pins(mapping
);
1891 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
1892 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
1896 lru_add_drain_all();
1897 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
1903 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
1904 start
, SHMEM_TAG_PINNED
) {
1906 page
= radix_tree_deref_slot(slot
);
1907 if (radix_tree_exception(page
)) {
1908 if (radix_tree_deref_retry(page
))
1915 page_count(page
) - page_mapcount(page
) != 1) {
1916 if (scan
< LAST_SCAN
)
1917 goto continue_resched
;
1920 * On the last scan, we clean up all those tags
1921 * we inserted; but make a note that we still
1922 * found pages pinned.
1927 spin_lock_irq(&mapping
->tree_lock
);
1928 radix_tree_tag_clear(&mapping
->page_tree
,
1929 iter
.index
, SHMEM_TAG_PINNED
);
1930 spin_unlock_irq(&mapping
->tree_lock
);
1932 if (need_resched()) {
1934 start
= iter
.index
+ 1;
1944 #define F_ALL_SEALS (F_SEAL_SEAL | \
1949 int shmem_add_seals(struct file
*file
, unsigned int seals
)
1951 struct inode
*inode
= file_inode(file
);
1952 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1957 * Sealing allows multiple parties to share a shmem-file but restrict
1958 * access to a specific subset of file operations. Seals can only be
1959 * added, but never removed. This way, mutually untrusted parties can
1960 * share common memory regions with a well-defined policy. A malicious
1961 * peer can thus never perform unwanted operations on a shared object.
1963 * Seals are only supported on special shmem-files and always affect
1964 * the whole underlying inode. Once a seal is set, it may prevent some
1965 * kinds of access to the file. Currently, the following seals are
1967 * SEAL_SEAL: Prevent further seals from being set on this file
1968 * SEAL_SHRINK: Prevent the file from shrinking
1969 * SEAL_GROW: Prevent the file from growing
1970 * SEAL_WRITE: Prevent write access to the file
1972 * As we don't require any trust relationship between two parties, we
1973 * must prevent seals from being removed. Therefore, sealing a file
1974 * only adds a given set of seals to the file, it never touches
1975 * existing seals. Furthermore, the "setting seals"-operation can be
1976 * sealed itself, which basically prevents any further seal from being
1979 * Semantics of sealing are only defined on volatile files. Only
1980 * anonymous shmem files support sealing. More importantly, seals are
1981 * never written to disk. Therefore, there's no plan to support it on
1985 if (file
->f_op
!= &shmem_file_operations
)
1987 if (!(file
->f_mode
& FMODE_WRITE
))
1989 if (seals
& ~(unsigned int)F_ALL_SEALS
)
1992 mutex_lock(&inode
->i_mutex
);
1994 if (info
->seals
& F_SEAL_SEAL
) {
1999 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2000 error
= mapping_deny_writable(file
->f_mapping
);
2004 error
= shmem_wait_for_pins(file
->f_mapping
);
2006 mapping_allow_writable(file
->f_mapping
);
2011 info
->seals
|= seals
;
2015 mutex_unlock(&inode
->i_mutex
);
2018 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2020 int shmem_get_seals(struct file
*file
)
2022 if (file
->f_op
!= &shmem_file_operations
)
2025 return SHMEM_I(file_inode(file
))->seals
;
2027 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2029 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2035 /* disallow upper 32bit */
2039 error
= shmem_add_seals(file
, arg
);
2042 error
= shmem_get_seals(file
);
2052 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2055 struct inode
*inode
= file_inode(file
);
2056 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2057 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2058 struct shmem_falloc shmem_falloc
;
2059 pgoff_t start
, index
, end
;
2062 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2065 mutex_lock(&inode
->i_mutex
);
2067 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2068 struct address_space
*mapping
= file
->f_mapping
;
2069 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2070 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2071 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2073 /* protected by i_mutex */
2074 if (info
->seals
& F_SEAL_WRITE
) {
2079 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2080 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2081 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2082 spin_lock(&inode
->i_lock
);
2083 inode
->i_private
= &shmem_falloc
;
2084 spin_unlock(&inode
->i_lock
);
2086 if ((u64
)unmap_end
> (u64
)unmap_start
)
2087 unmap_mapping_range(mapping
, unmap_start
,
2088 1 + unmap_end
- unmap_start
, 0);
2089 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2090 /* No need to unmap again: hole-punching leaves COWed pages */
2092 spin_lock(&inode
->i_lock
);
2093 inode
->i_private
= NULL
;
2094 wake_up_all(&shmem_falloc_waitq
);
2095 spin_unlock(&inode
->i_lock
);
2100 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2101 error
= inode_newsize_ok(inode
, offset
+ len
);
2105 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2110 start
= offset
>> PAGE_CACHE_SHIFT
;
2111 end
= (offset
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
2112 /* Try to avoid a swapstorm if len is impossible to satisfy */
2113 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2118 shmem_falloc
.waitq
= NULL
;
2119 shmem_falloc
.start
= start
;
2120 shmem_falloc
.next
= start
;
2121 shmem_falloc
.nr_falloced
= 0;
2122 shmem_falloc
.nr_unswapped
= 0;
2123 spin_lock(&inode
->i_lock
);
2124 inode
->i_private
= &shmem_falloc
;
2125 spin_unlock(&inode
->i_lock
);
2127 for (index
= start
; index
< end
; index
++) {
2131 * Good, the fallocate(2) manpage permits EINTR: we may have
2132 * been interrupted because we are using up too much memory.
2134 if (signal_pending(current
))
2136 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2139 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
,
2142 /* Remove the !PageUptodate pages we added */
2143 shmem_undo_range(inode
,
2144 (loff_t
)start
<< PAGE_CACHE_SHIFT
,
2145 (loff_t
)index
<< PAGE_CACHE_SHIFT
, true);
2150 * Inform shmem_writepage() how far we have reached.
2151 * No need for lock or barrier: we have the page lock.
2153 shmem_falloc
.next
++;
2154 if (!PageUptodate(page
))
2155 shmem_falloc
.nr_falloced
++;
2158 * If !PageUptodate, leave it that way so that freeable pages
2159 * can be recognized if we need to rollback on error later.
2160 * But set_page_dirty so that memory pressure will swap rather
2161 * than free the pages we are allocating (and SGP_CACHE pages
2162 * might still be clean: we now need to mark those dirty too).
2164 set_page_dirty(page
);
2166 page_cache_release(page
);
2170 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2171 i_size_write(inode
, offset
+ len
);
2172 inode
->i_ctime
= CURRENT_TIME
;
2174 spin_lock(&inode
->i_lock
);
2175 inode
->i_private
= NULL
;
2176 spin_unlock(&inode
->i_lock
);
2178 mutex_unlock(&inode
->i_mutex
);
2182 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2184 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2186 buf
->f_type
= TMPFS_MAGIC
;
2187 buf
->f_bsize
= PAGE_CACHE_SIZE
;
2188 buf
->f_namelen
= NAME_MAX
;
2189 if (sbinfo
->max_blocks
) {
2190 buf
->f_blocks
= sbinfo
->max_blocks
;
2192 buf
->f_bfree
= sbinfo
->max_blocks
-
2193 percpu_counter_sum(&sbinfo
->used_blocks
);
2195 if (sbinfo
->max_inodes
) {
2196 buf
->f_files
= sbinfo
->max_inodes
;
2197 buf
->f_ffree
= sbinfo
->free_inodes
;
2199 /* else leave those fields 0 like simple_statfs */
2204 * File creation. Allocate an inode, and we're done..
2207 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2209 struct inode
*inode
;
2210 int error
= -ENOSPC
;
2212 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2214 error
= simple_acl_create(dir
, inode
);
2217 error
= security_inode_init_security(inode
, dir
,
2219 shmem_initxattrs
, NULL
);
2220 if (error
&& error
!= -EOPNOTSUPP
)
2224 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2225 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2226 d_instantiate(dentry
, inode
);
2227 dget(dentry
); /* Extra count - pin the dentry in core */
2236 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2238 struct inode
*inode
;
2239 int error
= -ENOSPC
;
2241 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2243 error
= security_inode_init_security(inode
, dir
,
2245 shmem_initxattrs
, NULL
);
2246 if (error
&& error
!= -EOPNOTSUPP
)
2248 error
= simple_acl_create(dir
, inode
);
2251 d_tmpfile(dentry
, inode
);
2259 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2263 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2269 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2272 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2278 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2280 struct inode
*inode
= old_dentry
->d_inode
;
2284 * No ordinary (disk based) filesystem counts links as inodes;
2285 * but each new link needs a new dentry, pinning lowmem, and
2286 * tmpfs dentries cannot be pruned until they are unlinked.
2288 ret
= shmem_reserve_inode(inode
->i_sb
);
2292 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2293 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2295 ihold(inode
); /* New dentry reference */
2296 dget(dentry
); /* Extra pinning count for the created dentry */
2297 d_instantiate(dentry
, inode
);
2302 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2304 struct inode
*inode
= dentry
->d_inode
;
2306 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2307 shmem_free_inode(inode
->i_sb
);
2309 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2310 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2312 dput(dentry
); /* Undo the count from "create" - this does all the work */
2316 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2318 if (!simple_empty(dentry
))
2321 drop_nlink(dentry
->d_inode
);
2323 return shmem_unlink(dir
, dentry
);
2326 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2328 bool old_is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2329 bool new_is_dir
= S_ISDIR(new_dentry
->d_inode
->i_mode
);
2331 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2333 drop_nlink(old_dir
);
2336 drop_nlink(new_dir
);
2340 old_dir
->i_ctime
= old_dir
->i_mtime
=
2341 new_dir
->i_ctime
= new_dir
->i_mtime
=
2342 old_dentry
->d_inode
->i_ctime
=
2343 new_dentry
->d_inode
->i_ctime
= CURRENT_TIME
;
2349 * The VFS layer already does all the dentry stuff for rename,
2350 * we just have to decrement the usage count for the target if
2351 * it exists so that the VFS layer correctly free's it when it
2354 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
2356 struct inode
*inode
= old_dentry
->d_inode
;
2357 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2359 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
))
2362 if (flags
& RENAME_EXCHANGE
)
2363 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
2365 if (!simple_empty(new_dentry
))
2368 if (new_dentry
->d_inode
) {
2369 (void) shmem_unlink(new_dir
, new_dentry
);
2370 if (they_are_dirs
) {
2371 drop_nlink(new_dentry
->d_inode
);
2372 drop_nlink(old_dir
);
2374 } else if (they_are_dirs
) {
2375 drop_nlink(old_dir
);
2379 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2380 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2381 old_dir
->i_ctime
= old_dir
->i_mtime
=
2382 new_dir
->i_ctime
= new_dir
->i_mtime
=
2383 inode
->i_ctime
= CURRENT_TIME
;
2387 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2391 struct inode
*inode
;
2394 struct shmem_inode_info
*info
;
2396 len
= strlen(symname
) + 1;
2397 if (len
> PAGE_CACHE_SIZE
)
2398 return -ENAMETOOLONG
;
2400 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2404 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2405 shmem_initxattrs
, NULL
);
2407 if (error
!= -EOPNOTSUPP
) {
2414 info
= SHMEM_I(inode
);
2415 inode
->i_size
= len
-1;
2416 if (len
<= SHORT_SYMLINK_LEN
) {
2417 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
2418 if (!info
->symlink
) {
2422 inode
->i_op
= &shmem_short_symlink_operations
;
2424 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
2429 inode
->i_mapping
->a_ops
= &shmem_aops
;
2430 inode
->i_op
= &shmem_symlink_inode_operations
;
2431 kaddr
= kmap_atomic(page
);
2432 memcpy(kaddr
, symname
, len
);
2433 kunmap_atomic(kaddr
);
2434 SetPageUptodate(page
);
2435 set_page_dirty(page
);
2437 page_cache_release(page
);
2439 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2440 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2441 d_instantiate(dentry
, inode
);
2446 static void *shmem_follow_short_symlink(struct dentry
*dentry
, struct nameidata
*nd
)
2448 nd_set_link(nd
, SHMEM_I(dentry
->d_inode
)->symlink
);
2452 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
2454 struct page
*page
= NULL
;
2455 int error
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
2456 nd_set_link(nd
, error
? ERR_PTR(error
) : kmap(page
));
2462 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2464 if (!IS_ERR(nd_get_link(nd
))) {
2465 struct page
*page
= cookie
;
2467 mark_page_accessed(page
);
2468 page_cache_release(page
);
2472 #ifdef CONFIG_TMPFS_XATTR
2474 * Superblocks without xattr inode operations may get some security.* xattr
2475 * support from the LSM "for free". As soon as we have any other xattrs
2476 * like ACLs, we also need to implement the security.* handlers at
2477 * filesystem level, though.
2481 * Callback for security_inode_init_security() for acquiring xattrs.
2483 static int shmem_initxattrs(struct inode
*inode
,
2484 const struct xattr
*xattr_array
,
2487 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2488 const struct xattr
*xattr
;
2489 struct simple_xattr
*new_xattr
;
2492 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2493 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
2497 len
= strlen(xattr
->name
) + 1;
2498 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2500 if (!new_xattr
->name
) {
2505 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2506 XATTR_SECURITY_PREFIX_LEN
);
2507 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2510 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
2516 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2517 #ifdef CONFIG_TMPFS_POSIX_ACL
2518 &posix_acl_access_xattr_handler
,
2519 &posix_acl_default_xattr_handler
,
2524 static int shmem_xattr_validate(const char *name
)
2526 struct { const char *prefix
; size_t len
; } arr
[] = {
2527 { XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
},
2528 { XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
}
2532 for (i
= 0; i
< ARRAY_SIZE(arr
); i
++) {
2533 size_t preflen
= arr
[i
].len
;
2534 if (strncmp(name
, arr
[i
].prefix
, preflen
) == 0) {
2543 static ssize_t
shmem_getxattr(struct dentry
*dentry
, const char *name
,
2544 void *buffer
, size_t size
)
2546 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2550 * If this is a request for a synthetic attribute in the system.*
2551 * namespace use the generic infrastructure to resolve a handler
2552 * for it via sb->s_xattr.
2554 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2555 return generic_getxattr(dentry
, name
, buffer
, size
);
2557 err
= shmem_xattr_validate(name
);
2561 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
2564 static int shmem_setxattr(struct dentry
*dentry
, const char *name
,
2565 const void *value
, size_t size
, int flags
)
2567 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2571 * If this is a request for a synthetic attribute in the system.*
2572 * namespace use the generic infrastructure to resolve a handler
2573 * for it via sb->s_xattr.
2575 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2576 return generic_setxattr(dentry
, name
, value
, size
, flags
);
2578 err
= shmem_xattr_validate(name
);
2582 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
2585 static int shmem_removexattr(struct dentry
*dentry
, const char *name
)
2587 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2591 * If this is a request for a synthetic attribute in the system.*
2592 * namespace use the generic infrastructure to resolve a handler
2593 * for it via sb->s_xattr.
2595 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2596 return generic_removexattr(dentry
, name
);
2598 err
= shmem_xattr_validate(name
);
2602 return simple_xattr_remove(&info
->xattrs
, name
);
2605 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2607 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2608 return simple_xattr_list(&info
->xattrs
, buffer
, size
);
2610 #endif /* CONFIG_TMPFS_XATTR */
2612 static const struct inode_operations shmem_short_symlink_operations
= {
2613 .readlink
= generic_readlink
,
2614 .follow_link
= shmem_follow_short_symlink
,
2615 #ifdef CONFIG_TMPFS_XATTR
2616 .setxattr
= shmem_setxattr
,
2617 .getxattr
= shmem_getxattr
,
2618 .listxattr
= shmem_listxattr
,
2619 .removexattr
= shmem_removexattr
,
2623 static const struct inode_operations shmem_symlink_inode_operations
= {
2624 .readlink
= generic_readlink
,
2625 .follow_link
= shmem_follow_link
,
2626 .put_link
= shmem_put_link
,
2627 #ifdef CONFIG_TMPFS_XATTR
2628 .setxattr
= shmem_setxattr
,
2629 .getxattr
= shmem_getxattr
,
2630 .listxattr
= shmem_listxattr
,
2631 .removexattr
= shmem_removexattr
,
2635 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2637 return ERR_PTR(-ESTALE
);
2640 static int shmem_match(struct inode
*ino
, void *vfh
)
2644 inum
= (inum
<< 32) | fh
[1];
2645 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2648 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2649 struct fid
*fid
, int fh_len
, int fh_type
)
2651 struct inode
*inode
;
2652 struct dentry
*dentry
= NULL
;
2659 inum
= (inum
<< 32) | fid
->raw
[1];
2661 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2662 shmem_match
, fid
->raw
);
2664 dentry
= d_find_alias(inode
);
2671 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
2672 struct inode
*parent
)
2676 return FILEID_INVALID
;
2679 if (inode_unhashed(inode
)) {
2680 /* Unfortunately insert_inode_hash is not idempotent,
2681 * so as we hash inodes here rather than at creation
2682 * time, we need a lock to ensure we only try
2685 static DEFINE_SPINLOCK(lock
);
2687 if (inode_unhashed(inode
))
2688 __insert_inode_hash(inode
,
2689 inode
->i_ino
+ inode
->i_generation
);
2693 fh
[0] = inode
->i_generation
;
2694 fh
[1] = inode
->i_ino
;
2695 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2701 static const struct export_operations shmem_export_ops
= {
2702 .get_parent
= shmem_get_parent
,
2703 .encode_fh
= shmem_encode_fh
,
2704 .fh_to_dentry
= shmem_fh_to_dentry
,
2707 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2710 char *this_char
, *value
, *rest
;
2711 struct mempolicy
*mpol
= NULL
;
2715 while (options
!= NULL
) {
2716 this_char
= options
;
2719 * NUL-terminate this option: unfortunately,
2720 * mount options form a comma-separated list,
2721 * but mpol's nodelist may also contain commas.
2723 options
= strchr(options
, ',');
2724 if (options
== NULL
)
2727 if (!isdigit(*options
)) {
2734 if ((value
= strchr(this_char
,'=')) != NULL
) {
2738 "tmpfs: No value for mount option '%s'\n",
2743 if (!strcmp(this_char
,"size")) {
2744 unsigned long long size
;
2745 size
= memparse(value
,&rest
);
2747 size
<<= PAGE_SHIFT
;
2748 size
*= totalram_pages
;
2754 sbinfo
->max_blocks
=
2755 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2756 } else if (!strcmp(this_char
,"nr_blocks")) {
2757 sbinfo
->max_blocks
= memparse(value
, &rest
);
2760 } else if (!strcmp(this_char
,"nr_inodes")) {
2761 sbinfo
->max_inodes
= memparse(value
, &rest
);
2764 } else if (!strcmp(this_char
,"mode")) {
2767 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2770 } else if (!strcmp(this_char
,"uid")) {
2773 uid
= simple_strtoul(value
, &rest
, 0);
2776 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
2777 if (!uid_valid(sbinfo
->uid
))
2779 } else if (!strcmp(this_char
,"gid")) {
2782 gid
= simple_strtoul(value
, &rest
, 0);
2785 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
2786 if (!gid_valid(sbinfo
->gid
))
2788 } else if (!strcmp(this_char
,"mpol")) {
2791 if (mpol_parse_str(value
, &mpol
))
2794 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2799 sbinfo
->mpol
= mpol
;
2803 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2811 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2813 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2814 struct shmem_sb_info config
= *sbinfo
;
2815 unsigned long inodes
;
2816 int error
= -EINVAL
;
2819 if (shmem_parse_options(data
, &config
, true))
2822 spin_lock(&sbinfo
->stat_lock
);
2823 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2824 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2826 if (config
.max_inodes
< inodes
)
2829 * Those tests disallow limited->unlimited while any are in use;
2830 * but we must separately disallow unlimited->limited, because
2831 * in that case we have no record of how much is already in use.
2833 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2835 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2839 sbinfo
->max_blocks
= config
.max_blocks
;
2840 sbinfo
->max_inodes
= config
.max_inodes
;
2841 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2844 * Preserve previous mempolicy unless mpol remount option was specified.
2847 mpol_put(sbinfo
->mpol
);
2848 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2851 spin_unlock(&sbinfo
->stat_lock
);
2855 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2857 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2859 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2860 seq_printf(seq
, ",size=%luk",
2861 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2862 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2863 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2864 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2865 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2866 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
2867 seq_printf(seq
, ",uid=%u",
2868 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
2869 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
2870 seq_printf(seq
, ",gid=%u",
2871 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
2872 shmem_show_mpol(seq
, sbinfo
->mpol
);
2876 #define MFD_NAME_PREFIX "memfd:"
2877 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2878 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2880 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2882 SYSCALL_DEFINE2(memfd_create
,
2883 const char __user
*, uname
,
2884 unsigned int, flags
)
2886 struct shmem_inode_info
*info
;
2892 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
2895 /* length includes terminating zero */
2896 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
2899 if (len
> MFD_NAME_MAX_LEN
+ 1)
2902 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
2906 strcpy(name
, MFD_NAME_PREFIX
);
2907 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
2912 /* terminating-zero may have changed after strnlen_user() returned */
2913 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
2918 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
2924 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
2926 error
= PTR_ERR(file
);
2929 info
= SHMEM_I(file_inode(file
));
2930 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
2931 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
2932 if (flags
& MFD_ALLOW_SEALING
)
2933 info
->seals
&= ~F_SEAL_SEAL
;
2935 fd_install(fd
, file
);
2946 #endif /* CONFIG_TMPFS */
2948 static void shmem_put_super(struct super_block
*sb
)
2950 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2952 percpu_counter_destroy(&sbinfo
->used_blocks
);
2953 mpol_put(sbinfo
->mpol
);
2955 sb
->s_fs_info
= NULL
;
2958 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
2960 struct inode
*inode
;
2961 struct shmem_sb_info
*sbinfo
;
2964 /* Round up to L1_CACHE_BYTES to resist false sharing */
2965 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
2966 L1_CACHE_BYTES
), GFP_KERNEL
);
2970 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
2971 sbinfo
->uid
= current_fsuid();
2972 sbinfo
->gid
= current_fsgid();
2973 sb
->s_fs_info
= sbinfo
;
2977 * Per default we only allow half of the physical ram per
2978 * tmpfs instance, limiting inodes to one per page of lowmem;
2979 * but the internal instance is left unlimited.
2981 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
2982 sbinfo
->max_blocks
= shmem_default_max_blocks();
2983 sbinfo
->max_inodes
= shmem_default_max_inodes();
2984 if (shmem_parse_options(data
, sbinfo
, false)) {
2989 sb
->s_flags
|= MS_NOUSER
;
2991 sb
->s_export_op
= &shmem_export_ops
;
2992 sb
->s_flags
|= MS_NOSEC
;
2994 sb
->s_flags
|= MS_NOUSER
;
2997 spin_lock_init(&sbinfo
->stat_lock
);
2998 if (percpu_counter_init(&sbinfo
->used_blocks
, 0))
3000 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3002 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3003 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
3004 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
3005 sb
->s_magic
= TMPFS_MAGIC
;
3006 sb
->s_op
= &shmem_ops
;
3007 sb
->s_time_gran
= 1;
3008 #ifdef CONFIG_TMPFS_XATTR
3009 sb
->s_xattr
= shmem_xattr_handlers
;
3011 #ifdef CONFIG_TMPFS_POSIX_ACL
3012 sb
->s_flags
|= MS_POSIXACL
;
3015 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3018 inode
->i_uid
= sbinfo
->uid
;
3019 inode
->i_gid
= sbinfo
->gid
;
3020 sb
->s_root
= d_make_root(inode
);
3026 shmem_put_super(sb
);
3030 static struct kmem_cache
*shmem_inode_cachep
;
3032 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3034 struct shmem_inode_info
*info
;
3035 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3038 return &info
->vfs_inode
;
3041 static void shmem_destroy_callback(struct rcu_head
*head
)
3043 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3044 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3047 static void shmem_destroy_inode(struct inode
*inode
)
3049 if (S_ISREG(inode
->i_mode
))
3050 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3051 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3054 static void shmem_init_inode(void *foo
)
3056 struct shmem_inode_info
*info
= foo
;
3057 inode_init_once(&info
->vfs_inode
);
3060 static int shmem_init_inodecache(void)
3062 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3063 sizeof(struct shmem_inode_info
),
3064 0, SLAB_PANIC
, shmem_init_inode
);
3068 static void shmem_destroy_inodecache(void)
3070 kmem_cache_destroy(shmem_inode_cachep
);
3073 static const struct address_space_operations shmem_aops
= {
3074 .writepage
= shmem_writepage
,
3075 .set_page_dirty
= __set_page_dirty_no_writeback
,
3077 .write_begin
= shmem_write_begin
,
3078 .write_end
= shmem_write_end
,
3080 #ifdef CONFIG_MIGRATION
3081 .migratepage
= migrate_page
,
3083 .error_remove_page
= generic_error_remove_page
,
3086 static const struct file_operations shmem_file_operations
= {
3089 .llseek
= shmem_file_llseek
,
3090 .read
= new_sync_read
,
3091 .write
= new_sync_write
,
3092 .read_iter
= shmem_file_read_iter
,
3093 .write_iter
= generic_file_write_iter
,
3094 .fsync
= noop_fsync
,
3095 .splice_read
= shmem_file_splice_read
,
3096 .splice_write
= iter_file_splice_write
,
3097 .fallocate
= shmem_fallocate
,
3101 static const struct inode_operations shmem_inode_operations
= {
3102 .setattr
= shmem_setattr
,
3103 #ifdef CONFIG_TMPFS_XATTR
3104 .setxattr
= shmem_setxattr
,
3105 .getxattr
= shmem_getxattr
,
3106 .listxattr
= shmem_listxattr
,
3107 .removexattr
= shmem_removexattr
,
3108 .set_acl
= simple_set_acl
,
3112 static const struct inode_operations shmem_dir_inode_operations
= {
3114 .create
= shmem_create
,
3115 .lookup
= simple_lookup
,
3117 .unlink
= shmem_unlink
,
3118 .symlink
= shmem_symlink
,
3119 .mkdir
= shmem_mkdir
,
3120 .rmdir
= shmem_rmdir
,
3121 .mknod
= shmem_mknod
,
3122 .rename2
= shmem_rename2
,
3123 .tmpfile
= shmem_tmpfile
,
3125 #ifdef CONFIG_TMPFS_XATTR
3126 .setxattr
= shmem_setxattr
,
3127 .getxattr
= shmem_getxattr
,
3128 .listxattr
= shmem_listxattr
,
3129 .removexattr
= shmem_removexattr
,
3131 #ifdef CONFIG_TMPFS_POSIX_ACL
3132 .setattr
= shmem_setattr
,
3133 .set_acl
= simple_set_acl
,
3137 static const struct inode_operations shmem_special_inode_operations
= {
3138 #ifdef CONFIG_TMPFS_XATTR
3139 .setxattr
= shmem_setxattr
,
3140 .getxattr
= shmem_getxattr
,
3141 .listxattr
= shmem_listxattr
,
3142 .removexattr
= shmem_removexattr
,
3144 #ifdef CONFIG_TMPFS_POSIX_ACL
3145 .setattr
= shmem_setattr
,
3146 .set_acl
= simple_set_acl
,
3150 static const struct super_operations shmem_ops
= {
3151 .alloc_inode
= shmem_alloc_inode
,
3152 .destroy_inode
= shmem_destroy_inode
,
3154 .statfs
= shmem_statfs
,
3155 .remount_fs
= shmem_remount_fs
,
3156 .show_options
= shmem_show_options
,
3158 .evict_inode
= shmem_evict_inode
,
3159 .drop_inode
= generic_delete_inode
,
3160 .put_super
= shmem_put_super
,
3163 static const struct vm_operations_struct shmem_vm_ops
= {
3164 .fault
= shmem_fault
,
3165 .map_pages
= filemap_map_pages
,
3167 .set_policy
= shmem_set_policy
,
3168 .get_policy
= shmem_get_policy
,
3170 .remap_pages
= generic_file_remap_pages
,
3173 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3174 int flags
, const char *dev_name
, void *data
)
3176 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3179 static struct file_system_type shmem_fs_type
= {
3180 .owner
= THIS_MODULE
,
3182 .mount
= shmem_mount
,
3183 .kill_sb
= kill_litter_super
,
3184 .fs_flags
= FS_USERNS_MOUNT
,
3187 int __init
shmem_init(void)
3191 /* If rootfs called this, don't re-init */
3192 if (shmem_inode_cachep
)
3195 error
= bdi_init(&shmem_backing_dev_info
);
3199 error
= shmem_init_inodecache();
3203 error
= register_filesystem(&shmem_fs_type
);
3205 printk(KERN_ERR
"Could not register tmpfs\n");
3209 shm_mnt
= kern_mount(&shmem_fs_type
);
3210 if (IS_ERR(shm_mnt
)) {
3211 error
= PTR_ERR(shm_mnt
);
3212 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
3218 unregister_filesystem(&shmem_fs_type
);
3220 shmem_destroy_inodecache();
3222 bdi_destroy(&shmem_backing_dev_info
);
3224 shm_mnt
= ERR_PTR(error
);
3228 #else /* !CONFIG_SHMEM */
3231 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3233 * This is intended for small system where the benefits of the full
3234 * shmem code (swap-backed and resource-limited) are outweighed by
3235 * their complexity. On systems without swap this code should be
3236 * effectively equivalent, but much lighter weight.
3239 static struct file_system_type shmem_fs_type
= {
3241 .mount
= ramfs_mount
,
3242 .kill_sb
= kill_litter_super
,
3243 .fs_flags
= FS_USERNS_MOUNT
,
3246 int __init
shmem_init(void)
3248 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
3250 shm_mnt
= kern_mount(&shmem_fs_type
);
3251 BUG_ON(IS_ERR(shm_mnt
));
3256 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
3261 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
3266 void shmem_unlock_mapping(struct address_space
*mapping
)
3270 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
3272 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
3274 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
3276 #define shmem_vm_ops generic_file_vm_ops
3277 #define shmem_file_operations ramfs_file_operations
3278 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3279 #define shmem_acct_size(flags, size) 0
3280 #define shmem_unacct_size(flags, size) do {} while (0)
3282 #endif /* CONFIG_SHMEM */
3286 static struct dentry_operations anon_ops
= {
3287 .d_dname
= simple_dname
3290 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
3291 unsigned long flags
, unsigned int i_flags
)
3294 struct inode
*inode
;
3296 struct super_block
*sb
;
3299 if (IS_ERR(shm_mnt
))
3300 return ERR_CAST(shm_mnt
);
3302 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
3303 return ERR_PTR(-EINVAL
);
3305 if (shmem_acct_size(flags
, size
))
3306 return ERR_PTR(-ENOMEM
);
3308 res
= ERR_PTR(-ENOMEM
);
3310 this.len
= strlen(name
);
3311 this.hash
= 0; /* will go */
3312 sb
= shm_mnt
->mnt_sb
;
3313 path
.mnt
= mntget(shm_mnt
);
3314 path
.dentry
= d_alloc_pseudo(sb
, &this);
3317 d_set_d_op(path
.dentry
, &anon_ops
);
3319 res
= ERR_PTR(-ENOSPC
);
3320 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
3324 inode
->i_flags
|= i_flags
;
3325 d_instantiate(path
.dentry
, inode
);
3326 inode
->i_size
= size
;
3327 clear_nlink(inode
); /* It is unlinked */
3328 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
3332 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
3333 &shmem_file_operations
);
3340 shmem_unacct_size(flags
, size
);
3347 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3348 * kernel internal. There will be NO LSM permission checks against the
3349 * underlying inode. So users of this interface must do LSM checks at a
3350 * higher layer. The one user is the big_key implementation. LSM checks
3351 * are provided at the key level rather than the inode level.
3352 * @name: name for dentry (to be seen in /proc/<pid>/maps
3353 * @size: size to be set for the file
3354 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3356 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3358 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
3362 * shmem_file_setup - get an unlinked file living in tmpfs
3363 * @name: name for dentry (to be seen in /proc/<pid>/maps
3364 * @size: size to be set for the file
3365 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3367 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3369 return __shmem_file_setup(name
, size
, flags
, 0);
3371 EXPORT_SYMBOL_GPL(shmem_file_setup
);
3374 * shmem_zero_setup - setup a shared anonymous mapping
3375 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3377 int shmem_zero_setup(struct vm_area_struct
*vma
)
3380 loff_t size
= vma
->vm_end
- vma
->vm_start
;
3382 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
);
3384 return PTR_ERR(file
);
3388 vma
->vm_file
= file
;
3389 vma
->vm_ops
= &shmem_vm_ops
;
3394 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3395 * @mapping: the page's address_space
3396 * @index: the page index
3397 * @gfp: the page allocator flags to use if allocating
3399 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3400 * with any new page allocations done using the specified allocation flags.
3401 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3402 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3403 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3405 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3406 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3408 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
3409 pgoff_t index
, gfp_t gfp
)
3412 struct inode
*inode
= mapping
->host
;
3416 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
3417 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
3419 page
= ERR_PTR(error
);
3425 * The tiny !SHMEM case uses ramfs without swap
3427 return read_cache_page_gfp(mapping
, index
, gfp
);
3430 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);